[PATCH] D137639: [SLP]Fix PR58863: Mask index beyond mask size for non-power-2 insertelement analysis.

[Alexey Bataev via Phabricator via llvm-commits]

ABataev created this revision.
ABataev added reviewers: RKSimon, vdmitrie.
Herald added subscribers: vporpo, arphaman, hiraditya.
Herald added a project: All.
ABataev requested review of this revision.
Herald added a subscriber: pcwang-thead.
Herald added a project: LLVM.

Need to check if the insertelement mask size is reached during cost analysis to avoid compiler crash.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D137639

Files:
  llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
  llvm/test/Transforms/SLPVectorizer/slp-non-pow-2-insertelement.ll


Index: llvm/test/Transforms/SLPVectorizer/slp-non-pow-2-insertelement.ll
===================================================================
--- /dev/null
+++ llvm/test/Transforms/SLPVectorizer/slp-non-pow-2-insertelement.ll
@@ -0,0 +1,19 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -S -passes=slp-vectorizer < %s | FileCheck %s
+
+define void @PR58863() {
+; CHECK-LABEL: @PR58863(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[MUL_I:%.*]] = fmul float poison, poison
+; CHECK-NEXT:    [[MUL11_I:%.*]] = fmul float poison, poison
+; CHECK-NEXT:    [[I:%.*]] = insertelement <3 x float> <float poison, float 0.000000e+00, float poison>, float [[MUL_I]], i64 0
+; CHECK-NEXT:    [[I1:%.*]] = insertelement <3 x float> [[I]], float [[MUL11_I]], i64 2
+; CHECK-NEXT:    ret void
+;
+entry:
+  %mul.i = fmul float poison, poison
+  %mul11.i = fmul float poison, poison
+  %i = insertelement <3 x float> <float poison, float 0.000000e+00, float poison>, float %mul.i, i64 0
+  %i1 = insertelement <3 x float> %i, float %mul11.i, i64 2
+  ret void
+}
Index: llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
===================================================================
--- llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -6592,8 +6592,13 @@
              I <= End; ++I)
           if (Mask[I] != UndefMaskElem)
             Mask[I] = I + VecSz;
-        for (unsigned I = OffsetEnd + 1 - Offset; I < VecSz; ++I)
+        for (unsigned I = OffsetEnd + 1 - Offset; I < VecSz; ++I) {
+          if (I >= InMask.size()) {
+            Mask[I] = UndefMaskElem;
+            continue;
+          }
           Mask[I] = InMask.test(I) ? UndefMaskElem : I;
+        }
         Cost += TTI->getShuffleCost(TTI::SK_PermuteTwoSrc, InsertVecTy, Mask);
       }
     }


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D137639.473984.patch
Type: text/x-patch
Size: 1853 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20221108/54f58dce/attachment.bin>