[DropAssumes] Drop dereferenceable assumptions after vectorization. #166947

fhahn · 2025-11-07T15:01:15Z

This patch adds another run of DropUnnecessaryAssumes after vectorization, to clean up assumes that are not longer needed after this point.

The main example of such an assume is currently dereferenceable assumptions. This complements #166945, which avoids sinking code if it would mean remove a dereferenceable assumption.

There are a few additional cases where some unneeded assumes are left over after vectorization that also get cleaned up.

The main motivation is to work together with #166945, but there may be a better solution.

Adding another instance of this pass to the pipeline is not great, but compile-time impact seems in the noise: https://llvm-compile-time-tracker.com/compare.php?from=55e71fe08b6406ec7ce2c81ce042e48717acf204&to=85da4ee3a74126f557cdc74c7b40e048dacb3fc4&stat=instructions:u

llvmbot · 2025-11-07T15:01:48Z

@llvm/pr-subscribers-llvm-transforms

Author: Florian Hahn (fhahn)

Changes

This patch adds another run of DropUnnecessaryAssumes after vectorization, to clean up assumes that are not longer needed after this point.

The main example of such an assume is currently dereferenceable assumptions. This complements #166945, which avoids sinking code if it would mean remove a derferenceable assumption.

There are a few additional cases where some unneeded assumes are left over after vectorization that also get cleaned up.

The main motivation is to work together with #166945, but there may be a better solution.

Adding another instance of this pass to the pipeline is not great, but compile-time impact seems in the noise: https://llvm-compile-time-tracker.com/compare.php?from=55e71fe08b6406ec7ce2c81ce042e48717acf204&to=85da4ee3a74126f557cdc74c7b40e048dacb3fc4&stat=instructions:u

Patch is 20.80 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/166947.diff

14 Files Affected:

(modified) llvm/include/llvm/Passes/PassBuilder.h (+1-1)
(modified) llvm/include/llvm/Transforms/Scalar/DropUnnecessaryAssumes.h (+6)
(modified) llvm/lib/Passes/PassBuilder.cpp (+5)
(modified) llvm/lib/Passes/PassBuilderPipelines.cpp (+11-3)
(modified) llvm/lib/Passes/PassRegistry.def (+4-1)
(modified) llvm/lib/Transforms/Scalar/DropUnnecessaryAssumes.cpp (+6-1)
(modified) llvm/test/Other/new-pm-defaults.ll (+1)
(modified) llvm/test/Other/new-pm-lto-defaults.ll (+1)
(modified) llvm/test/Other/new-pm-thinlto-postlink-defaults.ll (+1)
(modified) llvm/test/Other/new-pm-thinlto-postlink-pgo-defaults.ll (+1)
(modified) llvm/test/Other/new-pm-thinlto-postlink-samplepgo-defaults.ll (+1)
(added) llvm/test/Transforms/DropUnnecessaryAssumes/dereferenceable.ll (+54)
(modified) llvm/test/Transforms/PhaseOrdering/AArch64/matrix-extract-insert.ll (-68)
(modified) llvm/test/Transforms/PhaseOrdering/AArch64/std-find.ll (-1)

diff --git a/llvm/include/llvm/Passes/PassBuilder.h b/llvm/include/llvm/Passes/PassBuilder.h
index 8538a8b2afe14..8fa21f2cb2dd6 100644
--- a/llvm/include/llvm/Passes/PassBuilder.h
+++ b/llvm/include/llvm/Passes/PassBuilder.h
@@ -742,7 +742,7 @@ class PassBuilder {
   void addRequiredLTOPreLinkPasses(ModulePassManager &MPM);
 
   void addVectorPasses(OptimizationLevel Level, FunctionPassManager &FPM,
-                       bool IsFullLTO);
+                       ThinOrFullLTOPhase LTOPhase);
 
   static std::optional<std::vector<PipelineElement>>
   parsePipelineText(StringRef Text);
diff --git a/llvm/include/llvm/Transforms/Scalar/DropUnnecessaryAssumes.h b/llvm/include/llvm/Transforms/Scalar/DropUnnecessaryAssumes.h
index 4ff442ff80c76..54ddcc09f7204 100644
--- a/llvm/include/llvm/Transforms/Scalar/DropUnnecessaryAssumes.h
+++ b/llvm/include/llvm/Transforms/Scalar/DropUnnecessaryAssumes.h
@@ -19,7 +19,13 @@ namespace llvm {
 
 struct DropUnnecessaryAssumesPass
     : public PassInfoMixin<DropUnnecessaryAssumesPass> {
+  DropUnnecessaryAssumesPass(bool DropDereferenceable = false)
+      : DropDereferenceable(DropDereferenceable) {}
+
   PreservedAnalyses run(Function &F, FunctionAnalysisManager &AM);
+
+private:
+  bool DropDereferenceable;
 };
 
