Skip to content

[AtomicExpandPass] Match isIdempotentRMW with InstcombineRMW #142277

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 4 commits into from
Jun 8, 2025
Merged

[AtomicExpandPass] Match isIdempotentRMW with InstcombineRMW #142277

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
0f73eee
Pre-commit tests (NFC)
AZero13 May 31, 2025
e3248f9
[AtomicExpandPass] Match isIdempotentRMW with InstcombineRMW
AZero13 May 31, 2025
c1e73c7
Address comments
AZero13 Jun 3, 2025
1573eb2
Remove copy-paste error
AZero13 Jun 5, 2025
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
13 changes: 10 additions & 3 deletions llvm/lib/CodeGen/AtomicExpandPass.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1570,20 +1570,27 @@ bool AtomicExpandImpl::expandAtomicCmpXchg(AtomicCmpXchgInst *CI) {
}

bool AtomicExpandImpl::isIdempotentRMW(AtomicRMWInst *RMWI) {
// TODO: Add floating point support.
auto C = dyn_cast<ConstantInt>(RMWI->getValOperand());
if (!C)
return false;

AtomicRMWInst::BinOp Op = RMWI->getOperation();
switch (Op) {
switch (RMWI->getOperation()) {
case AtomicRMWInst::Add:
case AtomicRMWInst::Sub:
case AtomicRMWInst::Or:
case AtomicRMWInst::Xor:
return C->isZero();
case AtomicRMWInst::And:
return C->isMinusOne();
// FIXME: we could also treat Min/Max/UMin/UMax by the INT_MIN/INT_MAX/...
case AtomicRMWInst::Min:
return C->isMaxValue(true);
case AtomicRMWInst::Max:
return C->isMinValue(true);
case AtomicRMWInst::UMin:
return C->isMaxValue(false);
case AtomicRMWInst::UMax:
return C->isMinValue(false);
default:
return false;
}
Expand Down
329 changes: 329 additions & 0 deletions llvm/test/CodeGen/X86/atomic-idempotent.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -622,4 +622,333 @@ define void @or8_nouse_seq_cst(ptr %p) #0 {
ret void
}

define void @atomic_umin_uint_max(ptr %addr) #0 {
; X64-LABEL: atomic_umin_uint_max:
; X64: # %bb.0:
; X64-NEXT: lock orl $0, -{{[0-9]+}}(%rsp)
; X64-NEXT: movl (%rdi), %eax
; X64-NEXT: retq
;
; X86-SSE2-LABEL: atomic_umin_uint_max:
; X86-SSE2: # %bb.0:
; X86-SSE2-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-SSE2-NEXT: mfence
; X86-SSE2-NEXT: movl (%eax), %eax
; X86-SSE2-NEXT: retl
;
; X86-SLM-LABEL: atomic_umin_uint_max:
; X86-SLM: # %bb.0:
; X86-SLM-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-SLM-NEXT: lock orl $0, (%esp)
; X86-SLM-NEXT: movl (%eax), %eax
; X86-SLM-NEXT: retl
;
; X86-ATOM-LABEL: atomic_umin_uint_max:
; X86-ATOM: # %bb.0:
; X86-ATOM-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-ATOM-NEXT: lock orl $0, (%esp)
; X86-ATOM-NEXT: movl (%eax), %eax
; X86-ATOM-NEXT: nop
; X86-ATOM-NEXT: nop
; X86-ATOM-NEXT: retl
atomicrmw umin ptr %addr, i32 -1 seq_cst
ret void
}

define void @atomic_umax_zero(ptr %addr) #0 {
; X64-LABEL: atomic_umax_zero:
; X64: # %bb.0:
; X64-NEXT: lock orl $0, -{{[0-9]+}}(%rsp)
; X64-NEXT: movl (%rdi), %eax
; X64-NEXT: retq
;
; X86-SSE2-LABEL: atomic_umax_zero:
; X86-SSE2: # %bb.0:
; X86-SSE2-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-SSE2-NEXT: mfence
; X86-SSE2-NEXT: movl (%eax), %eax
; X86-SSE2-NEXT: retl
;
; X86-SLM-LABEL: atomic_umax_zero:
; X86-SLM: # %bb.0:
; X86-SLM-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-SLM-NEXT: lock orl $0, (%esp)
; X86-SLM-NEXT: movl (%eax), %eax
; X86-SLM-NEXT: retl
;
; X86-ATOM-LABEL: atomic_umax_zero:
; X86-ATOM: # %bb.0:
; X86-ATOM-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-ATOM-NEXT: lock orl $0, (%esp)
; X86-ATOM-NEXT: movl (%eax), %eax
; X86-ATOM-NEXT: nop
; X86-ATOM-NEXT: nop
; X86-ATOM-NEXT: retl
atomicrmw umax ptr %addr, i32 0 seq_cst
ret void
}

