; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn-amd-amdhsa -mcpu=kaveri -verify-machineinstrs < %s | FileCheck -allow-deprecated-dag-overlap -enable-var-scope -check-prefixes=CIVI,CI %s
; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn-amd-amdhsa -mcpu=tonga -verify-machineinstrs < %s | FileCheck -allow-deprecated-dag-overlap -enable-var-scope -check-prefixes=CIVI,VI %s
; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -verify-machineinstrs < %s | FileCheck -allow-deprecated-dag-overlap -enable-var-scope -check-prefixes=GFX11 %s

; half args should be promoted to float for CI and lower.

define amdgpu_kernel void @load_f16_arg(ptr addrspace(1) %out, half %arg) #0 {
; CI-LABEL: load_f16_arg:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx2 s[0:1], s[8:9], 0x0
; CI-NEXT:    s_load_dword s2, s[8:9], 0x2
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    v_mov_b32_e32 v0, s0
; CI-NEXT:    v_mov_b32_e32 v1, s1
; CI-NEXT:    v_mov_b32_e32 v2, s2
; CI-NEXT:    flat_store_short v[0:1], v2
; CI-NEXT:    s_endpgm
;
; VI-LABEL: load_f16_arg:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx2 s[0:1], s[8:9], 0x0
; VI-NEXT:    s_load_dword s2, s[8:9], 0x8
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s0
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    v_mov_b32_e32 v2, s2
; VI-NEXT:    flat_store_short v[0:1], v2
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: load_f16_arg:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b32 s2, s[4:5], 0x8
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, s2
; GFX11-NEXT:    global_store_b16 v0, v1, s[0:1]
; GFX11-NEXT:    s_endpgm
  store half %arg, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @load_v2f16_arg(ptr addrspace(1) %out, <2 x half> %arg) #0 {
; CI-LABEL: load_v2f16_arg:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx2 s[0:1], s[8:9], 0x0
; CI-NEXT:    s_load_dword s2, s[8:9], 0x2
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    v_mov_b32_e32 v0, s0
; CI-NEXT:    v_mov_b32_e32 v1, s1
; CI-NEXT:    v_mov_b32_e32 v2, s2
; CI-NEXT:    flat_store_dword v[0:1], v2
; CI-NEXT:    s_endpgm
;
; VI-LABEL: load_v2f16_arg:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx2 s[0:1], s[8:9], 0x0
; VI-NEXT:    s_load_dword s2, s[8:9], 0x8
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s0
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    v_mov_b32_e32 v2, s2
; VI-NEXT:    flat_store_dword v[0:1], v2
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: load_v2f16_arg:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b32 s2, s[4:5], 0x8
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, s2
; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX11-NEXT:    s_endpgm
  store <2 x half> %arg, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @load_v3f16_arg(ptr addrspace(1) %out, <3 x half> %arg) #0 {
; CIVI-LABEL: load_v3f16_arg:
; CIVI:       ; %bb.0:
; CIVI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; CIVI-NEXT:    s_waitcnt lgkmcnt(0)
; CIVI-NEXT:    s_add_u32 s4, s0, 4
; CIVI-NEXT:    s_addc_u32 s5, s1, 0
; CIVI-NEXT:    v_mov_b32_e32 v2, s4
; CIVI-NEXT:    v_mov_b32_e32 v4, s3
; CIVI-NEXT:    v_mov_b32_e32 v0, s0
; CIVI-NEXT:    v_mov_b32_e32 v3, s5
; CIVI-NEXT:    v_mov_b32_e32 v1, s1
; CIVI-NEXT:    v_mov_b32_e32 v5, s2
; CIVI-NEXT:    flat_store_short v[2:3], v4
; CIVI-NEXT:    flat_store_dword v[0:1], v5
; CIVI-NEXT:    s_endpgm
;
; GFX11-LABEL: load_v3f16_arg:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, s3
; GFX11-NEXT:    v_mov_b32_e32 v2, s2
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    global_store_b16 v0, v1, s[0:1] offset:4
; GFX11-NEXT:    global_store_b32 v0, v2, s[0:1]
; GFX11-NEXT:    s_endpgm
  store <3 x half> %arg, ptr addrspace(1) %out
  ret void
}


; FIXME: Why not one load?
define amdgpu_kernel void @load_v4f16_arg(ptr addrspace(1) %out, <4 x half> %arg) #0 {
; CIVI-LABEL: load_v4f16_arg:
; CIVI:       ; %bb.0:
; CIVI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; CIVI-NEXT:    s_waitcnt lgkmcnt(0)
; CIVI-NEXT:    v_mov_b32_e32 v0, s0
; CIVI-NEXT:    v_mov_b32_e32 v2, s2
; CIVI-NEXT:    v_mov_b32_e32 v1, s1
; CIVI-NEXT:    v_mov_b32_e32 v3, s3
; CIVI-NEXT:    flat_store_dwordx2 v[0:1], v[2:3]
; CIVI-NEXT:    s_endpgm
;
; GFX11-LABEL: load_v4f16_arg:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-NEXT:    s_endpgm
  store <4 x half> %arg, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @load_v8f16_arg(ptr addrspace(1) %out, <8 x half> %arg) #0 {
; CI-LABEL: load_v8f16_arg:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx2 s[4:5], s[8:9], 0x0
; CI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x4
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    v_mov_b32_e32 v4, s4
; CI-NEXT:    v_mov_b32_e32 v0, s0
; CI-NEXT:    v_mov_b32_e32 v5, s5
; CI-NEXT:    v_mov_b32_e32 v1, s1
; CI-NEXT:    v_mov_b32_e32 v2, s2
; CI-NEXT:    v_mov_b32_e32 v3, s3
; CI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
; CI-NEXT:    s_endpgm
;
; VI-LABEL: load_v8f16_arg:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx2 s[4:5], s[8:9], 0x0
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x10
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v4, s4
; VI-NEXT:    v_mov_b32_e32 v0, s0
; VI-NEXT:    v_mov_b32_e32 v5, s5
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    v_mov_b32_e32 v2, s2
; VI-NEXT:    v_mov_b32_e32 v3, s3
; VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: load_v8f16_arg:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x10
; GFX11-NEXT:    s_load_b64 s[4:5], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v4, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v3, s3
; GFX11-NEXT:    v_dual_mov_b32 v1, s1 :: v_dual_mov_b32 v2, s2
; GFX11-NEXT:    global_store_b128 v4, v[0:3], s[4:5]
; GFX11-NEXT:    s_endpgm
  store <8 x half> %arg, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @extload_v2f16_arg(ptr addrspace(1) %out, <2 x half> %in) #0 {
; CI-LABEL: extload_v2f16_arg:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dword s2, s[8:9], 0x2
; CI-NEXT:    s_load_dwordx2 s[0:1], s[8:9], 0x0
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    s_lshr_b32 s3, s2, 16
; CI-NEXT:    v_cvt_f32_f16_e32 v1, s3
; CI-NEXT:    v_cvt_f32_f16_e32 v0, s2
; CI-NEXT:    v_mov_b32_e32 v3, s1
; CI-NEXT:    v_mov_b32_e32 v2, s0
; CI-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
; CI-NEXT:    s_endpgm
;
; VI-LABEL: extload_v2f16_arg:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dword s2, s[8:9], 0x8
; VI-NEXT:    s_load_dwordx2 s[0:1], s[8:9], 0x0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_lshr_b32 s3, s2, 16
; VI-NEXT:    v_cvt_f32_f16_e32 v1, s3
; VI-NEXT:    v_cvt_f32_f16_e32 v0, s2
; VI-NEXT:    v_mov_b32_e32 v3, s1
; VI-NEXT:    v_mov_b32_e32 v2, s0
; VI-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: extload_v2f16_arg:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b32 s2, s[4:5], 0x8
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_lshr_b32 s3, s2, 16
; GFX11-NEXT:    v_cvt_f32_f16_e32 v0, s2
; GFX11-NEXT:    v_cvt_f32_f16_e32 v1, s3
; GFX11-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-NEXT:    s_endpgm
  %fpext = fpext <2 x half> %in to <2 x float>
  store <2 x float> %fpext, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @extload_f16_to_f32_arg(ptr addrspace(1) %out, half %arg) #0 {
; CI-LABEL: extload_f16_to_f32_arg:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dword s2, s[8:9], 0x2
; CI-NEXT:    s_load_dwordx2 s[0:1], s[8:9], 0x0
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    v_cvt_f32_f16_e32 v2, s2
; CI-NEXT:    v_mov_b32_e32 v0, s0
; CI-NEXT:    v_mov_b32_e32 v1, s1
; CI-NEXT:    flat_store_dword v[0:1], v2
; CI-NEXT:    s_endpgm
;
; VI-LABEL: extload_f16_to_f32_arg:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dword s2, s[8:9], 0x8
; VI-NEXT:    s_load_dwordx2 s[0:1], s[8:9], 0x0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_cvt_f32_f16_e32 v2, s2
; VI-NEXT:    v_mov_b32_e32 v0, s0
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    flat_store_dword v[0:1], v2
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: extload_f16_to_f32_arg:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b32 s2, s[4:5], 0x8
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v0, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_cvt_f32_f16_e32 v1, s2
; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX11-NEXT:    s_endpgm
  %ext = fpext half %arg to float
  store float %ext, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @extload_v2f16_to_v2f32_arg(ptr addrspace(1) %out, <2 x half> %arg) #0 {
; CI-LABEL: extload_v2f16_to_v2f32_arg:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dword s2, s[8:9], 0x2
; CI-NEXT:    s_load_dwordx2 s[0:1], s[8:9], 0x0
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    s_lshr_b32 s3, s2, 16
; CI-NEXT:    v_cvt_f32_f16_e32 v1, s3
; CI-NEXT:    v_cvt_f32_f16_e32 v0, s2
; CI-NEXT:    v_mov_b32_e32 v3, s1
; CI-NEXT:    v_mov_b32_e32 v2, s0
; CI-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
; CI-NEXT:    s_endpgm
;
; VI-LABEL: extload_v2f16_to_v2f32_arg:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dword s2, s[8:9], 0x8
; VI-NEXT:    s_load_dwordx2 s[0:1], s[8:9], 0x0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_lshr_b32 s3, s2, 16
; VI-NEXT:    v_cvt_f32_f16_e32 v1, s3
; VI-NEXT:    v_cvt_f32_f16_e32 v0, s2
; VI-NEXT:    v_mov_b32_e32 v3, s1
; VI-NEXT:    v_mov_b32_e32 v2, s0
; VI-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: extload_v2f16_to_v2f32_arg:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b32 s2, s[4:5], 0x8
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_lshr_b32 s3, s2, 16
; GFX11-NEXT:    v_cvt_f32_f16_e32 v0, s2
; GFX11-NEXT:    v_cvt_f32_f16_e32 v1, s3
; GFX11-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-NEXT:    s_endpgm
  %ext = fpext <2 x half> %arg to <2 x float>
  store <2 x float> %ext, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @extload_v3f16_to_v3f32_arg(ptr addrspace(1) %out, <3 x half> %arg) #0 {
; CI-LABEL: extload_v3f16_to_v3f32_arg:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    s_lshr_b32 s4, s2, 16
; CI-NEXT:    v_cvt_f32_f16_e32 v2, s3
; CI-NEXT:    v_cvt_f32_f16_e32 v1, s4
; CI-NEXT:    v_cvt_f32_f16_e32 v0, s2
; CI-NEXT:    v_mov_b32_e32 v4, s1
; CI-NEXT:    v_mov_b32_e32 v3, s0
; CI-NEXT:    flat_store_dwordx3 v[3:4], v[0:2]
; CI-NEXT:    s_endpgm
;
; VI-LABEL: extload_v3f16_to_v3f32_arg:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_lshr_b32 s4, s2, 16
; VI-NEXT:    v_cvt_f32_f16_e32 v0, s2
; VI-NEXT:    v_cvt_f32_f16_e32 v1, s4
; VI-NEXT:    v_cvt_f32_f16_e32 v2, s3
; VI-NEXT:    v_mov_b32_e32 v4, s1
; VI-NEXT:    v_mov_b32_e32 v3, s0
; VI-NEXT:    flat_store_dwordx3 v[3:4], v[0:2]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: extload_v3f16_to_v3f32_arg:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v3, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_lshr_b32 s4, s2, 16
; GFX11-NEXT:    v_cvt_f32_f16_e32 v0, s2
; GFX11-NEXT:    v_cvt_f32_f16_e32 v1, s4
; GFX11-NEXT:    v_cvt_f32_f16_e32 v2, s3
; GFX11-NEXT:    global_store_b96 v3, v[0:2], s[0:1]
; GFX11-NEXT:    s_endpgm
  %ext = fpext <3 x half> %arg to <3 x float>
  store <3 x float> %ext, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @extload_v4f16_to_v4f32_arg(ptr addrspace(1) %out, <4 x half> %arg) #0 {
; CI-LABEL: extload_v4f16_to_v4f32_arg:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    s_lshr_b32 s4, s3, 16
; CI-NEXT:    s_lshr_b32 s5, s2, 16
; CI-NEXT:    v_cvt_f32_f16_e32 v2, s3
; CI-NEXT:    v_cvt_f32_f16_e32 v3, s4
; CI-NEXT:    v_cvt_f32_f16_e32 v1, s5
; CI-NEXT:    v_cvt_f32_f16_e32 v0, s2
; CI-NEXT:    v_mov_b32_e32 v5, s1
; CI-NEXT:    v_mov_b32_e32 v4, s0
; CI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
; CI-NEXT:    s_endpgm
;
; VI-LABEL: extload_v4f16_to_v4f32_arg:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_lshr_b32 s4, s3, 16
; VI-NEXT:    s_lshr_b32 s5, s2, 16
; VI-NEXT:    v_cvt_f32_f16_e32 v0, s2
; VI-NEXT:    v_cvt_f32_f16_e32 v3, s4
; VI-NEXT:    v_cvt_f32_f16_e32 v1, s5
; VI-NEXT:    v_cvt_f32_f16_e32 v2, s3
; VI-NEXT:    v_mov_b32_e32 v5, s1
; VI-NEXT:    v_mov_b32_e32 v4, s0
; VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: extload_v4f16_to_v4f32_arg:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v4, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_lshr_b32 s4, s3, 16
; GFX11-NEXT:    s_lshr_b32 s5, s2, 16
; GFX11-NEXT:    v_cvt_f32_f16_e32 v0, s2
; GFX11-NEXT:    v_cvt_f32_f16_e32 v3, s4
; GFX11-NEXT:    v_cvt_f32_f16_e32 v1, s5
; GFX11-NEXT:    v_cvt_f32_f16_e32 v2, s3
; GFX11-NEXT:    global_store_b128 v4, v[0:3], s[0:1]
; GFX11-NEXT:    s_endpgm
  %ext = fpext <4 x half> %arg to <4 x float>
  store <4 x float> %ext, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @extload_v8f16_to_v8f32_arg(ptr addrspace(1) %out, <8 x half> %arg) #0 {
; CI-LABEL: extload_v8f16_to_v8f32_arg:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x4
; CI-NEXT:    s_load_dwordx2 s[4:5], s[8:9], 0x0
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    s_lshr_b32 s6, s1, 16