 } // end namespace llvm
diff --git a/llvm/lib/Passes/PassBuilder.cpp b/llvm/lib/Passes/PassBuilder.cpp
index 40ceb6f6ae28f..e0babc4385aab 100644
--- a/llvm/lib/Passes/PassBuilder.cpp
+++ b/llvm/lib/Passes/PassBuilder.cpp
@@ -900,6 +900,11 @@ Expected<bool> parseEntryExitInstrumenterPassOptions(StringRef Params) {
                                             "EntryExitInstrumenter");
 }
 
+Expected<bool> parseDropUnnecessaryAssumesPassOptions(StringRef Params) {
+  return PassBuilder::parseSinglePassOption(Params, "drop-deref",
+                                            "DropUnnecessaryAssumes");
+}
+
 Expected<bool> parseLoopExtractorPassOptions(StringRef Params) {
   return PassBuilder::parseSinglePassOption(Params, "single", "LoopExtractor");
 }
diff --git a/llvm/lib/Passes/PassBuilderPipelines.cpp b/llvm/lib/Passes/PassBuilderPipelines.cpp
index 3f41618b18fcf..2fe963b3b68d9 100644
--- a/llvm/lib/Passes/PassBuilderPipelines.cpp
+++ b/llvm/lib/Passes/PassBuilderPipelines.cpp
@@ -1298,10 +1298,18 @@ PassBuilder::buildModuleSimplificationPipeline(OptimizationLevel Level,
 
 /// TODO: Should LTO cause any differences to this set of passes?
 void PassBuilder::addVectorPasses(OptimizationLevel Level,
-                                  FunctionPassManager &FPM, bool IsFullLTO) {
+                                  FunctionPassManager &FPM,
+                                  ThinOrFullLTOPhase LTOPhase) {
+  const bool IsFullLTO = LTOPhase == ThinOrFullLTOPhase::FullLTOPostLink;
+
   FPM.addPass(LoopVectorizePass(
       LoopVectorizeOptions(!PTO.LoopInterleaving, !PTO.LoopVectorization)));
 
+  // Drop dereferenceable assumes after vectorization, as they are no longer
+  // needed and can inhibit further optimization.
+  if (!isLTOPreLink(LTOPhase))
+    FPM.addPass(DropUnnecessaryAssumesPass(/*DropDereferenceable=*/true));
+
   FPM.addPass(InferAlignmentPass());
   if (IsFullLTO) {
     // The vectorizer may have significantly shortened a loop body; unroll
@@ -1572,7 +1580,7 @@ PassBuilder::buildModuleOptimizationPipeline(OptimizationLevel Level,
   // from the TargetLibraryInfo.
   OptimizePM.addPass(InjectTLIMappings());
 
-  addVectorPasses(Level, OptimizePM, /* IsFullLTO */ false);
+  addVectorPasses(Level, OptimizePM, LTOPhase);
 
   invokeVectorizerEndEPCallbacks(OptimizePM, Level);
 
@@ -2162,7 +2170,7 @@ PassBuilder::buildLTODefaultPipeline(OptimizationLevel Level,
 
   MainFPM.addPass(LoopDistributePass());
 
-  addVectorPasses(Level, MainFPM, /* IsFullLTO */ true);
+  addVectorPasses(Level, MainFPM, ThinOrFullLTOPhase::FullLTOPostLink);
 
   invokeVectorizerEndEPCallbacks(MainFPM, Level);
 
