; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=amdgcn -mcpu=gfx700 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX7 %s
; RUN: llc -mtriple=amdgcn -mcpu=gfx803 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX8 %s
; RUN: llc -mtriple=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9-NODL %s
; RUN: llc -mtriple=amdgcn -mcpu=gfx906 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9-DL %s

define amdgpu_kernel void @udot4_acc32(<4 x i8> addrspace(1)* %src1,
; GFX7-LABEL: udot4_acc32:
; GFX7:       ; %bb.0: ; %entry
; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; GFX7-NEXT:    s_movk_i32 s8, 0xff
; GFX7-NEXT:    s_mov_b32 s3, 0xf000
; GFX7-NEXT:    s_mov_b32 s2, -1
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    s_load_dword s4, s[4:5], 0x0
; GFX7-NEXT:    s_load_dword s5, s[6:7], 0x0
; GFX7-NEXT:    s_load_dword s6, s[0:1], 0x0
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    s_and_b32 s7, s4, s8
; GFX7-NEXT:    s_and_b32 s8, s5, s8
; GFX7-NEXT:    s_bfe_u32 s10, s5, 0x80008
; GFX7-NEXT:    v_mov_b32_e32 v0, s8
; GFX7-NEXT:    v_mov_b32_e32 v1, s6
; GFX7-NEXT:    s_bfe_u32 s12, s5, 0x80010
; GFX7-NEXT:    v_mad_u32_u24 v0, s7, v0, v1
; GFX7-NEXT:    s_bfe_u32 s9, s4, 0x80008
; GFX7-NEXT:    v_mov_b32_e32 v1, s10
; GFX7-NEXT:    s_bfe_u32 s11, s4, 0x80010
; GFX7-NEXT:    v_mad_u32_u24 v0, s9, v1, v0
; GFX7-NEXT:    v_mov_b32_e32 v1, s12
; GFX7-NEXT:    s_lshr_b32 s5, s5, 24
; GFX7-NEXT:    v_mad_u32_u24 v0, s11, v1, v0
; GFX7-NEXT:    s_lshr_b32 s4, s4, 24
; GFX7-NEXT:    v_mov_b32_e32 v1, s5
; GFX7-NEXT:    v_mad_u32_u24 v0, s4, v1, v0
; GFX7-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GFX7-NEXT:    s_endpgm
;
; GFX8-LABEL: udot4_acc32:
; GFX8:       ; %bb.0: ; %entry
; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX8-NEXT:    s_movk_i32 s2, 0xff
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    s_load_dword s3, s[4:5], 0x0
; GFX8-NEXT:    s_load_dword s4, s[6:7], 0x0
; GFX8-NEXT:    s_load_dword s5, s[0:1], 0x0
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    s_and_b32 s6, s3, s2
; GFX8-NEXT:    s_and_b32 s2, s4, s2
; GFX8-NEXT:    s_bfe_u32 s8, s4, 0x80008
; GFX8-NEXT:    v_mov_b32_e32 v0, s2
; GFX8-NEXT:    v_mov_b32_e32 v1, s5
; GFX8-NEXT:    s_bfe_u32 s10, s4, 0x80010
; GFX8-NEXT:    v_mad_u32_u24 v0, s6, v0, v1
; GFX8-NEXT:    s_bfe_u32 s7, s3, 0x80008
; GFX8-NEXT:    v_mov_b32_e32 v1, s8
; GFX8-NEXT:    s_bfe_u32 s9, s3, 0x80010
; GFX8-NEXT:    v_mad_u32_u24 v0, s7, v1, v0
; GFX8-NEXT:    v_mov_b32_e32 v1, s10
; GFX8-NEXT:    s_lshr_b32 s4, s4, 24
; GFX8-NEXT:    v_mad_u32_u24 v0, s9, v1, v0
; GFX8-NEXT:    s_lshr_b32 s3, s3, 24
; GFX8-NEXT:    v_mov_b32_e32 v1, s4
; GFX8-NEXT:    v_mad_u32_u24 v2, s3, v1, v0
; GFX8-NEXT:    v_mov_b32_e32 v0, s0
; GFX8-NEXT:    v_mov_b32_e32 v1, s1
; GFX8-NEXT:    flat_store_dword v[0:1], v2
; GFX8-NEXT:    s_endpgm
;
; GFX9-NODL-LABEL: udot4_acc32:
; GFX9-NODL:       ; %bb.0: ; %entry
; GFX9-NODL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX9-NODL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX9-NODL-NEXT:    s_movk_i32 s2, 0xff
; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NODL-NEXT:    s_load_dword s3, s[4:5], 0x0
; GFX9-NODL-NEXT:    s_load_dword s4, s[6:7], 0x0
; GFX9-NODL-NEXT:    s_load_dword s5, s[0:1], 0x0
; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NODL-NEXT:    s_and_b32 s6, s3, s2
; GFX9-NODL-NEXT:    s_and_b32 s2, s4, s2
; GFX9-NODL-NEXT:    s_bfe_u32 s8, s4, 0x80008
; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s2
; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s5
; GFX9-NODL-NEXT:    s_bfe_u32 s10, s4, 0x80010
; GFX9-NODL-NEXT:    v_mad_u32_u24 v0, s6, v0, v1
; GFX9-NODL-NEXT:    s_bfe_u32 s7, s3, 0x80008
; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s8
; GFX9-NODL-NEXT:    s_bfe_u32 s9, s3, 0x80010
; GFX9-NODL-NEXT:    v_mad_u32_u24 v0, s7, v1, v0
; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s10
; GFX9-NODL-NEXT:    s_lshr_b32 s4, s4, 24
; GFX9-NODL-NEXT:    v_mad_u32_u24 v0, s9, v1, v0
; GFX9-NODL-NEXT:    s_lshr_b32 s3, s3, 24
; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s4
; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s3, v1, v0
; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s0
; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s1
; GFX9-NODL-NEXT:    global_store_dword v[0:1], v2, off
; GFX9-NODL-NEXT:    s_endpgm
;
; GFX9-DL-LABEL: udot4_acc32:
; GFX9-DL:       ; %bb.0: ; %entry
; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
; GFX9-DL-NEXT:    s_load_dword s3, s[6:7], 0x0
; GFX9-DL-NEXT:    s_load_dword s4, s[0:1], 0x0
; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-DL-NEXT:    v_mov_b32_e32 v2, s3
; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s4
; GFX9-DL-NEXT:    v_dot4_u32_u8 v2, s2, v2, v3
; GFX9-DL-NEXT:    global_store_dword v[0:1], v2, off
; GFX9-DL-NEXT:    s_endpgm
                                       <4 x i8> addrspace(1)* %src2,
                                       i32 addrspace(1)* nocapture %dst) {
entry:
  %vec1 = load <4 x i8>, <4 x i8> addrspace(1)* %src1
  %vec2 = load <4 x i8>, <4 x i8> addrspace(1)* %src2

  %v1e0 = extractelement <4 x i8> %vec1, i64 0
  %cv1e0 = zext i8 %v1e0 to i32
  %v2e0 = extractelement <4 x i8> %vec2, i64 0
  %cv2e0 = zext i8 %v2e0 to i32
  %mul1 = mul nuw nsw i32 %cv1e0, %cv2e0

  %v1e1 = extractelement <4 x i8> %vec1, i64 1
  %cv1e1 = zext i8 %v1e1 to i32
  %v2e1 = extractelement <4 x i8> %vec2, i64 1
  %cv2e1 = zext i8 %v2e1 to i32
  %mul2 = mul nuw nsw i32 %cv1e1, %cv2e1

  %v1e2 = extractelement <4 x i8> %vec1, i64 2
  %cv1e2 = zext i8 %v1e2 to i32
  %v2e2 = extractelement <4 x i8> %vec2, i64 2
  %cv2e2 = zext i8 %v2e2 to i32
  %mul3 = mul nuw nsw i32 %cv1e2, %cv2e2

  %v1e3 = extractelement <4 x i8> %vec1, i64 3
  %cv1e3 = zext i8 %v1e3 to i32
  %v2e3 = extractelement <4 x i8> %vec2, i64 3
  %cv2e3 = zext i8 %v2e3 to i32
  %mul4 = mul nuw nsw i32 %cv1e3, %cv2e3

  %acc = load i32, i32 addrspace(1)* %dst, align 4
  %mad1 = add i32 %mul1, %acc
  %mad2 = add i32 %mad1, %mul2
  %mad3 = add i32 %mad2, %mul3
  %mad4 = add i32 %mad3, %mul4

  store i32 %mad4, i32 addrspace(1)* %dst, align 4
  ret void
}

define amdgpu_kernel void @udot4_acc16(<4 x i8> addrspace(1)* %src1,
; GFX7-LABEL: udot4_acc16:
; GFX7:       ; %bb.0: ; %entry
; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; GFX7-NEXT:    s_mov_b32 s3, 0xf000
; GFX7-NEXT:    s_mov_b32 s2, -1
; GFX7-NEXT:    s_movk_i32 s8, 0xff
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    s_load_dword s4, s[4:5], 0x0
; GFX7-NEXT:    buffer_load_ushort v0, off, s[0:3], 0
; GFX7-NEXT:    s_load_dword s5, s[6:7], 0x0
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    s_and_b32 s7, s4, s8
; GFX7-NEXT:    s_bfe_u32 s9, s4, 0x80008
; GFX7-NEXT:    s_and_b32 s6, s5, s8
; GFX7-NEXT:    s_bfe_u32 s8, s5, 0x80008
; GFX7-NEXT:    v_mov_b32_e32 v1, s6
; GFX7-NEXT:    s_bfe_u32 s10, s5, 0x80010
; GFX7-NEXT:    v_mov_b32_e32 v2, s8
; GFX7-NEXT:    s_bfe_u32 s11, s4, 0x80010
; GFX7-NEXT:    s_lshr_b32 s5, s5, 24
; GFX7-NEXT:    v_mov_b32_e32 v3, s10
; GFX7-NEXT:    s_lshr_b32 s4, s4, 24
; GFX7-NEXT:    s_waitcnt vmcnt(0)
; GFX7-NEXT:    v_mad_u32_u24 v0, s7, v1, v0
; GFX7-NEXT:    v_mad_u32_u24 v0, s9, v2, v0
; GFX7-NEXT:    v_mad_u32_u24 v0, s11, v3, v0
; GFX7-NEXT:    v_mov_b32_e32 v1, s5
; GFX7-NEXT:    v_mad_u32_u24 v0, s4, v1, v0
; GFX7-NEXT:    buffer_store_short v0, off, s[0:3], 0
; GFX7-NEXT:    s_endpgm
;
; GFX8-LABEL: udot4_acc16:
; GFX8:       ; %bb.0: ; %entry
; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_mov_b32_e32 v0, s0
; GFX8-NEXT:    v_mov_b32_e32 v1, s1
; GFX8-NEXT:    flat_load_ushort v2, v[0:1]
; GFX8-NEXT:    s_load_dword s1, s[4:5], 0x0
; GFX8-NEXT:    s_load_dword s2, s[6:7], 0x0
; GFX8-NEXT:    s_movk_i32 s0, 0xff
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    s_and_b32 s3, s1, s0
; GFX8-NEXT:    s_and_b32 s0, s2, s0
; GFX8-NEXT:    s_bfe_u32 s5, s2, 0x80008
; GFX8-NEXT:    v_mov_b32_e32 v3, s0
; GFX8-NEXT:    s_bfe_u32 s7, s2, 0x80010
; GFX8-NEXT:    s_bfe_u32 s4, s1, 0x80008
; GFX8-NEXT:    v_mov_b32_e32 v4, s5
; GFX8-NEXT:    s_bfe_u32 s6, s1, 0x80010
; GFX8-NEXT:    s_lshr_b32 s2, s2, 24
; GFX8-NEXT:    v_mov_b32_e32 v5, s7
; GFX8-NEXT:    s_lshr_b32 s1, s1, 24
; GFX8-NEXT:    s_waitcnt vmcnt(0)
; GFX8-NEXT:    v_mad_u32_u24 v2, s3, v3, v2
; GFX8-NEXT:    v_mad_u32_u24 v2, s4, v4, v2
; GFX8-NEXT:    v_mad_u32_u24 v2, s6, v5, v2
; GFX8-NEXT:    v_mov_b32_e32 v3, s2
; GFX8-NEXT:    v_mad_u32_u24 v2, s1, v3, v2
; GFX8-NEXT:    flat_store_short v[0:1], v2
; GFX8-NEXT:    s_endpgm
;
; GFX9-NODL-LABEL: udot4_acc16:
; GFX9-NODL:       ; %bb.0: ; %entry
; GFX9-NODL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX9-NODL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s0
; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s1
; GFX9-NODL-NEXT:    global_load_ushort v2, v[0:1], off
; GFX9-NODL-NEXT:    s_load_dword s1, s[4:5], 0x0
; GFX9-NODL-NEXT:    s_load_dword s2, s[6:7], 0x0
; GFX9-NODL-NEXT:    s_movk_i32 s0, 0xff
; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NODL-NEXT:    s_and_b32 s3, s1, s0
; GFX9-NODL-NEXT:    s_and_b32 s0, s2, s0
; GFX9-NODL-NEXT:    s_bfe_u32 s5, s2, 0x80008
; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s0
; GFX9-NODL-NEXT:    s_bfe_u32 s7, s2, 0x80010
; GFX9-NODL-NEXT:    s_bfe_u32 s4, s1, 0x80008
; GFX9-NODL-NEXT:    v_mov_b32_e32 v4, s5
; GFX9-NODL-NEXT:    s_bfe_u32 s6, s1, 0x80010
; GFX9-NODL-NEXT:    s_lshr_b32 s2, s2, 24
; GFX9-NODL-NEXT:    v_mov_b32_e32 v5, s7
; GFX9-NODL-NEXT:    s_lshr_b32 s1, s1, 24
; GFX9-NODL-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s3, v3, v2
; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s4, v4, v2
; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s6, v5, v2
; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s2
; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s1, v3, v2
; GFX9-NODL-NEXT:    global_store_short v[0:1], v2, off
; GFX9-NODL-NEXT:    s_endpgm
;
; GFX9-DL-LABEL: udot4_acc16:
; GFX9-DL:       ; %bb.0: ; %entry
; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
; GFX9-DL-NEXT:    s_load_dword s3, s[6:7], 0x0
; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
; GFX9-DL-NEXT:    global_load_ushort v2, v[0:1], off
; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s3
; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
; GFX9-DL-NEXT:    v_dot4_u32_u8 v2, s2, v3, v2
; GFX9-DL-NEXT:    global_store_short v[0:1], v2, off
; GFX9-DL-NEXT:    s_endpgm
                                       <4 x i8> addrspace(1)* %src2,
                                       i16 addrspace(1)* nocapture %dst) {
entry:
  %vec1 = load <4 x i8>, <4 x i8> addrspace(1)* %src1
  %vec2 = load <4 x i8>, <4 x i8> addrspace(1)* %src2

