; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -global-isel=0 -amdgpu-load-store-vectorizer=0 -mtriple=amdgcn -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX10,GFX10-SDAG %s
; RUN: llc -global-isel=1 -global-isel-abort=2 -amdgpu-load-store-vectorizer=0 -mtriple=amdgcn -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX10,GFX10-GISEL %s
; RUN: llc -global-isel=0 -amdgpu-load-store-vectorizer=0 -mtriple=amdgcn -mcpu=gfx1100 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX11,GFX11-SDAG %s
; RUN: llc -global-isel=1 -global-isel-abort=2 -amdgpu-load-store-vectorizer=0 -mtriple=amdgcn -mcpu=gfx1100 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX11,GFX11-GISEL %s
; RUN: llc -global-isel=0 -amdgpu-load-store-vectorizer=0 -mtriple=amdgcn -mcpu=gfx1200 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX12,GFX12-SDAG %s
; RUN: llc -global-isel=1 -global-isel-abort=2 -amdgpu-load-store-vectorizer=0 -mtriple=amdgcn -mcpu=gfx1200 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL %s

declare i32 @llvm.amdgcn.permlane16(i32, i32, i32, i32, i1, i1)
declare i32 @llvm.amdgcn.permlanex16(i32, i32, i32, i32, i1, i1)
declare i32 @llvm.amdgcn.workitem.id.x()
declare i32 @llvm.amdgcn.workitem.id.y()

define amdgpu_kernel void @v_permlane16_b32_vss_i32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlane16_b32_vss_i32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-NEXT:    s_load_dword s6, s[4:5], 0x34
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-NEXT:    v_permlane16_b32 v0, v0, s3, s6
; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlane16_b32_vss_i32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlane16_b32 v0, v0, s3, s4
; GFX11-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlane16_b32_vss_i32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlane16_b32 v0, v0, s3, s4
; GFX12-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-NEXT:    s_endpgm
  %v = call i32 @llvm.amdgcn.permlane16.i32(i32 %src0, i32 %src0, i32 %src1, i32 %src2, i1 false, i1 false)
  store i32 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_vss_f32(ptr addrspace(1) %out, float %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlane16_b32_vss_f32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-NEXT:    s_load_dword s6, s[4:5], 0x34
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-NEXT:    v_permlane16_b32 v0, v0, s3, s6
; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlane16_b32_vss_f32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlane16_b32 v0, v0, s3, s4
; GFX11-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlane16_b32_vss_f32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlane16_b32 v0, v0, s3, s4
; GFX12-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-NEXT:    s_endpgm
  %v = call float @llvm.amdgcn.permlane16.f32(float %src0, float %src0, i32 %src1, i32 %src2, i1 false, i1 false)
  store float %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_vss_i64(ptr addrspace(1) %out, i64 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlane16_b32_vss_i64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_permlane16_b32 v1, v1, s6, s7
; GFX10-SDAG-NEXT:    v_permlane16_b32 v0, v0, s6, s7
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_vss_i64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    v_permlane16_b32 v0, v0, s6, s7
; GFX10-GISEL-NEXT:    v_permlane16_b32 v1, v1, s6, s7
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlane16_b32_vss_i64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v1, v1, s4, s5
; GFX11-SDAG-NEXT:    v_permlane16_b32 v0, v0, s4, s5
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlane16_b32_vss_i64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v0, v0, s4, s5
; GFX11-GISEL-NEXT:    v_permlane16_b32 v1, v1, s4, s5
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlane16_b32_vss_i64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlane16_b32 v1, v1, s4, s5
; GFX12-SDAG-NEXT:    v_permlane16_b32 v0, v0, s4, s5
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlane16_b32_vss_i64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlane16_b32 v0, v0, s4, s5
; GFX12-GISEL-NEXT:    v_permlane16_b32 v1, v1, s4, s5
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %v = call i64 @llvm.amdgcn.permlane16.i64(i64 %src0, i64 %src0, i32 %src1, i32 %src2, i1 false, i1 false)
  store i64 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_vss_f64(ptr addrspace(1) %out, double %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlane16_b32_vss_f64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_permlane16_b32 v1, v1, s6, s7
; GFX10-SDAG-NEXT:    v_permlane16_b32 v0, v0, s6, s7
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_vss_f64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    v_permlane16_b32 v0, v0, s6, s7
; GFX10-GISEL-NEXT:    v_permlane16_b32 v1, v1, s6, s7
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlane16_b32_vss_f64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v1, v1, s4, s5
; GFX11-SDAG-NEXT:    v_permlane16_b32 v0, v0, s4, s5
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlane16_b32_vss_f64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v0, v0, s4, s5
; GFX11-GISEL-NEXT:    v_permlane16_b32 v1, v1, s4, s5
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlane16_b32_vss_f64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlane16_b32 v1, v1, s4, s5
; GFX12-SDAG-NEXT:    v_permlane16_b32 v0, v0, s4, s5
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlane16_b32_vss_f64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlane16_b32 v0, v0, s4, s5
; GFX12-GISEL-NEXT:    v_permlane16_b32 v1, v1, s4, s5
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %v = call double @llvm.amdgcn.permlane16.f64(double %src0, double %src0, i32 %src1, i32 %src2, i1 false, i1 false)
  store double %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_vii_i32(ptr addrspace(1) %out, i32 %src0) {
; GFX10-LABEL: v_permlane16_b32_vii_i32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dword s2, s[4:5], 0x2c
; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-NEXT:    v_permlane16_b32 v0, v0, 1, 2
; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlane16_b32_vii_i32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b32 s2, s[4:5], 0x2c
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlane16_b32 v0, v0, 1, 2
; GFX11-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlane16_b32_vii_i32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_load_b96 s[0:2], s[4:5], 0x24
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlane16_b32 v0, v0, 1, 2
; GFX12-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-NEXT:    s_endpgm
  %v = call i32 @llvm.amdgcn.permlane16.i32(i32 %src0, i32 %src0, i32 1, i32 2, i1 false, i1 false)
  store i32 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_vii_f32(ptr addrspace(1) %out, float %src0) {
; GFX10-LABEL: v_permlane16_b32_vii_f32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dword s2, s[4:5], 0x2c
; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-NEXT:    v_permlane16_b32 v0, v0, 1, 2
; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlane16_b32_vii_f32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b32 s2, s[4:5], 0x2c
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlane16_b32 v0, v0, 1, 2
; GFX11-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlane16_b32_vii_f32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_load_b96 s[0:2], s[4:5], 0x24
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlane16_b32 v0, v0, 1, 2
; GFX12-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-NEXT:    s_endpgm
  %v = call float @llvm.amdgcn.permlane16.f32(float %src0, float %src0, i32 1, i32 2, i1 false, i1 false)
  store float %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_vii_i64(ptr addrspace(1) %out, i64 %src0) {
; GFX10-SDAG-LABEL: v_permlane16_b32_vii_i64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_permlane16_b32 v1, v1, 1, 2
; GFX10-SDAG-NEXT:    v_permlane16_b32 v0, v0, 1, 2
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_vii_i64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    v_permlane16_b32 v0, v0, 1, 2
; GFX10-GISEL-NEXT:    v_permlane16_b32 v1, v1, 1, 2
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlane16_b32_vii_i64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v1, v1, 1, 2
; GFX11-SDAG-NEXT:    v_permlane16_b32 v0, v0, 1, 2
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlane16_b32_vii_i64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v0, v0, 1, 2
; GFX11-GISEL-NEXT:    v_permlane16_b32 v1, v1, 1, 2
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlane16_b32_vii_i64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlane16_b32 v1, v1, 1, 2
; GFX12-SDAG-NEXT:    v_permlane16_b32 v0, v0, 1, 2
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlane16_b32_vii_i64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlane16_b32 v0, v0, 1, 2
; GFX12-GISEL-NEXT:    v_permlane16_b32 v1, v1, 1, 2
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %v = call i64 @llvm.amdgcn.permlane16.i64(i64 %src0, i64 %src0, i32 1, i32 2, i1 false, i1 false)
  store i64 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_vii_f64(ptr addrspace(1) %out, double %src0) {
; GFX10-SDAG-LABEL: v_permlane16_b32_vii_f64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_permlane16_b32 v1, v1, 1, 2
; GFX10-SDAG-NEXT:    v_permlane16_b32 v0, v0, 1, 2
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_vii_f64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    v_permlane16_b32 v0, v0, 1, 2
; GFX10-GISEL-NEXT:    v_permlane16_b32 v1, v1, 1, 2
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlane16_b32_vii_f64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v1, v1, 1, 2
; GFX11-SDAG-NEXT:    v_permlane16_b32 v0, v0, 1, 2
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlane16_b32_vii_f64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v0, v0, 1, 2
; GFX11-GISEL-NEXT:    v_permlane16_b32 v1, v1, 1, 2
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlane16_b32_vii_f64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlane16_b32 v1, v1, 1, 2
; GFX12-SDAG-NEXT:    v_permlane16_b32 v0, v0, 1, 2
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlane16_b32_vii_f64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlane16_b32 v0, v0, 1, 2
; GFX12-GISEL-NEXT:    v_permlane16_b32 v1, v1, 1, 2
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %v = call double @llvm.amdgcn.permlane16.f64(double %src0, double %src0, i32 1, i32 2, i1 false, i1 false)
  store double %v, ptr addrspace(1) %out
  ret void
}