diff --git a/llvm/lib/Passes/PassRegistry.def b/llvm/lib/Passes/PassRegistry.def
index d870f99aad552..d8305fe5c8e73 100644
--- a/llvm/lib/Passes/PassRegistry.def
+++ b/llvm/lib/Passes/PassRegistry.def
@@ -432,7 +432,6 @@ FUNCTION_PASS("dot-post-dom", PostDomPrinter())
 FUNCTION_PASS("dot-post-dom-only", PostDomOnlyPrinter())
 FUNCTION_PASS("dse", DSEPass())
 FUNCTION_PASS("dwarf-eh-prepare", DwarfEHPreparePass(*TM))
-FUNCTION_PASS("drop-unnecessary-assumes", DropUnnecessaryAssumesPass())
 FUNCTION_PASS("expand-large-div-rem", ExpandLargeDivRemPass(*TM))
 FUNCTION_PASS("expand-memcmp", ExpandMemCmpPass(*TM))
 FUNCTION_PASS("expand-reductions", ExpandReductionsPass())
@@ -584,6 +583,10 @@ FUNCTION_PASS_WITH_PARAMS(
     "early-cse", "EarlyCSEPass",
     [](bool UseMemorySSA) { return EarlyCSEPass(UseMemorySSA); },
     parseEarlyCSEPassOptions, "memssa")
+FUNCTION_PASS_WITH_PARAMS(
+    "drop-unnecessary-assumes", "DropUnnecessaryAssumesPass",
+    [](bool DropDereferenceable) { return DropUnnecessaryAssumesPass(DropDereferenceable); },
+    parseDropUnnecessaryAssumesPassOptions, "drop-deref")
 FUNCTION_PASS_WITH_PARAMS(
     "ee-instrument", "EntryExitInstrumenterPass",
     [](bool PostInlining) { return EntryExitInstrumenterPass(PostInlining); },
diff --git a/llvm/lib/Transforms/Scalar/DropUnnecessaryAssumes.cpp b/llvm/lib/Transforms/Scalar/DropUnnecessaryAssumes.cpp
index a577f517d1e89..4a7144fe6c77a 100644
--- a/llvm/lib/Transforms/Scalar/DropUnnecessaryAssumes.cpp
+++ b/llvm/lib/Transforms/Scalar/DropUnnecessaryAssumes.cpp
@@ -78,11 +78,16 @@ DropUnnecessaryAssumesPass::run(Function &F, FunctionAnalysisManager &FAM) {
       SmallVector<OperandBundleDef> KeptBundles;
       unsigned NumBundles = Assume->getNumOperandBundles();
       for (unsigned I = 0; I != NumBundles; ++I) {
-        auto IsDead = [](OperandBundleUse Bundle) {
+        auto IsDead = [&](OperandBundleUse Bundle) {
           // "ignore" operand bundles are always dead.
           if (Bundle.getTagName() == "ignore")
             return true;
 