  %v1e0 = extractelement <4 x i8> %vec1, i64 0
  %cv1e0 = zext i8 %v1e0 to i16
  %v2e0 = extractelement <4 x i8> %vec2, i64 0
  %cv2e0 = zext i8 %v2e0 to i16
  %mul1 = mul nuw nsw i16 %cv1e0, %cv2e0

  %v1e1 = extractelement <4 x i8> %vec1, i64 1
  %cv1e1 = zext i8 %v1e1 to i16
  %v2e1 = extractelement <4 x i8> %vec2, i64 1
  %cv2e1 = zext i8 %v2e1 to i16
  %mul2 = mul nuw nsw i16 %cv1e1, %cv2e1

  %v1e2 = extractelement <4 x i8> %vec1, i64 2
  %cv1e2 = zext i8 %v1e2 to i16
  %v2e2 = extractelement <4 x i8> %vec2, i64 2
  %cv2e2 = zext i8 %v2e2 to i16
  %mul3 = mul nuw nsw i16 %cv1e2, %cv2e2

  %v1e3 = extractelement <4 x i8> %vec1, i64 3
  %cv1e3 = zext i8 %v1e3 to i16
  %v2e3 = extractelement <4 x i8> %vec2, i64 3
  %cv2e3 = zext i8 %v2e3 to i16
  %mul4 = mul nuw nsw i16 %cv1e3, %cv2e3

  %acc = load i16, i16 addrspace(1)* %dst, align 2
  %mad1 = add i16 %mul1, %acc
  %mad2 = add i16 %mad1, %mul2
  %mad3 = add i16 %mad2, %mul3
  %mad4 = add i16 %mad3, %mul4

  store i16 %mad4, i16 addrspace(1)* %dst, align 2
  ret void
}

define amdgpu_kernel void @udot4_acc8(<4 x i8> addrspace(1)* %src1,
; GFX7-LABEL: udot4_acc8:
; GFX7:       ; %bb.0: ; %entry
; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; GFX7-NEXT:    s_mov_b32 s3, 0xf000
; GFX7-NEXT:    s_mov_b32 s2, -1
; GFX7-NEXT:    s_movk_i32 s8, 0xff
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    s_load_dword s4, s[4:5], 0x0
; GFX7-NEXT:    buffer_load_ubyte v0, off, s[0:3], 0
; GFX7-NEXT:    s_load_dword s5, s[6:7], 0x0
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    s_and_b32 s7, s4, s8
; GFX7-NEXT:    s_bfe_u32 s9, s4, 0x80008
; GFX7-NEXT:    s_and_b32 s6, s5, s8
; GFX7-NEXT:    s_bfe_u32 s8, s5, 0x80008
; GFX7-NEXT:    v_mov_b32_e32 v1, s6
; GFX7-NEXT:    s_bfe_u32 s10, s5, 0x80010
; GFX7-NEXT:    v_mov_b32_e32 v2, s8
; GFX7-NEXT:    s_bfe_u32 s11, s4, 0x80010
; GFX7-NEXT:    s_lshr_b32 s5, s5, 24
; GFX7-NEXT:    v_mov_b32_e32 v3, s10
; GFX7-NEXT:    s_lshr_b32 s4, s4, 24
; GFX7-NEXT:    s_waitcnt vmcnt(0)
; GFX7-NEXT:    v_mad_u32_u24 v0, s7, v1, v0
; GFX7-NEXT:    v_mad_u32_u24 v0, s9, v2, v0
; GFX7-NEXT:    v_mad_u32_u24 v0, s11, v3, v0
; GFX7-NEXT:    v_mov_b32_e32 v1, s5
; GFX7-NEXT:    v_mad_u32_u24 v0, s4, v1, v0
; GFX7-NEXT:    buffer_store_byte v0, off, s[0:3], 0
; GFX7-NEXT:    s_endpgm
;
; GFX8-LABEL: udot4_acc8:
; GFX8:       ; %bb.0: ; %entry
; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX8-NEXT:    s_movk_i32 s2, 0xff
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_mov_b32_e32 v0, s0
; GFX8-NEXT:    v_mov_b32_e32 v1, s1
; GFX8-NEXT:    flat_load_ubyte v2, v[0:1]
; GFX8-NEXT:    s_load_dword s0, s[4:5], 0x0
; GFX8-NEXT:    s_load_dword s1, s[6:7], 0x0
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    s_bfe_u32 s4, s0, 0x80008
; GFX8-NEXT:    s_and_b32 s3, s1, s2
; GFX8-NEXT:    s_bfe_u32 s5, s1, 0x80008
; GFX8-NEXT:    s_and_b32 s2, s0, s2
; GFX8-NEXT:    v_mov_b32_e32 v3, s3
; GFX8-NEXT:    s_bfe_u32 s6, s1, 0x80010
; GFX8-NEXT:    v_mov_b32_e32 v4, s5
; GFX8-NEXT:    s_bfe_u32 s7, s0, 0x80010
; GFX8-NEXT:    s_lshr_b32 s1, s1, 24
; GFX8-NEXT:    v_mov_b32_e32 v5, s6
; GFX8-NEXT:    s_lshr_b32 s0, s0, 24
; GFX8-NEXT:    s_waitcnt vmcnt(0)
; GFX8-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
; GFX8-NEXT:    v_mad_u32_u24 v2, s4, v4, v2
; GFX8-NEXT:    v_mad_u32_u24 v2, s7, v5, v2
; GFX8-NEXT:    v_mov_b32_e32 v3, s1
; GFX8-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
; GFX8-NEXT:    flat_store_byte v[0:1], v2
; GFX8-NEXT:    s_endpgm
;
; GFX9-NODL-LABEL: udot4_acc8:
; GFX9-NODL:       ; %bb.0: ; %entry
; GFX9-NODL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX9-NODL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX9-NODL-NEXT:    s_movk_i32 s2, 0xff
; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s0
; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s1
; GFX9-NODL-NEXT:    global_load_ubyte v2, v[0:1], off
; GFX9-NODL-NEXT:    s_load_dword s0, s[4:5], 0x0
; GFX9-NODL-NEXT:    s_load_dword s1, s[6:7], 0x0
; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NODL-NEXT:    s_bfe_u32 s4, s0, 0x80008
; GFX9-NODL-NEXT:    s_and_b32 s3, s1, s2
; GFX9-NODL-NEXT:    s_bfe_u32 s5, s1, 0x80008
; GFX9-NODL-NEXT:    s_and_b32 s2, s0, s2
; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s3
; GFX9-NODL-NEXT:    s_bfe_u32 s6, s1, 0x80010
; GFX9-NODL-NEXT:    v_mov_b32_e32 v4, s5
; GFX9-NODL-NEXT:    s_bfe_u32 s7, s0, 0x80010
; GFX9-NODL-NEXT:    s_lshr_b32 s1, s1, 24
; GFX9-NODL-NEXT:    v_mov_b32_e32 v5, s6
; GFX9-NODL-NEXT:    s_lshr_b32 s0, s0, 24
; GFX9-NODL-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s4, v4, v2
; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s7, v5, v2
; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s1
; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
; GFX9-NODL-NEXT:    global_store_byte v[0:1], v2, off
; GFX9-NODL-NEXT:    s_endpgm
;
; GFX9-DL-LABEL: udot4_acc8:
; GFX9-DL:       ; %bb.0: ; %entry
; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
; GFX9-DL-NEXT:    s_load_dword s3, s[6:7], 0x0
; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
; GFX9-DL-NEXT:    global_load_ubyte v2, v[0:1], off
; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s3
; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
; GFX9-DL-NEXT:    v_dot4_u32_u8 v2, s2, v3, v2
; GFX9-DL-NEXT:    global_store_byte v[0:1], v2, off
; GFX9-DL-NEXT:    s_endpgm
                                      <4 x i8> addrspace(1)* %src2,
                                      i8 addrspace(1)* nocapture %dst) {
entry:
  %vec1 = load <4 x i8>, <4 x i8> addrspace(1)* %src1
  %vec2 = load <4 x i8>, <4 x i8> addrspace(1)* %src2

  %v1e0 = extractelement <4 x i8> %vec1, i64 0
  %v2e0 = extractelement <4 x i8> %vec2, i64 0
  %mul1 = mul nuw nsw i8 %v1e0, %v2e0

  %v1e1 = extractelement <4 x i8> %vec1, i64 1
  %v2e1 = extractelement <4 x i8> %vec2, i64 1
  %mul2 = mul nuw nsw i8 %v1e1, %v2e1

  %v1e2 = extractelement <4 x i8> %vec1, i64 2
  %v2e2 = extractelement <4 x i8> %vec2, i64 2
  %mul3 = mul nuw nsw i8 %v1e2, %v2e2

  %v1e3 = extractelement <4 x i8> %vec1, i64 3
  %v2e3 = extractelement <4 x i8> %vec2, i64 3
  %mul4 = mul nuw nsw i8 %v1e3, %v2e3

  %acc = load i8, i8 addrspace(1)* %dst, align 2
  %mad1 = add i8 %mul1, %acc
  %mad2 = add i8 %mad1, %mul2
  %mad3 = add i8 %mad2, %mul3
  %mad4 = add i8 %mad3, %mul4