define void @atomic_min_smax_char(ptr %addr) #0 {
; X64-LABEL: atomic_min_smax_char:
; X64: # %bb.0:
; X64-NEXT: lock orl $0, -{{[0-9]+}}(%rsp)
; X64-NEXT: movzbl (%rdi), %eax
; X64-NEXT: retq
;
; X86-SSE2-LABEL: atomic_min_smax_char:
; X86-SSE2: # %bb.0:
; X86-SSE2-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-SSE2-NEXT: mfence
; X86-SSE2-NEXT: movzbl (%eax), %eax
; X86-SSE2-NEXT: retl
;
; X86-SLM-LABEL: atomic_min_smax_char:
; X86-SLM: # %bb.0:
; X86-SLM-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-SLM-NEXT: lock orl $0, (%esp)
; X86-SLM-NEXT: movzbl (%eax), %eax
; X86-SLM-NEXT: retl
;
; X86-ATOM-LABEL: atomic_min_smax_char:
; X86-ATOM: # %bb.0:
; X86-ATOM-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-ATOM-NEXT: lock orl $0, (%esp)
; X86-ATOM-NEXT: movzbl (%eax), %eax
; X86-ATOM-NEXT: nop
; X86-ATOM-NEXT: nop
; X86-ATOM-NEXT: retl
atomicrmw min ptr %addr, i8 127 seq_cst
ret void
}

define void @atomic_max_smin_char(ptr %addr) #0 {
; X64-LABEL: atomic_max_smin_char:
; X64: # %bb.0:
; X64-NEXT: lock orl $0, -{{[0-9]+}}(%rsp)
; X64-NEXT: movzbl (%rdi), %eax
; X64-NEXT: retq
;
; X86-SSE2-LABEL: atomic_max_smin_char:
; X86-SSE2: # %bb.0:
; X86-SSE2-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-SSE2-NEXT: mfence
; X86-SSE2-NEXT: movzbl (%eax), %eax
; X86-SSE2-NEXT: retl
;
; X86-SLM-LABEL: atomic_max_smin_char:
; X86-SLM: # %bb.0:
; X86-SLM-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-SLM-NEXT: lock orl $0, (%esp)
; X86-SLM-NEXT: movzbl (%eax), %eax
; X86-SLM-NEXT: retl
;
; X86-ATOM-LABEL: atomic_max_smin_char:
; X86-ATOM: # %bb.0:
; X86-ATOM-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-ATOM-NEXT: lock orl $0, (%esp)
; X86-ATOM-NEXT: movzbl (%eax), %eax
; X86-ATOM-NEXT: nop
; X86-ATOM-NEXT: nop
; X86-ATOM-NEXT: retl
atomicrmw max ptr %addr, i8 -128 seq_cst
ret void
}

define void @atomic_min_umax_char(ptr %addr) #0 {
; X64-LABEL: atomic_min_umax_char:
; X64: # %bb.0:
; X64-NEXT: lock orl $0, -{{[0-9]+}}(%rsp)
; X64-NEXT: movzbl (%rdi), %eax
; X64-NEXT: retq
;
; X86-SSE2-LABEL: atomic_min_umax_char:
; X86-SSE2: # %bb.0:
; X86-SSE2-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-SSE2-NEXT: mfence
; X86-SSE2-NEXT: movzbl (%eax), %eax
; X86-SSE2-NEXT: retl
;
; X86-SLM-LABEL: atomic_min_umax_char:
; X86-SLM: # %bb.0:
; X86-SLM-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-SLM-NEXT: lock orl $0, (%esp)
; X86-SLM-NEXT: movzbl (%eax), %eax
; X86-SLM-NEXT: retl
;
; X86-ATOM-LABEL: atomic_min_umax_char:
; X86-ATOM: # %bb.0:
; X86-ATOM-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-ATOM-NEXT: lock orl $0, (%esp)
; X86-ATOM-NEXT: movzbl (%eax), %eax
; X86-ATOM-NEXT: nop
; X86-ATOM-NEXT: nop
; X86-ATOM-NEXT: retl
atomicrmw umin ptr %addr, i8 255 seq_cst
ret void
}

