; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=tahiti < %s | FileCheck -check-prefix=SI %s
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=tonga < %s | FileCheck -check-prefix=VI %s
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck -check-prefix=GFX900 %s
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx942 < %s | FileCheck -check-prefix=GFX942 %s
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 < %s | FileCheck -check-prefix=GFX1250 %s

define amdgpu_kernel void @s_insertelement_v2bf16_0(ptr addrspace(1) %out, ptr addrspace(4) %vec.ptr) #0 {
; SI-LABEL: s_insertelement_v2bf16_0:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_load_dword s4, s[2:3], 0x0
; SI-NEXT:    s_mov_b32 s3, 0x100f000
; SI-NEXT:    s_mov_b32 s2, -1
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_and_b32 s4, s4, 0xffff0000
; SI-NEXT:    s_or_b32 s4, s4, 0x40a0
; SI-NEXT:    v_mov_b32_e32 v0, s4
; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_insertelement_v2bf16_0:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    s_add_i32 s12, s12, s17
; VI-NEXT:    s_mov_b32 flat_scratch_lo, s13
; VI-NEXT:    s_lshr_b32 flat_scratch_hi, s12, 8
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_load_dword s2, s[2:3], 0x0
; VI-NEXT:    v_mov_b32_e32 v0, s0
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_and_b32 s0, s2, 0xffff0000
; VI-NEXT:    s_or_b32 s0, s0, 0x40a0
; VI-NEXT:    v_mov_b32_e32 v2, s0
; VI-NEXT:    flat_store_dword v[0:1], v2
; VI-NEXT:    s_endpgm
;
; GFX900-LABEL: s_insertelement_v2bf16_0:
; GFX900:       ; %bb.0:
; GFX900-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; GFX900-NEXT:    v_mov_b32_e32 v0, 0
; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
; GFX900-NEXT:    s_load_dword s2, s[2:3], 0x0
; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
; GFX900-NEXT:    s_lshr_b32 s2, s2, 16
; GFX900-NEXT:    s_pack_ll_b32_b16 s2, 0x40a0, s2
; GFX900-NEXT:    v_mov_b32_e32 v1, s2
; GFX900-NEXT:    global_store_dword v0, v1, s[0:1]
; GFX900-NEXT:    s_endpgm
;
; GFX942-LABEL: s_insertelement_v2bf16_0:
; GFX942:       ; %bb.0:
; GFX942-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
; GFX942-NEXT:    v_mov_b32_e32 v0, 0
; GFX942-NEXT:    s_waitcnt lgkmcnt(0)
; GFX942-NEXT:    s_load_dword s2, s[2:3], 0x0
; GFX942-NEXT:    s_waitcnt lgkmcnt(0)
; GFX942-NEXT:    s_lshr_b32 s2, s2, 16
; GFX942-NEXT:    s_pack_ll_b32_b16 s2, 0x40a0, s2
; GFX942-NEXT:    v_mov_b32_e32 v1, s2
; GFX942-NEXT:    global_store_dword v0, v1, s[0:1]
; GFX942-NEXT:    s_endpgm
;
; GFX1250-LABEL: s_insertelement_v2bf16_0:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    s_load_b32 s2, s[2:3], 0x0
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    s_lshr_b32 s2, s2, 16
; GFX1250-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX1250-NEXT:    s_pack_ll_b32_b16 s2, 0x40a0, s2
; GFX1250-NEXT:    v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, s2
; GFX1250-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX1250-NEXT:    s_endpgm
  %vec = load <2 x bfloat>, ptr addrspace(4) %vec.ptr
  %vecins = insertelement <2 x bfloat> %vec, bfloat 5.000000e+00, i32 0
  store <2 x bfloat> %vecins, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @s_insertelement_v2bf16_1(ptr addrspace(1) %out, ptr addrspace(4) %vec.ptr) #0 {
; SI-LABEL: s_insertelement_v2bf16_1:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_load_dword s4, s[2:3], 0x0
; SI-NEXT:    s_mov_b32 s3, 0x100f000
; SI-NEXT:    s_mov_b32 s2, -1
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_and_b32 s4, s4, 0xffff
; SI-NEXT:    s_or_b32 s4, s4, 0x40a00000
; SI-NEXT:    v_mov_b32_e32 v0, s4
; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: s_insertelement_v2bf16_1:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    s_add_i32 s12, s12, s17
; VI-NEXT:    s_mov_b32 flat_scratch_lo, s13
; VI-NEXT:    s_lshr_b32 flat_scratch_hi, s12, 8
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_load_dword s2, s[2:3], 0x0
; VI-NEXT:    v_mov_b32_e32 v0, s0
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_and_b32 s0, s2, 0xffff
; VI-NEXT:    s_or_b32 s0, s0, 0x40a00000
; VI-NEXT:    v_mov_b32_e32 v2, s0
; VI-NEXT:    flat_store_dword v[0:1], v2
; VI-NEXT:    s_endpgm
;
; GFX900-LABEL: s_insertelement_v2bf16_1:
; GFX900:       ; %bb.0:
; GFX900-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; GFX900-NEXT:    v_mov_b32_e32 v0, 0
; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
; GFX900-NEXT:    s_load_dword s2, s[2:3], 0x0
; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
; GFX900-NEXT:    s_pack_ll_b32_b16 s2, s2, 0x40a0
; GFX900-NEXT:    v_mov_b32_e32 v1, s2
; GFX900-NEXT:    global_store_dword v0, v1, s[0:1]
; GFX900-NEXT:    s_endpgm
;
; GFX942-LABEL: s_insertelement_v2bf16_1:
; GFX942:       ; %bb.0:
; GFX942-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
; GFX942-NEXT:    v_mov_b32_e32 v0, 0
; GFX942-NEXT:    s_waitcnt lgkmcnt(0)
; GFX942-NEXT:    s_load_dword s2, s[2:3], 0x0
; GFX942-NEXT:    s_waitcnt lgkmcnt(0)
; GFX942-NEXT:    s_pack_ll_b32_b16 s2, s2, 0x40a0
; GFX942-NEXT:    v_mov_b32_e32 v1, s2
; GFX942-NEXT:    global_store_dword v0, v1, s[0:1]
; GFX942-NEXT:    s_endpgm
;
; GFX1250-LABEL: s_insertelement_v2bf16_1:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    s_load_b32 s2, s[2:3], 0x0
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    s_pack_ll_b32_b16 s2, s2, 0x40a0
; GFX1250-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
; GFX1250-NEXT:    v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, s2
; GFX1250-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX1250-NEXT:    s_endpgm
  %vec = load <2 x bfloat>, ptr addrspace(4) %vec.ptr
  %vecins = insertelement <2 x bfloat> %vec, bfloat 5.000000e+00, i32 1
  store <2 x bfloat> %vecins, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @v_insertelement_v2bf16_0(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
; SI-LABEL: v_insertelement_v2bf16_0:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; SI-NEXT:    s_mov_b32 s7, 0x100f000
; SI-NEXT:    s_mov_b32 s6, 0
; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
; SI-NEXT:    v_mov_b32_e32 v1, 0
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b64 s[4:5], s[2:3]
; SI-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
; SI-NEXT:    s_mov_b64 s[2:3], s[6:7]
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_and_b32_e32 v2, 0xffff0000, v2
; SI-NEXT:    v_or_b32_e32 v2, 0x40a0, v2
; SI-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
; SI-NEXT:    s_endpgm
;
; VI-LABEL: v_insertelement_v2bf16_0:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
; VI-NEXT:    s_add_i32 s12, s12, s17
; VI-NEXT:    s_mov_b32 flat_scratch_lo, s13
; VI-NEXT:    s_lshr_b32 flat_scratch_hi, s12, 8
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT:    flat_load_dword v3, v[0:1]
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_and_b32_e32 v2, 0xffff0000, v3
; VI-NEXT:    v_or_b32_e32 v2, 0x40a0, v2
; VI-NEXT:    flat_store_dword v[0:1], v2
; VI-NEXT:    s_endpgm
;
; GFX900-LABEL: v_insertelement_v2bf16_0:
; GFX900:       ; %bb.0:
; GFX900-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; GFX900-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
; GFX900-NEXT:    v_mov_b32_e32 v2, 0x40a0
; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
; GFX900-NEXT:    global_load_dword v1, v0, s[2:3]
; GFX900-NEXT:    s_mov_b32 s2, 0xffff
; GFX900-NEXT:    s_waitcnt vmcnt(0)
; GFX900-NEXT:    v_bfi_b32 v1, s2, v2, v1
; GFX900-NEXT:    global_store_dword v0, v1, s[0:1]
; GFX900-NEXT:    s_endpgm
;
; GFX942-LABEL: v_insertelement_v2bf16_0:
; GFX942:       ; %bb.0:
; GFX942-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
; GFX942-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX942-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
; GFX942-NEXT:    v_mov_b32_e32 v2, 0x40a0
; GFX942-NEXT:    s_waitcnt lgkmcnt(0)
; GFX942-NEXT:    global_load_dword v1, v0, s[2:3]
; GFX942-NEXT:    s_mov_b32 s2, 0xffff
; GFX942-NEXT:    s_waitcnt vmcnt(0)
; GFX942-NEXT:    v_bfi_b32 v1, s2, v2, v1
; GFX942-NEXT:    global_store_dword v0, v1, s[0:1]
; GFX942-NEXT:    s_endpgm
;
; GFX1250-LABEL: v_insertelement_v2bf16_0:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX1250-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    global_load_b32 v1, v0, s[2:3] scale_offset
; GFX1250-NEXT:    s_wait_xcnt 0x0
; GFX1250-NEXT:    s_movk_i32 s2, 0x40a0
; GFX1250-NEXT:    s_wait_loadcnt 0x0
; GFX1250-NEXT:    v_bfi_b32 v1, 0xffff, s2, v1
; GFX1250-NEXT:    global_store_b32 v0, v1, s[0:1] scale_offset
; GFX1250-NEXT:    s_endpgm
  %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
  %tid.ext = sext i32 %tid to i64
  %in.gep = getelementptr inbounds <2 x bfloat>, ptr addrspace(1) %in, i64 %tid.ext
  %out.gep = getelementptr inbounds <2 x bfloat>, ptr addrspace(1) %out, i64 %tid.ext
  %vec = load <2 x bfloat>, ptr addrspace(1) %in.gep
  %vecins = insertelement <2 x bfloat> %vec, bfloat 5.000000e+00, i32 0
  store <2 x bfloat> %vecins, ptr addrspace(1) %out.gep
  ret void
}

define amdgpu_kernel void @v_insertelement_v2bf16_0_inlineimm(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
; SI-LABEL: v_insertelement_v2bf16_0_inlineimm:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; SI-NEXT:    s_mov_b32 s7, 0x100f000
; SI-NEXT:    s_mov_b32 s6, 0
; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
; SI-NEXT:    v_mov_b32_e32 v1, 0
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b64 s[4:5], s[2:3]
; SI-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
; SI-NEXT:    s_mov_b64 s[2:3], s[6:7]
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_and_b32_e32 v2, 0xffff0000, v2
; SI-NEXT:    v_or_b32_e32 v2, 53, v2
; SI-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
; SI-NEXT:    s_endpgm
;
; VI-LABEL: v_insertelement_v2bf16_0_inlineimm:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
; VI-NEXT:    s_add_i32 s12, s12, s17
; VI-NEXT:    s_mov_b32 flat_scratch_lo, s13
; VI-NEXT:    s_lshr_b32 flat_scratch_hi, s12, 8
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT:    flat_load_dword v3, v[0:1]
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_and_b32_e32 v2, 0xffff0000, v3
; VI-NEXT:    v_or_b32_e32 v2, 53, v2
; VI-NEXT:    flat_store_dword v[0:1], v2
; VI-NEXT:    s_endpgm
;
; GFX900-LABEL: v_insertelement_v2bf16_0_inlineimm:
; GFX900:       ; %bb.0:
; GFX900-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; GFX900-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
; GFX900-NEXT:    global_load_dword v1, v0, s[2:3]
; GFX900-NEXT:    s_mov_b32 s2, 0xffff
; GFX900-NEXT:    s_waitcnt vmcnt(0)
; GFX900-NEXT:    v_bfi_b32 v1, s2, 53, v1
; GFX900-NEXT:    global_store_dword v0, v1, s[0:1]
; GFX900-NEXT:    s_endpgm
;
; GFX942-LABEL: v_insertelement_v2bf16_0_inlineimm:
; GFX942:       ; %bb.0:
; GFX942-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
; GFX942-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX942-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
; GFX942-NEXT:    s_waitcnt lgkmcnt(0)
; GFX942-NEXT:    global_load_dword v1, v0, s[2:3]
; GFX942-NEXT:    s_mov_b32 s2, 0xffff
; GFX942-NEXT:    s_waitcnt vmcnt(0)
; GFX942-NEXT:    v_bfi_b32 v1, s2, 53, v1
; GFX942-NEXT:    global_store_dword v0, v1, s[0:1]
; GFX942-NEXT:    s_endpgm
;
; GFX1250-LABEL: v_insertelement_v2bf16_0_inlineimm:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX1250-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    global_load_b32 v1, v0, s[2:3] scale_offset
; GFX1250-NEXT:    s_wait_loadcnt 0x0
; GFX1250-NEXT:    v_bfi_b32 v1, 0xffff, 53, v1
; GFX1250-NEXT:    global_store_b32 v0, v1, s[0:1] scale_offset
; GFX1250-NEXT:    s_endpgm
  %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
  %tid.ext = sext i32 %tid to i64
  %in.gep = getelementptr inbounds <2 x bfloat>, ptr addrspace(1) %in, i64 %tid.ext
  %out.gep = getelementptr inbounds <2 x bfloat>, ptr addrspace(1) %out, i64 %tid.ext
  %vec = load <2 x bfloat>, ptr addrspace(1) %in.gep
  %vecins = insertelement <2 x bfloat> %vec, bfloat 0xR0035, i32 0
  store <2 x bfloat> %vecins, ptr addrspace(1) %out.gep
  ret void
}