; CI-NEXT:    s_lshr_b32 s7, s0, 16
; CI-NEXT:    s_lshr_b32 s8, s3, 16
; CI-NEXT:    v_cvt_f32_f16_e32 v3, s6
; CI-NEXT:    s_lshr_b32 s6, s2, 16
; CI-NEXT:    v_cvt_f32_f16_e32 v7, s8
; CI-NEXT:    v_cvt_f32_f16_e32 v5, s6
; CI-NEXT:    v_cvt_f32_f16_e32 v0, s0
; CI-NEXT:    v_cvt_f32_f16_e32 v6, s3
; CI-NEXT:    v_cvt_f32_f16_e32 v4, s2
; CI-NEXT:    s_add_u32 s0, s4, 16
; CI-NEXT:    v_cvt_f32_f16_e32 v2, s1
; CI-NEXT:    s_addc_u32 s1, s5, 0
; CI-NEXT:    v_cvt_f32_f16_e32 v1, s7
; CI-NEXT:    v_mov_b32_e32 v9, s1
; CI-NEXT:    v_mov_b32_e32 v8, s0
; CI-NEXT:    flat_store_dwordx4 v[8:9], v[4:7]
; CI-NEXT:    s_nop 0
; CI-NEXT:    v_mov_b32_e32 v4, s4
; CI-NEXT:    v_mov_b32_e32 v5, s5
; CI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
; CI-NEXT:    s_endpgm
;
; VI-LABEL: extload_v8f16_to_v8f32_arg:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x10
; VI-NEXT:    s_load_dwordx2 s[4:5], s[8:9], 0x0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_lshr_b32 s6, s1, 16
; VI-NEXT:    s_lshr_b32 s7, s0, 16
; VI-NEXT:    s_lshr_b32 s8, s3, 16
; VI-NEXT:    v_cvt_f32_f16_e32 v3, s6
; VI-NEXT:    s_lshr_b32 s6, s2, 16
; VI-NEXT:    v_cvt_f32_f16_e32 v7, s8
; VI-NEXT:    v_cvt_f32_f16_e32 v5, s6
; VI-NEXT:    v_cvt_f32_f16_e32 v0, s0
; VI-NEXT:    v_cvt_f32_f16_e32 v6, s3
; VI-NEXT:    v_cvt_f32_f16_e32 v4, s2
; VI-NEXT:    s_add_u32 s0, s4, 16
; VI-NEXT:    v_cvt_f32_f16_e32 v2, s1
; VI-NEXT:    s_addc_u32 s1, s5, 0
; VI-NEXT:    v_cvt_f32_f16_e32 v1, s7
; VI-NEXT:    v_mov_b32_e32 v9, s1
; VI-NEXT:    v_mov_b32_e32 v8, s0
; VI-NEXT:    flat_store_dwordx4 v[8:9], v[4:7]
; VI-NEXT:    s_nop 0
; VI-NEXT:    v_mov_b32_e32 v4, s4
; VI-NEXT:    v_mov_b32_e32 v5, s5
; VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: extload_v8f16_to_v8f32_arg:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x10
; GFX11-NEXT:    s_load_b64 s[4:5], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v8, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_lshr_b32 s8, s3, 16
; GFX11-NEXT:    s_lshr_b32 s9, s2, 16
; GFX11-NEXT:    s_lshr_b32 s6, s1, 16
; GFX11-NEXT:    s_lshr_b32 s7, s0, 16
; GFX11-NEXT:    v_cvt_f32_f16_e32 v6, s3
; GFX11-NEXT:    v_cvt_f32_f16_e32 v4, s2
; GFX11-NEXT:    v_cvt_f32_f16_e32 v7, s8
; GFX11-NEXT:    v_cvt_f32_f16_e32 v5, s9
; GFX11-NEXT:    v_cvt_f32_f16_e32 v2, s1
; GFX11-NEXT:    v_cvt_f32_f16_e32 v0, s0
; GFX11-NEXT:    v_cvt_f32_f16_e32 v3, s6
; GFX11-NEXT:    v_cvt_f32_f16_e32 v1, s7
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    global_store_b128 v8, v[4:7], s[4:5] offset:16
; GFX11-NEXT:    global_store_b128 v8, v[0:3], s[4:5]
; GFX11-NEXT:    s_endpgm
  %ext = fpext <8 x half> %arg to <8 x float>
  store <8 x float> %ext, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @extload_f16_to_f64_arg(ptr addrspace(1) %out, half %arg) #0 {
; CI-LABEL: extload_f16_to_f64_arg:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dword s0, s[8:9], 0x2
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    v_cvt_f32_f16_e32 v0, s0
; CI-NEXT:    s_load_dwordx2 s[0:1], s[8:9], 0x0
; CI-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    v_mov_b32_e32 v3, s1
; CI-NEXT:    v_mov_b32_e32 v2, s0
; CI-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
; CI-NEXT:    s_endpgm
;
; VI-LABEL: extload_f16_to_f64_arg:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dword s0, s[8:9], 0x8
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_cvt_f32_f16_e32 v0, s0
; VI-NEXT:    s_load_dwordx2 s[0:1], s[8:9], 0x0
; VI-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v3, s1
; VI-NEXT:    v_mov_b32_e32 v2, s0
; VI-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: extload_f16_to_f64_arg:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b32 s0, s[4:5], 0x8
; GFX11-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_cvt_f32_f16_e32 v0, s0
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x0
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-NEXT:    s_endpgm
  %ext = fpext half %arg to double
  store double %ext, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @extload_v2f16_to_v2f64_arg(ptr addrspace(1) %out, <2 x half> %arg) #0 {
; CI-LABEL: extload_v2f16_to_v2f64_arg:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dword s0, s[8:9], 0x2
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    s_lshr_b32 s1, s0, 16
; CI-NEXT:    v_cvt_f32_f16_e32 v0, s1
; CI-NEXT:    v_cvt_f32_f16_e32 v1, s0
; CI-NEXT:    s_load_dwordx2 s[0:1], s[8:9], 0x0
; CI-NEXT:    v_cvt_f64_f32_e32 v[2:3], v0
; CI-NEXT:    v_cvt_f64_f32_e32 v[0:1], v1
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    v_mov_b32_e32 v5, s1
; CI-NEXT:    v_mov_b32_e32 v4, s0
; CI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
; CI-NEXT:    s_endpgm
;
; VI-LABEL: extload_v2f16_to_v2f64_arg:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dword s0, s[8:9], 0x8
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_lshr_b32 s1, s0, 16
; VI-NEXT:    v_cvt_f32_f16_e32 v0, s1
; VI-NEXT:    v_cvt_f32_f16_e32 v1, s0
; VI-NEXT:    s_load_dwordx2 s[0:1], s[8:9], 0x0
; VI-NEXT:    v_cvt_f64_f32_e32 v[2:3], v0
; VI-NEXT:    v_cvt_f64_f32_e32 v[0:1], v1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v5, s1
; VI-NEXT:    v_mov_b32_e32 v4, s0
; VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: extload_v2f16_to_v2f64_arg:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b32 s0, s[4:5], 0x8
; GFX11-NEXT:    v_mov_b32_e32 v4, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_lshr_b32 s1, s0, 16
; GFX11-NEXT:    v_cvt_f32_f16_e32 v1, s0
; GFX11-NEXT:    v_cvt_f32_f16_e32 v0, s1
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x0
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_3)
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[2:3], v0
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[0:1], v1
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    global_store_b128 v4, v[0:3], s[0:1]
; GFX11-NEXT:    s_endpgm
  %ext = fpext <2 x half> %arg to <2 x double>
  store <2 x double> %ext, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @extload_v3f16_to_v3f64_arg(ptr addrspace(1) %out, <3 x half> %arg) #0 {
; CI-LABEL: extload_v3f16_to_v3f64_arg:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    v_cvt_f32_f16_e32 v0, s3
; CI-NEXT:    s_lshr_b32 s4, s2, 16
; CI-NEXT:    v_cvt_f32_f16_e32 v1, s2
; CI-NEXT:    v_cvt_f32_f16_e32 v2, s4
; CI-NEXT:    s_add_u32 s2, s0, 16
; CI-NEXT:    v_cvt_f64_f32_e32 v[4:5], v0
; CI-NEXT:    s_addc_u32 s3, s1, 0
; CI-NEXT:    v_cvt_f64_f32_e32 v[0:1], v1
; CI-NEXT:    v_cvt_f64_f32_e32 v[2:3], v2
; CI-NEXT:    v_mov_b32_e32 v7, s3
; CI-NEXT:    v_mov_b32_e32 v6, s2
; CI-NEXT:    flat_store_dwordx2 v[6:7], v[4:5]
; CI-NEXT:    v_mov_b32_e32 v5, s1
; CI-NEXT:    v_mov_b32_e32 v4, s0
; CI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
; CI-NEXT:    s_endpgm
;
; VI-LABEL: extload_v3f16_to_v3f64_arg:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_cvt_f32_f16_e32 v1, s3
; VI-NEXT:    s_lshr_b32 s4, s2, 16
; VI-NEXT:    v_cvt_f32_f16_e32 v0, s2
; VI-NEXT:    v_cvt_f32_f16_e32 v2, s4
; VI-NEXT:    s_add_u32 s2, s0, 16
; VI-NEXT:    v_cvt_f64_f32_e32 v[4:5], v1
; VI-NEXT:    s_addc_u32 s3, s1, 0
; VI-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; VI-NEXT:    v_cvt_f64_f32_e32 v[2:3], v2
; VI-NEXT:    v_mov_b32_e32 v7, s3
; VI-NEXT:    v_mov_b32_e32 v6, s2
; VI-NEXT:    flat_store_dwordx2 v[6:7], v[4:5]
; VI-NEXT:    v_mov_b32_e32 v5, s1
; VI-NEXT:    v_mov_b32_e32 v4, s0
; VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: extload_v3f16_to_v3f64_arg:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_lshr_b32 s4, s2, 16
; GFX11-NEXT:    v_cvt_f32_f16_e32 v0, s3
; GFX11-NEXT:    v_cvt_f32_f16_e32 v1, s4
; GFX11-NEXT:    v_cvt_f32_f16_e32 v6, s2
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[4:5], v0
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[2:3], v1
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[0:1], v6
; GFX11-NEXT:    v_mov_b32_e32 v6, 0
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    global_store_b64 v6, v[4:5], s[0:1] offset:16
; GFX11-NEXT:    global_store_b128 v6, v[0:3], s[0:1]
; GFX11-NEXT:    s_endpgm
  %ext = fpext <3 x half> %arg to <3 x double>
  store <3 x double> %ext, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @extload_v4f16_to_v4f64_arg(ptr addrspace(1) %out, <4 x half> %arg) #0 {
; CI-LABEL: extload_v4f16_to_v4f64_arg:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    s_lshr_b32 s4, s3, 16
; CI-NEXT:    v_cvt_f32_f16_e32 v0, s3
; CI-NEXT:    v_cvt_f32_f16_e32 v2, s4
; CI-NEXT:    s_lshr_b32 s5, s2, 16
; CI-NEXT:    v_cvt_f32_f16_e32 v4, s2
; CI-NEXT:    v_cvt_f32_f16_e32 v6, s5
; CI-NEXT:    s_add_u32 s2, s0, 16
; CI-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; CI-NEXT:    v_cvt_f64_f32_e32 v[2:3], v2
; CI-NEXT:    s_addc_u32 s3, s1, 0
; CI-NEXT:    v_cvt_f64_f32_e32 v[4:5], v4
; CI-NEXT:    v_cvt_f64_f32_e32 v[6:7], v6
; CI-NEXT:    v_mov_b32_e32 v9, s3
; CI-NEXT:    v_mov_b32_e32 v8, s2
; CI-NEXT:    flat_store_dwordx4 v[8:9], v[0:3]
; CI-NEXT:    s_nop 0
; CI-NEXT:    v_mov_b32_e32 v0, s0
; CI-NEXT:    v_mov_b32_e32 v1, s1
; CI-NEXT:    flat_store_dwordx4 v[0:1], v[4:7]
; CI-NEXT:    s_endpgm
;
; VI-LABEL: extload_v4f16_to_v4f64_arg:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_lshr_b32 s5, s3, 16
; VI-NEXT:    v_cvt_f32_f16_e32 v0, s3
; VI-NEXT:    v_cvt_f32_f16_e32 v2, s5
; VI-NEXT:    s_lshr_b32 s4, s2, 16
; VI-NEXT:    v_cvt_f32_f16_e32 v4, s2
; VI-NEXT:    v_cvt_f32_f16_e32 v6, s4
; VI-NEXT:    s_add_u32 s2, s0, 16
; VI-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; VI-NEXT:    v_cvt_f64_f32_e32 v[2:3], v2
; VI-NEXT:    s_addc_u32 s3, s1, 0
; VI-NEXT:    v_cvt_f64_f32_e32 v[4:5], v4
; VI-NEXT:    v_cvt_f64_f32_e32 v[6:7], v6
; VI-NEXT:    v_mov_b32_e32 v9, s3
; VI-NEXT:    v_mov_b32_e32 v8, s2
; VI-NEXT:    flat_store_dwordx4 v[8:9], v[0:3]
; VI-NEXT:    s_nop 0
; VI-NEXT:    v_mov_b32_e32 v0, s0
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    flat_store_dwordx4 v[0:1], v[4:7]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: extload_v4f16_to_v4f64_arg:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_lshr_b32 s5, s3, 16
; GFX11-NEXT:    s_lshr_b32 s4, s2, 16
; GFX11-NEXT:    v_cvt_f32_f16_e32 v2, s3
; GFX11-NEXT:    v_cvt_f32_f16_e32 v3, s5
; GFX11-NEXT:    v_cvt_f32_f16_e32 v0, s2
; GFX11-NEXT:    v_cvt_f32_f16_e32 v8, s4
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[4:5], v2
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[6:7], v3
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[2:3], v8
; GFX11-NEXT:    v_mov_b32_e32 v8, 0
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    global_store_b128 v8, v[4:7], s[0:1] offset:16
; GFX11-NEXT:    global_store_b128 v8, v[0:3], s[0:1]
; GFX11-NEXT:    s_endpgm
  %ext = fpext <4 x half> %arg to <4 x double>
  store <4 x double> %ext, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @extload_v8f16_to_v8f64_arg(ptr addrspace(1) %out, <8 x half> %arg) #0 {
; CI-LABEL: extload_v8f16_to_v8f64_arg:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x4
; CI-NEXT:    s_load_dwordx2 s[4:5], s[8:9], 0x0
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    s_lshr_b32 s6, s3, 16
; CI-NEXT:    v_cvt_f32_f16_e32 v0, s6
; CI-NEXT:    v_cvt_f32_f16_e32 v12, s3
; CI-NEXT:    s_lshr_b32 s7, s2, 16
; CI-NEXT:    s_lshr_b32 s8, s1, 16
; CI-NEXT:    s_lshr_b32 s6, s0, 16
; CI-NEXT:    v_cvt_f32_f16_e32 v1, s7
; CI-NEXT:    v_cvt_f32_f16_e32 v8, s2
; CI-NEXT:    v_cvt_f32_f16_e32 v9, s0
; CI-NEXT:    s_add_u32 s0, s4, 48
; CI-NEXT:    v_cvt_f32_f16_e32 v5, s1
; CI-NEXT:    v_cvt_f64_f32_e32 v[14:15], v0
; CI-NEXT:    v_cvt_f64_f32_e32 v[12:13], v12
; CI-NEXT:    s_addc_u32 s1, s5, 0
; CI-NEXT:    v_cvt_f32_f16_e32 v4, s8
; CI-NEXT:    v_mov_b32_e32 v17, s1
; CI-NEXT:    v_mov_b32_e32 v16, s0
; CI-NEXT:    s_add_u32 s0, s4, 32
; CI-NEXT:    v_cvt_f32_f16_e32 v2, s6
; CI-NEXT:    v_cvt_f64_f32_e32 v[10:11], v1
; CI-NEXT:    v_cvt_f64_f32_e32 v[0:1], v9
; CI-NEXT:    v_cvt_f64_f32_e32 v[8:9], v8
; CI-NEXT:    s_addc_u32 s1, s5, 0
; CI-NEXT:    flat_store_dwordx4 v[16:17], v[12:15]
; CI-NEXT:    v_cvt_f64_f32_e32 v[6:7], v4
; CI-NEXT:    v_mov_b32_e32 v13, s1
; CI-NEXT:    v_cvt_f64_f32_e32 v[4:5], v5
; CI-NEXT:    v_mov_b32_e32 v12, s0
; CI-NEXT:    s_add_u32 s0, s4, 16
; CI-NEXT:    s_addc_u32 s1, s5, 0
; CI-NEXT:    v_cvt_f64_f32_e32 v[2:3], v2
; CI-NEXT:    flat_store_dwordx4 v[12:13], v[8:11]
; CI-NEXT:    s_nop 0
; CI-NEXT:    v_mov_b32_e32 v9, s1
; CI-NEXT:    v_mov_b32_e32 v8, s0
; CI-NEXT:    flat_store_dwordx4 v[8:9], v[4:7]
; CI-NEXT:    s_nop 0
; CI-NEXT:    v_mov_b32_e32 v4, s4
; CI-NEXT:    v_mov_b32_e32 v5, s5
; CI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
; CI-NEXT:    s_endpgm
;
; VI-LABEL: extload_v8f16_to_v8f64_arg:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x10
; VI-NEXT:    s_load_dwordx2 s[4:5], s[8:9], 0x0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_lshr_b32 s6, s0, 16
; VI-NEXT:    s_lshr_b32 s8, s2, 16
; VI-NEXT:    s_lshr_b32 s9, s3, 16
; VI-NEXT:    v_cvt_f32_f16_e32 v0, s6
; VI-NEXT:    v_cvt_f32_f16_e32 v4, s8
; VI-NEXT:    v_cvt_f32_f16_e32 v5, s9
; VI-NEXT:    v_cvt_f32_f16_e32 v12, s3