; FIXME-GFX10PLUS: It is allowed to have both immediates as literals
define amdgpu_kernel void @v_permlane16_b32_vll_i32(ptr addrspace(1) %out, i32 %src0) {
; GFX10-LABEL: v_permlane16_b32_vll_i32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dword s2, s[4:5], 0x2c
; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-NEXT:    s_movk_i32 s2, 0x1234
; GFX10-NEXT:    v_permlane16_b32 v0, v0, s2, 0xc1d1
; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlane16_b32_vll_i32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b32 s2, s[4:5], 0x2c
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX11-NEXT:    s_movk_i32 s2, 0x1234
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
; GFX11-NEXT:    v_permlane16_b32 v0, v0, s2, 0xc1d1
; GFX11-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlane16_b32_vll_i32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_load_b96 s[0:2], s[4:5], 0x24
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX12-NEXT:    s_movk_i32 s2, 0x1234
; GFX12-NEXT:    s_wait_alu 0xfffe
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlane16_b32 v0, v0, s2, 0xc1d1
; GFX12-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-NEXT:    s_endpgm
  %v = call i32 @llvm.amdgcn.permlane16.i32(i32 %src0, i32 %src0, i32 4660, i32 49617, i1 false, i1 false)
  store i32 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_vll_i64(ptr addrspace(1) %out, i64 %src0) {
; GFX10-SDAG-LABEL: v_permlane16_b32_vll_i64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    s_movk_i32 s2, 0x1234
; GFX10-SDAG-NEXT:    v_permlane16_b32 v1, v1, s2, 0xc1d1
; GFX10-SDAG-NEXT:    v_permlane16_b32 v0, v0, s2, 0xc1d1
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_vll_i64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    s_movk_i32 s2, 0x1234
; GFX10-GISEL-NEXT:    v_permlane16_b32 v0, v0, s2, 0xc1d1
; GFX10-GISEL-NEXT:    v_permlane16_b32 v1, v1, s2, 0xc1d1
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlane16_b32_vll_i64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-SDAG-NEXT:    s_movk_i32 s2, 0x1234
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instid1(SALU_CYCLE_1)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v1, v1, s2, 0xc1d1
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v0, v0, s2, 0xc1d1
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlane16_b32_vll_i64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-GISEL-NEXT:    s_movk_i32 s2, 0x1234
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v0, v0, s2, 0xc1d1
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v1, v1, s2, 0xc1d1
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlane16_b32_vll_i64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-SDAG-NEXT:    s_movk_i32 s2, 0x1234
; GFX12-SDAG-NEXT:    s_wait_alu 0xfffe
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlane16_b32 v1, v1, s2, 0xc1d1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v0, v0, s2, 0xc1d1
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlane16_b32_vll_i64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX12-GISEL-NEXT:    s_movk_i32 s2, 0x1234
; GFX12-GISEL-NEXT:    s_wait_alu 0xfffe
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlane16_b32 v0, v0, s2, 0xc1d1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v1, v1, s2, 0xc1d1
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %v = call i64 @llvm.amdgcn.permlane16.i64(i64 %src0, i64 %src0, i32 4660, i32 49617, i1 false, i1 false)
  store i64 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_vll_f32(ptr addrspace(1) %out,float %src0) {
; GFX10-LABEL: v_permlane16_b32_vll_f32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dword s2, s[4:5], 0x2c
; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-NEXT:    s_movk_i32 s2, 0x1234
; GFX10-NEXT:    v_permlane16_b32 v0, v0, s2, 0xc1d1
; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlane16_b32_vll_f32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b32 s2, s[4:5], 0x2c
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX11-NEXT:    s_movk_i32 s2, 0x1234
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
; GFX11-NEXT:    v_permlane16_b32 v0, v0, s2, 0xc1d1
; GFX11-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlane16_b32_vll_f32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_load_b96 s[0:2], s[4:5], 0x24
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX12-NEXT:    s_movk_i32 s2, 0x1234
; GFX12-NEXT:    s_wait_alu 0xfffe
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlane16_b32 v0, v0, s2, 0xc1d1
; GFX12-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-NEXT:    s_endpgm
  %v = call float @llvm.amdgcn.permlane16.f32(float %src0, float %src0, i32 4660, i32 49617, i1 false, i1 false)
  store float %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_vll_f64(ptr addrspace(1) %out, double %src0) {
; GFX10-SDAG-LABEL: v_permlane16_b32_vll_f64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    s_movk_i32 s2, 0x1234
; GFX10-SDAG-NEXT:    v_permlane16_b32 v1, v1, s2, 0xc1d1
; GFX10-SDAG-NEXT:    v_permlane16_b32 v0, v0, s2, 0xc1d1
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_vll_f64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    s_movk_i32 s2, 0x1234
; GFX10-GISEL-NEXT:    v_permlane16_b32 v0, v0, s2, 0xc1d1
; GFX10-GISEL-NEXT:    v_permlane16_b32 v1, v1, s2, 0xc1d1
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlane16_b32_vll_f64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-SDAG-NEXT:    s_movk_i32 s2, 0x1234
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instid1(SALU_CYCLE_1)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v1, v1, s2, 0xc1d1
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v0, v0, s2, 0xc1d1
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlane16_b32_vll_f64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-GISEL-NEXT:    s_movk_i32 s2, 0x1234
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v0, v0, s2, 0xc1d1
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v1, v1, s2, 0xc1d1
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlane16_b32_vll_f64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-SDAG-NEXT:    s_movk_i32 s2, 0x1234
; GFX12-SDAG-NEXT:    s_wait_alu 0xfffe
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlane16_b32 v1, v1, s2, 0xc1d1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v0, v0, s2, 0xc1d1
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlane16_b32_vll_f64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX12-GISEL-NEXT:    s_movk_i32 s2, 0x1234
; GFX12-GISEL-NEXT:    s_wait_alu 0xfffe
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlane16_b32 v0, v0, s2, 0xc1d1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v1, v1, s2, 0xc1d1
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %v = call double @llvm.amdgcn.permlane16.f64(double %src0, double %src0, i32 4660, i32 49617, i1 false, i1 false)
  store double %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_vvv_i32(ptr addrspace(1) %out, i32 %src0) {
; GFX10-SDAG-LABEL: v_permlane16_b32_vvv_i32:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dword s2, s[4:5], 0x2c
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s3, v0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s2, v1
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-SDAG-NEXT:    v_permlane16_b32 v0, v0, s3, s2
; GFX10-SDAG-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_vvv_i32:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dword s2, s[4:5], 0x2c
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s3, v0
; GFX10-GISEL-NEXT:    s_mov_b32 null, 0
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v1
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_permlane16_b32 v0, v0, s3, s4
; GFX10-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlane16_b32_vvv_i32:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b32 s2, s[4:5], 0x2c
; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_and_b32_e32 v1, 0x3ff, v0
; GFX11-SDAG-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_4) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s3, v1
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v1, s2
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s2, v0
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, 0
; GFX11-SDAG-NEXT:    v_permlane16_b32 v1, v1, s3, s2
; GFX11-SDAG-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlane16_b32_vvv_i32:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b32 s2, s[4:5], 0x2c
; GFX11-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_and_b32_e32 v1, 0x3ff, v0
; GFX11-GISEL-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s3, v1
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v1, 0
; GFX11-GISEL-NEXT:    v_permlane16_b32 v0, v0, s3, s4
; GFX11-GISEL-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlane16_b32_vvv_i32:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_load_b96 s[0:2], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_and_b32_e32 v1, 0x3ff, v0
; GFX12-SDAG-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s3, v1
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v1, s2
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s2, v0
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, 0
; GFX12-SDAG-NEXT:    s_wait_alu 0xf1ff
; GFX12-SDAG-NEXT:    v_permlane16_b32 v1, v1, s3, s2
; GFX12-SDAG-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlane16_b32_vvv_i32:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_load_b96 s[0:2], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_and_b32_e32 v1, 0x3ff, v0
; GFX12-GISEL-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s3, v1
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v1, 0
; GFX12-GISEL-NEXT:    s_wait_alu 0xf1ff
; GFX12-GISEL-NEXT:    v_permlane16_b32 v0, v0, s3, s4
; GFX12-GISEL-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidy = call i32 @llvm.amdgcn.workitem.id.y()
  %v = call i32 @llvm.amdgcn.permlane16.i32(i32 %src0, i32 %src0, i32 %tidx, i32 %tidy, i1 false, i1 false)
  store i32 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_vvv_i64(ptr addrspace(1) %out, i64 %src0) {
; GFX10-SDAG-LABEL: v_permlane16_b32_vvv_i64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_mov_b32 null, 0
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s4, v0
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s5, v1
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_permlane16_b32 v1, v1, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v0, v0, s4, s5
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_vvv_i64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_mov_b32 null, 0
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s5, v1
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    v_permlane16_b32 v0, v0, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v1, v1, s4, s5
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlane16_b32_vvv_i64:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-NEXT:    v_and_b32_e32 v1, 0x3ff, v0
; GFX11-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX11-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_4) | instid1(VALU_DEP_2)
; GFX11-NEXT:    v_readfirstlane_b32 s5, v0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-NEXT:    v_readfirstlane_b32 s4, v1
; GFX11-NEXT:    v_mov_b32_e32 v1, s3
; GFX11-NEXT:    v_permlane16_b32 v0, v0, s4, s5
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11-NEXT:    v_permlane16_b32 v1, v1, s4, s5
; GFX11-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlane16_b32_vvv_i64:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-NEXT:    v_and_b32_e32 v1, 0x3ff, v0
; GFX12-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX12-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_4) | instid1(VALU_DEP_2)
; GFX12-NEXT:    v_readfirstlane_b32 s5, v0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-NEXT:    v_readfirstlane_b32 s4, v1
; GFX12-NEXT:    v_mov_b32_e32 v1, s3
; GFX12-NEXT:    v_permlane16_b32 v0, v0, s4, s5
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX12-NEXT:    v_permlane16_b32 v1, v1, s4, s5
; GFX12-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidy = call i32 @llvm.amdgcn.workitem.id.y()
  %v = call i64 @llvm.amdgcn.permlane16.i64(i64 %src0, i64 %src0, i32 %tidx, i32 %tidy, i1 false, i1 false)
  store i64 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_vvv_f32(ptr addrspace(1) %out, float %src0) {
; GFX10-SDAG-LABEL: v_permlane16_b32_vvv_f32:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dword s2, s[4:5], 0x2c
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s3, v0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s2, v1
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-SDAG-NEXT:    v_permlane16_b32 v0, v0, s3, s2
; GFX10-SDAG-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_vvv_f32:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dword s2, s[4:5], 0x2c
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s3, v0
; GFX10-GISEL-NEXT:    s_mov_b32 null, 0
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v1
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_permlane16_b32 v0, v0, s3, s4
; GFX10-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlane16_b32_vvv_f32:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b32 s2, s[4:5], 0x2c
; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_and_b32_e32 v1, 0x3ff, v0
; GFX11-SDAG-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_4) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s3, v1
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v1, s2
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s2, v0
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, 0
; GFX11-SDAG-NEXT:    v_permlane16_b32 v1, v1, s3, s2
; GFX11-SDAG-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlane16_b32_vvv_f32:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b32 s2, s[4:5], 0x2c
; GFX11-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_and_b32_e32 v1, 0x3ff, v0
; GFX11-GISEL-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s3, v1
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v1, 0
; GFX11-GISEL-NEXT:    v_permlane16_b32 v0, v0, s3, s4
; GFX11-GISEL-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlane16_b32_vvv_f32:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_load_b96 s[0:2], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_and_b32_e32 v1, 0x3ff, v0
; GFX12-SDAG-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s3, v1
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v1, s2
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s2, v0
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, 0
; GFX12-SDAG-NEXT:    s_wait_alu 0xf1ff
; GFX12-SDAG-NEXT:    v_permlane16_b32 v1, v1, s3, s2
; GFX12-SDAG-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlane16_b32_vvv_f32:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_load_b96 s[0:2], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_and_b32_e32 v1, 0x3ff, v0
; GFX12-GISEL-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s3, v1
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v1, 0
; GFX12-GISEL-NEXT:    s_wait_alu 0xf1ff
; GFX12-GISEL-NEXT:    v_permlane16_b32 v0, v0, s3, s4
; GFX12-GISEL-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidy = call i32 @llvm.amdgcn.workitem.id.y()
  %v = call float @llvm.amdgcn.permlane16.f32(float %src0, float %src0, i32 %tidx, i32 %tidy, i1 false, i1 false)
  store float %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_vvv_f64(ptr addrspace(1) %out, double %src0) {
; GFX10-SDAG-LABEL: v_permlane16_b32_vvv_f64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_mov_b32 null, 0
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s4, v0
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s5, v1
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_permlane16_b32 v1, v1, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v0, v0, s4, s5
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_vvv_f64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_mov_b32 null, 0
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s5, v1
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    v_permlane16_b32 v0, v0, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v1, v1, s4, s5
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlane16_b32_vvv_f64:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-NEXT:    v_and_b32_e32 v1, 0x3ff, v0
; GFX11-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX11-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_4) | instid1(VALU_DEP_2)
; GFX11-NEXT:    v_readfirstlane_b32 s5, v0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-NEXT:    v_readfirstlane_b32 s4, v1
; GFX11-NEXT:    v_mov_b32_e32 v1, s3
; GFX11-NEXT:    v_permlane16_b32 v0, v0, s4, s5
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11-NEXT:    v_permlane16_b32 v1, v1, s4, s5
; GFX11-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlane16_b32_vvv_f64:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-NEXT:    v_and_b32_e32 v1, 0x3ff, v0
; GFX12-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX12-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_4) | instid1(VALU_DEP_2)
; GFX12-NEXT:    v_readfirstlane_b32 s5, v0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-NEXT:    v_readfirstlane_b32 s4, v1
; GFX12-NEXT:    v_mov_b32_e32 v1, s3
; GFX12-NEXT:    v_permlane16_b32 v0, v0, s4, s5
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX12-NEXT:    v_permlane16_b32 v1, v1, s4, s5
; GFX12-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidy = call i32 @llvm.amdgcn.workitem.id.y()
  %v = call double @llvm.amdgcn.permlane16.f64(double %src0, double %src0, i32 %tidx, i32 %tidy, i1 false, i1 false)
  store double %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_vvs_i32(ptr addrspace(1) %out, i32 %src0, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlane16_b32_vvs_i32:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s2
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s2, v0
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, 0
; GFX10-SDAG-NEXT:    v_permlane16_b32 v1, v1, s2, s3
; GFX10-SDAG-NEXT:    global_store_dword v0, v1, s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_vvs_i32:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_mov_b32 null, 0
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_permlane16_b32 v0, v0, s4, s3
; GFX10-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlane16_b32_vvs_i32:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v1, s2 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s2, v0
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, 0
; GFX11-SDAG-NEXT:    v_permlane16_b32 v1, v1, s2, s3
; GFX11-SDAG-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlane16_b32_vvs_i32:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-GISEL-NEXT:    v_permlane16_b32 v0, v0, s4, s3
; GFX11-GISEL-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlane16_b32_vvs_i32:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v1, s2 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s2, v0
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, 0
; GFX12-SDAG-NEXT:    s_wait_alu 0xf1ff
; GFX12-SDAG-NEXT:    v_permlane16_b32 v1, v1, s2, s3
; GFX12-SDAG-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlane16_b32_vvs_i32:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-GISEL-NEXT:    v_permlane16_b32 v0, v0, s4, s3
; GFX12-GISEL-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %v = call i32 @llvm.amdgcn.permlane16.i32(i32 %src0, i32 %src0, i32 %tidx, i32 %src2, i1 false, i1 false)
  store i32 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_vvs_i64(ptr addrspace(1) %out, i64 %src0, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlane16_b32_vvs_i64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_load_dword s6, s[4:5], 0x34
; GFX10-SDAG-NEXT:    s_mov_b32 null, 0
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s4, v0
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_permlane16_b32 v1, v1, s4, s6
; GFX10-SDAG-NEXT:    v_permlane16_b32 v0, v0, s4, s6
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_vvs_i64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_load_dword s6, s[4:5], 0x34
; GFX10-GISEL-NEXT:    s_mov_b32 null, 0
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    v_permlane16_b32 v0, v0, s4, s6
; GFX10-GISEL-NEXT:    v_permlane16_b32 v1, v1, s4, s6
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlane16_b32_vvs_i64:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX11-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
; GFX11-NEXT:    v_readfirstlane_b32 s5, v0
; GFX11-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-NEXT:    v_permlane16_b32 v1, v1, s5, s4
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11-NEXT:    v_permlane16_b32 v0, v0, s5, s4
; GFX11-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlane16_b32_vvs_i64:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX12-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
; GFX12-NEXT:    v_readfirstlane_b32 s5, v0
; GFX12-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-NEXT:    v_permlane16_b32 v1, v1, s5, s4
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX12-NEXT:    v_permlane16_b32 v0, v0, s5, s4
; GFX12-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %v = call i64 @llvm.amdgcn.permlane16.i64(i64 %src0, i64 %src0, i32 %tidx, i32 %src2, i1 false, i1 false)
  store i64 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_vvs_f32(ptr addrspace(1) %out, float %src0, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlane16_b32_vvs_f32:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s2
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s2, v0
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, 0
; GFX10-SDAG-NEXT:    v_permlane16_b32 v1, v1, s2, s3
; GFX10-SDAG-NEXT:    global_store_dword v0, v1, s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_vvs_f32:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_mov_b32 null, 0
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_permlane16_b32 v0, v0, s4, s3
; GFX10-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlane16_b32_vvs_f32:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v1, s2 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s2, v0
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, 0
; GFX11-SDAG-NEXT:    v_permlane16_b32 v1, v1, s2, s3
; GFX11-SDAG-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlane16_b32_vvs_f32:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-GISEL-NEXT:    v_permlane16_b32 v0, v0, s4, s3
; GFX11-GISEL-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlane16_b32_vvs_f32:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v1, s2 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s2, v0
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, 0
; GFX12-SDAG-NEXT:    s_wait_alu 0xf1ff
; GFX12-SDAG-NEXT:    v_permlane16_b32 v1, v1, s2, s3
; GFX12-SDAG-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlane16_b32_vvs_f32:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-GISEL-NEXT:    v_permlane16_b32 v0, v0, s4, s3
; GFX12-GISEL-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %v = call float @llvm.amdgcn.permlane16.f32(float %src0, float %src0, i32 %tidx, i32 %src2, i1 false, i1 false)
  store float %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_vvs_f64(ptr addrspace(1) %out, double %src0, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlane16_b32_vvs_f64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_load_dword s6, s[4:5], 0x34
; GFX10-SDAG-NEXT:    s_mov_b32 null, 0
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s4, v0
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_permlane16_b32 v1, v1, s4, s6
; GFX10-SDAG-NEXT:    v_permlane16_b32 v0, v0, s4, s6
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_vvs_f64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_load_dword s6, s[4:5], 0x34
; GFX10-GISEL-NEXT:    s_mov_b32 null, 0
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    v_permlane16_b32 v0, v0, s4, s6
; GFX10-GISEL-NEXT:    v_permlane16_b32 v1, v1, s4, s6
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlane16_b32_vvs_f64:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX11-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
; GFX11-NEXT:    v_readfirstlane_b32 s5, v0
; GFX11-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-NEXT:    v_permlane16_b32 v1, v1, s5, s4
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11-NEXT:    v_permlane16_b32 v0, v0, s5, s4
; GFX11-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlane16_b32_vvs_f64:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX12-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
; GFX12-NEXT:    v_readfirstlane_b32 s5, v0
; GFX12-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-NEXT:    v_permlane16_b32 v1, v1, s5, s4
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX12-NEXT:    v_permlane16_b32 v0, v0, s5, s4
; GFX12-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %v = call double @llvm.amdgcn.permlane16.f64(double %src0, double %src0, i32 %tidx, i32 %src2, i1 false, i1 false)
  store double %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_vsv_i32(ptr addrspace(1) %out, i32 %src0, i32 %src1) {
; GFX10-SDAG-LABEL: v_permlane16_b32_vsv_i32:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s2, v1
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-SDAG-NEXT:    v_permlane16_b32 v0, v0, s3, s2
; GFX10-SDAG-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_vsv_i32:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_mov_b32 null, 0
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v1
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_permlane16_b32 v0, v0, s3, s4
; GFX10-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlane16_b32_vsv_i32:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v1, s2
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s2, v0
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, 0
; GFX11-SDAG-NEXT:    v_permlane16_b32 v1, v1, s3, s2
; GFX11-SDAG-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlane16_b32_vsv_i32:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v1, 0
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-GISEL-NEXT:    v_permlane16_b32 v0, v0, s3, s4
; GFX11-GISEL-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlane16_b32_vsv_i32:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v1, s2
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s2, v0
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, 0
; GFX12-SDAG-NEXT:    s_wait_alu 0xf1ff
; GFX12-SDAG-NEXT:    v_permlane16_b32 v1, v1, s3, s2
; GFX12-SDAG-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlane16_b32_vsv_i32:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v1, 0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-GISEL-NEXT:    v_permlane16_b32 v0, v0, s3, s4
; GFX12-GISEL-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidy = call i32 @llvm.amdgcn.workitem.id.y()
  %v = call i32 @llvm.amdgcn.permlane16.i32(i32 %src0, i32 %src0, i32 %src1, i32 %tidy, i1 false, i1 false)
  store i32 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_vsv_i64(ptr addrspace(1) %out, i64 %src0, i32 %src1) {
; GFX10-SDAG-LABEL: v_permlane16_b32_vsv_i64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_load_dword s6, s[4:5], 0x34
; GFX10-SDAG-NEXT:    s_mov_b32 null, 0
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s4, v1
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_permlane16_b32 v1, v1, s6, s4
; GFX10-SDAG-NEXT:    v_permlane16_b32 v0, v0, s6, s4
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_vsv_i64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_load_dword s6, s[4:5], 0x34
; GFX10-GISEL-NEXT:    s_mov_b32 null, 0
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v1
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    v_permlane16_b32 v0, v0, s6, s4
; GFX10-GISEL-NEXT:    v_permlane16_b32 v1, v1, s6, s4
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlane16_b32_vsv_i64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX11-SDAG-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s5, v0
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v1, s3 :: v_dual_mov_b32 v0, s2
; GFX11-SDAG-NEXT:    v_permlane16_b32 v1, v1, s4, s5
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v0, v0, s4, s5
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlane16_b32_vsv_i64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX11-GISEL-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s5, v0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-GISEL-NEXT:    v_permlane16_b32 v0, v0, s4, s5
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v1, v1, s4, s5
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlane16_b32_vsv_i64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX12-SDAG-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s5, v0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v1, s3 :: v_dual_mov_b32 v0, s2
; GFX12-SDAG-NEXT:    v_permlane16_b32 v1, v1, s4, s5
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlane16_b32 v0, v0, s4, s5
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlane16_b32_vsv_i64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX12-GISEL-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s5, v0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX12-GISEL-NEXT:    v_permlane16_b32 v0, v0, s4, s5
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlane16_b32 v1, v1, s4, s5
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidy = call i32 @llvm.amdgcn.workitem.id.y()
  %v = call i64 @llvm.amdgcn.permlane16.i64(i64 %src0, i64 %src0, i32 %src1, i32 %tidy, i1 false, i1 false)
  store i64 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_vsv_f32(ptr addrspace(1) %out, float %src0, i32 %src1) {
; GFX10-SDAG-LABEL: v_permlane16_b32_vsv_f32:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s2, v1
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-SDAG-NEXT:    v_permlane16_b32 v0, v0, s3, s2
; GFX10-SDAG-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_vsv_f32:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_mov_b32 null, 0
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v1
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_permlane16_b32 v0, v0, s3, s4
; GFX10-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlane16_b32_vsv_f32:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v1, s2
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s2, v0
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, 0
; GFX11-SDAG-NEXT:    v_permlane16_b32 v1, v1, s3, s2
; GFX11-SDAG-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlane16_b32_vsv_f32:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v1, 0
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-GISEL-NEXT:    v_permlane16_b32 v0, v0, s3, s4
; GFX11-GISEL-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlane16_b32_vsv_f32:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v1, s2
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s2, v0
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, 0
; GFX12-SDAG-NEXT:    s_wait_alu 0xf1ff
; GFX12-SDAG-NEXT:    v_permlane16_b32 v1, v1, s3, s2
; GFX12-SDAG-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlane16_b32_vsv_f32:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v1, 0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-GISEL-NEXT:    v_permlane16_b32 v0, v0, s3, s4
; GFX12-GISEL-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidy = call i32 @llvm.amdgcn.workitem.id.y()
  %v = call float @llvm.amdgcn.permlane16.f32(float %src0, float %src0, i32 %src1, i32 %tidy, i1 false, i1 false)
  store float %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_vsv_f64(ptr addrspace(1) %out, double %src0, i32 %src1) {
; GFX10-SDAG-LABEL: v_permlane16_b32_vsv_f64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_load_dword s6, s[4:5], 0x34
; GFX10-SDAG-NEXT:    s_mov_b32 null, 0
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s4, v1
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_permlane16_b32 v1, v1, s6, s4
; GFX10-SDAG-NEXT:    v_permlane16_b32 v0, v0, s6, s4
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_vsv_f64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_load_dword s6, s[4:5], 0x34
; GFX10-GISEL-NEXT:    s_mov_b32 null, 0
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v1
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    v_permlane16_b32 v0, v0, s6, s4
; GFX10-GISEL-NEXT:    v_permlane16_b32 v1, v1, s6, s4
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlane16_b32_vsv_f64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX11-SDAG-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s5, v0
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v1, s3 :: v_dual_mov_b32 v0, s2
; GFX11-SDAG-NEXT:    v_permlane16_b32 v1, v1, s4, s5
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v0, v0, s4, s5
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlane16_b32_vsv_f64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX11-GISEL-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s5, v0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-GISEL-NEXT:    v_permlane16_b32 v0, v0, s4, s5
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v1, v1, s4, s5
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlane16_b32_vsv_f64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX12-SDAG-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s5, v0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v1, s3 :: v_dual_mov_b32 v0, s2
; GFX12-SDAG-NEXT:    v_permlane16_b32 v1, v1, s4, s5
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlane16_b32 v0, v0, s4, s5
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlane16_b32_vsv_f64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX12-GISEL-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s5, v0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX12-GISEL-NEXT:    v_permlane16_b32 v0, v0, s4, s5
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlane16_b32 v1, v1, s4, s5
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidy = call i32 @llvm.amdgcn.workitem.id.y()
  %v = call double @llvm.amdgcn.permlane16.f64(double %src0, double %src0, i32 %src1, i32 %tidy, i1 false, i1 false)
  store double %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_vss_fi_i32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlane16_b32_vss_fi_i32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-NEXT:    s_load_dword s6, s[4:5], 0x34
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-NEXT:    v_permlane16_b32 v0, v0, s3, s6 op_sel:[1,0]
; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlane16_b32_vss_fi_i32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlane16_b32 v0, v0, s3, s4 op_sel:[1,0]
; GFX11-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlane16_b32_vss_fi_i32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlane16_b32 v0, v0, s3, s4 op_sel:[1,0]
; GFX12-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-NEXT:    s_endpgm
  %v = call i32 @llvm.amdgcn.permlane16.i32(i32 %src0, i32 %src0, i32 %src1, i32 %src2, i1 true, i1 false)
  store i32 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_vss_fi_i64(ptr addrspace(1) %out, i64 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlane16_b32_vss_fi_i64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_permlane16_b32 v1, v1, s6, s7 op_sel:[1,0]
; GFX10-SDAG-NEXT:    v_permlane16_b32 v0, v0, s6, s7 op_sel:[1,0]
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_vss_fi_i64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    v_permlane16_b32 v0, v0, s6, s7 op_sel:[1,0]
; GFX10-GISEL-NEXT:    v_permlane16_b32 v1, v1, s6, s7 op_sel:[1,0]
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlane16_b32_vss_fi_i64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v1, v1, s4, s5 op_sel:[1,0]
; GFX11-SDAG-NEXT:    v_permlane16_b32 v0, v0, s4, s5 op_sel:[1,0]
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlane16_b32_vss_fi_i64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v0, v0, s4, s5 op_sel:[1,0]
; GFX11-GISEL-NEXT:    v_permlane16_b32 v1, v1, s4, s5 op_sel:[1,0]
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlane16_b32_vss_fi_i64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlane16_b32 v1, v1, s4, s5 op_sel:[1,0]
; GFX12-SDAG-NEXT:    v_permlane16_b32 v0, v0, s4, s5 op_sel:[1,0]
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlane16_b32_vss_fi_i64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlane16_b32 v0, v0, s4, s5 op_sel:[1,0]
; GFX12-GISEL-NEXT:    v_permlane16_b32 v1, v1, s4, s5 op_sel:[1,0]
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %v = call i64 @llvm.amdgcn.permlane16.i64(i64 %src0, i64 %src0, i32 %src1, i32 %src2, i1 true, i1 false)
  store i64 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_vss_fi_f32(ptr addrspace(1) %out, float %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlane16_b32_vss_fi_f32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-NEXT:    s_load_dword s6, s[4:5], 0x34
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-NEXT:    v_permlane16_b32 v0, v0, s3, s6 op_sel:[1,0]
; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlane16_b32_vss_fi_f32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlane16_b32 v0, v0, s3, s4 op_sel:[1,0]
; GFX11-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlane16_b32_vss_fi_f32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlane16_b32 v0, v0, s3, s4 op_sel:[1,0]
; GFX12-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-NEXT:    s_endpgm
  %v = call float @llvm.amdgcn.permlane16.f32(float %src0, float %src0, i32 %src1, i32 %src2, i1 true, i1 false)
  store float %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_vss_fi_f64(ptr addrspace(1) %out, double %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlane16_b32_vss_fi_f64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_permlane16_b32 v1, v1, s6, s7 op_sel:[1,0]
; GFX10-SDAG-NEXT:    v_permlane16_b32 v0, v0, s6, s7 op_sel:[1,0]
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_vss_fi_f64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    v_permlane16_b32 v0, v0, s6, s7 op_sel:[1,0]
; GFX10-GISEL-NEXT:    v_permlane16_b32 v1, v1, s6, s7 op_sel:[1,0]
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlane16_b32_vss_fi_f64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v1, v1, s4, s5 op_sel:[1,0]
; GFX11-SDAG-NEXT:    v_permlane16_b32 v0, v0, s4, s5 op_sel:[1,0]
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlane16_b32_vss_fi_f64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v0, v0, s4, s5 op_sel:[1,0]
; GFX11-GISEL-NEXT:    v_permlane16_b32 v1, v1, s4, s5 op_sel:[1,0]
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlane16_b32_vss_fi_f64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlane16_b32 v1, v1, s4, s5 op_sel:[1,0]
; GFX12-SDAG-NEXT:    v_permlane16_b32 v0, v0, s4, s5 op_sel:[1,0]
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlane16_b32_vss_fi_f64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlane16_b32 v0, v0, s4, s5 op_sel:[1,0]
; GFX12-GISEL-NEXT:    v_permlane16_b32 v1, v1, s4, s5 op_sel:[1,0]
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %v = call double @llvm.amdgcn.permlane16.f64(double %src0, double %src0, i32 %src1, i32 %src2, i1 true, i1 false)
  store double %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_vss_bc_i32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlane16_b32_vss_bc_i32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-NEXT:    s_load_dword s6, s[4:5], 0x34
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-NEXT:    v_permlane16_b32 v0, v0, s3, s6 op_sel:[0,1]
; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlane16_b32_vss_bc_i32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlane16_b32 v0, v0, s3, s4 op_sel:[0,1]
; GFX11-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlane16_b32_vss_bc_i32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlane16_b32 v0, v0, s3, s4 op_sel:[0,1]
; GFX12-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-NEXT:    s_endpgm
  %v = call i32 @llvm.amdgcn.permlane16.i32(i32 %src0, i32 %src0, i32 %src1, i32 %src2, i1 false, i1 true)
  store i32 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_vss_bc_i64(ptr addrspace(1) %out, i64 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlane16_b32_vss_bc_i64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_permlane16_b32 v1, v1, s6, s7 op_sel:[0,1]
; GFX10-SDAG-NEXT:    v_permlane16_b32 v0, v0, s6, s7 op_sel:[0,1]
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_vss_bc_i64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    v_permlane16_b32 v0, v0, s6, s7 op_sel:[0,1]
; GFX10-GISEL-NEXT:    v_permlane16_b32 v1, v1, s6, s7 op_sel:[0,1]
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlane16_b32_vss_bc_i64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v1, v1, s4, s5 op_sel:[0,1]
; GFX11-SDAG-NEXT:    v_permlane16_b32 v0, v0, s4, s5 op_sel:[0,1]
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlane16_b32_vss_bc_i64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v0, v0, s4, s5 op_sel:[0,1]
; GFX11-GISEL-NEXT:    v_permlane16_b32 v1, v1, s4, s5 op_sel:[0,1]
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlane16_b32_vss_bc_i64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlane16_b32 v1, v1, s4, s5 op_sel:[0,1]
; GFX12-SDAG-NEXT:    v_permlane16_b32 v0, v0, s4, s5 op_sel:[0,1]
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlane16_b32_vss_bc_i64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlane16_b32 v0, v0, s4, s5 op_sel:[0,1]
; GFX12-GISEL-NEXT:    v_permlane16_b32 v1, v1, s4, s5 op_sel:[0,1]
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %v = call i64 @llvm.amdgcn.permlane16.i64(i64 %src0, i64 %src0, i32 %src1, i32 %src2, i1 false, i1 true)
  store i64 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_vss_bc_f32(ptr addrspace(1) %out, float %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlane16_b32_vss_bc_f32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-NEXT:    s_load_dword s6, s[4:5], 0x34
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-NEXT:    v_permlane16_b32 v0, v0, s3, s6 op_sel:[0,1]
; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlane16_b32_vss_bc_f32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlane16_b32 v0, v0, s3, s4 op_sel:[0,1]
; GFX11-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlane16_b32_vss_bc_f32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlane16_b32 v0, v0, s3, s4 op_sel:[0,1]
; GFX12-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-NEXT:    s_endpgm
  %v = call float @llvm.amdgcn.permlane16.f32(float %src0, float %src0, i32 %src1, i32 %src2, i1 false, i1 true)
  store float %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_vss_bc_f64(ptr addrspace(1) %out, double %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlane16_b32_vss_bc_f64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_permlane16_b32 v1, v1, s6, s7 op_sel:[0,1]
; GFX10-SDAG-NEXT:    v_permlane16_b32 v0, v0, s6, s7 op_sel:[0,1]
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_vss_bc_f64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    v_permlane16_b32 v0, v0, s6, s7 op_sel:[0,1]
; GFX10-GISEL-NEXT:    v_permlane16_b32 v1, v1, s6, s7 op_sel:[0,1]
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlane16_b32_vss_bc_f64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v1, v1, s4, s5 op_sel:[0,1]
; GFX11-SDAG-NEXT:    v_permlane16_b32 v0, v0, s4, s5 op_sel:[0,1]
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlane16_b32_vss_bc_f64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v0, v0, s4, s5 op_sel:[0,1]
; GFX11-GISEL-NEXT:    v_permlane16_b32 v1, v1, s4, s5 op_sel:[0,1]
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlane16_b32_vss_bc_f64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlane16_b32 v1, v1, s4, s5 op_sel:[0,1]
; GFX12-SDAG-NEXT:    v_permlane16_b32 v0, v0, s4, s5 op_sel:[0,1]
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlane16_b32_vss_bc_f64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlane16_b32 v0, v0, s4, s5 op_sel:[0,1]
; GFX12-GISEL-NEXT:    v_permlane16_b32 v1, v1, s4, s5 op_sel:[0,1]
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %v = call double @llvm.amdgcn.permlane16.f64(double %src0, double %src0, i32 %src1, i32 %src2, i1 false, i1 true)
  store double %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_vss_fi_bc_i32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlane16_b32_vss_fi_bc_i32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-NEXT:    s_load_dword s6, s[4:5], 0x34
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-NEXT:    v_permlane16_b32 v0, v0, s3, s6 op_sel:[1,1]
; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlane16_b32_vss_fi_bc_i32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlane16_b32 v0, v0, s3, s4 op_sel:[1,1]
; GFX11-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlane16_b32_vss_fi_bc_i32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlane16_b32 v0, v0, s3, s4 op_sel:[1,1]
; GFX12-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-NEXT:    s_endpgm
  %v = call i32 @llvm.amdgcn.permlane16.i32(i32 %src0, i32 %src0, i32 %src1, i32 %src2, i1 true, i1 true)
  store i32 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_vss_fi_bc_i64(ptr addrspace(1) %out, i64 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlane16_b32_vss_fi_bc_i64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_permlane16_b32 v1, v1, s6, s7 op_sel:[1,1]
; GFX10-SDAG-NEXT:    v_permlane16_b32 v0, v0, s6, s7 op_sel:[1,1]
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_vss_fi_bc_i64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    v_permlane16_b32 v0, v0, s6, s7 op_sel:[1,1]
; GFX10-GISEL-NEXT:    v_permlane16_b32 v1, v1, s6, s7 op_sel:[1,1]
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlane16_b32_vss_fi_bc_i64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v1, v1, s4, s5 op_sel:[1,1]
; GFX11-SDAG-NEXT:    v_permlane16_b32 v0, v0, s4, s5 op_sel:[1,1]
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlane16_b32_vss_fi_bc_i64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v0, v0, s4, s5 op_sel:[1,1]
; GFX11-GISEL-NEXT:    v_permlane16_b32 v1, v1, s4, s5 op_sel:[1,1]
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlane16_b32_vss_fi_bc_i64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlane16_b32 v1, v1, s4, s5 op_sel:[1,1]
; GFX12-SDAG-NEXT:    v_permlane16_b32 v0, v0, s4, s5 op_sel:[1,1]
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlane16_b32_vss_fi_bc_i64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlane16_b32 v0, v0, s4, s5 op_sel:[1,1]
; GFX12-GISEL-NEXT:    v_permlane16_b32 v1, v1, s4, s5 op_sel:[1,1]
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %v = call i64 @llvm.amdgcn.permlane16.i64(i64 %src0, i64 %src0, i32 %src1, i32 %src2, i1 true, i1 true)
  store i64 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_vss_fi_bc_f32(ptr addrspace(1) %out, float %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlane16_b32_vss_fi_bc_f32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-NEXT:    s_load_dword s6, s[4:5], 0x34
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-NEXT:    v_permlane16_b32 v0, v0, s3, s6 op_sel:[1,1]
; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlane16_b32_vss_fi_bc_f32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlane16_b32 v0, v0, s3, s4 op_sel:[1,1]
; GFX11-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlane16_b32_vss_fi_bc_f32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlane16_b32 v0, v0, s3, s4 op_sel:[1,1]
; GFX12-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-NEXT:    s_endpgm
  %v = call float @llvm.amdgcn.permlane16.f32(float %src0, float %src0, i32 %src1, i32 %src2, i1 true, i1 true)
  store float %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_vss_fi_bc_f64(ptr addrspace(1) %out, double %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlane16_b32_vss_fi_bc_f64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_permlane16_b32 v1, v1, s6, s7 op_sel:[1,1]
; GFX10-SDAG-NEXT:    v_permlane16_b32 v0, v0, s6, s7 op_sel:[1,1]
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_vss_fi_bc_f64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    v_permlane16_b32 v0, v0, s6, s7 op_sel:[1,1]
; GFX10-GISEL-NEXT:    v_permlane16_b32 v1, v1, s6, s7 op_sel:[1,1]
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlane16_b32_vss_fi_bc_f64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v1, v1, s4, s5 op_sel:[1,1]
; GFX11-SDAG-NEXT:    v_permlane16_b32 v0, v0, s4, s5 op_sel:[1,1]
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlane16_b32_vss_fi_bc_f64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v0, v0, s4, s5 op_sel:[1,1]
; GFX11-GISEL-NEXT:    v_permlane16_b32 v1, v1, s4, s5 op_sel:[1,1]
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlane16_b32_vss_fi_bc_f64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlane16_b32 v1, v1, s4, s5 op_sel:[1,1]
; GFX12-SDAG-NEXT:    v_permlane16_b32 v0, v0, s4, s5 op_sel:[1,1]
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlane16_b32_vss_fi_bc_f64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlane16_b32 v0, v0, s4, s5 op_sel:[1,1]
; GFX12-GISEL-NEXT:    v_permlane16_b32 v1, v1, s4, s5 op_sel:[1,1]
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %v = call double @llvm.amdgcn.permlane16.f64(double %src0, double %src0, i32 %src1, i32 %src2, i1 true, i1 true)
  store double %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vss_i32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlanex16_b32_vss_i32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-NEXT:    s_load_dword s6, s[4:5], 0x34
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-NEXT:    v_permlanex16_b32 v0, v0, s3, s6
; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlanex16_b32_vss_i32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlanex16_b32 v0, v0, s3, s4
; GFX11-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlanex16_b32_vss_i32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlanex16_b32 v0, v0, s3, s4
; GFX12-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-NEXT:    s_endpgm
  %v = call i32 @llvm.amdgcn.permlanex16.i32(i32 %src0, i32 %src0, i32 %src1, i32 %src2, i1 false, i1 false)
  store i32 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vss_f32(ptr addrspace(1) %out, float %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlanex16_b32_vss_f32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-NEXT:    s_load_dword s6, s[4:5], 0x34
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-NEXT:    v_permlanex16_b32 v0, v0, s3, s6
; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlanex16_b32_vss_f32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlanex16_b32 v0, v0, s3, s4
; GFX11-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlanex16_b32_vss_f32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlanex16_b32 v0, v0, s3, s4
; GFX12-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-NEXT:    s_endpgm
  %v = call float @llvm.amdgcn.permlanex16.f32(float %src0, float %src0, i32 %src1, i32 %src2, i1 false, i1 false)
  store float %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vss_i64(ptr addrspace(1) %out, i64 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_vss_i64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s6, s7
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s6, s7
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_vss_i64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s6, s7
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s6, s7
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlanex16_b32_vss_i64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s4, s5
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s4, s5
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlanex16_b32_vss_i64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s4, s5
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s4, s5
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlanex16_b32_vss_i64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s4, s5
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s4, s5
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlanex16_b32_vss_i64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s4, s5
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s4, s5
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %v = call i64 @llvm.amdgcn.permlanex16.i64(i64 %src0, i64 %src0, i32 %src1, i32 %src2, i1 false, i1 false)
  store i64 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vss_f64(ptr addrspace(1) %out, double %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_vss_f64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s6, s7
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s6, s7
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_vss_f64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s6, s7
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s6, s7
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlanex16_b32_vss_f64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s4, s5
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s4, s5
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlanex16_b32_vss_f64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s4, s5
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s4, s5
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlanex16_b32_vss_f64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s4, s5
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s4, s5
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlanex16_b32_vss_f64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s4, s5
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s4, s5
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %v = call double @llvm.amdgcn.permlanex16.f64(double %src0, double %src0, i32 %src1, i32 %src2, i1 false, i1 false)
  store double %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vii_i32(ptr addrspace(1) %out, i32 %src0) {
; GFX10-LABEL: v_permlanex16_b32_vii_i32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dword s2, s[4:5], 0x2c
; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-NEXT:    v_permlanex16_b32 v0, v0, 1, 2
; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlanex16_b32_vii_i32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b32 s2, s[4:5], 0x2c
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlanex16_b32 v0, v0, 1, 2
; GFX11-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlanex16_b32_vii_i32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_load_b96 s[0:2], s[4:5], 0x24
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlanex16_b32 v0, v0, 1, 2
; GFX12-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-NEXT:    s_endpgm
  %v = call i32 @llvm.amdgcn.permlanex16.i32(i32 %src0, i32 %src0, i32 1, i32 2, i1 false, i1 false)
  store i32 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vii_f32(ptr addrspace(1) %out, float %src0) {
; GFX10-LABEL: v_permlanex16_b32_vii_f32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dword s2, s[4:5], 0x2c
; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-NEXT:    v_permlanex16_b32 v0, v0, 1, 2
; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlanex16_b32_vii_f32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b32 s2, s[4:5], 0x2c
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlanex16_b32 v0, v0, 1, 2
; GFX11-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlanex16_b32_vii_f32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_load_b96 s[0:2], s[4:5], 0x24
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlanex16_b32 v0, v0, 1, 2
; GFX12-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-NEXT:    s_endpgm
  %v = call float @llvm.amdgcn.permlanex16.f32(float %src0, float %src0, i32 1, i32 2, i1 false, i1 false)
  store float %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vii_i64(ptr addrspace(1) %out, i64 %src0) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_vii_i64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v1, v1, 1, 2
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v0, v0, 1, 2
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_vii_i64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v0, v0, 1, 2
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v1, v1, 1, 2
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlanex16_b32_vii_i64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v1, v1, 1, 2
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v0, v0, 1, 2
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlanex16_b32_vii_i64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v0, v0, 1, 2
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v1, v1, 1, 2
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlanex16_b32_vii_i64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v1, v1, 1, 2
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v0, v0, 1, 2
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlanex16_b32_vii_i64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v0, v0, 1, 2
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v1, v1, 1, 2
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %v = call i64 @llvm.amdgcn.permlanex16.i64(i64 %src0, i64 %src0, i32 1, i32 2, i1 false, i1 false)
  store i64 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vii_f64(ptr addrspace(1) %out, double %src0) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_vii_f64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v1, v1, 1, 2
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v0, v0, 1, 2
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_vii_f64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v0, v0, 1, 2
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v1, v1, 1, 2
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlanex16_b32_vii_f64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v1, v1, 1, 2
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v0, v0, 1, 2
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlanex16_b32_vii_f64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v0, v0, 1, 2
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v1, v1, 1, 2
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlanex16_b32_vii_f64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v1, v1, 1, 2
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v0, v0, 1, 2
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlanex16_b32_vii_f64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v0, v0, 1, 2
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v1, v1, 1, 2
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %v = call double @llvm.amdgcn.permlanex16.f64(double %src0, double %src0, i32 1, i32 2, i1 false, i1 false)
  store double %v, ptr addrspace(1) %out
  ret void
}