define amdgpu_kernel void @v_insertelement_v2bf16_1(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
; SI-LABEL: v_insertelement_v2bf16_1:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; SI-NEXT:    s_mov_b32 s7, 0x100f000
; SI-NEXT:    s_mov_b32 s6, 0
; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
; SI-NEXT:    v_mov_b32_e32 v1, 0
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b64 s[4:5], s[2:3]
; SI-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
; SI-NEXT:    s_mov_b64 s[2:3], s[6:7]
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_and_b32_e32 v2, 0xffff, v2
; SI-NEXT:    v_or_b32_e32 v2, 0x40a00000, v2
; SI-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
; SI-NEXT:    s_endpgm
;
; VI-LABEL: v_insertelement_v2bf16_1:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
; VI-NEXT:    s_add_i32 s12, s12, s17
; VI-NEXT:    s_mov_b32 flat_scratch_lo, s13
; VI-NEXT:    s_lshr_b32 flat_scratch_hi, s12, 8
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT:    flat_load_dword v3, v[0:1]
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
; VI-NEXT:    v_mov_b32_e32 v2, 0x40a00000
; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_or_b32_sdwa v2, v3, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
; VI-NEXT:    flat_store_dword v[0:1], v2
; VI-NEXT:    s_endpgm
;
; GFX900-LABEL: v_insertelement_v2bf16_1:
; GFX900:       ; %bb.0:
; GFX900-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; GFX900-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
; GFX900-NEXT:    v_mov_b32_e32 v2, 0x5040100
; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
; GFX900-NEXT:    global_load_dword v1, v0, s[2:3]
; GFX900-NEXT:    s_movk_i32 s2, 0x40a0
; GFX900-NEXT:    s_waitcnt vmcnt(0)
; GFX900-NEXT:    v_perm_b32 v1, s2, v1, v2
; GFX900-NEXT:    global_store_dword v0, v1, s[0:1]
; GFX900-NEXT:    s_endpgm
;
; GFX942-LABEL: v_insertelement_v2bf16_1:
; GFX942:       ; %bb.0:
; GFX942-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
; GFX942-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX942-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
; GFX942-NEXT:    v_mov_b32_e32 v2, 0x5040100
; GFX942-NEXT:    s_waitcnt lgkmcnt(0)
; GFX942-NEXT:    global_load_dword v1, v0, s[2:3]
; GFX942-NEXT:    s_movk_i32 s2, 0x40a0
; GFX942-NEXT:    s_waitcnt vmcnt(0)
; GFX942-NEXT:    v_perm_b32 v1, s2, v1, v2
; GFX942-NEXT:    global_store_dword v0, v1, s[0:1]
; GFX942-NEXT:    s_endpgm
;
; GFX1250-LABEL: v_insertelement_v2bf16_1:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX1250-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    global_load_b32 v1, v0, s[2:3] scale_offset
; GFX1250-NEXT:    s_wait_xcnt 0x0
; GFX1250-NEXT:    s_movk_i32 s2, 0x40a0
; GFX1250-NEXT:    s_wait_loadcnt 0x0
; GFX1250-NEXT:    v_perm_b32 v1, s2, v1, 0x5040100
; GFX1250-NEXT:    global_store_b32 v0, v1, s[0:1] scale_offset
; GFX1250-NEXT:    s_endpgm
  %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
  %tid.ext = sext i32 %tid to i64
  %in.gep = getelementptr inbounds <2 x bfloat>, ptr addrspace(1) %in, i64 %tid.ext
  %out.gep = getelementptr inbounds <2 x bfloat>, ptr addrspace(1) %out, i64 %tid.ext
  %vec = load <2 x bfloat>, ptr addrspace(1) %in.gep
  %vecins = insertelement <2 x bfloat> %vec, bfloat 5.000000e+00, i32 1
  store <2 x bfloat> %vecins, ptr addrspace(1) %out.gep
  ret void
}

define amdgpu_kernel void @v_insertelement_v2bf16_1_inlineimm(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
; SI-LABEL: v_insertelement_v2bf16_1_inlineimm:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; SI-NEXT:    s_mov_b32 s7, 0x100f000
; SI-NEXT:    s_mov_b32 s6, 0
; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
; SI-NEXT:    v_mov_b32_e32 v1, 0
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b64 s[4:5], s[2:3]
; SI-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
; SI-NEXT:    s_mov_b64 s[2:3], s[6:7]
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_and_b32_e32 v2, 0xffff, v2
; SI-NEXT:    v_or_b32_e32 v2, 0x230000, v2
; SI-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
; SI-NEXT:    s_endpgm
;
; VI-LABEL: v_insertelement_v2bf16_1_inlineimm:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
; VI-NEXT:    s_add_i32 s12, s12, s17
; VI-NEXT:    s_mov_b32 flat_scratch_lo, s13
; VI-NEXT:    s_lshr_b32 flat_scratch_hi, s12, 8
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT:    flat_load_dword v3, v[0:1]
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
; VI-NEXT:    v_mov_b32_e32 v2, 0x230000
; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_or_b32_sdwa v2, v3, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
; VI-NEXT:    flat_store_dword v[0:1], v2
; VI-NEXT:    s_endpgm
;
; GFX900-LABEL: v_insertelement_v2bf16_1_inlineimm:
; GFX900:       ; %bb.0:
; GFX900-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; GFX900-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
; GFX900-NEXT:    v_mov_b32_e32 v2, 0x5040100
; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
; GFX900-NEXT:    global_load_dword v1, v0, s[2:3]
; GFX900-NEXT:    s_waitcnt vmcnt(0)
; GFX900-NEXT:    v_perm_b32 v1, 35, v1, v2
; GFX900-NEXT:    global_store_dword v0, v1, s[0:1]
; GFX900-NEXT:    s_endpgm
;
; GFX942-LABEL: v_insertelement_v2bf16_1_inlineimm:
; GFX942:       ; %bb.0:
; GFX942-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
; GFX942-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX942-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
; GFX942-NEXT:    v_mov_b32_e32 v2, 0x5040100
; GFX942-NEXT:    s_waitcnt lgkmcnt(0)
; GFX942-NEXT:    global_load_dword v1, v0, s[2:3]
; GFX942-NEXT:    s_waitcnt vmcnt(0)
; GFX942-NEXT:    v_perm_b32 v1, 35, v1, v2
; GFX942-NEXT:    global_store_dword v0, v1, s[0:1]
; GFX942-NEXT:    s_endpgm
;
; GFX1250-LABEL: v_insertelement_v2bf16_1_inlineimm:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX1250-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    global_load_b32 v1, v0, s[2:3] scale_offset
; GFX1250-NEXT:    s_wait_loadcnt 0x0
; GFX1250-NEXT:    v_perm_b32 v1, 35, v1, 0x5040100
; GFX1250-NEXT:    global_store_b32 v0, v1, s[0:1] scale_offset
; GFX1250-NEXT:    s_endpgm
  %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
  %tid.ext = sext i32 %tid to i64
  %in.gep = getelementptr inbounds <2 x bfloat>, ptr addrspace(1) %in, i64 %tid.ext
  %out.gep = getelementptr inbounds <2 x bfloat>, ptr addrspace(1) %out, i64 %tid.ext
  %vec = load <2 x bfloat>, ptr addrspace(1) %in.gep
  %vecins = insertelement <2 x bfloat> %vec, bfloat 0xR0023, i32 1
  store <2 x bfloat> %vecins, ptr addrspace(1) %out.gep
  ret void
}

define amdgpu_kernel void @v_insertelement_v2bf16_dynamic_vgpr(ptr addrspace(1) %out, ptr addrspace(1) %in, ptr addrspace(1) %idx.ptr) #0 {
; SI-LABEL: v_insertelement_v2bf16_dynamic_vgpr:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; SI-NEXT:    s_load_dwordx2 s[4:5], s[8:9], 0x4
; SI-NEXT:    s_mov_b32 s11, 0x100f000
; SI-NEXT:    s_mov_b32 s10, 0
; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
; SI-NEXT:    v_mov_b32_e32 v1, 0
; SI-NEXT:    s_mov_b64 s[6:7], s[10:11]
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
; SI-NEXT:    buffer_load_dword v3, v[0:1], s[8:11], 0 addr64
; SI-NEXT:    s_mov_b32 s4, 0x12341234
; SI-NEXT:    s_mov_b64 s[2:3], s[10:11]
; SI-NEXT:    s_waitcnt vmcnt(1)
; SI-NEXT:    v_lshlrev_b32_e32 v2, 4, v2
; SI-NEXT:    v_lshl_b32_e32 v2, 0xffff, v2
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_bfi_b32 v2, v2, s4, v3
; SI-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
; SI-NEXT:    s_endpgm
;
; VI-LABEL: v_insertelement_v2bf16_dynamic_vgpr:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    s_load_dwordx2 s[4:5], s[8:9], 0x10
; VI-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
; VI-NEXT:    s_add_i32 s12, s12, s17
; VI-NEXT:    s_mov_b32 flat_scratch_lo, s13
; VI-NEXT:    s_lshr_b32 flat_scratch_hi, s12, 8
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v1, s5
; VI-NEXT:    v_add_u32_e32 v0, vcc, s4, v2
; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT:    v_mov_b32_e32 v3, s3
; VI-NEXT:    flat_load_dword v4, v[0:1]
; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v3, vcc
; VI-NEXT:    flat_load_dword v3, v[0:1]
; VI-NEXT:    s_mov_b32 s2, 0xffff
; VI-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    s_mov_b32 s0, 0x12341234
; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT:    s_waitcnt vmcnt(1)
; VI-NEXT:    v_lshlrev_b32_e32 v2, 4, v4
; VI-NEXT:    v_lshlrev_b32_e64 v2, v2, s2
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_bfi_b32 v2, v2, s0, v3
; VI-NEXT:    flat_store_dword v[0:1], v2
; VI-NEXT:    s_endpgm
;
; GFX900-LABEL: v_insertelement_v2bf16_dynamic_vgpr:
; GFX900:       ; %bb.0:
; GFX900-NEXT:    s_load_dwordx2 s[4:5], s[8:9], 0x10
; GFX900-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; GFX900-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
; GFX900-NEXT:    global_load_dword v1, v0, s[4:5]
; GFX900-NEXT:    global_load_dword v2, v0, s[2:3]
; GFX900-NEXT:    s_mov_b32 s2, 0xffff
; GFX900-NEXT:    s_waitcnt vmcnt(1)
; GFX900-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
; GFX900-NEXT:    v_lshlrev_b32_e64 v1, v1, s2
; GFX900-NEXT:    s_mov_b32 s2, 0x12341234
; GFX900-NEXT:    s_waitcnt vmcnt(0)
; GFX900-NEXT:    v_bfi_b32 v1, v1, s2, v2
; GFX900-NEXT:    global_store_dword v0, v1, s[0:1]
; GFX900-NEXT:    s_endpgm
;
; GFX942-LABEL: v_insertelement_v2bf16_dynamic_vgpr:
; GFX942:       ; %bb.0:
; GFX942-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x10
; GFX942-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
; GFX942-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX942-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
; GFX942-NEXT:    s_waitcnt lgkmcnt(0)
; GFX942-NEXT:    global_load_dword v1, v0, s[6:7]
; GFX942-NEXT:    global_load_dword v2, v0, s[2:3]
; GFX942-NEXT:    s_mov_b32 s2, 0xffff
; GFX942-NEXT:    s_waitcnt vmcnt(1)
; GFX942-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
; GFX942-NEXT:    v_lshlrev_b32_e64 v1, v1, s2
; GFX942-NEXT:    s_mov_b32 s2, 0x12341234
; GFX942-NEXT:    s_waitcnt vmcnt(0)
; GFX942-NEXT:    v_bfi_b32 v1, v1, s2, v2
; GFX942-NEXT:    global_store_dword v0, v1, s[0:1]
; GFX942-NEXT:    s_endpgm
;
; GFX1250-LABEL: v_insertelement_v2bf16_dynamic_vgpr:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_clause 0x1
; GFX1250-NEXT:    s_load_b64 s[6:7], s[4:5], 0x10
; GFX1250-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX1250-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    s_clause 0x1
; GFX1250-NEXT:    global_load_b32 v1, v0, s[6:7] scale_offset
; GFX1250-NEXT:    global_load_b32 v2, v0, s[2:3] scale_offset
; GFX1250-NEXT:    s_wait_loadcnt 0x1
; GFX1250-NEXT:    v_lshlrev_b32_e32 v1, 4, v1
; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
; GFX1250-NEXT:    v_lshlrev_b32_e64 v1, v1, 0xffff
; GFX1250-NEXT:    s_wait_loadcnt 0x0
; GFX1250-NEXT:    v_bitop3_b32 v1, 0x12341234, v2, v1 bitop3:0xe4
; GFX1250-NEXT:    global_store_b32 v0, v1, s[0:1] scale_offset
; GFX1250-NEXT:    s_endpgm
  %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
  %tid.ext = sext i32 %tid to i64
  %in.gep = getelementptr inbounds <2 x bfloat>, ptr addrspace(1) %in, i64 %tid.ext
  %idx.gep = getelementptr inbounds i32, ptr addrspace(1) %idx.ptr, i64 %tid.ext
  %out.gep = getelementptr inbounds <2 x bfloat>, ptr addrspace(1) %out, i64 %tid.ext
  %idx = load i32, ptr addrspace(1) %idx.gep
  %vec = load <2 x bfloat>, ptr addrspace(1) %in.gep
  %vecins = insertelement <2 x bfloat> %vec, bfloat 0xR1234, i32 %idx
  store <2 x bfloat> %vecins, ptr addrspace(1) %out.gep
  ret void
}

