Sync

dataset/140481.json

@@ -0,0 +1,105 @@
+{
+  "bug_id": "140481",
+  "issue_url": "https://github.com/llvm/llvm-project/issues/140481",
+  "bug_type": "miscompilation",
+  "base_commit": "f72a8ee489368dd20c1392b122b0736aa7c8ada1",
+  "knowledge_cutoff": "2025-05-18T21:41:49Z",
+  "lit_test_dir": [
+    "llvm/test/Transforms/ConstraintElimination"
+  ],
+  "hints": {
+    "fix_commit": "287294d54d7a806e70b0061cf5ccc1fc2bd03eea",
+    "components": [
+      "ConstraintElimination"
+    ],
+    "bug_location_lineno": {
+      "llvm/lib/Transforms/Scalar/ConstraintElimination.cpp": [
+        [
+          64,
+          83
+        ],
+        [
+          366,
+          391
+        ],
+        [
+          467,
+          474
+        ],
+        [
+          488,
+          498
+        ],
+        [
+          533,
+          553
+        ],
+        [
+          557,
+          564
+        ],
+        [
+          593,
+          600
+        ],
+        [
+          603,
+          609
+        ],
+        [
+          611,
+          643
+        ]
+      ]
+    },
+    "bug_location_funcname": {
+      "llvm/lib/Transforms/Scalar/ConstraintElimination.cpp": [
+        "Decomposition",
+        "addWithOverflow",
+        "decomposeGEP",
+        "getContextInstForUse",
+        "multiplyWithOverflow",
+        "sub"
+      ]
+    }
+  },
+  "patch": "commit 287294d54d7a806e70b0061cf5ccc1fc2bd03eea\nAuthor: Yingwei Zheng <[email protected]>\nDate:   Thu May 22 11:31:04 2025 +0800\n\n    [ConstraintElim] Do not allow overflows in `Decomposition` (#140541)\n    \n    Consider the following case:\n    ```\n    define i1 @pr140481(i32 %x) {\n      %cond = icmp slt i32 %x, 0\n      call void @llvm.assume(i1 %cond)\n      %add = add nsw i32 %x, 5001000\n      %mul1 = mul nsw i32 %add, -5001000\n      %mul2 = mul nsw i32 %mul1, 5001000\n      %cmp2 = icmp sgt i32 %mul2, 0\n      ret i1 %cmp2\n    }\n    ```\n    Before this patch, `decompose(%mul2)` returns `-25010001000000 * %x +\n    4052193514966861312`.\n    Therefore, `%cmp2` will be simplified into true because `%x s< 0 &&\n    -25010001000000 * %x + 4052193514966861312 s<= 0` is unsat.\n    \n    It is incorrect since the offset `-25010001000000 * 5001000 ->\n    4052193514966861312` signed wraps.\n    This patch treats a decomposition as invalid if overflows occur when\n    computing coefficients.\n    \n    Closes https://github.com/llvm/llvm-project/issues/140481.\n\ndiff --git a/llvm/lib/Transforms/Scalar/ConstraintElimination.cpp b/llvm/lib/Transforms/Scalar/ConstraintElimination.cpp\nindex da5be383df15..cbad5dd35768 100644\n--- a/llvm/lib/Transforms/Scalar/ConstraintElimination.cpp\n+++ b/llvm/lib/Transforms/Scalar/ConstraintElimination.cpp\n@@ -64,20 +64,6 @@ static cl::opt<bool> DumpReproducers(\n static int64_t MaxConstraintValue = std::numeric_limits<int64_t>::max();\n static int64_t MinSignedConstraintValue = std::numeric_limits<int64_t>::min();\n \n-// A helper to multiply 2 signed integers where overflowing is allowed.\n-static int64_t multiplyWithOverflow(int64_t A, int64_t B) {\n-  int64_t Result;\n-  MulOverflow(A, B, Result);\n-  return Result;\n-}\n-\n-// A helper to add 2 signed integers where overflowing is allowed.\n-static int64_t addWithOverflow(int64_t A, int64_t B) {\n-  int64_t Result;\n-  AddOverflow(A, B, Result);\n-  return Result;\n-}\n-\n static Instruction *getContextInstForUse(Use &U) {\n   Instruction *UserI = cast<Instruction>(U.getUser());\n   if (auto *Phi = dyn_cast<PHINode>(UserI))\n@@ -366,26 +352,42 @@ struct Decomposition {\n   Decomposition(int64_t Offset, ArrayRef<DecompEntry> Vars)\n       : Offset(Offset), Vars(Vars) {}\n \n-  void add(int64_t OtherOffset) {\n-    Offset = addWithOverflow(Offset, OtherOffset);\n+  /// Add \\p OtherOffset and return true if the operation overflows, i.e. the\n+  /// new decomposition is invalid.\n+  [[nodiscard]] bool add(int64_t OtherOffset) {\n+    return AddOverflow(Offset, OtherOffset, Offset);\n   }\n \n-  void add(const Decomposition &Other) {\n-    add(Other.Offset);\n+  /// Add \\p Other and return true if the operation overflows, i.e. the new\n+  /// decomposition is invalid.\n+  [[nodiscard]] bool add(const Decomposition &Other) {\n+    if (add(Other.Offset))\n+      return true;\n     append_range(Vars, Other.Vars);\n+    return false;\n   }\n \n-  void sub(const Decomposition &Other) {\n+  /// Subtract \\p Other and return true if the operation overflows, i.e. the new\n+  /// decomposition is invalid.