; FIXME-GFX10PLUS: It is allowed to have both immediates as literals
define amdgpu_kernel void @v_permlanex16_b32_vll_i32(ptr addrspace(1) %out, i32 %src0) {
; GFX10-LABEL: v_permlanex16_b32_vll_i32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dword s2, s[4:5], 0x2c
; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-NEXT:    s_movk_i32 s2, 0x1234
; GFX10-NEXT:    v_permlanex16_b32 v0, v0, s2, 0xc1d1
; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlanex16_b32_vll_i32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b32 s2, s[4:5], 0x2c
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX11-NEXT:    s_movk_i32 s2, 0x1234
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
; GFX11-NEXT:    v_permlanex16_b32 v0, v0, s2, 0xc1d1
; GFX11-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlanex16_b32_vll_i32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_load_b96 s[0:2], s[4:5], 0x24
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX12-NEXT:    s_movk_i32 s2, 0x1234
; GFX12-NEXT:    s_wait_alu 0xfffe
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlanex16_b32 v0, v0, s2, 0xc1d1
; GFX12-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-NEXT:    s_endpgm
  %v = call i32 @llvm.amdgcn.permlanex16.i32(i32 %src0, i32 %src0, i32 4660, i32 49617, i1 false, i1 false)
  store i32 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vll_f32(ptr addrspace(1) %out, float %src0) {
; GFX10-LABEL: v_permlanex16_b32_vll_f32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dword s2, s[4:5], 0x2c
; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-NEXT:    s_movk_i32 s2, 0x1234
; GFX10-NEXT:    v_permlanex16_b32 v0, v0, s2, 0xc1d1
; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlanex16_b32_vll_f32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b32 s2, s[4:5], 0x2c
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX11-NEXT:    s_movk_i32 s2, 0x1234
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
; GFX11-NEXT:    v_permlanex16_b32 v0, v0, s2, 0xc1d1
; GFX11-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlanex16_b32_vll_f32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_load_b96 s[0:2], s[4:5], 0x24
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX12-NEXT:    s_movk_i32 s2, 0x1234
; GFX12-NEXT:    s_wait_alu 0xfffe
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlanex16_b32 v0, v0, s2, 0xc1d1
; GFX12-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-NEXT:    s_endpgm
  %v = call float @llvm.amdgcn.permlanex16.f32(float %src0, float %src0, i32 4660, i32 49617, i1 false, i1 false)
  store float %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vll_i64(ptr addrspace(1) %out, i64 %src0) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_vll_i64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    s_movk_i32 s2, 0x1234
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s2, 0xc1d1
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s2, 0xc1d1
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_vll_i64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    s_movk_i32 s2, 0x1234
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s2, 0xc1d1
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s2, 0xc1d1
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlanex16_b32_vll_i64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-SDAG-NEXT:    s_movk_i32 s2, 0x1234
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instid1(SALU_CYCLE_1)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s2, 0xc1d1
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s2, 0xc1d1
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlanex16_b32_vll_i64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-GISEL-NEXT:    s_movk_i32 s2, 0x1234
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s2, 0xc1d1
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s2, 0xc1d1
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlanex16_b32_vll_i64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-SDAG-NEXT:    s_movk_i32 s2, 0x1234
; GFX12-SDAG-NEXT:    s_wait_alu 0xfffe
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s2, 0xc1d1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s2, 0xc1d1
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlanex16_b32_vll_i64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX12-GISEL-NEXT:    s_movk_i32 s2, 0x1234
; GFX12-GISEL-NEXT:    s_wait_alu 0xfffe
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s2, 0xc1d1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s2, 0xc1d1
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %v = call i64 @llvm.amdgcn.permlanex16.i64(i64 %src0, i64 %src0, i32 4660, i32 49617, i1 false, i1 false)
  store i64 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vll_f64(ptr addrspace(1) %out, double %src0) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_vll_f64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    s_movk_i32 s2, 0x1234
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s2, 0xc1d1
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s2, 0xc1d1
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_vll_f64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    s_movk_i32 s2, 0x1234
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s2, 0xc1d1
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s2, 0xc1d1
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlanex16_b32_vll_f64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-SDAG-NEXT:    s_movk_i32 s2, 0x1234
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instid1(SALU_CYCLE_1)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s2, 0xc1d1
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s2, 0xc1d1
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlanex16_b32_vll_f64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-GISEL-NEXT:    s_movk_i32 s2, 0x1234
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s2, 0xc1d1
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s2, 0xc1d1
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlanex16_b32_vll_f64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-SDAG-NEXT:    s_movk_i32 s2, 0x1234
; GFX12-SDAG-NEXT:    s_wait_alu 0xfffe
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s2, 0xc1d1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s2, 0xc1d1
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlanex16_b32_vll_f64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX12-GISEL-NEXT:    s_movk_i32 s2, 0x1234
; GFX12-GISEL-NEXT:    s_wait_alu 0xfffe
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s2, 0xc1d1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s2, 0xc1d1
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %v = call double @llvm.amdgcn.permlanex16.f64(double %src0, double %src0, i32 4660, i32 49617, i1 false, i1 false)
  store double %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vvv_i32(ptr addrspace(1) %out, i32 %src0) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_vvv_i32:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dword s2, s[4:5], 0x2c
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s3, v0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s2, v1
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s3, s2
; GFX10-SDAG-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_vvv_i32:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dword s2, s[4:5], 0x2c
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s3, v0
; GFX10-GISEL-NEXT:    s_mov_b32 null, 0
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v1
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s3, s4
; GFX10-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlanex16_b32_vvv_i32:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b32 s2, s[4:5], 0x2c
; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_and_b32_e32 v1, 0x3ff, v0
; GFX11-SDAG-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_4) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s3, v1
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v1, s2
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s2, v0
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, 0
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s3, s2
; GFX11-SDAG-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlanex16_b32_vvv_i32:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b32 s2, s[4:5], 0x2c
; GFX11-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_and_b32_e32 v1, 0x3ff, v0
; GFX11-GISEL-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s3, v1
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v1, 0
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s3, s4
; GFX11-GISEL-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlanex16_b32_vvv_i32:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_load_b96 s[0:2], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_and_b32_e32 v1, 0x3ff, v0
; GFX12-SDAG-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s3, v1
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v1, s2
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s2, v0
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, 0
; GFX12-SDAG-NEXT:    s_wait_alu 0xf1ff
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s3, s2
; GFX12-SDAG-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlanex16_b32_vvv_i32:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_load_b96 s[0:2], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_and_b32_e32 v1, 0x3ff, v0
; GFX12-GISEL-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s3, v1
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v1, 0
; GFX12-GISEL-NEXT:    s_wait_alu 0xf1ff
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s3, s4
; GFX12-GISEL-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidy = call i32 @llvm.amdgcn.workitem.id.y()
  %v = call i32 @llvm.amdgcn.permlanex16.i32(i32 %src0, i32 %src0, i32 %tidx, i32 %tidy, i1 false, i1 false)
  store i32 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vvv_f32(ptr addrspace(1) %out, float %src0) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_vvv_f32:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dword s2, s[4:5], 0x2c
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s3, v0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s2, v1
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s3, s2
; GFX10-SDAG-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_vvv_f32:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dword s2, s[4:5], 0x2c
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s3, v0
; GFX10-GISEL-NEXT:    s_mov_b32 null, 0
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v1
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s3, s4
; GFX10-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlanex16_b32_vvv_f32:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b32 s2, s[4:5], 0x2c
; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_and_b32_e32 v1, 0x3ff, v0
; GFX11-SDAG-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_4) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s3, v1
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v1, s2
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s2, v0
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, 0
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s3, s2
; GFX11-SDAG-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlanex16_b32_vvv_f32:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b32 s2, s[4:5], 0x2c
; GFX11-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_and_b32_e32 v1, 0x3ff, v0
; GFX11-GISEL-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s3, v1
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v1, 0
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s3, s4
; GFX11-GISEL-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlanex16_b32_vvv_f32:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_load_b96 s[0:2], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_and_b32_e32 v1, 0x3ff, v0
; GFX12-SDAG-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s3, v1
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v1, s2
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s2, v0
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, 0
; GFX12-SDAG-NEXT:    s_wait_alu 0xf1ff
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s3, s2
; GFX12-SDAG-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlanex16_b32_vvv_f32:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_load_b96 s[0:2], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_and_b32_e32 v1, 0x3ff, v0
; GFX12-GISEL-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s3, v1
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v1, 0
; GFX12-GISEL-NEXT:    s_wait_alu 0xf1ff
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s3, s4
; GFX12-GISEL-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidy = call i32 @llvm.amdgcn.workitem.id.y()
  %v = call float @llvm.amdgcn.permlanex16.f32(float %src0, float %src0, i32 %tidx, i32 %tidy, i1 false, i1 false)
  store float %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vvv_i64(ptr addrspace(1) %out, i64 %src0) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_vvv_i64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_mov_b32 null, 0
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s4, v0
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s5, v1
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s4, s5
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_vvv_i64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_mov_b32 null, 0
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s5, v1
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s4, s5
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlanex16_b32_vvv_i64:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-NEXT:    v_and_b32_e32 v1, 0x3ff, v0
; GFX11-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX11-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_4) | instid1(VALU_DEP_2)
; GFX11-NEXT:    v_readfirstlane_b32 s5, v0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-NEXT:    v_readfirstlane_b32 s4, v1
; GFX11-NEXT:    v_mov_b32_e32 v1, s3
; GFX11-NEXT:    v_permlanex16_b32 v0, v0, s4, s5
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11-NEXT:    v_permlanex16_b32 v1, v1, s4, s5
; GFX11-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlanex16_b32_vvv_i64:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-NEXT:    v_and_b32_e32 v1, 0x3ff, v0
; GFX12-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX12-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_4) | instid1(VALU_DEP_2)
; GFX12-NEXT:    v_readfirstlane_b32 s5, v0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-NEXT:    v_readfirstlane_b32 s4, v1
; GFX12-NEXT:    v_mov_b32_e32 v1, s3
; GFX12-NEXT:    v_permlanex16_b32 v0, v0, s4, s5
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX12-NEXT:    v_permlanex16_b32 v1, v1, s4, s5
; GFX12-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidy = call i32 @llvm.amdgcn.workitem.id.y()
  %v = call i64 @llvm.amdgcn.permlanex16.i64(i64 %src0, i64 %src0, i32 %tidx, i32 %tidy, i1 false, i1 false)
  store i64 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vvv_f64(ptr addrspace(1) %out, double %src0) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_vvv_f64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_mov_b32 null, 0
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s4, v0
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s5, v1
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s4, s5
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_vvv_f64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_mov_b32 null, 0
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s5, v1
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s4, s5
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlanex16_b32_vvv_f64:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-NEXT:    v_and_b32_e32 v1, 0x3ff, v0
; GFX11-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX11-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_4) | instid1(VALU_DEP_2)
; GFX11-NEXT:    v_readfirstlane_b32 s5, v0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-NEXT:    v_readfirstlane_b32 s4, v1
; GFX11-NEXT:    v_mov_b32_e32 v1, s3
; GFX11-NEXT:    v_permlanex16_b32 v0, v0, s4, s5
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11-NEXT:    v_permlanex16_b32 v1, v1, s4, s5
; GFX11-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlanex16_b32_vvv_f64:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-NEXT:    v_and_b32_e32 v1, 0x3ff, v0
; GFX12-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX12-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_4) | instid1(VALU_DEP_2)
; GFX12-NEXT:    v_readfirstlane_b32 s5, v0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-NEXT:    v_readfirstlane_b32 s4, v1
; GFX12-NEXT:    v_mov_b32_e32 v1, s3
; GFX12-NEXT:    v_permlanex16_b32 v0, v0, s4, s5
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX12-NEXT:    v_permlanex16_b32 v1, v1, s4, s5
; GFX12-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidy = call i32 @llvm.amdgcn.workitem.id.y()
  %v = call double @llvm.amdgcn.permlanex16.f64(double %src0, double %src0, i32 %tidx, i32 %tidy, i1 false, i1 false)
  store double %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vvs_i32(ptr addrspace(1) %out, i32 %src0, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_vvs_i32:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s2
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s2, v0
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, 0
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s2, s3
; GFX10-SDAG-NEXT:    global_store_dword v0, v1, s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_vvs_i32:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_mov_b32 null, 0
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s4, s3
; GFX10-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlanex16_b32_vvs_i32:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v1, s2 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s2, v0
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, 0
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s2, s3
; GFX11-SDAG-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlanex16_b32_vvs_i32:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s4, s3
; GFX11-GISEL-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlanex16_b32_vvs_i32:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v1, s2 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s2, v0
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, 0
; GFX12-SDAG-NEXT:    s_wait_alu 0xf1ff
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s2, s3
; GFX12-SDAG-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlanex16_b32_vvs_i32:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s4, s3
; GFX12-GISEL-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %v = call i32 @llvm.amdgcn.permlanex16.i32(i32 %src0, i32 %src0, i32 %tidx, i32 %src2, i1 false, i1 false)
  store i32 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vvs_f32(ptr addrspace(1) %out, float %src0, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_vvs_f32:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s2
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s2, v0
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, 0
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s2, s3
; GFX10-SDAG-NEXT:    global_store_dword v0, v1, s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_vvs_f32:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_mov_b32 null, 0
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s4, s3
; GFX10-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlanex16_b32_vvs_f32:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v1, s2 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s2, v0
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, 0
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s2, s3
; GFX11-SDAG-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlanex16_b32_vvs_f32:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s4, s3
; GFX11-GISEL-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlanex16_b32_vvs_f32:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v1, s2 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s2, v0
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, 0
; GFX12-SDAG-NEXT:    s_wait_alu 0xf1ff
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s2, s3
; GFX12-SDAG-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlanex16_b32_vvs_f32:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s4, s3
; GFX12-GISEL-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %v = call float @llvm.amdgcn.permlanex16.f32(float %src0, float %src0, i32 %tidx, i32 %src2, i1 false, i1 false)
  store float %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vvs_i64(ptr addrspace(1) %out, i64 %src0, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_vvs_i64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_load_dword s6, s[4:5], 0x34
; GFX10-SDAG-NEXT:    s_mov_b32 null, 0
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s4, v0
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s4, s6
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s4, s6
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_vvs_i64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_load_dword s6, s[4:5], 0x34
; GFX10-GISEL-NEXT:    s_mov_b32 null, 0
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s4, s6
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s4, s6
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlanex16_b32_vvs_i64:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX11-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
; GFX11-NEXT:    v_readfirstlane_b32 s5, v0
; GFX11-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-NEXT:    v_permlanex16_b32 v1, v1, s5, s4
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11-NEXT:    v_permlanex16_b32 v0, v0, s5, s4
; GFX11-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlanex16_b32_vvs_i64:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX12-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
; GFX12-NEXT:    v_readfirstlane_b32 s5, v0
; GFX12-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-NEXT:    v_permlanex16_b32 v1, v1, s5, s4
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX12-NEXT:    v_permlanex16_b32 v0, v0, s5, s4
; GFX12-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %v = call i64 @llvm.amdgcn.permlanex16.i64(i64 %src0, i64 %src0, i32 %tidx, i32 %src2, i1 false, i1 false)
  store i64 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vvs_f64(ptr addrspace(1) %out, double %src0, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_vvs_f64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_load_dword s6, s[4:5], 0x34
; GFX10-SDAG-NEXT:    s_mov_b32 null, 0
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s4, v0
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s4, s6
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s4, s6
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_vvs_f64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_load_dword s6, s[4:5], 0x34
; GFX10-GISEL-NEXT:    s_mov_b32 null, 0
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s4, s6
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s4, s6
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlanex16_b32_vvs_f64:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX11-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
; GFX11-NEXT:    v_readfirstlane_b32 s5, v0
; GFX11-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-NEXT:    v_permlanex16_b32 v1, v1, s5, s4
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11-NEXT:    v_permlanex16_b32 v0, v0, s5, s4
; GFX11-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlanex16_b32_vvs_f64:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX12-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
; GFX12-NEXT:    v_readfirstlane_b32 s5, v0
; GFX12-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-NEXT:    v_permlanex16_b32 v1, v1, s5, s4
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX12-NEXT:    v_permlanex16_b32 v0, v0, s5, s4
; GFX12-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %v = call double @llvm.amdgcn.permlanex16.f64(double %src0, double %src0, i32 %tidx, i32 %src2, i1 false, i1 false)
  store double %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vsv_i32(ptr addrspace(1) %out, i32 %src0, i32 %src1) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_vsv_i32:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s2, v1
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s3, s2
; GFX10-SDAG-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_vsv_i32:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_mov_b32 null, 0
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v1
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s3, s4
; GFX10-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlanex16_b32_vsv_i32:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v1, s2
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s2, v0
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, 0
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s3, s2
; GFX11-SDAG-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlanex16_b32_vsv_i32:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v1, 0
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s3, s4
; GFX11-GISEL-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlanex16_b32_vsv_i32:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v1, s2
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s2, v0
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, 0
; GFX12-SDAG-NEXT:    s_wait_alu 0xf1ff
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s3, s2
; GFX12-SDAG-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlanex16_b32_vsv_i32:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v1, 0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s3, s4
; GFX12-GISEL-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidy = call i32 @llvm.amdgcn.workitem.id.y()
  %v = call i32 @llvm.amdgcn.permlanex16.i32(i32 %src0, i32 %src0, i32 %src1, i32 %tidy, i1 false, i1 false)
  store i32 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vsv_f32(ptr addrspace(1) %out, float %src0, i32 %src1) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_vsv_f32:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s2, v1
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s3, s2
; GFX10-SDAG-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_vsv_f32:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_mov_b32 null, 0
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v1
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s3, s4
; GFX10-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlanex16_b32_vsv_f32:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v1, s2
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s2, v0
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, 0
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s3, s2
; GFX11-SDAG-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlanex16_b32_vsv_f32:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v1, 0
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s3, s4
; GFX11-GISEL-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlanex16_b32_vsv_f32:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v1, s2
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s2, v0
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, 0
; GFX12-SDAG-NEXT:    s_wait_alu 0xf1ff
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s3, s2
; GFX12-SDAG-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlanex16_b32_vsv_f32:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v1, 0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s4, v0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s3, s4
; GFX12-GISEL-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidy = call i32 @llvm.amdgcn.workitem.id.y()
  %v = call float @llvm.amdgcn.permlanex16.f32(float %src0, float %src0, i32 %src1, i32 %tidy, i1 false, i1 false)
  store float %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vsv_i64(ptr addrspace(1) %out, i64 %src0, i32 %src1) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_vsv_i64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_load_dword s6, s[4:5], 0x34
; GFX10-SDAG-NEXT:    s_mov_b32 null, 0
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s4, v1
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s6, s4
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s6, s4
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_vsv_i64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_load_dword s6, s[4:5], 0x34
; GFX10-GISEL-NEXT:    s_mov_b32 null, 0
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v1
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s6, s4
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s6, s4
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlanex16_b32_vsv_i64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX11-SDAG-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s5, v0
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v1, s3 :: v_dual_mov_b32 v0, s2
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s4, s5
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s4, s5
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlanex16_b32_vsv_i64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX11-GISEL-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s5, v0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s4, s5
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s4, s5
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlanex16_b32_vsv_i64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX12-SDAG-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s5, v0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v1, s3 :: v_dual_mov_b32 v0, s2
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s4, s5
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s4, s5
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlanex16_b32_vsv_i64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX12-GISEL-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s5, v0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s4, s5
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s4, s5
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidy = call i32 @llvm.amdgcn.workitem.id.y()
  %v = call i64 @llvm.amdgcn.permlanex16.i64(i64 %src0, i64 %src0, i32 %src1, i32 %tidy, i1 false, i1 false)
  store i64 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vsv_f64(ptr addrspace(1) %out, double %src0, i32 %src1) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_vsv_f64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_load_dword s6, s[4:5], 0x34
; GFX10-SDAG-NEXT:    s_mov_b32 null, 0
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s4, v1
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s6, s4
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s6, s4
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_vsv_f64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_load_dword s6, s[4:5], 0x34
; GFX10-GISEL-NEXT:    s_mov_b32 null, 0
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v1
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s6, s4
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s6, s4
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlanex16_b32_vsv_f64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX11-SDAG-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s5, v0
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v1, s3 :: v_dual_mov_b32 v0, s2
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s4, s5
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s4, s5
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlanex16_b32_vsv_f64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX11-GISEL-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s5, v0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s4, s5
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s4, s5
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlanex16_b32_vsv_f64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX12-SDAG-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s5, v0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v1, s3 :: v_dual_mov_b32 v0, s2
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s4, s5
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s4, s5
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlanex16_b32_vsv_f64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX12-GISEL-NEXT:    v_bfe_u32 v0, v0, 10, 10
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s5, v0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s4, s5
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s4, s5
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidy = call i32 @llvm.amdgcn.workitem.id.y()
  %v = call double @llvm.amdgcn.permlanex16.f64(double %src0, double %src0, i32 %src1, i32 %tidy, i1 false, i1 false)
  store double %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vss_fi_i32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlanex16_b32_vss_fi_i32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-NEXT:    s_load_dword s6, s[4:5], 0x34
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-NEXT:    v_permlanex16_b32 v0, v0, s3, s6 op_sel:[1,0]
; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlanex16_b32_vss_fi_i32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlanex16_b32 v0, v0, s3, s4 op_sel:[1,0]
; GFX11-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlanex16_b32_vss_fi_i32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlanex16_b32 v0, v0, s3, s4 op_sel:[1,0]
; GFX12-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-NEXT:    s_endpgm
  %v = call i32 @llvm.amdgcn.permlanex16.i32(i32 %src0, i32 %src0, i32 %src1, i32 %src2, i1 true, i1 false)
  store i32 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vss_fi_f32(ptr addrspace(1) %out, float %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlanex16_b32_vss_fi_f32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-NEXT:    s_load_dword s6, s[4:5], 0x34
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-NEXT:    v_permlanex16_b32 v0, v0, s3, s6 op_sel:[1,0]
; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlanex16_b32_vss_fi_f32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlanex16_b32 v0, v0, s3, s4 op_sel:[1,0]
; GFX11-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlanex16_b32_vss_fi_f32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlanex16_b32 v0, v0, s3, s4 op_sel:[1,0]
; GFX12-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-NEXT:    s_endpgm
  %v = call float @llvm.amdgcn.permlanex16.f32(float %src0, float %src0, i32 %src1, i32 %src2, i1 true, i1 false)
  store float %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vss_fi_i64(ptr addrspace(1) %out, i64 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_vss_fi_i64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s6, s7 op_sel:[1,0]
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s6, s7 op_sel:[1,0]
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_vss_fi_i64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s6, s7 op_sel:[1,0]
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s6, s7 op_sel:[1,0]
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlanex16_b32_vss_fi_i64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s4, s5 op_sel:[1,0]
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s4, s5 op_sel:[1,0]
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlanex16_b32_vss_fi_i64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s4, s5 op_sel:[1,0]
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s4, s5 op_sel:[1,0]
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlanex16_b32_vss_fi_i64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s4, s5 op_sel:[1,0]
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s4, s5 op_sel:[1,0]
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlanex16_b32_vss_fi_i64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s4, s5 op_sel:[1,0]
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s4, s5 op_sel:[1,0]
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %v = call i64 @llvm.amdgcn.permlanex16.i64(i64 %src0, i64 %src0, i32 %src1, i32 %src2, i1 true, i1 false)
  store i64 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vss_fi_f64(ptr addrspace(1) %out, double %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_vss_fi_f64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s6, s7 op_sel:[1,0]
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s6, s7 op_sel:[1,0]
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_vss_fi_f64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s6, s7 op_sel:[1,0]
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s6, s7 op_sel:[1,0]
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlanex16_b32_vss_fi_f64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s4, s5 op_sel:[1,0]
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s4, s5 op_sel:[1,0]
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlanex16_b32_vss_fi_f64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s4, s5 op_sel:[1,0]
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s4, s5 op_sel:[1,0]
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlanex16_b32_vss_fi_f64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s4, s5 op_sel:[1,0]
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s4, s5 op_sel:[1,0]
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlanex16_b32_vss_fi_f64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s4, s5 op_sel:[1,0]
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s4, s5 op_sel:[1,0]
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %v = call double @llvm.amdgcn.permlanex16.f64(double %src0, double %src0, i32 %src1, i32 %src2, i1 true, i1 false)
  store double %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vss_bc_i32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlanex16_b32_vss_bc_i32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-NEXT:    s_load_dword s6, s[4:5], 0x34
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-NEXT:    v_permlanex16_b32 v0, v0, s3, s6 op_sel:[0,1]
; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlanex16_b32_vss_bc_i32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlanex16_b32 v0, v0, s3, s4 op_sel:[0,1]
; GFX11-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlanex16_b32_vss_bc_i32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlanex16_b32 v0, v0, s3, s4 op_sel:[0,1]
; GFX12-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-NEXT:    s_endpgm
  %v = call i32 @llvm.amdgcn.permlanex16.i32(i32 %src0, i32 %src0, i32 %src1, i32 %src2, i1 false, i1 true)
  store i32 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vss_bc_f32(ptr addrspace(1) %out, float %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlanex16_b32_vss_bc_f32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-NEXT:    s_load_dword s6, s[4:5], 0x34
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-NEXT:    v_permlanex16_b32 v0, v0, s3, s6 op_sel:[0,1]
; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlanex16_b32_vss_bc_f32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlanex16_b32 v0, v0, s3, s4 op_sel:[0,1]
; GFX11-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlanex16_b32_vss_bc_f32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlanex16_b32 v0, v0, s3, s4 op_sel:[0,1]
; GFX12-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-NEXT:    s_endpgm
  %v = call float @llvm.amdgcn.permlanex16.f32(float %src0, float %src0, i32 %src1, i32 %src2, i1 false, i1 true)
  store float %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vss_bc_i64(ptr addrspace(1) %out, i64 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_vss_bc_i64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s6, s7 op_sel:[0,1]
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s6, s7 op_sel:[0,1]
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_vss_bc_i64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s6, s7 op_sel:[0,1]
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s6, s7 op_sel:[0,1]
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlanex16_b32_vss_bc_i64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s4, s5 op_sel:[0,1]
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s4, s5 op_sel:[0,1]
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlanex16_b32_vss_bc_i64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s4, s5 op_sel:[0,1]
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s4, s5 op_sel:[0,1]
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlanex16_b32_vss_bc_i64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s4, s5 op_sel:[0,1]
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s4, s5 op_sel:[0,1]
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlanex16_b32_vss_bc_i64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s4, s5 op_sel:[0,1]
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s4, s5 op_sel:[0,1]
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %v = call i64 @llvm.amdgcn.permlanex16.i64(i64 %src0, i64 %src0, i32 %src1, i32 %src2, i1 false, i1 true)
  store i64 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vss_bc_f64(ptr addrspace(1) %out, double %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_vss_bc_f64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s6, s7 op_sel:[0,1]
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s6, s7 op_sel:[0,1]
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_vss_bc_f64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s6, s7 op_sel:[0,1]
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s6, s7 op_sel:[0,1]
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlanex16_b32_vss_bc_f64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s4, s5 op_sel:[0,1]
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s4, s5 op_sel:[0,1]
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlanex16_b32_vss_bc_f64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s4, s5 op_sel:[0,1]
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s4, s5 op_sel:[0,1]
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlanex16_b32_vss_bc_f64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s4, s5 op_sel:[0,1]
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s4, s5 op_sel:[0,1]
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlanex16_b32_vss_bc_f64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s4, s5 op_sel:[0,1]
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s4, s5 op_sel:[0,1]
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %v = call double @llvm.amdgcn.permlanex16.f64(double %src0, double %src0, i32 %src1, i32 %src2, i1 false, i1 true)
  store double %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vss_fi_bc_i32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlanex16_b32_vss_fi_bc_i32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-NEXT:    s_load_dword s6, s[4:5], 0x34
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-NEXT:    v_permlanex16_b32 v0, v0, s3, s6 op_sel:[1,1]
; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlanex16_b32_vss_fi_bc_i32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlanex16_b32 v0, v0, s3, s4 op_sel:[1,1]
; GFX11-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlanex16_b32_vss_fi_bc_i32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlanex16_b32 v0, v0, s3, s4 op_sel:[1,1]
; GFX12-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-NEXT:    s_endpgm
  %v = call i32 @llvm.amdgcn.permlanex16.i32(i32 %src0, i32 %src0, i32 %src1, i32 %src2, i1 true, i1 true)
  store i32 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vss_fi_bc_f32(ptr addrspace(1) %out, float %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlanex16_b32_vss_fi_bc_f32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-NEXT:    s_load_dword s6, s[4:5], 0x34
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-NEXT:    v_permlanex16_b32 v0, v0, s3, s6 op_sel:[1,1]
; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlanex16_b32_vss_fi_bc_f32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlanex16_b32 v0, v0, s3, s4 op_sel:[1,1]
; GFX11-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlanex16_b32_vss_fi_bc_f32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-NEXT:    s_load_b32 s4, s[4:5], 0x34
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlanex16_b32 v0, v0, s3, s4 op_sel:[1,1]
; GFX12-NEXT:    global_store_b32 v1, v0, s[0:1]
; GFX12-NEXT:    s_endpgm
  %v = call float @llvm.amdgcn.permlanex16.f32(float %src0, float %src0, i32 %src1, i32 %src2, i1 true, i1 true)
  store float %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vss_fi_bc_i64(ptr addrspace(1) %out, i64 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_vss_fi_bc_i64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s6, s7 op_sel:[1,1]
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s6, s7 op_sel:[1,1]
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_vss_fi_bc_i64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s6, s7 op_sel:[1,1]
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s6, s7 op_sel:[1,1]
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlanex16_b32_vss_fi_bc_i64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s4, s5 op_sel:[1,1]
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s4, s5 op_sel:[1,1]
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlanex16_b32_vss_fi_bc_i64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s4, s5 op_sel:[1,1]
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s4, s5 op_sel:[1,1]
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlanex16_b32_vss_fi_bc_i64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s4, s5 op_sel:[1,1]
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s4, s5 op_sel:[1,1]
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlanex16_b32_vss_fi_bc_i64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s4, s5 op_sel:[1,1]
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s4, s5 op_sel:[1,1]
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %v = call i64 @llvm.amdgcn.permlanex16.i64(i64 %src0, i64 %src0, i32 %src1, i32 %src2, i1 true, i1 true)
  store i64 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_vss_fi_bc_f64(ptr addrspace(1) %out, double %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_vss_fi_bc_f64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s6, s7 op_sel:[1,1]
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s6, s7 op_sel:[1,1]
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_vss_fi_bc_f64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x34
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s3
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s6, s7 op_sel:[1,1]
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s6, s7 op_sel:[1,1]
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlanex16_b32_vss_fi_bc_f64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s4, s5 op_sel:[1,1]
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s4, s5 op_sel:[1,1]
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlanex16_b32_vss_fi_bc_f64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s4, s5 op_sel:[1,1]
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s4, s5 op_sel:[1,1]
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlanex16_b32_vss_fi_bc_f64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, s3
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v0, s2
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s4, s5 op_sel:[1,1]
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s4, s5 op_sel:[1,1]
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlanex16_b32_vss_fi_bc_f64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s4, s5 op_sel:[1,1]
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s4, s5 op_sel:[1,1]
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX12-GISEL-NEXT:    s_endpgm
  %v = call double @llvm.amdgcn.permlanex16.f64(double %src0, double %src0, i32 %src1, i32 %src2, i1 true, i1 true)
  store double %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_tid_tid_i32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlane16_b32_tid_tid_i32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX10-NEXT:    global_store_dword v1, v0, s[2:3]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlane16_b32_tid_tid_i32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX11-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlane16_b32_tid_tid_i32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX12-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX12-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %v = call i32 @llvm.amdgcn.permlane16.i32(i32 %tidx, i32 %tidx, i32 %src1, i32 %src2, i1 false, i1 false)
  store i32 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_tid_tid_f32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlane16_b32_tid_tid_f32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX10-NEXT:    global_store_dword v1, v0, s[2:3]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlane16_b32_tid_tid_f32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX11-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlane16_b32_tid_tid_f32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX12-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX12-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_f32 = bitcast i32 %tidx to float
  %v = call float @llvm.amdgcn.permlane16.f32(float %tidx_f32, float %tidx_f32, i32 %src1, i32 %src2, i1 false, i1 false)
  store float %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_tid_tid_i64(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlane16_b32_tid_tid_i64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_permlane16_b32 v1, v1, s0, s1
; GFX10-SDAG-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_tid_tid_i64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX10-GISEL-NEXT:    v_permlane16_b32 v1, v1, s0, s1
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlane16_b32_tid_tid_i64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_3)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v1, v1, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlane16_b32_tid_tid_i64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_3)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v1, v1, s0, s1
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlane16_b32_tid_tid_i64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_3)
; GFX12-SDAG-NEXT:    v_permlane16_b32 v1, v1, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlane16_b32_tid_tid_i64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_3)
; GFX12-GISEL-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v1, v1, s0, s1
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_i64 = zext i32 %tidx to i64
  %v = call i64 @llvm.amdgcn.permlane16.i64(i64 %tidx_i64, i64 %tidx_i64, i32 %src1, i32 %src2, i1 false, i1 false)
  store i64 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_tid_tid_f64(ptr addrspace(1) %out, float %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlane16_b32_tid_tid_f64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_permlane16_b32 v1, v1, s0, s1
; GFX10-SDAG-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_tid_tid_f64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX10-GISEL-NEXT:    v_permlane16_b32 v1, v1, s0, s1
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlane16_b32_tid_tid_f64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v1, v1, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlane16_b32_tid_tid_f64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v1, v1, s0, s1
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlane16_b32_tid_tid_f64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlane16_b32 v1, v1, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlane16_b32_tid_tid_f64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v1, v1, s0, s1
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_f32 = bitcast i32 %tidx to float
  %tidx_f64 = fpext float %tidx_f32 to double
  %v = call double @llvm.amdgcn.permlane16.f64(double %tidx_f64, double %tidx_f64, i32 %src1, i32 %src2, i1 false, i1 false)
  store double %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_undef_tid_i32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlane16_b32_undef_tid_i32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX10-NEXT:    global_store_dword v1, v0, s[2:3]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlane16_b32_undef_tid_i32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX11-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlane16_b32_undef_tid_i32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX12-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX12-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %undef = freeze i32 poison
  %v = call i32 @llvm.amdgcn.permlane16.i32(i32 %undef, i32 %tidx, i32 %src1, i32 %src2, i1 false, i1 false)
  store i32 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_undef_tid_f32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlane16_b32_undef_tid_f32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX10-NEXT:    global_store_dword v1, v0, s[2:3]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlane16_b32_undef_tid_f32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX11-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlane16_b32_undef_tid_f32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX12-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX12-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_f32 = bitcast i32 %tidx to float
  %undef = freeze float poison
  %v = call float @llvm.amdgcn.permlane16.f32(float %undef, float %tidx_f32, i32 %src1, i32 %src2, i1 false, i1 false)
  store float %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_undef_tid_i64(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlane16_b32_undef_tid_i64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_permlane16_b32 v1, v2, s0, s1
; GFX10-SDAG-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_undef_tid_i64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX10-GISEL-NEXT:    v_permlane16_b32 v1, v2, s0, s1
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlane16_b32_undef_tid_i64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v1, v2, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlane16_b32_undef_tid_i64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v1, v2, s0, s1
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlane16_b32_undef_tid_i64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlane16_b32 v1, v2, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlane16_b32_undef_tid_i64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v1, v2, s0, s1
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_i64 = zext i32 %tidx to i64
  %undef = freeze i64 poison
  %v = call i64 @llvm.amdgcn.permlane16.i64(i64 %undef, i64 %tidx_i64, i32 %src1, i32 %src2, i1 false, i1 false)
  store i64 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_undef_tid_f64(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlane16_b32_undef_tid_f64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_permlane16_b32 v1, v1, s0, s1
; GFX10-SDAG-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_undef_tid_f64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX10-GISEL-NEXT:    v_permlane16_b32 v1, v1, s0, s1
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlane16_b32_undef_tid_f64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v1, v1, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlane16_b32_undef_tid_f64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v1, v1, s0, s1
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlane16_b32_undef_tid_f64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlane16_b32 v1, v1, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlane16_b32_undef_tid_f64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlane16_b32 v0, v0, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v1, v1, s0, s1
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_f32 = bitcast i32 %tidx to float
  %tidx_f64 = fpext float %tidx_f32 to double
  %undef = freeze double poison
  %v = call double @llvm.amdgcn.permlane16.f64(double %undef, double %tidx_f64, i32 %src1, i32 %src2, i1 false, i1 false)
  store double %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_i_tid_i32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlane16_b32_i_tid_i32:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, 0x3039
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_permlane16_b32 v1, v0, s0, s1
; GFX10-SDAG-NEXT:    global_store_dword v2, v1, s[2:3]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_i_tid_i32:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0x3039
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_permlane16_b32 v1, v0, s0, s1
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, 0
; GFX10-GISEL-NEXT:    global_store_dword v0, v1, s[2:3]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlane16_b32_i_tid_i32:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v1, 0x3039 :: v_dual_mov_b32 v2, 0
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v1, v0, s0, s1
; GFX11-SDAG-NEXT:    global_store_b32 v2, v1, s[2:3]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlane16_b32_i_tid_i32:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v1, 0x3039
; GFX11-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v1, v0, s0, s1
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v0, 0
; GFX11-GISEL-NEXT:    global_store_b32 v0, v1, s[2:3]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlane16_b32_i_tid_i32:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v1, 0x3039 :: v_dual_mov_b32 v2, 0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-SDAG-NEXT:    v_permlane16_b32 v1, v0, s0, s1
; GFX12-SDAG-NEXT:    global_store_b32 v2, v1, s[2:3]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlane16_b32_i_tid_i32:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v1, 0x3039
; GFX12-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-GISEL-NEXT:    v_permlane16_b32 v1, v0, s0, s1
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v0, 0
; GFX12-GISEL-NEXT:    global_store_b32 v0, v1, s[2:3]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %v = call i32 @llvm.amdgcn.permlane16.i32(i32 12345, i32 %tidx, i32 %src1, i32 %src2, i1 false, i1 false)
  store i32 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_i_tid_f32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlane16_b32_i_tid_f32:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, 0x449a5000
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_permlane16_b32 v1, v0, s0, s1
; GFX10-SDAG-NEXT:    global_store_dword v2, v1, s[2:3]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_i_tid_f32:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0x449a5000
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_permlane16_b32 v1, v0, s0, s1
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, 0
; GFX10-GISEL-NEXT:    global_store_dword v0, v1, s[2:3]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlane16_b32_i_tid_f32:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v1, 0x449a5000 :: v_dual_mov_b32 v2, 0
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v1, v0, s0, s1
; GFX11-SDAG-NEXT:    global_store_b32 v2, v1, s[2:3]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlane16_b32_i_tid_f32:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v1, 0x449a5000
; GFX11-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v1, v0, s0, s1
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v0, 0
; GFX11-GISEL-NEXT:    global_store_b32 v0, v1, s[2:3]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlane16_b32_i_tid_f32:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v1, 0x449a5000 :: v_dual_mov_b32 v2, 0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-SDAG-NEXT:    v_permlane16_b32 v1, v0, s0, s1
; GFX12-SDAG-NEXT:    global_store_b32 v2, v1, s[2:3]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlane16_b32_i_tid_f32:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v1, 0x449a5000
; GFX12-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-GISEL-NEXT:    v_permlane16_b32 v1, v0, s0, s1
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v0, 0
; GFX12-GISEL-NEXT:    global_store_b32 v0, v1, s[2:3]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_f32 = bitcast i32 %tidx to float
  %v = call float @llvm.amdgcn.permlane16.f32(float 1234.5, float %tidx_f32, i32 %src1, i32 %src2, i1 false, i1 false)
  store float %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_i_tid_i64(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlane16_b32_i_tid_i64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, 0x3039
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v3, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX10-SDAG-NEXT:    v_permlane16_b32 v1, v0, s0, s1
; GFX10-SDAG-NEXT:    global_store_dwordx2 v3, v[1:2], s[2:3]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_i_tid_i64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0x3039
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v3, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_permlane16_b32 v1, v0, s0, s1
; GFX10-GISEL-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX10-GISEL-NEXT:    global_store_dwordx2 v3, v[1:2], s[2:3]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlane16_b32_i_tid_i64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, 0x3039
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v3, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v1, v0, s0, s1
; GFX11-SDAG-NEXT:    global_store_b64 v3, v[1:2], s[2:3]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlane16_b32_i_tid_i64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v1, 0x3039 :: v_dual_mov_b32 v2, 0
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v3, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_3)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v1, v0, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX11-GISEL-NEXT:    global_store_b64 v3, v[1:2], s[2:3]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlane16_b32_i_tid_i64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, 0x3039
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v3, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v1, v0, s0, s1
; GFX12-SDAG-NEXT:    global_store_b64 v3, v[1:2], s[2:3]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlane16_b32_i_tid_i64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v1, 0x3039 :: v_dual_mov_b32 v2, 0
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v3, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_3)
; GFX12-GISEL-NEXT:    v_permlane16_b32 v1, v0, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX12-GISEL-NEXT:    global_store_b64 v3, v[1:2], s[2:3]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_i64 = zext i32 %tidx to i64
  %v = call i64 @llvm.amdgcn.permlane16.i64(i64 12345, i64 %tidx_i64, i32 %src1, i32 %src2, i1 false, i1 false)
  store i64 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_i_tid_f64(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlane16_b32_i_tid_f64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v3, 0x40934a00
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v4, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_permlane16_b32 v3, v1, s0, s1
; GFX10-SDAG-NEXT:    v_permlane16_b32 v2, v0, s0, s1
; GFX10-SDAG-NEXT:    global_store_dwordx2 v4, v[2:3], s[2:3]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_i_tid_f64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v3, 0x40934a00
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v4, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_permlane16_b32 v2, v0, s0, s1
; GFX10-GISEL-NEXT:    v_permlane16_b32 v3, v1, s0, s1
; GFX10-GISEL-NEXT:    global_store_dwordx2 v4, v[2:3], s[2:3]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlane16_b32_i_tid_f64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v3, 0x40934a00 :: v_dual_mov_b32 v2, 0
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v4, 0
; GFX11-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v3, v1, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v2, v0, s0, s1
; GFX11-SDAG-NEXT:    global_store_b64 v4, v[2:3], s[2:3]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlane16_b32_i_tid_f64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v3, 0x40934a00
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v4, 0
; GFX11-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v2, v0, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v3, v1, s0, s1
; GFX11-GISEL-NEXT:    global_store_b64 v4, v[2:3], s[2:3]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlane16_b32_i_tid_f64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v3, 0x40934a00 :: v_dual_mov_b32 v2, 0
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v4, 0
; GFX12-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlane16_b32 v3, v1, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v2, v0, s0, s1
; GFX12-SDAG-NEXT:    global_store_b64 v4, v[2:3], s[2:3]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlane16_b32_i_tid_f64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v3, 0x40934a00
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v4, 0
; GFX12-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlane16_b32 v2, v0, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v3, v1, s0, s1
; GFX12-GISEL-NEXT:    global_store_b64 v4, v[2:3], s[2:3]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_f32 = bitcast i32 %tidx to float
  %tidx_f64 = fpext float %tidx_f32 to double
  %v = call double @llvm.amdgcn.permlane16.f64(double 1234.5, double %tidx_f64, i32 %src1, i32 %src2, i1 false, i1 false)
  store double %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_i_tid_fi_i32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlane16_b32_i_tid_fi_i32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX10-NEXT:    global_store_dword v1, v0, s[2:3]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlane16_b32_i_tid_fi_i32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX11-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlane16_b32_i_tid_fi_i32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX12-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX12-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %undef = freeze i32 poison
  %v = call i32 @llvm.amdgcn.permlane16.i32(i32 %undef, i32 %tidx, i32 %src1, i32 %src2, i1 true, i1 false)
  store i32 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_i_tid_fi_f32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlane16_b32_i_tid_fi_f32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX10-NEXT:    global_store_dword v1, v0, s[2:3]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlane16_b32_i_tid_fi_f32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX11-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlane16_b32_i_tid_fi_f32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX12-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX12-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_f32 = bitcast i32 %tidx to float
  %undef = freeze float poison
  %v = call float @llvm.amdgcn.permlane16.f32(float %undef, float %tidx_f32, i32 %src1, i32 %src2, i1 true, i1 false)
  store float %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_i_tid_fi_i64(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlane16_b32_i_tid_fi_i64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_permlane16_b32 v1, v2, s0, s1 op_sel:[1,0]
; GFX10-SDAG-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_i_tid_fi_i64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX10-GISEL-NEXT:    v_permlane16_b32 v1, v2, s0, s1 op_sel:[1,0]
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlane16_b32_i_tid_fi_i64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v1, v2, s0, s1 op_sel:[1,0]
; GFX11-SDAG-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlane16_b32_i_tid_fi_i64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX11-GISEL-NEXT:    v_permlane16_b32 v1, v2, s0, s1 op_sel:[1,0]
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlane16_b32_i_tid_fi_i64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlane16_b32 v1, v2, s0, s1 op_sel:[1,0]
; GFX12-SDAG-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlane16_b32_i_tid_fi_i64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX12-GISEL-NEXT:    v_permlane16_b32 v1, v2, s0, s1 op_sel:[1,0]
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_i64 = zext i32 %tidx to i64
  %undef = freeze i64 poison
  %v = call i64 @llvm.amdgcn.permlane16.i64(i64 %undef, i64 %tidx_i64, i32 %src1, i32 %src2, i1 true, i1 false)
  store i64 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_i_tid_fi_f64(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlane16_b32_i_tid_fi_f64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_permlane16_b32 v1, v1, s0, s1 op_sel:[1,0]
; GFX10-SDAG-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_i_tid_fi_f64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX10-GISEL-NEXT:    v_permlane16_b32 v1, v1, s0, s1 op_sel:[1,0]
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlane16_b32_i_tid_fi_f64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v1, v1, s0, s1 op_sel:[1,0]
; GFX11-SDAG-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlane16_b32_i_tid_fi_f64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX11-GISEL-NEXT:    v_permlane16_b32 v1, v1, s0, s1 op_sel:[1,0]
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlane16_b32_i_tid_fi_f64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlane16_b32 v1, v1, s0, s1 op_sel:[1,0]
; GFX12-SDAG-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlane16_b32_i_tid_fi_f64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX12-GISEL-NEXT:    v_permlane16_b32 v1, v1, s0, s1 op_sel:[1,0]
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_f32 = bitcast i32 %tidx to float
  %tidx_f64 = fpext float %tidx_f32 to double
  %undef = freeze double poison
  %v = call double @llvm.amdgcn.permlane16.f64(double %undef, double %tidx_f64, i32 %src1, i32 %src2, i1 true, i1 false)
  store double %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_i_tid_bc_i32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlane16_b32_i_tid_bc_i32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX10-NEXT:    global_store_dword v1, v0, s[2:3]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlane16_b32_i_tid_bc_i32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX11-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlane16_b32_i_tid_bc_i32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX12-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX12-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %undef = freeze i32 poison
  %v = call i32 @llvm.amdgcn.permlane16.i32(i32 %undef, i32 %tidx, i32 %src1, i32 %src2, i1 false, i1 true)
  store i32 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_i_tid_bc_f32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlane16_b32_i_tid_bc_f32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX10-NEXT:    global_store_dword v1, v0, s[2:3]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlane16_b32_i_tid_bc_f32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX11-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlane16_b32_i_tid_bc_f32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX12-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX12-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_f32 = bitcast i32 %tidx to float
  %undef = freeze float poison
  %v = call float @llvm.amdgcn.permlane16.f32(float %undef, float %tidx_f32, i32 %src1, i32 %src2, i1 false, i1 true)
  store float %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_i_tid_bc_i64(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlane16_b32_i_tid_bc_i64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_permlane16_b32 v1, v2, s0, s1 op_sel:[0,1]
; GFX10-SDAG-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_i_tid_bc_i64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX10-GISEL-NEXT:    v_permlane16_b32 v1, v2, s0, s1 op_sel:[0,1]
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlane16_b32_i_tid_bc_i64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v1, v2, s0, s1 op_sel:[0,1]
; GFX11-SDAG-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlane16_b32_i_tid_bc_i64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX11-GISEL-NEXT:    v_permlane16_b32 v1, v2, s0, s1 op_sel:[0,1]
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlane16_b32_i_tid_bc_i64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlane16_b32 v1, v2, s0, s1 op_sel:[0,1]
; GFX12-SDAG-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlane16_b32_i_tid_bc_i64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX12-GISEL-NEXT:    v_permlane16_b32 v1, v2, s0, s1 op_sel:[0,1]
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_i64 = zext i32 %tidx to i64
  %undef = freeze i64 poison
  %v = call i64 @llvm.amdgcn.permlane16.i64(i64 %undef, i64 %tidx_i64, i32 %src1, i32 %src2, i1 false, i1 true)
  store i64 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_i_tid_bc_f64(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlane16_b32_i_tid_bc_f64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_permlane16_b32 v1, v1, s0, s1 op_sel:[0,1]
; GFX10-SDAG-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_i_tid_bc_f64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX10-GISEL-NEXT:    v_permlane16_b32 v1, v1, s0, s1 op_sel:[0,1]
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlane16_b32_i_tid_bc_f64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v1, v1, s0, s1 op_sel:[0,1]
; GFX11-SDAG-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlane16_b32_i_tid_bc_f64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX11-GISEL-NEXT:    v_permlane16_b32 v1, v1, s0, s1 op_sel:[0,1]
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlane16_b32_i_tid_bc_f64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlane16_b32 v1, v1, s0, s1 op_sel:[0,1]
; GFX12-SDAG-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlane16_b32_i_tid_bc_f64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX12-GISEL-NEXT:    v_permlane16_b32 v1, v1, s0, s1 op_sel:[0,1]
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_f32 = bitcast i32 %tidx to float
  %tidx_f64 = fpext float %tidx_f32 to double
  %undef = freeze double poison
  %v = call double @llvm.amdgcn.permlane16.f64(double %undef, double %tidx_f64, i32 %src1, i32 %src2, i1 false, i1 true)
  store double %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_i_tid_fi_bc_i32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlane16_b32_i_tid_fi_bc_i32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX10-NEXT:    global_store_dword v1, v0, s[2:3]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlane16_b32_i_tid_fi_bc_i32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX11-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlane16_b32_i_tid_fi_bc_i32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX12-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX12-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %undef = freeze i32 poison
  %v = call i32 @llvm.amdgcn.permlane16.i32(i32 %undef, i32 %tidx, i32 %src1, i32 %src2, i1 true, i1 true)
  store i32 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_i_tid_fi_bc_f32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlane16_b32_i_tid_fi_bc_f32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX10-NEXT:    global_store_dword v1, v0, s[2:3]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlane16_b32_i_tid_fi_bc_f32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX11-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlane16_b32_i_tid_fi_bc_f32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX12-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX12-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_f32 = bitcast i32 %tidx to float
  %undef = freeze float poison
  %v = call float @llvm.amdgcn.permlane16.f32(float %undef, float %tidx_f32, i32 %src1, i32 %src2, i1 true, i1 true)
  store float %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_i_tid_fi_bc_i64(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlane16_b32_i_tid_fi_bc_i64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_permlane16_b32 v1, v2, s0, s1 op_sel:[1,1]
; GFX10-SDAG-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_i_tid_fi_bc_i64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX10-GISEL-NEXT:    v_permlane16_b32 v1, v2, s0, s1 op_sel:[1,1]
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlane16_b32_i_tid_fi_bc_i64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v1, v2, s0, s1 op_sel:[1,1]
; GFX11-SDAG-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlane16_b32_i_tid_fi_bc_i64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX11-GISEL-NEXT:    v_permlane16_b32 v1, v2, s0, s1 op_sel:[1,1]
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlane16_b32_i_tid_fi_bc_i64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlane16_b32 v1, v2, s0, s1 op_sel:[1,1]
; GFX12-SDAG-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlane16_b32_i_tid_fi_bc_i64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX12-GISEL-NEXT:    v_permlane16_b32 v1, v2, s0, s1 op_sel:[1,1]
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_i64 = zext i32 %tidx to i64
  %undef = freeze i64 poison
  %v = call i64 @llvm.amdgcn.permlane16.i64(i64 %undef, i64 %tidx_i64, i32 %src1, i32 %src2, i1 true, i1 true)
  store i64 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlane16_b32_i_tid_fi_bc_f64(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlane16_b32_i_tid_fi_bc_f64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_permlane16_b32 v1, v1, s0, s1 op_sel:[1,1]
; GFX10-SDAG-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlane16_b32_i_tid_fi_bc_f64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX10-GISEL-NEXT:    v_permlane16_b32 v1, v1, s0, s1 op_sel:[1,1]
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlane16_b32_i_tid_fi_bc_f64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v1, v1, s0, s1 op_sel:[1,1]
; GFX11-SDAG-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlane16_b32_i_tid_fi_bc_f64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX11-GISEL-NEXT:    v_permlane16_b32 v1, v1, s0, s1 op_sel:[1,1]
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlane16_b32_i_tid_fi_bc_f64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlane16_b32 v1, v1, s0, s1 op_sel:[1,1]
; GFX12-SDAG-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlane16_b32_i_tid_fi_bc_f64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlane16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX12-GISEL-NEXT:    v_permlane16_b32 v1, v1, s0, s1 op_sel:[1,1]
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_f32 = bitcast i32 %tidx to float
  %tidx_f64 = fpext float %tidx_f32 to double
  %undef = freeze double poison
  %v = call double @llvm.amdgcn.permlane16.f64(double %undef, double %tidx_f64, i32 %src1, i32 %src2, i1 true, i1 true)
  store double %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_tid_tid_i32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlanex16_b32_tid_tid_i32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX10-NEXT:    global_store_dword v1, v0, s[2:3]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlanex16_b32_tid_tid_i32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX11-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlanex16_b32_tid_tid_i32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX12-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX12-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %v = call i32 @llvm.amdgcn.permlanex16.i32(i32 %tidx, i32 %tidx, i32 %src1, i32 %src2, i1 false, i1 false)
  store i32 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_tid_tid_f32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlanex16_b32_tid_tid_f32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX10-NEXT:    global_store_dword v1, v0, s[2:3]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlanex16_b32_tid_tid_f32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX11-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlanex16_b32_tid_tid_f32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX12-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX12-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_f32 = bitcast i32 %tidx to float
  %v = call float @llvm.amdgcn.permlanex16.f32(float %tidx_f32, float %tidx_f32, i32 %src1, i32 %src2, i1 false, i1 false)
  store float %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_tid_tid_i64(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_tid_tid_i64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s0, s1
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_tid_tid_i64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s0, s1
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlanex16_b32_tid_tid_i64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_3)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlanex16_b32_tid_tid_i64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_3)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s0, s1
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlanex16_b32_tid_tid_i64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_3)
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlanex16_b32_tid_tid_i64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_3)
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s0, s1
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_i64 = zext i32 %tidx to i64
  %v = call i64 @llvm.amdgcn.permlanex16.i64(i64 %tidx_i64, i64 %tidx_i64, i32 %src1, i32 %src2, i1 false, i1 false)
  store i64 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_tid_tid_f64(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_tid_tid_f64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s0, s1
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_tid_tid_f64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s0, s1
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlanex16_b32_tid_tid_f64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlanex16_b32_tid_tid_f64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s0, s1
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlanex16_b32_tid_tid_f64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlanex16_b32_tid_tid_f64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s0, s1
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_f32 = bitcast i32 %tidx to float
  %tidx_f64 = fpext float %tidx_f32 to double
  %v = call double @llvm.amdgcn.permlanex16.f64(double %tidx_f64, double %tidx_f64, i32 %src1, i32 %src2, i1 false, i1 false)
  store double %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_undef_tid_i32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlanex16_b32_undef_tid_i32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX10-NEXT:    global_store_dword v1, v0, s[2:3]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlanex16_b32_undef_tid_i32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX11-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlanex16_b32_undef_tid_i32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX12-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX12-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %undef = freeze i32 poison
  %v = call i32 @llvm.amdgcn.permlanex16.i32(i32 %undef, i32 %tidx, i32 %src1, i32 %src2, i1 false, i1 false)
  store i32 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_undef_tid_f32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlanex16_b32_undef_tid_f32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX10-NEXT:    global_store_dword v1, v0, s[2:3]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlanex16_b32_undef_tid_f32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX11-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlanex16_b32_undef_tid_f32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX12-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX12-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_f32 = bitcast i32 %tidx to float
  %undef = freeze float poison
  %v = call float @llvm.amdgcn.permlanex16.f32(float %undef, float %tidx_f32, i32 %src1, i32 %src2, i1 false, i1 false)
  store float %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_undef_tid_i64(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_undef_tid_i64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v1, v2, s0, s1
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_undef_tid_i64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v1, v2, s0, s1
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlanex16_b32_undef_tid_i64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v1, v2, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlanex16_b32_undef_tid_i64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v1, v2, s0, s1
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlanex16_b32_undef_tid_i64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v1, v2, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlanex16_b32_undef_tid_i64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v1, v2, s0, s1
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_i64 = zext i32 %tidx to i64
  %undef = freeze i64 poison
  %v = call i64 @llvm.amdgcn.permlanex16.i64(i64 %undef, i64 %tidx_i64, i32 %src1, i32 %src2, i1 false, i1 false)
  store i64 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_undef_tid_f64(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_undef_tid_f64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s0, s1
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_undef_tid_f64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s0, s1
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlanex16_b32_undef_tid_f64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlanex16_b32_undef_tid_f64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s0, s1
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlanex16_b32_undef_tid_f64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlanex16_b32_undef_tid_f64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s0, s1
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_f32 = bitcast i32 %tidx to float
  %tidx_f64 = fpext float %tidx_f32 to double
  %undef = freeze double poison
  %v = call double @llvm.amdgcn.permlanex16.f64(double %undef, double %tidx_f64, i32 %src1, i32 %src2, i1 false, i1 false)
  store double %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_i_tid_i32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_i_tid_i32:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, 0x3039
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v1, v0, s0, s1
; GFX10-SDAG-NEXT:    global_store_dword v2, v1, s[2:3]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_i_tid_i32:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0x3039
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v1, v0, s0, s1
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, 0
; GFX10-GISEL-NEXT:    global_store_dword v0, v1, s[2:3]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlanex16_b32_i_tid_i32:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v1, 0x3039 :: v_dual_mov_b32 v2, 0
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v1, v0, s0, s1
; GFX11-SDAG-NEXT:    global_store_b32 v2, v1, s[2:3]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlanex16_b32_i_tid_i32:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v1, 0x3039
; GFX11-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v1, v0, s0, s1
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v0, 0
; GFX11-GISEL-NEXT:    global_store_b32 v0, v1, s[2:3]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlanex16_b32_i_tid_i32:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v1, 0x3039 :: v_dual_mov_b32 v2, 0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v1, v0, s0, s1
; GFX12-SDAG-NEXT:    global_store_b32 v2, v1, s[2:3]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlanex16_b32_i_tid_i32:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v1, 0x3039
; GFX12-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v1, v0, s0, s1
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v0, 0
; GFX12-GISEL-NEXT:    global_store_b32 v0, v1, s[2:3]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %v = call i32 @llvm.amdgcn.permlanex16.i32(i32 12345, i32 %tidx, i32 %src1, i32 %src2, i1 false, i1 false)
  store i32 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_i_tid_f32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_i_tid_f32:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, 0x449a5000
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v1, v0, s0, s1
; GFX10-SDAG-NEXT:    global_store_dword v2, v1, s[2:3]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_i_tid_f32:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0x449a5000
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v1, v0, s0, s1
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, 0
; GFX10-GISEL-NEXT:    global_store_dword v0, v1, s[2:3]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlanex16_b32_i_tid_f32:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v1, 0x449a5000 :: v_dual_mov_b32 v2, 0
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v1, v0, s0, s1
; GFX11-SDAG-NEXT:    global_store_b32 v2, v1, s[2:3]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlanex16_b32_i_tid_f32:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v1, 0x449a5000
; GFX11-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v1, v0, s0, s1
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v0, 0
; GFX11-GISEL-NEXT:    global_store_b32 v0, v1, s[2:3]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlanex16_b32_i_tid_f32:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v1, 0x449a5000 :: v_dual_mov_b32 v2, 0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v1, v0, s0, s1
; GFX12-SDAG-NEXT:    global_store_b32 v2, v1, s[2:3]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlanex16_b32_i_tid_f32:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v1, 0x449a5000
; GFX12-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v1, v0, s0, s1
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v0, 0
; GFX12-GISEL-NEXT:    global_store_b32 v0, v1, s[2:3]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_f32 = bitcast i32 %tidx to float
  %v = call float @llvm.amdgcn.permlanex16.f32(float 1234.5, float %tidx_f32, i32 %src1, i32 %src2, i1 false, i1 false)
  store float %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_i_tid_i64(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_i_tid_i64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v1, 0x3039
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v3, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v1, v0, s0, s1
; GFX10-SDAG-NEXT:    global_store_dwordx2 v3, v[1:2], s[2:3]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_i_tid_i64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0x3039
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v3, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v1, v0, s0, s1
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX10-GISEL-NEXT:    global_store_dwordx2 v3, v[1:2], s[2:3]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlanex16_b32_i_tid_i64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, 0x3039
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v3, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v1, v0, s0, s1
; GFX11-SDAG-NEXT:    global_store_b64 v3, v[1:2], s[2:3]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlanex16_b32_i_tid_i64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v1, 0x3039 :: v_dual_mov_b32 v2, 0
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v3, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_3)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v1, v0, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX11-GISEL-NEXT:    global_store_b64 v3, v[1:2], s[2:3]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlanex16_b32_i_tid_i64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v1, 0x3039
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v3, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v1, v0, s0, s1
; GFX12-SDAG-NEXT:    global_store_b64 v3, v[1:2], s[2:3]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlanex16_b32_i_tid_i64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v1, 0x3039 :: v_dual_mov_b32 v2, 0
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v3, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_3)
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v1, v0, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX12-GISEL-NEXT:    global_store_b64 v3, v[1:2], s[2:3]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_i64 = zext i32 %tidx to i64
  %v = call i64 @llvm.amdgcn.permlanex16.i64(i64 12345, i64 %tidx_i64, i32 %src1, i32 %src2, i1 false, i1 false)
  store i64 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_i_tid_f64(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_i_tid_f64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v3, 0x40934a00
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v4, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v3, v1, s0, s1
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v2, v0, s0, s1
; GFX10-SDAG-NEXT:    global_store_dwordx2 v4, v[2:3], s[2:3]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_i_tid_f64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v3, 0x40934a00
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v4, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v2, v0, s0, s1
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v3, v1, s0, s1
; GFX10-GISEL-NEXT:    global_store_dwordx2 v4, v[2:3], s[2:3]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlanex16_b32_i_tid_f64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_dual_mov_b32 v3, 0x40934a00 :: v_dual_mov_b32 v2, 0
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v4, 0
; GFX11-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v3, v1, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v2, v0, s0, s1
; GFX11-SDAG-NEXT:    global_store_b64 v4, v[2:3], s[2:3]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlanex16_b32_i_tid_f64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v3, 0x40934a00
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v4, 0
; GFX11-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v2, v0, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v3, v1, s0, s1
; GFX11-GISEL-NEXT:    global_store_b64 v4, v[2:3], s[2:3]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlanex16_b32_i_tid_f64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_dual_mov_b32 v3, 0x40934a00 :: v_dual_mov_b32 v2, 0
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v4, 0
; GFX12-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v3, v1, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v2, v0, s0, s1
; GFX12-SDAG-NEXT:    global_store_b64 v4, v[2:3], s[2:3]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlanex16_b32_i_tid_f64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_dual_mov_b32 v2, 0 :: v_dual_mov_b32 v3, 0x40934a00
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v4, 0
; GFX12-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v2, v0, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v3, v1, s0, s1
; GFX12-GISEL-NEXT:    global_store_b64 v4, v[2:3], s[2:3]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_f32 = bitcast i32 %tidx to float
  %tidx_f64 = fpext float %tidx_f32 to double
  %v = call double @llvm.amdgcn.permlanex16.f64(double 1234.5, double %tidx_f64, i32 %src1, i32 %src2, i1 false, i1 false)
  store double %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_i_tid_fi_i32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlanex16_b32_i_tid_fi_i32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX10-NEXT:    global_store_dword v1, v0, s[2:3]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlanex16_b32_i_tid_fi_i32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX11-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlanex16_b32_i_tid_fi_i32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX12-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX12-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %undef = freeze i32 poison
  %v = call i32 @llvm.amdgcn.permlanex16.i32(i32 %undef, i32 %tidx, i32 %src1, i32 %src2, i1 true, i1 false)
  store i32 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_i_tid_fi_f32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlanex16_b32_i_tid_fi_f32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX10-NEXT:    global_store_dword v1, v0, s[2:3]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlanex16_b32_i_tid_fi_f32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX11-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlanex16_b32_i_tid_fi_f32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX12-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX12-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_f32 = bitcast i32 %tidx to float
  %undef = freeze float poison
  %v = call float @llvm.amdgcn.permlanex16.f32(float %undef, float %tidx_f32, i32 %src1, i32 %src2, i1 true, i1 false)
  store float %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_i_tid_fi_i64(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_i_tid_fi_i64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v1, v2, s0, s1 op_sel:[1,0]
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_i_tid_fi_i64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v1, v2, s0, s1 op_sel:[1,0]
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlanex16_b32_i_tid_fi_i64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v1, v2, s0, s1 op_sel:[1,0]
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlanex16_b32_i_tid_fi_i64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v1, v2, s0, s1 op_sel:[1,0]
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlanex16_b32_i_tid_fi_i64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v1, v2, s0, s1 op_sel:[1,0]
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlanex16_b32_i_tid_fi_i64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v1, v2, s0, s1 op_sel:[1,0]
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_i64 = zext i32 %tidx to i64
  %undef = freeze i64 poison
  %v = call i64 @llvm.amdgcn.permlanex16.i64(i64 %undef, i64 %tidx_i64, i32 %src1, i32 %src2, i1 true, i1 false)
  store i64 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_i_tid_fi_f64(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_i_tid_fi_f64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s0, s1 op_sel:[1,0]
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_i_tid_fi_f64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s0, s1 op_sel:[1,0]
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlanex16_b32_i_tid_fi_f64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s0, s1 op_sel:[1,0]
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlanex16_b32_i_tid_fi_f64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s0, s1 op_sel:[1,0]
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlanex16_b32_i_tid_fi_f64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s0, s1 op_sel:[1,0]
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlanex16_b32_i_tid_fi_f64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,0]
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s0, s1 op_sel:[1,0]
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_f32 = bitcast i32 %tidx to float
  %tidx_f64 = fpext float %tidx_f32 to double
  %undef = freeze double poison
  %v = call double @llvm.amdgcn.permlanex16.f64(double %undef, double %tidx_f64, i32 %src1, i32 %src2, i1 true, i1 false)
  store double %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_i_tid_bc_i32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlanex16_b32_i_tid_bc_i32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX10-NEXT:    global_store_dword v1, v0, s[2:3]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlanex16_b32_i_tid_bc_i32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX11-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlanex16_b32_i_tid_bc_i32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX12-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX12-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %undef = freeze i32 poison
  %v = call i32 @llvm.amdgcn.permlanex16.i32(i32 %undef, i32 %tidx, i32 %src1, i32 %src2, i1 false, i1 true)
  store i32 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_i_tid_bc_f32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlanex16_b32_i_tid_bc_f32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX10-NEXT:    global_store_dword v1, v0, s[2:3]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlanex16_b32_i_tid_bc_f32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX11-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlanex16_b32_i_tid_bc_f32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX12-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX12-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_f32 = bitcast i32 %tidx to float
  %undef = freeze float poison
  %v = call float @llvm.amdgcn.permlanex16.f32(float %undef, float %tidx_f32, i32 %src1, i32 %src2, i1 false, i1 true)
  store float %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_i_tid_bc_i64(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_i_tid_bc_i64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v1, v2, s0, s1 op_sel:[0,1]
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_i_tid_bc_i64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v1, v2, s0, s1 op_sel:[0,1]
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlanex16_b32_i_tid_bc_i64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v1, v2, s0, s1 op_sel:[0,1]
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlanex16_b32_i_tid_bc_i64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v1, v2, s0, s1 op_sel:[0,1]
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlanex16_b32_i_tid_bc_i64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v1, v2, s0, s1 op_sel:[0,1]
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlanex16_b32_i_tid_bc_i64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v1, v2, s0, s1 op_sel:[0,1]
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_i64 = zext i32 %tidx to i64
  %undef = freeze i64 poison
  %v = call i64 @llvm.amdgcn.permlanex16.i64(i64 %undef, i64 %tidx_i64, i32 %src1, i32 %src2, i1 false, i1 true)
  store i64 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_i_tid_bc_f64(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_i_tid_bc_f64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s0, s1 op_sel:[0,1]
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_i_tid_bc_f64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s0, s1 op_sel:[0,1]
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlanex16_b32_i_tid_bc_f64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s0, s1 op_sel:[0,1]
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlanex16_b32_i_tid_bc_f64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s0, s1 op_sel:[0,1]
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlanex16_b32_i_tid_bc_f64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s0, s1 op_sel:[0,1]
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlanex16_b32_i_tid_bc_f64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[0,1]
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s0, s1 op_sel:[0,1]
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_f32 = bitcast i32 %tidx to float
  %tidx_f64 = fpext float %tidx_f32 to double
  %undef = freeze double poison
  %v = call double @llvm.amdgcn.permlanex16.f64(double %undef, double %tidx_f64, i32 %src1, i32 %src2, i1 false, i1 true)
  store double %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_i_tid_fi_bc_i32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlanex16_b32_i_tid_fi_bc_i32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX10-NEXT:    global_store_dword v1, v0, s[2:3]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlanex16_b32_i_tid_fi_bc_i32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX11-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlanex16_b32_i_tid_fi_bc_i32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX12-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX12-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %undef = freeze i32 poison
  %v = call i32 @llvm.amdgcn.permlanex16.i32(i32 %undef, i32 %tidx, i32 %src1, i32 %src2, i1 true, i1 true)
  store i32 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_i_tid_fi_bc_f32(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlanex16_b32_i_tid_fi_bc_f32:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-NEXT:    v_mov_b32_e32 v1, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX10-NEXT:    global_store_dword v1, v0, s[2:3]
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlanex16_b32_i_tid_fi_bc_f32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX11-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlanex16_b32_i_tid_fi_bc_f32:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_clause 0x1
; GFX12-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX12-NEXT:    global_store_b32 v1, v0, s[2:3]
; GFX12-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_f32 = bitcast i32 %tidx to float
  %undef = freeze float poison
  %v = call float @llvm.amdgcn.permlanex16.f32(float %undef, float %tidx_f32, i32 %src1, i32 %src2, i1 true, i1 true)
  store float %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_i_tid_fi_bc_i64(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_i_tid_fi_bc_i64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v1, v2, s0, s1 op_sel:[1,1]
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_i_tid_fi_bc_i64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v1, v2, s0, s1 op_sel:[1,1]
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlanex16_b32_i_tid_fi_bc_i64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v1, v2, s0, s1 op_sel:[1,1]
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlanex16_b32_i_tid_fi_bc_i64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v1, v2, s0, s1 op_sel:[1,1]
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlanex16_b32_i_tid_fi_bc_i64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v1, v2, s0, s1 op_sel:[1,1]
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlanex16_b32_i_tid_fi_bc_i64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v1, v2, s0, s1 op_sel:[1,1]
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_i64 = zext i32 %tidx to i64
  %undef = freeze i64 poison
  %v = call i64 @llvm.amdgcn.permlanex16.i64(i64 %undef, i64 %tidx_i64, i32 %src1, i32 %src2, i1 true, i1 true)
  store i64 %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_b32_i_tid_fi_bc_f64(ptr addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlanex16_b32_i_tid_fi_bc_f64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX10-SDAG-NEXT:    s_clause 0x1
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-SDAG-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s0, s1 op_sel:[1,1]
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX10-SDAG-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-SDAG-NEXT:    s_endpgm
;
; GFX10-GISEL-LABEL: v_permlanex16_b32_i_tid_fi_bc_f64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX10-GISEL-NEXT:    s_clause 0x1
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x30
; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x24
; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s0, s1 op_sel:[1,1]
; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
; GFX10-GISEL-NEXT:    s_endpgm
;
; GFX11-SDAG-LABEL: v_permlanex16_b32_i_tid_fi_bc_f64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s0, s1 op_sel:[1,1]
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX11-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-SDAG-NEXT:    s_endpgm
;
; GFX11-GISEL-LABEL: v_permlanex16_b32_i_tid_fi_bc_f64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX11-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX11-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX11-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s0, s1 op_sel:[1,1]
; GFX11-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-GISEL-NEXT:    s_endpgm
;
; GFX12-SDAG-LABEL: v_permlanex16_b32_i_tid_fi_bc_f64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-SDAG-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-SDAG-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-SDAG-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v1, v1, s0, s1 op_sel:[1,1]
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX12-SDAG-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-SDAG-NEXT:    s_endpgm
;
; GFX12-GISEL-LABEL: v_permlanex16_b32_i_tid_fi_bc_f64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    s_load_b64 s[0:1], s[4:5], 0x30
; GFX12-GISEL-NEXT:    s_load_b64 s[2:3], s[4:5], 0x24
; GFX12-GISEL-NEXT:    v_mov_b32_e32 v2, 0
; GFX12-GISEL-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v0, v0, s0, s1 op_sel:[1,1]
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v1, v1, s0, s1 op_sel:[1,1]
; GFX12-GISEL-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX12-GISEL-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %tidx_f32 = bitcast i32 %tidx to float
  %tidx_f64 = fpext float %tidx_f32 to double
  %undef = freeze double poison
  %v = call double @llvm.amdgcn.permlanex16.f64(double %undef, double %tidx_f64, i32 %src1, i32 %src2, i1 true, i1 true)
  store double %v, ptr addrspace(1) %out
  ret void
}