define amdgpu_kernel void @v_insertelement_v4bf16_0(ptr addrspace(1) %out, ptr addrspace(1) %in, [8 x i32], i32 %val) #0 {
; SI-LABEL: v_insertelement_v4bf16_0:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; SI-NEXT:    s_mov_b32 s7, 0x100f000
; SI-NEXT:    s_mov_b32 s6, 0
; SI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; SI-NEXT:    v_mov_b32_e32 v1, 0
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b64 s[4:5], s[2:3]
; SI-NEXT:    buffer_load_dwordx2 v[2:3], v[0:1], s[4:7], 0 addr64
; SI-NEXT:    s_load_dword s8, s[8:9], 0xc
; SI-NEXT:    s_mov_b32 s4, 0xffff
; SI-NEXT:    s_mov_b64 s[2:3], s[6:7]
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    v_mov_b32_e32 v4, s8
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_bfi_b32 v2, s4, v4, v2
; SI-NEXT:    buffer_store_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64
; SI-NEXT:    s_endpgm
;
; VI-LABEL: v_insertelement_v4bf16_0:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    s_load_dword s4, s[8:9], 0x30
; VI-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
; VI-NEXT:    s_add_i32 s12, s12, s17
; VI-NEXT:    s_mov_b32 flat_scratch_lo, s13
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
; VI-NEXT:    s_lshr_b32 flat_scratch_hi, s12, 8
; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
; VI-NEXT:    v_mov_b32_e32 v3, s1
; VI-NEXT:    v_add_u32_e32 v2, vcc, s0, v2
; VI-NEXT:    v_mov_b32_e32 v4, 0x3020504
; VI-NEXT:    v_addc_u32_e32 v3, vcc, 0, v3, vcc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_perm_b32 v0, s4, v0, v4
; VI-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
; VI-NEXT:    s_endpgm
;
; GFX900-LABEL: v_insertelement_v4bf16_0:
; GFX900:       ; %bb.0:
; GFX900-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; GFX900-NEXT:    s_load_dword s4, s[8:9], 0x30
; GFX900-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
; GFX900-NEXT:    global_load_dwordx2 v[0:1], v2, s[2:3]
; GFX900-NEXT:    s_mov_b32 s2, 0xffff
; GFX900-NEXT:    v_mov_b32_e32 v3, s4
; GFX900-NEXT:    s_waitcnt vmcnt(0)
; GFX900-NEXT:    v_bfi_b32 v0, s2, v3, v0
; GFX900-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX900-NEXT:    s_endpgm
;
; GFX942-LABEL: v_insertelement_v4bf16_0:
; GFX942:       ; %bb.0:
; GFX942-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
; GFX942-NEXT:    s_load_dword s6, s[4:5], 0x30
; GFX942-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX942-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
; GFX942-NEXT:    s_waitcnt lgkmcnt(0)
; GFX942-NEXT:    global_load_dwordx2 v[0:1], v2, s[2:3]
; GFX942-NEXT:    s_mov_b32 s2, 0xffff
; GFX942-NEXT:    v_mov_b32_e32 v3, s6
; GFX942-NEXT:    s_waitcnt vmcnt(0)
; GFX942-NEXT:    v_bfi_b32 v0, s2, v3, v0
; GFX942-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX942-NEXT:    s_endpgm
;
; GFX1250-LABEL: v_insertelement_v4bf16_0:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX1250-NEXT:    v_and_b32_e32 v2, 0x3ff, v0
; GFX1250-NEXT:    s_wait_xcnt 0x0
; GFX1250-NEXT:    s_load_b32 s4, s[4:5], 0x30
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    global_load_b64 v[0:1], v2, s[2:3] scale_offset
; GFX1250-NEXT:    s_wait_loadcnt 0x0
; GFX1250-NEXT:    v_bfi_b32 v0, 0xffff, s4, v0
; GFX1250-NEXT:    global_store_b64 v2, v[0:1], s[0:1] scale_offset
; GFX1250-NEXT:    s_endpgm
  %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
  %tid.ext = sext i32 %tid to i64
  %in.gep = getelementptr inbounds <4 x bfloat>, ptr addrspace(1) %in, i64 %tid.ext
  %out.gep = getelementptr inbounds <4 x bfloat>, ptr addrspace(1) %out, i64 %tid.ext
  %vec = load <4 x bfloat>, ptr addrspace(1) %in.gep
  %val.trunc = trunc i32 %val to i16
  %val.cvt = bitcast i16 %val.trunc to bfloat
  %vecins = insertelement <4 x bfloat> %vec, bfloat %val.cvt, i32 0
  store <4 x bfloat> %vecins, ptr addrspace(1) %out.gep
  ret void
}

define amdgpu_kernel void @v_insertelement_v4bf16_1(ptr addrspace(1) %out, ptr addrspace(1) %in, i32 %val) #0 {
; SI-LABEL: v_insertelement_v4bf16_1:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; SI-NEXT:    s_mov_b32 s7, 0x100f000
; SI-NEXT:    s_mov_b32 s6, 0
; SI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; SI-NEXT:    v_mov_b32_e32 v1, 0
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b64 s[4:5], s[2:3]
; SI-NEXT:    buffer_load_dwordx2 v[2:3], v[0:1], s[4:7], 0 addr64
; SI-NEXT:    s_load_dword s8, s[8:9], 0x4
; SI-NEXT:    s_mov_b64 s[2:3], s[6:7]
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_lshl_b32 s4, s8, 16
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_and_b32_e32 v2, 0xffff, v2
; SI-NEXT:    v_or_b32_e32 v2, s4, v2
; SI-NEXT:    buffer_store_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64
; SI-NEXT:    s_endpgm
;
; VI-LABEL: v_insertelement_v4bf16_1:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    s_load_dword s4, s[8:9], 0x10
; VI-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
; VI-NEXT:    s_add_i32 s12, s12, s17
; VI-NEXT:    s_mov_b32 flat_scratch_lo, s13
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
; VI-NEXT:    s_lshr_b32 flat_scratch_hi, s12, 8
; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
; VI-NEXT:    v_mov_b32_e32 v3, s1
; VI-NEXT:    v_add_u32_e32 v2, vcc, s0, v2
; VI-NEXT:    v_mov_b32_e32 v4, 0x1000504
; VI-NEXT:    v_addc_u32_e32 v3, vcc, 0, v3, vcc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_perm_b32 v0, v0, s4, v4
; VI-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
; VI-NEXT:    s_endpgm
;
; GFX900-LABEL: v_insertelement_v4bf16_1:
; GFX900:       ; %bb.0:
; GFX900-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; GFX900-NEXT:    s_load_dword s4, s[8:9], 0x10
; GFX900-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
; GFX900-NEXT:    v_mov_b32_e32 v3, 0x5040100
; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
; GFX900-NEXT:    global_load_dwordx2 v[0:1], v2, s[2:3]
; GFX900-NEXT:    s_waitcnt vmcnt(0)
; GFX900-NEXT:    v_perm_b32 v0, s4, v0, v3
; GFX900-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX900-NEXT:    s_endpgm
;
; GFX942-LABEL: v_insertelement_v4bf16_1:
; GFX942:       ; %bb.0:
; GFX942-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
; GFX942-NEXT:    s_load_dword s6, s[4:5], 0x10
; GFX942-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX942-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
; GFX942-NEXT:    v_mov_b32_e32 v3, 0x5040100
; GFX942-NEXT:    s_waitcnt lgkmcnt(0)
; GFX942-NEXT:    global_load_dwordx2 v[0:1], v2, s[2:3]
; GFX942-NEXT:    s_waitcnt vmcnt(0)
; GFX942-NEXT:    v_perm_b32 v0, s6, v0, v3
; GFX942-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX942-NEXT:    s_endpgm
;
; GFX1250-LABEL: v_insertelement_v4bf16_1:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX1250-NEXT:    v_and_b32_e32 v2, 0x3ff, v0
; GFX1250-NEXT:    s_wait_xcnt 0x0
; GFX1250-NEXT:    s_load_b32 s4, s[4:5], 0x10
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    global_load_b64 v[0:1], v2, s[2:3] scale_offset
; GFX1250-NEXT:    s_wait_loadcnt 0x0
; GFX1250-NEXT:    v_perm_b32 v0, s4, v0, 0x5040100
; GFX1250-NEXT:    global_store_b64 v2, v[0:1], s[0:1] scale_offset
; GFX1250-NEXT:    s_endpgm
  %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
  %tid.ext = sext i32 %tid to i64
  %in.gep = getelementptr inbounds <4 x bfloat>, ptr addrspace(1) %in, i64 %tid.ext
  %out.gep = getelementptr inbounds <4 x bfloat>, ptr addrspace(1) %out, i64 %tid.ext
  %vec = load <4 x bfloat>, ptr addrspace(1) %in.gep
  %val.trunc = trunc i32 %val to i16
  %val.cvt = bitcast i16 %val.trunc to bfloat
  %vecins = insertelement <4 x bfloat> %vec, bfloat %val.cvt, i32 1
  store <4 x bfloat> %vecins, ptr addrspace(1) %out.gep
  ret void
}

define amdgpu_kernel void @v_insertelement_v4bf16_2(ptr addrspace(1) %out, ptr addrspace(1) %in, [8 x i32], i32 %val) #0 {
; SI-LABEL: v_insertelement_v4bf16_2:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; SI-NEXT:    s_mov_b32 s7, 0x100f000
; SI-NEXT:    s_mov_b32 s6, 0
; SI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; SI-NEXT:    v_mov_b32_e32 v1, 0
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b64 s[4:5], s[2:3]
; SI-NEXT:    buffer_load_dwordx2 v[2:3], v[0:1], s[4:7], 0 addr64
; SI-NEXT:    s_load_dword s8, s[8:9], 0xc
; SI-NEXT:    s_mov_b32 s4, 0xffff
; SI-NEXT:    s_mov_b64 s[2:3], s[6:7]
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    v_mov_b32_e32 v4, s8
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_bfi_b32 v3, s4, v4, v3
; SI-NEXT:    buffer_store_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64
; SI-NEXT:    s_endpgm
;
; VI-LABEL: v_insertelement_v4bf16_2:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    s_load_dword s4, s[8:9], 0x30
; VI-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
; VI-NEXT:    s_add_i32 s12, s12, s17
; VI-NEXT:    s_mov_b32 flat_scratch_lo, s13
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
; VI-NEXT:    s_lshr_b32 flat_scratch_hi, s12, 8
; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
; VI-NEXT:    v_mov_b32_e32 v3, s1
; VI-NEXT:    v_add_u32_e32 v2, vcc, s0, v2
; VI-NEXT:    v_mov_b32_e32 v4, 0x3020504
; VI-NEXT:    v_addc_u32_e32 v3, vcc, 0, v3, vcc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_perm_b32 v1, s4, v1, v4
; VI-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
; VI-NEXT:    s_endpgm
;
; GFX900-LABEL: v_insertelement_v4bf16_2:
; GFX900:       ; %bb.0:
; GFX900-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; GFX900-NEXT:    s_load_dword s4, s[8:9], 0x30
; GFX900-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
; GFX900-NEXT:    global_load_dwordx2 v[0:1], v2, s[2:3]
; GFX900-NEXT:    s_mov_b32 s2, 0xffff
; GFX900-NEXT:    v_mov_b32_e32 v3, s4
; GFX900-NEXT:    s_waitcnt vmcnt(0)
; GFX900-NEXT:    v_bfi_b32 v1, s2, v3, v1
; GFX900-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX900-NEXT:    s_endpgm
;
; GFX942-LABEL: v_insertelement_v4bf16_2:
; GFX942:       ; %bb.0:
; GFX942-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
; GFX942-NEXT:    s_load_dword s6, s[4:5], 0x30
; GFX942-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX942-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
; GFX942-NEXT:    s_waitcnt lgkmcnt(0)
; GFX942-NEXT:    global_load_dwordx2 v[0:1], v2, s[2:3]
; GFX942-NEXT:    s_mov_b32 s2, 0xffff
; GFX942-NEXT:    v_mov_b32_e32 v3, s6
; GFX942-NEXT:    s_waitcnt vmcnt(0)
; GFX942-NEXT:    v_bfi_b32 v1, s2, v3, v1
; GFX942-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX942-NEXT:    s_endpgm
;
; GFX1250-LABEL: v_insertelement_v4bf16_2:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX1250-NEXT:    v_and_b32_e32 v2, 0x3ff, v0
; GFX1250-NEXT:    s_wait_xcnt 0x0
; GFX1250-NEXT:    s_load_b32 s4, s[4:5], 0x30
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    global_load_b64 v[0:1], v2, s[2:3] scale_offset
; GFX1250-NEXT:    s_wait_loadcnt 0x0
; GFX1250-NEXT:    v_bfi_b32 v1, 0xffff, s4, v1
; GFX1250-NEXT:    global_store_b64 v2, v[0:1], s[0:1] scale_offset
; GFX1250-NEXT:    s_endpgm
  %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
  %tid.ext = sext i32 %tid to i64
  %in.gep = getelementptr inbounds <4 x bfloat>, ptr addrspace(1) %in, i64 %tid.ext
  %out.gep = getelementptr inbounds <4 x bfloat>, ptr addrspace(1) %out, i64 %tid.ext
  %vec = load <4 x bfloat>, ptr addrspace(1) %in.gep
  %val.trunc = trunc i32 %val to i16
  %val.cvt = bitcast i16 %val.trunc to bfloat
  %vecins = insertelement <4 x bfloat> %vec, bfloat %val.cvt, i32 2
  store <4 x bfloat> %vecins, ptr addrspace(1) %out.gep
  ret void
}

define amdgpu_kernel void @v_insertelement_v4bf16_3(ptr addrspace(1) %out, ptr addrspace(1) %in, i32 %val) #0 {
; SI-LABEL: v_insertelement_v4bf16_3:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; SI-NEXT:    s_mov_b32 s7, 0x100f000
; SI-NEXT:    s_mov_b32 s6, 0
; SI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; SI-NEXT:    v_mov_b32_e32 v1, 0
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b64 s[4:5], s[2:3]
; SI-NEXT:    buffer_load_dwordx2 v[2:3], v[0:1], s[4:7], 0 addr64
; SI-NEXT:    s_load_dword s8, s[8:9], 0x4
; SI-NEXT:    s_mov_b64 s[2:3], s[6:7]
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_lshl_b32 s4, s8, 16
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_and_b32_e32 v3, 0xffff, v3
; SI-NEXT:    v_or_b32_e32 v3, s4, v3
; SI-NEXT:    buffer_store_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64
; SI-NEXT:    s_endpgm
;
; VI-LABEL: v_insertelement_v4bf16_3:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    s_load_dword s4, s[8:9], 0x10
; VI-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
; VI-NEXT:    s_add_i32 s12, s12, s17
; VI-NEXT:    s_mov_b32 flat_scratch_lo, s13
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
; VI-NEXT:    s_lshr_b32 flat_scratch_hi, s12, 8
; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
; VI-NEXT:    v_mov_b32_e32 v3, s1
; VI-NEXT:    v_add_u32_e32 v2, vcc, s0, v2
; VI-NEXT:    v_mov_b32_e32 v4, 0x1000504
; VI-NEXT:    v_addc_u32_e32 v3, vcc, 0, v3, vcc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_perm_b32 v1, v1, s4, v4
; VI-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
; VI-NEXT:    s_endpgm
;
; GFX900-LABEL: v_insertelement_v4bf16_3:
; GFX900:       ; %bb.0:
; GFX900-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; GFX900-NEXT:    s_load_dword s4, s[8:9], 0x10
; GFX900-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
; GFX900-NEXT:    v_mov_b32_e32 v3, 0x5040100
; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
; GFX900-NEXT:    global_load_dwordx2 v[0:1], v2, s[2:3]
; GFX900-NEXT:    s_waitcnt vmcnt(0)
; GFX900-NEXT:    v_perm_b32 v1, s4, v1, v3
; GFX900-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX900-NEXT:    s_endpgm
;
; GFX942-LABEL: v_insertelement_v4bf16_3:
; GFX942:       ; %bb.0:
; GFX942-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
; GFX942-NEXT:    s_load_dword s6, s[4:5], 0x10
; GFX942-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX942-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
; GFX942-NEXT:    v_mov_b32_e32 v3, 0x5040100
; GFX942-NEXT:    s_waitcnt lgkmcnt(0)
; GFX942-NEXT:    global_load_dwordx2 v[0:1], v2, s[2:3]
; GFX942-NEXT:    s_waitcnt vmcnt(0)
; GFX942-NEXT:    v_perm_b32 v1, s6, v1, v3
; GFX942-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX942-NEXT:    s_endpgm
;
; GFX1250-LABEL: v_insertelement_v4bf16_3:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX1250-NEXT:    v_and_b32_e32 v2, 0x3ff, v0
; GFX1250-NEXT:    s_wait_xcnt 0x0
; GFX1250-NEXT:    s_load_b32 s4, s[4:5], 0x10
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    global_load_b64 v[0:1], v2, s[2:3] scale_offset
; GFX1250-NEXT:    s_wait_loadcnt 0x0
; GFX1250-NEXT:    v_perm_b32 v1, s4, v1, 0x5040100
; GFX1250-NEXT:    global_store_b64 v2, v[0:1], s[0:1] scale_offset
; GFX1250-NEXT:    s_endpgm
  %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
  %tid.ext = sext i32 %tid to i64
  %in.gep = getelementptr inbounds <4 x bfloat>, ptr addrspace(1) %in, i64 %tid.ext
  %out.gep = getelementptr inbounds <4 x bfloat>, ptr addrspace(1) %out, i64 %tid.ext
  %vec = load <4 x bfloat>, ptr addrspace(1) %in.gep
  %val.trunc = trunc i32 %val to i16
  %val.cvt = bitcast i16 %val.trunc to bfloat
  %vecins = insertelement <4 x bfloat> %vec, bfloat %val.cvt, i32 3
  store <4 x bfloat> %vecins, ptr addrspace(1) %out.gep
  ret void
}

