; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefixes=SI %s
; RUN: llc -mtriple=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefixes=VI %s
; RUN: llc -mtriple=amdgcn -mcpu=gfx900 -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9 %s
; RUN: llc -mtriple=r600 -mcpu=redwood < %s | FileCheck -check-prefix=EG %s
; RUN: llc -mtriple=r600 -mcpu=cayman < %s | FileCheck -check-prefix=CM %s

; Signed 24-bit multiply is not supported on pre-Cayman GPUs.
define amdgpu_kernel void @test_smul24_i32(ptr addrspace(1) %out, i32 %a, i32 %b) #0 {
; SI-LABEL: test_smul24_i32:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x9
; SI-NEXT:    s_mov_b32 s7, 0xf000
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_bfe_i32 s2, s2, 0x180000
; SI-NEXT:    s_bfe_i32 s3, s3, 0x180000
; SI-NEXT:    s_mul_i32 s2, s2, s3
; SI-NEXT:    s_mov_b32 s6, -1
; SI-NEXT:    s_mov_b32 s4, s0
; SI-NEXT:    s_mov_b32 s5, s1
; SI-NEXT:    v_mov_b32_e32 v0, s2
; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: test_smul24_i32:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; VI-NEXT:    s_mov_b32 s7, 0xf000
; VI-NEXT:    s_mov_b32 s6, -1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s4, s0
; VI-NEXT:    s_mov_b32 s5, s1
; VI-NEXT:    s_bfe_i32 s0, s2, 0x180000
; VI-NEXT:    s_bfe_i32 s1, s3, 0x180000
; VI-NEXT:    s_mul_i32 s0, s0, s1
; VI-NEXT:    v_mov_b32_e32 v0, s0
; VI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: test_smul24_i32:
; GFX9:       ; %bb.0: ; %entry
; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX9-NEXT:    s_mov_b32 s7, 0xf000
; GFX9-NEXT:    s_mov_b32 s6, -1
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    s_mov_b32 s4, s0
; GFX9-NEXT:    s_mov_b32 s5, s1
; GFX9-NEXT:    s_bfe_i32 s0, s2, 0x180000
; GFX9-NEXT:    s_bfe_i32 s1, s3, 0x180000
; GFX9-NEXT:    s_mul_i32 s0, s0, s1
; GFX9-NEXT:    v_mov_b32_e32 v0, s0
; GFX9-NEXT:    buffer_store_dword v0, off, s[4:7], 0
; GFX9-NEXT:    s_endpgm
;
; EG-LABEL: test_smul24_i32:
; EG:       ; %bb.0: ; %entry
; EG-NEXT:    ALU 8, @4, KC0[CB0:0-32], KC1[]
; EG-NEXT:    MEM_RAT_CACHELESS STORE_RAW T1.X, T0.X, 1
; EG-NEXT:    CF_END
; EG-NEXT:    PAD
; EG-NEXT:    ALU clause starting at 4:
; EG-NEXT:     LSHL T0.W, KC0[2].Z, literal.x,
; EG-NEXT:     LSHL * T1.W, KC0[2].W, literal.x,
; EG-NEXT:    8(1.121039e-44), 0(0.000000e+00)
; EG-NEXT:     ASHR T1.W, PS, literal.x,
; EG-NEXT:     ASHR * T0.W, PV.W, literal.x,
; EG-NEXT:    8(1.121039e-44), 0(0.000000e+00)
; EG-NEXT:     LSHR T0.X, KC0[2].Y, literal.x,
; EG-NEXT:     MULLO_INT * T1.X, PS, PV.W,
; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
;
; CM-LABEL: test_smul24_i32:
; CM:       ; %bb.0: ; %entry
; CM-NEXT:    ALU 10, @4, KC0[CB0:0-32], KC1[]
; CM-NEXT:    MEM_RAT_CACHELESS STORE_DWORD T1.X, T0.X
; CM-NEXT:    CF_END
; CM-NEXT:    PAD
; CM-NEXT:    ALU clause starting at 4:
; CM-NEXT:     LSHL T0.Z, KC0[2].Z, literal.x,
; CM-NEXT:     LSHL * T0.W, KC0[2].W, literal.x,
; CM-NEXT:    8(1.121039e-44), 0(0.000000e+00)
; CM-NEXT:     LSHR T0.X, KC0[2].Y, literal.x,
; CM-NEXT:     ASHR T1.Z, PV.W, literal.y,
; CM-NEXT:     ASHR * T0.W, PV.Z, literal.y,
; CM-NEXT:    2(2.802597e-45), 8(1.121039e-44)
; CM-NEXT:     MULLO_INT T1.X, T0.W, T1.Z,
; CM-NEXT:     MULLO_INT T1.Y (MASKED), T0.W, T1.Z,
; CM-NEXT:     MULLO_INT T1.Z (MASKED), T0.W, T1.Z,
; CM-NEXT:     MULLO_INT * T1.W (MASKED), T0.W, T1.Z,
entry:
  %a.shl = shl i32 %a, 8
  %a.24 = ashr i32 %a.shl, 8
  %b.shl = shl i32 %b, 8
  %b.24 = ashr i32 %b.shl, 8
  %mul24 = mul i32 %a.24, %b.24
  store i32 %mul24, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @test_smulhi24_i64(ptr addrspace(1) %out, i32 %a, i32 %b) #0 {
; SI-LABEL: test_smulhi24_i64:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x9
; SI-NEXT:    s_mov_b32 s7, 0xf000
; SI-NEXT:    s_mov_b32 s6, -1
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s4, s0
; SI-NEXT:    s_mov_b32 s5, s1
; SI-NEXT:    v_mov_b32_e32 v0, s3
; SI-NEXT:    v_mul_hi_i32_i24_e32 v0, s2, v0
; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: test_smulhi24_i64:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; VI-NEXT:    s_mov_b32 s7, 0xf000
; VI-NEXT:    s_mov_b32 s6, -1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    v_mov_b32_e32 v0, s3
; VI-NEXT:    s_mov_b32 s4, s0
; VI-NEXT:    s_mov_b32 s5, s1
; VI-NEXT:    v_mul_hi_i32_i24_e32 v0, s2, v0
; VI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: test_smulhi24_i64:
; GFX9:       ; %bb.0: ; %entry
; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; GFX9-NEXT:    s_mov_b32 s7, 0xf000
; GFX9-NEXT:    s_mov_b32 s6, -1
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    s_mov_b32 s4, s0