; VI-NEXT:    s_lshr_b32 s7, s1, 16
; VI-NEXT:    v_cvt_f64_f32_e32 v[2:3], v0
; VI-NEXT:    v_cvt_f32_f16_e32 v0, s0
; VI-NEXT:    v_cvt_f32_f16_e32 v8, s2
; VI-NEXT:    s_add_u32 s0, s4, 48
; VI-NEXT:    v_cvt_f64_f32_e32 v[10:11], v4
; VI-NEXT:    v_cvt_f64_f32_e32 v[14:15], v5
; VI-NEXT:    v_cvt_f32_f16_e32 v4, s1
; VI-NEXT:    v_cvt_f64_f32_e32 v[12:13], v12
; VI-NEXT:    s_addc_u32 s1, s5, 0
; VI-NEXT:    v_cvt_f32_f16_e32 v1, s7
; VI-NEXT:    v_mov_b32_e32 v17, s1
; VI-NEXT:    v_mov_b32_e32 v16, s0
; VI-NEXT:    s_add_u32 s0, s4, 32
; VI-NEXT:    v_cvt_f64_f32_e32 v[8:9], v8
; VI-NEXT:    s_addc_u32 s1, s5, 0
; VI-NEXT:    flat_store_dwordx4 v[16:17], v[12:15]
; VI-NEXT:    v_cvt_f64_f32_e32 v[6:7], v1
; VI-NEXT:    v_mov_b32_e32 v13, s1
; VI-NEXT:    v_cvt_f64_f32_e32 v[4:5], v4
; VI-NEXT:    v_mov_b32_e32 v12, s0
; VI-NEXT:    s_add_u32 s0, s4, 16
; VI-NEXT:    s_addc_u32 s1, s5, 0
; VI-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; VI-NEXT:    flat_store_dwordx4 v[12:13], v[8:11]
; VI-NEXT:    s_nop 0
; VI-NEXT:    v_mov_b32_e32 v9, s1
; VI-NEXT:    v_mov_b32_e32 v8, s0
; VI-NEXT:    flat_store_dwordx4 v[8:9], v[4:7]
; VI-NEXT:    s_nop 0
; VI-NEXT:    v_mov_b32_e32 v4, s4
; VI-NEXT:    v_mov_b32_e32 v5, s5
; VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: extload_v8f16_to_v8f64_arg:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x10
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_lshr_b32 s9, s3, 16
; GFX11-NEXT:    s_lshr_b32 s8, s2, 16
; GFX11-NEXT:    s_lshr_b32 s7, s1, 16
; GFX11-NEXT:    v_cvt_f32_f16_e32 v6, s3
; GFX11-NEXT:    v_cvt_f32_f16_e32 v11, s9
; GFX11-NEXT:    s_lshr_b32 s6, s0, 16
; GFX11-NEXT:    v_cvt_f32_f16_e32 v3, s2
; GFX11-NEXT:    v_cvt_f32_f16_e32 v10, s8
; GFX11-NEXT:    v_cvt_f32_f16_e32 v2, s1
; GFX11-NEXT:    v_cvt_f32_f16_e32 v7, s7
; GFX11-NEXT:    v_cvt_f32_f16_e32 v0, s0
; GFX11-NEXT:    v_cvt_f32_f16_e32 v16, s6
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[12:13], v6
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[14:15], v11
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[8:9], v3
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[10:11], v10
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[4:5], v2
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[6:7], v7
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[2:3], v16
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v16, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_clause 0x3
; GFX11-NEXT:    global_store_b128 v16, v[12:15], s[0:1] offset:48
; GFX11-NEXT:    global_store_b128 v16, v[8:11], s[0:1] offset:32
; GFX11-NEXT:    global_store_b128 v16, v[4:7], s[0:1] offset:16
; GFX11-NEXT:    global_store_b128 v16, v[0:3], s[0:1]
; GFX11-NEXT:    s_endpgm
  %ext = fpext <8 x half> %arg to <8 x double>
  store <8 x double> %ext, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @global_load_store_f16(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
; CIVI-LABEL: global_load_store_f16:
; CIVI:       ; %bb.0:
; CIVI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; CIVI-NEXT:    s_waitcnt lgkmcnt(0)
; CIVI-NEXT:    v_mov_b32_e32 v0, s2
; CIVI-NEXT:    v_mov_b32_e32 v1, s3
; CIVI-NEXT:    flat_load_ushort v2, v[0:1]
; CIVI-NEXT:    v_mov_b32_e32 v0, s0
; CIVI-NEXT:    v_mov_b32_e32 v1, s1
; CIVI-NEXT:    s_waitcnt vmcnt(0)
; CIVI-NEXT:    flat_store_short v[0:1], v2
; CIVI-NEXT:    s_endpgm
;
; GFX11-LABEL: global_load_store_f16:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v0, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    global_load_u16 v1, v0, s[2:3]
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    global_store_b16 v0, v1, s[0:1]
; GFX11-NEXT:    s_endpgm
  %val = load half, ptr addrspace(1) %in
  store half %val, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @global_load_store_v2f16(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
; CIVI-LABEL: global_load_store_v2f16:
; CIVI:       ; %bb.0:
; CIVI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; CIVI-NEXT:    s_waitcnt lgkmcnt(0)
; CIVI-NEXT:    v_mov_b32_e32 v0, s2
; CIVI-NEXT:    v_mov_b32_e32 v1, s3
; CIVI-NEXT:    flat_load_dword v2, v[0:1]
; CIVI-NEXT:    v_mov_b32_e32 v0, s0
; CIVI-NEXT:    v_mov_b32_e32 v1, s1
; CIVI-NEXT:    s_waitcnt vmcnt(0)
; CIVI-NEXT:    flat_store_dword v[0:1], v2
; CIVI-NEXT:    s_endpgm
;
; GFX11-LABEL: global_load_store_v2f16:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v0, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX11-NEXT:    s_endpgm
  %val = load <2 x half>, ptr addrspace(1) %in
  store <2 x half> %val, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @global_load_store_v4f16(ptr addrspace(1) %in, ptr addrspace(1) %out) #0 {
; CIVI-LABEL: global_load_store_v4f16:
; CIVI:       ; %bb.0:
; CIVI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; CIVI-NEXT:    s_waitcnt lgkmcnt(0)
; CIVI-NEXT:    v_mov_b32_e32 v0, s0
; CIVI-NEXT:    v_mov_b32_e32 v1, s1
; CIVI-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
; CIVI-NEXT:    v_mov_b32_e32 v2, s2
; CIVI-NEXT:    v_mov_b32_e32 v3, s3
; CIVI-NEXT:    s_waitcnt vmcnt(0)
; CIVI-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
; CIVI-NEXT:    s_endpgm
;
; GFX11-LABEL: global_load_store_v4f16:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    global_load_b64 v[0:1], v2, s[0:1]
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    global_store_b64 v2, v[0:1], s[2:3]
; GFX11-NEXT:    s_endpgm
  %val = load <4 x half>, ptr addrspace(1) %in
  store <4 x half> %val, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @global_load_store_v8f16(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
; CIVI-LABEL: global_load_store_v8f16:
; CIVI:       ; %bb.0:
; CIVI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; CIVI-NEXT:    s_waitcnt lgkmcnt(0)
; CIVI-NEXT:    v_mov_b32_e32 v0, s2
; CIVI-NEXT:    v_mov_b32_e32 v1, s3
; CIVI-NEXT:    flat_load_dwordx4 v[0:3], v[0:1]
; CIVI-NEXT:    v_mov_b32_e32 v4, s0
; CIVI-NEXT:    v_mov_b32_e32 v5, s1
; CIVI-NEXT:    s_waitcnt vmcnt(0)
; CIVI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
; CIVI-NEXT:    s_endpgm
;
; GFX11-LABEL: global_load_store_v8f16:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v4, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    global_load_b128 v[0:3], v4, s[2:3]
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    global_store_b128 v4, v[0:3], s[0:1]
; GFX11-NEXT:    s_endpgm
  %val = load <8 x half>, ptr addrspace(1) %in
  store <8 x half> %val, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @global_extload_f16_to_f32(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
; CIVI-LABEL: global_extload_f16_to_f32:
; CIVI:       ; %bb.0:
; CIVI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; CIVI-NEXT:    s_waitcnt lgkmcnt(0)
; CIVI-NEXT:    v_mov_b32_e32 v0, s2
; CIVI-NEXT:    v_mov_b32_e32 v1, s3
; CIVI-NEXT:    flat_load_ushort v0, v[0:1]
; CIVI-NEXT:    v_mov_b32_e32 v1, s1
; CIVI-NEXT:    s_waitcnt vmcnt(0)
; CIVI-NEXT:    v_cvt_f32_f16_e32 v2, v0
; CIVI-NEXT:    v_mov_b32_e32 v0, s0
; CIVI-NEXT:    flat_store_dword v[0:1], v2
; CIVI-NEXT:    s_endpgm
;
; GFX11-LABEL: global_extload_f16_to_f32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v0, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    global_load_u16 v1, v0, s[2:3]
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    v_cvt_f32_f16_e32 v1, v1
; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX11-NEXT:    s_endpgm
  %val = load half, ptr addrspace(1) %in
  %cvt = fpext half %val to float
  store float %cvt, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @global_extload_v2f16_to_v2f32(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
; CI-LABEL: global_extload_v2f16_to_v2f32:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    v_mov_b32_e32 v0, s2
; CI-NEXT:    v_mov_b32_e32 v1, s3
; CI-NEXT:    flat_load_dword v1, v[0:1]
; CI-NEXT:    v_mov_b32_e32 v2, s0
; CI-NEXT:    v_mov_b32_e32 v3, s1
; CI-NEXT:    s_waitcnt vmcnt(0)
; CI-NEXT:    v_cvt_f32_f16_e32 v0, v1
; CI-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
; CI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; CI-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
; CI-NEXT:    s_endpgm
;
; VI-LABEL: global_extload_v2f16_to_v2f32:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s2
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    flat_load_dword v1, v[0:1]
; VI-NEXT:    v_mov_b32_e32 v2, s0
; VI-NEXT:    v_mov_b32_e32 v3, s1
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_cvt_f32_f16_e32 v0, v1
; VI-NEXT:    v_cvt_f32_f16_sdwa v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; VI-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: global_extload_v2f16_to_v2f32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    global_load_b32 v0, v2, s[2:3]
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    v_lshrrev_b32_e32 v1, 16, v0
; GFX11-NEXT:    v_cvt_f32_f16_e32 v0, v0
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11-NEXT:    v_cvt_f32_f16_e32 v1, v1
; GFX11-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-NEXT:    s_endpgm
  %val = load <2 x half>, ptr addrspace(1) %in
  %cvt = fpext <2 x half> %val to <2 x float>
  store <2 x float> %cvt, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @global_extload_v3f16_to_v3f32(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
; CI-LABEL: global_extload_v3f16_to_v3f32:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    v_mov_b32_e32 v0, s2
; CI-NEXT:    v_mov_b32_e32 v1, s3
; CI-NEXT:    flat_load_dwordx2 v[1:2], v[0:1]
; CI-NEXT:    v_mov_b32_e32 v3, s0
; CI-NEXT:    v_mov_b32_e32 v4, s1
; CI-NEXT:    s_waitcnt vmcnt(0)
; CI-NEXT:    v_cvt_f32_f16_e32 v0, v1
; CI-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
; CI-NEXT:    v_cvt_f32_f16_e32 v2, v2
; CI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; CI-NEXT:    flat_store_dwordx3 v[3:4], v[0:2]
; CI-NEXT:    s_endpgm
;
; VI-LABEL: global_extload_v3f16_to_v3f32:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s2
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    flat_load_dwordx2 v[1:2], v[0:1]
; VI-NEXT:    v_mov_b32_e32 v3, s0
; VI-NEXT:    v_mov_b32_e32 v4, s1
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_cvt_f32_f16_e32 v0, v1
; VI-NEXT:    v_cvt_f32_f16_e32 v2, v2
; VI-NEXT:    v_cvt_f32_f16_sdwa v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; VI-NEXT:    flat_store_dwordx3 v[3:4], v[0:2]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: global_extload_v3f16_to_v3f32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v3, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    global_load_b64 v[0:1], v3, s[2:3]
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    v_lshrrev_b32_e32 v4, 16, v0
; GFX11-NEXT:    v_cvt_f32_f16_e32 v0, v0
; GFX11-NEXT:    v_cvt_f32_f16_e32 v2, v1
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
; GFX11-NEXT:    v_cvt_f32_f16_e32 v1, v4
; GFX11-NEXT:    global_store_b96 v3, v[0:2], s[0:1]
; GFX11-NEXT:    s_endpgm
  %val = load <3 x half>, ptr addrspace(1) %in
  %cvt = fpext <3 x half> %val to <3 x float>
  store <3 x float> %cvt, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @global_extload_v4f16_to_v4f32(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
; CI-LABEL: global_extload_v4f16_to_v4f32:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    v_mov_b32_e32 v0, s2
; CI-NEXT:    v_mov_b32_e32 v1, s3
; CI-NEXT:    flat_load_dwordx2 v[3:4], v[0:1]
; CI-NEXT:    v_mov_b32_e32 v5, s1
; CI-NEXT:    s_waitcnt vmcnt(0)
; CI-NEXT:    v_cvt_f32_f16_e32 v2, v4
; CI-NEXT:    v_lshrrev_b32_e32 v1, 16, v4
; CI-NEXT:    v_lshrrev_b32_e32 v4, 16, v3
; CI-NEXT:    v_cvt_f32_f16_e32 v0, v3
; CI-NEXT:    v_cvt_f32_f16_e32 v3, v1
; CI-NEXT:    v_cvt_f32_f16_e32 v1, v4
; CI-NEXT:    v_mov_b32_e32 v4, s0
; CI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
; CI-NEXT:    s_endpgm
;
; VI-LABEL: global_extload_v4f16_to_v4f32:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s2
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    flat_load_dwordx2 v[4:5], v[0:1]
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_cvt_f32_f16_e32 v0, v4
; VI-NEXT:    v_cvt_f32_f16_e32 v2, v5
; VI-NEXT:    v_cvt_f32_f16_sdwa v3, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; VI-NEXT:    v_cvt_f32_f16_sdwa v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; VI-NEXT:    v_mov_b32_e32 v4, s0
; VI-NEXT:    v_mov_b32_e32 v5, s1
; VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: global_extload_v4f16_to_v4f32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v4, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    global_load_b64 v[0:1], v4, s[2:3]
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
; GFX11-NEXT:    v_lshrrev_b32_e32 v5, 16, v0
; GFX11-NEXT:    v_cvt_f32_f16_e32 v0, v0
; GFX11-NEXT:    v_cvt_f32_f16_e32 v2, v1
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
; GFX11-NEXT:    v_cvt_f32_f16_e32 v3, v3
; GFX11-NEXT:    v_cvt_f32_f16_e32 v1, v5
; GFX11-NEXT:    global_store_b128 v4, v[0:3], s[0:1]
; GFX11-NEXT:    s_endpgm
  %val = load <4 x half>, ptr addrspace(1) %in