\n+  [[nodiscard]] bool sub(const Decomposition &Other) {\n     Decomposition Tmp = Other;\n-    Tmp.mul(-1);\n-    add(Tmp.Offset);\n+    if (Tmp.mul(-1))\n+      return true;\n+    if (add(Tmp.Offset))\n+      return true;\n     append_range(Vars, Tmp.Vars);\n+    return false;\n   }\n \n-  void mul(int64_t Factor) {\n-    Offset = multiplyWithOverflow(Offset, Factor);\n+  /// Multiply all coefficients by \\p Factor and return true if the operation\n+  /// overflows, i.e. the new decomposition is invalid.\n+  [[nodiscard]] bool mul(int64_t Factor) {\n+    if (MulOverflow(Offset, Factor, Offset))\n+      return true;\n     for (auto &Var : Vars)\n-      Var.Coefficient = multiplyWithOverflow(Var.Coefficient, Factor);\n+      if (MulOverflow(Var.Coefficient, Factor, Var.Coefficient))\n+        return true;\n+    return false;\n   }\n };\n \n@@ -467,8 +469,10 @@ static Decomposition decomposeGEP(GEPOperator &GEP,\n   Decomposition Result(ConstantOffset.getSExtValue(), DecompEntry(1, BasePtr));\n   for (auto [Index, Scale] : VariableOffsets) {\n     auto IdxResult = decompose(Index, Preconditions, IsSigned, DL);\n-    IdxResult.mul(Scale.getSExtValue());\n-    Result.add(IdxResult);\n+    if (IdxResult.mul(Scale.getSExtValue()))\n+      return &GEP;\n+    if (Result.add(IdxResult))\n+      return &GEP;\n \n     if (!NW.hasNoUnsignedWrap()) {\n       // Try to prove nuw from nusw and nneg.\n@@ -488,11 +492,13 @@ static Decomposition decompose(Value *V,\n                                SmallVectorImpl<ConditionTy> &Preconditions,\n                                bool IsSigned, const DataLayout &DL) {\n \n-  auto MergeResults = [&Preconditions, IsSigned, &DL](Value *A, Value *B,\n-                                                      bool IsSignedB) {\n+  auto MergeResults = [&Preconditions, IsSigned,\n+                       &DL](Value *A, Value *B,\n+                            bool IsSignedB) -> std::optional<Decomposition> {\n     auto ResA = decompose(A, Preconditions, IsSigned, DL);\n     auto ResB = decompose(B, Preconditions, IsSignedB, DL);\n-    ResA.add(ResB);\n+    if (ResA.add(ResB))\n+      return std::nullopt;\n     return ResA;\n   };\n \n@@ -533,21 +539,26 @@ static Decomposition decompose(Value *V,\n         V = Op0;\n     }\n \n-    if (match(V, m_NSWAdd(m_Value(Op0), m_Value(Op1))))\n-      return MergeResults(Op0, Op1, IsSigned);\n+    if (match(V, m_NSWAdd(m_Value(Op0), m_Value(Op1)))) {\n+      if (auto Decomp = MergeResults(Op0, Op1, IsSigned))\n+        return *Decomp;\n+      return {V, IsKnownNonNegative};\n+    }\n \n     if (match(V, m_NSWSub(m_Value(Op0), m_Value(Op1)))) {\n       auto ResA = decompose(Op0, Preconditions, IsSigned, DL);\n       auto ResB = decompose(Op1, Preconditions, IsSigned, DL);\n-      ResA.sub(ResB);\n-      return ResA;\n+      if (!ResA.sub(ResB))\n+        return ResA;\n+      return {V, IsKnownNonNegative};\n     }\n \n     ConstantInt *CI;\n     if (match(V, m_NSWMul(m_Value(Op0), m_ConstantInt(CI))) && canUseSExt(CI)) {\n       auto Result = decompose(Op0, Preconditions, IsSigned, DL);\n-      Result.mul(CI->getSExtValue());\n-      return Result;\n+      if (!Result.mul(CI->getSExtValue()))\n+        return Result;\n+      return {V, IsKnownNonNegative};\n     }\n \n     // (shl nsw x, shift) is (mul nsw x, (1<<shift)), with the exception of\n@@ -557,8 +568,9 @@ static Decomposition decompose(Value *V,\n       if (Shift < Ty->getIntegerBitWidth() - 1) {\n         assert(Shift < 64 && \"Would overflow\");\n         auto Result = decompose(Op0, Preconditions, IsSigned, DL);\n-        Result.mul(int64_t(1) << Shift);\n-        return Result;\n+        if (!Result.mul(int64_t(1) << Shift))\n+          return Result;\n+        return {V, IsKnownNonNegative};\n       }\n     }\n \n@@ -593,8 +605,11 @@ static Decomposition decompose(Value *V,\n   Value *Op1;\n   ConstantInt *CI;\n   if (match(V, m_NUWAdd(m_Value(Op0), m_Value(Op1)))) {\n-    return MergeResults(Op0, Op1, IsSigned);\n+    if (auto Decomp = MergeResults(Op0, Op1, IsSigned))\n+      return *Decomp;\n+    return {V, IsKnownNonNegative};\n   }\n+\n   if (match(V, m_NSWAdd(m_Value(Op0), m_Value(Op1)))) {\n     if (!isKnownNonNegative(Op0, DL))\n       Preconditions.emplace_back(CmpInst::ICMP_SGE, Op0,\n@@ -603,7 +618,9 @@ static Decomposition decompose(Value *V,\n       Preconditions.emplace_back(CmpInst::ICMP_SGE, Op1,\n                                  ConstantInt::get(Op1->getType(), 0));\n \n-    return MergeResults(Op0, Op1, IsSigned);\n+    if (auto Decomp = MergeResults(Op0, Op1, IsSigned))\n+      return *Decomp;\n+    return {V, IsKnownNonNegative};\n   }\n \n   if (match(V, m_Add(m_Value(Op0), m_ConstantInt(CI))) && CI->isNegative() &&\n@@ -611,33 +628,41 @@ static Decomposition decompose(Value *V,\n     Preconditions.emplace_back(\n         CmpInst::ICMP_UGE, Op0,\n         ConstantInt::get(Op0->getType(), CI->getSExtValue() * -1));\n-    return MergeResults(Op0, CI, true);\n+    if (auto Decomp = MergeResults(Op0, CI, true))\n+      return *Decomp;\n+    return {V, IsKnownNonNegative};\n   }\n \n   // Decompose or as an add if there are no common bits between the operands.