define void @v_permlane16_half(ptr addrspace(1) %out, half %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlane16_half:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT:    v_readfirstlane_b32 s4, v3
; GFX10-NEXT:    v_readfirstlane_b32 s5, v4
; GFX10-NEXT:    v_permlane16_b32 v2, v2, s4, s5
; GFX10-NEXT:    global_store_short v[0:1], v2, off
; GFX10-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-LABEL: v_permlane16_half:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT:    v_readfirstlane_b32 s0, v3
; GFX11-NEXT:    v_readfirstlane_b32 s1, v4
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX11-NEXT:    global_store_b16 v[0:1], v2, off
; GFX11-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-LABEL: v_permlane16_half:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT:    s_wait_expcnt 0x0
; GFX12-NEXT:    s_wait_samplecnt 0x0
; GFX12-NEXT:    s_wait_bvhcnt 0x0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_readfirstlane_b32 s0, v3
; GFX12-NEXT:    v_readfirstlane_b32 s1, v4
; GFX12-NEXT:    s_wait_alu 0xf1ff
; GFX12-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX12-NEXT:    global_store_b16 v[0:1], v2, off
; GFX12-NEXT:    s_setpc_b64 s[30:31]
  %v = call half @llvm.amdgcn.permlane16.f16(half %src0, half %src0, i32 %src1, i32 %src2, i1 false, i1 false)
  store half %v, ptr addrspace(1) %out
  ret void
}