  %cvt = fpext <4 x half> %val to <4 x float>
  store <4 x float> %cvt, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @global_extload_v8f16_to_v8f32(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
; CI-LABEL: global_extload_v8f16_to_v8f32:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    v_mov_b32_e32 v0, s2
; CI-NEXT:    v_mov_b32_e32 v1, s3
; CI-NEXT:    flat_load_dwordx4 v[0:3], v[0:1]
; CI-NEXT:    s_add_u32 s2, s0, 16
; CI-NEXT:    s_addc_u32 s3, s1, 0
; CI-NEXT:    v_mov_b32_e32 v13, s1
; CI-NEXT:    v_mov_b32_e32 v12, s0
; CI-NEXT:    s_waitcnt vmcnt(0)
; CI-NEXT:    v_cvt_f32_f16_e32 v10, v3
; CI-NEXT:    v_cvt_f32_f16_e32 v8, v2
; CI-NEXT:    v_lshrrev_b32_e32 v3, 16, v3
; CI-NEXT:    v_lshrrev_b32_e32 v2, 16, v2
; CI-NEXT:    v_cvt_f32_f16_e32 v6, v1
; CI-NEXT:    v_cvt_f32_f16_e32 v4, v0
; CI-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
; CI-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
; CI-NEXT:    v_cvt_f32_f16_e32 v11, v3
; CI-NEXT:    v_cvt_f32_f16_e32 v9, v2
; CI-NEXT:    v_cvt_f32_f16_e32 v7, v1
; CI-NEXT:    v_cvt_f32_f16_e32 v5, v0
; CI-NEXT:    v_mov_b32_e32 v0, s2
; CI-NEXT:    v_mov_b32_e32 v1, s3
; CI-NEXT:    flat_store_dwordx4 v[0:1], v[8:11]
; CI-NEXT:    flat_store_dwordx4 v[12:13], v[4:7]
; CI-NEXT:    s_endpgm
;
; VI-LABEL: global_extload_v8f16_to_v8f32:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s2
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    flat_load_dwordx4 v[0:3], v[0:1]
; VI-NEXT:    s_add_u32 s2, s0, 16
; VI-NEXT:    s_addc_u32 s3, s1, 0
; VI-NEXT:    v_mov_b32_e32 v13, s1
; VI-NEXT:    v_mov_b32_e32 v12, s0
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_cvt_f32_f16_e32 v10, v3
; VI-NEXT:    v_cvt_f32_f16_e32 v8, v2
; VI-NEXT:    v_cvt_f32_f16_sdwa v11, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; VI-NEXT:    v_cvt_f32_f16_sdwa v9, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; VI-NEXT:    v_cvt_f32_f16_e32 v6, v1
; VI-NEXT:    v_cvt_f32_f16_e32 v4, v0
; VI-NEXT:    v_cvt_f32_f16_sdwa v7, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; VI-NEXT:    v_cvt_f32_f16_sdwa v5, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; VI-NEXT:    v_mov_b32_e32 v0, s2
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    flat_store_dwordx4 v[0:1], v[8:11]
; VI-NEXT:    flat_store_dwordx4 v[12:13], v[4:7]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: global_extload_v8f16_to_v8f32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v12, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    global_load_b128 v[0:3], v12, s[2:3]
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    v_lshrrev_b32_e32 v5, 16, v3
; GFX11-NEXT:    v_lshrrev_b32_e32 v9, 16, v2
; GFX11-NEXT:    v_cvt_f32_f16_e32 v6, v1
; GFX11-NEXT:    v_cvt_f32_f16_e32 v4, v0
; GFX11-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
; GFX11-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
; GFX11-NEXT:    v_cvt_f32_f16_e32 v10, v3
; GFX11-NEXT:    v_cvt_f32_f16_e32 v8, v2
; GFX11-NEXT:    v_cvt_f32_f16_e32 v11, v5
; GFX11-NEXT:    v_cvt_f32_f16_e32 v9, v9
; GFX11-NEXT:    v_cvt_f32_f16_e32 v7, v1
; GFX11-NEXT:    v_cvt_f32_f16_e32 v5, v0
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    global_store_b128 v12, v[8:11], s[0:1] offset:16
; GFX11-NEXT:    global_store_b128 v12, v[4:7], s[0:1]
; GFX11-NEXT:    s_endpgm
  %val = load <8 x half>, ptr addrspace(1) %in
  %cvt = fpext <8 x half> %val to <8 x float>
  store <8 x float> %cvt, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @global_extload_v16f16_to_v16f32(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
; CI-LABEL: global_extload_v16f16_to_v16f32:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    s_add_u32 s4, s2, 16
; CI-NEXT:    v_mov_b32_e32 v5, s3
; CI-NEXT:    s_addc_u32 s5, s3, 0
; CI-NEXT:    v_mov_b32_e32 v0, s4
; CI-NEXT:    v_mov_b32_e32 v4, s2
; CI-NEXT:    v_mov_b32_e32 v1, s5
; CI-NEXT:    flat_load_dwordx4 v[0:3], v[0:1]
; CI-NEXT:    flat_load_dwordx4 v[4:7], v[4:5]
; CI-NEXT:    s_add_u32 s2, s0, 16
; CI-NEXT:    s_addc_u32 s3, s1, 0
; CI-NEXT:    v_mov_b32_e32 v14, s3
; CI-NEXT:    v_mov_b32_e32 v13, s2
; CI-NEXT:    s_add_u32 s2, s0, 48
; CI-NEXT:    s_addc_u32 s3, s1, 0
; CI-NEXT:    s_waitcnt vmcnt(1)
; CI-NEXT:    v_cvt_f32_f16_e32 v8, v1
; CI-NEXT:    s_waitcnt vmcnt(0)
; CI-NEXT:    v_cvt_f32_f16_e32 v11, v7
; CI-NEXT:    v_cvt_f32_f16_e32 v9, v6
; CI-NEXT:    v_lshrrev_b32_e32 v7, 16, v7
; CI-NEXT:    v_lshrrev_b32_e32 v6, 16, v6
; CI-NEXT:    v_cvt_f32_f16_e32 v12, v7
; CI-NEXT:    v_cvt_f32_f16_e32 v10, v6
; CI-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
; CI-NEXT:    v_lshrrev_b32_e32 v16, 16, v5
; CI-NEXT:    v_lshrrev_b32_e32 v17, 16, v4
; CI-NEXT:    flat_store_dwordx4 v[13:14], v[9:12]
; CI-NEXT:    v_cvt_f32_f16_e32 v6, v0
; CI-NEXT:    v_cvt_f32_f16_e32 v12, v3
; CI-NEXT:    v_lshrrev_b32_e32 v3, 16, v3
; CI-NEXT:    v_lshrrev_b32_e32 v7, 16, v0
; CI-NEXT:    v_cvt_f32_f16_e32 v10, v2
; CI-NEXT:    v_lshrrev_b32_e32 v11, 16, v2
; CI-NEXT:    v_cvt_f32_f16_e32 v2, v5
; CI-NEXT:    v_cvt_f32_f16_e32 v0, v4
; CI-NEXT:    v_mov_b32_e32 v5, s1
; CI-NEXT:    v_cvt_f32_f16_e32 v9, v1
; CI-NEXT:    v_cvt_f32_f16_e32 v13, v3
; CI-NEXT:    v_cvt_f32_f16_e32 v3, v16
; CI-NEXT:    v_cvt_f32_f16_e32 v1, v17
; CI-NEXT:    v_mov_b32_e32 v4, s0
; CI-NEXT:    s_add_u32 s0, s0, 32
; CI-NEXT:    v_cvt_f32_f16_e32 v11, v11
; CI-NEXT:    s_addc_u32 s1, s1, 0
; CI-NEXT:    v_cvt_f32_f16_e32 v7, v7
; CI-NEXT:    v_mov_b32_e32 v15, s3
; CI-NEXT:    v_mov_b32_e32 v17, s1
; CI-NEXT:    v_mov_b32_e32 v14, s2
; CI-NEXT:    v_mov_b32_e32 v16, s0
; CI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
; CI-NEXT:    flat_store_dwordx4 v[14:15], v[10:13]
; CI-NEXT:    flat_store_dwordx4 v[16:17], v[6:9]
; CI-NEXT:    s_endpgm
;
; VI-LABEL: global_extload_v16f16_to_v16f32:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s2
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    flat_load_dwordx4 v[0:3], v[0:1]
; VI-NEXT:    s_add_u32 s2, s2, 16
; VI-NEXT:    s_addc_u32 s3, s3, 0
; VI-NEXT:    v_mov_b32_e32 v5, s3
; VI-NEXT:    v_mov_b32_e32 v4, s2
; VI-NEXT:    flat_load_dwordx4 v[4:7], v[4:5]
; VI-NEXT:    s_add_u32 s2, s0, 16
; VI-NEXT:    s_addc_u32 s3, s1, 0
; VI-NEXT:    v_mov_b32_e32 v19, s3
; VI-NEXT:    v_mov_b32_e32 v18, s2
; VI-NEXT:    s_add_u32 s2, s0, 48
; VI-NEXT:    v_mov_b32_e32 v17, s1
; VI-NEXT:    s_addc_u32 s3, s1, 0
; VI-NEXT:    v_mov_b32_e32 v16, s0
; VI-NEXT:    s_add_u32 s0, s0, 32
; VI-NEXT:    s_addc_u32 s1, s1, 0
; VI-NEXT:    v_mov_b32_e32 v21, s3
; VI-NEXT:    v_mov_b32_e32 v20, s2
; VI-NEXT:    s_waitcnt vmcnt(1)
; VI-NEXT:    v_cvt_f32_f16_e32 v14, v3
; VI-NEXT:    v_cvt_f32_f16_e32 v12, v2
; VI-NEXT:    v_cvt_f32_f16_sdwa v15, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; VI-NEXT:    v_cvt_f32_f16_sdwa v13, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; VI-NEXT:    v_cvt_f32_f16_e32 v10, v1
; VI-NEXT:    v_cvt_f32_f16_e32 v8, v0
; VI-NEXT:    v_cvt_f32_f16_sdwa v11, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; VI-NEXT:    v_cvt_f32_f16_sdwa v9, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; VI-NEXT:    flat_store_dwordx4 v[18:19], v[12:15]
; VI-NEXT:    s_waitcnt vmcnt(1)
; VI-NEXT:    v_cvt_f32_f16_e32 v2, v5
; VI-NEXT:    v_cvt_f32_f16_e32 v14, v7
; VI-NEXT:    v_cvt_f32_f16_e32 v12, v6
; VI-NEXT:    v_cvt_f32_f16_sdwa v15, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; VI-NEXT:    v_cvt_f32_f16_sdwa v13, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; VI-NEXT:    v_cvt_f32_f16_e32 v0, v4
; VI-NEXT:    v_cvt_f32_f16_sdwa v3, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; VI-NEXT:    v_cvt_f32_f16_sdwa v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; VI-NEXT:    v_mov_b32_e32 v5, s1
; VI-NEXT:    v_mov_b32_e32 v4, s0
; VI-NEXT:    flat_store_dwordx4 v[16:17], v[8:11]
; VI-NEXT:    flat_store_dwordx4 v[20:21], v[12:15]
; VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: global_extload_v16f16_to_v16f32:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v20, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    global_load_b128 v[0:3], v20, s[2:3]
; GFX11-NEXT:    global_load_b128 v[4:7], v20, s[2:3] offset:16
; GFX11-NEXT:    s_waitcnt vmcnt(1)
; GFX11-NEXT:    v_cvt_f32_f16_e32 v10, v1
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    v_cvt_f32_f16_e32 v18, v7
; GFX11-NEXT:    v_cvt_f32_f16_e32 v16, v6
; GFX11-NEXT:    v_lshrrev_b32_e32 v7, 16, v7
; GFX11-NEXT:    v_lshrrev_b32_e32 v6, 16, v6
; GFX11-NEXT:    v_cvt_f32_f16_e32 v8, v0
; GFX11-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
; GFX11-NEXT:    v_lshrrev_b32_e32 v9, 16, v0
; GFX11-NEXT:    v_cvt_f32_f16_e32 v14, v3
; GFX11-NEXT:    v_cvt_f32_f16_e32 v12, v2
; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v3
; GFX11-NEXT:    v_lshrrev_b32_e32 v13, 16, v2
; GFX11-NEXT:    v_cvt_f32_f16_e32 v2, v5
; GFX11-NEXT:    v_cvt_f32_f16_e32 v0, v4
; GFX11-NEXT:    v_lshrrev_b32_e32 v5, 16, v5
; GFX11-NEXT:    v_lshrrev_b32_e32 v4, 16, v4
; GFX11-NEXT:    v_cvt_f32_f16_e32 v19, v7
; GFX11-NEXT:    v_cvt_f32_f16_e32 v17, v6
; GFX11-NEXT:    v_cvt_f32_f16_e32 v11, v1
; GFX11-NEXT:    v_cvt_f32_f16_e32 v15, v3
; GFX11-NEXT:    v_cvt_f32_f16_e32 v3, v5
; GFX11-NEXT:    v_cvt_f32_f16_e32 v1, v4
; GFX11-NEXT:    v_cvt_f32_f16_e32 v13, v13
; GFX11-NEXT:    v_cvt_f32_f16_e32 v9, v9
; GFX11-NEXT:    s_clause 0x3
; GFX11-NEXT:    global_store_b128 v20, v[16:19], s[0:1] offset:48
; GFX11-NEXT:    global_store_b128 v20, v[0:3], s[0:1] offset:32
; GFX11-NEXT:    global_store_b128 v20, v[12:15], s[0:1] offset:16
; GFX11-NEXT:    global_store_b128 v20, v[8:11], s[0:1]
; GFX11-NEXT:    s_endpgm
  %val = load <16 x half>, ptr addrspace(1) %in
  %cvt = fpext <16 x half> %val to <16 x float>
  store <16 x float> %cvt, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @global_extload_f16_to_f64(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
; CIVI-LABEL: global_extload_f16_to_f64:
; CIVI:       ; %bb.0:
; CIVI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; CIVI-NEXT:    s_waitcnt lgkmcnt(0)
; CIVI-NEXT:    v_mov_b32_e32 v0, s2
; CIVI-NEXT:    v_mov_b32_e32 v1, s3
; CIVI-NEXT:    flat_load_ushort v0, v[0:1]
; CIVI-NEXT:    v_mov_b32_e32 v2, s0
; CIVI-NEXT:    v_mov_b32_e32 v3, s1
; CIVI-NEXT:    s_waitcnt vmcnt(0)
; CIVI-NEXT:    v_cvt_f32_f16_e32 v0, v0
; CIVI-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; CIVI-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
; CIVI-NEXT:    s_endpgm
;
; GFX11-LABEL: global_extload_f16_to_f64:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    global_load_u16 v0, v2, s[2:3]
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    v_cvt_f32_f16_e32 v0, v0
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX11-NEXT:    global_store_b64 v2, v[0:1], s[0:1]
; GFX11-NEXT:    s_endpgm
  %val = load half, ptr addrspace(1) %in
  %cvt = fpext half %val to double
  store double %cvt, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @global_extload_v2f16_to_v2f64(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
; CI-LABEL: global_extload_v2f16_to_v2f64:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    v_mov_b32_e32 v0, s2
; CI-NEXT:    v_mov_b32_e32 v1, s3
; CI-NEXT:    flat_load_dword v0, v[0:1]
; CI-NEXT:    v_mov_b32_e32 v4, s0
; CI-NEXT:    v_mov_b32_e32 v5, s1
; CI-NEXT:    s_waitcnt vmcnt(0)
; CI-NEXT:    v_lshrrev_b32_e32 v1, 16, v0
; CI-NEXT:    v_cvt_f32_f16_e32 v0, v0
; CI-NEXT:    v_cvt_f32_f16_e32 v2, v1
; CI-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; CI-NEXT:    v_cvt_f64_f32_e32 v[2:3], v2
; CI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
; CI-NEXT:    s_endpgm
;
; VI-LABEL: global_extload_v2f16_to_v2f64:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s2
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    flat_load_dword v0, v[0:1]
; VI-NEXT:    v_mov_b32_e32 v4, s0
; VI-NEXT:    v_mov_b32_e32 v5, s1
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_cvt_f32_f16_e32 v1, v0
; VI-NEXT:    v_cvt_f32_f16_sdwa v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; VI-NEXT:    v_cvt_f64_f32_e32 v[0:1], v1
; VI-NEXT:    v_cvt_f64_f32_e32 v[2:3], v2
; VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: global_extload_v2f16_to_v2f64:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v4, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    global_load_b32 v0, v4, s[2:3]
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    v_lshrrev_b32_e32 v1, 16, v0
; GFX11-NEXT:    v_cvt_f32_f16_e32 v0, v0
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-NEXT:    v_cvt_f32_f16_e32 v2, v1
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[2:3], v2
; GFX11-NEXT:    global_store_b128 v4, v[0:3], s[0:1]
; GFX11-NEXT:    s_endpgm
  %val = load <2 x half>, ptr addrspace(1) %in
  %cvt = fpext <2 x half> %val to <2 x double>
  store <2 x double> %cvt, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @global_extload_v3f16_to_v3f64(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
; CI-LABEL: global_extload_v3f16_to_v3f64:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    v_mov_b32_e32 v0, s2
; CI-NEXT:    v_mov_b32_e32 v1, s3
; CI-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