; GFX9-NEXT:    s_mov_b32 s5, s1
; GFX9-NEXT:    s_bfe_i32 s0, s2, 0x180000
; GFX9-NEXT:    s_bfe_i32 s1, s3, 0x180000
; GFX9-NEXT:    s_mul_hi_i32 s0, s0, s1
; GFX9-NEXT:    v_mov_b32_e32 v0, s0
; GFX9-NEXT:    buffer_store_dword v0, off, s[4:7], 0
; GFX9-NEXT:    s_endpgm
;
; EG-LABEL: test_smulhi24_i64:
; EG:       ; %bb.0: ; %entry
; EG-NEXT:    ALU 8, @4, KC0[CB0:0-32], KC1[]
; EG-NEXT:    MEM_RAT_CACHELESS STORE_RAW T1.X, T0.X, 1
; EG-NEXT:    CF_END
; EG-NEXT:    PAD
; EG-NEXT:    ALU clause starting at 4:
; EG-NEXT:     LSHL T0.W, KC0[2].Z, literal.x,
; EG-NEXT:     LSHL * T1.W, KC0[2].W, literal.x,
; EG-NEXT:    8(1.121039e-44), 0(0.000000e+00)
; EG-NEXT:     ASHR T1.W, PS, literal.x,
; EG-NEXT:     ASHR * T0.W, PV.W, literal.x,
; EG-NEXT:    8(1.121039e-44), 0(0.000000e+00)
; EG-NEXT:     LSHR T0.X, KC0[2].Y, literal.x,
; EG-NEXT:     MULHI_INT * T1.X, PS, PV.W,
; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
;
; CM-LABEL: test_smulhi24_i64:
; CM:       ; %bb.0: ; %entry
; CM-NEXT:    ALU 5, @4, KC0[CB0:0-32], KC1[]
; CM-NEXT:    MEM_RAT_CACHELESS STORE_DWORD T1.X, T0.X
; CM-NEXT:    CF_END
; CM-NEXT:    PAD
; CM-NEXT:    ALU clause starting at 4:
; CM-NEXT:     LSHR * T0.X, KC0[2].Y, literal.x,
; CM-NEXT:    2(2.802597e-45), 0(0.000000e+00)
; CM-NEXT:     MULHI_INT24 T1.X, KC0[2].Z, KC0[2].W,
; CM-NEXT:     MULHI_INT24 T1.Y (MASKED), KC0[2].Z, KC0[2].W,
; CM-NEXT:     MULHI_INT24 T1.Z (MASKED), KC0[2].Z, KC0[2].W,
; CM-NEXT:     MULHI_INT24 * T1.W (MASKED), KC0[2].Z, KC0[2].W,
entry:
  %a.shl = shl i32 %a, 8
  %a.24 = ashr i32 %a.shl, 8
  %b.shl = shl i32 %b, 8
  %b.24 = ashr i32 %b.shl, 8
  %a.24.i64 = sext i32 %a.24 to i64
  %b.24.i64 = sext i32 %b.24 to i64
  %mul48 = mul i64 %a.24.i64, %b.24.i64
  %mul48.hi = lshr i64 %mul48, 32
  %mul24hi = trunc i64 %mul48.hi to i32
  store i32 %mul24hi, ptr addrspace(1) %out
  ret void
}

define i64 @test_smul48_i64(i64 %lhs, i64 %rhs) {
; SI-LABEL: test_smul48_i64:
; SI:       ; %bb.0:
; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT:    v_mul_hi_i32_i24_e32 v1, v0, v2
; SI-NEXT:    v_mul_i32_i24_e32 v0, v0, v2
; SI-NEXT:    s_setpc_b64 s[30:31]
;
; VI-LABEL: test_smul48_i64:
; VI:       ; %bb.0:
; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT:    v_mul_hi_i32_i24_e32 v1, v0, v2
; VI-NEXT:    v_mul_i32_i24_e32 v0, v0, v2
; VI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX9-LABEL: test_smul48_i64:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT:    v_mul_hi_i32_i24_e32 v1, v0, v2
; GFX9-NEXT:    v_mul_i32_i24_e32 v0, v0, v2
; GFX9-NEXT:    s_setpc_b64 s[30:31]
;
; EG-LABEL: test_smul48_i64:
; EG:       ; %bb.0:
; EG-NEXT:    CF_END
; EG-NEXT:    PAD
;
; CM-LABEL: test_smul48_i64:
; CM:       ; %bb.0:
; CM-NEXT:    CF_END
; CM-NEXT:    PAD
  %shl.lhs = shl i64 %lhs, 40
  %lhs24 = ashr i64 %shl.lhs, 40
  %shl.rhs = shl i64 %rhs, 40
  %rhs24 = ashr i64 %shl.rhs, 40
  %mul = mul i64 %lhs24, %rhs24
  ret i64 %mul
}

define <2 x i64> @test_smul48_v2i64(<2 x i64> %lhs, <2 x i64> %rhs) {
; SI-LABEL: test_smul48_v2i64:
; SI:       ; %bb.0:
; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; SI-NEXT:    v_mul_hi_i32_i24_e32 v1, v0, v4
; SI-NEXT:    v_mul_i32_i24_e32 v0, v0, v4
; SI-NEXT:    v_mul_hi_i32_i24_e32 v3, v2, v6
; SI-NEXT:    v_mul_i32_i24_e32 v2, v2, v6
; SI-NEXT:    s_setpc_b64 s[30:31]
;
; VI-LABEL: test_smul48_v2i64:
; VI:       ; %bb.0:
; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT:    v_mul_hi_i32_i24_e32 v1, v0, v4
; VI-NEXT:    v_mul_i32_i24_e32 v0, v0, v4
; VI-NEXT:    v_mul_hi_i32_i24_e32 v3, v2, v6
; VI-NEXT:    v_mul_i32_i24_e32 v2, v2, v6
; VI-NEXT:    s_setpc_b64 s[30:31]
;
; GFX9-LABEL: test_smul48_v2i64:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT:    v_mul_hi_i32_i24_e32 v1, v0, v4
; GFX9-NEXT:    v_mul_i32_i24_e32 v0, v0, v4
; GFX9-NEXT:    v_mul_hi_i32_i24_e32 v3, v2, v6
; GFX9-NEXT:    v_mul_i32_i24_e32 v2, v2, v6
; GFX9-NEXT:    s_setpc_b64 s[30:31]
;
; EG-LABEL: test_smul48_v2i64:
; EG:       ; %bb.0:
; EG-NEXT:    CF_END
; EG-NEXT:    PAD
;
; CM-LABEL: test_smul48_v2i64:
; CM:       ; %bb.0:
; CM-NEXT:    CF_END
; CM-NEXT:    PAD
  %shl.lhs = shl <2 x i64> %lhs, <i64 40, i64 40>
  %lhs24 = ashr <2 x i64> %shl.lhs, <i64 40, i64 40>
  %shl.rhs = shl <2 x i64> %rhs, <i64 40, i64 40>
  %rhs24 = ashr <2 x i64> %shl.rhs, <i64 40, i64 40>
  %mul = mul <2 x i64> %lhs24, %rhs24
  ret <2 x i64> %mul
}

; This requires handling of the original 64-bit mul node to eliminate
; unnecessary extension instructions because after legalization they
; will not be removed by SimplifyDemandedBits because there are
; multiple uses by the separate mul and mulhi.