  store i8 %mad4, i8 addrspace(1)* %dst, align 2
  ret void
}

; TODO: Generate udot4?
define amdgpu_kernel void @udot2_8(<4 x i8> addrspace(1)* %src1,
; GFX7-LABEL: udot2_8:
; GFX7:       ; %bb.0: ; %entry
; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; GFX7-NEXT:    s_mov_b32 s3, 0xf000
; GFX7-NEXT:    s_mov_b32 s2, -1
; GFX7-NEXT:    s_movk_i32 s8, 0xff
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    s_load_dword s4, s[4:5], 0x0
; GFX7-NEXT:    buffer_load_ubyte v0, off, s[0:3], 0
; GFX7-NEXT:    s_load_dword s5, s[6:7], 0x0
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    s_and_b32 s7, s4, s8
; GFX7-NEXT:    s_bfe_u32 s4, s4, 0x80008
; GFX7-NEXT:    s_and_b32 s6, s5, s8
; GFX7-NEXT:    v_mov_b32_e32 v1, s6
; GFX7-NEXT:    s_bfe_u32 s5, s5, 0x80008
; GFX7-NEXT:    s_waitcnt vmcnt(0)
; GFX7-NEXT:    v_mad_u32_u24 v0, s7, v1, v0
; GFX7-NEXT:    v_mov_b32_e32 v1, s5
; GFX7-NEXT:    v_mad_u32_u24 v0, s4, v1, v0
; GFX7-NEXT:    buffer_store_byte v0, off, s[0:3], 0
; GFX7-NEXT:    s_endpgm
;
; GFX8-LABEL: udot2_8:
; GFX8:       ; %bb.0: ; %entry
; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX8-NEXT:    s_movk_i32 s2, 0xff
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_mov_b32_e32 v0, s0
; GFX8-NEXT:    v_mov_b32_e32 v1, s1
; GFX8-NEXT:    flat_load_ubyte v2, v[0:1]
; GFX8-NEXT:    s_load_dword s0, s[4:5], 0x0
; GFX8-NEXT:    s_load_dword s1, s[6:7], 0x0
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    s_and_b32 s3, s1, s2
; GFX8-NEXT:    s_and_b32 s2, s0, s2
; GFX8-NEXT:    v_mov_b32_e32 v3, s3
; GFX8-NEXT:    s_bfe_u32 s1, s1, 0x80008
; GFX8-NEXT:    s_bfe_u32 s0, s0, 0x80008
; GFX8-NEXT:    s_waitcnt vmcnt(0)
; GFX8-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
; GFX8-NEXT:    v_mov_b32_e32 v3, s1
; GFX8-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
; GFX8-NEXT:    flat_store_byte v[0:1], v2
; GFX8-NEXT:    s_endpgm
;
; GFX9-NODL-LABEL: udot2_8:
; GFX9-NODL:       ; %bb.0: ; %entry
; GFX9-NODL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX9-NODL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX9-NODL-NEXT:    s_movk_i32 s2, 0xff
; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s0
; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s1
; GFX9-NODL-NEXT:    global_load_ubyte v2, v[0:1], off
; GFX9-NODL-NEXT:    s_load_dword s0, s[4:5], 0x0
; GFX9-NODL-NEXT:    s_load_dword s1, s[6:7], 0x0
; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NODL-NEXT:    s_and_b32 s3, s1, s2
; GFX9-NODL-NEXT:    s_and_b32 s2, s0, s2
; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s3
; GFX9-NODL-NEXT:    s_bfe_u32 s1, s1, 0x80008
; GFX9-NODL-NEXT:    s_bfe_u32 s0, s0, 0x80008
; GFX9-NODL-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s1
; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
; GFX9-NODL-NEXT:    global_store_byte v[0:1], v2, off
; GFX9-NODL-NEXT:    s_endpgm
;
; GFX9-DL-LABEL: udot2_8:
; GFX9-DL:       ; %bb.0: ; %entry
; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX9-DL-NEXT:    s_movk_i32 s2, 0xff
; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
; GFX9-DL-NEXT:    global_load_ubyte v2, v[0:1], off
; GFX9-DL-NEXT:    s_load_dword s0, s[4:5], 0x0
; GFX9-DL-NEXT:    s_load_dword s1, s[6:7], 0x0
; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-DL-NEXT:    s_and_b32 s3, s1, s2
; GFX9-DL-NEXT:    s_and_b32 s2, s0, s2
; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s3
; GFX9-DL-NEXT:    s_bfe_u32 s1, s1, 0x80008
; GFX9-DL-NEXT:    s_bfe_u32 s0, s0, 0x80008
; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s1
; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
; GFX9-DL-NEXT:    global_store_byte v[0:1], v2, off
; GFX9-DL-NEXT:    s_endpgm
                                   <4 x i8> addrspace(1)* %src2,
                                   i8 addrspace(1)* nocapture %dst) {
entry:
  %vec1 = load <4 x i8>, <4 x i8> addrspace(1)* %src1
  %vec2 = load <4 x i8>, <4 x i8> addrspace(1)* %src2

  %v1e0 = extractelement <4 x i8> %vec1, i64 0
  %v2e0 = extractelement <4 x i8> %vec2, i64 0
  %mul1 = mul nuw nsw i8 %v1e0, %v2e0

  %v1e1 = extractelement <4 x i8> %vec1, i64 1
  %v2e1 = extractelement <4 x i8> %vec2, i64 1
  %mul2 = mul nuw nsw i8 %v1e1, %v2e1

  %acc = load i8, i8 addrspace(1)* %dst, align 2
  %mad1 = add i8 %mul1, %acc
  %mad2 = add i8 %mad1, %mul2
  store i8 %mad2, i8 addrspace(1)* %dst, align 2
  ret void
}

define amdgpu_kernel void @udot4_CommutationInsideMAD(<4 x i8> addrspace(1)* %src1,
; GFX7-LABEL: udot4_CommutationInsideMAD:
; GFX7:       ; %bb.0: ; %entry
; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; GFX7-NEXT:    s_mov_b32 s3, 0xf000
; GFX7-NEXT:    s_mov_b32 s2, -1
; GFX7-NEXT:    s_movk_i32 s8, 0xff
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    s_load_dword s4, s[4:5], 0x0
; GFX7-NEXT:    buffer_load_ubyte v0, off, s[0:3], 0
; GFX7-NEXT:    s_load_dword s5, s[6:7], 0x0
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    s_and_b32 s6, s4, s8
; GFX7-NEXT:    v_mov_b32_e32 v1, s6
; GFX7-NEXT:    s_and_b32 s7, s5, s8
; GFX7-NEXT:    s_bfe_u32 s8, s4, 0x80008
; GFX7-NEXT:    s_bfe_u32 s10, s4, 0x80010
; GFX7-NEXT:    s_bfe_u32 s9, s5, 0x80008
; GFX7-NEXT:    v_mov_b32_e32 v2, s8
; GFX7-NEXT:    s_bfe_u32 s11, s5, 0x80010
; GFX7-NEXT:    s_lshr_b32 s4, s4, 24
; GFX7-NEXT:    v_mov_b32_e32 v3, s10
; GFX7-NEXT:    s_lshr_b32 s5, s5, 24
; GFX7-NEXT:    s_waitcnt vmcnt(0)
; GFX7-NEXT:    v_mad_u32_u24 v0, s7, v1, v0
; GFX7-NEXT:    v_mad_u32_u24 v0, s9, v2, v0
; GFX7-NEXT:    v_mad_u32_u24 v0, s11, v3, v0
; GFX7-NEXT:    v_mov_b32_e32 v1, s4
; GFX7-NEXT:    v_mad_u32_u24 v0, s5, v1, v0
; GFX7-NEXT:    buffer_store_byte v0, off, s[0:3], 0
; GFX7-NEXT:    s_endpgm
;
; GFX8-LABEL: udot4_CommutationInsideMAD:
; GFX8:       ; %bb.0: ; %entry
; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_mov_b32_e32 v0, s0
; GFX8-NEXT:    v_mov_b32_e32 v1, s1
; GFX8-NEXT:    flat_load_ubyte v2, v[0:1]
; GFX8-NEXT:    s_load_dword s1, s[4:5], 0x0
; GFX8-NEXT:    s_load_dword s2, s[6:7], 0x0
; GFX8-NEXT:    s_movk_i32 s0, 0xff
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    s_and_b32 s3, s1, s0
; GFX8-NEXT:    s_bfe_u32 s5, s1, 0x80008
; GFX8-NEXT:    s_and_b32 s0, s2, s0
; GFX8-NEXT:    v_mov_b32_e32 v3, s3
; GFX8-NEXT:    s_bfe_u32 s6, s1, 0x80010
; GFX8-NEXT:    s_bfe_u32 s4, s2, 0x80008
; GFX8-NEXT:    v_mov_b32_e32 v4, s5
; GFX8-NEXT:    s_bfe_u32 s7, s2, 0x80010
; GFX8-NEXT:    s_lshr_b32 s1, s1, 24
; GFX8-NEXT:    v_mov_b32_e32 v5, s6
; GFX8-NEXT:    s_lshr_b32 s2, s2, 24
; GFX8-NEXT:    s_waitcnt vmcnt(0)
; GFX8-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
; GFX8-NEXT:    v_mad_u32_u24 v2, s4, v4, v2
; GFX8-NEXT:    v_mad_u32_u24 v2, s7, v5, v2
; GFX8-NEXT:    v_mov_b32_e32 v3, s1
; GFX8-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
; GFX8-NEXT:    flat_store_byte v[0:1], v2
; GFX8-NEXT:    s_endpgm
;
; GFX9-NODL-LABEL: udot4_CommutationInsideMAD:
; GFX9-NODL:       ; %bb.0: ; %entry
; GFX9-NODL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX9-NODL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s0
; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s1
; GFX9-NODL-NEXT:    global_load_ubyte v2, v[0:1], off
; GFX9-NODL-NEXT:    s_load_dword s1, s[4:5], 0x0
; GFX9-NODL-NEXT:    s_load_dword s2, s[6:7], 0x0
; GFX9-NODL-NEXT:    s_movk_i32 s0, 0xff
; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NODL-NEXT:    s_and_b32 s3, s1, s0
; GFX9-NODL-NEXT:    s_bfe_u32 s5, s1, 0x80008
; GFX9-NODL-NEXT:    s_and_b32 s0, s2, s0
; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s3
; GFX9-NODL-NEXT:    s_bfe_u32 s6, s1, 0x80010
; GFX9-NODL-NEXT:    s_bfe_u32 s4, s2, 0x80008
; GFX9-NODL-NEXT:    v_mov_b32_e32 v4, s5
; GFX9-NODL-NEXT:    s_bfe_u32 s7, s2, 0x80010
; GFX9-NODL-NEXT:    s_lshr_b32 s1, s1, 24
; GFX9-NODL-NEXT:    v_mov_b32_e32 v5, s6
; GFX9-NODL-NEXT:    s_lshr_b32 s2, s2, 24
; GFX9-NODL-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s4, v4, v2
; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s7, v5, v2
; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s1
; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
; GFX9-NODL-NEXT:    global_store_byte v[0:1], v2, off
; GFX9-NODL-NEXT:    s_endpgm
;
; GFX9-DL-LABEL: udot4_CommutationInsideMAD:
; GFX9-DL:       ; %bb.0: ; %entry
; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
; GFX9-DL-NEXT:    s_load_dword s3, s[6:7], 0x0
; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
; GFX9-DL-NEXT:    global_load_ubyte v2, v[0:1], off
; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s2
; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
; GFX9-DL-NEXT:    v_dot4_u32_u8 v2, s3, v3, v2
; GFX9-DL-NEXT:    global_store_byte v[0:1], v2, off
; GFX9-DL-NEXT:    s_endpgm
                                                      <4 x i8> addrspace(1)* %src2,
                                                      i8 addrspace(1)* nocapture %dst) {
entry:
  %vec1 = load <4 x i8>, <4 x i8> addrspace(1)* %src1
  %vec2 = load <4 x i8>, <4 x i8> addrspace(1)* %src2

  %v1e0 = extractelement <4 x i8> %vec1, i64 0
  %v2e0 = extractelement <4 x i8> %vec2, i64 0
  %mul1 = mul nuw nsw i8 %v2e0, %v1e0

  %v1e1 = extractelement <4 x i8> %vec1, i64 1
  %v2e1 = extractelement <4 x i8> %vec2, i64 1
  %mul2 = mul nuw nsw i8 %v2e1, %v1e1

  %v1e2 = extractelement <4 x i8> %vec1, i64 2
  %v2e2 = extractelement <4 x i8> %vec2, i64 2
  %mul3 = mul nuw nsw i8 %v2e2, %v1e2

  %v1e3 = extractelement <4 x i8> %vec1, i64 3
  %v2e3 = extractelement <4 x i8> %vec2, i64 3
  %mul4 = mul nuw nsw i8 %v2e3, %v1e3

  %acc = load i8, i8 addrspace(1)* %dst, align 2
  %mad1 = add i8 %acc, %mul1
  %mad2 = add i8 %mul2, %mad1
  %mad3 = add i8 %mul3, %mad2
  %mad4 = add i8 %mul4, %mad3