\n-  if (match(V, m_DisjointOr(m_Value(Op0), m_ConstantInt(CI))))\n-    return MergeResults(Op0, CI, IsSigned);\n+  if (match(V, m_DisjointOr(m_Value(Op0), m_ConstantInt(CI)))) {\n+    if (auto Decomp = MergeResults(Op0, CI, IsSigned))\n+      return *Decomp;\n+    return {V, IsKnownNonNegative};\n+  }\n \n   if (match(V, m_NUWShl(m_Value(Op1), m_ConstantInt(CI))) && canUseSExt(CI)) {\n     if (CI->getSExtValue() < 0 || CI->getSExtValue() >= 64)\n       return {V, IsKnownNonNegative};\n     auto Result = decompose(Op1, Preconditions, IsSigned, DL);\n-    Result.mul(int64_t{1} << CI->getSExtValue());\n-    return Result;\n+    if (!Result.mul(int64_t{1} << CI->getSExtValue()))\n+      return Result;\n+    return {V, IsKnownNonNegative};\n   }\n \n   if (match(V, m_NUWMul(m_Value(Op1), m_ConstantInt(CI))) && canUseSExt(CI) &&\n       (!CI->isNegative())) {\n     auto Result = decompose(Op1, Preconditions, IsSigned, DL);\n-    Result.mul(CI->getSExtValue());\n-    return Result;\n+    if (!Result.mul(CI->getSExtValue()))\n+      return Result;\n+    return {V, IsKnownNonNegative};\n   }\n \n   if (match(V, m_NUWSub(m_Value(Op0), m_Value(Op1)))) {\n     auto ResA = decompose(Op0, Preconditions, IsSigned, DL);\n     auto ResB = decompose(Op1, Preconditions, IsSigned, DL);\n-    ResA.sub(ResB);\n-    return ResA;\n+    if (!ResA.sub(ResB))\n+      return ResA;\n+    return {V, IsKnownNonNegative};\n   }\n \n   return {V, IsKnownNonNegative};\n",
+  "tests": [
+    {
+      "file": "llvm/test/Transforms/ConstraintElimination/constraint-overflow.ll",
+      "commands": [
+        "opt -passes=constraint-elimination -S %s"
+      ],
+      "tests": [
+        {
+          "test_name": "pr140481",
+          "test_body": "; Function Attrs: nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: write)\ndeclare void @llvm.assume(i1 noundef) #0\n\ndefine i1 @pr140481(i32 %x) {\nentry:\n  %cond = icmp slt i32 %x, 0\n  call void @llvm.assume(i1 %cond)\n  %add = add nsw i32 %x, 5001000\n  %mul1 = mul nsw i32 %add, -5001000\n  %mul2 = mul nsw i32 %mul1, 5001000\n  %cmp2 = icmp sgt i32 %mul2, 0\n  ret i1 %cmp2\n}\n\nattributes #0 = { nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: write) }\n"
+        }
+      ]
+    }
+  ],
+  "issue": {
+    "title": "wrong code at -O{s,2,3} on x86_64-linux-gnu",
+    "body": "Compiler Explorer: https://godbolt.org/z/KGW6a53xf\n\nIt appears to be a regression from 16.0.0, and affects 17.0.1 and later. \n\n```\n[590] % clangtk -v\nclang version 21.0.0git (https://github.com/llvm/llvm-project.git fb86b3d96b73f4e628288b180ef4e038da8b7bc1)\nTarget: x86_64-unknown-linux-gnu\nThread model: posix\nInstalledDir: /local/home/suz/suz-local/software/local/clang-trunk/bin\nBuild config: +assertions\nFound candidate GCC installation: /usr/lib/gcc/x86_64-linux-gnu/11\nFound candidate GCC installation: /usr/lib/gcc/x86_64-linux-gnu/12\nSelected GCC installation: /usr/lib/gcc/x86_64-linux-gnu/12\nCandidate multilib: .;@m64\nSelected multilib: .;@m64\nFound CUDA installation: /usr/local/cuda, version 12.1\n[591] % \n[591] % clangtk -O1 small.c; ./a.out\n[592] % \n[592] % clangtk -O3 small.c\n[593] % ./a.out\nAborted\n[594] % \n[594] % cat small.c\nint a = 1, b, c;\nint main() {\n  b = -5001001 * a + 5001000;\n  while (b >= 5001001)\n    b = a + 5001000;\n  c = -5001000 * b - 5001001;\n  if (5001000 * c >= b)\n    __builtin_abort();\n  return 0;\n}\n```",
+    "author": "zhendongsu",
+    "labels": [
+      "miscompilation",
+      "llvm:transforms"
+    ],
+    "comments": [
+      {
+        "author": "hstk30-hw",
+        "body": "https://godbolt.org/z/noce5aa3Y CC @nikic take a look"
+      },
+      {
+        "author": "dtcxzyw",
+        "body": "Reproducer: https://alive2.llvm.org/ce/z/JVTQCY\n"
+      },
+      {
+        "author": "dtcxzyw",
+        "body": "Further reduced version: https://alive2.llvm.org/ce/z/3_XCwc"
+      }
+    ]
+  },
+  "verified": true
+}

dataset/141237.json

@@ -0,0 +1,80 @@
+{
+  "bug_id": "141237",
+  "issue_url": "https://github.com/llvm/llvm-project/issues/141237",
+  "bug_type": "crash",
+  "base_commit": "af2a957ce30e3d91e17e2194e4be0a6b6481e4ba",
+  "knowledge_cutoff": "2025-05-23T15:00:13Z",
+  "lit_test_dir": [
+    "llvm/test/Transforms/LoopVectorize"
+  ],
+  "hints": {
+    "fix_commit": "a9b2998e315af64b7a68606af9064db425699c39",