define void @v_permlanex16_half(ptr addrspace(1) %out, half %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlanex16_half:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT:    v_readfirstlane_b32 s4, v3
; GFX10-NEXT:    v_readfirstlane_b32 s5, v4
; GFX10-NEXT:    v_permlanex16_b32 v2, v2, s4, s5
; GFX10-NEXT:    global_store_short v[0:1], v2, off
; GFX10-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-LABEL: v_permlanex16_half:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT:    v_readfirstlane_b32 s0, v3
; GFX11-NEXT:    v_readfirstlane_b32 s1, v4
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX11-NEXT:    global_store_b16 v[0:1], v2, off
; GFX11-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-LABEL: v_permlanex16_half:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT:    s_wait_expcnt 0x0
; GFX12-NEXT:    s_wait_samplecnt 0x0
; GFX12-NEXT:    s_wait_bvhcnt 0x0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_readfirstlane_b32 s0, v3
; GFX12-NEXT:    v_readfirstlane_b32 s1, v4
; GFX12-NEXT:    s_wait_alu 0xf1ff
; GFX12-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX12-NEXT:    global_store_b16 v[0:1], v2, off
; GFX12-NEXT:    s_setpc_b64 s[30:31]
  %v = call half @llvm.amdgcn.permlanex16.f16(half %src0, half %src0, i32 %src1, i32 %src2, i1 false, i1 false)
  store half %v, ptr addrspace(1) %out
  ret void
}