  store i8 %mad4, i8 addrspace(1)* %dst, align 2
  ret void
}

; TODO: Support commutation accross the adds.
define amdgpu_kernel void @udot4_CommutationAccrossMADs(<4 x i8> addrspace(1)* %src1,
; GFX7-LABEL: udot4_CommutationAccrossMADs:
; GFX7:       ; %bb.0: ; %entry
; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; GFX7-NEXT:    s_mov_b32 s3, 0xf000
; GFX7-NEXT:    s_mov_b32 s2, -1
; GFX7-NEXT:    s_movk_i32 s8, 0xff
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    s_load_dword s4, s[4:5], 0x0
; GFX7-NEXT:    buffer_load_ubyte v0, off, s[0:3], 0
; GFX7-NEXT:    s_load_dword s5, s[6:7], 0x0
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    s_and_b32 s6, s4, s8
; GFX7-NEXT:    s_bfe_u32 s10, s4, 0x80010
; GFX7-NEXT:    s_and_b32 s7, s5, s8
; GFX7-NEXT:    s_bfe_u32 s8, s4, 0x80008
; GFX7-NEXT:    s_bfe_u32 s9, s5, 0x80008
; GFX7-NEXT:    v_mov_b32_e32 v1, s8
; GFX7-NEXT:    v_mov_b32_e32 v2, s6
; GFX7-NEXT:    s_bfe_u32 s11, s5, 0x80010
; GFX7-NEXT:    s_lshr_b32 s4, s4, 24
; GFX7-NEXT:    v_mov_b32_e32 v3, s10
; GFX7-NEXT:    s_lshr_b32 s5, s5, 24
; GFX7-NEXT:    s_waitcnt vmcnt(0)
; GFX7-NEXT:    v_mad_u32_u24 v0, s9, v1, v0
; GFX7-NEXT:    v_mad_u32_u24 v0, s7, v2, v0
; GFX7-NEXT:    v_mad_u32_u24 v0, s11, v3, v0
; GFX7-NEXT:    v_mov_b32_e32 v1, s4
; GFX7-NEXT:    v_mad_u32_u24 v0, s5, v1, v0
; GFX7-NEXT:    buffer_store_byte v0, off, s[0:3], 0
; GFX7-NEXT:    s_endpgm
;
; GFX8-LABEL: udot4_CommutationAccrossMADs:
; GFX8:       ; %bb.0: ; %entry
; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_mov_b32_e32 v0, s0
; GFX8-NEXT:    v_mov_b32_e32 v1, s1
; GFX8-NEXT:    flat_load_ubyte v2, v[0:1]
; GFX8-NEXT:    s_load_dword s1, s[4:5], 0x0
; GFX8-NEXT:    s_load_dword s2, s[6:7], 0x0
; GFX8-NEXT:    s_movk_i32 s0, 0xff
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    s_bfe_u32 s4, s1, 0x80008
; GFX8-NEXT:    s_and_b32 s3, s2, s0
; GFX8-NEXT:    s_and_b32 s0, s1, s0
; GFX8-NEXT:    s_bfe_u32 s5, s2, 0x80008
; GFX8-NEXT:    v_mov_b32_e32 v3, s4
; GFX8-NEXT:    s_bfe_u32 s6, s1, 0x80010
; GFX8-NEXT:    v_mov_b32_e32 v4, s0
; GFX8-NEXT:    s_bfe_u32 s7, s2, 0x80010
; GFX8-NEXT:    s_lshr_b32 s1, s1, 24
; GFX8-NEXT:    v_mov_b32_e32 v5, s6
; GFX8-NEXT:    s_lshr_b32 s2, s2, 24
; GFX8-NEXT:    s_waitcnt vmcnt(0)
; GFX8-NEXT:    v_mad_u32_u24 v2, s5, v3, v2
; GFX8-NEXT:    v_mad_u32_u24 v2, s3, v4, v2
; GFX8-NEXT:    v_mad_u32_u24 v2, s7, v5, v2
; GFX8-NEXT:    v_mov_b32_e32 v3, s1
; GFX8-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
; GFX8-NEXT:    flat_store_byte v[0:1], v2
; GFX8-NEXT:    s_endpgm
;
; GFX9-NODL-LABEL: udot4_CommutationAccrossMADs:
; GFX9-NODL:       ; %bb.0: ; %entry
; GFX9-NODL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX9-NODL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s0
; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s1
; GFX9-NODL-NEXT:    global_load_ubyte v2, v[0:1], off
; GFX9-NODL-NEXT:    s_load_dword s1, s[4:5], 0x0
; GFX9-NODL-NEXT:    s_load_dword s2, s[6:7], 0x0
; GFX9-NODL-NEXT:    s_movk_i32 s0, 0xff
; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NODL-NEXT:    s_bfe_u32 s4, s1, 0x80008
; GFX9-NODL-NEXT:    s_and_b32 s3, s2, s0
; GFX9-NODL-NEXT:    s_and_b32 s0, s1, s0
; GFX9-NODL-NEXT:    s_bfe_u32 s5, s2, 0x80008
; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s4
; GFX9-NODL-NEXT:    s_bfe_u32 s6, s1, 0x80010
; GFX9-NODL-NEXT:    v_mov_b32_e32 v4, s0
; GFX9-NODL-NEXT:    s_bfe_u32 s7, s2, 0x80010
; GFX9-NODL-NEXT:    s_lshr_b32 s1, s1, 24
; GFX9-NODL-NEXT:    v_mov_b32_e32 v5, s6
; GFX9-NODL-NEXT:    s_lshr_b32 s2, s2, 24
; GFX9-NODL-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s5, v3, v2
; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s3, v4, v2
; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s7, v5, v2
; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s1
; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
; GFX9-NODL-NEXT:    global_store_byte v[0:1], v2, off
; GFX9-NODL-NEXT:    s_endpgm
;
; GFX9-DL-LABEL: udot4_CommutationAccrossMADs:
; GFX9-DL:       ; %bb.0: ; %entry
; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
; GFX9-DL-NEXT:    global_load_ubyte v2, v[0:1], off
; GFX9-DL-NEXT:    s_load_dword s1, s[4:5], 0x0
; GFX9-DL-NEXT:    s_load_dword s2, s[6:7], 0x0
; GFX9-DL-NEXT:    s_movk_i32 s0, 0xff
; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-DL-NEXT:    s_bfe_u32 s4, s1, 0x80008
; GFX9-DL-NEXT:    s_and_b32 s3, s2, s0
; GFX9-DL-NEXT:    s_and_b32 s0, s1, s0
; GFX9-DL-NEXT:    s_bfe_u32 s5, s2, 0x80008
; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s4
; GFX9-DL-NEXT:    s_bfe_u32 s6, s1, 0x80010
; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s0
; GFX9-DL-NEXT:    s_bfe_u32 s7, s2, 0x80010
; GFX9-DL-NEXT:    s_lshr_b32 s1, s1, 24
; GFX9-DL-NEXT:    v_mov_b32_e32 v5, s6
; GFX9-DL-NEXT:    s_lshr_b32 s2, s2, 24
; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s5, v3, v2
; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s3, v4, v2
; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s7, v5, v2
; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s1
; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s2, v3, v2
; GFX9-DL-NEXT:    global_store_byte v[0:1], v2, off
; GFX9-DL-NEXT:    s_endpgm
                                                        <4 x i8> addrspace(1)* %src2,
                                                        i8 addrspace(1)* nocapture %dst) {
entry:
  %vec1 = load <4 x i8>, <4 x i8> addrspace(1)* %src1
  %vec2 = load <4 x i8>, <4 x i8> addrspace(1)* %src2

  %v1e0 = extractelement <4 x i8> %vec1, i64 0
  %v2e0 = extractelement <4 x i8> %vec2, i64 0
  %mul1 = mul nuw nsw i8 %v2e0, %v1e0

  %v1e1 = extractelement <4 x i8> %vec1, i64 1
  %v2e1 = extractelement <4 x i8> %vec2, i64 1
  %mul2 = mul nuw nsw i8 %v2e1, %v1e1

  %v1e2 = extractelement <4 x i8> %vec1, i64 2
  %v2e2 = extractelement <4 x i8> %vec2, i64 2
  %mul3 = mul nuw nsw i8 %v2e2, %v1e2

  %v1e3 = extractelement <4 x i8> %vec1, i64 3
  %v2e3 = extractelement <4 x i8> %vec2, i64 3
  %mul4 = mul nuw nsw i8 %v2e3, %v1e3

  %acc = load i8, i8 addrspace(1)* %dst, align 2
  %mad1 = add i8 %acc, %mul2
  %mad2 = add i8 %mad1, %mul1
  %mad3 = add i8 %mad2, %mul3
  %mad4 = add i8 %mad3, %mul4

  store i8 %mad4, i8 addrspace(1)* %dst, align 2
  ret void
}

define amdgpu_kernel void @udot4_multiuse_mul1(<4 x i8> addrspace(1)* %src1,
; GFX7-LABEL: udot4_multiuse_mul1:
; GFX7:       ; %bb.0: ; %entry
; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; GFX7-NEXT:    s_movk_i32 s8, 0xff
; GFX7-NEXT:    s_mov_b32 s3, 0xf000
; GFX7-NEXT:    s_mov_b32 s2, -1
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    s_load_dword s4, s[4:5], 0x0
; GFX7-NEXT:    s_load_dword s5, s[6:7], 0x0
; GFX7-NEXT:    s_load_dword s6, s[0:1], 0x0
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    s_and_b32 s7, s4, s8
; GFX7-NEXT:    s_and_b32 s8, s5, s8
; GFX7-NEXT:    s_bfe_u32 s10, s5, 0x80008
; GFX7-NEXT:    v_mov_b32_e32 v0, s8
; GFX7-NEXT:    v_mov_b32_e32 v1, s6
; GFX7-NEXT:    s_bfe_u32 s9, s4, 0x80008
; GFX7-NEXT:    v_mad_u32_u24 v1, s7, v0, v1
; GFX7-NEXT:    v_mov_b32_e32 v2, s10
; GFX7-NEXT:    s_bfe_u32 s12, s5, 0x80010
; GFX7-NEXT:    v_mad_u32_u24 v1, s9, v2, v1
; GFX7-NEXT:    s_bfe_u32 s11, s4, 0x80010
; GFX7-NEXT:    v_mad_u32_u24 v0, s7, v0, v1
; GFX7-NEXT:    v_mov_b32_e32 v1, s12
; GFX7-NEXT:    s_lshr_b32 s5, s5, 24
; GFX7-NEXT:    v_mad_u32_u24 v0, s11, v1, v0
; GFX7-NEXT:    s_lshr_b32 s4, s4, 24
; GFX7-NEXT:    v_mov_b32_e32 v1, s5
; GFX7-NEXT:    v_mad_u32_u24 v0, s4, v1, v0
; GFX7-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GFX7-NEXT:    s_endpgm
;
; GFX8-LABEL: udot4_multiuse_mul1:
; GFX8:       ; %bb.0: ; %entry
; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX8-NEXT:    s_movk_i32 s2, 0xff
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    s_load_dword s3, s[4:5], 0x0
; GFX8-NEXT:    s_load_dword s4, s[6:7], 0x0
; GFX8-NEXT:    s_load_dword s5, s[0:1], 0x0
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    s_and_b32 s6, s3, s2
; GFX8-NEXT:    s_and_b32 s2, s4, s2
; GFX8-NEXT:    s_bfe_u32 s8, s4, 0x80008
; GFX8-NEXT:    v_mov_b32_e32 v0, s2
; GFX8-NEXT:    v_mov_b32_e32 v1, s5
; GFX8-NEXT:    s_bfe_u32 s7, s3, 0x80008
; GFX8-NEXT:    v_mad_u32_u24 v1, s6, v0, v1
; GFX8-NEXT:    v_mov_b32_e32 v2, s8
; GFX8-NEXT:    s_bfe_u32 s10, s4, 0x80010
; GFX8-NEXT:    v_mad_u32_u24 v1, s7, v2, v1
; GFX8-NEXT:    s_bfe_u32 s9, s3, 0x80010
; GFX8-NEXT:    v_mad_u32_u24 v0, s6, v0, v1
; GFX8-NEXT:    v_mov_b32_e32 v1, s10
; GFX8-NEXT:    s_lshr_b32 s4, s4, 24
; GFX8-NEXT:    v_mad_u32_u24 v0, s9, v1, v0
; GFX8-NEXT:    s_lshr_b32 s3, s3, 24
; GFX8-NEXT:    v_mov_b32_e32 v1, s4
; GFX8-NEXT:    v_mad_u32_u24 v2, s3, v1, v0
; GFX8-NEXT:    v_mov_b32_e32 v0, s0
; GFX8-NEXT:    v_mov_b32_e32 v1, s1
; GFX8-NEXT:    flat_store_dword v[0:1], v2
; GFX8-NEXT:    s_endpgm
;
; GFX9-NODL-LABEL: udot4_multiuse_mul1:
; GFX9-NODL:       ; %bb.0: ; %entry
; GFX9-NODL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX9-NODL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX9-NODL-NEXT:    s_movk_i32 s2, 0xff
; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NODL-NEXT:    s_load_dword s3, s[4:5], 0x0
; GFX9-NODL-NEXT:    s_load_dword s4, s[6:7], 0x0
; GFX9-NODL-NEXT:    s_load_dword s5, s[0:1], 0x0
; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NODL-NEXT:    s_and_b32 s6, s3, s2
; GFX9-NODL-NEXT:    s_and_b32 s2, s4, s2
; GFX9-NODL-NEXT:    s_bfe_u32 s8, s4, 0x80008
; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s2
; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s5
; GFX9-NODL-NEXT:    s_bfe_u32 s7, s3, 0x80008
; GFX9-NODL-NEXT:    v_mad_u32_u24 v1, s6, v0, v1
; GFX9-NODL-NEXT:    v_mov_b32_e32 v2, s8
; GFX9-NODL-NEXT:    s_bfe_u32 s10, s4, 0x80010
; GFX9-NODL-NEXT:    v_mad_u32_u24 v1, s7, v2, v1
; GFX9-NODL-NEXT:    s_bfe_u32 s9, s3, 0x80010
; GFX9-NODL-NEXT:    v_mad_u32_u24 v0, s6, v0, v1
; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s10
; GFX9-NODL-NEXT:    s_lshr_b32 s4, s4, 24
; GFX9-NODL-NEXT:    v_mad_u32_u24 v0, s9, v1, v0
; GFX9-NODL-NEXT:    s_lshr_b32 s3, s3, 24
; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s4
; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s3, v1, v0
; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s0
; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s1
; GFX9-NODL-NEXT:    global_store_dword v[0:1], v2, off
; GFX9-NODL-NEXT:    s_endpgm
;
; GFX9-DL-LABEL: udot4_multiuse_mul1:
; GFX9-DL:       ; %bb.0: ; %entry
; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX9-DL-NEXT:    s_movk_i32 s2, 0xff
; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-DL-NEXT:    s_load_dword s3, s[4:5], 0x0
; GFX9-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
; GFX9-DL-NEXT:    s_load_dword s5, s[0:1], 0x0
; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-DL-NEXT:    s_and_b32 s6, s3, s2
; GFX9-DL-NEXT:    s_and_b32 s2, s4, s2
; GFX9-DL-NEXT:    s_bfe_u32 s8, s4, 0x80008
; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s2
; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s5
; GFX9-DL-NEXT:    s_bfe_u32 s7, s3, 0x80008
; GFX9-DL-NEXT:    v_mad_u32_u24 v1, s6, v0, v1
; GFX9-DL-NEXT:    v_mov_b32_e32 v2, s8
; GFX9-DL-NEXT:    s_bfe_u32 s10, s4, 0x80010
; GFX9-DL-NEXT:    v_mad_u32_u24 v1, s7, v2, v1
; GFX9-DL-NEXT:    s_bfe_u32 s9, s3, 0x80010
; GFX9-DL-NEXT:    v_mad_u32_u24 v0, s6, v0, v1
; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s10
; GFX9-DL-NEXT:    s_lshr_b32 s4, s4, 24
; GFX9-DL-NEXT:    v_mad_u32_u24 v0, s9, v1, v0
; GFX9-DL-NEXT:    s_lshr_b32 s3, s3, 24
; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s4
; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s3, v1, v0
; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
; GFX9-DL-NEXT:    global_store_dword v[0:1], v2, off
; GFX9-DL-NEXT:    s_endpgm
                                               <4 x i8> addrspace(1)* %src2,
                                               i32 addrspace(1)* nocapture %dst) {
entry:
  %vec1 = load <4 x i8>, <4 x i8> addrspace(1)* %src1
  %vec2 = load <4 x i8>, <4 x i8> addrspace(1)* %src2

