[llvm] 81279bf - [X86] Fix missing check of rotate <-> shift equivilence (Issue 108722)

Noah Goldstein via llvm-commits llvm-commits at lists.llvm.org
Sun Sep 15 16:15:20 PDT 2024


Author: Noah Goldstein
Date: 2024-09-15T16:15:05-07:00
New Revision: 81279bf97f187eee0446af00d8ae9ec32a22e878

URL: https://github.com/llvm/llvm-project/commit/81279bf97f187eee0446af00d8ae9ec32a22e878
DIFF: https://github.com/llvm/llvm-project/commit/81279bf97f187eee0446af00d8ae9ec32a22e878.diff

LOG: [X86] Fix missing check of rotate <-> shift equivilence (Issue 108722)

Previous code was checking that rotate and shift where equivilent when
transforming shift -> rotate but not the other way around.

Closes #108767

Added: 
    

Modified: 
    llvm/lib/Target/X86/X86ISelLowering.cpp
    llvm/test/CodeGen/X86/cmp-shiftX-maskX.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 053b2354f6aeb6..d73c49124315a9 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -3489,7 +3489,7 @@ unsigned X86TargetLowering::preferedOpcodeForCmpEqPiecesOfOperand(
 
   // We prefer rotate for vectors of if we won't get a zext mask with SRL
   // (PreferRotate will be set in the latter case).
-  if (PreferRotate || VT.isVector())
+  if (PreferRotate || !MayTransformRotate || VT.isVector())
     return ShiftOpc;
 
   // Non-vector type and we have a zext mask with SRL.

diff  --git a/llvm/test/CodeGen/X86/cmp-shiftX-maskX.ll b/llvm/test/CodeGen/X86/cmp-shiftX-maskX.ll
index 799b5417857575..227de9ad0ab690 100644
--- a/llvm/test/CodeGen/X86/cmp-shiftX-maskX.ll
+++ b/llvm/test/CodeGen/X86/cmp-shiftX-maskX.ll
@@ -997,8 +997,8 @@ define i1 @shr_to_rotate_eq_i32_s5(i32 %x) {
 define i32 @issue108722(i32 %0) {
 ; CHECK-NOBMI-LABEL: issue108722:
 ; CHECK-NOBMI:       # %bb.0:
-; CHECK-NOBMI-NEXT:    movzbl %dil, %ecx
-; CHECK-NOBMI-NEXT:    shrl $24, %edi
+; CHECK-NOBMI-NEXT:    movl %edi, %ecx
+; CHECK-NOBMI-NEXT:    roll $24, %ecx
 ; CHECK-NOBMI-NEXT:    xorl %eax, %eax
 ; CHECK-NOBMI-NEXT:    cmpl %edi, %ecx
 ; CHECK-NOBMI-NEXT:    sete %al


        


More information about the llvm-commits mailing list