; CI-NEXT:    s_add_u32 s2, s0, 16
; CI-NEXT:    s_addc_u32 s3, s1, 0
; CI-NEXT:    v_mov_b32_e32 v7, s3
; CI-NEXT:    v_mov_b32_e32 v6, s2
; CI-NEXT:    s_waitcnt vmcnt(0)
; CI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; CI-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
; CI-NEXT:    v_cvt_f32_f16_e32 v0, v0
; CI-NEXT:    v_cvt_f32_f16_e32 v2, v2
; CI-NEXT:    v_cvt_f64_f32_e32 v[4:5], v1
; CI-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; CI-NEXT:    v_cvt_f64_f32_e32 v[2:3], v2
; CI-NEXT:    flat_store_dwordx2 v[6:7], v[4:5]
; CI-NEXT:    v_mov_b32_e32 v5, s1
; CI-NEXT:    v_mov_b32_e32 v4, s0
; CI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
; CI-NEXT:    s_endpgm
;
; VI-LABEL: global_extload_v3f16_to_v3f64:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s2
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
; VI-NEXT:    s_add_u32 s2, s0, 16
; VI-NEXT:    s_addc_u32 s3, s1, 0
; VI-NEXT:    v_mov_b32_e32 v5, s1
; VI-NEXT:    v_mov_b32_e32 v4, s0
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_cvt_f32_f16_e32 v3, v1
; VI-NEXT:    v_cvt_f32_f16_e32 v2, v0
; VI-NEXT:    v_cvt_f32_f16_sdwa v8, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; VI-NEXT:    v_cvt_f64_f32_e32 v[6:7], v3
; VI-NEXT:    v_cvt_f64_f32_e32 v[0:1], v2
; VI-NEXT:    v_cvt_f64_f32_e32 v[2:3], v8
; VI-NEXT:    v_mov_b32_e32 v9, s3
; VI-NEXT:    v_mov_b32_e32 v8, s2
; VI-NEXT:    flat_store_dwordx2 v[8:9], v[6:7]
; VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: global_extload_v3f16_to_v3f64:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v6, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    global_load_b64 v[0:1], v6, s[2:3]
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
; GFX11-NEXT:    v_cvt_f32_f16_e32 v3, v1
; GFX11-NEXT:    v_cvt_f32_f16_e32 v0, v0
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
; GFX11-NEXT:    v_cvt_f32_f16_e32 v2, v2
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[4:5], v3
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[2:3], v2
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    global_store_b64 v6, v[4:5], s[0:1] offset:16
; GFX11-NEXT:    global_store_b128 v6, v[0:3], s[0:1]
; GFX11-NEXT:    s_endpgm
  %val = load <3 x half>, ptr addrspace(1) %in
  %cvt = fpext <3 x half> %val to <3 x double>
  store <3 x double> %cvt, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @global_extload_v4f16_to_v4f64(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
; CI-LABEL: global_extload_v4f16_to_v4f64:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    v_mov_b32_e32 v0, s2
; CI-NEXT:    v_mov_b32_e32 v1, s3
; CI-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
; CI-NEXT:    s_add_u32 s2, s0, 16
; CI-NEXT:    s_addc_u32 s3, s1, 0
; CI-NEXT:    v_mov_b32_e32 v9, s1
; CI-NEXT:    v_mov_b32_e32 v8, s0
; CI-NEXT:    s_waitcnt vmcnt(0)
; CI-NEXT:    v_lshrrev_b32_e32 v2, 16, v1
; CI-NEXT:    v_cvt_f32_f16_e32 v3, v1
; CI-NEXT:    v_cvt_f32_f16_e32 v1, v0
; CI-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
; CI-NEXT:    v_cvt_f32_f16_e32 v2, v2
; CI-NEXT:    v_cvt_f32_f16_e32 v10, v0
; CI-NEXT:    v_cvt_f64_f32_e32 v[4:5], v3
; CI-NEXT:    v_cvt_f64_f32_e32 v[0:1], v1
; CI-NEXT:    v_cvt_f64_f32_e32 v[6:7], v2
; CI-NEXT:    v_cvt_f64_f32_e32 v[2:3], v10
; CI-NEXT:    v_mov_b32_e32 v11, s3
; CI-NEXT:    v_mov_b32_e32 v10, s2
; CI-NEXT:    flat_store_dwordx4 v[10:11], v[4:7]
; CI-NEXT:    flat_store_dwordx4 v[8:9], v[0:3]
; CI-NEXT:    s_endpgm
;
; VI-LABEL: global_extload_v4f16_to_v4f64:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s2
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
; VI-NEXT:    s_add_u32 s2, s0, 16
; VI-NEXT:    s_addc_u32 s3, s1, 0
; VI-NEXT:    v_mov_b32_e32 v9, s1
; VI-NEXT:    v_mov_b32_e32 v8, s0
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_cvt_f32_f16_e32 v3, v1
; VI-NEXT:    v_cvt_f32_f16_sdwa v6, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; VI-NEXT:    v_cvt_f32_f16_e32 v2, v0
; VI-NEXT:    v_cvt_f32_f16_sdwa v10, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; VI-NEXT:    v_cvt_f64_f32_e32 v[4:5], v3
; VI-NEXT:    v_cvt_f64_f32_e32 v[6:7], v6
; VI-NEXT:    v_cvt_f64_f32_e32 v[0:1], v2
; VI-NEXT:    v_cvt_f64_f32_e32 v[2:3], v10
; VI-NEXT:    v_mov_b32_e32 v11, s3
; VI-NEXT:    v_mov_b32_e32 v10, s2
; 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
;
; GFX11-LABEL: global_extload_v4f16_to_v4f64:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v8, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    global_load_b64 v[0:1], v8, s[2:3]
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v1
; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v0
; GFX11-NEXT:    v_cvt_f32_f16_e32 v4, v1
; GFX11-NEXT:    v_cvt_f32_f16_e32 v0, v0
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
; GFX11-NEXT:    v_cvt_f32_f16_e32 v2, v2
; GFX11-NEXT:    v_cvt_f32_f16_e32 v3, v3
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[4:5], v4
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[0:1], v0
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[6:7], v2
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[2:3], v3
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    global_store_b128 v8, v[4:7], s[0:1] offset:16
; GFX11-NEXT:    global_store_b128 v8, v[0:3], s[0:1]
; GFX11-NEXT:    s_endpgm
  %val = load <4 x half>, ptr addrspace(1) %in
  %cvt = fpext <4 x half> %val to <4 x double>
  store <4 x double> %cvt, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @global_extload_v8f16_to_v8f64(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
; CI-LABEL: global_extload_v8f16_to_v8f64:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    v_mov_b32_e32 v0, s2
; CI-NEXT:    v_mov_b32_e32 v1, s3
; CI-NEXT:    flat_load_dwordx4 v[0:3], v[0:1]
; CI-NEXT:    s_add_u32 s2, s0, 48
; CI-NEXT:    s_addc_u32 s3, s1, 0
; CI-NEXT:    v_mov_b32_e32 v7, s3
; CI-NEXT:    v_mov_b32_e32 v6, s2
; CI-NEXT:    s_add_u32 s2, s0, 32
; CI-NEXT:    v_mov_b32_e32 v13, s1
; CI-NEXT:    s_addc_u32 s3, s1, 0
; CI-NEXT:    v_mov_b32_e32 v12, s0
; CI-NEXT:    s_add_u32 s0, s0, 16
; CI-NEXT:    v_mov_b32_e32 v15, s3
; CI-NEXT:    s_addc_u32 s1, s1, 0
; CI-NEXT:    v_mov_b32_e32 v14, s2
; CI-NEXT:    s_waitcnt vmcnt(0)
; CI-NEXT:    v_lshrrev_b32_e32 v4, 16, v3
; CI-NEXT:    v_cvt_f32_f16_e32 v3, v3
; CI-NEXT:    v_lshrrev_b32_e32 v5, 16, v2
; CI-NEXT:    v_cvt_f32_f16_e32 v8, v2
; CI-NEXT:    v_cvt_f32_f16_e32 v2, v4
; CI-NEXT:    v_lshrrev_b32_e32 v9, 16, v1
; CI-NEXT:    v_cvt_f32_f16_e32 v10, v1
; CI-NEXT:    v_lshrrev_b32_e32 v11, 16, v0
; CI-NEXT:    v_cvt_f32_f16_e32 v4, v0
; CI-NEXT:    v_cvt_f32_f16_e32 v16, v5
; CI-NEXT:    v_cvt_f64_f32_e32 v[0:1], v3
; CI-NEXT:    v_cvt_f64_f32_e32 v[2:3], v2
; CI-NEXT:    v_cvt_f32_f16_e32 v17, v9
; CI-NEXT:    v_cvt_f32_f16_e32 v18, v11
; CI-NEXT:    v_cvt_f64_f32_e32 v[8:9], v8
; CI-NEXT:    flat_store_dwordx4 v[6:7], v[0:3]
; CI-NEXT:    v_cvt_f64_f32_e32 v[4:5], v4
; CI-NEXT:    v_cvt_f64_f32_e32 v[0:1], v10
; CI-NEXT:    v_cvt_f64_f32_e32 v[10:11], v16
; CI-NEXT:    v_cvt_f64_f32_e32 v[2:3], v17
; CI-NEXT:    v_cvt_f64_f32_e32 v[6:7], v18
; CI-NEXT:    v_mov_b32_e32 v17, s1
; CI-NEXT:    v_mov_b32_e32 v16, s0
; CI-NEXT:    flat_store_dwordx4 v[14:15], v[8:11]
; CI-NEXT:    flat_store_dwordx4 v[16:17], v[0:3]
; CI-NEXT:    flat_store_dwordx4 v[12:13], v[4:7]
; CI-NEXT:    s_endpgm
;
; VI-LABEL: global_extload_v8f16_to_v8f64:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s2
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    flat_load_dwordx4 v[0:3], v[0:1]
; VI-NEXT:    s_add_u32 s2, s0, 48
; VI-NEXT:    s_addc_u32 s3, s1, 0
; VI-NEXT:    v_mov_b32_e32 v8, s3
; VI-NEXT:    v_mov_b32_e32 v7, s2
; VI-NEXT:    s_add_u32 s2, s0, 32
; VI-NEXT:    v_mov_b32_e32 v13, s1
; VI-NEXT:    s_addc_u32 s3, s1, 0
; VI-NEXT:    v_mov_b32_e32 v12, s0
; VI-NEXT:    s_add_u32 s0, s0, 16
; VI-NEXT:    v_mov_b32_e32 v15, s3
; VI-NEXT:    s_addc_u32 s1, s1, 0
; VI-NEXT:    v_mov_b32_e32 v14, s2
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_cvt_f32_f16_e32 v9, v0
; VI-NEXT:    v_cvt_f32_f16_sdwa v16, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; VI-NEXT:    v_cvt_f32_f16_e32 v0, v3
; VI-NEXT:    v_cvt_f32_f16_sdwa v5, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; VI-NEXT:    v_cvt_f32_f16_e32 v10, v1
; VI-NEXT:    v_cvt_f32_f16_e32 v11, v2
; VI-NEXT:    v_cvt_f64_f32_e32 v[3:4], v0
; VI-NEXT:    v_cvt_f64_f32_e32 v[5:6], v5
; VI-NEXT:    v_cvt_f32_f16_sdwa v2, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; VI-NEXT:    v_cvt_f32_f16_sdwa v17, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; VI-NEXT:    v_cvt_f64_f32_e32 v[0:1], v9
; VI-NEXT:    flat_store_dwordx4 v[7:8], v[3:6]
; VI-NEXT:    v_cvt_f64_f32_e32 v[8:9], v11
; VI-NEXT:    v_cvt_f64_f32_e32 v[4:5], v10
; VI-NEXT:    v_cvt_f64_f32_e32 v[10:11], v2
; VI-NEXT:    v_cvt_f64_f32_e32 v[6:7], v17
; VI-NEXT:    v_cvt_f64_f32_e32 v[2:3], v16
; VI-NEXT:    v_mov_b32_e32 v17, s1
; VI-NEXT:    v_mov_b32_e32 v16, s0
; VI-NEXT:    flat_store_dwordx4 v[14:15], v[8:11]
; VI-NEXT:    flat_store_dwordx4 v[16:17], v[4:7]
; VI-NEXT:    flat_store_dwordx4 v[12:13], v[0:3]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: global_extload_v8f16_to_v8f64:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v16, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    global_load_b128 v[0:3], v16, s[2:3]
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    v_cvt_f32_f16_e32 v4, v0
; GFX11-NEXT:    v_lshrrev_b32_e32 v5, 16, v0
; GFX11-NEXT:    v_cvt_f32_f16_e32 v6, v1
; GFX11-NEXT:    v_lshrrev_b32_e32 v9, 16, v3
; GFX11-NEXT:    v_cvt_f32_f16_e32 v8, v2
; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v2
; GFX11-NEXT:    v_lshrrev_b32_e32 v7, 16, v1
; GFX11-NEXT:    v_cvt_f32_f16_e32 v3, v3
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[0:1], v4
; GFX11-NEXT:    v_cvt_f32_f16_e32 v17, v5
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[4:5], v6
; GFX11-NEXT:    v_cvt_f32_f16_e32 v6, v9
; GFX11-NEXT:    v_cvt_f32_f16_e32 v2, v2
; GFX11-NEXT:    v_cvt_f32_f16_e32 v7, v7
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[12:13], v3
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[8:9], v8
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[14:15], v6
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[10:11], v2
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[6:7], v7
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[2:3], v17
; GFX11-NEXT:    s_clause 0x3
; GFX11-NEXT:    global_store_b128 v16, v[12:15], s[0:1] offset:48
; GFX11-NEXT:    global_store_b128 v16, v[8:11], s[0:1] offset:32
; GFX11-NEXT:    global_store_b128 v16, v[4:7], s[0:1] offset:16
; GFX11-NEXT:    global_store_b128 v16, v[0:3], s[0:1]
; GFX11-NEXT:    s_endpgm
  %val = load <8 x half>, ptr addrspace(1) %in
  %cvt = fpext <8 x half> %val to <8 x double>
  store <8 x double> %cvt, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @global_extload_v16f16_to_v16f64(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
; CI-LABEL: global_extload_v16f16_to_v16f64:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    v_mov_b32_e32 v0, s2
; CI-NEXT:    v_mov_b32_e32 v1, s3
; CI-NEXT:    flat_load_dwordx4 v[0:3], v[0:1]
; CI-NEXT:    s_add_u32 s2, s2, 16
; CI-NEXT:    s_addc_u32 s3, s3, 0
; CI-NEXT:    v_mov_b32_e32 v5, s3
; CI-NEXT:    v_mov_b32_e32 v4, s2
; CI-NEXT:    flat_load_dwordx4 v[4:7], v[4:5]
; CI-NEXT:    s_add_u32 s2, s0, 48
; CI-NEXT:    s_addc_u32 s3, s1, 0
; CI-NEXT:    v_mov_b32_e32 v15, s3
; CI-NEXT:    v_mov_b32_e32 v14, s2
; CI-NEXT:    s_add_u32 s2, s0, 32
; CI-NEXT:    s_addc_u32 s3, s1, 0
; CI-NEXT:    v_mov_b32_e32 v17, s3
; CI-NEXT:    v_mov_b32_e32 v16, s2
; CI-NEXT:    s_add_u32 s2, s0, 16
; CI-NEXT:    s_addc_u32 s3, s1, 0
; CI-NEXT:    v_mov_b32_e32 v19, s3
; CI-NEXT:    v_mov_b32_e32 v18, s2
; CI-NEXT:    s_add_u32 s2, s0, 0x70
; CI-NEXT:    s_addc_u32 s3, s1, 0
; CI-NEXT:    v_mov_b32_e32 v13, s1
; CI-NEXT:    v_mov_b32_e32 v12, s0
; CI-NEXT:    s_waitcnt vmcnt(1)
; CI-NEXT:    v_lshrrev_b32_e32 v8, 16, v3
; CI-NEXT:    v_cvt_f32_f16_e32 v3, v3
; CI-NEXT:    v_cvt_f32_f16_e32 v10, v8
; CI-NEXT:    s_waitcnt vmcnt(0)
; CI-NEXT:    v_lshrrev_b32_e32 v20, 16, v5
; CI-NEXT:    v_cvt_f64_f32_e32 v[8:9], v3
; CI-NEXT:    v_lshrrev_b32_e32 v3, 16, v2
; CI-NEXT:    v_cvt_f64_f32_e32 v[10:11], v10
; CI-NEXT:    v_cvt_f32_f16_e32 v2, v2
; CI-NEXT:    v_cvt_f32_f16_e32 v3, v3
; CI-NEXT:    v_cvt_f32_f16_e32 v21, v5
; CI-NEXT:    flat_store_dwordx4 v[14:15], v[8:11]
; CI-NEXT:    v_mov_b32_e32 v15, s3
; CI-NEXT:    v_cvt_f64_f32_e32 v[8:9], v2
; CI-NEXT:    v_cvt_f64_f32_e32 v[10:11], v3
; CI-NEXT:    v_lshrrev_b32_e32 v2, 16, v1
; CI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; CI-NEXT:    v_cvt_f32_f16_e32 v2, v2
; CI-NEXT:    flat_store_dwordx4 v[16:17], v[8:11]
; CI-NEXT:    v_mov_b32_e32 v14, s2
; CI-NEXT:    v_lshrrev_b32_e32 v8, 16, v0
; CI-NEXT:    v_cvt_f32_f16_e32 v9, v0
; CI-NEXT:    v_cvt_f64_f32_e32 v[0:1], v1
; CI-NEXT:    v_cvt_f64_f32_e32 v[2:3], v2
; CI-NEXT:    v_cvt_f32_f16_e32 v8, v8
; CI-NEXT:    v_lshrrev_b32_e32 v10, 16, v7
; CI-NEXT:    v_cvt_f32_f16_e32 v7, v7