+          // "dereferenceable" operand bundles are only dropped if requested
+          // (e.g., after loop vectorization has run).
+          if (Bundle.getTagName() == "dereferenceable")
+            return DropDereferenceable;
+
           // Bundles without arguments do not affect any specific values.
           // Always keep them for now.
           if (Bundle.Inputs.empty())
diff --git a/llvm/test/Other/new-pm-defaults.ll b/llvm/test/Other/new-pm-defaults.ll
index 62975a3cf8ac4..b59d4cf6af998 100644
--- a/llvm/test/Other/new-pm-defaults.ll
+++ b/llvm/test/Other/new-pm-defaults.ll
@@ -261,6 +261,7 @@
 ; CHECK-O-NEXT: Running analysis: LoopAccessAnalysis on foo
 ; CHECK-O-NEXT: Running pass: InjectTLIMappings
 ; CHECK-O-NEXT: Running pass: LoopVectorizePass
+; CHECK-DEFAULT-NEXT: Running pass: DropUnnecessaryAssumesPass
 ; CHECK-O-NEXT: Running pass: InferAlignmentPass
 ; CHECK-O-NEXT: Running pass: LoopLoadEliminationPass
 ; CHECK-O-NEXT: Running pass: InstCombinePass
diff --git a/llvm/test/Other/new-pm-lto-defaults.ll b/llvm/test/Other/new-pm-lto-defaults.ll
index f595dfe1d6845..c865d77c86d77 100644
--- a/llvm/test/Other/new-pm-lto-defaults.ll
+++ b/llvm/test/Other/new-pm-lto-defaults.ll
@@ -129,6 +129,7 @@
 ; CHECK-O23SZ-NEXT: Running analysis: LoopAccessAnalysis on foo
 ; CHECK-O23SZ-NEXT: Running pass: LoopVectorizePass on foo
 ; CHECK-O23SZ-NEXT: Running analysis: DemandedBitsAnalysis on foo
+; CHECK-O23SZ-NEXT: Running pass: DropUnnecessaryAssumesPass on foo
 ; CHECK-O23SZ-NEXT: Running pass: InferAlignmentPass on foo
 ; CHECK-O23SZ-NEXT: Running pass: LoopUnrollPass on foo
 ; CHECK-O23SZ-NEXT: WarnMissedTransformationsPass on foo
diff --git a/llvm/test/Other/new-pm-thinlto-postlink-defaults.ll b/llvm/test/Other/new-pm-thinlto-postlink-defaults.ll
index 012a1ab5802b5..c1d8b42505c84 100644
--- a/llvm/test/Other/new-pm-thinlto-postlink-defaults.ll
+++ b/llvm/test/Other/new-pm-thinlto-postlink-defaults.ll
@@ -180,6 +180,7 @@
 ; CHECK-POSTLINK-O-NEXT: Running analysis: LoopAccessAnalysis on foo
 ; CHECK-POSTLINK-O-NEXT: Running pass: InjectTLIMappings
 ; CHECK-POSTLINK-O-NEXT: Running pass: LoopVectorizePass
+; CHECK-POSTLINK-O-NEXT: Running pass: DropUnnecessaryAssumesPass
 ; CHECK-POSTLINK-O-NEXT: Running pass: InferAlignmentPass
 ; CHECK-POSTLINK-O-NEXT: Running pass: LoopLoadEliminationPass
 ; CHECK-POSTLINK-O-NEXT: Running pass: InstCombinePass
diff --git a/llvm/test/Other/new-pm-thinlto-postlink-pgo-defaults.ll b/llvm/test/Other/new-pm-thinlto-postlink-pgo-defaults.ll
index e021ff3124b60..45f090252eaf7 100644
--- a/llvm/test/Other/new-pm-thinlto-postlink-pgo-defaults.ll
+++ b/llvm/test/Other/new-pm-thinlto-postlink-pgo-defaults.ll
@@ -165,6 +165,7 @@
 ; CHECK-O-NEXT: Running analysis: LoopAccessAnalysis on foo
 ; CHECK-O-NEXT: Running pass: InjectTLIMappings
 ; CHECK-O-NEXT: Running pass: LoopVectorizePass
+; CHECK-O-NEXT: Running pass: DropUnnecessaryAssumesPass
 ; CHECK-O-NEXT: Running pass: InferAlignmentPass
 ; CHECK-O-NEXT: Running pass: LoopLoadEliminationPass
 ; CHECK-O-NEXT: Running pass: InstCombinePass
diff --git a/llvm/test/Other/new-pm-thinlto-postlink-samplepgo-defaults.ll b/llvm/test/Other/new-pm-thinlto-postlink-samplepgo-defaults.ll
index 20f94bc2e0f6c..4c330f44d30cc 100644
--- a/llvm/test/Other/new-pm-thinlto-postlink-samplepgo-defaults.ll
+++ b/llvm/test/Other/new-pm-thinlto-postlink-samplepgo-defaults.ll
@@ -174,6 +174,7 @@
 ; CHECK-O-NEXT: Running analysis: LoopAccessAnalysis
 ; CHECK-O-NEXT: Running pass: InjectTLIMappings
 ; CHECK-O-NEXT: Running pass: LoopVectorizePass
+; CHECK-O-NEXT: Running pass: DropUnnecessaryAssumesPass