define amdgpu_kernel void @v_insertelement_v4bf16_dynamic_sgpr(ptr addrspace(1) %out, ptr addrspace(1) %in, i32 %val, i32 %idxval) #0 {
; SI-LABEL: v_insertelement_v4bf16_dynamic_sgpr:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; SI-NEXT:    s_mov_b32 s7, 0x100f000
; SI-NEXT:    s_mov_b32 s6, 0
; SI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
; SI-NEXT:    v_mov_b32_e32 v1, 0
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b64 s[4:5], s[2:3]
; SI-NEXT:    buffer_load_dwordx2 v[2:3], v[0:1], s[4:7], 0 addr64
; SI-NEXT:    s_load_dwordx2 s[8:9], s[8:9], 0x4
; SI-NEXT:    s_mov_b64 s[2:3], s[6:7]
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_lshl_b32 s4, s8, 16
; SI-NEXT:    s_and_b32 s5, s8, 0xffff
; SI-NEXT:    s_lshl_b32 s6, s9, 4
; SI-NEXT:    s_or_b32 s7, s5, s4
; SI-NEXT:    s_lshl_b64 s[4:5], 0xffff, s6
; SI-NEXT:    v_mov_b32_e32 v4, s7
; SI-NEXT:    v_mov_b32_e32 v5, s7
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_bfi_b32 v3, s5, v4, v3
; SI-NEXT:    v_bfi_b32 v2, s4, v5, v2
; SI-NEXT:    buffer_store_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64
; SI-NEXT:    s_endpgm
;
; VI-LABEL: v_insertelement_v4bf16_dynamic_sgpr:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    s_load_dwordx2 s[4:5], s[8:9], 0x10
; VI-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
; VI-NEXT:    s_add_i32 s12, s12, s17
; VI-NEXT:    s_mov_b32 flat_scratch_lo, s13
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
; VI-NEXT:    s_lshr_b32 flat_scratch_hi, s12, 8
; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
; VI-NEXT:    v_mov_b32_e32 v3, s1
; VI-NEXT:    s_lshl_b32 s1, s4, 16
; VI-NEXT:    s_and_b32 s2, s4, 0xffff
; VI-NEXT:    s_lshl_b32 s3, s5, 4
; VI-NEXT:    s_or_b32 s2, s2, s1
; VI-NEXT:    v_add_u32_e32 v2, vcc, s0, v2
; VI-NEXT:    s_lshl_b64 s[0:1], 0xffff, s3
; VI-NEXT:    v_mov_b32_e32 v4, s2
; VI-NEXT:    v_mov_b32_e32 v5, s2
; VI-NEXT:    v_addc_u32_e32 v3, vcc, 0, v3, vcc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_bfi_b32 v1, s1, v4, v1
; VI-NEXT:    v_bfi_b32 v0, s0, v5, v0
; VI-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
; VI-NEXT:    s_endpgm
;
; GFX900-LABEL: v_insertelement_v4bf16_dynamic_sgpr:
; GFX900:       ; %bb.0:
; GFX900-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; GFX900-NEXT:    s_load_dwordx2 s[4:5], s[8:9], 0x10
; GFX900-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
; GFX900-NEXT:    global_load_dwordx2 v[0:1], v2, s[2:3]
; GFX900-NEXT:    s_lshl_b32 s2, s5, 4
; GFX900-NEXT:    s_pack_ll_b32_b16 s4, s4, s4
; GFX900-NEXT:    s_lshl_b64 s[2:3], 0xffff, s2
; GFX900-NEXT:    v_mov_b32_e32 v3, s4
; GFX900-NEXT:    v_mov_b32_e32 v4, s4
; GFX900-NEXT:    s_waitcnt vmcnt(0)
; GFX900-NEXT:    v_bfi_b32 v1, s3, v3, v1
; GFX900-NEXT:    v_bfi_b32 v0, s2, v4, v0
; GFX900-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX900-NEXT:    s_endpgm
;
; GFX942-LABEL: v_insertelement_v4bf16_dynamic_sgpr:
; GFX942:       ; %bb.0:
; GFX942-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
; GFX942-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x10
; GFX942-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX942-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
; GFX942-NEXT:    s_waitcnt lgkmcnt(0)
; GFX942-NEXT:    global_load_dwordx2 v[0:1], v2, s[2:3]
; GFX942-NEXT:    s_lshl_b32 s2, s7, 4
; GFX942-NEXT:    s_pack_ll_b32_b16 s4, s6, s6
; GFX942-NEXT:    s_lshl_b64 s[2:3], 0xffff, s2
; GFX942-NEXT:    v_mov_b32_e32 v3, s4
; GFX942-NEXT:    v_mov_b32_e32 v4, s4
; GFX942-NEXT:    s_waitcnt vmcnt(0)
; GFX942-NEXT:    v_bfi_b32 v1, s3, v3, v1
; GFX942-NEXT:    v_bfi_b32 v0, s2, v4, v0
; GFX942-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
; GFX942-NEXT:    s_endpgm
;
; GFX1250-LABEL: v_insertelement_v4bf16_dynamic_sgpr:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX1250-NEXT:    v_and_b32_e32 v2, 0x3ff, v0
; GFX1250-NEXT:    s_wait_xcnt 0x0
; GFX1250-NEXT:    s_load_b64 s[4:5], s[4:5], 0x10
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    global_load_b64 v[0:1], v2, s[2:3] scale_offset
; GFX1250-NEXT:    s_wait_xcnt 0x0
; GFX1250-NEXT:    s_lshl_b32 s2, s5, 4
; GFX1250-NEXT:    s_pack_ll_b32_b16 s4, s4, s4
; GFX1250-NEXT:    s_lshl_b64 s[2:3], 0xffff, s2
; GFX1250-NEXT:    s_wait_loadcnt 0x0
; GFX1250-NEXT:    v_bfi_b32 v1, s3, s4, v1
; GFX1250-NEXT:    v_bfi_b32 v0, s2, s4, v0
; GFX1250-NEXT:    global_store_b64 v2, v[0:1], s[0:1] scale_offset
; GFX1250-NEXT:    s_endpgm
  %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
  %tid.ext = sext i32 %tid to i64
  %in.gep = getelementptr inbounds <4 x bfloat>, ptr addrspace(1) %in, i64 %tid.ext
  %out.gep = getelementptr inbounds <4 x bfloat>, ptr addrspace(1) %out, i64 %tid.ext
  %vec = load <4 x bfloat>, ptr addrspace(1) %in.gep
  %val.trunc = trunc i32 %val to i16
  %val.cvt = bitcast i16 %val.trunc to bfloat
  %vecins = insertelement <4 x bfloat> %vec, bfloat %val.cvt, i32 %idxval
  store <4 x bfloat> %vecins, ptr addrspace(1) %out.gep
  ret void
}

define amdgpu_kernel void @v_insertelement_v8bf16_3(ptr addrspace(1) %out, ptr addrspace(1) %in, i32 %val) {
; SI-LABEL: v_insertelement_v8bf16_3:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; SI-NEXT:    s_mov_b32 s7, 0x100f000
; SI-NEXT:    s_mov_b32 s6, 0
; SI-NEXT:    v_lshlrev_b32_e32 v4, 4, v0
; SI-NEXT:    v_mov_b32_e32 v5, 0
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b64 s[4:5], s[2:3]
; SI-NEXT:    buffer_load_dwordx4 v[0:3], v[4:5], s[4:7], 0 addr64
; SI-NEXT:    s_load_dword s8, s[8:9], 0x4
; SI-NEXT:    s_mov_b64 s[2:3], s[6:7]
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_lshl_b32 s4, s8, 16
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_and_b32_e32 v1, 0xffff, v1
; SI-NEXT:    v_or_b32_e32 v1, s4, v1
; SI-NEXT:    buffer_store_dwordx4 v[0:3], v[4:5], s[0:3], 0 addr64
; SI-NEXT:    s_endpgm
;
; VI-LABEL: v_insertelement_v8bf16_3:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    s_load_dword s4, s[8:9], 0x10
; VI-NEXT:    v_lshlrev_b32_e32 v4, 4, v0
; VI-NEXT:    s_add_i32 s12, s12, s17
; VI-NEXT:    s_mov_b32 flat_scratch_lo, s13
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v4
; VI-NEXT:    s_lshr_b32 flat_scratch_hi, s12, 8
; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT:    flat_load_dwordx4 v[0:3], v[0:1]
; VI-NEXT:    v_add_u32_e32 v4, vcc, s0, v4
; VI-NEXT:    s_lshl_b32 s0, s4, 16
; VI-NEXT:    v_mov_b32_e32 v5, s1
; VI-NEXT:    v_mov_b32_e32 v6, s0
; VI-NEXT:    v_addc_u32_e32 v5, vcc, 0, v5, vcc
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_or_b32_sdwa v1, v1, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
; VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
; VI-NEXT:    s_endpgm
;
; GFX900-LABEL: v_insertelement_v8bf16_3:
; GFX900:       ; %bb.0:
; GFX900-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; GFX900-NEXT:    s_load_dword s4, s[8:9], 0x10
; GFX900-NEXT:    v_lshlrev_b32_e32 v4, 4, v0
; GFX900-NEXT:    v_mov_b32_e32 v5, 0x5040100
; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
; GFX900-NEXT:    global_load_dwordx4 v[0:3], v4, s[2:3]
; GFX900-NEXT:    s_waitcnt vmcnt(0)
; GFX900-NEXT:    v_perm_b32 v1, s4, v1, v5
; GFX900-NEXT:    global_store_dwordx4 v4, v[0:3], s[0:1]
; GFX900-NEXT:    s_endpgm
;
; GFX942-LABEL: v_insertelement_v8bf16_3:
; GFX942:       ; %bb.0:
; GFX942-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
; GFX942-NEXT:    s_load_dword s6, s[4:5], 0x10
; GFX942-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX942-NEXT:    v_lshlrev_b32_e32 v4, 4, v0
; GFX942-NEXT:    v_mov_b32_e32 v5, 0x5040100
; GFX942-NEXT:    s_waitcnt lgkmcnt(0)
; GFX942-NEXT:    global_load_dwordx4 v[0:3], v4, s[2:3]
; GFX942-NEXT:    s_waitcnt vmcnt(0)
; GFX942-NEXT:    v_perm_b32 v1, s6, v1, v5
; GFX942-NEXT:    global_store_dwordx4 v4, v[0:3], s[0:1]
; GFX942-NEXT:    s_endpgm
;
; GFX1250-LABEL: v_insertelement_v8bf16_3:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX1250-NEXT:    v_and_b32_e32 v4, 0x3ff, v0
; GFX1250-NEXT:    s_wait_xcnt 0x0
; GFX1250-NEXT:    s_load_b32 s4, s[4:5], 0x10
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    global_load_b128 v[0:3], v4, s[2:3] scale_offset
; GFX1250-NEXT:    s_wait_loadcnt 0x0
; GFX1250-NEXT:    v_perm_b32 v1, s4, v1, 0x5040100
; GFX1250-NEXT:    global_store_b128 v4, v[0:3], s[0:1] scale_offset
; GFX1250-NEXT:    s_endpgm
  %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
  %tid.ext = sext i32 %tid to i64
  %in.gep = getelementptr inbounds <8 x bfloat>, ptr addrspace(1) %in, i64 %tid.ext
  %out.gep = getelementptr inbounds <8 x bfloat>, ptr addrspace(1) %out, i64 %tid.ext
  %vec = load <8 x bfloat>, ptr addrspace(1) %in.gep
  %val.trunc = trunc i32 %val to i16
  %val.cvt = bitcast i16 %val.trunc to bfloat
  %vecins = insertelement <8 x bfloat> %vec, bfloat %val.cvt, i32 3
  store <8 x bfloat> %vecins, ptr addrspace(1) %out.gep
  ret void
}

