Add umin, smin, umax, smax to isIdempotentRMW

Author: AZero13

llvm/lib/CodeGen/AtomicExpandPass.cpp

@@ -1580,10 +1580,16 @@ bool AtomicExpandImpl::isIdempotentRMW(AtomicRMWInst *RMWI) {
   case AtomicRMWInst::Sub:
   case AtomicRMWInst::Or:
   case AtomicRMWInst::Xor:
+  case AtomicRMWInst::UMax: // umax(x, 0) == x
     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::Max: // max(x, INT_MIN) == x
+    return C->isMinValue(/*isSigned=*/true);
+  case AtomicRMWInst::Min: // min(x, INT_MAX) == x
+    return C->isMaxValue(/*isSigned=*/true);
+  case AtomicRMWInst::UMin: // umin(x, UINT_MAX) == x
+    return C->isMaxValue(/*isSigned=*/false);
   default:
     return false;
   }

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
 }
@@ -784,26 +814,32 @@ 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
 }