  %v1e0 = extractelement <4 x i8> %vec1, i64 0
  %cv1e0 = zext i8 %v1e0 to i32
  %v2e0 = extractelement <4 x i8> %vec2, i64 0
  %cv2e0 = zext i8 %v2e0 to i32
  %mul1 = mul nuw nsw i32 %cv1e0, %cv2e0

  %v1e1 = extractelement <4 x i8> %vec1, i64 1
  %cv1e1 = zext i8 %v1e1 to i32
  %v2e1 = extractelement <4 x i8> %vec2, i64 1
  %cv2e1 = zext i8 %v2e1 to i32
  %mul2 = mul nuw nsw i32 %cv1e1, %cv2e1

  %v1e2 = extractelement <4 x i8> %vec1, i64 2
  %cv1e2 = zext i8 %v1e2 to i32
  %v2e2 = extractelement <4 x i8> %vec2, i64 2
  %cv2e2 = zext i8 %v2e2 to i32
  %mul3 = mul nuw nsw i32 %cv1e2, %cv2e2

  %v1e3 = extractelement <4 x i8> %vec1, i64 3
  %cv1e3 = zext i8 %v1e3 to i32
  %v2e3 = extractelement <4 x i8> %vec2, i64 3
  %cv2e3 = zext i8 %v2e3 to i32
  %mul4 = mul nuw nsw i32 %cv1e3, %cv2e3

  %acc = load i32, i32 addrspace(1)* %dst, align 4
  %add = add i32 %mul1, %acc
  %add1 = add i32 %mul2, %add
  %add2 = add i32 %add1, %mul1
  %add3 = add i32 %add2, %mul3
  %add4 = add i32 %add3, %mul4

  store i32 %add4, i32 addrspace(1)* %dst, align 4
  ret void
}

define amdgpu_kernel void @udot4_multiuse_add1(<4 x i8> addrspace(1)* %src1,
; GFX7-LABEL: udot4_multiuse_add1:
; GFX7:       ; %bb.0: ; %entry
; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; GFX7-NEXT:    s_movk_i32 s8, 0xff
; GFX7-NEXT:    s_mov_b32 s3, 0xf000
; GFX7-NEXT:    s_mov_b32 s2, -1
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    s_load_dword s4, s[4:5], 0x0
; GFX7-NEXT:    s_load_dword s5, s[6:7], 0x0
; GFX7-NEXT:    s_load_dword s6, s[0:1], 0x0
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    s_and_b32 s7, s4, s8
; GFX7-NEXT:    s_bfe_u32 s10, s5, 0x80008
; GFX7-NEXT:    s_and_b32 s8, s5, s8
; GFX7-NEXT:    s_bfe_u32 s9, s4, 0x80008
; GFX7-NEXT:    v_mov_b32_e32 v0, s10
; GFX7-NEXT:    v_mov_b32_e32 v1, s6
; GFX7-NEXT:    v_mad_u32_u24 v0, s9, v0, v1
; GFX7-NEXT:    s_bfe_u32 s12, s5, 0x80010
; GFX7-NEXT:    v_mov_b32_e32 v2, s8
; GFX7-NEXT:    s_bfe_u32 s11, s4, 0x80010
; GFX7-NEXT:    v_add_i32_e32 v1, vcc, s6, v0
; GFX7-NEXT:    v_mad_u32_u24 v0, s7, v2, v0
; GFX7-NEXT:    v_mov_b32_e32 v2, s12
; GFX7-NEXT:    s_lshr_b32 s5, s5, 24
; GFX7-NEXT:    v_mad_u32_u24 v0, s11, v2, v0
; GFX7-NEXT:    s_lshr_b32 s4, s4, 24
; GFX7-NEXT:    v_mov_b32_e32 v2, s5
; GFX7-NEXT:    v_mad_u32_u24 v0, s4, v2, v0
; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
; GFX7-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GFX7-NEXT:    s_endpgm
;
; GFX8-LABEL: udot4_multiuse_add1:
; GFX8:       ; %bb.0: ; %entry
; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX8-NEXT:    s_movk_i32 s2, 0xff
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    s_load_dword s3, s[4:5], 0x0
; GFX8-NEXT:    s_load_dword s4, s[6:7], 0x0
; GFX8-NEXT:    s_load_dword s5, s[0:1], 0x0
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    s_and_b32 s6, s3, s2
; GFX8-NEXT:    s_bfe_u32 s8, s4, 0x80008
; GFX8-NEXT:    s_and_b32 s2, s4, s2
; GFX8-NEXT:    s_bfe_u32 s7, s3, 0x80008
; GFX8-NEXT:    v_mov_b32_e32 v0, s8
; GFX8-NEXT:    v_mov_b32_e32 v1, s5
; GFX8-NEXT:    v_mad_u32_u24 v0, s7, v0, v1
; GFX8-NEXT:    s_bfe_u32 s10, s4, 0x80010
; GFX8-NEXT:    v_mov_b32_e32 v2, s2
; GFX8-NEXT:    s_bfe_u32 s9, s3, 0x80010
; GFX8-NEXT:    v_add_u32_e32 v1, vcc, s5, v0
; GFX8-NEXT:    v_mad_u32_u24 v0, s6, v2, v0
; GFX8-NEXT:    v_mov_b32_e32 v2, s10
; GFX8-NEXT:    s_lshr_b32 s4, s4, 24
; GFX8-NEXT:    v_mad_u32_u24 v0, s9, v2, v0
; GFX8-NEXT:    s_lshr_b32 s3, s3, 24
; GFX8-NEXT:    v_mov_b32_e32 v2, s4
; GFX8-NEXT:    v_mad_u32_u24 v0, s3, v2, v0
; GFX8-NEXT:    v_add_u32_e32 v2, vcc, v0, v1
; GFX8-NEXT:    v_mov_b32_e32 v0, s0
; GFX8-NEXT:    v_mov_b32_e32 v1, s1
; GFX8-NEXT:    flat_store_dword v[0:1], v2
; GFX8-NEXT:    s_endpgm
;
; GFX9-NODL-LABEL: udot4_multiuse_add1:
; GFX9-NODL:       ; %bb.0: ; %entry
; GFX9-NODL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX9-NODL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX9-NODL-NEXT:    s_movk_i32 s2, 0xff
; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NODL-NEXT:    s_load_dword s3, s[4:5], 0x0
; GFX9-NODL-NEXT:    s_load_dword s4, s[6:7], 0x0
; GFX9-NODL-NEXT:    s_load_dword s5, s[0:1], 0x0
; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NODL-NEXT:    s_and_b32 s6, s3, s2
; GFX9-NODL-NEXT:    s_bfe_u32 s8, s4, 0x80008
; GFX9-NODL-NEXT:    s_and_b32 s2, s4, s2
; GFX9-NODL-NEXT:    s_bfe_u32 s7, s3, 0x80008
; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s8
; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s5
; GFX9-NODL-NEXT:    v_mad_u32_u24 v0, s7, v0, v1
; GFX9-NODL-NEXT:    s_bfe_u32 s10, s4, 0x80010
; GFX9-NODL-NEXT:    v_mov_b32_e32 v2, s2
; GFX9-NODL-NEXT:    s_bfe_u32 s9, s3, 0x80010
; GFX9-NODL-NEXT:    v_add_u32_e32 v1, s5, v0
; GFX9-NODL-NEXT:    v_mad_u32_u24 v0, s6, v2, v0
; GFX9-NODL-NEXT:    v_mov_b32_e32 v2, s10
; GFX9-NODL-NEXT:    s_lshr_b32 s4, s4, 24
; GFX9-NODL-NEXT:    v_mad_u32_u24 v0, s9, v2, v0
; GFX9-NODL-NEXT:    s_lshr_b32 s3, s3, 24
; GFX9-NODL-NEXT:    v_mov_b32_e32 v2, s4
; GFX9-NODL-NEXT:    v_mad_u32_u24 v0, s3, v2, v0
; GFX9-NODL-NEXT:    v_add_u32_e32 v2, v0, v1
; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s0
; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s1
; GFX9-NODL-NEXT:    global_store_dword v[0:1], v2, off
; GFX9-NODL-NEXT:    s_endpgm
;
; GFX9-DL-LABEL: udot4_multiuse_add1:
; GFX9-DL:       ; %bb.0: ; %entry
; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX9-DL-NEXT:    s_movk_i32 s2, 0xff
; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-DL-NEXT:    s_load_dword s3, s[4:5], 0x0
; GFX9-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
; GFX9-DL-NEXT:    s_load_dword s5, s[0:1], 0x0
; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-DL-NEXT:    s_and_b32 s6, s3, s2
; GFX9-DL-NEXT:    s_bfe_u32 s8, s4, 0x80008
; GFX9-DL-NEXT:    s_and_b32 s2, s4, s2
; GFX9-DL-NEXT:    s_bfe_u32 s7, s3, 0x80008
; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s8
; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s5
; GFX9-DL-NEXT:    v_mad_u32_u24 v0, s7, v0, v1
; GFX9-DL-NEXT:    s_bfe_u32 s10, s4, 0x80010
; GFX9-DL-NEXT:    v_mov_b32_e32 v2, s2
; GFX9-DL-NEXT:    s_bfe_u32 s9, s3, 0x80010
; GFX9-DL-NEXT:    v_add_u32_e32 v1, s5, v0
; GFX9-DL-NEXT:    v_mad_u32_u24 v0, s6, v2, v0
; GFX9-DL-NEXT:    v_mov_b32_e32 v2, s10
; GFX9-DL-NEXT:    s_lshr_b32 s4, s4, 24
; GFX9-DL-NEXT:    v_mad_u32_u24 v0, s9, v2, v0
; GFX9-DL-NEXT:    s_lshr_b32 s3, s3, 24
; GFX9-DL-NEXT:    v_mov_b32_e32 v2, s4
; GFX9-DL-NEXT:    v_mad_u32_u24 v0, s3, v2, v0
; GFX9-DL-NEXT:    v_add_u32_e32 v2, v0, v1
; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
; GFX9-DL-NEXT:    global_store_dword v[0:1], v2, off
; GFX9-DL-NEXT:    s_endpgm
                                               <4 x i8> addrspace(1)* %src2,
                                               i32 addrspace(1)* nocapture %dst) {
entry:
  %vec1 = load <4 x i8>, <4 x i8> addrspace(1)* %src1
  %vec2 = load <4 x i8>, <4 x i8> addrspace(1)* %src2

  %v1e0 = extractelement <4 x i8> %vec1, i64 0
  %cv1e0 = zext i8 %v1e0 to i32
  %v2e0 = extractelement <4 x i8> %vec2, i64 0
  %cv2e0 = zext i8 %v2e0 to i32
  %mul1 = mul nuw nsw i32 %cv1e0, %cv2e0

  %v1e1 = extractelement <4 x i8> %vec1, i64 1
  %cv1e1 = zext i8 %v1e1 to i32
  %v2e1 = extractelement <4 x i8> %vec2, i64 1
  %cv2e1 = zext i8 %v2e1 to i32
  %mul2 = mul nuw nsw i32 %cv1e1, %cv2e1

  %v1e2 = extractelement <4 x i8> %vec1, i64 2
  %cv1e2 = zext i8 %v1e2 to i32
  %v2e2 = extractelement <4 x i8> %vec2, i64 2
  %cv2e2 = zext i8 %v2e2 to i32
  %mul3 = mul nuw nsw i32 %cv1e2, %cv2e2

  %v1e3 = extractelement <4 x i8> %vec1, i64 3
  %cv1e3 = zext i8 %v1e3 to i32
  %v2e3 = extractelement <4 x i8> %vec2, i64 3
  %cv2e3 = zext i8 %v2e3 to i32
  %mul4 = mul nuw nsw i32 %cv1e3, %cv2e3