+    "components": [
+      "LoopVectorize"
+    ],
+    "bug_location_lineno": {
+      "llvm/lib/Transforms/Vectorize/LoopVectorize.cpp": [
+        [
+          7082,
+          7087
+        ],
+        [
+          7315,
+          7321
+        ],
+        [
+          7352,
+          7357
+        ],
+        [
+          7477,
+          7485
+        ]
+      ],
+      "llvm/lib/Transforms/Vectorize/VPlanHelpers.h": [
+        [
+          364,
+          369
+        ]
+      ]
+    },
+    "bug_location_funcname": {
+      "llvm/lib/Transforms/Vectorize/LoopVectorize.cpp": [
+        "LoopVectorizationPlanner::computeBestVF",
+        "VPCostContext::getLegacyCost",
+        "VPCostContext::skipCostComputation",
+        "planContainsAdditionalSimplifications"
+      ]
+    }
+  },
+  "patch": "commit a9b2998e315af64b7a68606af9064db425699c39\nAuthor: Florian Hahn <[email protected]>\nDate:   Sat May 24 11:09:27 2025 +0100\n\n    [VPlan] Skip cost assert if VPlan converted to single-scalar recipes.\n    \n    Check if a VPlan transform converted recipes to single-scalar\n    VPReplicateRecipes (after 07c085af3efcd67503232f99a1652efc6e54c1a9). If\n    that's the case, the legacy cost model incorrectly overestimates the cost.\n    \n    Fixes https://github.com/llvm/llvm-project/issues/141237.\n\ndiff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp\nindex 275b3d567856..8a35afbb73f3 100644\n--- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp\n+++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp\n@@ -7082,6 +7082,11 @@ InstructionCost VPCostContext::getLegacyCost(Instruction *UI,\n   return CM.getInstructionCost(UI, VF);\n }\n \n+bool VPCostContext::isLegacyUniformAfterVectorization(Instruction *I,\n+                                                      ElementCount VF) const {\n+  return CM.isUniformAfterVectorization(I, VF);\n+}\n+\n bool VPCostContext::skipCostComputation(Instruction *UI, bool IsVector) const {\n   return CM.ValuesToIgnore.contains(UI) ||\n          (IsVector && CM.VecValuesToIgnore.contains(UI)) ||\n@@ -7315,7 +7320,8 @@ InstructionCost LoopVectorizationPlanner::cost(VPlan &Plan,\n /// cost-model did not account for.\n static bool planContainsAdditionalSimplifications(VPlan &Plan,\n                                                   VPCostContext &CostCtx,\n-                                                  Loop *TheLoop) {\n+                                                  Loop *TheLoop,\n+                                                  ElementCount VF) {\n   // First collect all instructions for the recipes in Plan.\n   auto GetInstructionForCost = [](const VPRecipeBase *R) -> Instruction * {\n     if (auto *S = dyn_cast<VPSingleDefRecipe>(R))\n@@ -7352,6 +7358,16 @@ static bool planContainsAdditionalSimplifications(VPlan &Plan,\n       // comparing against the legacy cost isn't desirable.\n       if (isa<VPPartialReductionRecipe>(&R))\n         return true;\n+\n+      /// If a VPlan transform folded a recipe to one producing a single-scalar,\n+      /// but the original instruction wasn't uniform-after-vectorization in the\n+      /// legacy cost model, the legacy cost overestimates the actual cost.\n+      if (auto *RepR = dyn_cast<VPReplicateRecipe>(&R)) {\n+        if (RepR->isSingleScalar() &&\n+            !CostCtx.isLegacyUniformAfterVectorization(\n+                RepR->getUnderlyingInstr(), VF))\n+          return true;\n+      }\n       if (Instruction *UI = GetInstructionForCost(&R)) {\n         // If we adjusted the predicate of the recipe, the cost in the legacy\n         // cost model may be different.\n@@ -7477,9 +7493,10 @@ VectorizationFactor LoopVectorizationPlanner::computeBestVF() {\n   // legacy cost model doesn't properly model costs for such loops.\n   assert((BestFactor.Width == LegacyVF.Width || BestPlan.hasEarlyExit() ||\n           planContainsAdditionalSimplifications(getPlanFor(BestFactor.Width),\n-                                                CostCtx, OrigLoop) ||\n-          planContainsAdditionalSimplifications(getPlanFor(LegacyVF.Width),\n-                                                CostCtx, OrigLoop)) &&\n+                                                CostCtx, OrigLoop,\n+                                                BestFactor.Width) ||\n+          planContainsAdditionalSimplifications(\n+              getPlanFor(LegacyVF.Width), CostCtx, OrigLoop, LegacyVF.Width)) &&\n          \" VPlan cost model and legacy cost model disagreed\");\n   assert((BestFactor.Width.isScalar() || BestFactor.ScalarCost > 0) &&\n          \"when vectorizing, the scalar cost must be computed.