define void @v_permlane16_bfloat(ptr addrspace(1) %out, bfloat %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlane16_bfloat:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT:    v_readfirstlane_b32 s4, v3
; GFX10-NEXT:    v_readfirstlane_b32 s5, v4
; GFX10-NEXT:    v_permlane16_b32 v2, v2, s4, s5
; GFX10-NEXT:    global_store_short v[0:1], v2, off
; GFX10-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-LABEL: v_permlane16_bfloat:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT:    v_readfirstlane_b32 s0, v3
; GFX11-NEXT:    v_readfirstlane_b32 s1, v4
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX11-NEXT:    global_store_b16 v[0:1], v2, off
; GFX11-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-LABEL: v_permlane16_bfloat:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT:    s_wait_expcnt 0x0
; GFX12-NEXT:    s_wait_samplecnt 0x0
; GFX12-NEXT:    s_wait_bvhcnt 0x0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_readfirstlane_b32 s0, v3
; GFX12-NEXT:    v_readfirstlane_b32 s1, v4
; GFX12-NEXT:    s_wait_alu 0xf1ff
; GFX12-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX12-NEXT:    global_store_b16 v[0:1], v2, off
; GFX12-NEXT:    s_setpc_b64 s[30:31]
  %v = call bfloat @llvm.amdgcn.permlane16.f16(bfloat %src0, bfloat %src0, i32 %src1, i32 %src2, i1 false, i1 false)
  store bfloat %v, ptr addrspace(1) %out
  ret void
}

define void @v_permlanex16_bfloat(ptr addrspace(1) %out, bfloat %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlanex16_bfloat:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT:    v_readfirstlane_b32 s4, v3
; GFX10-NEXT:    v_readfirstlane_b32 s5, v4
; GFX10-NEXT:    v_permlanex16_b32 v2, v2, s4, s5
; GFX10-NEXT:    global_store_short v[0:1], v2, off
; GFX10-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-LABEL: v_permlanex16_bfloat:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT:    v_readfirstlane_b32 s0, v3
; GFX11-NEXT:    v_readfirstlane_b32 s1, v4
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX11-NEXT:    global_store_b16 v[0:1], v2, off
; GFX11-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-LABEL: v_permlanex16_bfloat:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT:    s_wait_expcnt 0x0
; GFX12-NEXT:    s_wait_samplecnt 0x0
; GFX12-NEXT:    s_wait_bvhcnt 0x0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_readfirstlane_b32 s0, v3
; GFX12-NEXT:    v_readfirstlane_b32 s1, v4
; GFX12-NEXT:    s_wait_alu 0xf1ff
; GFX12-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX12-NEXT:    global_store_b16 v[0:1], v2, off
; GFX12-NEXT:    s_setpc_b64 s[30:31]
  %v = call bfloat @llvm.amdgcn.permlanex16.f16(bfloat %src0, bfloat %src0, i32 %src1, i32 %src2, i1 false, i1 false)
  store bfloat %v, ptr addrspace(1) %out
  ret void
}

define void @v_permlane16_i16(ptr addrspace(1) %out, i16 %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlane16_i16:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT:    v_readfirstlane_b32 s4, v3
; GFX10-NEXT:    v_readfirstlane_b32 s5, v4
; GFX10-NEXT:    v_permlane16_b32 v2, v2, s4, s5
; GFX10-NEXT:    global_store_short v[0:1], v2, off
; GFX10-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-LABEL: v_permlane16_i16:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT:    v_readfirstlane_b32 s0, v3
; GFX11-NEXT:    v_readfirstlane_b32 s1, v4
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX11-NEXT:    global_store_b16 v[0:1], v2, off
; GFX11-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-LABEL: v_permlane16_i16:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT:    s_wait_expcnt 0x0
; GFX12-NEXT:    s_wait_samplecnt 0x0
; GFX12-NEXT:    s_wait_bvhcnt 0x0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_readfirstlane_b32 s0, v3
; GFX12-NEXT:    v_readfirstlane_b32 s1, v4
; GFX12-NEXT:    s_wait_alu 0xf1ff
; GFX12-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX12-NEXT:    global_store_b16 v[0:1], v2, off
; GFX12-NEXT:    s_setpc_b64 s[30:31]
  %v = call i16 @llvm.amdgcn.permlane16.i16(i16 %src0, i16 %src0, i32 %src1, i32 %src2, i1 false, i1 false)
  store i16 %v, ptr addrspace(1) %out
  ret void
}

define void @v_permlanex16_i16(ptr addrspace(1) %out, i16 %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlanex16_i16:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT:    v_readfirstlane_b32 s4, v3
; GFX10-NEXT:    v_readfirstlane_b32 s5, v4
; GFX10-NEXT:    v_permlanex16_b32 v2, v2, s4, s5
; GFX10-NEXT:    global_store_short v[0:1], v2, off
; GFX10-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-LABEL: v_permlanex16_i16:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT:    v_readfirstlane_b32 s0, v3
; GFX11-NEXT:    v_readfirstlane_b32 s1, v4
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX11-NEXT:    global_store_b16 v[0:1], v2, off
; GFX11-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-LABEL: v_permlanex16_i16:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT:    s_wait_expcnt 0x0
; GFX12-NEXT:    s_wait_samplecnt 0x0
; GFX12-NEXT:    s_wait_bvhcnt 0x0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_readfirstlane_b32 s0, v3
; GFX12-NEXT:    v_readfirstlane_b32 s1, v4
; GFX12-NEXT:    s_wait_alu 0xf1ff
; GFX12-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX12-NEXT:    global_store_b16 v[0:1], v2, off
; GFX12-NEXT:    s_setpc_b64 s[30:31]
  %v = call i16 @llvm.amdgcn.permlanex16.i16(i16 %src0, i16 %src0, i32 %src1, i32 %src2, i1 false, i1 false)
  store i16 %v, ptr addrspace(1) %out
  ret void
}

define void @v_permlane16_v2f16(ptr addrspace(1) %out, <2 x half> %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlane16_v2f16:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT:    v_readfirstlane_b32 s4, v3
; GFX10-NEXT:    v_readfirstlane_b32 s5, v4
; GFX10-NEXT:    v_permlane16_b32 v2, v2, s4, s5
; GFX10-NEXT:    global_store_dword v[0:1], v2, off
; GFX10-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-LABEL: v_permlane16_v2f16:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT:    v_readfirstlane_b32 s0, v3
; GFX11-NEXT:    v_readfirstlane_b32 s1, v4
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX11-NEXT:    global_store_b32 v[0:1], v2, off
; GFX11-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-LABEL: v_permlane16_v2f16:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT:    s_wait_expcnt 0x0
; GFX12-NEXT:    s_wait_samplecnt 0x0
; GFX12-NEXT:    s_wait_bvhcnt 0x0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_readfirstlane_b32 s0, v3
; GFX12-NEXT:    v_readfirstlane_b32 s1, v4
; GFX12-NEXT:    s_wait_alu 0xf1ff
; GFX12-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX12-NEXT:    global_store_b32 v[0:1], v2, off
; GFX12-NEXT:    s_setpc_b64 s[30:31]
  %v = call <2 x half> @llvm.amdgcn.permlane16.v2f16(<2 x half> %src0, <2 x half> %src0, i32 %src1, i32 %src2, i1 false, i1 false)
  store <2 x half> %v, ptr addrspace(1) %out
  ret void
}

define void @v_permlanex16_v2f16(ptr addrspace(1) %out, <2 x half> %src0, i32 %src1, i32 %src2) {
; GFX10-LABEL: v_permlanex16_v2f16:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT:    v_readfirstlane_b32 s4, v3
; GFX10-NEXT:    v_readfirstlane_b32 s5, v4
; GFX10-NEXT:    v_permlanex16_b32 v2, v2, s4, s5
; GFX10-NEXT:    global_store_dword v[0:1], v2, off
; GFX10-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-LABEL: v_permlanex16_v2f16:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT:    v_readfirstlane_b32 s0, v3
; GFX11-NEXT:    v_readfirstlane_b32 s1, v4
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX11-NEXT:    global_store_b32 v[0:1], v2, off
; GFX11-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-LABEL: v_permlanex16_v2f16:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT:    s_wait_expcnt 0x0
; GFX12-NEXT:    s_wait_samplecnt 0x0
; GFX12-NEXT:    s_wait_bvhcnt 0x0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_readfirstlane_b32 s0, v3
; GFX12-NEXT:    v_readfirstlane_b32 s1, v4
; GFX12-NEXT:    s_wait_alu 0xf1ff
; GFX12-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX12-NEXT:    global_store_b32 v[0:1], v2, off
; GFX12-NEXT:    s_setpc_b64 s[30:31]
  %v = call <2 x half> @llvm.amdgcn.permlanex16.v2f16(<2 x half> %src0, <2 x half> %src0, i32 %src1, i32 %src2, i1 false, i1 false)
  store <2 x half> %v, ptr addrspace(1) %out
  ret void
}

define void @v_permlane16_v2f32(ptr addrspace(1) %out, <2 x float> %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlane16_v2f32:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s4, v4
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s5, v5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v3, v3, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v2, v2, s4, s5
; GFX10-SDAG-NEXT:    global_store_dwordx2 v[0:1], v[2:3], off
; GFX10-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX10-GISEL-LABEL: v_permlane16_v2f32:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v4
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s5, v5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v2, v2, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v3, v3, s4, s5
; GFX10-GISEL-NEXT:    global_store_dwordx2 v[0:1], v[2:3], off
; GFX10-GISEL-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-SDAG-LABEL: v_permlane16_v2f32:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s0, v4
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s1, v5
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v3, v3, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX11-SDAG-NEXT:    global_store_b64 v[0:1], v[2:3], off
; GFX11-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-GISEL-LABEL: v_permlane16_v2f32:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s0, v4
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s1, v5
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v3, v3, s0, s1
; GFX11-GISEL-NEXT:    global_store_b64 v[0:1], v[2:3], off
; GFX11-GISEL-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-SDAG-LABEL: v_permlane16_v2f32:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-SDAG-NEXT:    s_wait_expcnt 0x0
; GFX12-SDAG-NEXT:    s_wait_samplecnt 0x0
; GFX12-SDAG-NEXT:    s_wait_bvhcnt 0x0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s0, v4
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s1, v5
; GFX12-SDAG-NEXT:    s_wait_alu 0xf1ff
; GFX12-SDAG-NEXT:    v_permlane16_b32 v3, v3, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX12-SDAG-NEXT:    global_store_b64 v[0:1], v[2:3], off
; GFX12-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-GISEL-LABEL: v_permlane16_v2f32:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-GISEL-NEXT:    s_wait_expcnt 0x0
; GFX12-GISEL-NEXT:    s_wait_samplecnt 0x0
; GFX12-GISEL-NEXT:    s_wait_bvhcnt 0x0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s0, v4
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s1, v5
; GFX12-GISEL-NEXT:    s_wait_alu 0xf1ff
; GFX12-GISEL-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v3, v3, s0, s1
; GFX12-GISEL-NEXT:    global_store_b64 v[0:1], v[2:3], off
; GFX12-GISEL-NEXT:    s_setpc_b64 s[30:31]
  %v = call <2 x float> @llvm.amdgcn.permlane16.v2f32(<2 x float> %src0, <2 x float> %src0, i32 %src1, i32 %src2, i1 false, i1 false)
  store <2 x float> %v, ptr addrspace(1) %out
  ret void
}

