xref: /llvm-project/llvm/test/CodeGen/X86/amx-tile-avx512-internals.ll (revision 48803bc8c7be25745a0e623e6753261c07281b06) 
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+amx-tile,+amx-bf16,+avx10.2-512, \
3; RUN: -mattr=+amx-avx512 -verify-machineinstrs | FileCheck %s
4
5define void @test_amx(i8* %pointer, i8* %base, i32 %index, i64 %stride) {
6; CHECK-LABEL: test_amx:
7; CHECK:       # %bb.0:
8; CHECK-NEXT:    vxorps %xmm0, %xmm0, %xmm0
9; CHECK-NEXT:    vmovups %zmm0, -{{[0-9]+}}(%rsp)
10; CHECK-NEXT:    movb $1, -{{[0-9]+}}(%rsp)
11; CHECK-NEXT:    movb $8, -{{[0-9]+}}(%rsp)
12; CHECK-NEXT:    movw $8, -{{[0-9]+}}(%rsp)
13; CHECK-NEXT:    ldtilecfg -{{[0-9]+}}(%rsp)
14; CHECK-NEXT:    movw $8, %ax
15; CHECK-NEXT:    tileloadd (%rsi,%rcx), %tmm0
16; CHECK-NEXT:    tcvtrowd2ps %edx, %tmm0, %zmm0
17; CHECK-NEXT:    tcvtrowd2ps $16, %tmm0, %zmm0
18; CHECK-NEXT:    tcvtrowps2bf16h %edx, %tmm0, %zmm0
19; CHECK-NEXT:    tcvtrowps2bf16h $16, %tmm0, %zmm0
20; CHECK-NEXT:    tcvtrowps2bf16l %edx, %tmm0, %zmm0
21; CHECK-NEXT:    tcvtrowps2bf16l $16, %tmm0, %zmm0
22; CHECK-NEXT:    tcvtrowps2phh %edx, %tmm0, %zmm0
23; CHECK-NEXT:    tcvtrowps2phh $16, %tmm0, %zmm0
24; CHECK-NEXT:    tcvtrowps2phl %edx, %tmm0, %zmm0
25; CHECK-NEXT:    tcvtrowps2phl $16, %tmm0, %zmm0
26; CHECK-NEXT:    tilemovrow %edx, %tmm0, %zmm0
27; CHECK-NEXT:    tilemovrow $16, %tmm0, %zmm0
28; CHECK-NEXT:    tilestored %tmm0, (%rdi,%rcx)
29; CHECK-NEXT:    tilerelease
30; CHECK-NEXT:    vzeroupper
31; CHECK-NEXT:    retq
32
33  %a = call x86_amx @llvm.x86.tileloadd64.internal(i16 8, i16 8, i8* %base, i64 %stride)
34  call <16 x float> @llvm.x86.tcvtrowd2ps.internal(i16 8, i16 8, x86_amx %a, i32 %index)
35  call <16 x float> @llvm.x86.tcvtrowd2ps.internal(i16 8, i16 8, x86_amx %a, i32 16)
36  call <32 x bfloat> @llvm.x86.tcvtrowps2bf16h.internal(i16 8, i16 8, x86_amx %a, i32 %index)
37  call <32 x bfloat> @llvm.x86.tcvtrowps2bf16h.internal(i16 8, i16 8, x86_amx %a, i32 16)
38  call <32 x bfloat> @llvm.x86.tcvtrowps2bf16l.internal(i16 8, i16 8, x86_amx %a, i32 %index)
39  call <32 x bfloat> @llvm.x86.tcvtrowps2bf16l.internal(i16 8, i16 8, x86_amx %a, i32 16)
40  call <32 x half> @llvm.x86.tcvtrowps2phh.internal(i16 8, i16 8, x86_amx %a, i32 %index)
41  call <32 x half> @llvm.x86.tcvtrowps2phh.internal(i16 8, i16 8, x86_amx %a, i32 16)
42  call <32 x half> @llvm.x86.tcvtrowps2phl.internal(i16 8, i16 8, x86_amx %a, i32 %index)
43  call <32 x half> @llvm.x86.tcvtrowps2phl.internal(i16 8, i16 8, x86_amx %a, i32 16)
44  call <16 x i32> @llvm.x86.tilemovrow.internal(i16 8, i16 8, x86_amx %a, i32 %index)
45  call <16 x i32> @llvm.x86.tilemovrow.internal(i16 8, i16 8, x86_amx %a, i32 16)
46
47  call void @llvm.x86.tilestored64.internal(i16 8, i16 8, i8* %pointer, i64 %stride, x86_amx %a)
48  ret void
49}
50
51declare x86_amx @llvm.x86.tilezero.internal(i16, i16)
52declare x86_amx @llvm.x86.tileloadd64.internal(i16, i16, i8*, i64)
53declare x86_amx @llvm.x86.tileloaddt164.internal(i16, i16, i8*, i64)
54declare void @llvm.x86.tilestored64.internal(i16, i16, i8*, i64, x86_amx)
55
56declare <16 x float> @llvm.x86.tcvtrowd2ps.internal(i16, i16, x86_amx, i32)
57declare <32 x bfloat> @llvm.x86.tcvtrowps2bf16h.internal(i16, i16, x86_amx, i32)
58declare <32 x bfloat> @llvm.x86.tcvtrowps2bf16l.internal(i16, i16, x86_amx, i32)
59declare <32 x half> @llvm.x86.tcvtrowps2phh.internal(i16, i16, x86_amx, i32)
60declare <32 x half> @llvm.x86.tcvtrowps2phl.internal(i16, i16, x86_amx, i32)
61declare <16 x i32> @llvm.x86.tilemovrow.internal(i16, i16, x86_amx, i32)
62