\");\ndiff --git a/llvm/lib/Transforms/Vectorize/VPlanHelpers.h b/llvm/lib/Transforms/Vectorize/VPlanHelpers.h\nindex 1d42c8f5f373..0446991ebfff 100644\n--- a/llvm/lib/Transforms/Vectorize/VPlanHelpers.h\n+++ b/llvm/lib/Transforms/Vectorize/VPlanHelpers.h\n@@ -364,6 +364,11 @@ struct VPCostContext {\n \n   /// Returns the OperandInfo for \\p V, if it is a live-in.\n   TargetTransformInfo::OperandValueInfo getOperandInfo(VPValue *V) const;\n+\n+  /// Return true if \\p I is considered uniform-after-vectorization in the\n+  /// legacy cost model for \\p VF. Only used to check for additional VPlan\n+  /// simplifications.\n+  bool isLegacyUniformAfterVectorization(Instruction *I, ElementCount VF) const;\n };\n \n /// This class can be used to assign names to VPValues. For VPValues without\n",
+  "tests": [
+    {
+      "file": "llvm/test/Transforms/LoopVectorize/X86/uniform_load.ll",
+      "commands": [
+        "opt -passes=loop-vectorize -S %s"
+      ],
+      "tests": [
+        {
+          "test_name": "foo",
+          "test_body": "target datalayout = \"e-m:e-i64:64-f80:128-n8:16:32:64-S128\"\ntarget triple = \"x86_64-unknown-linux-gnu\"\n\n@inc = external global float, align 4\n\ndefine void @foo(ptr noalias captures(none) %A, i64 %N) #0 {\nentry:\n  br label %loop\n\nloop:                                             ; preds = %loop, %entry\n  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]\n  %l.inc = load float, ptr @inc, align 4\n  %gep.A = getelementptr inbounds float, ptr %A, i64 %iv\n  %l.A = load float, ptr %gep.A, align 4\n  %add = fadd float %l.inc, %l.A\n  store float %add, ptr %gep.A, align 4\n  %iv.next = add nuw nsw i64 %iv, 1\n  %ec = icmp eq i64 %iv.next, 32\n  br i1 %ec, label %exit, label %loop\n\nexit:                                             ; preds = %loop\n  ret void\n}\n\nattributes #0 = { \"target-cpu\"=\"core-avx2\" }\n"
+        },
+        {
+          "test_name": "uniform_load_can_fold_users",
+          "test_body": "target datalayout = \"e-m:e-i64:64-f80:128-n8:16:32:64-S128\"\ntarget triple = \"x86_64-unknown-linux-gnu\"\n\ndefine void @uniform_load_can_fold_users(ptr noalias %src, ptr %dst, i64 %start, double %d) {\nentry:\n  br label %loop\n\nloop:                                             ; preds = %loop, %entry\n  %iv.1 = phi i64 [ 0, %entry ], [ %iv.1.next, %loop ]\n  %iv.2 = phi i64 [ %start, %entry ], [ %iv.2.next, %loop ]\n  %l = load double, ptr %src, align 8\n  %m = fmul double %l, 9.000000e+00\n  %div = fdiv double %m, %d\n  %sub = sub i64 %iv.1, 1\n  %gep.1 = getelementptr double, ptr %dst, i64 %iv.1\n  %gep.2 = getelementptr double, ptr %gep.1, i64 %sub\n  store double %div, ptr %gep.2, align 8\n  %iv.1.next = add i64 %iv.1, 1\n  %iv.2.next = add i64 %iv.2, -1\n  %ec = icmp sgt i64 %iv.2, 0\n  br i1 %ec, label %loop, label %exit\n\nexit:                                             ; preds = %loop\n  ret void\n}\n"
+        }
+      ]
+    }
+  ],
+  "issue": {
+    "title": "Crash in LoopVectorizationPlanner::computeBestVF()",
+    "body": "Reduced reproducer:\n```\ntarget datalayout = \"e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128\"\ntarget triple = \"x86_64-unknown-linux-gnu\"\n\ndefine void @reduced_test_(ptr %0, i64 %1, ptr %.sroa.067.0.copyload) {\n._crit_edge103:\n  br label %2\n\n2:                                                ; preds = %2, %._crit_edge103\n  %indvars.iv = phi i64 [ 0, %._crit_edge103 ], [ %indvars.iv.next, %2 ]\n  %3 = phi i64 [ %1, %._crit_edge103 ], [ %10, %2 ]\n  %4 = load double, ptr %0, align 8\n  %5 = fmul double %4, 0.000000e+00\n  %6 = fdiv double %5, 0.000000e+00\n  %7 = sub i64 %indvars.iv, %1\n  %8 = getelementptr double, ptr %.sroa.067.0.copyload, i64 %indvars.iv\n  %9 = getelementptr double, ptr %8, i64 %7\n  store double %6, ptr %9, align 8\n  %indvars.iv.next = add i64 %indvars.iv, 1\n  %10 = add i64 %3, -1\n  %11 = icmp sgt i64 %3, 0\n  br i1 %11, label %2, label %._crit_edge106\n\n._crit_edge106:                                   ; preds = %2\n  ret void\n}\n```\nCrash:\n```\n$ opt -O2 reduced.ll \nWARNING: You're attempting to print out a bitcode file.\nThis is inadvisable as it may cause display problems. If\nyou REALLY want to taste LLVM bitcode first-hand, you\ncan force output with the `-f' option.\n\nopt: .../llvm/lib/Transforms/Vectorize/LoopVectorize.cpp:7478: llvm::VectorizationFactor llvm::LoopVectorizationPlanner::computeBestVF(): Assertion `(BestFactor.Width == LegacyVF.Width || BestPlan.hasEarlyExit() || planContainsAdditionalSimplifications(getPlanFor(BestFactor.Width), CostCtx, OrigLoop) || planContainsAdditionalSimplifications(getPlanFor(LegacyVF.Width), CostCtx, OrigLoop)) && \" VPlan cost model and legacy cost model disagreed\"' failed.\nPLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace.\nStack dump:\n0.      Program arguments: opt -O2 reduced.ll\n1.      Running pass \"function<eager-inv>(float2int,lower-constant-intrinsics,loop(loop-rotate<header-duplication;no-prepare-for-lto>,loop-deletion),loop-distribute,inject-tli-mappings,loop-vectorize<no-interleave-forced-only;no-vectorize-forced-only;>,infer-alignment,loop-load-elim,instcombine<max-iterations=1;no-verify-fixpoint>,simplifycfg<bonus-inst-threshold=1;forward-switch-cond;switch-range-to-icmp;switch-to-lookup;no-keep-loops;hoist-common-insts;no-hoist-loads-stores-with-cond-faulting;sink-common-insts;speculate-blocks;simplify-cond-branch;no-speculate-unpredictables>,slp-vectorizer,vector-combine,instcombine<max-iterations=1;no-verify-fixpoint>,loop-unroll<O2>,transform-warning,sroa<preserve-cfg>,infer-alignment,instcombine<max-iterations=1;no-verify-fixpoint>,loop-mssa(licm<allowspeculation>),alignment-from-assumptions,loop-sink,instsimplify,div-rem-pairs,tailcallelim,simplifycfg<bonus-inst-threshold=1;no-forward-switch-cond;switch-range-to-icmp;no-switch-to-lookup;keep-loops;no-hoist-common-insts;hoist-loads-stores-with-cond-faulting;no-sink-common-insts;speculate-blocks;simplify-cond-branch;speculate-unpredictables>)\" on module \"reduced.ll\"\n2.      Running pass \"loop-vectorize<no-interleave-forced-only;no-vectorize-forced-only;>\" on function \"reduced_test_\"\n #0 0x0000000001b16948 llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) (.../llvm-5162/bin/opt+0x1b16948)\n #1 0x0000000001b13bd4 SignalHandler(int, siginfo_t*, void*) Signals.cpp:0:0\n #2 0x00007ce1f6c45330 (/lib/x86_64-linux-gnu/libc.so.6+0x45330)\n #3 0x00007ce1f6c9eb2c __pthread_kill_implementation ./nptl/pthread_kill.c:44:76\n #4 0x00007ce1f6c9eb2c __pthread_kill_internal ./nptl/pthread_kill.c:78:10\n #5 0x00007ce1f6c9eb2c pthread_kill ./nptl/pthread_kill.c:89:10\n #6 0x00007ce1f6c4527e raise ./signal/../sysdeps/posix/raise.c:27:6\n #7 0x00007ce1f6c288ff abort ./stdlib/abort.c:81:7\n #8 0x00007ce1f6c2881b _nl_load_domain ./intl/loadmsgcat.c:1177:9\n #9 0x00007ce1f6c3b517 (/lib/x86_64-linux-gnu/libc.so.6+0x3b517)\n#10 0x0000000003b73b36 llvm::LoopVectorizationPlanner::computeBestVF() (.../llvm-5162/bin/opt+0x3b73b36)\n#11 0x0000000003b7556a llvm::LoopVectorizePass::processLoop(llvm::Loop*) (.../llvm-5162/bin/opt+0x3b7556a)\n#12 0x0000000003b78140 llvm::LoopVectorizePass::runImpl(llvm::Function&) (.../llvm-5162/bin/opt+0x3b78140)\n#13 0x0000000003b78793 llvm::LoopVectorizePass::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (opt+0x3b78793)\n#14 0x00000000036215ae llvm::detail::PassModel<llvm::Function, llvm::LoopVectorizePass, llvm::AnalysisManager<llvm::Function>>::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) PassBuilder.cpp:0:0\n#15 0x0000000001de4030 llvm::PassManager<llvm::Function, llvm::AnalysisManager<llvm::Function>>::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (opt+0x1de4030)\n#16 0x0000000002e4da6e llvm::detail::PassModel<llvm::Function, llvm::PassManager<llvm::Function, llvm::AnalysisManager<llvm::Function>>, llvm::AnalysisManager<llvm::Function>>::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) NVPTXTargetMachine.cpp:0:0\n#17 0x0000000001de4543 llvm::ModuleToFunctionPassAdaptor::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (opt+0x1de4543)\n#18 0x0000000002e4da2e llvm::detail::PassModel<llvm::Module, llvm::ModuleToFunctionPassAdaptor, llvm::AnalysisManager<llvm::Module>>::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) NVPTXTargetMachine.cpp:0:0\n#19 0x0000000001de2160 llvm::PassManager<llvm::Module, llvm::AnalysisManager<llvm::Module>>::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (opt+0x1de2160)\n#20 0x00000000034ad61a llvm::runPassPipeline(llvm::StringRef, llvm::Module&, llvm::TargetMachine*, llvm::TargetLibraryInfoImpl*, llvm::ToolOutputFile*, llvm::ToolOutputFile*, llvm::ToolOutputFile*, llvm::StringRef, llvm::ArrayRef<llvm::PassPlugin>, llvm::ArrayRef<std::function<void (llvm::PassBuilder&)>>, llvm::opt_tool::OutputKind, llvm::opt_tool::VerifierKind, bool, bool, bool, bool, bool, bool, bool) (opt+0x34ad61a)\n...\nAborted\n```",
+    "author": "eugeneepshteyn",
+    "labels": [
+      "vectorizers",
+      "crash"
+    ],
+    "comments": []
+  },
+  "verified": true
+}

dataset/141265.json

@@ -0,0 +1,56 @@
+{
+  "bug_id": "141265",
+  "issue_url": "https://github.com/llvm/llvm-project/issues/141265",
+  "bug_type": "crash",
+  "base_commit": "69f2ff3e9be5e786529a409e6f06f942096e8dbb",
+  "knowledge_cutoff": "2025-05-23T17:57:35Z",
+  "lit_test_dir": [
+    "llvm/test/Transforms/SLPVectorizer"