; CI-NEXT:    flat_store_dwordx4 v[18:19], v[0:3]
; CI-NEXT:    v_lshrrev_b32_e32 v11, 16, v6
; CI-NEXT:    v_cvt_f64_f32_e32 v[0:1], v9
; CI-NEXT:    v_cvt_f64_f32_e32 v[2:3], v8
; CI-NEXT:    v_cvt_f32_f16_e32 v8, v10
; CI-NEXT:    s_add_u32 s2, s0, 0x60
; CI-NEXT:    v_cvt_f32_f16_e32 v6, v6
; CI-NEXT:    v_cvt_f32_f16_e32 v10, v11
; CI-NEXT:    s_addc_u32 s3, s1, 0
; CI-NEXT:    v_lshrrev_b32_e32 v5, 16, v4
; CI-NEXT:    flat_store_dwordx4 v[12:13], v[0:3]
; CI-NEXT:    v_mov_b32_e32 v17, s3
; CI-NEXT:    v_cvt_f64_f32_e32 v[0:1], v7
; CI-NEXT:    v_cvt_f64_f32_e32 v[2:3], v8
; CI-NEXT:    v_cvt_f32_f16_e32 v7, v20
; CI-NEXT:    v_cvt_f32_f16_e32 v4, v4
; CI-NEXT:    v_cvt_f32_f16_e32 v12, v5
; CI-NEXT:    v_mov_b32_e32 v16, s2
; CI-NEXT:    s_add_u32 s2, s0, 0x50
; CI-NEXT:    s_addc_u32 s3, s1, 0
; CI-NEXT:    v_cvt_f64_f32_e32 v[8:9], v6
; CI-NEXT:    v_cvt_f64_f32_e32 v[10:11], v10
; CI-NEXT:    s_add_u32 s0, s0, 64
; CI-NEXT:    flat_store_dwordx4 v[14:15], v[0:3]
; CI-NEXT:    s_addc_u32 s1, s1, 0
; CI-NEXT:    v_cvt_f64_f32_e32 v[0:1], v21
; CI-NEXT:    v_cvt_f64_f32_e32 v[2:3], v7
; CI-NEXT:    v_cvt_f64_f32_e32 v[4:5], v4
; CI-NEXT:    v_cvt_f64_f32_e32 v[6:7], v12
; CI-NEXT:    v_mov_b32_e32 v19, s3
; CI-NEXT:    v_mov_b32_e32 v13, s1
; CI-NEXT:    v_mov_b32_e32 v18, s2
; CI-NEXT:    v_mov_b32_e32 v12, s0
; CI-NEXT:    flat_store_dwordx4 v[16:17], v[8:11]
; CI-NEXT:    flat_store_dwordx4 v[18:19], v[0:3]
; CI-NEXT:    flat_store_dwordx4 v[12:13], v[4:7]
; CI-NEXT:    s_endpgm
;
; VI-LABEL: global_extload_v16f16_to_v16f64:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s2
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    flat_load_dwordx4 v[4:7], v[0:1]
; VI-NEXT:    s_add_u32 s2, s2, 16
; VI-NEXT:    s_addc_u32 s3, s3, 0
; VI-NEXT:    v_mov_b32_e32 v0, s2
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    flat_load_dwordx4 v[0:3], v[0:1]
; VI-NEXT:    s_add_u32 s2, s0, 48
; VI-NEXT:    s_addc_u32 s3, s1, 0
; VI-NEXT:    v_mov_b32_e32 v14, s3
; VI-NEXT:    v_mov_b32_e32 v13, s2
; VI-NEXT:    s_add_u32 s2, s0, 32
; VI-NEXT:    s_addc_u32 s3, s1, 0
; VI-NEXT:    v_mov_b32_e32 v16, s3
; VI-NEXT:    v_mov_b32_e32 v15, s2
; VI-NEXT:    s_add_u32 s2, s0, 16
; VI-NEXT:    s_addc_u32 s3, s1, 0
; VI-NEXT:    v_mov_b32_e32 v18, s3
; VI-NEXT:    v_mov_b32_e32 v17, s2
; VI-NEXT:    s_add_u32 s2, s0, 0x50
; VI-NEXT:    v_mov_b32_e32 v12, s1
; VI-NEXT:    s_addc_u32 s3, s1, 0
; VI-NEXT:    v_mov_b32_e32 v11, s0
; VI-NEXT:    s_waitcnt vmcnt(1)
; VI-NEXT:    v_cvt_f32_f16_e32 v8, v7
; VI-NEXT:    v_cvt_f32_f16_sdwa v9, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; VI-NEXT:    v_cvt_f64_f32_e32 v[7:8], v8
; VI-NEXT:    v_cvt_f64_f32_e32 v[9:10], v9
; VI-NEXT:    flat_store_dwordx4 v[13:14], v[7:10]
; VI-NEXT:    s_nop 0
; VI-NEXT:    v_cvt_f32_f16_e32 v7, v6
; VI-NEXT:    v_cvt_f32_f16_sdwa v8, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; VI-NEXT:    s_waitcnt vmcnt(1)
; VI-NEXT:    v_cvt_f32_f16_e32 v10, v2
; VI-NEXT:    v_mov_b32_e32 v14, s3
; VI-NEXT:    v_cvt_f64_f32_e32 v[6:7], v7
; VI-NEXT:    v_cvt_f64_f32_e32 v[8:9], v8
; VI-NEXT:    v_mov_b32_e32 v13, s2
; VI-NEXT:    s_add_u32 s2, s0, 64
; VI-NEXT:    s_addc_u32 s3, s1, 0
; VI-NEXT:    flat_store_dwordx4 v[15:16], v[6:9]
; VI-NEXT:    v_mov_b32_e32 v16, s3
; VI-NEXT:    v_cvt_f32_f16_e32 v6, v5
; VI-NEXT:    v_cvt_f32_f16_sdwa v7, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; VI-NEXT:    v_cvt_f32_f16_e32 v8, v4
; VI-NEXT:    v_cvt_f32_f16_sdwa v9, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; VI-NEXT:    v_cvt_f64_f32_e32 v[4:5], v6
; VI-NEXT:    v_cvt_f64_f32_e32 v[6:7], v7
; VI-NEXT:    v_mov_b32_e32 v15, s2
; VI-NEXT:    s_add_u32 s2, s0, 0x70
; VI-NEXT:    s_addc_u32 s3, s1, 0
; VI-NEXT:    flat_store_dwordx4 v[17:18], v[4:7]
; VI-NEXT:    v_cvt_f32_f16_sdwa v17, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; VI-NEXT:    v_cvt_f64_f32_e32 v[4:5], v8
; VI-NEXT:    v_cvt_f64_f32_e32 v[6:7], v9
; VI-NEXT:    v_cvt_f32_f16_sdwa v9, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; VI-NEXT:    v_cvt_f32_f16_sdwa v8, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; VI-NEXT:    v_cvt_f32_f16_e32 v2, v1
; VI-NEXT:    flat_store_dwordx4 v[11:12], v[4:7]
; VI-NEXT:    v_cvt_f32_f16_sdwa v11, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
; VI-NEXT:    v_cvt_f32_f16_e32 v7, v3
; VI-NEXT:    v_cvt_f64_f32_e32 v[3:4], v9
; VI-NEXT:    v_cvt_f32_f16_e32 v9, v0
; VI-NEXT:    v_cvt_f64_f32_e32 v[1:2], v2
; VI-NEXT:    v_cvt_f64_f32_e32 v[5:6], v10
; VI-NEXT:    v_cvt_f64_f32_e32 v[11:12], v11
; VI-NEXT:    v_cvt_f64_f32_e32 v[9:10], v9
; VI-NEXT:    s_add_u32 s0, s0, 0x60
; VI-NEXT:    flat_store_dwordx4 v[13:14], v[1:4]
; VI-NEXT:    s_addc_u32 s1, s1, 0
; VI-NEXT:    v_cvt_f64_f32_e32 v[0:1], v7
; VI-NEXT:    v_cvt_f64_f32_e32 v[2:3], v17
; VI-NEXT:    v_cvt_f64_f32_e32 v[7:8], v8
; VI-NEXT:    v_mov_b32_e32 v20, s3
; VI-NEXT:    v_mov_b32_e32 v14, s1
; VI-NEXT:    v_mov_b32_e32 v19, s2
; VI-NEXT:    v_mov_b32_e32 v13, s0
; VI-NEXT:    flat_store_dwordx4 v[15:16], v[9:12]
; VI-NEXT:    flat_store_dwordx4 v[19:20], v[0:3]
; VI-NEXT:    flat_store_dwordx4 v[13:14], v[5:8]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: global_extload_v16f16_to_v16f64:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v32, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    global_load_b128 v[0:3], v32, s[2:3]
; GFX11-NEXT:    global_load_b128 v[4:7], v32, s[2:3] offset:16
; GFX11-NEXT:    s_waitcnt vmcnt(1)
; GFX11-NEXT:    v_cvt_f32_f16_e32 v10, v1
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    v_lshrrev_b32_e32 v23, 16, v5
; GFX11-NEXT:    v_lshrrev_b32_e32 v11, 16, v1
; GFX11-NEXT:    v_lshrrev_b32_e32 v19, 16, v4
; GFX11-NEXT:    v_cvt_f32_f16_e32 v15, v7
; GFX11-NEXT:    v_lshrrev_b32_e32 v7, 16, v7
; GFX11-NEXT:    v_cvt_f32_f16_e32 v14, v6
; GFX11-NEXT:    v_lshrrev_b32_e32 v6, 16, v6
; GFX11-NEXT:    v_cvt_f32_f16_e32 v13, v3
; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v3
; GFX11-NEXT:    v_cvt_f32_f16_e32 v12, v2
; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v2
; GFX11-NEXT:    v_cvt_f32_f16_e32 v18, v4
; GFX11-NEXT:    v_cvt_f32_f16_e32 v22, v5
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[4:5], v10
; GFX11-NEXT:    v_cvt_f32_f16_e32 v10, v23
; GFX11-NEXT:    v_cvt_f32_f16_e32 v34, v11
; GFX11-NEXT:    v_cvt_f32_f16_e32 v11, v19
; GFX11-NEXT:    v_lshrrev_b32_e32 v9, 16, v0
; GFX11-NEXT:    v_cvt_f32_f16_e32 v7, v7
; GFX11-NEXT:    v_cvt_f32_f16_e32 v6, v6
; GFX11-NEXT:    v_cvt_f32_f16_e32 v8, v0
; GFX11-NEXT:    v_cvt_f32_f16_e32 v3, v3
; GFX11-NEXT:    v_cvt_f32_f16_e32 v2, v2
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[28:29], v22
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[30:31], v10
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[24:25], v18
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[26:27], v11
; GFX11-NEXT:    v_cvt_f32_f16_e32 v33, v9
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[20:21], v15
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[22:23], v7
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[16:17], v14
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[18:19], v6
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[0:1], v8
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[8:9], v12
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[12:13], v13
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[14:15], v3
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[10:11], v2
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[6:7], v34
; GFX11-NEXT:    v_cvt_f64_f32_e32 v[2:3], v33
; GFX11-NEXT:    s_clause 0x7
; GFX11-NEXT:    global_store_b128 v32, v[28:31], s[0:1] offset:80
; GFX11-NEXT:    global_store_b128 v32, v[24:27], s[0:1] offset:64
; GFX11-NEXT:    global_store_b128 v32, v[20:23], s[0:1] offset:112
; GFX11-NEXT:    global_store_b128 v32, v[16:19], s[0:1] offset:96
; GFX11-NEXT:    global_store_b128 v32, v[12:15], s[0:1] offset:48
; GFX11-NEXT:    global_store_b128 v32, v[8:11], s[0:1] offset:32
; GFX11-NEXT:    global_store_b128 v32, v[4:7], s[0:1] offset:16
; GFX11-NEXT:    global_store_b128 v32, v[0:3], s[0:1]
; GFX11-NEXT:    s_endpgm
  %val = load <16 x half>, ptr addrspace(1) %in
  %cvt = fpext <16 x half> %val to <16 x double>
  store <16 x double> %cvt, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @global_truncstore_f32_to_f16(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
; CIVI-LABEL: global_truncstore_f32_to_f16:
; CIVI:       ; %bb.0:
; CIVI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; CIVI-NEXT:    s_waitcnt lgkmcnt(0)
; CIVI-NEXT:    v_mov_b32_e32 v0, s2
; CIVI-NEXT:    v_mov_b32_e32 v1, s3
; CIVI-NEXT:    flat_load_dword v0, v[0:1]
; CIVI-NEXT:    v_mov_b32_e32 v1, s1
; CIVI-NEXT:    s_waitcnt vmcnt(0)
; CIVI-NEXT:    v_cvt_f16_f32_e32 v2, v0
; CIVI-NEXT:    v_mov_b32_e32 v0, s0
; CIVI-NEXT:    flat_store_short v[0:1], v2
; CIVI-NEXT:    s_endpgm
;
; GFX11-LABEL: global_truncstore_f32_to_f16:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v0, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    global_load_b32 v1, v0, s[2:3]
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    v_cvt_f16_f32_e32 v1, v1
; GFX11-NEXT:    global_store_b16 v0, v1, s[0:1]
; GFX11-NEXT:    s_endpgm
  %val = load float, ptr addrspace(1) %in
  %cvt = fptrunc float %val to half
  store half %cvt, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @global_truncstore_v2f32_to_v2f16(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
; CI-LABEL: global_truncstore_v2f32_to_v2f16:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    v_mov_b32_e32 v0, s2
; CI-NEXT:    v_mov_b32_e32 v1, s3
; CI-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
; CI-NEXT:    s_waitcnt vmcnt(0)
; CI-NEXT:    v_cvt_f16_f32_e32 v2, v1
; CI-NEXT:    v_cvt_f16_f32_e32 v3, v0
; CI-NEXT:    v_mov_b32_e32 v0, s0
; CI-NEXT:    v_mov_b32_e32 v1, s1
; CI-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
; CI-NEXT:    v_or_b32_e32 v2, v3, v2
; CI-NEXT:    flat_store_dword v[0:1], v2
; CI-NEXT:    s_endpgm
;
; VI-LABEL: global_truncstore_v2f32_to_v2f16:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s2
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_cvt_f16_f32_sdwa v2, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD
; VI-NEXT:    v_cvt_f16_f32_e32 v3, v0
; VI-NEXT:    v_mov_b32_e32 v0, s0
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    v_or_b32_e32 v2, v3, v2
; VI-NEXT:    flat_store_dword v[0:1], v2
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: global_truncstore_v2f32_to_v2f16:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v2, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    global_load_b64 v[0:1], v2, s[2:3]
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    v_cvt_f16_f32_e32 v1, v1
; GFX11-NEXT:    v_cvt_f16_f32_e32 v0, v0
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT:    v_pack_b32_f16 v0, v0, v1
; GFX11-NEXT:    global_store_b32 v2, v0, s[0:1]
; GFX11-NEXT:    s_endpgm
  %val = load <2 x float>, ptr addrspace(1) %in
  %cvt = fptrunc <2 x float> %val to <2 x half>
  store <2 x half> %cvt, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @global_truncstore_v3f32_to_v3f16(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
; CI-LABEL: global_truncstore_v3f32_to_v3f16:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    v_mov_b32_e32 v0, s2
; CI-NEXT:    v_mov_b32_e32 v1, s3
; CI-NEXT:    flat_load_dwordx3 v[0:2], v[0:1]
; CI-NEXT:    s_add_u32 s2, s0, 4
; CI-NEXT:    s_addc_u32 s3, s1, 0
; CI-NEXT:    s_waitcnt vmcnt(0)
; CI-NEXT:    v_cvt_f16_f32_e32 v3, v1
; CI-NEXT:    v_cvt_f16_f32_e32 v2, v2
; CI-NEXT:    v_cvt_f16_f32_e32 v4, v0
; CI-NEXT:    v_mov_b32_e32 v0, s2
; CI-NEXT:    v_mov_b32_e32 v1, s3
; CI-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
; CI-NEXT:    flat_store_short v[0:1], v2
; CI-NEXT:    v_mov_b32_e32 v0, s0
; CI-NEXT:    v_or_b32_e32 v2, v4, v3
; CI-NEXT:    v_mov_b32_e32 v1, s1
; CI-NEXT:    flat_store_dword v[0:1], v2
; CI-NEXT:    s_endpgm
;
; VI-LABEL: global_truncstore_v3f32_to_v3f16:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s2
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    flat_load_dwordx3 v[0:2], v[0:1]
; VI-NEXT:    s_add_u32 s2, s0, 4
; VI-NEXT:    s_addc_u32 s3, s1, 0
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_cvt_f16_f32_e32 v2, v2
; VI-NEXT:    v_cvt_f16_f32_sdwa v3, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD
; VI-NEXT:    v_cvt_f16_f32_e32 v4, v0
; VI-NEXT:    v_mov_b32_e32 v0, s2
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    flat_store_short v[0:1], v2
; VI-NEXT:    v_mov_b32_e32 v0, s0
; VI-NEXT:    v_or_b32_e32 v3, v4, v3
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    flat_store_dword v[0:1], v3
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: global_truncstore_v3f32_to_v3f16:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v3, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    global_load_b96 v[0:2], v3, s[2:3]
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    v_cvt_f16_f32_e32 v1, v1
; GFX11-NEXT:    v_cvt_f16_f32_e32 v0, v0