define amdgpu_kernel void @v_insertelement_v8bf16_dynamic(ptr addrspace(1) %out, ptr addrspace(1) %in, i32 %val, i32 %n) {
; SI-LABEL: v_insertelement_v8bf16_dynamic:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; SI-NEXT:    s_mov_b32 s7, 0x100f000
; SI-NEXT:    s_mov_b32 s6, 0
; SI-NEXT:    v_lshlrev_b32_e32 v4, 4, v0
; SI-NEXT:    v_mov_b32_e32 v5, 0
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b64 s[4:5], s[2:3]
; SI-NEXT:    buffer_load_dwordx4 v[0:3], v[4:5], s[4:7], 0 addr64
; SI-NEXT:    s_load_dwordx2 s[8:9], s[8:9], 0x4
; SI-NEXT:    s_mov_b64 s[2:3], s[6:7]
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_cmp_eq_u32 s9, 6
; SI-NEXT:    v_mov_b32_e32 v6, s8
; SI-NEXT:    s_cselect_b64 vcc, -1, 0
; SI-NEXT:    s_cmp_eq_u32 s9, 7
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_cndmask_b32_e32 v7, v3, v6, vcc
; SI-NEXT:    v_lshrrev_b32_e32 v3, 16, v3
; SI-NEXT:    s_cselect_b64 vcc, -1, 0
; SI-NEXT:    s_cmp_eq_u32 s9, 4
; SI-NEXT:    v_cndmask_b32_e32 v3, v3, v6, vcc
; SI-NEXT:    s_cselect_b64 vcc, -1, 0
; SI-NEXT:    s_cmp_eq_u32 s9, 5
; SI-NEXT:    v_lshrrev_b32_e32 v8, 16, v2
; SI-NEXT:    v_cndmask_b32_e32 v2, v2, v6, vcc
; SI-NEXT:    s_cselect_b64 vcc, -1, 0
; SI-NEXT:    s_cmp_eq_u32 s9, 2
; SI-NEXT:    v_and_b32_e32 v7, 0xffff, v7
; SI-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
; SI-NEXT:    v_cndmask_b32_e32 v8, v8, v6, vcc
; SI-NEXT:    s_cselect_b64 vcc, -1, 0
; SI-NEXT:    s_cmp_eq_u32 s9, 3
; SI-NEXT:    v_lshrrev_b32_e32 v9, 16, v1
; SI-NEXT:    v_and_b32_e32 v2, 0xffff, v2
; SI-NEXT:    v_or_b32_e32 v3, v7, v3
; SI-NEXT:    v_lshlrev_b32_e32 v7, 16, v8
; SI-NEXT:    v_cndmask_b32_e32 v1, v1, v6, vcc
; SI-NEXT:    s_cselect_b64 vcc, -1, 0
; SI-NEXT:    s_cmp_eq_u32 s9, 0
; SI-NEXT:    v_or_b32_e32 v2, v2, v7
; SI-NEXT:    v_cndmask_b32_e32 v7, v9, v6, vcc
; SI-NEXT:    s_cselect_b64 vcc, -1, 0
; SI-NEXT:    s_cmp_eq_u32 s9, 1
; SI-NEXT:    v_lshrrev_b32_e32 v10, 16, v0
; SI-NEXT:    v_cndmask_b32_e32 v0, v0, v6, vcc
; SI-NEXT:    s_cselect_b64 vcc, -1, 0
; SI-NEXT:    v_cndmask_b32_e32 v6, v10, v6, vcc
; SI-NEXT:    v_and_b32_e32 v1, 0xffff, v1
; SI-NEXT:    v_lshlrev_b32_e32 v7, 16, v7
; SI-NEXT:    v_and_b32_e32 v0, 0xffff, v0
; SI-NEXT:    v_lshlrev_b32_e32 v6, 16, v6
; SI-NEXT:    v_or_b32_e32 v1, v1, v7
; SI-NEXT:    v_or_b32_e32 v0, v0, v6
; SI-NEXT:    buffer_store_dwordx4 v[0:3], v[4:5], s[0:3], 0 addr64
; SI-NEXT:    s_endpgm
;
; VI-LABEL: v_insertelement_v8bf16_dynamic:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    s_load_dwordx2 s[4:5], s[8:9], 0x10
; VI-NEXT:    v_lshlrev_b32_e32 v4, 4, v0
; VI-NEXT:    s_add_i32 s12, s12, s17
; VI-NEXT:    s_mov_b32 flat_scratch_lo, s13
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v4
; VI-NEXT:    s_lshr_b32 flat_scratch_hi, s12, 8
; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT:    flat_load_dwordx4 v[0:3], v[0:1]
; VI-NEXT:    v_mov_b32_e32 v5, s1
; VI-NEXT:    v_add_u32_e32 v4, vcc, s0, v4
; VI-NEXT:    s_cmp_eq_u32 s5, 6
; VI-NEXT:    v_addc_u32_e32 v5, vcc, 0, v5, vcc
; VI-NEXT:    v_mov_b32_e32 v6, s4
; VI-NEXT:    s_cselect_b64 vcc, -1, 0
; VI-NEXT:    s_cmp_eq_u32 s5, 7
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_cndmask_b32_e32 v7, v3, v6, vcc
; VI-NEXT:    v_lshrrev_b32_e32 v3, 16, v3
; VI-NEXT:    s_cselect_b64 vcc, -1, 0
; VI-NEXT:    s_cmp_eq_u32 s5, 4
; VI-NEXT:    v_cndmask_b32_e32 v3, v3, v6, vcc
; VI-NEXT:    s_cselect_b64 vcc, -1, 0
; VI-NEXT:    s_cmp_eq_u32 s5, 5
; VI-NEXT:    v_lshrrev_b32_e32 v8, 16, v2
; VI-NEXT:    v_cndmask_b32_e32 v2, v2, v6, vcc
; VI-NEXT:    s_cselect_b64 vcc, -1, 0
; VI-NEXT:    s_cmp_eq_u32 s5, 2
; VI-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
; VI-NEXT:    v_cndmask_b32_e32 v8, v8, v6, vcc
; VI-NEXT:    s_cselect_b64 vcc, -1, 0
; VI-NEXT:    s_cmp_eq_u32 s5, 3
; VI-NEXT:    v_lshrrev_b32_e32 v9, 16, v1
; VI-NEXT:    v_or_b32_sdwa v3, v7, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
; VI-NEXT:    v_lshlrev_b32_e32 v7, 16, v8
; VI-NEXT:    v_cndmask_b32_e32 v1, v1, v6, vcc
; VI-NEXT:    s_cselect_b64 vcc, -1, 0
; VI-NEXT:    s_cmp_eq_u32 s5, 0
; VI-NEXT:    v_or_b32_sdwa v2, v2, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
; VI-NEXT:    v_cndmask_b32_e32 v7, v9, v6, vcc
; VI-NEXT:    s_cselect_b64 vcc, -1, 0
; VI-NEXT:    s_cmp_eq_u32 s5, 1
; VI-NEXT:    v_cndmask_b32_e32 v8, v0, v6, vcc
; VI-NEXT:    s_cselect_b64 vcc, -1, 0
; VI-NEXT:    v_lshlrev_b32_e32 v7, 16, v7
; VI-NEXT:    v_cndmask_b32_sdwa v0, v0, v6, vcc dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; VI-NEXT:    v_or_b32_sdwa v1, v1, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
; VI-NEXT:    v_or_b32_sdwa v0, v8, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
; VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
; VI-NEXT:    s_endpgm
;
; GFX900-LABEL: v_insertelement_v8bf16_dynamic:
; GFX900:       ; %bb.0:
; GFX900-NEXT:    s_load_dwordx4 s[16:19], s[8:9], 0x0
; GFX900-NEXT:    s_load_dwordx2 s[12:13], s[8:9], 0x10
; GFX900-NEXT:    v_lshlrev_b32_e32 v4, 4, v0
; GFX900-NEXT:    s_mov_b32 s14, 0x5040100
; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
; GFX900-NEXT:    global_load_dwordx4 v[0:3], v4, s[18:19]
; GFX900-NEXT:    s_cmp_eq_u32 s13, 6
; GFX900-NEXT:    s_cselect_b64 vcc, -1, 0
; GFX900-NEXT:    s_cmp_eq_u32 s13, 7
; GFX900-NEXT:    s_cselect_b64 s[0:1], -1, 0
; GFX900-NEXT:    s_cmp_eq_u32 s13, 4
; GFX900-NEXT:    s_cselect_b64 s[2:3], -1, 0
; GFX900-NEXT:    s_cmp_eq_u32 s13, 5
; GFX900-NEXT:    s_cselect_b64 s[4:5], -1, 0
; GFX900-NEXT:    s_cmp_eq_u32 s13, 2
; GFX900-NEXT:    v_mov_b32_e32 v5, s12
; GFX900-NEXT:    s_cselect_b64 s[6:7], -1, 0
; GFX900-NEXT:    s_cmp_eq_u32 s13, 3
; GFX900-NEXT:    s_cselect_b64 s[8:9], -1, 0
; GFX900-NEXT:    s_cmp_eq_u32 s13, 0
; GFX900-NEXT:    s_cselect_b64 s[10:11], -1, 0
; GFX900-NEXT:    s_cmp_eq_u32 s13, 1
; GFX900-NEXT:    s_cselect_b64 s[12:13], -1, 0
; GFX900-NEXT:    s_waitcnt vmcnt(0)
; GFX900-NEXT:    v_cndmask_b32_e32 v6, v3, v5, vcc
; GFX900-NEXT:    s_mov_b64 vcc, s[0:1]
; GFX900-NEXT:    v_cndmask_b32_sdwa v3, v3, v5, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; GFX900-NEXT:    s_mov_b64 vcc, s[4:5]
; GFX900-NEXT:    v_cndmask_b32_e64 v7, v2, v5, s[2:3]
; GFX900-NEXT:    v_cndmask_b32_sdwa v2, v2, v5, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; GFX900-NEXT:    s_mov_b64 vcc, s[8:9]
; GFX900-NEXT:    v_cndmask_b32_e64 v8, v1, v5, s[6:7]
; GFX900-NEXT:    v_cndmask_b32_sdwa v1, v1, v5, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; GFX900-NEXT:    s_mov_b64 vcc, s[12:13]
; GFX900-NEXT:    v_cndmask_b32_e64 v9, v0, v5, s[10:11]
; GFX900-NEXT:    v_cndmask_b32_sdwa v0, v0, v5, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; GFX900-NEXT:    v_perm_b32 v3, v3, v6, s14
; GFX900-NEXT:    v_perm_b32 v2, v2, v7, s14
; GFX900-NEXT:    v_perm_b32 v1, v1, v8, s14
; GFX900-NEXT:    v_perm_b32 v0, v0, v9, s14
; GFX900-NEXT:    global_store_dwordx4 v4, v[0:3], s[16:17]
; GFX900-NEXT:    s_endpgm
;
; GFX942-LABEL: v_insertelement_v8bf16_dynamic:
; GFX942:       ; %bb.0:
; GFX942-NEXT:    s_load_dwordx4 s[16:19], s[4:5], 0x0
; GFX942-NEXT:    s_load_dwordx2 s[12:13], s[4:5], 0x10
; GFX942-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX942-NEXT:    v_lshlrev_b32_e32 v4, 4, v0
; GFX942-NEXT:    s_mov_b32 s14, 0x5040100
; GFX942-NEXT:    s_waitcnt lgkmcnt(0)
; GFX942-NEXT:    global_load_dwordx4 v[0:3], v4, s[18:19]
; GFX942-NEXT:    s_cmp_eq_u32 s13, 6
; GFX942-NEXT:    s_cselect_b64 vcc, -1, 0
; GFX942-NEXT:    s_cmp_eq_u32 s13, 7
; GFX942-NEXT:    s_cselect_b64 s[0:1], -1, 0
; GFX942-NEXT:    s_cmp_eq_u32 s13, 4
; GFX942-NEXT:    s_cselect_b64 s[2:3], -1, 0
; GFX942-NEXT:    s_cmp_eq_u32 s13, 5
; GFX942-NEXT:    s_cselect_b64 s[4:5], -1, 0
; GFX942-NEXT:    s_cmp_eq_u32 s13, 2
; GFX942-NEXT:    v_mov_b32_e32 v5, s12
; GFX942-NEXT:    s_cselect_b64 s[6:7], -1, 0
; GFX942-NEXT:    s_cmp_eq_u32 s13, 3
; GFX942-NEXT:    s_cselect_b64 s[8:9], -1, 0
; GFX942-NEXT:    s_cmp_eq_u32 s13, 0
; GFX942-NEXT:    s_cselect_b64 s[10:11], -1, 0
; GFX942-NEXT:    s_cmp_eq_u32 s13, 1
; GFX942-NEXT:    s_cselect_b64 s[12:13], -1, 0
; GFX942-NEXT:    s_waitcnt vmcnt(0)
; GFX942-NEXT:    v_cndmask_b32_e32 v6, v3, v5, vcc
; GFX942-NEXT:    s_mov_b64 vcc, s[0:1]
; GFX942-NEXT:    v_cndmask_b32_sdwa v3, v3, v5, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; GFX942-NEXT:    s_mov_b64 vcc, s[4:5]
; GFX942-NEXT:    v_cndmask_b32_e64 v7, v2, v5, s[2:3]
; GFX942-NEXT:    v_cndmask_b32_sdwa v2, v2, v5, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; GFX942-NEXT:    s_mov_b64 vcc, s[8:9]
; GFX942-NEXT:    v_cndmask_b32_e64 v8, v1, v5, s[6:7]
; GFX942-NEXT:    v_cndmask_b32_sdwa v1, v1, v5, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; GFX942-NEXT:    s_mov_b64 vcc, s[12:13]
; GFX942-NEXT:    v_cndmask_b32_e64 v9, v0, v5, s[10:11]
; GFX942-NEXT:    v_cndmask_b32_sdwa v0, v0, v5, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; GFX942-NEXT:    v_perm_b32 v3, v3, v6, s14
; GFX942-NEXT:    v_perm_b32 v2, v2, v7, s14
; GFX942-NEXT:    v_perm_b32 v1, v1, v8, s14
; GFX942-NEXT:    v_perm_b32 v0, v0, v9, s14
; GFX942-NEXT:    global_store_dwordx4 v4, v[0:3], s[16:17]
; GFX942-NEXT:    s_endpgm
;
; GFX1250-LABEL: v_insertelement_v8bf16_dynamic:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX1250-NEXT:    v_and_b32_e32 v4, 0x3ff, v0
; GFX1250-NEXT:    s_wait_xcnt 0x0
; GFX1250-NEXT:    s_load_b64 s[4:5], s[4:5], 0x10
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    global_load_b128 v[0:3], v4, s[2:3] scale_offset
; GFX1250-NEXT:    s_cmp_eq_u32 s5, 6
; GFX1250-NEXT:    s_wait_xcnt 0x0
; GFX1250-NEXT:    s_cselect_b32 s2, -1, 0
; GFX1250-NEXT:    s_cmp_eq_u32 s5, 7
; GFX1250-NEXT:    s_wait_loadcnt 0x0
; GFX1250-NEXT:    v_cndmask_b32_e64 v5, v3, s4, s2
; GFX1250-NEXT:    s_cselect_b32 s2, -1, 0
; GFX1250-NEXT:    s_cmp_eq_u32 s5, 4
; GFX1250-NEXT:    v_dual_lshrrev_b32 v3, 16, v3 :: v_dual_lshrrev_b32 v6, 16, v2
; GFX1250-NEXT:    s_cselect_b32 s3, -1, 0
; GFX1250-NEXT:    s_cmp_eq_u32 s5, 5
; GFX1250-NEXT:    v_cndmask_b32_e64 v2, v2, s4, s3
; GFX1250-NEXT:    s_cselect_b32 s3, -1, 0
; GFX1250-NEXT:    s_cmp_eq_u32 s5, 2
; GFX1250-NEXT:    v_dual_lshrrev_b32 v7, 16, v1 :: v_dual_lshrrev_b32 v8, 16, v0
; GFX1250-NEXT:    v_cndmask_b32_e64 v3, v3, s4, s2
; GFX1250-NEXT:    s_cselect_b32 s2, -1, 0
; GFX1250-NEXT:    s_cmp_eq_u32 s5, 3
; GFX1250-NEXT:    v_cndmask_b32_e64 v1, v1, s4, s2
; GFX1250-NEXT:    s_cselect_b32 s2, -1, 0
; GFX1250-NEXT:    s_cmp_eq_u32 s5, 0
; GFX1250-NEXT:    v_cndmask_b32_e64 v7, v7, s4, s2
; GFX1250-NEXT:    s_cselect_b32 s2, -1, 0
; GFX1250-NEXT:    s_cmp_eq_u32 s5, 1
; GFX1250-NEXT:    v_cndmask_b32_e64 v0, v0, s4, s2
; GFX1250-NEXT:    s_cselect_b32 s2, -1, 0
; GFX1250-NEXT:    v_cndmask_b32_e64 v6, v6, s4, s3
; GFX1250-NEXT:    v_cndmask_b32_e64 v8, v8, s4, s2
; GFX1250-NEXT:    v_perm_b32 v3, v3, v5, 0x5040100
; GFX1250-NEXT:    v_perm_b32 v1, v7, v1, 0x5040100
; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
; GFX1250-NEXT:    v_perm_b32 v2, v6, v2, 0x5040100
; GFX1250-NEXT:    v_perm_b32 v0, v8, v0, 0x5040100
; GFX1250-NEXT:    global_store_b128 v4, v[0:3], s[0:1] scale_offset
; GFX1250-NEXT:    s_endpgm
  %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
  %tid.ext = sext i32 %tid to i64
  %in.gep = getelementptr inbounds <8 x bfloat>, ptr addrspace(1) %in, i64 %tid.ext
  %out.gep = getelementptr inbounds <8 x bfloat>, ptr addrspace(1) %out, i64 %tid.ext
  %vec = load <8 x bfloat>, ptr addrspace(1) %in.gep
  %val.trunc = trunc i32 %val to i16
  %val.cvt = bitcast i16 %val.trunc to bfloat
  %vecins = insertelement <8 x bfloat> %vec, bfloat %val.cvt, i32 %n
  store <8 x bfloat> %vecins, ptr addrspace(1) %out.gep
  ret void
}