 ; CHECK-O-NEXT: Running pass: InferAlignmentPass
 ; CHECK-O-NEXT: Running pass: LoopLoadEliminationPass
 ; CHECK-O-NEXT: Running pass: InstCombinePass
diff --git a/llvm/test/Transforms/DropUnnecessaryAssumes/dereferenceable.ll b/llvm/test/Transforms/DropUnnecessaryAssumes/dereferenceable.ll
new file mode 100644
index 0000000000000..43fa08c070828
--- /dev/null
+++ b/llvm/test/Transforms/DropUnnecessaryAssumes/dereferenceable.ll
@@ -0,0 +1,54 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
+; RUN: opt -passes='drop-unnecessary-assumes' -S %s | FileCheck %s
+; RUN: opt -passes='drop-unnecessary-assumes<drop-deref>' -S %s | FileCheck --check-prefix=DROP-DEREF %s
+
+declare void @use(ptr)
+
+define i8 @test_dereferenceable_assume_ptr_not_used(ptr %p, i64 %size) {
+; CHECK-LABEL: define i8 @test_dereferenceable_assume_ptr_not_used(
+; CHECK-SAME: ptr [[P:%.*]], i64 [[SIZE:%.*]]) {
+; CHECK-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(ptr [[P]], i64 [[SIZE]]) ]
+; CHECK-NEXT:    ret i8 0
+;
+; DROP-DEREF-LABEL: define i8 @test_dereferenceable_assume_ptr_not_used(
+; DROP-DEREF-SAME: ptr [[P:%.*]], i64 [[SIZE:%.*]]) {
+; DROP-DEREF-NEXT:    ret i8 0
+;
+  call void @llvm.assume(i1 true) [ "dereferenceable"(ptr %p, i64 %size) ]
+  ret i8 0
+}
+
+define i8 @test_dereferenceable_assume_ptr_used_variable_size(ptr %p, i64 %size) {
+; CHECK-LABEL: define i8 @test_dereferenceable_assume_ptr_used_variable_size(
+; CHECK-SAME: ptr [[P:%.*]], i64 [[SIZE:%.*]]) {
+; CHECK-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(ptr [[P]], i64 [[SIZE]]) ]
+; CHECK-NEXT:    [[VAL:%.*]] = load i8, ptr [[P]], align 1
+; CHECK-NEXT:    ret i8 [[VAL]]
+;
+; DROP-DEREF-LABEL: define i8 @test_dereferenceable_assume_ptr_used_variable_size(
+; DROP-DEREF-SAME: ptr [[P:%.*]], i64 [[SIZE:%.*]]) {
+; DROP-DEREF-NEXT:    [[VAL:%.*]] = load i8, ptr [[P]], align 1
+; DROP-DEREF-NEXT:    ret i8 [[VAL]]
+;
+  call void @llvm.assume(i1 true) [ "dereferenceable"(ptr %p, i64 %size) ]
+  %val = load i8, ptr %p
+  ret i8 %val
+}
+
+define i8 @test_dereferenceable_with_align_ptr_used(ptr %p, i64 %size) {
+; CHECK-LABEL: define i8 @test_dereferenceable_with_align_ptr_used(
+; CHECK-SAME: ptr [[P:%.*]], i64 [[SIZE:%.*]]) {
+; CHECK-NEXT:    call void @llvm.assume(i1 true) [ "dereferenceable"(ptr [[P]], i64 [[SIZE]]), "align"(ptr [[P]], i64 8) ]
+; CHECK-NEXT:    [[VAL:%.*]] = load i8, ptr [[P]], align 1
+; CHECK-NEXT:    ret i8 [[VAL]]
+;
+; DROP-DEREF-LABEL: define i8 @test_dereferenceable_with_align_ptr_used(
+; DROP-DEREF-SAME: ptr [[P:%.*]], i64 [[SIZE:%.*]]) {
+; DROP-DEREF-NEXT:    call void @llvm.assume(i1 true) [ "align"(ptr [[P]], i64 8) ]
+; DROP-DEREF-NEXT:    [[VAL:%.*]] = load i8, ptr [[P]], align 1
+; DROP-DEREF-NEXT:    ret i8 [[VAL]]
+;
+  call void @llvm.assume(i1 true) [ "dereferenceable"(ptr %p, i64 %size), "align"(ptr %p, i64 8) ]
+  %val = load i8, ptr %p
+  ret i8 %val
+}
diff --git a/llvm/test/Transforms/PhaseOrdering/AArch64/matrix-extract-insert.ll b/llvm/test/Transforms/PhaseOrdering/AArch64/matrix-extract-insert.ll
index e3765ed541e7a..75276c0412647 100644
--- a/llvm/test/Transforms/PhaseOrdering/AArch64/matrix-extract-insert.ll
+++ b/llvm/test/Transforms/PhaseOrdering/AArch64/matrix-extract-insert.ll
@@ -106,23 +106,6 @@ define void @matrix_extract_insert_loop(i32 %i, ptr nonnull align 8 dereferencea
 ; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
 ; CHECK:       vector.body:
 ; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NEXT:    [[TMP4:%.*]] = or disjoint i64 [[INDEX]], 1
-; CHECK-NEXT:    [[TMP5:%.*]] = or disjoint i64 [[INDEX]], 2