; GFX11-NEXT:    v_cvt_f16_f32_e32 v2, v2
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX11-NEXT:    v_pack_b32_f16 v0, v0, v1
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    global_store_b16 v3, v2, s[0:1] offset:4
; GFX11-NEXT:    global_store_b32 v3, v0, s[0:1]
; GFX11-NEXT:    s_endpgm
  %val = load <3 x float>, ptr addrspace(1) %in
  %cvt = fptrunc <3 x float> %val to <3 x half>
  store <3 x half> %cvt, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @global_truncstore_v4f32_to_v4f16(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
; CI-LABEL: global_truncstore_v4f32_to_v4f16:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    v_mov_b32_e32 v0, s2
; CI-NEXT:    v_mov_b32_e32 v1, s3
; CI-NEXT:    flat_load_dwordx4 v[0:3], v[0:1]
; CI-NEXT:    v_mov_b32_e32 v4, s0
; CI-NEXT:    v_mov_b32_e32 v5, s1
; CI-NEXT:    s_waitcnt vmcnt(0)
; CI-NEXT:    v_cvt_f16_f32_e32 v3, v3
; CI-NEXT:    v_cvt_f16_f32_e32 v1, v1
; CI-NEXT:    v_cvt_f16_f32_e32 v2, v2
; CI-NEXT:    v_cvt_f16_f32_e32 v0, v0
; CI-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
; CI-NEXT:    v_lshlrev_b32_e32 v6, 16, v1
; CI-NEXT:    v_or_b32_e32 v1, v2, v3
; CI-NEXT:    v_or_b32_e32 v0, v0, v6
; CI-NEXT:    flat_store_dwordx2 v[4:5], v[0:1]
; CI-NEXT:    s_endpgm
;
; VI-LABEL: global_truncstore_v4f32_to_v4f16:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s2
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    flat_load_dwordx4 v[0:3], v[0:1]
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_cvt_f16_f32_sdwa v3, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD
; VI-NEXT:    v_cvt_f16_f32_e32 v2, v2
; VI-NEXT:    v_cvt_f16_f32_sdwa v4, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD
; VI-NEXT:    v_cvt_f16_f32_e32 v5, v0
; VI-NEXT:    v_mov_b32_e32 v0, s0
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    v_or_b32_e32 v3, v2, v3
; VI-NEXT:    v_or_b32_e32 v2, v5, v4
; VI-NEXT:    flat_store_dwordx2 v[0:1], v[2:3]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: global_truncstore_v4f32_to_v4f16:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v4, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    global_load_b128 v[0:3], v4, s[2:3]
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    v_cvt_f16_f32_e32 v3, v3
; GFX11-NEXT:    v_cvt_f16_f32_e32 v2, v2
; GFX11-NEXT:    v_cvt_f16_f32_e32 v5, v1
; GFX11-NEXT:    v_cvt_f16_f32_e32 v0, v0
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-NEXT:    v_pack_b32_f16 v1, v2, v3
; GFX11-NEXT:    v_pack_b32_f16 v0, v0, v5
; GFX11-NEXT:    global_store_b64 v4, v[0:1], s[0:1]
; GFX11-NEXT:    s_endpgm
  %val = load <4 x float>, ptr addrspace(1) %in
  %cvt = fptrunc <4 x float> %val to <4 x half>
  store <4 x half> %cvt, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @global_truncstore_v8f32_to_v8f16(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
; CI-LABEL: global_truncstore_v8f32_to_v8f16:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    v_mov_b32_e32 v0, s2
; CI-NEXT:    v_mov_b32_e32 v1, s3
; CI-NEXT:    s_add_u32 s2, s2, 16
; CI-NEXT:    s_addc_u32 s3, s3, 0
; CI-NEXT:    v_mov_b32_e32 v5, s3
; CI-NEXT:    v_mov_b32_e32 v4, s2
; CI-NEXT:    flat_load_dwordx4 v[0:3], v[0:1]
; CI-NEXT:    flat_load_dwordx4 v[4:7], v[4:5]
; CI-NEXT:    v_mov_b32_e32 v8, s0
; CI-NEXT:    v_mov_b32_e32 v9, s1
; CI-NEXT:    s_waitcnt vmcnt(1)
; CI-NEXT:    v_cvt_f16_f32_e32 v3, v3
; CI-NEXT:    v_cvt_f16_f32_e32 v1, v1
; CI-NEXT:    s_waitcnt vmcnt(0)
; CI-NEXT:    v_cvt_f16_f32_e32 v7, v7
; CI-NEXT:    v_cvt_f16_f32_e32 v5, v5
; CI-NEXT:    v_cvt_f16_f32_e32 v2, v2
; CI-NEXT:    v_cvt_f16_f32_e32 v0, v0
; CI-NEXT:    v_cvt_f16_f32_e32 v6, v6
; CI-NEXT:    v_cvt_f16_f32_e32 v4, v4
; CI-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
; CI-NEXT:    v_lshlrev_b32_e32 v10, 16, v1
; CI-NEXT:    v_lshlrev_b32_e32 v7, 16, v7
; CI-NEXT:    v_lshlrev_b32_e32 v5, 16, v5
; CI-NEXT:    v_or_b32_e32 v1, v2, v3
; CI-NEXT:    v_or_b32_e32 v0, v0, v10
; CI-NEXT:    v_or_b32_e32 v3, v6, v7
; CI-NEXT:    v_or_b32_e32 v2, v4, v5
; CI-NEXT:    flat_store_dwordx4 v[8:9], v[0:3]
; CI-NEXT:    s_endpgm
;
; VI-LABEL: global_truncstore_v8f32_to_v8f16:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s2
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    s_add_u32 s2, s2, 16
; VI-NEXT:    s_addc_u32 s3, s3, 0
; VI-NEXT:    v_mov_b32_e32 v5, s3
; VI-NEXT:    v_mov_b32_e32 v4, s2
; 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 v8, s0
; VI-NEXT:    v_mov_b32_e32 v9, s1
; VI-NEXT:    s_waitcnt vmcnt(1)
; VI-NEXT:    v_cvt_f16_f32_sdwa v3, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD
; VI-NEXT:    v_cvt_f16_f32_e32 v2, v2
; VI-NEXT:    v_cvt_f16_f32_sdwa v10, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD
; VI-NEXT:    v_cvt_f16_f32_e32 v0, v0
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_cvt_f16_f32_sdwa v7, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD
; VI-NEXT:    v_cvt_f16_f32_e32 v6, v6
; VI-NEXT:    v_cvt_f16_f32_sdwa v5, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD
; VI-NEXT:    v_cvt_f16_f32_e32 v4, v4
; VI-NEXT:    v_or_b32_e32 v1, v2, v3
; VI-NEXT:    v_or_b32_e32 v0, v0, v10
; VI-NEXT:    v_or_b32_e32 v3, v6, v7
; VI-NEXT:    v_or_b32_e32 v2, v4, v5
; VI-NEXT:    flat_store_dwordx4 v[8:9], v[0:3]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: global_truncstore_v8f32_to_v8f16:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v8, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    global_load_b128 v[0:3], v8, s[2:3] offset:16
; GFX11-NEXT:    global_load_b128 v[4:7], v8, s[2:3]
; GFX11-NEXT:    s_waitcnt vmcnt(1)
; GFX11-NEXT:    v_cvt_f16_f32_e32 v3, v3
; GFX11-NEXT:    v_cvt_f16_f32_e32 v2, v2
; GFX11-NEXT:    v_cvt_f16_f32_e32 v1, v1
; GFX11-NEXT:    v_cvt_f16_f32_e32 v0, v0
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    v_cvt_f16_f32_e32 v7, v7
; GFX11-NEXT:    v_cvt_f16_f32_e32 v6, v6
; GFX11-NEXT:    v_cvt_f16_f32_e32 v5, v5
; GFX11-NEXT:    v_cvt_f16_f32_e32 v4, v4
; GFX11-NEXT:    v_pack_b32_f16 v3, v2, v3
; GFX11-NEXT:    v_pack_b32_f16 v2, v0, v1
; GFX11-NEXT:    v_pack_b32_f16 v1, v6, v7
; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_4)
; GFX11-NEXT:    v_pack_b32_f16 v0, v4, v5
; GFX11-NEXT:    global_store_b128 v8, v[0:3], s[0:1]
; GFX11-NEXT:    s_endpgm
  %val = load <8 x float>, ptr addrspace(1) %in
  %cvt = fptrunc <8 x float> %val to <8 x half>
  store <8 x half> %cvt, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @global_truncstore_v16f32_to_v16f16(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
; CI-LABEL: global_truncstore_v16f32_to_v16f16:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    s_add_u32 s4, s2, 32
; CI-NEXT:    s_addc_u32 s5, s3, 0
; CI-NEXT:    v_mov_b32_e32 v0, s4
; CI-NEXT:    v_mov_b32_e32 v1, s5
; CI-NEXT:    s_add_u32 s4, s2, 48
; CI-NEXT:    s_addc_u32 s5, s3, 0
; CI-NEXT:    v_mov_b32_e32 v9, s3
; CI-NEXT:    v_mov_b32_e32 v4, s4
; CI-NEXT:    v_mov_b32_e32 v8, s2
; CI-NEXT:    s_add_u32 s2, s2, 16
; CI-NEXT:    flat_load_dwordx4 v[0:3], v[0:1]
; CI-NEXT:    v_mov_b32_e32 v5, s5
; CI-NEXT:    flat_load_dwordx4 v[4:7], v[4:5]
; CI-NEXT:    s_addc_u32 s3, s3, 0
; CI-NEXT:    v_mov_b32_e32 v13, s3
; CI-NEXT:    v_mov_b32_e32 v12, s2
; CI-NEXT:    flat_load_dwordx4 v[8:11], v[8:9]
; CI-NEXT:    flat_load_dwordx4 v[12:15], v[12:13]
; CI-NEXT:    s_add_u32 s2, s0, 16
; CI-NEXT:    s_addc_u32 s3, s1, 0
; CI-NEXT:    s_waitcnt vmcnt(3)
; CI-NEXT:    v_cvt_f16_f32_e32 v3, v3
; CI-NEXT:    v_cvt_f16_f32_e32 v2, v2
; CI-NEXT:    v_cvt_f16_f32_e32 v1, v1
; CI-NEXT:    s_waitcnt vmcnt(2)
; CI-NEXT:    v_cvt_f16_f32_e32 v7, v7
; CI-NEXT:    v_cvt_f16_f32_e32 v16, v5
; CI-NEXT:    v_cvt_f16_f32_e32 v0, v0
; CI-NEXT:    v_cvt_f16_f32_e32 v6, v6
; CI-NEXT:    v_cvt_f16_f32_e32 v17, v4
; CI-NEXT:    s_waitcnt vmcnt(1)
; CI-NEXT:    v_cvt_f16_f32_e32 v11, v11
; CI-NEXT:    v_cvt_f16_f32_e32 v9, v9
; CI-NEXT:    s_waitcnt vmcnt(0)
; CI-NEXT:    v_cvt_f16_f32_e32 v15, v15
; CI-NEXT:    v_cvt_f16_f32_e32 v13, v13
; CI-NEXT:    v_cvt_f16_f32_e32 v10, v10
; CI-NEXT:    v_cvt_f16_f32_e32 v8, v8
; CI-NEXT:    v_cvt_f16_f32_e32 v14, v14
; CI-NEXT:    v_cvt_f16_f32_e32 v12, v12
; CI-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
; CI-NEXT:    v_mov_b32_e32 v5, s3
; CI-NEXT:    v_lshlrev_b32_e32 v18, 16, v1
; CI-NEXT:    v_or_b32_e32 v1, v2, v3
; CI-NEXT:    v_lshlrev_b32_e32 v2, 16, v7
; CI-NEXT:    v_lshlrev_b32_e32 v7, 16, v16
; CI-NEXT:    v_mov_b32_e32 v4, s2
; CI-NEXT:    v_or_b32_e32 v0, v0, v18
; CI-NEXT:    v_or_b32_e32 v3, v6, v2
; CI-NEXT:    v_or_b32_e32 v2, v17, v7
; CI-NEXT:    v_lshlrev_b32_e32 v6, 16, v11
; CI-NEXT:    v_lshlrev_b32_e32 v7, 16, v9
; CI-NEXT:    v_lshlrev_b32_e32 v9, 16, v15
; CI-NEXT:    v_lshlrev_b32_e32 v11, 16, v13
; CI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
; CI-NEXT:    v_mov_b32_e32 v5, s1
; CI-NEXT:    v_or_b32_e32 v1, v10, v6
; CI-NEXT:    v_or_b32_e32 v0, v8, v7
; CI-NEXT:    v_or_b32_e32 v3, v14, v9
; CI-NEXT:    v_or_b32_e32 v2, v12, v11
; CI-NEXT:    v_mov_b32_e32 v4, s0
; CI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
; CI-NEXT:    s_endpgm
;
; VI-LABEL: global_truncstore_v16f32_to_v16f16:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_add_u32 s4, s2, 32
; VI-NEXT:    s_addc_u32 s5, s3, 0
; VI-NEXT:    v_mov_b32_e32 v0, s4
; VI-NEXT:    v_mov_b32_e32 v1, s5
; VI-NEXT:    s_add_u32 s4, s2, 48
; VI-NEXT:    s_addc_u32 s5, s3, 0
; VI-NEXT:    v_mov_b32_e32 v9, s3
; VI-NEXT:    v_mov_b32_e32 v4, s4
; VI-NEXT:    v_mov_b32_e32 v8, s2
; VI-NEXT:    s_add_u32 s2, s2, 16
; VI-NEXT:    v_mov_b32_e32 v5, s5
; VI-NEXT:    s_addc_u32 s3, s3, 0
; 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 v13, s3
; VI-NEXT:    v_mov_b32_e32 v12, s2
; VI-NEXT:    flat_load_dwordx4 v[8:11], v[8:9]
; VI-NEXT:    flat_load_dwordx4 v[12:15], v[12:13]
; VI-NEXT:    s_add_u32 s2, s0, 16
; VI-NEXT:    s_addc_u32 s3, s1, 0
; VI-NEXT:    s_waitcnt vmcnt(3)
; VI-NEXT:    v_cvt_f16_f32_sdwa v3, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD
; VI-NEXT:    v_cvt_f16_f32_e32 v2, v2
; VI-NEXT:    v_cvt_f16_f32_sdwa v16, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD
; VI-NEXT:    v_cvt_f16_f32_e32 v0, v0
; VI-NEXT:    s_waitcnt vmcnt(2)
; VI-NEXT:    v_cvt_f16_f32_sdwa v7, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD
; VI-NEXT:    v_cvt_f16_f32_e32 v6, v6
; VI-NEXT:    v_cvt_f16_f32_sdwa v17, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD
; VI-NEXT:    v_cvt_f16_f32_e32 v18, v4
; VI-NEXT:    s_waitcnt vmcnt(1)
; VI-NEXT:    v_cvt_f16_f32_sdwa v11, v11 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD
; VI-NEXT:    v_cvt_f16_f32_e32 v10, v10
; VI-NEXT:    v_cvt_f16_f32_sdwa v9, v9 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD
; VI-NEXT:    v_cvt_f16_f32_e32 v8, v8
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_cvt_f16_f32_sdwa v15, v15 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD
; VI-NEXT:    v_cvt_f16_f32_e32 v14, v14
; VI-NEXT:    v_cvt_f16_f32_sdwa v13, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD
; VI-NEXT:    v_cvt_f16_f32_e32 v12, v12
; VI-NEXT:    v_mov_b32_e32 v5, s3
; VI-NEXT:    v_mov_b32_e32 v4, s2
; VI-NEXT:    v_or_b32_e32 v1, v2, v3
; VI-NEXT:    v_or_b32_e32 v0, v0, v16
; VI-NEXT:    v_or_b32_e32 v3, v6, v7
; VI-NEXT:    v_or_b32_e32 v2, v18, v17
; VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
; VI-NEXT:    v_mov_b32_e32 v5, s1
; VI-NEXT:    v_or_b32_e32 v1, v10, v11
; VI-NEXT:    v_or_b32_e32 v0, v8, v9
; VI-NEXT:    v_or_b32_e32 v3, v14, v15
; VI-NEXT:    v_or_b32_e32 v2, v12, v13
; VI-NEXT:    v_mov_b32_e32 v4, s0
; VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: global_truncstore_v16f32_to_v16f16:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v16, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_clause 0x3
; GFX11-NEXT:    global_load_b128 v[0:3], v16, s[2:3] offset:16
; GFX11-NEXT:    global_load_b128 v[4:7], v16, s[2:3]
; GFX11-NEXT:    global_load_b128 v[8:11], v16, s[2:3] offset:48
; GFX11-NEXT:    global_load_b128 v[12:15], v16, s[2:3] offset:32
; GFX11-NEXT:    s_waitcnt vmcnt(3)
; GFX11-NEXT:    v_cvt_f16_f32_e32 v3, v3
; GFX11-NEXT:    v_cvt_f16_f32_e32 v2, v2
; GFX11-NEXT:    v_cvt_f16_f32_e32 v1, v1
; GFX11-NEXT:    v_cvt_f16_f32_e32 v0, v0
; GFX11-NEXT:    s_waitcnt vmcnt(2)
; GFX11-NEXT:    v_cvt_f16_f32_e32 v7, v7
; GFX11-NEXT:    v_cvt_f16_f32_e32 v6, v6
; GFX11-NEXT:    v_cvt_f16_f32_e32 v17, v5
; GFX11-NEXT:    v_cvt_f16_f32_e32 v18, v4
; GFX11-NEXT:    s_waitcnt vmcnt(1)
; GFX11-NEXT:    v_cvt_f16_f32_e32 v4, v11
; GFX11-NEXT:    v_cvt_f16_f32_e32 v5, v10
; GFX11-NEXT:    v_cvt_f16_f32_e32 v9, v9
; GFX11-NEXT:    v_cvt_f16_f32_e32 v8, v8
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    v_cvt_f16_f32_e32 v10, v15
; GFX11-NEXT:    v_cvt_f16_f32_e32 v11, v14
; GFX11-NEXT:    v_cvt_f16_f32_e32 v13, v13
; GFX11-NEXT:    v_cvt_f16_f32_e32 v12, v12
; GFX11-NEXT:    v_pack_b32_f16 v3, v2, v3
; GFX11-NEXT:    v_pack_b32_f16 v2, v0, v1
; GFX11-NEXT:    v_pack_b32_f16 v1, v6, v7
; GFX11-NEXT:    v_pack_b32_f16 v7, v5, v4
; GFX11-NEXT:    v_pack_b32_f16 v6, v8, v9
; GFX11-NEXT:    v_pack_b32_f16 v5, v11, v10