define amdgpu_kernel void @v_insertelement_v16bf16_3(ptr addrspace(1) %out, ptr addrspace(1) %in, i32 %val) {
; SI-LABEL: v_insertelement_v16bf16_3:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; SI-NEXT:    s_mov_b32 s7, 0x100f000
; SI-NEXT:    s_mov_b32 s6, 0
; SI-NEXT:    v_lshlrev_b32_e32 v8, 5, v0
; SI-NEXT:    v_mov_b32_e32 v9, 0
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b64 s[4:5], s[2:3]
; SI-NEXT:    buffer_load_dwordx4 v[0:3], v[8:9], s[4:7], 0 addr64
; SI-NEXT:    buffer_load_dwordx4 v[4:7], v[8:9], s[4:7], 0 addr64 offset:16
; SI-NEXT:    s_load_dword s8, s[8:9], 0x4
; SI-NEXT:    s_mov_b64 s[2:3], s[6:7]
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_lshl_b32 s4, s8, 16
; SI-NEXT:    s_waitcnt vmcnt(1)
; SI-NEXT:    v_and_b32_e32 v1, 0xffff, v1
; SI-NEXT:    v_or_b32_e32 v1, s4, v1
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    buffer_store_dwordx4 v[4:7], v[8:9], s[0:3], 0 addr64 offset:16
; SI-NEXT:    buffer_store_dwordx4 v[0:3], v[8:9], s[0:3], 0 addr64
; SI-NEXT:    s_endpgm
;
; VI-LABEL: v_insertelement_v16bf16_3:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    s_load_dword s4, s[8:9], 0x10
; VI-NEXT:    v_lshlrev_b32_e32 v8, 5, v0
; VI-NEXT:    s_add_i32 s12, s12, s17
; VI-NEXT:    s_mov_b32 flat_scratch_lo, s13
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v8
; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT:    v_add_u32_e32 v4, vcc, 16, v0
; VI-NEXT:    s_lshr_b32 flat_scratch_hi, s12, 8
; VI-NEXT:    v_addc_u32_e32 v5, vcc, 0, v1, vcc
; VI-NEXT:    flat_load_dwordx4 v[0:3], v[0:1]
; VI-NEXT:    flat_load_dwordx4 v[4:7], v[4:5]
; VI-NEXT:    v_mov_b32_e32 v9, s1
; VI-NEXT:    v_add_u32_e32 v8, vcc, s0, v8
; VI-NEXT:    v_addc_u32_e32 v9, vcc, 0, v9, vcc
; VI-NEXT:    s_lshl_b32 s1, s4, 16
; VI-NEXT:    v_add_u32_e32 v10, vcc, 16, v8
; VI-NEXT:    v_mov_b32_e32 v12, s1
; VI-NEXT:    v_addc_u32_e32 v11, vcc, 0, v9, vcc
; VI-NEXT:    s_waitcnt vmcnt(1)
; VI-NEXT:    v_or_b32_sdwa v1, v1, v12 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    flat_store_dwordx4 v[10:11], v[4:7]
; VI-NEXT:    flat_store_dwordx4 v[8:9], v[0:3]
; VI-NEXT:    s_endpgm
;
; GFX900-LABEL: v_insertelement_v16bf16_3:
; GFX900:       ; %bb.0:
; GFX900-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; GFX900-NEXT:    s_load_dword s4, s[8:9], 0x10
; GFX900-NEXT:    v_lshlrev_b32_e32 v8, 5, v0
; GFX900-NEXT:    v_mov_b32_e32 v9, 0x5040100
; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
; GFX900-NEXT:    global_load_dwordx4 v[0:3], v8, s[2:3]
; GFX900-NEXT:    global_load_dwordx4 v[4:7], v8, s[2:3] offset:16
; GFX900-NEXT:    s_waitcnt vmcnt(1)
; GFX900-NEXT:    v_perm_b32 v1, s4, v1, v9
; GFX900-NEXT:    s_waitcnt vmcnt(0)
; GFX900-NEXT:    global_store_dwordx4 v8, v[4:7], s[0:1] offset:16
; GFX900-NEXT:    global_store_dwordx4 v8, v[0:3], s[0:1]
; GFX900-NEXT:    s_endpgm
;
; GFX942-LABEL: v_insertelement_v16bf16_3:
; GFX942:       ; %bb.0:
; GFX942-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
; GFX942-NEXT:    s_load_dword s6, s[4:5], 0x10
; GFX942-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX942-NEXT:    v_lshlrev_b32_e32 v4, 5, v0
; GFX942-NEXT:    v_mov_b32_e32 v5, 0x5040100
; GFX942-NEXT:    s_waitcnt lgkmcnt(0)
; GFX942-NEXT:    global_load_dwordx4 v[0:3], v4, s[2:3]
; GFX942-NEXT:    global_load_dwordx4 v[6:9], v4, s[2:3] offset:16
; GFX942-NEXT:    s_waitcnt vmcnt(1)
; GFX942-NEXT:    v_perm_b32 v1, s6, v1, v5
; GFX942-NEXT:    s_waitcnt vmcnt(0)
; GFX942-NEXT:    global_store_dwordx4 v4, v[6:9], s[0:1] offset:16
; GFX942-NEXT:    global_store_dwordx4 v4, v[0:3], s[0:1]
; GFX942-NEXT:    s_endpgm
;
; GFX1250-LABEL: v_insertelement_v16bf16_3:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX1250-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX1250-NEXT:    s_wait_xcnt 0x0
; GFX1250-NEXT:    s_load_b32 s4, s[4:5], 0x10
; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX1250-NEXT:    v_lshlrev_b32_e32 v8, 5, v0
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    s_clause 0x1
; GFX1250-NEXT:    global_load_b128 v[0:3], v8, s[2:3]
; GFX1250-NEXT:    global_load_b128 v[4:7], v8, s[2:3] offset:16
; GFX1250-NEXT:    s_wait_loadcnt 0x1
; GFX1250-NEXT:    v_perm_b32 v1, s4, v1, 0x5040100
; GFX1250-NEXT:    s_wait_loadcnt 0x0
; GFX1250-NEXT:    s_clause 0x1
; GFX1250-NEXT:    global_store_b128 v8, v[4:7], s[0:1] offset:16
; GFX1250-NEXT:    global_store_b128 v8, v[0:3], s[0:1]
; GFX1250-NEXT:    s_endpgm
  %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
  %tid.ext = sext i32 %tid to i64
  %in.gep = getelementptr inbounds <16 x bfloat>, ptr addrspace(1) %in, i64 %tid.ext
  %out.gep = getelementptr inbounds <16 x bfloat>, ptr addrspace(1) %out, i64 %tid.ext
  %vec = load <16 x bfloat>, ptr addrspace(1) %in.gep
  %val.trunc = trunc i32 %val to i16
  %val.cvt = bitcast i16 %val.trunc to bfloat
  %vecins = insertelement <16 x bfloat> %vec, bfloat %val.cvt, i32 3
  store <16 x bfloat> %vecins, ptr addrspace(1) %out.gep
  ret void
}