define amdgpu_kernel void @test_smul24_i64(ptr addrspace(1) %out, [8 x i32], i32 %a, [8 x i32], i32 %b) #0 {
; SI-LABEL: test_smul24_i64:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x9
; SI-NEXT:    s_load_dword s6, s[4:5], 0x13
; SI-NEXT:    s_load_dword s4, s[4:5], 0x1c
; SI-NEXT:    s_mov_b32 s3, 0xf000
; SI-NEXT:    s_mov_b32 s2, -1
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_bfe_i32 s5, s6, 0x180000
; SI-NEXT:    s_bfe_i32 s4, s4, 0x180000
; SI-NEXT:    v_mov_b32_e32 v0, s5
; SI-NEXT:    s_mul_i32 s5, s4, s5
; SI-NEXT:    v_mul_hi_i32_i24_e32 v1, s4, v0
; SI-NEXT:    v_mov_b32_e32 v0, s5
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: test_smul24_i64:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dword s6, s[4:5], 0x4c
; VI-NEXT:    s_load_dword s7, s[4:5], 0x70
; VI-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_bfe_i32 s4, s6, 0x180000
; VI-NEXT:    s_bfe_i32 s5, s7, 0x180000
; VI-NEXT:    v_mov_b32_e32 v0, s4
; VI-NEXT:    v_mul_hi_i32_i24_e32 v1, s5, v0
; VI-NEXT:    v_mul_i32_i24_e32 v0, s5, v0
; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: test_smul24_i64:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_load_dword s6, s[4:5], 0x4c
; GFX9-NEXT:    s_load_dword s7, s[4:5], 0x70
; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
; GFX9-NEXT:    s_mov_b32 s3, 0xf000
; GFX9-NEXT:    s_mov_b32 s2, -1
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    s_bfe_i32 s4, s6, 0x180000
; GFX9-NEXT:    s_bfe_i32 s5, s7, 0x180000
; GFX9-NEXT:    s_mul_hi_i32 s6, s5, s4
; GFX9-NEXT:    s_mul_i32 s5, s5, s4
; GFX9-NEXT:    v_mov_b32_e32 v0, s5
; GFX9-NEXT:    v_mov_b32_e32 v1, s6
; GFX9-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; GFX9-NEXT:    s_endpgm
;
; EG-LABEL: test_smul24_i64:
; EG:       ; %bb.0:
; EG-NEXT:    ALU 9, @4, KC0[CB0:0-32], KC1[]
; EG-NEXT:    MEM_RAT_CACHELESS STORE_RAW T0.XY, T1.X, 1
; EG-NEXT:    CF_END
; EG-NEXT:    PAD
; EG-NEXT:    ALU clause starting at 4:
; EG-NEXT:     LSHL T0.W, KC0[4].Z, literal.x,
; EG-NEXT:     LSHL * T1.W, KC0[6].W, literal.x,
; EG-NEXT:    8(1.121039e-44), 0(0.000000e+00)
; EG-NEXT:     ASHR T1.W, PS, literal.x,
; EG-NEXT:     ASHR * T0.W, PV.W, literal.x,
; EG-NEXT:    8(1.121039e-44), 0(0.000000e+00)
; EG-NEXT:     MULHI_INT * T0.Y, PV.W, PS,
; EG-NEXT:     LSHR T1.X, KC0[2].Y, literal.x,
; EG-NEXT:     MULLO_INT * T0.X, T1.W, T0.W,
; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
;
; CM-LABEL: test_smul24_i64:
; CM:       ; %bb.0:
; CM-NEXT:    ALU 14, @4, KC0[CB0:0-32], KC1[]
; CM-NEXT:    MEM_RAT_CACHELESS STORE_DWORD T1, T0.X
; CM-NEXT:    CF_END
; CM-NEXT:    PAD
; CM-NEXT:    ALU clause starting at 4:
; CM-NEXT:     LSHL T0.Z, KC0[4].Z, literal.x,
; CM-NEXT:     LSHL * T0.W, KC0[6].W, literal.x,
; CM-NEXT:    8(1.121039e-44), 0(0.000000e+00)
; CM-NEXT:     LSHR T0.X, KC0[2].Y, literal.x,
; CM-NEXT:     ASHR T1.Z, PV.W, literal.y,
; CM-NEXT:     ASHR * T0.W, PV.Z, literal.y,
; CM-NEXT:    2(2.802597e-45), 8(1.121039e-44)
; CM-NEXT:     MULLO_INT T1.X, T1.Z, T0.W,
; CM-NEXT:     MULLO_INT T1.Y (MASKED), T1.Z, T0.W,
; CM-NEXT:     MULLO_INT T1.Z (MASKED), T1.Z, T0.W,
; CM-NEXT:     MULLO_INT * T1.W (MASKED), T1.Z, T0.W,
; CM-NEXT:     MULHI_INT24 T1.X (MASKED), KC0[6].W, KC0[4].Z,
; CM-NEXT:     MULHI_INT24 T1.Y, KC0[6].W, KC0[4].Z,
; CM-NEXT:     MULHI_INT24 T1.Z (MASKED), KC0[6].W, KC0[4].Z,
; CM-NEXT:     MULHI_INT24 * T1.W (MASKED), KC0[6].W, KC0[4].Z,
  %shl.i = shl i32 %a, 8
  %shr.i = ashr i32 %shl.i, 8
  %conv.i = sext i32 %shr.i to i64
  %shl1.i = shl i32 %b, 8
  %shr2.i = ashr i32 %shl1.i, 8
  %conv3.i = sext i32 %shr2.i to i64
  %mul.i = mul i64 %conv3.i, %conv.i
  store i64 %mul.i, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @test_smul24_i64_square(ptr addrspace(1) %out, i32 %a, i32 %b) #0 {
