Skip to content

[X86][AMX] Fix a bug after #83628 #91207

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
May 15, 2024
Merged

[X86][AMX] Fix a bug after #83628 #91207

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
609666e
[X86][AMX] Fix a bug after #83628
phoebewang May 6, 2024
52b7547
Add crash test
phoebewang May 15, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions llvm/lib/Target/X86/X86LowerTileCopy.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -146,7 +146,7 @@ bool X86LowerTileCopy::runOnMachineFunction(MachineFunction &MF) {
addFrameReference(BuildMI(MBB, MI, DL, TII->get(Opc)), TileSS)
.addReg(SrcReg, getKillRegState(SrcMO.isKill()));
MachineOperand &MO = NewMI->getOperand(2);
MO.setReg(GR64Cand);
MO.setReg(GR64Cand ? GR64Cand : X86::RAX);
MO.setIsKill(true);
// tileloadd (%sp, %idx), %tmm
Opc = GET_EGPR_IF_ENABLED(X86::TILELOADD);
Expand All @@ -157,7 +157,7 @@ bool X86LowerTileCopy::runOnMachineFunction(MachineFunction &MF) {
// restore %rax
// mov (%sp) %rax
addFrameReference(
BuildMI(MBB, MI, DL, TII->get(X86::MOV64rm), GR64Cand), StrideSS);
BuildMI(MBB, MI, DL, TII->get(X86::MOV64rm), X86::RAX), StrideSS);
}
MI.eraseFromParent();
Changed = true;
Expand Down
153 changes: 153 additions & 0 deletions llvm/test/CodeGen/X86/AMX/amx-tile-basic.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -51,3 +51,156 @@ declare x86_amx @llvm.x86.tdpbusd.internal(i16, i16, i16, x86_amx, x86_amx, x86_
declare x86_amx @llvm.x86.tdpbuud.internal(i16, i16, i16, x86_amx, x86_amx, x86_amx)
declare x86_amx @llvm.x86.tdpbf16ps.internal(i16, i16, i16, x86_amx, x86_amx, x86_amx)
declare void @llvm.x86.tilestored64.internal(i16, i16, ptr, i64, x86_amx)