define amdgpu_kernel void @v_insertelement_v16bf16_dynamic(ptr addrspace(1) %out, ptr addrspace(1) %in, i32 %val, i32 %n) {
; SI-LABEL: v_insertelement_v16bf16_dynamic:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx4 s[12:15], s[8:9], 0x0
; SI-NEXT:    s_load_dwordx2 s[6:7], s[8:9], 0x4
; SI-NEXT:    s_mov_b32 s3, 0x100f000
; SI-NEXT:    s_mov_b32 s2, 0
; SI-NEXT:    v_lshlrev_b32_e32 v4, 5, v0
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b64 s[0:1], s[14:15]
; SI-NEXT:    v_mov_b32_e32 v5, 0
; SI-NEXT:    buffer_load_dwordx4 v[7:10], v[4:5], s[0:3], 0 addr64
; SI-NEXT:    buffer_load_dwordx4 v[0:3], v[4:5], s[0:3], 0 addr64 offset:16
; SI-NEXT:    s_cmp_eq_u32 s7, 6
; SI-NEXT:    v_mov_b32_e32 v6, s6
; SI-NEXT:    s_cselect_b64 vcc, -1, 0
; SI-NEXT:    s_cmp_eq_u32 s7, 7
; SI-NEXT:    s_mov_b64 s[14:15], s[2:3]
; SI-NEXT:    s_waitcnt vmcnt(1)
; SI-NEXT:    v_cndmask_b32_e32 v11, v10, v6, vcc
; SI-NEXT:    s_cselect_b64 vcc, -1, 0
; SI-NEXT:    s_cmp_eq_u32 s7, 4
; SI-NEXT:    s_cselect_b64 s[0:1], -1, 0
; SI-NEXT:    s_cmp_eq_u32 s7, 5
; SI-NEXT:    v_lshrrev_b32_e32 v10, 16, v10
; SI-NEXT:    v_lshrrev_b32_e32 v12, 16, v9
; SI-NEXT:    v_cndmask_b32_e64 v9, v9, v6, s[0:1]
; SI-NEXT:    s_cselect_b64 s[0:1], -1, 0
; SI-NEXT:    s_cmp_eq_u32 s7, 2
; SI-NEXT:    s_cselect_b64 s[2:3], -1, 0
; SI-NEXT:    s_cmp_eq_u32 s7, 3
; SI-NEXT:    v_cndmask_b32_e32 v10, v10, v6, vcc
; SI-NEXT:    v_lshrrev_b32_e32 v13, 16, v8
; SI-NEXT:    v_cndmask_b32_e64 v8, v8, v6, s[2:3]
; SI-NEXT:    s_cselect_b64 s[2:3], -1, 0
; SI-NEXT:    s_cmp_eq_u32 s7, 0
; SI-NEXT:    v_and_b32_e32 v11, 0xffff, v11
; SI-NEXT:    v_lshlrev_b32_e32 v10, 16, v10
; SI-NEXT:    v_cndmask_b32_e64 v12, v12, v6, s[0:1]
; SI-NEXT:    s_cselect_b64 s[4:5], -1, 0
; SI-NEXT:    v_or_b32_e32 v10, v11, v10
; SI-NEXT:    v_lshlrev_b32_e32 v11, 16, v12
; SI-NEXT:    v_cndmask_b32_e64 v12, v13, v6, s[2:3]
; SI-NEXT:    s_cmp_eq_u32 s7, 1
; SI-NEXT:    v_lshrrev_b32_e32 v14, 16, v7
; SI-NEXT:    v_and_b32_e32 v8, 0xffff, v8
; SI-NEXT:    v_lshlrev_b32_e32 v12, 16, v12
; SI-NEXT:    s_cselect_b64 vcc, -1, 0
; SI-NEXT:    s_cmp_eq_u32 s7, 14
; SI-NEXT:    v_cndmask_b32_e64 v7, v7, v6, s[4:5]
; SI-NEXT:    v_or_b32_e32 v8, v8, v12
; SI-NEXT:    v_cndmask_b32_e32 v12, v14, v6, vcc
; SI-NEXT:    s_cselect_b64 vcc, -1, 0
; SI-NEXT:    s_cmp_eq_u32 s7, 15
; SI-NEXT:    s_waitcnt vmcnt(0)
; SI-NEXT:    v_lshrrev_b32_e32 v15, 16, v3
; SI-NEXT:    v_and_b32_e32 v7, 0xffff, v7
; SI-NEXT:    v_lshlrev_b32_e32 v12, 16, v12
; SI-NEXT:    v_cndmask_b32_e32 v3, v3, v6, vcc
; SI-NEXT:    s_cselect_b64 vcc, -1, 0
; SI-NEXT:    s_cmp_eq_u32 s7, 12
; SI-NEXT:    v_or_b32_e32 v7, v7, v12
; SI-NEXT:    v_cndmask_b32_e32 v12, v15, v6, vcc
; SI-NEXT:    s_cselect_b64 vcc, -1, 0
; SI-NEXT:    s_cmp_eq_u32 s7, 13
; SI-NEXT:    v_lshrrev_b32_e32 v16, 16, v2
; SI-NEXT:    v_and_b32_e32 v3, 0xffff, v3
; SI-NEXT:    v_lshlrev_b32_e32 v12, 16, v12
; SI-NEXT:    v_cndmask_b32_e32 v2, v2, v6, vcc
; SI-NEXT:    s_cselect_b64 vcc, -1, 0
; SI-NEXT:    s_cmp_eq_u32 s7, 10
; SI-NEXT:    v_or_b32_e32 v3, v3, v12
; SI-NEXT:    v_cndmask_b32_e32 v12, v16, v6, vcc
; SI-NEXT:    s_cselect_b64 vcc, -1, 0
; SI-NEXT:    s_cmp_eq_u32 s7, 11
; SI-NEXT:    v_lshrrev_b32_e32 v17, 16, v1
; SI-NEXT:    v_and_b32_e32 v2, 0xffff, v2
; SI-NEXT:    v_lshlrev_b32_e32 v12, 16, v12
; SI-NEXT:    v_cndmask_b32_e32 v1, v1, v6, vcc
; SI-NEXT:    s_cselect_b64 vcc, -1, 0
; SI-NEXT:    s_cmp_eq_u32 s7, 8
; SI-NEXT:    v_and_b32_e32 v9, 0xffff, v9
; SI-NEXT:    v_or_b32_e32 v2, v2, v12
; SI-NEXT:    v_cndmask_b32_e32 v12, v17, v6, vcc
; SI-NEXT:    s_cselect_b64 vcc, -1, 0
; SI-NEXT:    s_cmp_eq_u32 s7, 9
; SI-NEXT:    v_or_b32_e32 v9, v9, v11
; SI-NEXT:    v_lshrrev_b32_e32 v11, 16, v0
; SI-NEXT:    v_cndmask_b32_e32 v0, v0, v6, vcc
; SI-NEXT:    s_cselect_b64 vcc, -1, 0
; SI-NEXT:    v_cndmask_b32_e32 v6, v11, v6, vcc
; SI-NEXT:    v_and_b32_e32 v1, 0xffff, v1
; SI-NEXT:    v_lshlrev_b32_e32 v12, 16, v12
; SI-NEXT:    v_and_b32_e32 v0, 0xffff, v0
; SI-NEXT:    v_lshlrev_b32_e32 v6, 16, v6
; SI-NEXT:    v_or_b32_e32 v1, v1, v12
; SI-NEXT:    v_or_b32_e32 v0, v0, v6
; SI-NEXT:    buffer_store_dwordx4 v[0:3], v[4:5], s[12:15], 0 addr64 offset:16
; SI-NEXT:    buffer_store_dwordx4 v[7:10], v[4:5], s[12:15], 0 addr64
; SI-NEXT:    s_endpgm
;
; VI-LABEL: v_insertelement_v16bf16_dynamic:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    s_load_dwordx2 s[6:7], s[8:9], 0x10
; VI-NEXT:    v_lshlrev_b32_e32 v8, 5, v0
; VI-NEXT:    s_add_i32 s12, s12, s17
; VI-NEXT:    s_mov_b32 flat_scratch_lo, s13
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s3
; VI-NEXT:    v_add_u32_e32 v4, vcc, s2, v8
; VI-NEXT:    v_addc_u32_e32 v5, vcc, 0, v0, vcc
; VI-NEXT:    v_add_u32_e32 v0, vcc, 16, v4
; VI-NEXT:    s_lshr_b32 flat_scratch_hi, s12, 8
; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v5, vcc
; VI-NEXT:    flat_load_dwordx4 v[0:3], v[0:1]
; VI-NEXT:    flat_load_dwordx4 v[4:7], v[4:5]
; VI-NEXT:    v_mov_b32_e32 v9, s1
; VI-NEXT:    v_add_u32_e32 v8, vcc, s0, v8
; VI-NEXT:    v_addc_u32_e32 v9, vcc, 0, v9, vcc
; VI-NEXT:    v_add_u32_e32 v10, vcc, 16, v8
; VI-NEXT:    s_cmp_eq_u32 s7, 14
; VI-NEXT:    v_addc_u32_e32 v11, vcc, 0, v9, vcc
; VI-NEXT:    v_mov_b32_e32 v12, s6
; VI-NEXT:    s_cselect_b64 vcc, -1, 0
; VI-NEXT:    s_cmp_eq_u32 s7, 15
; VI-NEXT:    s_waitcnt vmcnt(1)
; VI-NEXT:    v_cndmask_b32_e32 v13, v3, v12, vcc
; VI-NEXT:    s_cselect_b64 vcc, -1, 0
; VI-NEXT:    s_cmp_eq_u32 s7, 12
; VI-NEXT:    s_cselect_b64 s[0:1], -1, 0
; VI-NEXT:    s_cmp_eq_u32 s7, 13
; VI-NEXT:    v_lshrrev_b32_e32 v14, 16, v2
; VI-NEXT:    v_cndmask_b32_e64 v2, v2, v12, s[0:1]
; VI-NEXT:    s_cselect_b64 s[0:1], -1, 0
; VI-NEXT:    s_cmp_eq_u32 s7, 10
; VI-NEXT:    s_cselect_b64 s[2:3], -1, 0
; VI-NEXT:    s_cmp_eq_u32 s7, 11
; VI-NEXT:    v_lshrrev_b32_e32 v15, 16, v1
; VI-NEXT:    v_cndmask_b32_e64 v1, v1, v12, s[2:3]
; VI-NEXT:    s_cselect_b64 s[2:3], -1, 0
; VI-NEXT:    s_cmp_eq_u32 s7, 8
; VI-NEXT:    v_lshrrev_b32_e32 v3, 16, v3
; VI-NEXT:    s_cselect_b64 s[4:5], -1, 0
; VI-NEXT:    v_cndmask_b32_e64 v15, v15, v12, s[2:3]
; VI-NEXT:    s_cmp_eq_u32 s7, 9
; VI-NEXT:    v_lshrrev_b32_e32 v16, 16, v0
; VI-NEXT:    v_cndmask_b32_e32 v3, v3, v12, vcc
; VI-NEXT:    v_lshlrev_b32_e32 v15, 16, v15
; VI-NEXT:    s_cselect_b64 vcc, -1, 0
; VI-NEXT:    s_cmp_eq_u32 s7, 6
; VI-NEXT:    v_or_b32_sdwa v1, v1, v15 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
; VI-NEXT:    v_cndmask_b32_e32 v15, v16, v12, vcc
; VI-NEXT:    s_cselect_b64 vcc, -1, 0
; VI-NEXT:    s_cmp_eq_u32 s7, 7
; VI-NEXT:    v_cndmask_b32_e64 v0, v0, v12, s[4:5]
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_lshrrev_b32_e32 v17, 16, v7
; VI-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
; VI-NEXT:    v_cndmask_b32_e64 v14, v14, v12, s[0:1]
; VI-NEXT:    v_lshlrev_b32_e32 v15, 16, v15
; VI-NEXT:    v_cndmask_b32_e32 v7, v7, v12, vcc
; VI-NEXT:    s_cselect_b64 vcc, -1, 0
; VI-NEXT:    s_cmp_eq_u32 s7, 4
; VI-NEXT:    v_or_b32_sdwa v3, v13, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
; VI-NEXT:    v_lshlrev_b32_e32 v13, 16, v14
; VI-NEXT:    v_or_b32_sdwa v0, v0, v15 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
; VI-NEXT:    v_cndmask_b32_e32 v15, v17, v12, vcc
; VI-NEXT:    s_cselect_b64 vcc, -1, 0
; VI-NEXT:    s_cmp_eq_u32 s7, 5
; VI-NEXT:    v_or_b32_sdwa v2, v2, v13 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
; VI-NEXT:    v_lshrrev_b32_e32 v13, 16, v6
; VI-NEXT:    v_cndmask_b32_e32 v6, v6, v12, vcc
; VI-NEXT:    s_cselect_b64 vcc, -1, 0
; VI-NEXT:    s_cmp_eq_u32 s7, 2
; VI-NEXT:    v_cndmask_b32_e32 v13, v13, v12, vcc
; VI-NEXT:    s_cselect_b64 vcc, -1, 0
; VI-NEXT:    s_cmp_eq_u32 s7, 3
; VI-NEXT:    v_lshrrev_b32_e32 v14, 16, v5
; VI-NEXT:    v_cndmask_b32_e32 v5, v5, v12, vcc
; VI-NEXT:    s_cselect_b64 vcc, -1, 0
; VI-NEXT:    s_cmp_eq_u32 s7, 0
; VI-NEXT:    v_lshlrev_b32_e32 v13, 16, v13
; VI-NEXT:    v_cndmask_b32_e32 v14, v14, v12, vcc
; VI-NEXT:    s_cselect_b64 vcc, -1, 0
; VI-NEXT:    s_cmp_eq_u32 s7, 1
; VI-NEXT:    v_or_b32_sdwa v6, v6, v13 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
; VI-NEXT:    v_lshrrev_b32_e32 v13, 16, v4
; VI-NEXT:    v_cndmask_b32_e32 v4, v4, v12, vcc
; VI-NEXT:    s_cselect_b64 vcc, -1, 0
; VI-NEXT:    v_cndmask_b32_e32 v12, v13, v12, vcc
; VI-NEXT:    v_lshlrev_b32_e32 v15, 16, v15
; VI-NEXT:    v_lshlrev_b32_e32 v14, 16, v14
; VI-NEXT:    v_lshlrev_b32_e32 v12, 16, v12
; VI-NEXT:    v_or_b32_sdwa v7, v7, v15 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
; VI-NEXT:    v_or_b32_sdwa v5, v5, v14 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
; VI-NEXT:    v_or_b32_sdwa v4, v4, v12 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
; VI-NEXT:    flat_store_dwordx4 v[8:9], v[4:7]
; VI-NEXT:    flat_store_dwordx4 v[10:11], v[0:3]
; VI-NEXT:    s_endpgm
;
; GFX900-LABEL: v_insertelement_v16bf16_dynamic:
; GFX900:       ; %bb.0:
; GFX900-NEXT:    s_load_dwordx4 s[36:39], s[8:9], 0x0
; GFX900-NEXT:    s_load_dwordx2 s[28:29], s[8:9], 0x10
; GFX900-NEXT:    v_lshlrev_b32_e32 v8, 5, v0
; GFX900-NEXT:    s_mov_b32 s30, 0x5040100
; GFX900-NEXT:    s_waitcnt lgkmcnt(0)
; GFX900-NEXT:    global_load_dwordx4 v[0:3], v8, s[38:39]
; GFX900-NEXT:    global_load_dwordx4 v[4:7], v8, s[38:39] offset:16
; GFX900-NEXT:    s_cmp_eq_u32 s29, 6
; GFX900-NEXT:    s_cselect_b64 vcc, -1, 0
; GFX900-NEXT:    s_cmp_eq_u32 s29, 7
; GFX900-NEXT:    s_cselect_b64 s[0:1], -1, 0
; GFX900-NEXT:    s_cmp_eq_u32 s29, 4
; GFX900-NEXT:    s_cselect_b64 s[2:3], -1, 0
; GFX900-NEXT:    s_cmp_eq_u32 s29, 5
; GFX900-NEXT:    s_cselect_b64 s[4:5], -1, 0
; GFX900-NEXT:    s_cmp_eq_u32 s29, 2
; GFX900-NEXT:    s_cselect_b64 s[6:7], -1, 0
; GFX900-NEXT:    s_cmp_eq_u32 s29, 3
; GFX900-NEXT:    s_cselect_b64 s[8:9], -1, 0
; GFX900-NEXT:    s_cmp_eq_u32 s29, 0
; GFX900-NEXT:    s_cselect_b64 s[10:11], -1, 0
; GFX900-NEXT:    s_cmp_eq_u32 s29, 1
; GFX900-NEXT:    s_cselect_b64 s[12:13], -1, 0
; GFX900-NEXT:    s_cmp_eq_u32 s29, 14
; GFX900-NEXT:    v_mov_b32_e32 v9, s28
; GFX900-NEXT:    s_cselect_b64 s[14:15], -1, 0
; GFX900-NEXT:    s_cmp_eq_u32 s29, 15
; GFX900-NEXT:    s_cselect_b64 s[16:17], -1, 0
; GFX900-NEXT:    s_cmp_eq_u32 s29, 12
; GFX900-NEXT:    s_cselect_b64 s[18:19], -1, 0
; GFX900-NEXT:    s_cmp_eq_u32 s29, 13
; GFX900-NEXT:    s_cselect_b64 s[20:21], -1, 0
; GFX900-NEXT:    s_cmp_eq_u32 s29, 10
; GFX900-NEXT:    s_cselect_b64 s[22:23], -1, 0
; GFX900-NEXT:    s_cmp_eq_u32 s29, 11
; GFX900-NEXT:    s_cselect_b64 s[24:25], -1, 0
; GFX900-NEXT:    s_cmp_eq_u32 s29, 8
; GFX900-NEXT:    s_cselect_b64 s[26:27], -1, 0
; GFX900-NEXT:    s_cmp_eq_u32 s29, 9
; GFX900-NEXT:    s_cselect_b64 s[28:29], -1, 0
