[llvm] 543f948 - [AMDGPU] Preserve exact flag for lshr (#146744)

[via llvm-commits]

Author: LU-JOHN
Date: 2025-07-07T23:26:45+09:00
New Revision: 543f94831245497b675dadb6a20ebc309d289465

URL: https://github.com/llvm/llvm-project/commit/543f94831245497b675dadb6a20ebc309d289465
DIFF: https://github.com/llvm/llvm-project/commit/543f94831245497b675dadb6a20ebc309d289465.diff

LOG: [AMDGPU] Preserve exact flag for lshr (#146744)

When reducing 64-bit lshr to 32-bit preserve exact flag.

Alive2 verification:  https://alive2.llvm.org/ce/z/LcnX7V

---------

Signed-off-by: John Lu <John.Lu at amd.com>

Added: 
    

Modified: 
    llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
    llvm/test/CodeGen/AMDGPU/srl64_reduce_flags.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
index c4a917d93cba1..e32accaba85fc 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
@@ -4328,7 +4328,8 @@ SDValue AMDGPUTargetLowering::performSrlCombine(SDNode *N,
     Hi = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, LHSSL, TargetType, SplitLHS, One);
   }
 
-  SDValue NewShift = DAG.getNode(ISD::SRL, SL, TargetType, Hi, ShiftAmt);
+  SDValue NewShift =
+      DAG.getNode(ISD::SRL, SL, TargetType, Hi, ShiftAmt, N->getFlags());
 
   SDValue Vec;
   if (VT.isVector()) {

diff  --git a/llvm/test/CodeGen/AMDGPU/srl64_reduce_flags.ll b/llvm/test/CodeGen/AMDGPU/srl64_reduce_flags.ll
index c79a94b837a31..ca4b72872a03b 100644
--- a/llvm/test/CodeGen/AMDGPU/srl64_reduce_flags.ll
+++ b/llvm/test/CodeGen/AMDGPU/srl64_reduce_flags.ll
@@ -30,7 +30,7 @@ define i64 @srl_exact(i64 %arg0, i64 %shift_amt) {
   ; CHECK-NEXT:   [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[DEF3]]
   ; CHECK-NEXT:   [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, killed [[COPY4]], %subreg.sub1
   ; CHECK-NEXT:   [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub0
-  ; CHECK-NEXT:   [[V_LSHRREV_B32_e64_:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_e64 killed [[COPY5]], killed [[COPY3]], implicit $exec
+  ; CHECK-NEXT:   [[V_LSHRREV_B32_e64_:%[0-9]+]]:vgpr_32 = exact V_LSHRREV_B32_e64 killed [[COPY5]], killed [[COPY3]], implicit $exec
   ; CHECK-NEXT:   [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
   ; CHECK-NEXT:   $vgpr0 = COPY [[V_LSHRREV_B32_e64_]]
   ; CHECK-NEXT:   $vgpr1 = COPY [[V_MOV_B32_e32_]]