; GFX11-NEXT:    v_pack_b32_f16 v4, v12, v13
; GFX11-NEXT:    v_pack_b32_f16 v0, v18, v17
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    global_store_b128 v16, v[4:7], s[0:1] offset:16
; GFX11-NEXT:    global_store_b128 v16, v[0:3], s[0:1]
; GFX11-NEXT:    s_endpgm
  %val = load <16 x float>, ptr addrspace(1) %in
  %cvt = fptrunc <16 x float> %val to <16 x half>
  store <16 x half> %cvt, ptr addrspace(1) %out
  ret void
}

; FIXME: Unsafe math should fold conversions away
define amdgpu_kernel void @fadd_f16(ptr addrspace(1) %out, half %a, half %b) #0 {
; CI-LABEL: fadd_f16:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dword s0, s[8:9], 0x2
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    v_cvt_f32_f16_e32 v0, s0
; CI-NEXT:    s_lshr_b32 s0, s0, 16
; CI-NEXT:    v_cvt_f32_f16_e32 v1, s0
; CI-NEXT:    s_load_dwordx2 s[0:1], s[8:9], 0x0
; CI-NEXT:    v_add_f32_e32 v0, v0, v1
; CI-NEXT:    v_cvt_f16_f32_e32 v2, v0
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    v_mov_b32_e32 v0, s0
; CI-NEXT:    v_mov_b32_e32 v1, s1
; CI-NEXT:    flat_store_short v[0:1], v2
; CI-NEXT:    s_endpgm
;
; VI-LABEL: fadd_f16:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dword s2, s[8:9], 0x8
; VI-NEXT:    s_load_dwordx2 s[0:1], s[8:9], 0x0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_lshr_b32 s3, s2, 16
; VI-NEXT:    v_mov_b32_e32 v0, s3
; VI-NEXT:    v_add_f16_e32 v2, s2, v0
; VI-NEXT:    v_mov_b32_e32 v0, s0
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    flat_store_short v[0:1], v2
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: fadd_f16:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b32 s2, s[4:5], 0x8
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v0, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    s_lshr_b32 s3, s2, 16
; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT:    v_add_f16_e64 v1, s2, s3
; GFX11-NEXT:    global_store_b16 v0, v1, s[0:1]
; GFX11-NEXT:    s_endpgm
   %add = fadd half %a, %b
   store half %add, ptr addrspace(1) %out, align 4
   ret void
}

define amdgpu_kernel void @fadd_v2f16(ptr addrspace(1) %out, <2 x half> %a, <2 x half> %b) #0 {
; CI-LABEL: fadd_v2f16:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    s_lshr_b32 s4, s2, 16
; CI-NEXT:    v_cvt_f32_f16_e32 v0, s2
; CI-NEXT:    s_lshr_b32 s2, s3, 16
; CI-NEXT:    v_cvt_f32_f16_e32 v1, s3
; CI-NEXT:    v_cvt_f32_f16_e32 v2, s4
; CI-NEXT:    v_cvt_f32_f16_e32 v3, s2
; CI-NEXT:    v_add_f32_e32 v0, v0, v1
; CI-NEXT:    v_cvt_f16_f32_e32 v0, v0
; CI-NEXT:    v_add_f32_e32 v1, v2, v3
; CI-NEXT:    v_cvt_f16_f32_e32 v1, v1
; CI-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
; CI-NEXT:    v_or_b32_e32 v2, v0, v1
; CI-NEXT:    v_mov_b32_e32 v0, s0
; CI-NEXT:    v_mov_b32_e32 v1, s1
; CI-NEXT:    flat_store_dword v[0:1], v2
; CI-NEXT:    s_endpgm
;
; VI-LABEL: fadd_v2f16:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_lshr_b32 s4, s3, 16
; VI-NEXT:    s_lshr_b32 s5, s2, 16
; VI-NEXT:    v_mov_b32_e32 v0, s3
; VI-NEXT:    v_mov_b32_e32 v1, s4
; VI-NEXT:    v_mov_b32_e32 v2, s5
; VI-NEXT:    v_add_f16_e32 v0, s2, v0
; VI-NEXT:    v_add_f16_sdwa v1, v2, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT:    v_or_b32_e32 v2, v0, v1
; VI-NEXT:    v_mov_b32_e32 v0, s0
; VI-NEXT:    v_mov_b32_e32 v1, s1
; VI-NEXT:    flat_store_dword v[0:1], v2
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: fadd_v2f16:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v0, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_pk_add_f16 v1, s2, s3
; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
; GFX11-NEXT:    s_endpgm
  %add = fadd <2 x half> %a, %b
  store <2 x half> %add, ptr addrspace(1) %out, align 8
  ret void
}

define amdgpu_kernel void @fadd_v4f16(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
; CI-LABEL: fadd_v4f16:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    v_mov_b32_e32 v0, s2
; CI-NEXT:    v_mov_b32_e32 v1, s3
; CI-NEXT:    flat_load_dwordx4 v[0:3], v[0:1]
; CI-NEXT:    v_mov_b32_e32 v4, s0
; CI-NEXT:    v_mov_b32_e32 v5, s1
; CI-NEXT:    s_waitcnt vmcnt(0)
; CI-NEXT:    v_cvt_f32_f16_e32 v6, v0
; CI-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
; CI-NEXT:    v_cvt_f32_f16_e32 v7, v1
; CI-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
; CI-NEXT:    v_cvt_f32_f16_e32 v8, v2
; CI-NEXT:    v_lshrrev_b32_e32 v2, 16, v2
; CI-NEXT:    v_cvt_f32_f16_e32 v9, v3
; CI-NEXT:    v_lshrrev_b32_e32 v3, 16, v3
; CI-NEXT:    v_cvt_f32_f16_e32 v0, v0
; CI-NEXT:    v_cvt_f32_f16_e32 v1, v1
; CI-NEXT:    v_cvt_f32_f16_e32 v3, v3
; CI-NEXT:    v_cvt_f32_f16_e32 v2, v2
; CI-NEXT:    v_add_f32_e32 v7, v7, v9
; CI-NEXT:    v_add_f32_e32 v6, v6, v8
; CI-NEXT:    v_add_f32_e32 v1, v1, v3
; CI-NEXT:    v_add_f32_e32 v0, v0, v2
; CI-NEXT:    v_cvt_f16_f32_e32 v1, v1
; CI-NEXT:    v_cvt_f16_f32_e32 v0, v0
; CI-NEXT:    v_cvt_f16_f32_e32 v2, v7
; CI-NEXT:    v_cvt_f16_f32_e32 v3, v6
; CI-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
; CI-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
; CI-NEXT:    v_or_b32_e32 v1, v2, v1
; CI-NEXT:    v_or_b32_e32 v0, v3, v0
; CI-NEXT:    flat_store_dwordx2 v[4:5], v[0:1]
; CI-NEXT:    s_endpgm
;
; VI-LABEL: fadd_v4f16:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s2
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    flat_load_dwordx4 v[0:3], v[0:1]
; VI-NEXT:    v_mov_b32_e32 v4, s0
; VI-NEXT:    v_mov_b32_e32 v5, s1
; VI-NEXT:    s_waitcnt vmcnt(0)
; VI-NEXT:    v_add_f16_sdwa v6, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
; VI-NEXT:    v_add_f16_e32 v1, v1, v3
; VI-NEXT:    v_add_f16_sdwa v3, v0, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
; VI-NEXT:    v_add_f16_e32 v0, v0, v2
; VI-NEXT:    v_or_b32_e32 v1, v1, v6
; VI-NEXT:    v_or_b32_e32 v0, v0, v3
; VI-NEXT:    flat_store_dwordx2 v[4:5], v[0:1]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: fadd_v4f16:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v4, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    global_load_b128 v[0:3], v4, s[2:3]
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    v_pk_add_f16 v1, v1, v3
; GFX11-NEXT:    v_pk_add_f16 v0, v0, v2
; GFX11-NEXT:    global_store_b64 v4, v[0:1], s[0:1]
; GFX11-NEXT:    s_endpgm
  %b_ptr = getelementptr <4 x half>, ptr addrspace(1) %in, i32 1
  %a = load <4 x half>, ptr addrspace(1) %in, align 16
  %b = load <4 x half>, ptr addrspace(1) %b_ptr, align 16
  %result = fadd <4 x half> %a, %b
  store <4 x half> %result, ptr addrspace(1) %out, align 16
  ret void
}

define amdgpu_kernel void @fadd_v8f16(ptr addrspace(1) %out, <8 x half> %a, <8 x half> %b) #0 {
; CI-LABEL: fadd_v8f16:
; CI:       ; %bb.0:
; CI-NEXT:    s_load_dwordx8 s[0:7], s[8:9], 0x4
; CI-NEXT:    s_load_dwordx2 s[8:9], s[8:9], 0x0
; CI-NEXT:    s_waitcnt lgkmcnt(0)
; CI-NEXT:    s_lshr_b32 s10, s0, 16
; CI-NEXT:    v_cvt_f32_f16_e32 v4, s0
; CI-NEXT:    s_lshr_b32 s0, s4, 16
; CI-NEXT:    v_cvt_f32_f16_e32 v8, s0
; CI-NEXT:    s_lshr_b32 s0, s5, 16
; CI-NEXT:    s_lshr_b32 s11, s1, 16
; CI-NEXT:    v_cvt_f32_f16_e32 v9, s0
; CI-NEXT:    s_lshr_b32 s0, s6, 16
; CI-NEXT:    s_lshr_b32 s12, s2, 16
; CI-NEXT:    v_cvt_f32_f16_e32 v0, s10
; CI-NEXT:    v_cvt_f32_f16_e32 v1, s11
; CI-NEXT:    s_lshr_b32 s10, s3, 16
; CI-NEXT:    v_cvt_f32_f16_e32 v10, s0
; CI-NEXT:    s_lshr_b32 s0, s7, 16
; CI-NEXT:    v_cvt_f32_f16_e32 v2, s12
; CI-NEXT:    v_cvt_f32_f16_e32 v3, s10
; CI-NEXT:    v_cvt_f32_f16_e32 v5, s1
; CI-NEXT:    v_cvt_f32_f16_e32 v11, s0
; CI-NEXT:    v_cvt_f32_f16_e32 v12, s4
; CI-NEXT:    v_cvt_f32_f16_e32 v13, s5
; CI-NEXT:    v_cvt_f32_f16_e32 v6, s2
; CI-NEXT:    v_cvt_f32_f16_e32 v7, s3
; CI-NEXT:    v_cvt_f32_f16_e32 v14, s7
; CI-NEXT:    v_cvt_f32_f16_e32 v15, s6
; CI-NEXT:    v_add_f32_e32 v1, v1, v9
; CI-NEXT:    v_add_f32_e32 v0, v0, v8
; CI-NEXT:    v_add_f32_e32 v3, v3, v11
; CI-NEXT:    v_add_f32_e32 v2, v2, v10
; CI-NEXT:    v_cvt_f16_f32_e32 v1, v1
; CI-NEXT:    v_add_f32_e32 v5, v5, v13
; CI-NEXT:    v_cvt_f16_f32_e32 v0, v0
; CI-NEXT:    v_add_f32_e32 v4, v4, v12
; CI-NEXT:    v_cvt_f16_f32_e32 v3, v3
; CI-NEXT:    v_add_f32_e32 v7, v7, v14
; CI-NEXT:    v_cvt_f16_f32_e32 v2, v2
; CI-NEXT:    v_add_f32_e32 v6, v6, v15
; CI-NEXT:    v_cvt_f16_f32_e32 v5, v5
; CI-NEXT:    v_cvt_f16_f32_e32 v4, v4
; CI-NEXT:    v_cvt_f16_f32_e32 v7, v7
; CI-NEXT:    v_cvt_f16_f32_e32 v6, v6
; CI-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
; CI-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
; CI-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
; CI-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
; CI-NEXT:    v_or_b32_e32 v1, v5, v1
; CI-NEXT:    v_or_b32_e32 v0, v4, v0
; CI-NEXT:    v_mov_b32_e32 v4, s8
; CI-NEXT:    v_or_b32_e32 v3, v7, v3
; CI-NEXT:    v_or_b32_e32 v2, v6, v2
; CI-NEXT:    v_mov_b32_e32 v5, s9
; CI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
; CI-NEXT:    s_endpgm
;
; VI-LABEL: fadd_v8f16:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dwordx8 s[0:7], s[8:9], 0x10
; VI-NEXT:    s_load_dwordx2 s[8:9], s[8:9], 0x0
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_lshr_b32 s10, s7, 16
; VI-NEXT:    s_lshr_b32 s11, s3, 16
; VI-NEXT:    v_mov_b32_e32 v0, s7
; VI-NEXT:    v_mov_b32_e32 v1, s10
; VI-NEXT:    v_mov_b32_e32 v2, s11
; VI-NEXT:    v_add_f16_sdwa v1, v2, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT:    v_add_f16_e32 v0, s3, v0
; VI-NEXT:    s_lshr_b32 s3, s6, 16
; VI-NEXT:    s_lshr_b32 s7, s2, 16
; VI-NEXT:    v_or_b32_e32 v3, v0, v1
; VI-NEXT:    v_mov_b32_e32 v0, s3
; VI-NEXT:    v_mov_b32_e32 v1, s7
; VI-NEXT:    v_add_f16_sdwa v0, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT:    v_mov_b32_e32 v1, s6
; VI-NEXT:    v_add_f16_e32 v1, s2, v1
; VI-NEXT:    s_lshr_b32 s2, s5, 16
; VI-NEXT:    s_lshr_b32 s3, s1, 16
; VI-NEXT:    v_or_b32_e32 v2, v1, v0
; VI-NEXT:    v_mov_b32_e32 v0, s2
; VI-NEXT:    v_mov_b32_e32 v1, s3
; VI-NEXT:    v_add_f16_sdwa v0, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT:    v_mov_b32_e32 v1, s5
; VI-NEXT:    v_add_f16_e32 v1, s1, v1
; VI-NEXT:    s_lshr_b32 s1, s4, 16
; VI-NEXT:    s_lshr_b32 s2, s0, 16
; VI-NEXT:    v_or_b32_e32 v1, v1, v0
; VI-NEXT:    v_mov_b32_e32 v0, s1
; VI-NEXT:    v_mov_b32_e32 v4, s2
; VI-NEXT:    v_add_f16_sdwa v0, v4, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT:    v_mov_b32_e32 v4, s4
; VI-NEXT:    v_add_f16_e32 v4, s0, v4
; VI-NEXT:    v_or_b32_e32 v0, v4, v0
; VI-NEXT:    v_mov_b32_e32 v4, s8
; VI-NEXT:    v_mov_b32_e32 v5, s9
; VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
; VI-NEXT:    s_endpgm
;
; GFX11-LABEL: fadd_v8f16:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_clause 0x1
; GFX11-NEXT:    s_load_b256 s[8:15], s[4:5], 0x10
; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v4, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    v_pk_add_f16 v3, s11, s15
; GFX11-NEXT:    v_pk_add_f16 v2, s10, s14
; GFX11-NEXT:    v_pk_add_f16 v1, s9, s13
; GFX11-NEXT:    v_pk_add_f16 v0, s8, s12
; GFX11-NEXT:    global_store_b128 v4, v[0:3], s[0:1]
; GFX11-NEXT:    s_endpgm
  %add = fadd <8 x half> %a, %b
  store <8 x half> %add, ptr addrspace(1) %out, align 32
  ret void
}

define amdgpu_kernel void @test_bitcast_from_half(ptr addrspace(1) %in, ptr addrspace(1) %out) #0 {
; CIVI-LABEL: test_bitcast_from_half:
; CIVI:       ; %bb.0:
; CIVI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; CIVI-NEXT:    s_waitcnt lgkmcnt(0)
; CIVI-NEXT:    v_mov_b32_e32 v0, s0
; CIVI-NEXT:    v_mov_b32_e32 v1, s1
; CIVI-NEXT:    flat_load_ushort v2, v[0:1]
; CIVI-NEXT:    v_mov_b32_e32 v0, s2
; CIVI-NEXT:    v_mov_b32_e32 v1, s3
; CIVI-NEXT:    s_waitcnt vmcnt(0)
; CIVI-NEXT:    flat_store_short v[0:1], v2
; CIVI-NEXT:    s_endpgm
;
; GFX11-LABEL: test_bitcast_from_half:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v0, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    global_load_u16 v1, v0, s[0:1]
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    global_store_b16 v0, v1, s[2:3]
; GFX11-NEXT:    s_endpgm
  %val = load half, ptr addrspace(1) %in
  %val_int = bitcast half %val to i16
  store i16 %val_int, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @test_bitcast_to_half(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
; CIVI-LABEL: test_bitcast_to_half:
; CIVI:       ; %bb.0:
; CIVI-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
; CIVI-NEXT:    s_waitcnt lgkmcnt(0)
; CIVI-NEXT:    v_mov_b32_e32 v0, s2
; CIVI-NEXT:    v_mov_b32_e32 v1, s3
; CIVI-NEXT:    flat_load_ushort v2, v[0:1]
; CIVI-NEXT:    v_mov_b32_e32 v0, s0
; CIVI-NEXT:    v_mov_b32_e32 v1, s1
; CIVI-NEXT:    s_waitcnt vmcnt(0)
; CIVI-NEXT:    flat_store_short v[0:1], v2
; CIVI-NEXT:    s_endpgm
;
; GFX11-LABEL: test_bitcast_to_half:
; GFX11:       ; %bb.0:
; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
; GFX11-NEXT:    v_mov_b32_e32 v0, 0
; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-NEXT:    global_load_u16 v1, v0, s[2:3]
; GFX11-NEXT:    s_waitcnt vmcnt(0)
; GFX11-NEXT:    global_store_b16 v0, v1, s[0:1]
; GFX11-NEXT:    s_endpgm
  %val = load i16, ptr addrspace(1) %in
  %val_fp = bitcast i16 %val to half
  store half %val_fp, ptr addrspace(1) %out
  ret void
}

attributes #0 = { nounwind }