define void @PR90954(ptr %0, ptr %1, i32 %2) {
; CHECK-LABEL: PR90954:
; CHECK: # %bb.0:
; CHECK-NEXT: pushq %rbp
; CHECK-NEXT: .cfi_def_cfa_offset 16
; CHECK-NEXT: .cfi_offset %rbp, -16
; CHECK-NEXT: movq %rsp, %rbp
; CHECK-NEXT: .cfi_def_cfa_register %rbp
; CHECK-NEXT: pushq %r15
; CHECK-NEXT: pushq %r14
; CHECK-NEXT: pushq %r13
; CHECK-NEXT: pushq %r12
; CHECK-NEXT: pushq %rbx
; CHECK-NEXT: andq $-1024, %rsp # imm = 0xFC00
; CHECK-NEXT: subq $5120, %rsp # imm = 0x1400
; CHECK-NEXT: .cfi_offset %rbx, -56
; CHECK-NEXT: .cfi_offset %r12, -48
; CHECK-NEXT: .cfi_offset %r13, -40
; CHECK-NEXT: .cfi_offset %r14, -32
; CHECK-NEXT: .cfi_offset %r15, -24
; CHECK-NEXT: vxorps %xmm0, %xmm0, %xmm0
; CHECK-NEXT: vmovups %zmm0, {{[0-9]+}}(%rsp)
; CHECK-NEXT: movb $1, {{[0-9]+}}(%rsp)
; CHECK-NEXT: movb $16, {{[0-9]+}}(%rsp)
; CHECK-NEXT: movw $64, {{[0-9]+}}(%rsp)
; CHECK-NEXT: movb $16, {{[0-9]+}}(%rsp)
; CHECK-NEXT: movw $64, {{[0-9]+}}(%rsp)
; CHECK-NEXT: movb $16, {{[0-9]+}}(%rsp)
; CHECK-NEXT: movw $64, {{[0-9]+}}(%rsp)
; CHECK-NEXT: ldtilecfg {{[0-9]+}}(%rsp)
; CHECK-NEXT: shll $4, %edx
; CHECK-NEXT: xorl %eax, %eax
; CHECK-NEXT: movw $64, %cx
; CHECK-NEXT: movw $16, %di
; CHECK-NEXT: movb $1, %r8b
; CHECK-NEXT: movl $64, %r9d
; CHECK-NEXT: leaq {{[0-9]+}}(%rsp), %r10
; CHECK-NEXT: leaq {{[0-9]+}}(%rsp), %r11
; CHECK-NEXT: xorl %ebx, %ebx
; CHECK-NEXT: xorl %r14d, %r14d
; CHECK-NEXT: jmp .LBB1_1
; CHECK-NEXT: .p2align 4, 0x90
; CHECK-NEXT: .LBB1_5: # in Loop: Header=BB1_1 Depth=1
; CHECK-NEXT: incq %r14
; CHECK-NEXT: addl %edx, %ebx
; CHECK-NEXT: .LBB1_1: # =>This Loop Header: Depth=1
; CHECK-NEXT: # Child Loop BB1_2 Depth 2
; CHECK-NEXT: movslq %ebx, %r15
; CHECK-NEXT: leaq (%rsi,%r15,4), %r15
; CHECK-NEXT: xorl %r12d, %r12d
; CHECK-NEXT: xorl %r13d, %r13d
; CHECK-NEXT: jmp .LBB1_2
; CHECK-NEXT: .p2align 4, 0x90
; CHECK-NEXT: .LBB1_4: # in Loop: Header=BB1_2 Depth=2
; CHECK-NEXT: tilestored %tmm1, (%r15,%rax)
; CHECK-NEXT: incq %r13
; CHECK-NEXT: addq $64, %r15
; CHECK-NEXT: decq %r12
; CHECK-NEXT: je .LBB1_5
; CHECK-NEXT: .LBB1_2: # Parent Loop BB1_1 Depth=1
; CHECK-NEXT: # => This Inner Loop Header: Depth=2
; CHECK-NEXT: tilezero %tmm0
; CHECK-NEXT: tilezero %tmm1
; CHECK-NEXT: testb %r8b, %r8b
; CHECK-NEXT: jne .LBB1_4
; CHECK-NEXT: # %bb.3: # in Loop: Header=BB1_2 Depth=2
; CHECK-NEXT: vmovaps %zmm0, {{[0-9]+}}(%rsp)
; CHECK-NEXT: vmovaps %zmm0, {{[0-9]+}}(%rsp)
; CHECK-NEXT: vmovaps %zmm0, {{[0-9]+}}(%rsp)
; CHECK-NEXT: vmovaps %zmm0, {{[0-9]+}}(%rsp)
; CHECK-NEXT: vmovaps %zmm0, {{[0-9]+}}(%rsp)
; CHECK-NEXT: vmovaps %zmm0, {{[0-9]+}}(%rsp)
; CHECK-NEXT: vmovaps %zmm0, {{[0-9]+}}(%rsp)
; CHECK-NEXT: vmovaps %zmm0, {{[0-9]+}}(%rsp)
; CHECK-NEXT: vmovaps %zmm0, {{[0-9]+}}(%rsp)
; CHECK-NEXT: vmovaps %zmm0, {{[0-9]+}}(%rsp)
; CHECK-NEXT: vmovaps %zmm0, {{[0-9]+}}(%rsp)
; CHECK-NEXT: vmovaps %zmm0, {{[0-9]+}}(%rsp)
; CHECK-NEXT: vmovaps %zmm0, {{[0-9]+}}(%rsp)
; CHECK-NEXT: vmovaps %zmm0, {{[0-9]+}}(%rsp)
; CHECK-NEXT: vmovaps %zmm0, {{[0-9]+}}(%rsp)
; CHECK-NEXT: vmovaps %zmm0, {{[0-9]+}}(%rsp)
; CHECK-NEXT: tileloadd (%r10,%r9), %tmm1
; CHECK-NEXT: vmovaps %zmm0, {{[0-9]+}}(%rsp)
; CHECK-NEXT: vmovaps %zmm0, {{[0-9]+}}(%rsp)
; CHECK-NEXT: vmovaps %zmm0, {{[0-9]+}}(%rsp)
; CHECK-NEXT: vmovaps %zmm0, {{[0-9]+}}(%rsp)
; CHECK-NEXT: vmovaps %zmm0, {{[0-9]+}}(%rsp)
; CHECK-NEXT: vmovaps %zmm0, {{[0-9]+}}(%rsp)
; CHECK-NEXT: vmovaps %zmm0, {{[0-9]+}}(%rsp)
; CHECK-NEXT: vmovaps %zmm0, {{[0-9]+}}(%rsp)
; CHECK-NEXT: vmovaps %zmm0, {{[0-9]+}}(%rsp)
; CHECK-NEXT: vmovaps %zmm0, {{[0-9]+}}(%rsp)
; CHECK-NEXT: vmovaps %zmm0, {{[0-9]+}}(%rsp)
; CHECK-NEXT: vmovaps %zmm0, {{[0-9]+}}(%rsp)
; CHECK-NEXT: vmovaps %zmm0, {{[0-9]+}}(%rsp)
; CHECK-NEXT: vmovaps %zmm0, {{[0-9]+}}(%rsp)
; CHECK-NEXT: vmovaps %zmm0, {{[0-9]+}}(%rsp)
; CHECK-NEXT: vmovaps %zmm0, {{[0-9]+}}(%rsp)
; CHECK-NEXT: tileloadd (%r11,%r9), %tmm2
; CHECK-NEXT: tdpbf16ps %tmm2, %tmm1, %tmm0
; CHECK-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
; CHECK-NEXT: movabsq $64, %rax
; CHECK-NEXT: tilestored %tmm0, 3072(%rsp,%rax) # 1024-byte Folded Spill
; CHECK-NEXT: tileloadd {{[-0-9]+}}(%r{{[sb]}}p), %tmm1 # 1024-byte Folded Reload
; CHECK-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
; CHECK-NEXT: jmp .LBB1_4
%4 = shl i32 %2, 4
%5 = icmp eq i64 0, 0
br label %6