define void @atomic_max_umin_char(ptr %addr) #0 {
; X64-LABEL: atomic_max_umin_char:
; X64: # %bb.0:
; X64-NEXT: lock orl $0, -{{[0-9]+}}(%rsp)
; X64-NEXT: movzbl (%rdi), %eax
; X64-NEXT: retq
;
; X86-SSE2-LABEL: atomic_max_umin_char:
; X86-SSE2: # %bb.0:
; X86-SSE2-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-SSE2-NEXT: mfence
; X86-SSE2-NEXT: movzbl (%eax), %eax
; X86-SSE2-NEXT: retl
;
; X86-SLM-LABEL: atomic_max_umin_char:
; X86-SLM: # %bb.0:
; X86-SLM-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-SLM-NEXT: lock orl $0, (%esp)
; X86-SLM-NEXT: movzbl (%eax), %eax
; X86-SLM-NEXT: retl
;
; X86-ATOM-LABEL: atomic_max_umin_char:
; X86-ATOM: # %bb.0:
; X86-ATOM-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-ATOM-NEXT: lock orl $0, (%esp)
; X86-ATOM-NEXT: movzbl (%eax), %eax
; X86-ATOM-NEXT: nop
; X86-ATOM-NEXT: nop
; X86-ATOM-NEXT: retl
atomicrmw umax ptr %addr, i8 0 seq_cst
ret void
}

; TODO: Add floating point support.
define void @atomic_fadd_zero(ptr %addr) #0 {
; X64-LABEL: atomic_fadd_zero:
; X64: # %bb.0:
; X64-NEXT: movl (%rdi), %eax
; X64-NEXT: .p2align 4
; X64-NEXT: .LBB21_1: # %atomicrmw.start
; X64-NEXT: # =>This Inner Loop Header: Depth=1
; X64-NEXT: lock cmpxchgl %eax, (%rdi)
; X64-NEXT: jne .LBB21_1
; X64-NEXT: # %bb.2: # %atomicrmw.end
; X64-NEXT: retq
;
; X86-SSE2-LABEL: atomic_fadd_zero:
; X86-SSE2: # %bb.0:
; X86-SSE2-NEXT: movl {{[0-9]+}}(%esp), %ecx
; X86-SSE2-NEXT: movd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; X86-SSE2-NEXT: .p2align 4
; X86-SSE2-NEXT: .LBB21_1: # %atomicrmw.start
; X86-SSE2-NEXT: # =>This Inner Loop Header: Depth=1
; X86-SSE2-NEXT: movd %xmm0, %eax
; X86-SSE2-NEXT: lock cmpxchgl %eax, (%ecx)
; X86-SSE2-NEXT: movd %eax, %xmm0
; X86-SSE2-NEXT: jne .LBB21_1
; X86-SSE2-NEXT: # %bb.2: # %atomicrmw.end
; X86-SSE2-NEXT: retl
;
; X86-SLM-LABEL: atomic_fadd_zero:
; X86-SLM: # %bb.0:
; X86-SLM-NEXT: subl $8, %esp
; X86-SLM-NEXT: movl {{[0-9]+}}(%esp), %ecx
; X86-SLM-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; X86-SLM-NEXT: .p2align 4
; X86-SLM-NEXT: .LBB21_1: # %atomicrmw.start
; X86-SLM-NEXT: # =>This Inner Loop Header: Depth=1
; X86-SLM-NEXT: movss %xmm0, (%esp)
; X86-SLM-NEXT: movl (%esp), %eax
; X86-SLM-NEXT: lock cmpxchgl %eax, (%ecx)
; X86-SLM-NEXT: movl %eax, {{[0-9]+}}(%esp)
; X86-SLM-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; X86-SLM-NEXT: jne .LBB21_1
; X86-SLM-NEXT: # %bb.2: # %atomicrmw.end
; X86-SLM-NEXT: addl $8, %esp
; X86-SLM-NEXT: retl
;
; X86-ATOM-LABEL: atomic_fadd_zero:
; X86-ATOM: # %bb.0:
; X86-ATOM-NEXT: leal -{{[0-9]+}}(%esp), %esp
; X86-ATOM-NEXT: movl {{[0-9]+}}(%esp), %ecx
; X86-ATOM-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; X86-ATOM-NEXT: .p2align 4
; X86-ATOM-NEXT: .LBB21_1: # %atomicrmw.start
; X86-ATOM-NEXT: # =>This Inner Loop Header: Depth=1
; X86-ATOM-NEXT: movss %xmm0, (%esp)
; X86-ATOM-NEXT: movl (%esp), %eax
; X86-ATOM-NEXT: lock cmpxchgl %eax, (%ecx)
; X86-ATOM-NEXT: movl %eax, {{[0-9]+}}(%esp)
; X86-ATOM-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; X86-ATOM-NEXT: jne .LBB21_1
; X86-ATOM-NEXT: # %bb.2: # %atomicrmw.end
; X86-ATOM-NEXT: leal {{[0-9]+}}(%esp), %esp
; X86-ATOM-NEXT: retl
atomicrmw fadd ptr %addr, float -0.0 monotonic
ret void
}