  %acc = load i32, i32 addrspace(1)* %dst, align 4
  %add1 = add i32 %mul2, %acc
  %add = add i32 %add1, %acc
  %add2 = add i32 %add1, %mul1
  %add3 = add i32 %add2, %mul3
  %add4 = add i32 %add3, %mul4
  %res = add i32 %add4, %add
  store i32 %res, i32 addrspace(1)* %dst, align 4
  ret void
}

define amdgpu_kernel void @notdot4_mixedtypes(<4 x i8> addrspace(1)* %src1,
; GFX7-LABEL: notdot4_mixedtypes:
; GFX7:       ; %bb.0: ; %entry
; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; GFX7-NEXT:    s_mov_b32 s3, 0xf000
; GFX7-NEXT:    s_mov_b32 s2, -1
; GFX7-NEXT:    s_mov_b32 s8, 0xffff
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    s_load_dword s4, s[4:5], 0x0
; GFX7-NEXT:    buffer_load_ushort v0, off, s[0:3], 0
; GFX7-NEXT:    s_load_dword s5, s[6:7], 0x0
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    s_sext_i32_i8 s6, s4
; GFX7-NEXT:    s_bfe_u32 s10, s4, 0x80008
; GFX7-NEXT:    s_sext_i32_i8 s7, s5
; GFX7-NEXT:    s_bfe_u32 s9, s5, 0x80008
; GFX7-NEXT:    s_and_b32 s7, s7, s8
; GFX7-NEXT:    v_mov_b32_e32 v1, s9
; GFX7-NEXT:    s_bfe_u32 s11, s5, 0x80010
; GFX7-NEXT:    s_and_b32 s6, s6, s8
; GFX7-NEXT:    v_mov_b32_e32 v3, s7
; GFX7-NEXT:    s_bfe_u32 s12, s4, 0x80010
; GFX7-NEXT:    s_lshr_b32 s5, s5, 24
; GFX7-NEXT:    v_mov_b32_e32 v2, s11
; GFX7-NEXT:    s_lshr_b32 s4, s4, 24
; GFX7-NEXT:    s_waitcnt vmcnt(0)
; GFX7-NEXT:    v_mad_u32_u24 v0, s10, v1, v0
; GFX7-NEXT:    v_mad_u32_u24 v0, s6, v3, v0
; GFX7-NEXT:    v_mad_u32_u24 v0, s12, v2, v0
; GFX7-NEXT:    v_mov_b32_e32 v1, s5
; GFX7-NEXT:    v_mad_u32_u24 v0, s4, v1, v0
; GFX7-NEXT:    buffer_store_short v0, off, s[0:3], 0
; GFX7-NEXT:    s_endpgm
;
; GFX8-LABEL: notdot4_mixedtypes:
; GFX8:       ; %bb.0: ; %entry
; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX8-NEXT:    s_mov_b32 s2, 0xffff
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_mov_b32_e32 v0, s0
; GFX8-NEXT:    v_mov_b32_e32 v1, s1
; GFX8-NEXT:    flat_load_ushort v2, v[0:1]
; GFX8-NEXT:    s_load_dword s0, s[4:5], 0x0
; GFX8-NEXT:    s_load_dword s1, s[6:7], 0x0
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    s_bfe_i32 s3, s0, 0x80000
; GFX8-NEXT:    s_bfe_u32 s6, s1, 0x80008
; GFX8-NEXT:    s_bfe_i32 s4, s1, 0x80000
; GFX8-NEXT:    s_and_b32 s3, s2, s3
; GFX8-NEXT:    s_and_b32 s2, s2, s4
; GFX8-NEXT:    s_bfe_u32 s5, s0, 0x80008
; GFX8-NEXT:    v_mov_b32_e32 v3, s6
; GFX8-NEXT:    s_bfe_u32 s8, s1, 0x80010
; GFX8-NEXT:    v_mov_b32_e32 v5, s2
; GFX8-NEXT:    s_bfe_u32 s7, s0, 0x80010
; GFX8-NEXT:    s_lshr_b32 s1, s1, 24
; GFX8-NEXT:    v_mov_b32_e32 v4, s8
; GFX8-NEXT:    s_lshr_b32 s0, s0, 24
; GFX8-NEXT:    s_waitcnt vmcnt(0)
; GFX8-NEXT:    v_mad_u32_u24 v2, s5, v3, v2
; GFX8-NEXT:    v_mad_u32_u24 v2, s3, v5, v2
; GFX8-NEXT:    v_mad_u32_u24 v2, s7, v4, v2
; GFX8-NEXT:    v_mov_b32_e32 v3, s1
; GFX8-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
; GFX8-NEXT:    flat_store_short v[0:1], v2
; GFX8-NEXT:    s_endpgm
;
; GFX9-NODL-LABEL: notdot4_mixedtypes:
; GFX9-NODL:       ; %bb.0: ; %entry
; GFX9-NODL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX9-NODL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX9-NODL-NEXT:    s_mov_b32 s2, 0xffff
; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s0
; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s1
; GFX9-NODL-NEXT:    global_load_ushort v2, v[0:1], off
; GFX9-NODL-NEXT:    s_load_dword s0, s[4:5], 0x0
; GFX9-NODL-NEXT:    s_load_dword s1, s[6:7], 0x0
; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NODL-NEXT:    s_bfe_i32 s3, s0, 0x80000
; GFX9-NODL-NEXT:    s_bfe_u32 s6, s1, 0x80008
; GFX9-NODL-NEXT:    s_bfe_i32 s4, s1, 0x80000
; GFX9-NODL-NEXT:    s_and_b32 s3, s2, s3
; GFX9-NODL-NEXT:    s_and_b32 s2, s2, s4
; GFX9-NODL-NEXT:    s_bfe_u32 s5, s0, 0x80008
; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s6
; GFX9-NODL-NEXT:    s_bfe_u32 s8, s1, 0x80010
; GFX9-NODL-NEXT:    v_mov_b32_e32 v5, s2
; GFX9-NODL-NEXT:    s_bfe_u32 s7, s0, 0x80010
; GFX9-NODL-NEXT:    s_lshr_b32 s1, s1, 24
; GFX9-NODL-NEXT:    v_mov_b32_e32 v4, s8
; GFX9-NODL-NEXT:    s_lshr_b32 s0, s0, 24
; GFX9-NODL-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s5, v3, v2
; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s3, v5, v2
; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s7, v4, v2
; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s1
; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
; GFX9-NODL-NEXT:    global_store_short v[0:1], v2, off
; GFX9-NODL-NEXT:    s_endpgm
;
; GFX9-DL-LABEL: notdot4_mixedtypes:
; GFX9-DL:       ; %bb.0: ; %entry
; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX9-DL-NEXT:    s_mov_b32 s2, 0xffff
; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
; GFX9-DL-NEXT:    global_load_ushort v2, v[0:1], off
; GFX9-DL-NEXT:    s_load_dword s0, s[4:5], 0x0
; GFX9-DL-NEXT:    s_load_dword s1, s[6:7], 0x0
; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-DL-NEXT:    s_bfe_i32 s3, s0, 0x80000
; GFX9-DL-NEXT:    s_bfe_u32 s6, s1, 0x80008
; GFX9-DL-NEXT:    s_bfe_i32 s4, s1, 0x80000
; GFX9-DL-NEXT:    s_and_b32 s3, s2, s3
; GFX9-DL-NEXT:    s_and_b32 s2, s2, s4
; GFX9-DL-NEXT:    s_bfe_u32 s5, s0, 0x80008
; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s6
; GFX9-DL-NEXT:    s_bfe_u32 s8, s1, 0x80010
; GFX9-DL-NEXT:    v_mov_b32_e32 v5, s2
; GFX9-DL-NEXT:    s_bfe_u32 s7, s0, 0x80010
; GFX9-DL-NEXT:    s_lshr_b32 s1, s1, 24
; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s8
; GFX9-DL-NEXT:    s_lshr_b32 s0, s0, 24
; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s5, v3, v2
; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s3, v5, v2
; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s7, v4, v2
; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s1
; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s0, v3, v2
; GFX9-DL-NEXT:    global_store_short v[0:1], v2, off
; GFX9-DL-NEXT:    s_endpgm
                                              <4 x i8> addrspace(1)* %src2,
                                              i16 addrspace(1)* nocapture %dst) {
entry:
  %vec1 = load <4 x i8>, <4 x i8> addrspace(1)* %src1
  %vec2 = load <4 x i8>, <4 x i8> addrspace(1)* %src2

  %v1e0 = extractelement <4 x i8> %vec1, i64 0
  %cv1e0 = sext i8 %v1e0 to i16
  %v2e0 = extractelement <4 x i8> %vec2, i64 0
  %cv2e0 = sext i8 %v2e0 to i16
  %mul1 = mul nuw nsw i16 %cv1e0, %cv2e0

  %v1e1 = extractelement <4 x i8> %vec1, i64 1
  %cv1e1 = zext i8 %v1e1 to i16
  %v2e1 = extractelement <4 x i8> %vec2, i64 1
  %cv2e1 = zext i8 %v2e1 to i16
  %mul2 = mul nuw nsw i16 %cv1e1, %cv2e1

  %v1e2 = extractelement <4 x i8> %vec1, i64 2
  %cv1e2 = zext i8 %v1e2 to i16
  %v2e2 = extractelement <4 x i8> %vec2, i64 2
  %cv2e2 = zext i8 %v2e2 to i16
  %mul3 = mul nuw nsw i16 %cv1e2, %cv2e2

  %v1e3 = extractelement <4 x i8> %vec1, i64 3
  %cv1e3 = zext i8 %v1e3 to i16
  %v2e3 = extractelement <4 x i8> %vec2, i64 3
  %cv2e3 = zext i8 %v2e3 to i16
  %mul4 = mul nuw nsw i16 %cv1e3, %cv2e3

  %acc = load i16, i16 addrspace(1)* %dst, align 2
  %add1 = add i16 %mul2, %acc
  %add2 = add i16 %add1, %mul1
  %add3 = add i16 %add2, %mul3
  %add4 = add i16 %add3, %mul4