-; CHECK-NEXT:    [[TMP6:%.*]] = or disjoint i64 [[INDEX]], 3
-; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <2 x i64> poison, i64 [[INDEX]], i64 0
-; CHECK-NEXT:    [[TMP8:%.*]] = insertelement <2 x i64> [[TMP7]], i64 [[TMP4]], i64 1
-; CHECK-NEXT:    [[TMP9:%.*]] = insertelement <2 x i64> poison, i64 [[TMP5]], i64 0
-; CHECK-NEXT:    [[TMP10:%.*]] = insertelement <2 x i64> [[TMP9]], i64 [[TMP6]], i64 1
-; CHECK-NEXT:    [[TMP11:%.*]] = icmp ult <2 x i64> [[TMP8]], splat (i64 225)
-; CHECK-NEXT:    [[TMP13:%.*]] = extractelement <2 x i1> [[TMP11]], i64 0
-; CHECK-NEXT:    [[TMP14:%.*]] = extractelement <2 x i1> [[TMP11]], i64 1
-; CHECK-NEXT:    [[TMP12:%.*]] = icmp ult <2 x i64> [[TMP10]], splat (i64 225)
-; CHECK-NEXT:    [[TMP15:%.*]] = extractelement <2 x i1> [[TMP12]], i64 0
-; CHECK-NEXT:    [[TMP16:%.*]] = extractelement <2 x i1> [[TMP12]], i64 1
-; CHECK-NEXT:    tail call void @llvm.assume(i1 [[TMP13]])
-; CHECK-NEXT:    tail call void @llvm.assume(i1 [[TMP14]])
-; CHECK-NEXT:    tail call void @llvm.assume(i1 [[TMP15]])
-; CHECK-NEXT:    tail call void @llvm.assume(i1 [[TMP16]])
 ; CHECK-NEXT:    [[TMP17:%.*]] = getelementptr inbounds nuw double, ptr [[A]], i64 [[INDEX]]
 ; CHECK-NEXT:    [[TMP18:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP17]], i64 16
 ; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <2 x double>, ptr [[TMP17]], align 8, !alias.scope [[META0:![0-9]+]]
@@ -182,23 +165,6 @@ define void @matrix_extract_insert_loop(i32 %i, ptr nonnull align 8 dereferencea
 ; CHECK:       vector.body.1:
 ; CHECK-NEXT:    [[INDEX_1:%.*]] = phi i64 [ 0, [[VECTOR_PH_1]] ], [ [[INDEX_NEXT_1:%.*]], [[VECTOR_BODY_1]] ]
 ; CHECK-NEXT:    [[TMP33:%.*]] = add nuw nsw i64 [[INDEX_1]], 15
-; CHECK-NEXT:    [[TMP34:%.*]] = add nuw nsw i64 [[INDEX_1]], 16
-; CHECK-NEXT:    [[TMP35:%.*]] = insertelement <2 x i64> poison, i64 [[TMP33]], i64 0
-; CHECK-NEXT:    [[TMP36:%.*]] = insertelement <2 x i64> [[TMP35]], i64 [[TMP34]], i64 1
-; CHECK-NEXT:    [[TMP37:%.*]] = add nuw nsw i64 [[INDEX_1]], 17
-; CHECK-NEXT:    [[TMP38:%.*]] = add nuw nsw i64 [[INDEX_1]], 18
-; CHECK-NEXT:    [[TMP39:%.*]] = insertelement <2 x i64> poison, i64 [[TMP37]], i64 0
-; CHECK-NEXT:    [[TMP40:%.*]] = insertelement <2 x i64> [[TMP39]], i64 [[TMP38]], i64 1
-; CHECK-NEXT:    [[TMP41:%.*]] = icmp ult <2 x i64> [[TMP36]], splat (i64 225)
-; CHECK-NEXT:    [[TMP43:%.*]] = extractelement <2 x i1> [[TMP41]], i64 0
-; CHECK-NEXT:    [[TMP44:%.*]] = extractelement <2 x i1> [[TMP41]], i64 1
-; CHECK-NEXT:    [[TMP42:%.*]] = icmp ult <2 x i64> [[TMP40]], splat (i64 225)
-; CHECK-NEXT:    [[TMP45:%.*]] = extractelement <2 x i1> [[TMP42]], i64 0
-; CHECK-NEXT:    [[TMP46:%.*]] = extractelement <2 x i1> [[TMP42]], i64 1
-; CHECK-NEXT:    tail call void @llvm.assume(i1 [[TMP43]])
-; CHECK-NEXT:    tail call void @llvm.assume(i1 [[TMP44]])
-; CHECK-NEXT:    tail call void @llvm.assume(i1 [[TMP45]])
-; CHECK-NEXT:    tail call void @llvm.assume(i1 [[TMP46]])
 ; CHECK-NEXT:    [[TMP47:%.*]] = getelementptr inbounds nuw double, ptr [[A]], i64 [[TMP33]]
 ; CHECK-NEXT:    [[TMP48:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP47]], i64 16
 ; CHECK-NEXT:    [[WIDE_LOAD_1:%.*]] = load <2 x double>, ptr [[TMP47]], align 8, !alias.scope [[META0]]
@@ -259,23 +225,6 @@ define void @matrix_extract_insert_loop(i32 %i, ptr nonnull align 8 dereferencea
 ; CHECK:       vector.body.2:
 ; CHECK-NEXT:    [[INDEX_2:%.*]] = phi i64 [ 0, [[VECTOR_PH_2]] ], [ [[INDEX_NEXT_2:%.*]], [[VECTOR_BODY_2]] ]
 ; CHECK-NEXT:    [[TMP64:%.*]] = add nuw nsw i64 [[INDEX_2]], 30
