[AtomicExpandPass] Match isIdempotentRMW with InstcombineRMW

AZero13 · AZero13 · commit e3248f96760a · 2025-05-31T12:04:30.000-04:00
Add umin, smin, umax, smax to isIdempotentRMW
diff --git a/llvm/lib/CodeGen/AtomicExpandPass.cpp b/llvm/lib/CodeGen/AtomicExpandPass.cpp
@@ -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;
   }
diff --git a/llvm/test/CodeGen/X86/atomic-idempotent.ll b/llvm/test/CodeGen/X86/atomic-idempotent.ll
@@ -629,26 +629,32 @@ define void @atomic_umin_uint_max(ptr %addr) {
 ;
 ; X64-LABEL: atomic_umin_uint_max:
 ; X64:       # %bb.0:
+; X64-NEXT:    lock orl $0, -{{[0-9]+}}(%rsp)
 ; X64-NEXT:    movl (%rdi), %eax
-; X64-NEXT:    .p2align 4
-; X64-NEXT:  .LBB15_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    lock cmpxchgl %eax, (%rdi)
-; X64-NEXT:    jne .LBB15_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
 ; X64-NEXT:    retq
 ;
-; X86-LABEL: atomic_umin_uint_max:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movl (%ecx), %eax
-; X86-NEXT:    .p2align 4
-; X86-NEXT:  .LBB15_1: # %atomicrmw.start
-; X86-NEXT:    # =>This Inner Loop Header: Depth=1
-; X86-NEXT:    lock cmpxchgl %eax, (%ecx)
-; X86-NEXT:    jne .LBB15_1
-; X86-NEXT:  # %bb.2: # %atomicrmw.end
-; X86-NEXT:    retl
+; 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
 }
@@ -660,26 +666,32 @@ define void @atomic_umax_zero(ptr %addr) {
 ;
 ; X64-LABEL: atomic_umax_zero:
 ; X64:       # %bb.0:
+; X64-NEXT:    lock orl $0, -{{[0-9]+}}(%rsp)
 ; X64-NEXT:    movl (%rdi), %eax
-; X64-NEXT:    .p2align 4
-; X64-NEXT:  .LBB16_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    lock cmpxchgl %eax, (%rdi)
-; X64-NEXT:    jne .LBB16_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
 ; X64-NEXT:    retq
 ;
-; X86-LABEL: atomic_umax_zero:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movl (%ecx), %eax
-; X86-NEXT:    .p2align 4
-; X86-NEXT:  .LBB16_1: # %atomicrmw.start
-; X86-NEXT:    # =>This Inner Loop Header: Depth=1
-; X86-NEXT:    lock cmpxchgl %eax, (%ecx)
-; X86-NEXT:    jne .LBB16_1
-; X86-NEXT:  # %bb.2: # %atomicrmw.end
-; X86-NEXT:    retl
+; 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
 }
@@ -691,26 +703,32 @@ define void @atomic_min_smax_char(ptr %addr) {
 ;
 ; X64-LABEL: atomic_min_smax_char:
 ; X64:       # %bb.0:
+; X64-NEXT:    lock orl $0, -{{[0-9]+}}(%rsp)
 ; X64-NEXT:    movzbl (%rdi), %eax
-; X64-NEXT:    .p2align 4
-; X64-NEXT:  .LBB17_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    lock cmpxchgb %al, (%rdi)
-; X64-NEXT:    jne .LBB17_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
 ; X64-NEXT:    retq
 ;
-; X86-LABEL: atomic_min_smax_char:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movzbl (%ecx), %eax
-; X86-NEXT:    .p2align 4
-; X86-NEXT:  .LBB17_1: # %atomicrmw.start
-; X86-NEXT:    # =>This Inner Loop Header: Depth=1
-; X86-NEXT:    lock cmpxchgb %al, (%ecx)
-; X86-NEXT:    jne .LBB17_1