define void @v_permlanex16_v2f32(ptr addrspace(1) %out, <2 x float> %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlanex16_v2f32:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s4, v4
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s5, v5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v3, v3, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v2, v2, s4, s5
; GFX10-SDAG-NEXT:    global_store_dwordx2 v[0:1], v[2:3], off
; GFX10-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX10-GISEL-LABEL: v_permlanex16_v2f32:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v4
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s5, v5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v2, v2, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v3, v3, s4, s5
; GFX10-GISEL-NEXT:    global_store_dwordx2 v[0:1], v[2:3], off
; GFX10-GISEL-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-SDAG-LABEL: v_permlanex16_v2f32:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s0, v4
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s1, v5
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v3, v3, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX11-SDAG-NEXT:    global_store_b64 v[0:1], v[2:3], off
; GFX11-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-GISEL-LABEL: v_permlanex16_v2f32:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s0, v4
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s1, v5
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v3, v3, s0, s1
; GFX11-GISEL-NEXT:    global_store_b64 v[0:1], v[2:3], off
; GFX11-GISEL-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-SDAG-LABEL: v_permlanex16_v2f32:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-SDAG-NEXT:    s_wait_expcnt 0x0
; GFX12-SDAG-NEXT:    s_wait_samplecnt 0x0
; GFX12-SDAG-NEXT:    s_wait_bvhcnt 0x0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s0, v4
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s1, v5
; GFX12-SDAG-NEXT:    s_wait_alu 0xf1ff
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v3, v3, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX12-SDAG-NEXT:    global_store_b64 v[0:1], v[2:3], off
; GFX12-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-GISEL-LABEL: v_permlanex16_v2f32:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-GISEL-NEXT:    s_wait_expcnt 0x0
; GFX12-GISEL-NEXT:    s_wait_samplecnt 0x0
; GFX12-GISEL-NEXT:    s_wait_bvhcnt 0x0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s0, v4
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s1, v5
; GFX12-GISEL-NEXT:    s_wait_alu 0xf1ff
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v3, v3, s0, s1
; GFX12-GISEL-NEXT:    global_store_b64 v[0:1], v[2:3], off
; GFX12-GISEL-NEXT:    s_setpc_b64 s[30:31]
  %v = call <2 x float> @llvm.amdgcn.permlanex16.v2f32(<2 x float> %src0, <2 x float> %src0, i32 %src1, i32 %src2, i1 false, i1 false)
  store <2 x float> %v, ptr addrspace(1) %out
  ret void
}

define void @v_permlane16_v7i32(ptr addrspace(1) %out, <7 x i32> %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlane16_v7i32:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s4, v9
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s5, v10
; GFX10-SDAG-NEXT:    v_permlane16_b32 v8, v8, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v7, v7, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v6, v6, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v5, v5, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v4, v4, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v3, v3, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v2, v2, s4, s5
; GFX10-SDAG-NEXT:    global_store_dwordx3 v[0:1], v[6:8], off offset:16
; GFX10-SDAG-NEXT:    global_store_dwordx4 v[0:1], v[2:5], off
; GFX10-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX10-GISEL-LABEL: v_permlane16_v7i32:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v9
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s5, v10
; GFX10-GISEL-NEXT:    v_permlane16_b32 v2, v2, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v3, v3, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v4, v4, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v5, v5, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v6, v6, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v7, v7, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v8, v8, s4, s5
; GFX10-GISEL-NEXT:    global_store_dwordx4 v[0:1], v[2:5], off
; GFX10-GISEL-NEXT:    global_store_dwordx3 v[0:1], v[6:8], off offset:16
; GFX10-GISEL-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-SDAG-LABEL: v_permlane16_v7i32:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s0, v9
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s1, v10
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v8, v8, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v7, v7, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v6, v6, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v5, v5, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v4, v4, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v3, v3, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    global_store_b96 v[0:1], v[6:8], off offset:16
; GFX11-SDAG-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX11-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-GISEL-LABEL: v_permlane16_v7i32:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s0, v9
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s1, v10
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v3, v3, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v4, v4, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v5, v5, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v6, v6, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v7, v7, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v8, v8, s0, s1
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX11-GISEL-NEXT:    global_store_b96 v[0:1], v[6:8], off offset:16
; GFX11-GISEL-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-SDAG-LABEL: v_permlane16_v7i32:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-SDAG-NEXT:    s_wait_expcnt 0x0
; GFX12-SDAG-NEXT:    s_wait_samplecnt 0x0
; GFX12-SDAG-NEXT:    s_wait_bvhcnt 0x0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s0, v9
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s1, v10
; GFX12-SDAG-NEXT:    s_wait_alu 0xf1ff
; GFX12-SDAG-NEXT:    v_permlane16_b32 v8, v8, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v7, v7, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v6, v6, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v5, v5, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v4, v4, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v3, v3, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    global_store_b96 v[0:1], v[6:8], off offset:16
; GFX12-SDAG-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX12-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-GISEL-LABEL: v_permlane16_v7i32:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-GISEL-NEXT:    s_wait_expcnt 0x0
; GFX12-GISEL-NEXT:    s_wait_samplecnt 0x0
; GFX12-GISEL-NEXT:    s_wait_bvhcnt 0x0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s0, v9
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s1, v10
; GFX12-GISEL-NEXT:    s_wait_alu 0xf1ff
; GFX12-GISEL-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v3, v3, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v4, v4, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v5, v5, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v6, v6, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v7, v7, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v8, v8, s0, s1
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX12-GISEL-NEXT:    global_store_b96 v[0:1], v[6:8], off offset:16
; GFX12-GISEL-NEXT:    s_setpc_b64 s[30:31]
  %v = call <7 x i32> @llvm.amdgcn.permlane16.v7i32(<7 x i32> %src0, <7 x i32> %src0, i32 %src1, i32 %src2, i1 false, i1 false)
  store <7 x i32> %v, ptr addrspace(1) %out
  ret void
}

define void @v_permlanex16_v7i32(ptr addrspace(1) %out, <7 x i32> %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlanex16_v7i32:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s4, v9
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s5, v10
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v8, v8, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v7, v7, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v6, v6, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v5, v5, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v4, v4, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v3, v3, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v2, v2, s4, s5
; GFX10-SDAG-NEXT:    global_store_dwordx3 v[0:1], v[6:8], off offset:16
; GFX10-SDAG-NEXT:    global_store_dwordx4 v[0:1], v[2:5], off
; GFX10-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX10-GISEL-LABEL: v_permlanex16_v7i32:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v9
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s5, v10
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v2, v2, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v3, v3, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v4, v4, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v5, v5, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v6, v6, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v7, v7, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v8, v8, s4, s5
; GFX10-GISEL-NEXT:    global_store_dwordx4 v[0:1], v[2:5], off
; GFX10-GISEL-NEXT:    global_store_dwordx3 v[0:1], v[6:8], off offset:16
; GFX10-GISEL-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-SDAG-LABEL: v_permlanex16_v7i32:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s0, v9
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s1, v10
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v8, v8, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v7, v7, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v6, v6, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v5, v5, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v4, v4, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v3, v3, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    global_store_b96 v[0:1], v[6:8], off offset:16
; GFX11-SDAG-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX11-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-GISEL-LABEL: v_permlanex16_v7i32:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s0, v9
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s1, v10
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v3, v3, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v4, v4, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v5, v5, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v6, v6, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v7, v7, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v8, v8, s0, s1
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX11-GISEL-NEXT:    global_store_b96 v[0:1], v[6:8], off offset:16
; GFX11-GISEL-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-SDAG-LABEL: v_permlanex16_v7i32:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-SDAG-NEXT:    s_wait_expcnt 0x0
; GFX12-SDAG-NEXT:    s_wait_samplecnt 0x0
; GFX12-SDAG-NEXT:    s_wait_bvhcnt 0x0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s0, v9
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s1, v10
; GFX12-SDAG-NEXT:    s_wait_alu 0xf1ff
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v8, v8, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v7, v7, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v6, v6, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v5, v5, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v4, v4, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v3, v3, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    global_store_b96 v[0:1], v[6:8], off offset:16
; GFX12-SDAG-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX12-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-GISEL-LABEL: v_permlanex16_v7i32:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-GISEL-NEXT:    s_wait_expcnt 0x0
; GFX12-GISEL-NEXT:    s_wait_samplecnt 0x0
; GFX12-GISEL-NEXT:    s_wait_bvhcnt 0x0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s0, v9
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s1, v10
; GFX12-GISEL-NEXT:    s_wait_alu 0xf1ff
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v3, v3, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v4, v4, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v5, v5, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v6, v6, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v7, v7, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v8, v8, s0, s1
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX12-GISEL-NEXT:    global_store_b96 v[0:1], v[6:8], off offset:16
; GFX12-GISEL-NEXT:    s_setpc_b64 s[30:31]
  %v = call <7 x i32> @llvm.amdgcn.permlanex16.v7i32(<7 x i32> %src0, <7 x i32> %src0, i32 %src1, i32 %src2, i1 false, i1 false)
  store <7 x i32> %v, ptr addrspace(1) %out
  ret void
}

define void @v_permlane16_v8i16(ptr addrspace(1) %out, <8 x i16> %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlane16_v8i16:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s4, v6
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s5, v7
; GFX10-SDAG-NEXT:    v_permlane16_b32 v5, v5, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v4, v4, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v3, v3, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v2, v2, s4, s5
; GFX10-SDAG-NEXT:    global_store_dwordx4 v[0:1], v[2:5], off
; GFX10-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX10-GISEL-LABEL: v_permlane16_v8i16:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v6
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s5, v7
; GFX10-GISEL-NEXT:    v_permlane16_b32 v2, v2, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v3, v3, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v4, v4, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v5, v5, s4, s5
; GFX10-GISEL-NEXT:    global_store_dwordx4 v[0:1], v[2:5], off
; GFX10-GISEL-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-SDAG-LABEL: v_permlane16_v8i16:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s0, v6
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s1, v7
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v5, v5, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v4, v4, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v3, v3, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX11-SDAG-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX11-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-GISEL-LABEL: v_permlane16_v8i16:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s0, v6
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s1, v7
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v3, v3, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v4, v4, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v5, v5, s0, s1
; GFX11-GISEL-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX11-GISEL-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-SDAG-LABEL: v_permlane16_v8i16:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-SDAG-NEXT:    s_wait_expcnt 0x0
; GFX12-SDAG-NEXT:    s_wait_samplecnt 0x0
; GFX12-SDAG-NEXT:    s_wait_bvhcnt 0x0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s0, v6
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s1, v7
; GFX12-SDAG-NEXT:    s_wait_alu 0xf1ff
; GFX12-SDAG-NEXT:    v_permlane16_b32 v5, v5, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v4, v4, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v3, v3, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX12-SDAG-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX12-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-GISEL-LABEL: v_permlane16_v8i16:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-GISEL-NEXT:    s_wait_expcnt 0x0
; GFX12-GISEL-NEXT:    s_wait_samplecnt 0x0
; GFX12-GISEL-NEXT:    s_wait_bvhcnt 0x0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s0, v6
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s1, v7
; GFX12-GISEL-NEXT:    s_wait_alu 0xf1ff
; GFX12-GISEL-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v3, v3, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v4, v4, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v5, v5, s0, s1
; GFX12-GISEL-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX12-GISEL-NEXT:    s_setpc_b64 s[30:31]
  %v = call <8 x i16> @llvm.amdgcn.permlane16.v8i16(<8 x i16> %src0, <8 x i16> %src0, i32 %src1, i32 %src2, i1 false, i1 false)
  store <8 x i16> %v, ptr addrspace(1) %out
  ret void
}

define void @v_permlanex16_v8i16(ptr addrspace(1) %out, <8 x i16> %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlanex16_v8i16:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s4, v6
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s5, v7
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v5, v5, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v4, v4, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v3, v3, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v2, v2, s4, s5
; GFX10-SDAG-NEXT:    global_store_dwordx4 v[0:1], v[2:5], off
; GFX10-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX10-GISEL-LABEL: v_permlanex16_v8i16:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v6
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s5, v7
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v2, v2, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v3, v3, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v4, v4, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v5, v5, s4, s5
; GFX10-GISEL-NEXT:    global_store_dwordx4 v[0:1], v[2:5], off
; GFX10-GISEL-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-SDAG-LABEL: v_permlanex16_v8i16:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s0, v6
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s1, v7
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v5, v5, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v4, v4, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v3, v3, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX11-SDAG-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX11-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-GISEL-LABEL: v_permlanex16_v8i16:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s0, v6
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s1, v7
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v3, v3, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v4, v4, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v5, v5, s0, s1
; GFX11-GISEL-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX11-GISEL-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-SDAG-LABEL: v_permlanex16_v8i16:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-SDAG-NEXT:    s_wait_expcnt 0x0
; GFX12-SDAG-NEXT:    s_wait_samplecnt 0x0
; GFX12-SDAG-NEXT:    s_wait_bvhcnt 0x0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s0, v6
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s1, v7
; GFX12-SDAG-NEXT:    s_wait_alu 0xf1ff
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v5, v5, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v4, v4, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v3, v3, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX12-SDAG-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX12-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-GISEL-LABEL: v_permlanex16_v8i16:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-GISEL-NEXT:    s_wait_expcnt 0x0
; GFX12-GISEL-NEXT:    s_wait_samplecnt 0x0
; GFX12-GISEL-NEXT:    s_wait_bvhcnt 0x0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s0, v6
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s1, v7
; GFX12-GISEL-NEXT:    s_wait_alu 0xf1ff
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v3, v3, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v4, v4, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v5, v5, s0, s1
; GFX12-GISEL-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX12-GISEL-NEXT:    s_setpc_b64 s[30:31]
  %v = call <8 x i16> @llvm.amdgcn.permlanex16.v8i16(<8 x i16> %src0, <8 x i16> %src0, i32 %src1, i32 %src2, i1 false, i1 false)
  store <8 x i16> %v, ptr addrspace(1) %out
  ret void
}

define void @v_permlane16_v2i64(ptr addrspace(1) %out, <2 x i64> %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlane16_v2i64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s4, v6
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s5, v7
; GFX10-SDAG-NEXT:    v_permlane16_b32 v5, v5, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v4, v4, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v3, v3, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v2, v2, s4, s5
; GFX10-SDAG-NEXT:    global_store_dwordx4 v[0:1], v[2:5], off
; GFX10-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX10-GISEL-LABEL: v_permlane16_v2i64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v6
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s5, v7
; GFX10-GISEL-NEXT:    v_permlane16_b32 v2, v2, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v3, v3, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v4, v4, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v5, v5, s4, s5
; GFX10-GISEL-NEXT:    global_store_dwordx4 v[0:1], v[2:5], off
; GFX10-GISEL-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-SDAG-LABEL: v_permlane16_v2i64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s0, v6
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s1, v7
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v5, v5, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v4, v4, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v3, v3, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX11-SDAG-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX11-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-GISEL-LABEL: v_permlane16_v2i64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s0, v6
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s1, v7
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v3, v3, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v4, v4, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v5, v5, s0, s1
; GFX11-GISEL-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX11-GISEL-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-SDAG-LABEL: v_permlane16_v2i64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-SDAG-NEXT:    s_wait_expcnt 0x0
; GFX12-SDAG-NEXT:    s_wait_samplecnt 0x0
; GFX12-SDAG-NEXT:    s_wait_bvhcnt 0x0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s0, v6
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s1, v7
; GFX12-SDAG-NEXT:    s_wait_alu 0xf1ff
; GFX12-SDAG-NEXT:    v_permlane16_b32 v5, v5, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v4, v4, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v3, v3, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX12-SDAG-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX12-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-GISEL-LABEL: v_permlane16_v2i64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-GISEL-NEXT:    s_wait_expcnt 0x0
; GFX12-GISEL-NEXT:    s_wait_samplecnt 0x0
; GFX12-GISEL-NEXT:    s_wait_bvhcnt 0x0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s0, v6
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s1, v7
; GFX12-GISEL-NEXT:    s_wait_alu 0xf1ff
; GFX12-GISEL-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v3, v3, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v4, v4, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v5, v5, s0, s1
; GFX12-GISEL-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX12-GISEL-NEXT:    s_setpc_b64 s[30:31]
  %v = call <2 x i64> @llvm.amdgcn.permlane16.v2i64(<2 x i64> %src0, <2 x i64> %src0, i32 %src1, i32 %src2, i1 false, i1 false)
  store <2 x i64> %v, ptr addrspace(1) %out
  ret void
}

define void @v_permlane16_v3i64(ptr addrspace(1) %out, <3 x i64> %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlane16_v3i64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s4, v8
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s5, v9
; GFX10-SDAG-NEXT:    v_permlane16_b32 v7, v7, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v6, v6, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v5, v5, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v4, v4, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v3, v3, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v2, v2, s4, s5
; GFX10-SDAG-NEXT:    global_store_dwordx2 v[0:1], v[6:7], off offset:16
; GFX10-SDAG-NEXT:    global_store_dwordx4 v[0:1], v[2:5], off
; GFX10-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX10-GISEL-LABEL: v_permlane16_v3i64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v8
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s5, v9
; GFX10-GISEL-NEXT:    v_permlane16_b32 v2, v2, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v3, v3, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v4, v4, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v5, v5, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v6, v6, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v7, v7, s4, s5
; GFX10-GISEL-NEXT:    global_store_dwordx4 v[0:1], v[2:5], off
; GFX10-GISEL-NEXT:    global_store_dwordx2 v[0:1], v[6:7], off offset:16
; GFX10-GISEL-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-SDAG-LABEL: v_permlane16_v3i64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s0, v8
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s1, v9
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v7, v7, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v6, v6, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v5, v5, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v4, v4, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v3, v3, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    global_store_b64 v[0:1], v[6:7], off offset:16
; GFX11-SDAG-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX11-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-GISEL-LABEL: v_permlane16_v3i64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s0, v8
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s1, v9
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v3, v3, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v4, v4, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v5, v5, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v6, v6, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v7, v7, s0, s1
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX11-GISEL-NEXT:    global_store_b64 v[0:1], v[6:7], off offset:16
; GFX11-GISEL-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-SDAG-LABEL: v_permlane16_v3i64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-SDAG-NEXT:    s_wait_expcnt 0x0
; GFX12-SDAG-NEXT:    s_wait_samplecnt 0x0
; GFX12-SDAG-NEXT:    s_wait_bvhcnt 0x0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s0, v8
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s1, v9
; GFX12-SDAG-NEXT:    s_wait_alu 0xf1ff
; GFX12-SDAG-NEXT:    v_permlane16_b32 v7, v7, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v6, v6, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v5, v5, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v4, v4, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v3, v3, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    global_store_b64 v[0:1], v[6:7], off offset:16
; GFX12-SDAG-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX12-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-GISEL-LABEL: v_permlane16_v3i64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-GISEL-NEXT:    s_wait_expcnt 0x0
; GFX12-GISEL-NEXT:    s_wait_samplecnt 0x0
; GFX12-GISEL-NEXT:    s_wait_bvhcnt 0x0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s0, v8
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s1, v9
; GFX12-GISEL-NEXT:    s_wait_alu 0xf1ff
; GFX12-GISEL-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v3, v3, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v4, v4, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v5, v5, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v6, v6, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v7, v7, s0, s1
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX12-GISEL-NEXT:    global_store_b64 v[0:1], v[6:7], off offset:16
; GFX12-GISEL-NEXT:    s_setpc_b64 s[30:31]
  %v = call <3 x i64> @llvm.amdgcn.permlane16.v3i64(<3 x i64> %src0, <3 x i64> %src0, i32 %src1, i32 %src2, i1 false, i1 false)
  store <3 x i64> %v, ptr addrspace(1) %out
  ret void
}

define void @v_permlane16_v4f64(ptr addrspace(1) %out, <4 x double> %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlane16_v4f64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s4, v10
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s5, v11
; GFX10-SDAG-NEXT:    v_permlane16_b32 v9, v9, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v8, v8, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v7, v7, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v6, v6, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v5, v5, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v4, v4, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v3, v3, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v2, v2, s4, s5
; GFX10-SDAG-NEXT:    global_store_dwordx4 v[0:1], v[6:9], off offset:16
; GFX10-SDAG-NEXT:    global_store_dwordx4 v[0:1], v[2:5], off
; GFX10-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX10-GISEL-LABEL: v_permlane16_v4f64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v10
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s5, v11
; GFX10-GISEL-NEXT:    v_permlane16_b32 v2, v2, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v3, v3, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v4, v4, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v5, v5, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v6, v6, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v7, v7, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v8, v8, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v9, v9, s4, s5
; GFX10-GISEL-NEXT:    global_store_dwordx4 v[0:1], v[2:5], off
; GFX10-GISEL-NEXT:    global_store_dwordx4 v[0:1], v[6:9], off offset:16
; GFX10-GISEL-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-SDAG-LABEL: v_permlane16_v4f64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s0, v10
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s1, v11
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v9, v9, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v8, v8, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v7, v7, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v6, v6, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v5, v5, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v4, v4, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v3, v3, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    global_store_b128 v[0:1], v[6:9], off offset:16
; GFX11-SDAG-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX11-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-GISEL-LABEL: v_permlane16_v4f64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s0, v10
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s1, v11
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v3, v3, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v4, v4, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v5, v5, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v6, v6, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v7, v7, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v8, v8, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v9, v9, s0, s1
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX11-GISEL-NEXT:    global_store_b128 v[0:1], v[6:9], off offset:16
; GFX11-GISEL-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-SDAG-LABEL: v_permlane16_v4f64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-SDAG-NEXT:    s_wait_expcnt 0x0
; GFX12-SDAG-NEXT:    s_wait_samplecnt 0x0
; GFX12-SDAG-NEXT:    s_wait_bvhcnt 0x0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s0, v10
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s1, v11
; GFX12-SDAG-NEXT:    s_wait_alu 0xf1ff
; GFX12-SDAG-NEXT:    v_permlane16_b32 v9, v9, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v8, v8, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v7, v7, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v6, v6, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v5, v5, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v4, v4, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v3, v3, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    global_store_b128 v[0:1], v[6:9], off offset:16
; GFX12-SDAG-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX12-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-GISEL-LABEL: v_permlane16_v4f64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-GISEL-NEXT:    s_wait_expcnt 0x0
; GFX12-GISEL-NEXT:    s_wait_samplecnt 0x0
; GFX12-GISEL-NEXT:    s_wait_bvhcnt 0x0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s0, v10
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s1, v11
; GFX12-GISEL-NEXT:    s_wait_alu 0xf1ff
; GFX12-GISEL-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v3, v3, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v4, v4, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v5, v5, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v6, v6, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v7, v7, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v8, v8, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v9, v9, s0, s1
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX12-GISEL-NEXT:    global_store_b128 v[0:1], v[6:9], off offset:16
; GFX12-GISEL-NEXT:    s_setpc_b64 s[30:31]
  %v = call <4 x double> @llvm.amdgcn.permlane16.v4f64(<4 x double> %src0, <4 x double> %src0, i32 %src1, i32 %src2, i1 false, i1 false)
  store <4 x double> %v, ptr addrspace(1) %out
  ret void
}