; SI-LABEL: test_smul24_i64_square:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dword s6, s[4:5], 0xb
; SI-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x9
; SI-NEXT:    s_mov_b32 s3, 0xf000
; SI-NEXT:    s_mov_b32 s2, -1
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_bfe_i32 s4, s6, 0x180000
; SI-NEXT:    s_mul_i32 s5, s4, s4
; SI-NEXT:    v_mul_hi_i32_i24_e64 v1, s4, s4
; SI-NEXT:    v_mov_b32_e32 v0, s5
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: test_smul24_i64_square:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dword s6, s[4:5], 0x2c
; VI-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_bfe_i32 s4, s6, 0x180000
; VI-NEXT:    v_mul_hi_i32_i24_e64 v1, s4, s4
; VI-NEXT:    v_mul_i32_i24_e64 v0, s4, s4
; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: test_smul24_i64_square:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_load_dword s6, s[4:5], 0x2c
; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
; GFX9-NEXT:    s_mov_b32 s3, 0xf000
; GFX9-NEXT:    s_mov_b32 s2, -1
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    s_bfe_i32 s4, s6, 0x180000
; GFX9-NEXT:    s_mul_hi_i32 s5, s4, s4
; GFX9-NEXT:    s_mul_i32 s4, s4, s4
; GFX9-NEXT:    v_mov_b32_e32 v0, s4
; GFX9-NEXT:    v_mov_b32_e32 v1, s5
; GFX9-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; GFX9-NEXT:    s_endpgm
;
; EG-LABEL: test_smul24_i64_square:
; EG:       ; %bb.0:
; EG-NEXT:    ALU 7, @4, KC0[CB0:0-32], KC1[]
; EG-NEXT:    MEM_RAT_CACHELESS STORE_RAW T0.XY, T1.X, 1
; EG-NEXT:    CF_END
; EG-NEXT:    PAD
; EG-NEXT:    ALU clause starting at 4:
; EG-NEXT:     LSHL * T0.W, KC0[2].Z, literal.x,
; EG-NEXT:    8(1.121039e-44), 0(0.000000e+00)
; EG-NEXT:     ASHR * T0.W, PV.W, literal.x,
; EG-NEXT:    8(1.121039e-44), 0(0.000000e+00)
; EG-NEXT:     MULHI_INT * T0.Y, PV.W, PV.W,
; EG-NEXT:     LSHR T1.X, KC0[2].Y, literal.x,
; EG-NEXT:     MULLO_INT * T0.X, T0.W, T0.W,
; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
;
; CM-LABEL: test_smul24_i64_square:
; CM:       ; %bb.0:
; CM-NEXT:    ALU 12, @4, KC0[CB0:0-32], KC1[]
; CM-NEXT:    MEM_RAT_CACHELESS STORE_DWORD T1, T0.X
; CM-NEXT:    CF_END
; CM-NEXT:    PAD
; CM-NEXT:    ALU clause starting at 4:
; CM-NEXT:     LSHL * T0.W, KC0[2].Z, literal.x,
; CM-NEXT:    8(1.121039e-44), 0(0.000000e+00)
; CM-NEXT:     LSHR T0.X, KC0[2].Y, literal.x,
; CM-NEXT:     ASHR * T0.W, PV.W, literal.y,
; CM-NEXT:    2(2.802597e-45), 8(1.121039e-44)
; CM-NEXT:     MULLO_INT T1.X, T0.W, T0.W,
; CM-NEXT:     MULLO_INT T1.Y (MASKED), T0.W, T0.W,
; CM-NEXT:     MULLO_INT T1.Z (MASKED), T0.W, T0.W,
; CM-NEXT:     MULLO_INT * T1.W (MASKED), T0.W, T0.W,
; CM-NEXT:     MULHI_INT24 T1.X (MASKED), KC0[2].Z, KC0[2].Z,
; CM-NEXT:     MULHI_INT24 T1.Y, KC0[2].Z, KC0[2].Z,
; CM-NEXT:     MULHI_INT24 T1.Z (MASKED), KC0[2].Z, KC0[2].Z,
; CM-NEXT:     MULHI_INT24 * T1.W (MASKED), KC0[2].Z, KC0[2].Z,
  %shl.i = shl i32 %a, 8
  %shr.i = ashr i32 %shl.i, 8
  %conv.i = sext i32 %shr.i to i64
  %mul.i = mul i64 %conv.i, %conv.i
  store i64 %mul.i, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @test_smul24_i33(ptr addrspace(1) %out, i33 %a, i33 %b) #0 {
; SI-LABEL: test_smul24_i33:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x9
; SI-NEXT:    s_load_dwordx2 s[8:9], s[4:5], 0xd
; SI-NEXT:    s_mov_b32 s7, 0xf000
; SI-NEXT:    s_mov_b32 s6, -1
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s4, s0
; SI-NEXT:    s_mov_b32 s5, s1
; SI-NEXT:    s_bfe_i32 s0, s8, 0x180000
; SI-NEXT:    s_bfe_i32 s1, s2, 0x180000
; SI-NEXT:    v_mov_b32_e32 v0, s0
; SI-NEXT:    s_mul_i32 s0, s1, s0
; SI-NEXT:    v_mul_hi_i32_i24_e32 v1, s1, v0
; SI-NEXT:    v_mov_b32_e32 v0, s0
; SI-NEXT:    v_lshl_b64 v[0:1], v[0:1], 31
; SI-NEXT:    v_ashr_i64 v[0:1], v[0:1], 31
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: test_smul24_i33:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; VI-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x34