  store i16 %add4, i16 addrspace(1)* %dst, align 2
  ret void
}

; TODO: cleanup s_lshr_b32 and support this pattern.
define amdgpu_kernel void @udot4_acc32_vecMul(<4 x i8> addrspace(1)* %src1,
; GFX7-LABEL: udot4_acc32_vecMul:
; GFX7:       ; %bb.0: ; %entry
; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; GFX7-NEXT:    s_movk_i32 s12, 0xff
; GFX7-NEXT:    s_mov_b32 s3, 0xf000
; GFX7-NEXT:    s_mov_b32 s2, -1
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    s_load_dword s4, s[4:5], 0x0
; GFX7-NEXT:    s_load_dword s5, s[6:7], 0x0
; GFX7-NEXT:    s_load_dword s6, s[0:1], 0x0
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    s_lshr_b32 s7, s4, 24
; GFX7-NEXT:    s_lshr_b32 s9, s5, 24
; GFX7-NEXT:    s_bfe_u32 s10, s5, 0x80008
; GFX7-NEXT:    s_bfe_u32 s13, s5, 0x80010
; GFX7-NEXT:    s_and_b32 s5, s5, s12
; GFX7-NEXT:    s_bfe_u32 s8, s4, 0x80008
; GFX7-NEXT:    s_bfe_u32 s11, s4, 0x80010
; GFX7-NEXT:    s_and_b32 s4, s4, s12
; GFX7-NEXT:    v_mov_b32_e32 v0, s5
; GFX7-NEXT:    v_mov_b32_e32 v1, s6
; GFX7-NEXT:    v_mad_u32_u24 v0, s4, v0, v1
; GFX7-NEXT:    v_mov_b32_e32 v1, s10
; GFX7-NEXT:    v_mad_u32_u24 v0, s8, v1, v0
; GFX7-NEXT:    v_mov_b32_e32 v1, s13
; GFX7-NEXT:    v_mad_u32_u24 v0, s11, v1, v0
; GFX7-NEXT:    v_mov_b32_e32 v1, s9
; GFX7-NEXT:    v_mad_u32_u24 v0, s7, v1, v0
; GFX7-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GFX7-NEXT:    s_endpgm
;
; GFX8-LABEL: udot4_acc32_vecMul:
; GFX8:       ; %bb.0: ; %entry
; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX8-NEXT:    s_movk_i32 s2, 0xff
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    s_load_dword s3, s[4:5], 0x0
; GFX8-NEXT:    s_load_dword s4, s[6:7], 0x0
; GFX8-NEXT:    s_load_dword s5, s[0:1], 0x0
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    s_lshr_b32 s6, s3, 24
; GFX8-NEXT:    s_bfe_u32 s8, s3, 0x80010
; GFX8-NEXT:    v_lshrrev_b16_e64 v0, 8, s3
; GFX8-NEXT:    s_and_b32 s3, s3, s2
; GFX8-NEXT:    s_and_b32 s2, s4, s2
; GFX8-NEXT:    v_mov_b32_e32 v2, s2
; GFX8-NEXT:    v_mov_b32_e32 v3, s5
; GFX8-NEXT:    s_bfe_u32 s9, s4, 0x80010
; GFX8-NEXT:    v_lshrrev_b16_e64 v1, 8, s4
; GFX8-NEXT:    v_mad_u32_u24 v2, s3, v2, v3
; GFX8-NEXT:    v_mad_u32_u24 v0, v0, v1, v2
; GFX8-NEXT:    v_mov_b32_e32 v1, s9
; GFX8-NEXT:    s_lshr_b32 s7, s4, 24
; GFX8-NEXT:    v_mad_u32_u24 v0, s8, v1, v0
; GFX8-NEXT:    v_mov_b32_e32 v1, s7
; GFX8-NEXT:    v_mad_u32_u24 v2, s6, v1, v0
; GFX8-NEXT:    v_mov_b32_e32 v0, s0
; GFX8-NEXT:    v_mov_b32_e32 v1, s1
; GFX8-NEXT:    flat_store_dword v[0:1], v2
; GFX8-NEXT:    s_endpgm
;
; GFX9-NODL-LABEL: udot4_acc32_vecMul:
; GFX9-NODL:       ; %bb.0: ; %entry
; GFX9-NODL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX9-NODL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX9-NODL-NEXT:    s_movk_i32 s2, 0xff
; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NODL-NEXT:    s_load_dword s3, s[4:5], 0x0
; GFX9-NODL-NEXT:    s_load_dword s4, s[6:7], 0x0
; GFX9-NODL-NEXT:    s_load_dword s5, s[0:1], 0x0
; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NODL-NEXT:    s_lshr_b32 s6, s3, 24
; GFX9-NODL-NEXT:    s_bfe_u32 s8, s3, 0x80010
; GFX9-NODL-NEXT:    v_lshrrev_b16_e64 v0, 8, s3
; GFX9-NODL-NEXT:    s_and_b32 s3, s3, s2
; GFX9-NODL-NEXT:    s_and_b32 s2, s4, s2
; GFX9-NODL-NEXT:    v_mov_b32_e32 v2, s2
; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s5
; GFX9-NODL-NEXT:    s_bfe_u32 s9, s4, 0x80010
; GFX9-NODL-NEXT:    v_lshrrev_b16_e64 v1, 8, s4
; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s3, v2, v3
; GFX9-NODL-NEXT:    v_mad_u32_u24 v0, v0, v1, v2
; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s9
; GFX9-NODL-NEXT:    s_lshr_b32 s7, s4, 24
; GFX9-NODL-NEXT:    v_mad_u32_u24 v0, s8, v1, v0
; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s7
; GFX9-NODL-NEXT:    v_mad_u32_u24 v2, s6, v1, v0
; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s0
; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s1
; GFX9-NODL-NEXT:    global_store_dword v[0:1], v2, off
; GFX9-NODL-NEXT:    s_endpgm
;
; GFX9-DL-LABEL: udot4_acc32_vecMul:
; GFX9-DL:       ; %bb.0: ; %entry
; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX9-DL-NEXT:    s_movk_i32 s2, 0xff
; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-DL-NEXT:    s_load_dword s3, s[4:5], 0x0
; GFX9-DL-NEXT:    s_load_dword s4, s[6:7], 0x0
; GFX9-DL-NEXT:    s_load_dword s5, s[0:1], 0x0
; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-DL-NEXT:    s_lshr_b32 s6, s3, 24
; GFX9-DL-NEXT:    s_bfe_u32 s8, s3, 0x80010
; GFX9-DL-NEXT:    v_lshrrev_b16_e64 v0, 8, s3
; GFX9-DL-NEXT:    s_and_b32 s3, s3, s2
; GFX9-DL-NEXT:    s_and_b32 s2, s4, s2
; GFX9-DL-NEXT:    v_mov_b32_e32 v2, s2
; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s5
; GFX9-DL-NEXT:    s_bfe_u32 s9, s4, 0x80010
; GFX9-DL-NEXT:    v_lshrrev_b16_e64 v1, 8, s4
; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s3, v2, v3
; GFX9-DL-NEXT:    v_mad_u32_u24 v0, v0, v1, v2
; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s9
; GFX9-DL-NEXT:    s_lshr_b32 s7, s4, 24
; GFX9-DL-NEXT:    v_mad_u32_u24 v0, s8, v1, v0
; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s7
; GFX9-DL-NEXT:    v_mad_u32_u24 v2, s6, v1, v0
; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
; GFX9-DL-NEXT:    global_store_dword v[0:1], v2, off
; GFX9-DL-NEXT:    s_endpgm
                                              <4 x i8> addrspace(1)* %src2,
                                              i32 addrspace(1)* nocapture %dst) {
entry:
  %vec1 = load <4 x i8>, <4 x i8> addrspace(1)* %src1
  %vec2 = load <4 x i8>, <4 x i8> addrspace(1)* %src2

  %cvec1 = zext <4 x i8> %vec1 to <4 x i32>
  %cvec2 = zext <4 x i8> %vec2 to <4 x i32>

  %mul = mul <4 x i32> %cvec1, %cvec2
  %mul0 = extractelement <4 x i32> %mul, i64 0
  %mul1 = extractelement <4 x i32> %mul, i64 1
  %mul2 = extractelement <4 x i32> %mul, i64 2
  %mul3 = extractelement <4 x i32> %mul, i64 3

  %acc = load i32, i32 addrspace(1)* %dst, align 4
  %add1 = add i32 %mul0, %acc
  %add2 = add i32 %add1, %mul1
  %add3 = add i32 %add2, %mul2
  %add4 = add i32 %add3, %mul3

  store i32 %add4, i32 addrspace(1)* %dst, align 4
  ret void
}

; TODO: This pattern should be recognized.
define amdgpu_kernel void @udot4_acc16_vecMul(<4 x i8> addrspace(1)* %src1,
; GFX7-LABEL: udot4_acc16_vecMul:
; GFX7:       ; %bb.0: ; %entry
; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; GFX7-NEXT:    s_mov_b32 s3, 0xf000
; GFX7-NEXT:    s_mov_b32 s2, -1
; GFX7-NEXT:    s_movk_i32 s8, 0xff
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    s_load_dword s4, s[4:5], 0x0
; GFX7-NEXT:    buffer_load_ushort v0, off, s[0:3], 0
; GFX7-NEXT:    s_load_dword s5, s[6:7], 0x0
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    s_and_b32 s11, s4, s8
; GFX7-NEXT:    s_bfe_u32 s6, s4, 0x80008
; GFX7-NEXT:    s_bfe_u32 s9, s5, 0x80008
; GFX7-NEXT:    s_lshr_b32 s10, s5, 24
; GFX7-NEXT:    s_and_b32 s8, s5, s8
; GFX7-NEXT:    v_mov_b32_e32 v4, s9
; GFX7-NEXT:    s_lshr_b32 s7, s4, 24
; GFX7-NEXT:    v_mov_b32_e32 v2, s10
; GFX7-NEXT:    s_bfe_u32 s5, s5, 0x80010
; GFX7-NEXT:    v_mov_b32_e32 v3, s8
; GFX7-NEXT:    v_mul_u32_u24_e32 v2, s7, v2
; GFX7-NEXT:    v_mul_u32_u24_e32 v4, s6, v4
; GFX7-NEXT:    s_bfe_u32 s4, s4, 0x80010
; GFX7-NEXT:    v_mov_b32_e32 v1, s5
; GFX7-NEXT:    v_mul_u32_u24_e32 v1, s4, v1
; GFX7-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
; GFX7-NEXT:    v_mul_u32_u24_e32 v3, s11, v3
; GFX7-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
; GFX7-NEXT:    v_or_b32_e32 v1, v1, v2
; GFX7-NEXT:    v_or_b32_e32 v2, v3, v4
; GFX7-NEXT:    v_lshrrev_b32_e32 v3, 16, v2
; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 16, v1
; GFX7-NEXT:    s_waitcnt vmcnt(0)
; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v3, v0
; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v4, v0
; GFX7-NEXT:    buffer_store_short v0, off, s[0:3], 0
; GFX7-NEXT:    s_endpgm
;
; GFX8-LABEL: udot4_acc16_vecMul:
; GFX8:       ; %bb.0: ; %entry
; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_mov_b32_e32 v0, s0
; GFX8-NEXT:    v_mov_b32_e32 v1, s1
; GFX8-NEXT:    flat_load_ushort v2, v[0:1]
; GFX8-NEXT:    s_load_dword s1, s[4:5], 0x0
; GFX8-NEXT:    s_load_dword s2, s[6:7], 0x0
; GFX8-NEXT:    s_movk_i32 s0, 0xff
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    s_and_b32 s6, s1, s0
; GFX8-NEXT:    s_and_b32 s0, s2, s0
; GFX8-NEXT:    v_mov_b32_e32 v5, s0
; GFX8-NEXT:    s_bfe_u32 s7, s2, 0x80010
; GFX8-NEXT:    v_lshrrev_b16_e64 v3, 8, s2
; GFX8-NEXT:    v_lshrrev_b16_e64 v4, 8, s1
; GFX8-NEXT:    s_lshr_b32 s4, s2, 24
; GFX8-NEXT:    s_bfe_u32 s5, s1, 0x80010
; GFX8-NEXT:    v_mov_b32_e32 v6, s7
; GFX8-NEXT:    s_lshr_b32 s3, s1, 24
; GFX8-NEXT:    s_waitcnt vmcnt(0)
; GFX8-NEXT:    v_mad_u32_u24 v2, s6, v5, v2
; GFX8-NEXT:    v_mad_u32_u24 v2, v4, v3, v2
; GFX8-NEXT:    v_mad_u32_u24 v2, s5, v6, v2
; GFX8-NEXT:    v_mov_b32_e32 v3, s4
; GFX8-NEXT:    v_mad_u32_u24 v2, s3, v3, v2
; GFX8-NEXT:    flat_store_short v[0:1], v2
; GFX8-NEXT:    s_endpgm
;
; GFX9-NODL-LABEL: udot4_acc16_vecMul:
; GFX9-NODL:       ; %bb.0: ; %entry
; GFX9-NODL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX9-NODL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, 0xffff
; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NODL-NEXT:    s_load_dword s2, s[4:5], 0x0
; GFX9-NODL-NEXT:    s_load_dword s3, s[6:7], 0x0
; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NODL-NEXT:    s_lshr_b32 s5, s2, 16
; GFX9-NODL-NEXT:    s_lshr_b32 s7, s3, 16
; GFX9-NODL-NEXT:    s_lshr_b32 s4, s2, 24
; GFX9-NODL-NEXT:    v_and_b32_sdwa v4, v0, s5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
; GFX9-NODL-NEXT:    s_lshr_b32 s6, s3, 24
; GFX9-NODL-NEXT:    v_and_b32_sdwa v3, v0, s7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
; GFX9-NODL-NEXT:    v_lshl_or_b32 v3, s6, 16, v3
; GFX9-NODL-NEXT:    v_lshl_or_b32 v4, s4, 16, v4
; GFX9-NODL-NEXT:    v_pk_mul_lo_u16 v3, v4, v3
; GFX9-NODL-NEXT:    v_and_b32_sdwa v4, v0, s3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
; GFX9-NODL-NEXT:    v_lshrrev_b16_e64 v2, 8, s3
; GFX9-NODL-NEXT:    v_lshrrev_b16_e64 v1, 8, s2
; GFX9-NODL-NEXT:    v_and_b32_sdwa v0, v0, s2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
; GFX9-NODL-NEXT:    v_lshl_or_b32 v2, v2, 16, v4
; GFX9-NODL-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
; GFX9-NODL-NEXT:    v_pk_mul_lo_u16 v2, v0, v2
; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s0
; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s1
; GFX9-NODL-NEXT:    global_load_ushort v4, v[0:1], off
; GFX9-NODL-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NODL-NEXT:    v_add_u32_e32 v4, v2, v4
; GFX9-NODL-NEXT:    v_add_u32_sdwa v2, v4, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
; GFX9-NODL-NEXT:    v_add_u32_e32 v2, v2, v3
; GFX9-NODL-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
; GFX9-NODL-NEXT:    global_store_short v[0:1], v2, off
; GFX9-NODL-NEXT:    s_endpgm
;
; GFX9-DL-LABEL: udot4_acc16_vecMul:
; GFX9-DL:       ; %bb.0: ; %entry
; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX9-DL-NEXT:    v_mov_b32_e32 v0, 0xffff
; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
; GFX9-DL-NEXT:    s_load_dword s3, s[6:7], 0x0
; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-DL-NEXT:    s_lshr_b32 s5, s2, 16
; GFX9-DL-NEXT:    s_lshr_b32 s7, s3, 16
; GFX9-DL-NEXT:    s_lshr_b32 s4, s2, 24
; GFX9-DL-NEXT:    v_and_b32_sdwa v4, v0, s5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
; GFX9-DL-NEXT:    s_lshr_b32 s6, s3, 24
; GFX9-DL-NEXT:    v_and_b32_sdwa v3, v0, s7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
; GFX9-DL-NEXT:    v_lshl_or_b32 v3, s6, 16, v3
; GFX9-DL-NEXT:    v_lshl_or_b32 v4, s4, 16, v4
; GFX9-DL-NEXT:    v_pk_mul_lo_u16 v3, v4, v3
; GFX9-DL-NEXT:    v_and_b32_sdwa v4, v0, s3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
; GFX9-DL-NEXT:    v_lshrrev_b16_e64 v2, 8, s3
; GFX9-DL-NEXT:    v_lshrrev_b16_e64 v1, 8, s2
; GFX9-DL-NEXT:    v_and_b32_sdwa v0, v0, s2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
; GFX9-DL-NEXT:    v_lshl_or_b32 v2, v2, 16, v4
; GFX9-DL-NEXT:    v_lshl_or_b32 v0, v1, 16, v0
; GFX9-DL-NEXT:    v_pk_mul_lo_u16 v2, v0, v2
; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
; GFX9-DL-NEXT:    global_load_ushort v4, v[0:1], off
; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
; GFX9-DL-NEXT:    v_add_u32_e32 v4, v2, v4
; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v4, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
; GFX9-DL-NEXT:    v_add_u32_e32 v2, v2, v3
; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
; GFX9-DL-NEXT:    global_store_short v[0:1], v2, off
; GFX9-DL-NEXT:    s_endpgm
                                              <4 x i8> addrspace(1)* %src2,
                                              i16 addrspace(1)* nocapture %dst) {
entry:
  %vec1 = load <4 x i8>, <4 x i8> addrspace(1)* %src1
  %vec2 = load <4 x i8>, <4 x i8> addrspace(1)* %src2