; GFX900-NEXT:    s_waitcnt vmcnt(1)
; GFX900-NEXT:    v_cndmask_b32_e32 v10, v3, v9, vcc
; GFX900-NEXT:    s_mov_b64 vcc, s[0:1]
; GFX900-NEXT:    v_cndmask_b32_sdwa v3, v3, v9, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; GFX900-NEXT:    s_mov_b64 vcc, s[4:5]
; GFX900-NEXT:    v_cndmask_b32_e64 v11, v2, v9, s[2:3]
; GFX900-NEXT:    v_cndmask_b32_sdwa v2, v2, v9, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; GFX900-NEXT:    s_mov_b64 vcc, s[8:9]
; GFX900-NEXT:    v_cndmask_b32_e64 v12, v1, v9, s[6:7]
; GFX900-NEXT:    v_cndmask_b32_sdwa v1, v1, v9, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; GFX900-NEXT:    s_mov_b64 vcc, s[12:13]
; GFX900-NEXT:    v_cndmask_b32_e64 v13, v0, v9, s[10:11]
; GFX900-NEXT:    v_cndmask_b32_sdwa v0, v0, v9, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; GFX900-NEXT:    s_mov_b64 vcc, s[16:17]
; GFX900-NEXT:    s_waitcnt vmcnt(0)
; GFX900-NEXT:    v_cndmask_b32_e64 v14, v7, v9, s[14:15]
; GFX900-NEXT:    v_cndmask_b32_sdwa v7, v7, v9, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; GFX900-NEXT:    s_mov_b64 vcc, s[20:21]
; GFX900-NEXT:    v_cndmask_b32_e64 v15, v6, v9, s[18:19]
; GFX900-NEXT:    v_cndmask_b32_sdwa v6, v6, v9, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; GFX900-NEXT:    s_mov_b64 vcc, s[24:25]
; GFX900-NEXT:    v_perm_b32 v3, v3, v10, s30
; GFX900-NEXT:    v_cndmask_b32_e64 v10, v5, v9, s[22:23]
; GFX900-NEXT:    v_cndmask_b32_sdwa v5, v5, v9, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; GFX900-NEXT:    s_mov_b64 vcc, s[28:29]
; GFX900-NEXT:    v_perm_b32 v2, v2, v11, s30
; GFX900-NEXT:    v_cndmask_b32_e64 v11, v4, v9, s[26:27]
; GFX900-NEXT:    v_cndmask_b32_sdwa v4, v4, v9, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; GFX900-NEXT:    v_perm_b32 v7, v7, v14, s30
; GFX900-NEXT:    v_perm_b32 v6, v6, v15, s30
; GFX900-NEXT:    v_perm_b32 v5, v5, v10, s30
; GFX900-NEXT:    v_perm_b32 v4, v4, v11, s30
; GFX900-NEXT:    v_perm_b32 v1, v1, v12, s30
; GFX900-NEXT:    v_perm_b32 v0, v0, v13, s30
; GFX900-NEXT:    global_store_dwordx4 v8, v[4:7], s[36:37] offset:16
; GFX900-NEXT:    global_store_dwordx4 v8, v[0:3], s[36:37]
; GFX900-NEXT:    s_endpgm
;
; GFX942-LABEL: v_insertelement_v16bf16_dynamic:
; GFX942:       ; %bb.0:
; GFX942-NEXT:    s_load_dwordx4 s[36:39], s[4:5], 0x0
; GFX942-NEXT:    s_load_dwordx2 s[28:29], s[4:5], 0x10
; GFX942-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX942-NEXT:    v_lshlrev_b32_e32 v8, 5, v0
; GFX942-NEXT:    s_mov_b32 s30, 0x5040100
; GFX942-NEXT:    s_waitcnt lgkmcnt(0)
; GFX942-NEXT:    global_load_dwordx4 v[0:3], v8, s[38:39]
; GFX942-NEXT:    global_load_dwordx4 v[4:7], v8, s[38:39] offset:16
; GFX942-NEXT:    s_cmp_eq_u32 s29, 6
; GFX942-NEXT:    s_cselect_b64 vcc, -1, 0
; GFX942-NEXT:    s_cmp_eq_u32 s29, 7
; GFX942-NEXT:    s_cselect_b64 s[0:1], -1, 0
; GFX942-NEXT:    s_cmp_eq_u32 s29, 4
; GFX942-NEXT:    s_cselect_b64 s[2:3], -1, 0
; GFX942-NEXT:    s_cmp_eq_u32 s29, 5
; GFX942-NEXT:    s_cselect_b64 s[4:5], -1, 0
; GFX942-NEXT:    s_cmp_eq_u32 s29, 2
; GFX942-NEXT:    s_cselect_b64 s[6:7], -1, 0
; GFX942-NEXT:    s_cmp_eq_u32 s29, 3
; GFX942-NEXT:    s_cselect_b64 s[8:9], -1, 0
; GFX942-NEXT:    s_cmp_eq_u32 s29, 0
; GFX942-NEXT:    s_cselect_b64 s[10:11], -1, 0
; GFX942-NEXT:    s_cmp_eq_u32 s29, 1
; GFX942-NEXT:    s_cselect_b64 s[12:13], -1, 0
; GFX942-NEXT:    s_cmp_eq_u32 s29, 14
; GFX942-NEXT:    v_mov_b32_e32 v9, s28
; GFX942-NEXT:    s_cselect_b64 s[14:15], -1, 0
; GFX942-NEXT:    s_cmp_eq_u32 s29, 15
; GFX942-NEXT:    s_cselect_b64 s[16:17], -1, 0
; GFX942-NEXT:    s_cmp_eq_u32 s29, 12
; GFX942-NEXT:    s_cselect_b64 s[18:19], -1, 0
; GFX942-NEXT:    s_cmp_eq_u32 s29, 13
; GFX942-NEXT:    s_cselect_b64 s[20:21], -1, 0
; GFX942-NEXT:    s_cmp_eq_u32 s29, 10
; GFX942-NEXT:    s_cselect_b64 s[22:23], -1, 0
; GFX942-NEXT:    s_cmp_eq_u32 s29, 11
; GFX942-NEXT:    s_cselect_b64 s[24:25], -1, 0
; GFX942-NEXT:    s_cmp_eq_u32 s29, 8
; GFX942-NEXT:    s_cselect_b64 s[26:27], -1, 0
; GFX942-NEXT:    s_cmp_eq_u32 s29, 9
; GFX942-NEXT:    s_cselect_b64 s[28:29], -1, 0
; GFX942-NEXT:    s_waitcnt vmcnt(1)
; GFX942-NEXT:    v_cndmask_b32_e32 v10, v3, v9, vcc
; GFX942-NEXT:    s_mov_b64 vcc, s[0:1]
; GFX942-NEXT:    v_cndmask_b32_sdwa v3, v3, v9, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; GFX942-NEXT:    s_mov_b64 vcc, s[4:5]
; GFX942-NEXT:    v_cndmask_b32_e64 v11, v2, v9, s[2:3]
; GFX942-NEXT:    v_cndmask_b32_sdwa v2, v2, v9, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; GFX942-NEXT:    s_mov_b64 vcc, s[8:9]
; GFX942-NEXT:    v_cndmask_b32_e64 v12, v1, v9, s[6:7]
; GFX942-NEXT:    v_cndmask_b32_sdwa v1, v1, v9, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; GFX942-NEXT:    s_mov_b64 vcc, s[12:13]
; GFX942-NEXT:    v_cndmask_b32_e64 v13, v0, v9, s[10:11]
; GFX942-NEXT:    v_cndmask_b32_sdwa v0, v0, v9, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; GFX942-NEXT:    s_mov_b64 vcc, s[16:17]
; GFX942-NEXT:    s_waitcnt vmcnt(0)
; GFX942-NEXT:    v_cndmask_b32_e64 v14, v7, v9, s[14:15]
; GFX942-NEXT:    v_cndmask_b32_sdwa v7, v7, v9, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; GFX942-NEXT:    s_mov_b64 vcc, s[20:21]
; GFX942-NEXT:    v_cndmask_b32_e64 v15, v6, v9, s[18:19]
; GFX942-NEXT:    v_cndmask_b32_sdwa v6, v6, v9, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; GFX942-NEXT:    s_mov_b64 vcc, s[24:25]
; GFX942-NEXT:    v_cndmask_b32_e64 v16, v5, v9, s[22:23]
; GFX942-NEXT:    v_cndmask_b32_sdwa v5, v5, v9, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; GFX942-NEXT:    s_mov_b64 vcc, s[28:29]
; GFX942-NEXT:    v_cndmask_b32_e64 v17, v4, v9, s[26:27]
; GFX942-NEXT:    v_cndmask_b32_sdwa v4, v4, v9, vcc dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; GFX942-NEXT:    v_perm_b32 v7, v7, v14, s30
; GFX942-NEXT:    v_perm_b32 v6, v6, v15, s30
; GFX942-NEXT:    v_perm_b32 v5, v5, v16, s30
; GFX942-NEXT:    v_perm_b32 v4, v4, v17, s30
; GFX942-NEXT:    v_perm_b32 v3, v3, v10, s30
; GFX942-NEXT:    v_perm_b32 v2, v2, v11, s30
; GFX942-NEXT:    v_perm_b32 v1, v1, v12, s30
; GFX942-NEXT:    v_perm_b32 v0, v0, v13, s30
; GFX942-NEXT:    global_store_dwordx4 v8, v[4:7], s[36:37] offset:16
; GFX942-NEXT:    global_store_dwordx4 v8, v[0:3], s[36:37]
; GFX942-NEXT:    s_endpgm
;
; GFX1250-LABEL: v_insertelement_v16bf16_dynamic:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX1250-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
; GFX1250-NEXT:    s_wait_xcnt 0x0
; GFX1250-NEXT:    s_load_b64 s[4:5], s[4:5], 0x10
; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX1250-NEXT:    v_lshlrev_b32_e32 v8, 5, v0
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    s_clause 0x1
; GFX1250-NEXT:    global_load_b128 v[0:3], v8, s[2:3]
; GFX1250-NEXT:    global_load_b128 v[4:7], v8, s[2:3] offset:16
; GFX1250-NEXT:    s_cmp_eq_u32 s5, 6
; GFX1250-NEXT:    s_wait_xcnt 0x0
; GFX1250-NEXT:    s_cselect_b32 s2, -1, 0
; GFX1250-NEXT:    s_cmp_eq_u32 s5, 7
; GFX1250-NEXT:    s_wait_loadcnt 0x1
; GFX1250-NEXT:    v_cndmask_b32_e64 v9, v3, s4, s2
; GFX1250-NEXT:    s_cselect_b32 s2, -1, 0
; GFX1250-NEXT:    s_cmp_eq_u32 s5, 4
; GFX1250-NEXT:    v_lshrrev_b32_e32 v3, 16, v3
; GFX1250-NEXT:    s_cselect_b32 s3, -1, 0
; GFX1250-NEXT:    s_cmp_eq_u32 s5, 5
; GFX1250-NEXT:    v_dual_lshrrev_b32 v10, 16, v2 :: v_dual_lshrrev_b32 v11, 16, v1
; GFX1250-NEXT:    v_cndmask_b32_e64 v2, v2, s4, s3
; GFX1250-NEXT:    s_cselect_b32 s3, -1, 0
; GFX1250-NEXT:    s_cmp_eq_u32 s5, 2
; GFX1250-NEXT:    v_cndmask_b32_e64 v3, v3, s4, s2
; GFX1250-NEXT:    s_cselect_b32 s2, -1, 0
; GFX1250-NEXT:    s_cmp_eq_u32 s5, 3
; GFX1250-NEXT:    v_cndmask_b32_e64 v1, v1, s4, s2
; GFX1250-NEXT:    s_cselect_b32 s2, -1, 0
; GFX1250-NEXT:    s_cmp_eq_u32 s5, 0
; GFX1250-NEXT:    s_wait_loadcnt 0x0
; GFX1250-NEXT:    v_dual_lshrrev_b32 v12, 16, v0 :: v_dual_lshrrev_b32 v13, 16, v7
; GFX1250-NEXT:    v_cndmask_b32_e64 v11, v11, s4, s2
; GFX1250-NEXT:    s_cselect_b32 s2, -1, 0
; GFX1250-NEXT:    s_cmp_eq_u32 s5, 1
; GFX1250-NEXT:    v_cndmask_b32_e64 v0, v0, s4, s2
; GFX1250-NEXT:    s_cselect_b32 s2, -1, 0
; GFX1250-NEXT:    s_cmp_eq_u32 s5, 14
; GFX1250-NEXT:    v_cndmask_b32_e64 v10, v10, s4, s3
; GFX1250-NEXT:    v_perm_b32 v3, v3, v9, 0x5040100
; GFX1250-NEXT:    v_cndmask_b32_e64 v9, v12, s4, s2
; GFX1250-NEXT:    s_cselect_b32 s2, -1, 0
; GFX1250-NEXT:    s_cmp_eq_u32 s5, 15
; GFX1250-NEXT:    v_cndmask_b32_e64 v7, v7, s4, s2
; GFX1250-NEXT:    s_cselect_b32 s2, -1, 0
; GFX1250-NEXT:    s_cmp_eq_u32 s5, 12
; GFX1250-NEXT:    v_dual_lshrrev_b32 v14, 16, v6 :: v_dual_lshrrev_b32 v15, 16, v5
; GFX1250-NEXT:    v_perm_b32 v2, v10, v2, 0x5040100
; GFX1250-NEXT:    v_cndmask_b32_e64 v10, v13, s4, s2
; GFX1250-NEXT:    s_cselect_b32 s2, -1, 0
; GFX1250-NEXT:    s_cmp_eq_u32 s5, 13
; GFX1250-NEXT:    v_cndmask_b32_e64 v6, v6, s4, s2
; GFX1250-NEXT:    s_cselect_b32 s2, -1, 0
; GFX1250-NEXT:    s_cmp_eq_u32 s5, 10
; GFX1250-NEXT:    v_cndmask_b32_e64 v12, v14, s4, s2
; GFX1250-NEXT:    s_cselect_b32 s2, -1, 0
; GFX1250-NEXT:    s_cmp_eq_u32 s5, 11
; GFX1250-NEXT:    v_cndmask_b32_e64 v5, v5, s4, s2
; GFX1250-NEXT:    s_cselect_b32 s2, -1, 0
; GFX1250-NEXT:    s_cmp_eq_u32 s5, 8
; GFX1250-NEXT:    v_lshrrev_b32_e32 v16, 16, v4
; GFX1250-NEXT:    v_cndmask_b32_e64 v13, v15, s4, s2
; GFX1250-NEXT:    s_cselect_b32 s2, -1, 0
; GFX1250-NEXT:    s_cmp_eq_u32 s5, 9
; GFX1250-NEXT:    v_cndmask_b32_e64 v4, v4, s4, s2
; GFX1250-NEXT:    s_cselect_b32 s2, -1, 0
; GFX1250-NEXT:    v_perm_b32 v7, v10, v7, 0x5040100
; GFX1250-NEXT:    v_cndmask_b32_e64 v14, v16, s4, s2
; GFX1250-NEXT:    v_perm_b32 v6, v12, v6, 0x5040100
; GFX1250-NEXT:    v_perm_b32 v5, v13, v5, 0x5040100
; GFX1250-NEXT:    v_perm_b32 v1, v11, v1, 0x5040100
; GFX1250-NEXT:    v_perm_b32 v0, v9, v0, 0x5040100
; GFX1250-NEXT:    v_perm_b32 v4, v14, v4, 0x5040100
; GFX1250-NEXT:    s_clause 0x1
; GFX1250-NEXT:    global_store_b128 v8, v[4:7], s[0:1] offset:16
; GFX1250-NEXT:    global_store_b128 v8, v[0:3], s[0:1]
; GFX1250-NEXT:    s_endpgm
  %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
  %tid.ext = sext i32 %tid to i64
  %in.gep = getelementptr inbounds <16 x bfloat>, ptr addrspace(1) %in, i64 %tid.ext
  %out.gep = getelementptr inbounds <16 x bfloat>, ptr addrspace(1) %out, i64 %tid.ext
  %vec = load <16 x bfloat>, ptr addrspace(1) %in.gep
  %val.trunc = trunc i32 %val to i16
  %val.cvt = bitcast i16 %val.trunc to bfloat
  %vecins = insertelement <16 x bfloat> %vec, bfloat %val.cvt, i32 %n
  store <16 x bfloat> %vecins, ptr addrspace(1) %out.gep
  ret void
}

declare i32 @llvm.amdgcn.workitem.id.x() #1

attributes #0 = { nounwind }
attributes #1 = { nounwind readnone }