+  ],
+  "hints": {
+    "fix_commit": "aa452b65fc7ebfee6f7e5b9d08aa418d532c7b88",
+    "components": [
+      "SLPVectorizer"
+    ],
+    "bug_location_lineno": {
+      "llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp": [
+        [
+          18535,
+          18540
+        ]
+      ]
+    },
+    "bug_location_funcname": {
+      "llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp": [
+        "BoUpSLP::vectorizeTree"
+      ]
+    }
+  },
+  "patch": "commit aa452b65fc7ebfee6f7e5b9d08aa418d532c7b88\nAuthor: Alexey Bataev <[email protected]>\nDate:   Sat May 24 07:20:41 2025 -0700\n\n    [SLP]Restore insertion points after gathers vectorization\n    \n    Restore insertion points after gathers vectorization to avoid a crash in\n    a root node vectorization.\n    \n    Fixes #141265\n\ndiff --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp\nindex 0f86c572639c..831703b375d9 100644\n--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp\n+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp\n@@ -18535,6 +18535,7 @@ Value *BoUpSLP::vectorizeTree(\n     }\n   }\n   for (auto &Entry : GatherEntries) {\n+    IRBuilderBase::InsertPointGuard Guard(Builder);\n     Builder.SetInsertPoint(Entry.second);\n     Builder.SetCurrentDebugLocation(Entry.second->getDebugLoc());\n     (void)vectorizeTree(Entry.first);\n",
+  "tests": [
+    {
+      "file": "llvm/test/Transforms/SLPVectorizer/RISCV/gather-insert-point-restore.ll",
+      "commands": [
+        "opt -S --passes=slp-vectorizer -mtriple=riscv64-unknown-linux-gnu -mattr=+v < %s"
+      ],
+      "tests": [
+        {
+          "test_name": "<module>",
+          "test_body": "\ndefine i16 @test(ptr %i) {\n;\nentry:\n  %gep.us154 = getelementptr i8, ptr %i, i64 132860\n  %gep.us154.1 = getelementptr i8, ptr %i, i64 137774\n  %gep.us154.2 = getelementptr i8, ptr %i, i64 142688\n  %gep.us154.3 = getelementptr i8, ptr %i, i64 147602\n  %gep.us154.4 = getelementptr i8, ptr %i, i64 152516\n  %gep.us154.5 = getelementptr i8, ptr %i, i64 157430\n  br label %for.cond5.us\n\nfor.cond5.us:\n  %0 = load i16, ptr %gep.us154, align 2\n  %1 = load i16, ptr %gep.us154.1, align 2\n  %2 = load i16, ptr %gep.us154.2, align 2\n  %3 = load i16, ptr %gep.us154.3, align 2\n  %4 = load i16, ptr %gep.us154.4, align 2\n  %5 = load i16, ptr %gep.us154.5, align 2\n  %6 = call i16 @llvm.umax.i16(i16 %5, i16 0)\n  %7 = call i16 @llvm.umax.i16(i16 %0, i16 %6)\n  %8 = call i16 @llvm.umax.i16(i16 %1, i16 %7)\n  %9 = call i16 @llvm.umax.i16(i16 %2, i16 %8)\n  %10 = call i16 @llvm.umax.i16(i16 %3, i16 %9)\n  %11 = call i16 @llvm.umax.i16(i16 %2, i16 %10)\n  %12 = call i16 @llvm.umax.i16(i16 %3, i16 %11)\n  %13 = call i16 @llvm.umax.i16(i16 %4, i16 %12)\n  %14 = load i16, ptr %gep.us154, align 2\n  %15 = call i16 @llvm.umax.i16(i16 %14, i16 %13)\n  %16 = load i16, ptr %gep.us154.1, align 2\n  %17 = call i16 @llvm.umax.i16(i16 %16, i16 %15)\n  %18 = call i16 @llvm.umax.i16(i16 %4, i16 %17)\n  ret i16 %18\n}\n\ndeclare i16 @llvm.umax.i16(i16, i16) #1"
+        }
+      ]
+    }
+  ],
+  "issue": {
+    "title": "[SLPVectorizer] Instruction does not dominate all uses!",
+    "body": "Testcase:\n```llvm ir\ntarget datalayout = \"e-m:e-p:64:64-i64:64-i128:128-n32:64-S128\"\ntarget triple = \"riscv64-unknown-linux-gnu\"\n\ndefine i16 @c(ptr %i) #0 {\nentry:\n  %gep.us154 = getelementptr i8, ptr %i, i64 132860\n  %gep.us154.1 = getelementptr i8, ptr %i, i64 137774\n  %gep.us154.2 = getelementptr i8, ptr %i, i64 142688\n  %gep.us154.3 = getelementptr i8, ptr %i, i64 147602\n  %gep.us154.4 = getelementptr i8, ptr %i, i64 152516\n  %gep.us154.5 = getelementptr i8, ptr %i, i64 157430\n  br label %for.cond5.us\n\nfor.cond5.us:                                     ; preds = %entry\n  %0 = load i16, ptr %gep.us154, align 2\n  %1 = load i16, ptr %gep.us154.1, align 2\n  %2 = load i16, ptr %gep.us154.2, align 2\n  %3 = load i16, ptr %gep.us154.3, align 2\n  %4 = load i16, ptr %gep.us154.4, align 2\n  %5 = load i16, ptr %gep.us154.5, align 2\n  %6 = call i16 @llvm.umax.i16(i16 %5, i16 0)\n  %7 = call i16 @llvm.umax.i16(i16 %0, i16 %6)\n  %8 = call i16 @llvm.umax.i16(i16 %1, i16 %7)\n  %9 = call i16 @llvm.umax.i16(i16 %2, i16 %8)\n  %10 = call i16 @llvm.umax.i16(i16 %3, i16 %9)\n  %11 = call i16 @llvm.umax.i16(i16 %2, i16 %10)\n  %12 = call i16 @llvm.umax.i16(i16 %3, i16 %11)\n  %13 = call i16 @llvm.umax.i16(i16 %4, i16 %12)\n  %14 = load i16, ptr %gep.us154, align 2\n  %15 = call i16 @llvm.umax.i16(i16 %14, i16 %13)\n  %16 = load i16, ptr %gep.us154.1, align 2\n  %17 = call i16 @llvm.umax.i16(i16 %16, i16 %15)\n  %18 = call i16 @llvm.umax.i16(i16 %4, i16 %17)\n  ret i16 %18\n}\n\n; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)\ndeclare i16 @llvm.umax.i16(i16, i16) #1\n\n; uselistorder directives\nuselistorder ptr @llvm.umax.i16, { 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0 }\n\nattributes #0 = { \"target-features\"=\"+v\" }\nattributes #1 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }\n```\n\nCommand/backtrace:\n```\n$ /scratch/ewlu/daily-upstream-build/build-gcv/build-llvm-linux/bin/opt reduced.ll --passes=slp-vectorizer\nWARNING: You're attempting to print out a bitcode file.