-; CHECK-NEXT:    [[TMP65:%.*]] = add nuw nsw i64 [[INDEX_2]], 31
-; CHECK-NEXT:    [[TMP66:%.*]] = insertelement <2 x i64> poison, i64 [[TMP64]], i64 0
-; CHECK-NEXT:    [[TMP67:%.*]] = insertelement <2 x i64> [[TMP66]], i64 [[TMP65]], i64 1
-; CHECK-NEXT:    [[TMP68:%.*]] = add nuw nsw i64 [[INDEX_2]], 32
-; CHECK-NEXT:    [[TMP69:%.*]] = add nuw nsw i64 [[INDEX_2]], 33
-; CHECK-NEXT:    [[TMP70:%.*]] = insertelement <2 x i64> poison, i64 [[TMP68]], i64 0
-; CHECK-NEXT:    [[TMP71:%.*]] = insertelement <2 x i64> [[TMP70]], i64 [[TMP69]], i64 1
-; CHECK-NEXT:    [[TMP72:%.*]] = icmp ult <2 x i64> [[TMP67]], splat (i64 225)
-; CHECK-NEXT:    [[TMP74:%.*]] = extractelement <2 x i1> [[TMP72]], i64 0
-; CHECK-NEXT:    [[TMP75:%.*]] = extractelement <2 x i1> [[TMP72]], i64 1
-; CHECK-NEXT:    [[TMP73:%.*]] = icmp ult <2 x i64> [[TMP71]], splat (i64 225)
-; CHECK-NEXT:    [[TMP76:%.*]] = extractelement <2 x i1> [[TMP73]], i64 0
-; CHECK-NEXT:    [[TMP77:%.*]] = extractelement <2 x i1> [[TMP73]], i64 1
-; CHECK-NEXT:    tail call void @llvm.assume(i1 [[TMP74]])
-; CHECK-NEXT:    tail call void @llvm.assume(i1 [[TMP75]])
-; CHECK-NEXT:    tail call void @llvm.assume(i1 [[TMP76]])
-; CHECK-NEXT:    tail call void @llvm.assume(i1 [[TMP77]])
 ; CHECK-NEXT:    [[TMP78:%.*]] = getelementptr inbounds nuw double, ptr [[A]], i64 [[TMP64]]
 ; CHECK-NEXT:    [[TMP79:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP78]], i64 16
 ; CHECK-NEXT:    [[WIDE_LOAD_2:%.*]] = load <2 x double>, ptr [[TMP78]], align 8, !alias.scope [[META0]]
@@ -336,23 +285,6 @@ define void @matrix_extract_insert_loop(i32 %i, ptr nonnull align 8 dereferencea
 ; CHECK:       vector.body.3:
 ; CHECK-NEXT:    [[INDEX_3:%.*]] = phi i64 [ 0, [[VECTOR_PH_3]] ], [ [[INDEX_NEXT_3:%.*]], [[VECTOR_BODY_3]] ]
 ; CHECK-NEXT:    [[TMP95:%.*]] = add nuw nsw i64 [[INDEX_3]], 45
-; CHECK-NEXT:    [[TMP96:%.*]] = add nuw nsw i64 [[INDEX_3]], 46
-; CHECK-NEXT:    [[TMP97:%.*]] = insertelement <2 x i64> poison, i64 [[TMP95]], i64 0
-; CHECK-NEXT:    [[TMP98:%.*]] = insertelement <2 x i64> [[TMP97]], i64 [[TMP96]], i64 1
-; CHECK-NEXT:    [[TMP99:%.*]] = add nuw nsw i64 [[INDEX_3]], 47
-; CHECK-NEXT:    [[TMP100:%.*]] = add nuw nsw i64 [[INDEX_3]], 48
-; CHECK-NEXT:    [[TMP101:%.*]] = insertelement <2 x i64> poison, i64 [[TMP99]], i64 0
-; CHECK-NEXT:    [[TMP102:%.*]] = insertelement <2 x i64> [[TMP101]], i64 [[TMP100]], i64 1
-; CHECK-NEXT:    [[TMP103:%.*]] = icmp ult <2 x i64> [[TMP98]], splat (i64 225)
-; CHECK-NEXT:    [[TMP105:%.*]] = extractelement <2 x i1> [[TMP103]], i64 0
-; CHECK-NEXT:    [[TMP106:%.*]] = extractelement <2 x i1> [[TMP103]], i64 1
-; CHECK-NEXT:    [[TMP104:%.*]] = icmp ult <2 x i64> [[TMP102]], splat (i64 225)
-; CHECK-NEXT:    [[TMP107:%.*]] = extractelement <2 x i1> [[TMP104]], i64 0
-; CHECK-NEXT:    [[TMP108:%.*]] = extractelement <2 x i1> [[TMP104]], i64 1
-; CHECK-NEXT:    tail call void @llvm.assume(i1 [[TMP105]])
-; CHECK-NEXT:    tail call void @llvm.assume(i1 [[TMP106]])
-; CHECK-NEXT:    tail call void @llvm.assume(i1 [[TMP107]])
-; CHECK-NEXT:    tail call void @llvm.assume(i1 [[TMP108]])
 ; CHECK-NEXT:    [[TMP109:%.*]] = getelementptr inbounds nuw double, ptr [[A]], i64 [[TMP95]]
 ; CHECK-NEXT:    [[TMP110:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP109]], i64 16
 ; CHECK-NEXT:    [[WIDE_LOAD_3:%.*]] = load <2 x double>, ptr [[TMP109]], align 8, !alias.scope [[META0]]
diff --git a/llvm/test/Transforms/PhaseOrdering/AArch64/std-find.ll b/llvm/test/Transforms/PhaseOrdering/AArch64/std-find.ll
index 33e3e83770e7f..823f8dcf83ce2 100644
--- a/llvm/test/T...
[truncated]