; VI-NEXT:    s_mov_b32 s7, 0xf000
; VI-NEXT:    s_mov_b32 s6, -1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_bfe_i32 s2, s2, 0x180000
; VI-NEXT:    s_bfe_i32 s3, s4, 0x180000
; VI-NEXT:    v_mov_b32_e32 v0, s3
; VI-NEXT:    v_mul_hi_i32_i24_e32 v1, s2, v0
; VI-NEXT:    v_mul_i32_i24_e32 v0, s2, v0
; VI-NEXT:    v_lshlrev_b64 v[0:1], 31, v[0:1]
; VI-NEXT:    s_mov_b32 s4, s0
; VI-NEXT:    v_ashrrev_i64 v[0:1], 31, v[0:1]
; VI-NEXT:    s_mov_b32 s5, s1
; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: test_smul24_i33:
; GFX9:       ; %bb.0: ; %entry
; GFX9-NEXT:    s_load_dword s6, s[4:5], 0x2c
; GFX9-NEXT:    s_load_dword s7, s[4:5], 0x34
; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
; GFX9-NEXT:    s_mov_b32 s3, 0xf000
; GFX9-NEXT:    s_mov_b32 s2, -1
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    s_bfe_i32 s4, s6, 0x180000
; GFX9-NEXT:    s_bfe_i32 s6, s7, 0x180000
; GFX9-NEXT:    s_mul_hi_i32 s5, s4, s6
; GFX9-NEXT:    s_mul_i32 s4, s4, s6
; GFX9-NEXT:    s_lshl_b64 s[4:5], s[4:5], 31
; GFX9-NEXT:    s_ashr_i64 s[4:5], s[4:5], 31
; GFX9-NEXT:    v_mov_b32_e32 v0, s4
; GFX9-NEXT:    v_mov_b32_e32 v1, s5
; GFX9-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; GFX9-NEXT:    s_endpgm
;
; EG-LABEL: test_smul24_i33:
; EG:       ; %bb.0: ; %entry
; EG-NEXT:    ALU 10, @4, KC0[CB0:0-32], KC1[]
; EG-NEXT:    MEM_RAT_CACHELESS STORE_RAW T1.XY, T2.X, 1
; EG-NEXT:    CF_END
; EG-NEXT:    PAD
; EG-NEXT:    ALU clause starting at 4:
; EG-NEXT:     LSHL T0.W, KC0[2].W, literal.x,
; EG-NEXT:     LSHL * T1.W, KC0[3].Y, literal.x,
; EG-NEXT:    8(1.121039e-44), 0(0.000000e+00)
; EG-NEXT:     ASHR T1.W, PS, literal.x,
; EG-NEXT:     ASHR * T0.W, PV.W, literal.x,
; EG-NEXT:    8(1.121039e-44), 0(0.000000e+00)
; EG-NEXT:     MULHI_INT * T0.X, PS, PV.W,
; EG-NEXT:     MULLO_INT * T1.X, T0.W, T1.W,
; EG-NEXT:     LSHR T2.X, KC0[2].Y, literal.x,
; EG-NEXT:     BFE_INT * T1.Y, T0.X, 0.0, 1,
; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
;
; CM-LABEL: test_smul24_i33:
; CM:       ; %bb.0: ; %entry
; CM-NEXT:    ALU 16, @4, KC0[CB0:0-32], KC1[]
; CM-NEXT:    MEM_RAT_CACHELESS STORE_DWORD T1, T2.X
; CM-NEXT:    CF_END
; CM-NEXT:    PAD
; CM-NEXT:    ALU clause starting at 4:
; CM-NEXT:     LSHL T0.Z, KC0[2].W, literal.x,
; CM-NEXT:     LSHL * T0.W, KC0[3].Y, literal.x,
; CM-NEXT:    8(1.121039e-44), 0(0.000000e+00)
; CM-NEXT:     ASHR T1.Z, PV.W, literal.x,
; CM-NEXT:     ASHR * T0.W, PV.Z, literal.x,
; CM-NEXT:    8(1.121039e-44), 0(0.000000e+00)
; CM-NEXT:     MULHI_INT24 T0.X, KC0[2].W, KC0[3].Y,
; CM-NEXT:     MULHI_INT24 T0.Y (MASKED), KC0[2].W, KC0[3].Y,
; CM-NEXT:     MULHI_INT24 T0.Z (MASKED), KC0[2].W, KC0[3].Y,
; CM-NEXT:     MULHI_INT24 * T0.W (MASKED), KC0[2].W, KC0[3].Y,
; CM-NEXT:     MULLO_INT T1.X, T0.W, T1.Z,
; CM-NEXT:     MULLO_INT T1.Y (MASKED), T0.W, T1.Z,
; CM-NEXT:     MULLO_INT T1.Z (MASKED), T0.W, T1.Z,
; CM-NEXT:     MULLO_INT * T1.W (MASKED), T0.W, T1.Z,
; CM-NEXT:     LSHR T2.X, KC0[2].Y, literal.x,
; CM-NEXT:     BFE_INT * T1.Y, T0.X, 0.0, 1,
; CM-NEXT:    2(2.802597e-45), 0(0.000000e+00)
entry:
  %a.shl = shl i33 %a, 9
  %a.24 = ashr i33 %a.shl, 9
  %b.shl = shl i33 %b, 9
  %b.24 = ashr i33 %b.shl, 9
  %mul24 = mul i33 %a.24, %b.24
  %ext = sext i33 %mul24 to i64
  store i64 %ext, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @test_smulhi24_i33(ptr addrspace(1) %out, i33 %a, i33 %b) {
; SI-LABEL: test_smulhi24_i33:
; SI:       ; %bb.0: ; %entry
; SI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x9
; SI-NEXT:    s_load_dwordx2 s[8:9], s[4:5], 0xd
; SI-NEXT:    s_mov_b32 s7, 0xf000
; SI-NEXT:    s_mov_b32 s6, -1
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_mov_b32 s4, s0
; SI-NEXT:    s_mov_b32 s5, s1