\nThis is inadvisable as it may cause display problems. If\nyou REALLY want to taste LLVM bitcode first-hand, you\ncan force output with the `-f' option.\n\nInstruction does not dominate all uses!\n  %6 = call <4 x i16> @llvm.experimental.vp.strided.load.v4i16.p0.i64(ptr align 2 %gep.us154.2, i64 4914, <4 x i1> splat (i1 true), i32 4)\n  %4 = call <8 x i16> @llvm.vector.insert.v8i16.v4i16(<8 x i16> poison, <4 x i16> %6, i64 0)\nInstruction does not dominate all uses!\n  %7 = call <4 x i16> @llvm.masked.gather.v4i16.v4p0(<4 x ptr> %3, i32 2, <4 x i1> splat (i1 true), <4 x i16> poison)\n  %5 = call <8 x i16> @llvm.vector.insert.v8i16.v4i16(<8 x i16> %4, <4 x i16> %7, i64 4)\nLLVM ERROR: Broken module found, compilation aborted!\nPLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace.\nStack dump:\n0.      Program arguments: /scratch/ewlu/daily-upstream-build/build-gcv/build-llvm-linux/bin/opt reduced.ll --passes=slp-vectorizer\n1.      Running pass \"verify\" on module \"reduced.ll\"\n #0 0x00005de624576b42 llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) (/scratch/ewlu/daily-upstream-build/build-gcv/build-llvm-linux/bin/opt+0x3588b42)\n #1 0x00005de624573baf llvm::sys::RunSignalHandlers() (/scratch/ewlu/daily-upstream-build/build-gcv/build-llvm-linux/bin/opt+0x3585baf)\n #2 0x00005de624573cf4 SignalHandler(int, siginfo_t*, void*) Signals.cpp:0:0\n #3 0x0000723bce445330 (/lib/x86_64-linux-gnu/libc.so.6+0x45330)\n #4 0x0000723bce49eb2c __pthread_kill_implementation ./nptl/pthread_kill.c:44:76\n #5 0x0000723bce49eb2c __pthread_kill_internal ./nptl/pthread_kill.c:78:10\n #6 0x0000723bce49eb2c pthread_kill ./nptl/pthread_kill.c:89:10\n #7 0x0000723bce44527e raise ./signal/../sysdeps/posix/raise.c:27:6\n #8 0x0000723bce4288ff abort ./stdlib/abort.c:81:7\n #9 0x00005de6216a55f8 llvm::GlobPattern::create(llvm::StringRef, std::optional<unsigned long>) (.cold) GlobPattern.cpp:0:0\n#10 0x00005de6244a24f5 (/scratch/ewlu/daily-upstream-build/build-gcv/build-llvm-linux/bin/opt+0x34b44f5)\n#11 0x00005de6243a4a90 llvm::VerifierPass::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (/scratch/ewlu/daily-upstream-build/build-gcv/build-llvm-linux/bin/opt+0x33b6a90)\n#12 0x00005de621795e15 llvm::detail::PassModel<llvm::Module, llvm::VerifierPass, llvm::AnalysisManager<llvm::Module>>::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (/scratch/ewlu/daily-upstream-build/build-gcv/build-llvm-linux/bin/opt+0x7a7e15)\n#13 0x00005de62435f33d llvm::PassManager<llvm::Module, llvm::AnalysisManager<llvm::Module>>::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (/scratch/ewlu/daily-upstream-build/build-gcv/build-llvm-linux/bin/opt+0x337133d)\n#14 0x00005de6217a2051 llvm::runPassPipeline(llvm::StringRef, llvm::Module&, llvm::TargetMachine*, llvm::TargetLibraryInfoImpl*, llvm::ToolOutputFile*, llvm::ToolOutputFile*, llvm::ToolOutputFile*, llvm::StringRef, llvm::ArrayRef<llvm::PassPlugin>, llvm::ArrayRef<std::function<void (llvm::PassBuilder&)>>, llvm::opt_tool::OutputKind, llvm::opt_tool::VerifierKind, bool, bool, bool, bool, bool, bool, bool) (/scratch/ewlu/daily-upstream-build/build-gcv/build-llvm-linux/bin/opt+0x7b4051)\n#15 0x00005de621793b8c optMain (/scratch/ewlu/daily-upstream-build/build-gcv/build-llvm-linux/bin/opt+0x7a5b8c)\n#16 0x0000723bce42a1ca __libc_start_call_main ./csu/../sysdeps/nptl/libc_start_call_main.h:74:3\n#17 0x0000723bce42a28b call_init ./csu/../csu/libc-start.c:128:20\n#18 0x0000723bce42a28b __libc_start_main ./csu/../csu/libc-start.c:347:5\n#19 0x00005de62178a1e5 _start (/scratch/ewlu/daily-upstream-build/build-gcv/build-llvm-linux/bin/opt+0x79c1e5)\nAborted\n```\n\nGodbolt: https://godbolt.org/z/hq5nGcnrd\n\nFound via fuzzer: Reduced from C testcase in https://github.com/llvm/llvm-project/issues/141262",
+    "author": "ewlu",
+    "labels": [
+      "llvm:SLPVectorizer",
+      "crash",
+      "generated by fuzzer"
+    ],
+    "comments": []
+  },
+  "verified": true
+}

scripts/extract_from_issues.py

@@ -34,7 +34,7 @@ session.headers.update(
 )
 
 issue_id_begin = 76663  # Since 2024-01-01
-issue_id_end = 140749
+issue_id_end = 141464
 
 
 def wait(progress):