6: ; preds = %31, %3
%7 = phi i64 [ 0, %3 ], [ %32, %31 ]
%8 = trunc nuw nsw i64 %7 to i32
%9 = mul i32 %4, %8
%10 = mul i32 0, %8
%11 = sext i32 %9 to i64
%12 = getelementptr inbounds i32, ptr %1, i64 %11
br label %13

13: ; preds = %25, %6
%14 = phi i64 [ %29, %25 ], [ 0, %6 ]
%15 = tail call x86_amx @llvm.x86.tilezero.internal(i16 16, i16 64)
%16 = tail call <256 x i32> @llvm.x86.cast.tile.to.vector.v256i32(x86_amx %15)
%17 = shl nsw i64 %14, 4
%18 = getelementptr i32, ptr %0, i64 %17
br i1 %5, label %25, label %19

19: ; preds = %13
%20 = tail call x86_amx @llvm.x86.cast.vector.to.tile.v256i32(<256 x i32> %16)
%21 = tail call x86_amx @llvm.x86.cast.vector.to.tile.v256i32(<256 x i32> zeroinitializer)
%22 = tail call x86_amx @llvm.x86.cast.vector.to.tile.v256i32(<256 x i32> zeroinitializer)
%23 = tail call x86_amx @llvm.x86.tdpbf16ps.internal(i16 16, i16 64, i16 64, x86_amx %20, x86_amx %21, x86_amx %22)
%24 = tail call noundef <256 x i32> @llvm.x86.cast.tile.to.vector.v256i32(x86_amx %23)
br label %25

25: ; preds = %19, %13
%26 = phi <256 x i32> [ undef, %13 ], [ %24, %19 ]
%27 = getelementptr inbounds i32, ptr %12, i64 %17
%28 = tail call x86_amx @llvm.x86.cast.vector.to.tile.v256i32(<256 x i32> %26)
tail call void @llvm.x86.tilestored64.internal(i16 16, i16 64, ptr %27, i64 0, x86_amx %28)
%29 = add nuw nsw i64 %14, 1
%30 = icmp eq i64 %29, 0
br i1 %30, label %31, label %13

31: ; preds = %25
%32 = add nuw nsw i64 %7, 1
br label %6
}

declare x86_amx @llvm.x86.cast.vector.to.tile.v256i32(<256 x i32>)
declare <256 x i32> @llvm.x86.cast.tile.to.vector.v256i32(x86_amx)
Loading