-; X86-NEXT:  # %bb.2: # %atomicrmw.end
-; X86-NEXT:    retl
+; 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
 }
@@ -722,26 +740,32 @@ define void @atomic_max_smin_char(ptr %addr) {
 ;
 ; X64-LABEL: atomic_max_smin_char:
 ; X64:       # %bb.0:
+; X64-NEXT:    lock orl $0, -{{[0-9]+}}(%rsp)
 ; X64-NEXT:    movzbl (%rdi), %eax
-; X64-NEXT:    .p2align 4
-; X64-NEXT:  .LBB18_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    lock cmpxchgb %al, (%rdi)
-; X64-NEXT:    jne .LBB18_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
 ; X64-NEXT:    retq
 ;
-; X86-LABEL: atomic_max_smin_char:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movzbl (%ecx), %eax
-; X86-NEXT:    .p2align 4
-; X86-NEXT:  .LBB18_1: # %atomicrmw.start
-; X86-NEXT:    # =>This Inner Loop Header: Depth=1
-; X86-NEXT:    lock cmpxchgb %al, (%ecx)
-; X86-NEXT:    jne .LBB18_1
-; X86-NEXT:  # %bb.2: # %atomicrmw.end
-; X86-NEXT:    retl
+; 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
 }
@@ -753,26 +777,32 @@ define void @atomic_min_umax_char(ptr %addr) {
 ;
 ; X64-LABEL: atomic_min_umax_char:
 ; X64:       # %bb.0:
+; X64-NEXT:    lock orl $0, -{{[0-9]+}}(%rsp)
 ; X64-NEXT:    movzbl (%rdi), %eax
-; X64-NEXT:    .p2align 4
-; X64-NEXT:  .LBB19_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    lock cmpxchgb %al, (%rdi)
-; X64-NEXT:    jne .LBB19_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
 ; X64-NEXT:    retq
 ;
-; X86-LABEL: atomic_min_umax_char:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movzbl (%ecx), %eax
-; X86-NEXT:    .p2align 4
-; X86-NEXT:  .LBB19_1: # %atomicrmw.start
-; X86-NEXT:    # =>This Inner Loop Header: Depth=1
-; X86-NEXT:    lock cmpxchgb %al, (%ecx)
-; X86-NEXT:    jne .LBB19_1
-; X86-NEXT:  # %bb.2: # %atomicrmw.end
-; X86-NEXT:    retl
+; 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
 }
@@ -784,30 +814,37 @@ define void @atomic_max_umin_char(ptr %addr) {
 ;
 ; X64-LABEL: atomic_max_umin_char:
 ; X64:       # %bb.0:
+; X64-NEXT:    lock orl $0, -{{[0-9]+}}(%rsp)
 ; X64-NEXT:    movzbl (%rdi), %eax
-; X64-NEXT:    .p2align 4
-; X64-NEXT:  .LBB20_1: # %atomicrmw.start
-; X64-NEXT:    # =>This Inner Loop Header: Depth=1
-; X64-NEXT:    lock cmpxchgb %al, (%rdi)
-; X64-NEXT:    jne .LBB20_1
-; X64-NEXT:  # %bb.2: # %atomicrmw.end
 ; X64-NEXT:    retq
 ;
-; X86-LABEL: atomic_max_umin_char:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movzbl (%ecx), %eax
-; X86-NEXT:    .p2align 4
-; X86-NEXT:  .LBB20_1: # %atomicrmw.start
-; X86-NEXT:    # =>This Inner Loop Header: Depth=1
-; X86-NEXT:    lock cmpxchgb %al, (%ecx)
-; X86-NEXT:    jne .LBB20_1
-; X86-NEXT:  # %bb.2: # %atomicrmw.end
-; X86-NEXT:    retl
+; 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) {
 ; CHECK-LABEL: @atomic_fadd_zero(
 ; CHECK-NEXT:    [[RES:%.*]] = atomicrmw fadd ptr [[ADDR:%.*]], float -0.000000e+00 monotonic, align 4