  %cvec1 = zext <4 x i8> %vec1 to <4 x i16>
  %cvec2 = zext <4 x i8> %vec2 to <4 x i16>

  %mul = mul <4 x i16> %cvec1, %cvec2
  %mul0 = extractelement <4 x i16> %mul, i64 0
  %mul1 = extractelement <4 x i16> %mul, i64 1
  %mul2 = extractelement <4 x i16> %mul, i64 2
  %mul3 = extractelement <4 x i16> %mul, i64 3

  %acc = load i16, i16 addrspace(1)* %dst, align 4
  %add1 = add i16 %mul0, %acc
  %add2 = add i16 %add1, %mul1
  %add3 = add i16 %add2, %mul2
  %add4 = add i16 %add3, %mul3

  store i16 %add4, i16 addrspace(1)* %dst, align 4
  ret void
}

; TODO: Support this pattern.
define amdgpu_kernel void @udot4_acc8_vecMul(<4 x i8> addrspace(1)* %src1,
; GFX7-LABEL: udot4_acc8_vecMul:
; GFX7:       ; %bb.0: ; %entry
; GFX7-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
; GFX7-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; GFX7-NEXT:    s_mov_b32 s3, 0xf000
; GFX7-NEXT:    s_mov_b32 s2, -1
; GFX7-NEXT:    s_movk_i32 s8, 0xff
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    s_load_dword s4, s[4:5], 0x0
; GFX7-NEXT:    buffer_load_ubyte v0, off, s[0:3], 0
; GFX7-NEXT:    s_load_dword s5, s[6:7], 0x0
; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
; GFX7-NEXT:    s_bfe_u32 s6, s4, 0x80008
; GFX7-NEXT:    s_lshr_b32 s7, s4, 16
; GFX7-NEXT:    s_bfe_u32 s10, s5, 0x80008
; GFX7-NEXT:    s_lshr_b32 s11, s5, 16
; GFX7-NEXT:    s_lshr_b32 s12, s5, 24
; GFX7-NEXT:    v_mov_b32_e32 v2, s11
; GFX7-NEXT:    v_mov_b32_e32 v3, s10
; GFX7-NEXT:    s_lshr_b32 s9, s4, 24
; GFX7-NEXT:    v_mov_b32_e32 v1, s12
; GFX7-NEXT:    s_mul_i32 s4, s4, s5
; GFX7-NEXT:    v_mul_u32_u24_e32 v1, s9, v1
; GFX7-NEXT:    v_mul_u32_u24_e32 v2, s7, v2
; GFX7-NEXT:    v_mul_u32_u24_e32 v3, s6, v3
; GFX7-NEXT:    s_and_b32 s4, s4, s8
; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
; GFX7-NEXT:    v_and_b32_e32 v2, s8, v2
; GFX7-NEXT:    v_lshlrev_b32_e32 v3, 8, v3
; GFX7-NEXT:    v_or_b32_e32 v1, v2, v1
; GFX7-NEXT:    v_or_b32_e32 v2, s4, v3
; GFX7-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
; GFX7-NEXT:    v_and_b32_e32 v2, 0xffff, v2
; GFX7-NEXT:    v_or_b32_e32 v1, v2, v1
; GFX7-NEXT:    v_lshrrev_b32_e32 v2, 8, v1
; GFX7-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
; GFX7-NEXT:    v_lshrrev_b32_e32 v4, 24, v1
; GFX7-NEXT:    s_waitcnt vmcnt(0)
; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v3, v0
; GFX7-NEXT:    v_add_i32_e32 v0, vcc, v4, v0
; GFX7-NEXT:    buffer_store_byte v0, off, s[0:3], 0
; GFX7-NEXT:    s_endpgm
;
; GFX8-LABEL: udot4_acc8_vecMul:
; GFX8:       ; %bb.0: ; %entry
; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    v_mov_b32_e32 v0, s0
; GFX8-NEXT:    v_mov_b32_e32 v1, s1
; GFX8-NEXT:    flat_load_ubyte v2, v[0:1]
; GFX8-NEXT:    s_load_dword s1, s[4:5], 0x0
; GFX8-NEXT:    s_load_dword s2, s[6:7], 0x0
; GFX8-NEXT:    s_movk_i32 s0, 0xff
; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
; GFX8-NEXT:    s_lshr_b32 s3, s1, 24
; GFX8-NEXT:    s_lshr_b32 s4, s2, 24
; GFX8-NEXT:    s_and_b32 s6, s1, s0
; GFX8-NEXT:    s_bfe_u32 s7, s2, 0x80010
; GFX8-NEXT:    s_and_b32 s0, s2, s0
; GFX8-NEXT:    v_mov_b32_e32 v3, s1
; GFX8-NEXT:    v_mov_b32_e32 v4, s2
; GFX8-NEXT:    v_mul_u32_u24_sdwa v3, v3, v4 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:BYTE_1
; GFX8-NEXT:    v_mov_b32_e32 v4, s0
; GFX8-NEXT:    s_bfe_u32 s5, s1, 0x80010
; GFX8-NEXT:    v_mov_b32_e32 v5, s7
; GFX8-NEXT:    v_mov_b32_e32 v6, s4
; GFX8-NEXT:    v_mov_b32_e32 v7, s3
; GFX8-NEXT:    v_mul_u32_u24_e32 v4, s6, v4
; GFX8-NEXT:    v_mul_u32_u24_e32 v5, s5, v5
; GFX8-NEXT:    v_mul_u32_u24_sdwa v6, v7, v6 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; GFX8-NEXT:    v_or_b32_sdwa v5, v5, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
; GFX8-NEXT:    v_or_b32_sdwa v3, v4, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
; GFX8-NEXT:    v_or_b32_sdwa v3, v3, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
; GFX8-NEXT:    v_lshrrev_b32_e32 v4, 8, v3
; GFX8-NEXT:    s_waitcnt vmcnt(0)
; GFX8-NEXT:    v_add_u32_e32 v2, vcc, v2, v3
; GFX8-NEXT:    v_add_u32_e32 v2, vcc, v4, v2
; GFX8-NEXT:    v_add_u32_sdwa v2, vcc, v3, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
; GFX8-NEXT:    v_add_u32_sdwa v2, vcc, v3, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:DWORD
; GFX8-NEXT:    flat_store_byte v[0:1], v2
; GFX8-NEXT:    s_endpgm
;
; GFX9-NODL-LABEL: udot4_acc8_vecMul:
; GFX9-NODL:       ; %bb.0: ; %entry
; GFX9-NODL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX9-NODL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NODL-NEXT:    s_load_dword s2, s[4:5], 0x0
; GFX9-NODL-NEXT:    s_load_dword s3, s[6:7], 0x0
; GFX9-NODL-NEXT:    v_mov_b32_e32 v0, s0
; GFX9-NODL-NEXT:    v_mov_b32_e32 v1, s1
; GFX9-NODL-NEXT:    global_load_ubyte v2, v[0:1], off
; GFX9-NODL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NODL-NEXT:    s_lshr_b32 s0, s2, 16
; GFX9-NODL-NEXT:    v_mov_b32_e32 v3, s3
; GFX9-NODL-NEXT:    v_mov_b32_e32 v4, s3
; GFX9-NODL-NEXT:    s_lshr_b32 s1, s3, 16
; GFX9-NODL-NEXT:    s_lshr_b32 s4, s3, 24
; GFX9-NODL-NEXT:    v_mul_lo_u16_e32 v3, s2, v3
; GFX9-NODL-NEXT:    v_mul_lo_u16_sdwa v4, s2, v4 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:BYTE_1
; GFX9-NODL-NEXT:    v_mov_b32_e32 v5, s1
; GFX9-NODL-NEXT:    v_or_b32_sdwa v3, v3, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
; GFX9-NODL-NEXT:    s_lshr_b32 s5, s2, 24
; GFX9-NODL-NEXT:    v_mov_b32_e32 v4, s4
; GFX9-NODL-NEXT:    v_mul_lo_u16_sdwa v4, s5, v4 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; GFX9-NODL-NEXT:    v_mul_lo_u16_e32 v5, s0, v5
; GFX9-NODL-NEXT:    v_or_b32_sdwa v4, v5, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
; GFX9-NODL-NEXT:    v_or_b32_sdwa v3, v3, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
; GFX9-NODL-NEXT:    v_lshrrev_b32_e32 v4, 8, v3
; GFX9-NODL-NEXT:    s_waitcnt vmcnt(0)
; GFX9-NODL-NEXT:    v_add_u32_e32 v2, v3, v2
; GFX9-NODL-NEXT:    v_add_u32_e32 v2, v2, v4
; GFX9-NODL-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
; GFX9-NODL-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
; GFX9-NODL-NEXT:    global_store_byte v[0:1], v2, off
; GFX9-NODL-NEXT:    s_endpgm
;
; GFX9-DL-LABEL: udot4_acc8_vecMul:
; GFX9-DL:       ; %bb.0: ; %entry
; GFX9-DL-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
; GFX9-DL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-DL-NEXT:    s_load_dword s2, s[4:5], 0x0
; GFX9-DL-NEXT:    s_load_dword s3, s[6:7], 0x0
; GFX9-DL-NEXT:    v_mov_b32_e32 v0, s0
; GFX9-DL-NEXT:    v_mov_b32_e32 v1, s1
; GFX9-DL-NEXT:    global_load_ubyte v2, v[0:1], off
; GFX9-DL-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-DL-NEXT:    s_lshr_b32 s0, s2, 16
; GFX9-DL-NEXT:    v_mov_b32_e32 v3, s3
; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s3
; GFX9-DL-NEXT:    s_lshr_b32 s1, s3, 16
; GFX9-DL-NEXT:    s_lshr_b32 s4, s3, 24
; GFX9-DL-NEXT:    v_mul_lo_u16_e32 v3, s2, v3
; GFX9-DL-NEXT:    v_mul_lo_u16_sdwa v4, s2, v4 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:BYTE_1
; GFX9-DL-NEXT:    v_mov_b32_e32 v5, s1
; GFX9-DL-NEXT:    v_or_b32_sdwa v3, v3, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
; GFX9-DL-NEXT:    s_lshr_b32 s5, s2, 24
; GFX9-DL-NEXT:    v_mov_b32_e32 v4, s4
; GFX9-DL-NEXT:    v_mul_lo_u16_sdwa v4, s5, v4 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; GFX9-DL-NEXT:    v_mul_lo_u16_e32 v5, s0, v5
; GFX9-DL-NEXT:    v_or_b32_sdwa v4, v5, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
; GFX9-DL-NEXT:    v_or_b32_sdwa v3, v3, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
; GFX9-DL-NEXT:    v_lshrrev_b32_e32 v4, 8, v3
; GFX9-DL-NEXT:    s_waitcnt vmcnt(0)
; GFX9-DL-NEXT:    v_add_u32_e32 v2, v3, v2
; GFX9-DL-NEXT:    v_add_u32_e32 v2, v2, v4
; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
; GFX9-DL-NEXT:    v_add_u32_sdwa v2, v2, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_3
; GFX9-DL-NEXT:    global_store_byte v[0:1], v2, off
; GFX9-DL-NEXT:    s_endpgm
                                             <4 x i8> addrspace(1)* %src2,
                                             i8 addrspace(1)* nocapture %dst) {
entry:
  %vec1 = load <4 x i8>, <4 x i8> addrspace(1)* %src1
  %vec2 = load <4 x i8>, <4 x i8> addrspace(1)* %src2

  %mul = mul <4 x i8> %vec1, %vec2
  %mul0 = extractelement <4 x i8> %mul, i64 0
  %mul1 = extractelement <4 x i8> %mul, i64 1
  %mul2 = extractelement <4 x i8> %mul, i64 2
  %mul3 = extractelement <4 x i8> %mul, i64 3

  %acc = load i8, i8 addrspace(1)* %dst, align 4
  %add1 = add i8 %mul0, %acc
  %add2 = add i8 %add1, %mul1
  %add3 = add i8 %add2, %mul2
  %add4 = add i8 %add3, %mul3

  store i8 %add4, i8 addrspace(1)* %dst, align 4
  ret void
}