define void @v_permlane16_v8f64(ptr addrspace(1) %out, <8 x double> %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlane16_v8f64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s4, v18
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s5, v19
; GFX10-SDAG-NEXT:    v_permlane16_b32 v17, v17, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v16, v16, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v15, v15, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v14, v14, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v13, v13, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v12, v12, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v11, v11, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v10, v10, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v9, v9, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v8, v8, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v7, v7, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v6, v6, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v5, v5, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v4, v4, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v3, v3, s4, s5
; GFX10-SDAG-NEXT:    v_permlane16_b32 v2, v2, s4, s5
; GFX10-SDAG-NEXT:    global_store_dwordx4 v[0:1], v[14:17], off offset:48
; GFX10-SDAG-NEXT:    global_store_dwordx4 v[0:1], v[10:13], off offset:32
; GFX10-SDAG-NEXT:    global_store_dwordx4 v[0:1], v[6:9], off offset:16
; GFX10-SDAG-NEXT:    global_store_dwordx4 v[0:1], v[2:5], off
; GFX10-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX10-GISEL-LABEL: v_permlane16_v8f64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v18
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s5, v19
; GFX10-GISEL-NEXT:    v_permlane16_b32 v2, v2, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v3, v3, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v4, v4, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v5, v5, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v6, v6, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v7, v7, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v8, v8, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v9, v9, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v10, v10, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v11, v11, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v12, v12, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v13, v13, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v14, v14, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v15, v15, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v16, v16, s4, s5
; GFX10-GISEL-NEXT:    v_permlane16_b32 v17, v17, s4, s5
; GFX10-GISEL-NEXT:    global_store_dwordx4 v[0:1], v[2:5], off
; GFX10-GISEL-NEXT:    global_store_dwordx4 v[0:1], v[6:9], off offset:16
; GFX10-GISEL-NEXT:    global_store_dwordx4 v[0:1], v[10:13], off offset:32
; GFX10-GISEL-NEXT:    global_store_dwordx4 v[0:1], v[14:17], off offset:48
; GFX10-GISEL-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-SDAG-LABEL: v_permlane16_v8f64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s0, v18
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s1, v19
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-SDAG-NEXT:    v_permlane16_b32 v17, v17, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v16, v16, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v15, v15, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v14, v14, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v13, v13, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v12, v12, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v11, v11, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v10, v10, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v9, v9, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v8, v8, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v7, v7, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v6, v6, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v5, v5, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v4, v4, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v3, v3, s0, s1
; GFX11-SDAG-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX11-SDAG-NEXT:    s_clause 0x3
; GFX11-SDAG-NEXT:    global_store_b128 v[0:1], v[14:17], off offset:48
; GFX11-SDAG-NEXT:    global_store_b128 v[0:1], v[10:13], off offset:32
; GFX11-SDAG-NEXT:    global_store_b128 v[0:1], v[6:9], off offset:16
; GFX11-SDAG-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX11-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-GISEL-LABEL: v_permlane16_v8f64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s0, v18
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s1, v19
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-GISEL-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v3, v3, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v4, v4, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v5, v5, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v6, v6, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v7, v7, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v8, v8, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v9, v9, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v10, v10, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v11, v11, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v12, v12, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v13, v13, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v14, v14, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v15, v15, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v16, v16, s0, s1
; GFX11-GISEL-NEXT:    v_permlane16_b32 v17, v17, s0, s1
; GFX11-GISEL-NEXT:    s_clause 0x3
; GFX11-GISEL-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX11-GISEL-NEXT:    global_store_b128 v[0:1], v[6:9], off offset:16
; GFX11-GISEL-NEXT:    global_store_b128 v[0:1], v[10:13], off offset:32
; GFX11-GISEL-NEXT:    global_store_b128 v[0:1], v[14:17], off offset:48
; GFX11-GISEL-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-SDAG-LABEL: v_permlane16_v8f64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-SDAG-NEXT:    s_wait_expcnt 0x0
; GFX12-SDAG-NEXT:    s_wait_samplecnt 0x0
; GFX12-SDAG-NEXT:    s_wait_bvhcnt 0x0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s0, v18
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s1, v19
; GFX12-SDAG-NEXT:    s_wait_alu 0xf1ff
; GFX12-SDAG-NEXT:    v_permlane16_b32 v17, v17, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v16, v16, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v15, v15, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v14, v14, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v13, v13, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v12, v12, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v11, v11, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v10, v10, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v9, v9, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v8, v8, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v7, v7, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v6, v6, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v5, v5, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v4, v4, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v3, v3, s0, s1
; GFX12-SDAG-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX12-SDAG-NEXT:    s_clause 0x3
; GFX12-SDAG-NEXT:    global_store_b128 v[0:1], v[14:17], off offset:48
; GFX12-SDAG-NEXT:    global_store_b128 v[0:1], v[10:13], off offset:32
; GFX12-SDAG-NEXT:    global_store_b128 v[0:1], v[6:9], off offset:16
; GFX12-SDAG-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX12-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-GISEL-LABEL: v_permlane16_v8f64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-GISEL-NEXT:    s_wait_expcnt 0x0
; GFX12-GISEL-NEXT:    s_wait_samplecnt 0x0
; GFX12-GISEL-NEXT:    s_wait_bvhcnt 0x0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s0, v18
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s1, v19
; GFX12-GISEL-NEXT:    s_wait_alu 0xf1ff
; GFX12-GISEL-NEXT:    v_permlane16_b32 v2, v2, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v3, v3, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v4, v4, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v5, v5, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v6, v6, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v7, v7, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v8, v8, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v9, v9, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v10, v10, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v11, v11, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v12, v12, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v13, v13, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v14, v14, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v15, v15, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v16, v16, s0, s1
; GFX12-GISEL-NEXT:    v_permlane16_b32 v17, v17, s0, s1
; GFX12-GISEL-NEXT:    s_clause 0x3
; GFX12-GISEL-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX12-GISEL-NEXT:    global_store_b128 v[0:1], v[6:9], off offset:16
; GFX12-GISEL-NEXT:    global_store_b128 v[0:1], v[10:13], off offset:32
; GFX12-GISEL-NEXT:    global_store_b128 v[0:1], v[14:17], off offset:48
; GFX12-GISEL-NEXT:    s_setpc_b64 s[30:31]
  %v = call <8 x double> @llvm.amdgcn.permlane16.v8f64(<8 x double> %src0, <8 x double> %src0, i32 %src1, i32 %src2, i1 false, i1 false)
  store <8 x double> %v, ptr addrspace(1) %out
  ret void
}

define void @v_permlanex16_v2i64(ptr addrspace(1) %out, <2 x i64> %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlanex16_v2i64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s4, v6
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s5, v7
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v5, v5, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v4, v4, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v3, v3, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v2, v2, s4, s5
; GFX10-SDAG-NEXT:    global_store_dwordx4 v[0:1], v[2:5], off
; GFX10-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX10-GISEL-LABEL: v_permlanex16_v2i64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v6
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s5, v7
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v2, v2, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v3, v3, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v4, v4, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v5, v5, s4, s5
; GFX10-GISEL-NEXT:    global_store_dwordx4 v[0:1], v[2:5], off
; GFX10-GISEL-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-SDAG-LABEL: v_permlanex16_v2i64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s0, v6
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s1, v7
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v5, v5, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v4, v4, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v3, v3, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX11-SDAG-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX11-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-GISEL-LABEL: v_permlanex16_v2i64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s0, v6
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s1, v7
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v3, v3, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v4, v4, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v5, v5, s0, s1
; GFX11-GISEL-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX11-GISEL-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-SDAG-LABEL: v_permlanex16_v2i64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-SDAG-NEXT:    s_wait_expcnt 0x0
; GFX12-SDAG-NEXT:    s_wait_samplecnt 0x0
; GFX12-SDAG-NEXT:    s_wait_bvhcnt 0x0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s0, v6
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s1, v7
; GFX12-SDAG-NEXT:    s_wait_alu 0xf1ff
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v5, v5, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v4, v4, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v3, v3, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX12-SDAG-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX12-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-GISEL-LABEL: v_permlanex16_v2i64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-GISEL-NEXT:    s_wait_expcnt 0x0
; GFX12-GISEL-NEXT:    s_wait_samplecnt 0x0
; GFX12-GISEL-NEXT:    s_wait_bvhcnt 0x0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s0, v6
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s1, v7
; GFX12-GISEL-NEXT:    s_wait_alu 0xf1ff
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v3, v3, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v4, v4, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v5, v5, s0, s1
; GFX12-GISEL-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX12-GISEL-NEXT:    s_setpc_b64 s[30:31]
  %v = call <2 x i64> @llvm.amdgcn.permlanex16.v2i64(<2 x i64> %src0, <2 x i64> %src0, i32 %src1, i32 %src2, i1 false, i1 false)
  store <2 x i64> %v, ptr addrspace(1) %out
  ret void
}

define void @v_permlanex16_v3i64(ptr addrspace(1) %out, <3 x i64> %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlanex16_v3i64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s4, v8
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s5, v9
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v7, v7, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v6, v6, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v5, v5, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v4, v4, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v3, v3, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v2, v2, s4, s5
; GFX10-SDAG-NEXT:    global_store_dwordx2 v[0:1], v[6:7], off offset:16
; GFX10-SDAG-NEXT:    global_store_dwordx4 v[0:1], v[2:5], off
; GFX10-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX10-GISEL-LABEL: v_permlanex16_v3i64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v8
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s5, v9
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v2, v2, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v3, v3, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v4, v4, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v5, v5, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v6, v6, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v7, v7, s4, s5
; GFX10-GISEL-NEXT:    global_store_dwordx4 v[0:1], v[2:5], off
; GFX10-GISEL-NEXT:    global_store_dwordx2 v[0:1], v[6:7], off offset:16
; GFX10-GISEL-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-SDAG-LABEL: v_permlanex16_v3i64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s0, v8
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s1, v9
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v7, v7, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v6, v6, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v5, v5, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v4, v4, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v3, v3, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    global_store_b64 v[0:1], v[6:7], off offset:16
; GFX11-SDAG-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX11-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-GISEL-LABEL: v_permlanex16_v3i64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s0, v8
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s1, v9
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v3, v3, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v4, v4, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v5, v5, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v6, v6, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v7, v7, s0, s1
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX11-GISEL-NEXT:    global_store_b64 v[0:1], v[6:7], off offset:16
; GFX11-GISEL-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-SDAG-LABEL: v_permlanex16_v3i64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-SDAG-NEXT:    s_wait_expcnt 0x0
; GFX12-SDAG-NEXT:    s_wait_samplecnt 0x0
; GFX12-SDAG-NEXT:    s_wait_bvhcnt 0x0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s0, v8
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s1, v9
; GFX12-SDAG-NEXT:    s_wait_alu 0xf1ff
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v7, v7, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v6, v6, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v5, v5, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v4, v4, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v3, v3, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    global_store_b64 v[0:1], v[6:7], off offset:16
; GFX12-SDAG-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX12-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-GISEL-LABEL: v_permlanex16_v3i64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-GISEL-NEXT:    s_wait_expcnt 0x0
; GFX12-GISEL-NEXT:    s_wait_samplecnt 0x0
; GFX12-GISEL-NEXT:    s_wait_bvhcnt 0x0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s0, v8
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s1, v9
; GFX12-GISEL-NEXT:    s_wait_alu 0xf1ff
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v3, v3, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v4, v4, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v5, v5, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v6, v6, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v7, v7, s0, s1
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX12-GISEL-NEXT:    global_store_b64 v[0:1], v[6:7], off offset:16
; GFX12-GISEL-NEXT:    s_setpc_b64 s[30:31]
  %v = call <3 x i64> @llvm.amdgcn.permlanex16.v3i64(<3 x i64> %src0, <3 x i64> %src0, i32 %src1, i32 %src2, i1 false, i1 false)
  store <3 x i64> %v, ptr addrspace(1) %out
  ret void
}

define void @v_permlanex16_v4f64(ptr addrspace(1) %out, <4 x double> %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlanex16_v4f64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s4, v10
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s5, v11
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v9, v9, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v8, v8, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v7, v7, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v6, v6, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v5, v5, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v4, v4, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v3, v3, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v2, v2, s4, s5
; GFX10-SDAG-NEXT:    global_store_dwordx4 v[0:1], v[6:9], off offset:16
; GFX10-SDAG-NEXT:    global_store_dwordx4 v[0:1], v[2:5], off
; GFX10-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX10-GISEL-LABEL: v_permlanex16_v4f64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v10
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s5, v11
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v2, v2, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v3, v3, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v4, v4, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v5, v5, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v6, v6, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v7, v7, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v8, v8, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v9, v9, s4, s5
; GFX10-GISEL-NEXT:    global_store_dwordx4 v[0:1], v[2:5], off
; GFX10-GISEL-NEXT:    global_store_dwordx4 v[0:1], v[6:9], off offset:16
; GFX10-GISEL-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-SDAG-LABEL: v_permlanex16_v4f64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s0, v10
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s1, v11
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v9, v9, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v8, v8, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v7, v7, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v6, v6, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v5, v5, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v4, v4, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v3, v3, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX11-SDAG-NEXT:    s_clause 0x1
; GFX11-SDAG-NEXT:    global_store_b128 v[0:1], v[6:9], off offset:16
; GFX11-SDAG-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX11-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-GISEL-LABEL: v_permlanex16_v4f64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s0, v10
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s1, v11
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v3, v3, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v4, v4, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v5, v5, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v6, v6, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v7, v7, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v8, v8, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v9, v9, s0, s1
; GFX11-GISEL-NEXT:    s_clause 0x1
; GFX11-GISEL-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX11-GISEL-NEXT:    global_store_b128 v[0:1], v[6:9], off offset:16
; GFX11-GISEL-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-SDAG-LABEL: v_permlanex16_v4f64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-SDAG-NEXT:    s_wait_expcnt 0x0
; GFX12-SDAG-NEXT:    s_wait_samplecnt 0x0
; GFX12-SDAG-NEXT:    s_wait_bvhcnt 0x0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s0, v10
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s1, v11
; GFX12-SDAG-NEXT:    s_wait_alu 0xf1ff
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v9, v9, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v8, v8, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v7, v7, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v6, v6, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v5, v5, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v4, v4, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v3, v3, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX12-SDAG-NEXT:    s_clause 0x1
; GFX12-SDAG-NEXT:    global_store_b128 v[0:1], v[6:9], off offset:16
; GFX12-SDAG-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX12-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-GISEL-LABEL: v_permlanex16_v4f64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-GISEL-NEXT:    s_wait_expcnt 0x0
; GFX12-GISEL-NEXT:    s_wait_samplecnt 0x0
; GFX12-GISEL-NEXT:    s_wait_bvhcnt 0x0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s0, v10
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s1, v11
; GFX12-GISEL-NEXT:    s_wait_alu 0xf1ff
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v3, v3, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v4, v4, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v5, v5, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v6, v6, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v7, v7, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v8, v8, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v9, v9, s0, s1
; GFX12-GISEL-NEXT:    s_clause 0x1
; GFX12-GISEL-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX12-GISEL-NEXT:    global_store_b128 v[0:1], v[6:9], off offset:16
; GFX12-GISEL-NEXT:    s_setpc_b64 s[30:31]
  %v = call <4 x double> @llvm.amdgcn.permlanex16.v4f64(<4 x double> %src0, <4 x double> %src0, i32 %src1, i32 %src2, i1 false, i1 false)
  store <4 x double> %v, ptr addrspace(1) %out
  ret void
}

define void @v_permlanex16_v8f64(ptr addrspace(1) %out, <8 x double> %src0, i32 %src1, i32 %src2) {
; GFX10-SDAG-LABEL: v_permlanex16_v8f64:
; GFX10-SDAG:       ; %bb.0:
; GFX10-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s4, v18
; GFX10-SDAG-NEXT:    v_readfirstlane_b32 s5, v19
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v17, v17, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v16, v16, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v15, v15, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v14, v14, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v13, v13, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v12, v12, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v11, v11, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v10, v10, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v9, v9, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v8, v8, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v7, v7, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v6, v6, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v5, v5, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v4, v4, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v3, v3, s4, s5
; GFX10-SDAG-NEXT:    v_permlanex16_b32 v2, v2, s4, s5
; GFX10-SDAG-NEXT:    global_store_dwordx4 v[0:1], v[14:17], off offset:48
; GFX10-SDAG-NEXT:    global_store_dwordx4 v[0:1], v[10:13], off offset:32
; GFX10-SDAG-NEXT:    global_store_dwordx4 v[0:1], v[6:9], off offset:16
; GFX10-SDAG-NEXT:    global_store_dwordx4 v[0:1], v[2:5], off
; GFX10-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX10-GISEL-LABEL: v_permlanex16_v8f64:
; GFX10-GISEL:       ; %bb.0:
; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s4, v18
; GFX10-GISEL-NEXT:    v_readfirstlane_b32 s5, v19
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v2, v2, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v3, v3, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v4, v4, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v5, v5, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v6, v6, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v7, v7, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v8, v8, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v9, v9, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v10, v10, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v11, v11, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v12, v12, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v13, v13, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v14, v14, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v15, v15, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v16, v16, s4, s5
; GFX10-GISEL-NEXT:    v_permlanex16_b32 v17, v17, s4, s5
; GFX10-GISEL-NEXT:    global_store_dwordx4 v[0:1], v[2:5], off
; GFX10-GISEL-NEXT:    global_store_dwordx4 v[0:1], v[6:9], off offset:16
; GFX10-GISEL-NEXT:    global_store_dwordx4 v[0:1], v[10:13], off offset:32
; GFX10-GISEL-NEXT:    global_store_dwordx4 v[0:1], v[14:17], off offset:48
; GFX10-GISEL-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-SDAG-LABEL: v_permlanex16_v8f64:
; GFX11-SDAG:       ; %bb.0:
; GFX11-SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s0, v18
; GFX11-SDAG-NEXT:    v_readfirstlane_b32 s1, v19
; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v17, v17, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v16, v16, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v15, v15, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v14, v14, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v13, v13, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v12, v12, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v11, v11, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v10, v10, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v9, v9, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v8, v8, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v7, v7, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v6, v6, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v5, v5, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v4, v4, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v3, v3, s0, s1
; GFX11-SDAG-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX11-SDAG-NEXT:    s_clause 0x3
; GFX11-SDAG-NEXT:    global_store_b128 v[0:1], v[14:17], off offset:48
; GFX11-SDAG-NEXT:    global_store_b128 v[0:1], v[10:13], off offset:32
; GFX11-SDAG-NEXT:    global_store_b128 v[0:1], v[6:9], off offset:16
; GFX11-SDAG-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX11-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX11-GISEL-LABEL: v_permlanex16_v8f64:
; GFX11-GISEL:       ; %bb.0:
; GFX11-GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s0, v18
; GFX11-GISEL-NEXT:    v_readfirstlane_b32 s1, v19
; GFX11-GISEL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v3, v3, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v4, v4, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v5, v5, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v6, v6, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v7, v7, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v8, v8, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v9, v9, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v10, v10, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v11, v11, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v12, v12, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v13, v13, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v14, v14, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v15, v15, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v16, v16, s0, s1
; GFX11-GISEL-NEXT:    v_permlanex16_b32 v17, v17, s0, s1
; GFX11-GISEL-NEXT:    s_clause 0x3
; GFX11-GISEL-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX11-GISEL-NEXT:    global_store_b128 v[0:1], v[6:9], off offset:16
; GFX11-GISEL-NEXT:    global_store_b128 v[0:1], v[10:13], off offset:32
; GFX11-GISEL-NEXT:    global_store_b128 v[0:1], v[14:17], off offset:48
; GFX11-GISEL-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-SDAG-LABEL: v_permlanex16_v8f64:
; GFX12-SDAG:       ; %bb.0:
; GFX12-SDAG-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-SDAG-NEXT:    s_wait_expcnt 0x0
; GFX12-SDAG-NEXT:    s_wait_samplecnt 0x0
; GFX12-SDAG-NEXT:    s_wait_bvhcnt 0x0
; GFX12-SDAG-NEXT:    s_wait_kmcnt 0x0
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s0, v18
; GFX12-SDAG-NEXT:    v_readfirstlane_b32 s1, v19
; GFX12-SDAG-NEXT:    s_wait_alu 0xf1ff
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v17, v17, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v16, v16, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v15, v15, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v14, v14, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v13, v13, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v12, v12, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v11, v11, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v10, v10, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v9, v9, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v8, v8, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v7, v7, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v6, v6, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v5, v5, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v4, v4, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v3, v3, s0, s1
; GFX12-SDAG-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX12-SDAG-NEXT:    s_clause 0x3
; GFX12-SDAG-NEXT:    global_store_b128 v[0:1], v[14:17], off offset:48
; GFX12-SDAG-NEXT:    global_store_b128 v[0:1], v[10:13], off offset:32
; GFX12-SDAG-NEXT:    global_store_b128 v[0:1], v[6:9], off offset:16
; GFX12-SDAG-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX12-SDAG-NEXT:    s_setpc_b64 s[30:31]
;
; GFX12-GISEL-LABEL: v_permlanex16_v8f64:
; GFX12-GISEL:       ; %bb.0:
; GFX12-GISEL-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-GISEL-NEXT:    s_wait_expcnt 0x0
; GFX12-GISEL-NEXT:    s_wait_samplecnt 0x0
; GFX12-GISEL-NEXT:    s_wait_bvhcnt 0x0
; GFX12-GISEL-NEXT:    s_wait_kmcnt 0x0
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s0, v18
; GFX12-GISEL-NEXT:    v_readfirstlane_b32 s1, v19
; GFX12-GISEL-NEXT:    s_wait_alu 0xf1ff
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v2, v2, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v3, v3, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v4, v4, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v5, v5, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v6, v6, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v7, v7, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v8, v8, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v9, v9, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v10, v10, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v11, v11, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v12, v12, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v13, v13, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v14, v14, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v15, v15, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v16, v16, s0, s1
; GFX12-GISEL-NEXT:    v_permlanex16_b32 v17, v17, s0, s1
; GFX12-GISEL-NEXT:    s_clause 0x3
; GFX12-GISEL-NEXT:    global_store_b128 v[0:1], v[2:5], off
; GFX12-GISEL-NEXT:    global_store_b128 v[0:1], v[6:9], off offset:16
; GFX12-GISEL-NEXT:    global_store_b128 v[0:1], v[10:13], off offset:32
; GFX12-GISEL-NEXT:    global_store_b128 v[0:1], v[14:17], off offset:48
; GFX12-GISEL-NEXT:    s_setpc_b64 s[30:31]
  %v = call <8 x double> @llvm.amdgcn.permlanex16.v8f64(<8 x double> %src0, <8 x double> %src0, i32 %src1, i32 %src2, i1 false, i1 false)
  store <8 x double> %v, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_permlanex16_convergent(ptr addrspace(1) %out, i32 %src0, i32 %pattern_lo, i32 %pattern_hi) {
; GFX10-LABEL: v_permlanex16_convergent:
; GFX10:       ; %bb.0:
; GFX10-NEXT:    s_clause 0x1
; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x2c
; GFX10-NEXT:    s_load_dword s2, s[4:5], 0x34
; GFX10-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    v_mov_b32_e32 v1, s0
; GFX10-NEXT:    v_permlanex16_b32 v1, v1, s1, s2
; GFX10-NEXT:    s_and_saveexec_b32 s0, vcc_lo
; GFX10-NEXT:    s_cbranch_execz .LBB142_2
; GFX10-NEXT:  ; %bb.1: ; %t
; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
; GFX10-NEXT:    v_mov_b32_e32 v0, 0
; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
; GFX10-NEXT:    global_store_dword v0, v1, s[0:1]
; GFX10-NEXT:  .LBB142_2: ; %f
; GFX10-NEXT:    s_endpgm
;
; GFX11-LABEL: v_permlanex16_convergent:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x2c
; GFX11-NEXT:    s_load_b32 s2, s[4:5], 0x34
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v1, s0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX11-NEXT:    s_mov_b32 s0, exec_lo
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-NEXT:    v_permlanex16_b32 v1, v1, s1, s2
; GFX11-NEXT:    v_cmpx_eq_u32_e32 0, v0
; GFX11-NEXT:    s_cbranch_execz .LBB142_2
; GFX11-NEXT:  ; %bb.1: ; %t
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
; GFX11-NEXT:    v_mov_b32_e32 v0, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX11-NEXT:  .LBB142_2: ; %f
; GFX11-NEXT:    s_endpgm
;
; GFX12-LABEL: v_permlanex16_convergent:
; GFX12:       ; %bb.0:
; GFX12-NEXT:    s_load_b96 s[0:2], s[4:5], 0x2c
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_dual_mov_b32 v1, s0 :: v_dual_and_b32 v0, 0x3ff, v0
; GFX12-NEXT:    s_mov_b32 s0, exec_lo
; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-NEXT:    v_permlanex16_b32 v1, v1, s1, s2
; GFX12-NEXT:    v_cmpx_eq_u32_e32 0, v0
; GFX12-NEXT:    s_cbranch_execz .LBB142_2
; GFX12-NEXT:  ; %bb.1: ; %t
; GFX12-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
; GFX12-NEXT:    v_mov_b32_e32 v0, 0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX12-NEXT:  .LBB142_2: ; %f
; GFX12-NEXT:    s_endpgm
  %tidx = call i32 @llvm.amdgcn.workitem.id.x()
  %v = call i32 @llvm.amdgcn.permlanex16.i32(i32 %src0, i32 %src0, i32 %pattern_lo, i32 %pattern_hi, i1 false, i1 false)
  %select = icmp eq i32 %tidx, 0
  br i1 %select, label %t, label %f
t:
  store i32 %v, ptr addrspace(1) %out
  br label %f
f:
  ret void
}