define void @atomic_fsub_zero(ptr %addr) #0 {
; X64-LABEL: atomic_fsub_zero:
; X64: # %bb.0:
; X64-NEXT: movl (%rdi), %eax
; X64-NEXT: .p2align 4
; X64-NEXT: .LBB22_1: # %atomicrmw.start
; X64-NEXT: # =>This Inner Loop Header: Depth=1
; X64-NEXT: lock cmpxchgl %eax, (%rdi)
; X64-NEXT: jne .LBB22_1
; X64-NEXT: # %bb.2: # %atomicrmw.end
; X64-NEXT: retq
;
; X86-SSE2-LABEL: atomic_fsub_zero:
; X86-SSE2: # %bb.0:
; X86-SSE2-NEXT: movl {{[0-9]+}}(%esp), %ecx
; X86-SSE2-NEXT: movd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; X86-SSE2-NEXT: .p2align 4
; X86-SSE2-NEXT: .LBB22_1: # %atomicrmw.start
; X86-SSE2-NEXT: # =>This Inner Loop Header: Depth=1
; X86-SSE2-NEXT: movd %xmm0, %eax
; X86-SSE2-NEXT: lock cmpxchgl %eax, (%ecx)
; X86-SSE2-NEXT: movd %eax, %xmm0
; X86-SSE2-NEXT: jne .LBB22_1
; X86-SSE2-NEXT: # %bb.2: # %atomicrmw.end
; X86-SSE2-NEXT: retl
;
; X86-SLM-LABEL: atomic_fsub_zero:
; X86-SLM: # %bb.0:
; X86-SLM-NEXT: subl $8, %esp
; X86-SLM-NEXT: movl {{[0-9]+}}(%esp), %ecx
; X86-SLM-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; X86-SLM-NEXT: .p2align 4
; X86-SLM-NEXT: .LBB22_1: # %atomicrmw.start
; X86-SLM-NEXT: # =>This Inner Loop Header: Depth=1
; X86-SLM-NEXT: movss %xmm0, (%esp)
; X86-SLM-NEXT: movl (%esp), %eax
; X86-SLM-NEXT: lock cmpxchgl %eax, (%ecx)
; X86-SLM-NEXT: movl %eax, {{[0-9]+}}(%esp)
; X86-SLM-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; X86-SLM-NEXT: jne .LBB22_1
; X86-SLM-NEXT: # %bb.2: # %atomicrmw.end
; X86-SLM-NEXT: addl $8, %esp
; X86-SLM-NEXT: retl
;
; X86-ATOM-LABEL: atomic_fsub_zero:
; X86-ATOM: # %bb.0:
; X86-ATOM-NEXT: leal -{{[0-9]+}}(%esp), %esp
; X86-ATOM-NEXT: movl {{[0-9]+}}(%esp), %ecx
; X86-ATOM-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; X86-ATOM-NEXT: .p2align 4
; X86-ATOM-NEXT: .LBB22_1: # %atomicrmw.start
; X86-ATOM-NEXT: # =>This Inner Loop Header: Depth=1
; X86-ATOM-NEXT: movss %xmm0, (%esp)
; X86-ATOM-NEXT: movl (%esp), %eax
; X86-ATOM-NEXT: lock cmpxchgl %eax, (%ecx)
; X86-ATOM-NEXT: movl %eax, {{[0-9]+}}(%esp)
; X86-ATOM-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; X86-ATOM-NEXT: jne .LBB22_1
; X86-ATOM-NEXT: # %bb.2: # %atomicrmw.end
; X86-ATOM-NEXT: leal {{[0-9]+}}(%esp), %esp
; X86-ATOM-NEXT: retl
atomicrmw fsub ptr %addr, float 0.0 release
ret void
}

attributes #0 = { nounwind }
Loading