[llvm] [AMDGPU] Preserve exact flag for lshr (PR #146744)

[via llvm-commits]

https://github.com/LU-JOHN updated https://github.com/llvm/llvm-project/pull/146744

>From 9125754fb380d6acae0abfea6d82bfe60b3b2582 Mon Sep 17 00:00:00 2001
From: John Lu <John.Lu at amd.com>
Date: Wed, 2 Jul 2025 10:08:01 -0500
Subject: [PATCH 1/2] Preserve exact flag for lshr

Signed-off-by: John Lu <John.Lu at amd.com>
---
 llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
index d75c7a178b4a8..8f9c0530e09c5 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()) {

>From 20867d047ad367696601e639c4aa5746249ee254 Mon Sep 17 00:00:00 2001
From: John Lu <John.Lu at amd.com>
Date: Mon, 7 Jul 2025 08:26:29 -0500
Subject: [PATCH 2/2] Update flags in test

Signed-off-by: John Lu <John.Lu at amd.com>
---
 llvm/test/CodeGen/AMDGPU/srl64_reduce_flags.ll | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

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_]]