; SI-NEXT:    v_mov_b32_e32 v0, s8
; SI-NEXT:    v_mul_hi_i32_i24_e32 v0, s2, v0
; SI-NEXT:    v_and_b32_e32 v0, 1, v0
; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: test_smulhi24_i33:
; VI:       ; %bb.0: ; %entry
; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
; VI-NEXT:    s_load_dwordx2 s[8:9], s[4:5], 0x34
; VI-NEXT:    s_mov_b32 s7, 0xf000
; VI-NEXT:    s_mov_b32 s6, -1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_mov_b32 s4, s0
; VI-NEXT:    v_mov_b32_e32 v0, s8
; VI-NEXT:    v_mul_hi_i32_i24_e32 v0, s2, v0
; VI-NEXT:    s_mov_b32 s5, s1
; VI-NEXT:    v_and_b32_e32 v0, 1, v0
; VI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: test_smulhi24_i33:
; GFX9:       ; %bb.0: ; %entry
; GFX9-NEXT:    s_load_dword s6, s[4:5], 0x2c
; GFX9-NEXT:    s_load_dword s7, s[4:5], 0x34
; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
; GFX9-NEXT:    s_mov_b32 s3, 0xf000
; GFX9-NEXT:    s_mov_b32 s2, -1
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    s_bfe_i32 s4, s6, 0x180000
; GFX9-NEXT:    s_bfe_i32 s5, s7, 0x180000
; GFX9-NEXT:    s_mul_hi_i32 s4, s4, s5
; GFX9-NEXT:    s_and_b32 s4, s4, 1
; GFX9-NEXT:    v_mov_b32_e32 v0, s4
; GFX9-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GFX9-NEXT:    s_endpgm
;
; EG-LABEL: test_smulhi24_i33:
; EG:       ; %bb.0: ; %entry
; EG-NEXT:    ALU 9, @4, KC0[CB0:0-32], KC1[]
; EG-NEXT:    MEM_RAT_CACHELESS STORE_RAW T0.X, T1.X, 1
; EG-NEXT:    CF_END
; EG-NEXT:    PAD
; EG-NEXT:    ALU clause starting at 4:
; EG-NEXT:     LSHL T0.W, KC0[2].W, literal.x,
; EG-NEXT:     LSHL * T1.W, KC0[3].Y, literal.x,
; EG-NEXT:    8(1.121039e-44), 0(0.000000e+00)
; EG-NEXT:     ASHR T1.W, PS, literal.x,
; EG-NEXT:     ASHR * T0.W, PV.W, literal.x,
; EG-NEXT:    8(1.121039e-44), 0(0.000000e+00)
; EG-NEXT:     MULHI_INT * T0.X, PS, PV.W,
; EG-NEXT:     AND_INT T0.X, PS, 1,
; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
;
; CM-LABEL: test_smulhi24_i33:
; CM:       ; %bb.0: ; %entry
; CM-NEXT:    ALU 6, @4, KC0[CB0:0-32], KC1[]
; CM-NEXT:    MEM_RAT_CACHELESS STORE_DWORD T0.X, T1.X
; CM-NEXT:    CF_END
; CM-NEXT:    PAD
; CM-NEXT:    ALU clause starting at 4:
; CM-NEXT:     MULHI_INT24 T0.X, KC0[2].W, KC0[3].Y,
; CM-NEXT:     MULHI_INT24 T0.Y (MASKED), KC0[2].W, KC0[3].Y,
; CM-NEXT:     MULHI_INT24 T0.Z (MASKED), KC0[2].W, KC0[3].Y,
; CM-NEXT:     MULHI_INT24 * T0.W (MASKED), KC0[2].W, KC0[3].Y,
; CM-NEXT:     AND_INT * T0.X, PV.X, 1,
; CM-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
; CM-NEXT:    2(2.802597e-45), 0(0.000000e+00)
entry:
  %tmp0 = shl i33 %a, 9
  %a_24 = ashr i33 %tmp0, 9
  %tmp1 = shl i33 %b, 9
  %b_24 = ashr i33 %tmp1, 9
  %tmp2 = mul i33 %a_24, %b_24
  %hi = lshr i33 %tmp2, 32
  %trunc = trunc i33 %hi to i32

  store i32 %trunc, ptr addrspace(1) %out
  ret void
}

define amdgpu_kernel void @simplify_i24_crash(ptr addrspace(1) %out, i32 %arg0, <2 x i32> %arg1, <2 x i32> %arg2) {
; SI-LABEL: simplify_i24_crash:
; SI:       ; %bb.0: ; %bb
; SI-NEXT:    s_load_dword s0, s[4:5], 0xb
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_cmp_lg_u32 s0, 0
; SI-NEXT:    s_cbranch_scc0 .LBB8_2
; SI-NEXT:  ; %bb.1: ; %bb7
; SI-NEXT:    s_endpgm
; SI-NEXT:  .LBB8_2: ; %bb11
; SI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0xd
; SI-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x9
; SI-NEXT:    s_mov_b32 s7, 0xf000
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_bfe_i32 s0, s0, 0x180000
; SI-NEXT:    s_bfe_i32 s1, s2, 0x180000
; SI-NEXT:    s_mul_i32 s0, s0, s1
; SI-NEXT:    s_mov_b32 s6, -1
; SI-NEXT:    v_mov_b32_e32 v0, s0
; SI-NEXT:    v_mov_b32_e32 v1, s0
; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: simplify_i24_crash:
; VI:       ; %bb.0: ; %bb
; VI-NEXT:    s_load_dword s0, s[4:5], 0x2c
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_cmp_lg_u32 s0, 0
