[llvm] 78281fd - Revert "[AMDGPU] InstCombine llvm.amdgcn.ds.bpermute with uniform arguments (#129895)"

[Jay Foad via llvm-commits]

Author: Jay Foad
Date: 2025-03-06T15:05:19Z
New Revision: 78281fd12c9f5c46b95d9399ae72ca357812ecee

URL: https://github.com/llvm/llvm-project/commit/78281fd12c9f5c46b95d9399ae72ca357812ecee
DIFF: https://github.com/llvm/llvm-project/commit/78281fd12c9f5c46b95d9399ae72ca357812ecee.diff

LOG: Revert "[AMDGPU] InstCombine llvm.amdgcn.ds.bpermute with uniform arguments (#129895)"

This reverts commit be5149a3158cbce3051629e450950ccb96926365.

It caused build failures in the openmp-offload-amdgpu-runtime buildbot
and others.

Added: 
    

Modified: 
    llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp
    llvm/test/Transforms/InstCombine/AMDGPU/amdgcn-intrinsics.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp b/llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp
index 8925f2e974efa..ebe740f884ea6 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp
@@ -1118,11 +1118,9 @@ GCNTTIImpl::instCombineIntrinsic(InstCombiner &IC, IntrinsicInst &II) const {
   }
   case Intrinsic::amdgcn_permlane64:
   case Intrinsic::amdgcn_readfirstlane:
-  case Intrinsic::amdgcn_readlane:
-  case Intrinsic::amdgcn_ds_bpermute: {
-    // If the data argument is uniform these intrinsics return it unchanged.
-    unsigned SrcIdx = IID == Intrinsic::amdgcn_ds_bpermute ? 1 : 0;
-    const Use &Src = II.getArgOperandUse(SrcIdx);
+  case Intrinsic::amdgcn_readlane: {
+    // If the first argument is uniform these intrinsics return it unchanged.
+    const Use &Src = II.getArgOperandUse(0);
     if (isTriviallyUniform(Src))
       return IC.replaceInstUsesWith(II, Src.get());
 
@@ -1154,22 +1152,6 @@ GCNTTIImpl::instCombineIntrinsic(InstCombiner &IC, IntrinsicInst &II) const {
       }
     }
 
-    // If the lane argument of bpermute is uniform, change it to readlane. This
-    // generates better code and can enable further optimizations because
-    // readlane is AlwaysUniform.
-    if (IID == Intrinsic::amdgcn_ds_bpermute) {
-      const Use &Lane = II.getArgOperandUse(0);
-      if (isTriviallyUniform(Lane)) {
-        Value *NewLane = IC.Builder.CreateLShr(Lane, 2);
-        Function *NewDecl = Intrinsic::getOrInsertDeclaration(
-            II.getModule(), Intrinsic::amdgcn_readlane, II.getType());
-        II.setCalledFunction(NewDecl);
-        II.setOperand(0, Src);
-        II.setOperand(1, NewLane);
-        return ⅈ
-      }
-    }
-
     return std::nullopt;
   }
   case Intrinsic::amdgcn_writelane: {

diff  --git a/llvm/test/Transforms/InstCombine/AMDGPU/amdgcn-intrinsics.ll b/llvm/test/Transforms/InstCombine/AMDGPU/amdgcn-intrinsics.ll
index 843b436aa1b0f..3c190efca7acf 100644
--- a/llvm/test/Transforms/InstCombine/AMDGPU/amdgcn-intrinsics.ll
+++ b/llvm/test/Transforms/InstCombine/AMDGPU/amdgcn-intrinsics.ll
@@ -6583,42 +6583,3 @@ define i32 @prng_poison_i32() {
   %prng = call i32 @llvm.amdgcn.prng.b32(i32 poison)
   ret i32 %prng
 }
-
-; --------------------------------------------------------------------
-; llvm.amdgcn.ds.bpermute
-; --------------------------------------------------------------------
-
-define amdgpu_kernel void @ds_bpermute_uniform_src(ptr addrspace(1) %out, i32 %lane) {
-; CHECK-LABEL: @ds_bpermute_uniform_src(
-; CHECK-NEXT:    store i32 7, ptr addrspace(1) [[OUT:%.*]], align 4
-; CHECK-NEXT:    ret void
-;
-  %v = call i32 @llvm.amdgcn.ds.bpermute(i32 %lane, i32 7)
-  store i32 %v, ptr addrspace(1) %out
-  ret void
-}
-
-define amdgpu_kernel void @ds_bpermute_constant_lane(ptr addrspace(1) %out, i32 %src) {
-; CHECK-LABEL: @ds_bpermute_constant_lane(
-; CHECK-NEXT:    [[V:%.*]] = call i32 @llvm.amdgcn.readlane.i32(i32 [[SRC:%.*]], i32 7)
-; CHECK-NEXT:    store i32 [[V]], ptr addrspace(1) [[OUT:%.*]], align 4
-; CHECK-NEXT:    ret void
-;
-  %v = call i32 @llvm.amdgcn.ds.bpermute(i32 28, i32 %src)
-  store i32 %v, ptr addrspace(1) %out
-  ret void
-}
-
-define amdgpu_kernel void @ds_bpermute_uniform_lane(ptr addrspace(1) %out, i32 %lanearg, i32 %src) {
-; CHECK-LABEL: @ds_bpermute_uniform_lane(
-; CHECK-NEXT:    [[LANE:%.*]] = call i32 @llvm.amdgcn.readfirstlane.i32(i32 [[LANEARG:%.*]])
-; CHECK-NEXT:    [[TMP1:%.*]] = lshr i32 [[LANE]], 2
-; CHECK-NEXT:    [[V:%.*]] = call i32 @llvm.amdgcn.readlane.i32(i32 [[SRC:%.*]], i32 [[TMP1]])
-; CHECK-NEXT:    store i32 [[V]], ptr addrspace(1) [[OUT:%.*]], align 4
-; CHECK-NEXT:    ret void
-;
-  %lane = call i32 @llvm.amdgcn.readfirstlane(i32 %lanearg)
-  %v = call i32 @llvm.amdgcn.ds.bpermute(i32 %lane, i32 %src)
-  store i32 %v, ptr addrspace(1) %out
-  ret void
-}