[llvm] [LV] Don't treat the pointer operand of histograms as scalar uses (PR #215769)

Wed Aug 12 03:21:35 PDT 2026

llvmorg-github-actions[bot] wrote:




@llvm/pr-subscribers-llvm-transforms

Author: Benjamin Maxwell (MacDue)

<details>
<summary>Changes</summary>



---
Full diff: https://github.com/llvm/llvm-project/pull/215769.diff


3 Files Affected:

- (modified) llvm/lib/Transforms/Vectorize/LoopVectorize.cpp (+4-3) 
- (modified) llvm/test/Transforms/LoopVectorize/AArch64/sve2-histcnt-epilogue.ll (+1-6) 
- (modified) llvm/test/Transforms/LoopVectorize/histograms.ll (+7-18) 


``````````diff

diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
index d892ff2a0fa2b..c4cae190e9f63 100644
--- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -2228,8 +2228,8 @@ void LoopVectorizationCostModel::collectLoopScalars(ElementCount VF) {
 
   // A helper that returns true if the use of Ptr by MemAccess will be scalar.
   // The pointer operands of loads and stores will be scalar as long as the
-  // memory access is not a gather or scatter operation. The value operand of a
-  // store will remain scalar if the store is scalarized.
+  // memory access is not a gather/scatter or histogram operation. The value
+  // operand of a store will remain scalar if the store is scalarized.
   auto IsScalarUse = [&](Instruction *MemAccess, Value *Ptr) {
     InstWidening WideningDecision = getWideningDecision(MemAccess, VF);
     assert(WideningDecision != CM_Unknown &&
@@ -2239,7 +2239,8 @@ void LoopVectorizationCostModel::collectLoopScalars(ElementCount VF) {
         return WideningDecision == CM_Scalarize;
     assert(Ptr == getLoadStorePointerOperand(MemAccess) &&
            "Ptr is neither a value or pointer operand");
-    return WideningDecision != CM_GatherScatter;
+    return WideningDecision != CM_GatherScatter &&
+           !Legal->getHistogramInfo(MemAccess);
   };
 
   // A helper that returns true if the given value is a getelementptr
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/sve2-histcnt-epilogue.ll b/llvm/test/Transforms/LoopVectorize/AArch64/sve2-histcnt-epilogue.ll
index f14e913c9e15a..3c1854e7fff25 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/sve2-histcnt-epilogue.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/sve2-histcnt-epilogue.ll
@@ -46,12 +46,7 @@ define void @simple_histogram(ptr noalias %buckets, ptr readonly %indices, i64 %
 ; CHECK-NEXT:    [[TMP18:%.*]] = getelementptr inbounds i32, ptr [[INDICES]], i64 [[INDEX4]]
 ; CHECK-NEXT:    [[WIDE_LOAD5:%.*]] = load <2 x i32>, ptr [[TMP18]], align 4
 ; CHECK-NEXT:    [[TMP16:%.*]] = zext <2 x i32> [[WIDE_LOAD5]] to <2 x i64>
-; CHECK-NEXT:    [[TMP17:%.*]] = extractelement <2 x i64> [[TMP16]], i64 0
-; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds i32, ptr [[BUCKETS]], i64 [[TMP17]]
-; CHECK-NEXT:    [[TMP10:%.*]] = extractelement <2 x i64> [[TMP16]], i64 1
-; CHECK-NEXT:    [[TMP19:%.*]] = getelementptr inbounds i32, ptr [[BUCKETS]], i64 [[TMP10]]
-; CHECK-NEXT:    [[TMP20:%.*]] = insertelement <2 x ptr> poison, ptr [[TMP9]], i64 0
-; CHECK-NEXT:    [[TMP13:%.*]] = insertelement <2 x ptr> [[TMP20]], ptr [[TMP19]], i64 1
+; CHECK-NEXT:    [[TMP13:%.*]] = getelementptr inbounds i32, ptr [[BUCKETS]], <2 x i64> [[TMP16]]
 ; CHECK-NEXT:    call void @llvm.experimental.vector.histogram.add.v2p0.i32(<2 x ptr> [[TMP13]], i32 1, <2 x i1> splat (i1 true))
 ; CHECK-NEXT:    [[INDEX_NEXT6]] = add nuw i64 [[INDEX4]], 2
 ; CHECK-NEXT:    [[TMP21:%.*]] = icmp eq i64 [[INDEX_NEXT6]], [[N_VEC3]]
diff --git a/llvm/test/Transforms/LoopVectorize/histograms.ll b/llvm/test/Transforms/LoopVectorize/histograms.ll
index 895ff9e9c3b20..cf2ea77355d40 100644
--- a/llvm/test/Transforms/LoopVectorize/histograms.ll
+++ b/llvm/test/Transforms/LoopVectorize/histograms.ll
@@ -13,12 +13,7 @@ define void @simple_histogram(ptr noalias %buckets, ptr readonly %indices) {
 ; CHECK-NEXT:    [[TMP0:%.*]] = getelementptr inbounds i32, ptr [[INDICES]], i64 [[IV]]
 ; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <2 x i32>, ptr [[TMP0]], align 4
 ; CHECK-NEXT:    [[TMP1:%.*]] = zext <2 x i32> [[WIDE_LOAD]] to <2 x i64>
-; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <2 x i64> [[TMP1]], i64 0
-; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i32, ptr [[BUCKETS]], i64 [[TMP2]]
-; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <2 x i64> [[TMP1]], i64 1
-; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds i32, ptr [[BUCKETS]], i64 [[TMP4]]
-; CHECK-NEXT:    [[TMP6:%.*]] = insertelement <2 x ptr> poison, ptr [[TMP3]], i64 0
-; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <2 x ptr> [[TMP6]], ptr [[TMP5]], i64 1
+; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i32, ptr [[BUCKETS]], <2 x i64> [[TMP1]]
 ; CHECK-NEXT:    call void @llvm.experimental.vector.histogram.add.v2p0.i32(<2 x ptr> [[TMP7]], i32 1, <2 x i1> splat (i1 true))
 ; CHECK-NEXT:    [[IV_NEXT]] = add nuw i64 [[IV]], 2
 ; CHECK-NEXT:    [[TMP8:%.*]] = icmp eq i64 [[IV_NEXT]], 1000
@@ -62,13 +57,10 @@ define void @histogram_extra_use_of_bucket_value(ptr noalias %buckets, ptr reado
 ; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds i32, ptr [[INDICES]], i64 [[INDEX]]
 ; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <2 x i32>, ptr [[TMP1]], align 4
 ; CHECK-NEXT:    [[TMP2:%.*]] = zext <2 x i32> [[WIDE_LOAD]] to <2 x i64>
-; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <2 x i64> [[TMP2]], i64 0
-; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds i32, ptr [[BUCKETS]], i64 [[TMP3]]
-; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x i64> [[TMP2]], i64 1
-; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds i32, ptr [[BUCKETS]], i64 [[TMP5]]
-; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <2 x ptr> poison, ptr [[TMP4]], i64 0
-; CHECK-NEXT:    [[TMP8:%.*]] = insertelement <2 x ptr> [[TMP7]], ptr [[TMP6]], i64 1
+; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr inbounds i32, ptr [[BUCKETS]], <2 x i64> [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <2 x ptr> [[TMP8]], i64 0
 ; CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP4]], align 4
+; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <2 x ptr> [[TMP8]], i64 1
 ; CHECK-NEXT:    [[TMP10:%.*]] = load i32, ptr [[TMP6]], align 4
 ; CHECK-NEXT:    [[TMP11:%.*]] = insertelement <2 x i32> poison, i32 [[TMP9]], i64 0
 ; CHECK-NEXT:    [[TMP12:%.*]] = insertelement <2 x i32> [[TMP11]], i32 [[TMP10]], i64 1
@@ -119,13 +111,10 @@ define void @histogram_extra_use_of_update(ptr noalias %buckets, ptr readonly %i
 ; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds i32, ptr [[INDICES]], i64 [[INDEX]]
 ; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <2 x i32>, ptr [[TMP1]], align 4
 ; CHECK-NEXT:    [[TMP2:%.*]] = zext <2 x i32> [[WIDE_LOAD]] to <2 x i64>
-; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <2 x i64> [[TMP2]], i64 0
-; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds i32, ptr [[BUCKETS]], i64 [[TMP3]]
-; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x i64> [[TMP2]], i64 1
-; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds i32, ptr [[BUCKETS]], i64 [[TMP5]]
-; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <2 x ptr> poison, ptr [[TMP4]], i64 0
-; CHECK-NEXT:    [[TMP8:%.*]] = insertelement <2 x ptr> [[TMP7]], ptr [[TMP6]], i64 1
+; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr inbounds i32, ptr [[BUCKETS]], <2 x i64> [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <2 x ptr> [[TMP8]], i64 0
 ; CHECK-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP4]], align 4
+; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <2 x ptr> [[TMP8]], i64 1
 ; CHECK-NEXT:    [[TMP10:%.*]] = load i32, ptr [[TMP6]], align 4
 ; CHECK-NEXT:    [[TMP11:%.*]] = insertelement <2 x i32> poison, i32 [[TMP9]], i64 0
 ; CHECK-NEXT:    [[TMP12:%.*]] = insertelement <2 x i32> [[TMP11]], i32 [[TMP10]], i64 1

``````````

</details>


https://github.com/llvm/llvm-project/pull/215769