; VI-NEXT:    s_cbranch_scc0 .LBB8_2
; VI-NEXT:  ; %bb.1: ; %bb7
; VI-NEXT:    s_endpgm
; VI-NEXT:  .LBB8_2: ; %bb11
; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x34
; VI-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x24
; VI-NEXT:    s_mov_b32 s7, 0xf000
; VI-NEXT:    s_mov_b32 s6, -1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_bfe_i32 s0, s0, 0x180000
; VI-NEXT:    s_bfe_i32 s1, s2, 0x180000
; VI-NEXT:    s_mul_i32 s0, s0, s1
; VI-NEXT:    v_mov_b32_e32 v0, s0
; VI-NEXT:    v_mov_b32_e32 v1, s0
; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: simplify_i24_crash:
; GFX9:       ; %bb.0: ; %bb
; GFX9-NEXT:    s_load_dword s0, s[4:5], 0x2c
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    s_cmp_lg_u32 s0, 0
; GFX9-NEXT:    s_cbranch_scc0 .LBB8_2
; GFX9-NEXT:  ; %bb.1: ; %bb7
; GFX9-NEXT:    s_endpgm
; GFX9-NEXT:  .LBB8_2: ; %bb11
; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x34
; GFX9-NEXT:    s_load_dwordx2 s[8:9], s[4:5], 0x24
; GFX9-NEXT:    s_mov_b32 s11, 0xf000
; GFX9-NEXT:    s_mov_b32 s10, -1
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    s_bfe_i32 s0, s0, 0x180000
; GFX9-NEXT:    s_bfe_i32 s1, s2, 0x180000
; GFX9-NEXT:    s_mul_i32 s0, s0, s1
; GFX9-NEXT:    v_mov_b32_e32 v0, s0
; GFX9-NEXT:    v_mov_b32_e32 v1, s0
; GFX9-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
; GFX9-NEXT:    s_endpgm
;
; EG-LABEL: simplify_i24_crash:
; EG:       ; %bb.0: ; %bb
; EG-NEXT:    ALU_PUSH_BEFORE 1, @6, KC0[CB0:0-32], KC1[]
; EG-NEXT:    JUMP @5 POP:1
; EG-NEXT:    ALU 12, @8, KC0[CB0:0-32], KC1[]
; EG-NEXT:    MEM_RAT_CACHELESS STORE_RAW T0.XY, T1.X, 0
; EG-NEXT:    POP @5 POP:1
; EG-NEXT:    CF_END
; EG-NEXT:    ALU clause starting at 6:
; EG-NEXT:     SETNE_INT * T0.W, KC0[2].Z, 0.0,
; EG-NEXT:     PRED_SETE_INT * ExecMask,PredicateBit (MASKED), PV.W, 0.0,
; EG-NEXT:    ALU clause starting at 8:
; EG-NEXT:     MOV T0.W, KC0[3].Y,
; EG-NEXT:     MOV * T1.W, KC0[2].W,
; EG-NEXT:     LSHL T1.W, PS, literal.x,
; EG-NEXT:     LSHL * T0.W, PV.W, literal.x,
; EG-NEXT:    8(1.121039e-44), 0(0.000000e+00)
; EG-NEXT:     ASHR T0.W, PS, literal.x,
; EG-NEXT:     ASHR * T1.W, PV.W, literal.x,
; EG-NEXT:    8(1.121039e-44), 0(0.000000e+00)
; EG-NEXT:     MOV T2.W, KC0[2].Y,
; EG-NEXT:     MULLO_INT * T0.X, PS, PV.W,
; EG-NEXT:     LSHR T1.X, PV.W, literal.x,
; EG-NEXT:     MOV * T0.Y, PS,
; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
;
; CM-LABEL: simplify_i24_crash:
; CM:       ; %bb.0: ; %bb
; CM-NEXT:    ALU_PUSH_BEFORE 1, @6, KC0[CB0:0-32], KC1[]
; CM-NEXT:    JUMP @5 POP:1
; CM-NEXT:    ALU 15, @8, KC0[CB0:0-32], KC1[]
; CM-NEXT:    MEM_RAT_CACHELESS STORE_DWORD T0, T1.X
; CM-NEXT:    POP @5 POP:1
; CM-NEXT:    CF_END
; CM-NEXT:    ALU clause starting at 6:
; CM-NEXT:     SETNE_INT * T0.W, KC0[2].Z, 0.0,
; CM-NEXT:     PRED_SETE_INT * ExecMask,PredicateBit (MASKED), PV.W, 0.0,
; CM-NEXT:    ALU clause starting at 8:
; CM-NEXT:     MOV T0.Z, KC0[3].Y,
; CM-NEXT:     MOV * T0.W, KC0[2].W,
; CM-NEXT:     LSHL T1.Z, PV.W, literal.x,
; CM-NEXT:     LSHL * T0.W, PV.Z, literal.x,
; CM-NEXT:    8(1.121039e-44), 0(0.000000e+00)
; CM-NEXT:     MOV T0.Y, KC0[2].Y,
; CM-NEXT:     ASHR T0.Z, PV.W, literal.x,
; CM-NEXT:     ASHR * T0.W, PV.Z, literal.x,
; CM-NEXT:    8(1.121039e-44), 0(0.000000e+00)
; CM-NEXT:     MULLO_INT T0.X, T0.W, T0.Z,
; CM-NEXT:     MULLO_INT T0.Y (MASKED), T0.W, T0.Z,
; CM-NEXT:     MULLO_INT T0.Z (MASKED), T0.W, T0.Z,
; CM-NEXT:     MULLO_INT * T0.W (MASKED), T0.W, T0.Z,
; CM-NEXT:     LSHR T1.X, T0.Y, literal.x,
; CM-NEXT:     MOV * T0.Y, PV.X,
; CM-NEXT:    2(2.802597e-45), 0(0.000000e+00)
bb:
  %cmp = icmp eq i32 %arg0, 0
  br i1 %cmp, label %bb11, label %bb7

bb11:
  %tmp14 = shufflevector <2 x i32> %arg1, <2 x i32> poison, <2 x i32> zeroinitializer
  %tmp16 = shufflevector <2 x i32> %arg2, <2 x i32> poison, <2 x i32> zeroinitializer
  %tmp17 = shl <2 x i32> %tmp14, <i32 8, i32 8>
  %tmp18 = ashr <2 x i32> %tmp17, <i32 8, i32 8>
  %tmp19 = shl <2 x i32> %tmp16, <i32 8, i32 8>
  %tmp20 = ashr <2 x i32> %tmp19, <i32 8, i32 8>
  %tmp21 = mul <2 x i32> %tmp18, %tmp20
  store <2 x i32> %tmp21, ptr addrspace(1) %out
  br label %bb7

bb7:
  ret void

}

define amdgpu_kernel void @test_umul_i24(ptr addrspace(1) %out, i32 %arg) {
; SI-LABEL: test_umul_i24:
; SI:       ; %bb.0:
; SI-NEXT:    s_load_dword s2, s[4:5], 0xb
; SI-NEXT:    v_mov_b32_e32 v0, 0xff803fe1
; SI-NEXT:    s_mov_b64 s[0:1], 0
; SI-NEXT:    s_mov_b32 s3, 0xf000
; SI-NEXT:    s_waitcnt lgkmcnt(0)
; SI-NEXT:    s_lshr_b32 s2, s2, 9
; SI-NEXT:    v_mul_hi_u32 v0, s2, v0
; SI-NEXT:    s_mul_i32 s2, s2, 0xff803fe1
; SI-NEXT:    v_alignbit_b32 v0, v0, s2, 1
; SI-NEXT:    s_mov_b32 s2, -1
; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; SI-NEXT:    s_endpgm
;
; VI-LABEL: test_umul_i24:
; VI:       ; %bb.0:
; VI-NEXT:    s_load_dword s0, s[4:5], 0x2c
; VI-NEXT:    v_mov_b32_e32 v0, 0xff803fe1
; VI-NEXT:    s_mov_b32 s3, 0xf000
; VI-NEXT:    s_mov_b32 s2, -1
; VI-NEXT:    s_waitcnt lgkmcnt(0)
; VI-NEXT:    s_lshr_b32 s0, s0, 9
; VI-NEXT:    v_mad_u64_u32 v[0:1], s[0:1], s0, v0, 0
; VI-NEXT:    s_mov_b64 s[0:1], 0
; VI-NEXT:    v_alignbit_b32 v0, v1, v0, 1
; VI-NEXT:    s_nop 2
; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; VI-NEXT:    s_endpgm
;
; GFX9-LABEL: test_umul_i24:
; GFX9:       ; %bb.0:
; GFX9-NEXT:    s_load_dword s2, s[4:5], 0x2c
; GFX9-NEXT:    s_mov_b64 s[0:1], 0
; GFX9-NEXT:    s_mov_b32 s3, 0xf000
; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
; GFX9-NEXT:    s_lshr_b32 s2, s2, 9
; GFX9-NEXT:    s_mul_hi_u32 s4, s2, 0xff803fe1
; GFX9-NEXT:    s_mul_i32 s2, s2, 0xff803fe1
; GFX9-NEXT:    v_mov_b32_e32 v0, s2
; GFX9-NEXT:    v_alignbit_b32 v0, s4, v0, 1
; GFX9-NEXT:    s_mov_b32 s2, -1
; GFX9-NEXT:    buffer_store_dword v0, off, s[0:3], 0
; GFX9-NEXT:    s_endpgm
;
; EG-LABEL: test_umul_i24:
; EG:       ; %bb.0:
; EG-NEXT:    ALU 8, @4, KC0[CB0:0-32], KC1[]
; EG-NEXT:    MEM_RAT_CACHELESS STORE_RAW T0.X, T1.X, 1
; EG-NEXT:    CF_END
; EG-NEXT:    PAD
; EG-NEXT:    ALU clause starting at 4:
; EG-NEXT:     LSHR * T0.W, KC0[2].Z, literal.x,
; EG-NEXT:    9(1.261169e-44), 0(0.000000e+00)
; EG-NEXT:     MULHI * T0.X, PV.W, literal.x,
; EG-NEXT:    -8372255(nan), 0(0.000000e+00)
; EG-NEXT:     MULLO_INT * T0.Y, T0.W, literal.x,
; EG-NEXT:    -8372255(nan), 0(0.000000e+00)
; EG-NEXT:     BIT_ALIGN_INT T0.X, T0.X, PS, 1,
; EG-NEXT:     MOV * T1.X, literal.x,
; EG-NEXT:    0(0.000000e+00), 0(0.000000e+00)
;
; CM-LABEL: test_umul_i24:
; CM:       ; %bb.0:
; CM-NEXT:    ALU 14, @4, KC0[CB0:0-32], KC1[]
; CM-NEXT:    MEM_RAT_CACHELESS STORE_DWORD T0.X, T1.X
; CM-NEXT:    CF_END
; CM-NEXT:    PAD
; CM-NEXT:    ALU clause starting at 4:
; CM-NEXT:     LSHR * T0.W, KC0[2].Z, literal.x,
; CM-NEXT:    9(1.261169e-44), 0(0.000000e+00)
; CM-NEXT:     MULHI T0.X, T0.W, literal.x,
; CM-NEXT:     MULHI T0.Y (MASKED), T0.W, literal.x,
; CM-NEXT:     MULHI T0.Z (MASKED), T0.W, literal.x,
; CM-NEXT:     MULHI * T0.W (MASKED), T0.W, literal.x,
; CM-NEXT:    -8372255(nan), 0(0.000000e+00)
; CM-NEXT:     MULLO_INT T0.X (MASKED), T0.W, literal.x,
; CM-NEXT:     MULLO_INT T0.Y, T0.W, literal.x,
; CM-NEXT:     MULLO_INT T0.Z (MASKED), T0.W, literal.x,
; CM-NEXT:     MULLO_INT * T0.W (MASKED), T0.W, literal.x,
; CM-NEXT:    -8372255(nan), 0(0.000000e+00)
; CM-NEXT:     BIT_ALIGN_INT * T0.X, T0.X, PV.Y, 1,
; CM-NEXT:     MOV * T1.X, literal.x,
; CM-NEXT:    0(0.000000e+00), 0(0.000000e+00)
  %i = lshr i32 %arg, 9
  %i1 = zext i32 %i to i64
  %i2 = mul i64 %i1, 4286595041
  %i3 = lshr i64 %i2, 1
  %i4 = trunc i64 %i3 to i32
  store i32 %i4, ptr addrspace(1) null, align 4
  ret void
}

attributes #0 = { nounwind }