dtcxzyw

LGTM

antoniofrighetto

LGTM too, thanks.

…rization. (#166947) This patch adds another run of DropUnnecessaryAssumes after vectorization, to clean up assumes that are not longer needed after this point. The main example of such an assume is currently dereferenceable assumptions. This complements llvm/llvm-project#166945, which avoids sinking code if it would mean remove a dereferenceable assumption. There are a few additional cases where some unneeded assumes are left over after vectorization that also get cleaned up. The main motivation is to work together with llvm/llvm-project#166945, but there may be a better solution. Adding another instance of this pass to the pipeline is not great, but compile-time impact seems in the noise: https://llvm-compile-time-tracker.com/compare.php?from=55e71fe08b6406ec7ce2c81ce042e48717acf204&to=85da4ee3a74126f557cdc74c7b40e048dacb3fc4&stat=instructions:u PR: llvm/llvm-project#166947

* main: (1028 commits) [clang][DebugInfo] Attach `DISubprogram` to additional call variants (llvm#166202) [C2y] Claim nonconformance to WG14 N3348 (llvm#166966) [X86] 2012-01-10-UndefExceptionEdge.ll - regenerate test checks (llvm#167307) Remove unused standard headers: <string>, <optional>, <numeric>, <tuple> (llvm#167232) [DebugInfo] Add Verifier check for incorrectly-scoped retainedNodes (llvm#166855) [VPlan] Don't apply predication discount to non-originally-predicated blocks (llvm#160449) [libc++] Avoid overloaded `operator,` for (`T`, `Iter`) cases (llvm#161049) [tools][llc] Make save-stats.ll test target independent (llvm#167238) [AArch64] Fallback to PRFUM for PRFM with negative or unaligned offset (llvm#166756) [X86] ldexp-avx512.ll - add v8f16/v16f16/v32f16 test coverage for llvm#165694 (llvm#167294) [DropAssumes] Drop dereferenceable assumptions after vectorization. (llvm#166947) [VPlan] Simplify branch-cond with getVectorTripCount (llvm#155604) Remove unused <algorithm> inclusion (llvm#166942) [AArch64] Combine subtract with borrow to SBC. (llvm#165271) [AArch64][SVE] Avoid redundant extend of unsigned i8/i16 extracts. (llvm#165863) [SPIRV] Fix failing assertion in SPIRVAsmPrinter (llvm#166909) [libc++] Merge insert/emplace(const_iterator, Args...) implementations (llvm#166470) [libc++] Replace __libcpp_is_final with a variable template (llvm#167137) [gn build] Port 152bda7 [libc++] Replace the last uses of __tuple_types with __type_list (llvm#167214) ...

artagnon

Very nice patch, thanks! This is probably the best test diff for DropAssumes that I've seen!

fhahn added 2 commits November 7, 2025 14:53

[DropAssumes] Add tests with dereferenceable assumptions.

e4cde4d

[DropAssumes] Drop dereferenceable assumptions after vectorization.

b964d08

llvmbot added the llvm:transforms label Nov 7, 2025

fhahn requested review from antoniofrighetto, dtcxzyw, goldsteinn and nikic November 7, 2025 15:01

dtcxzyw approved these changes Nov 7, 2025

View reviewed changes

antoniofrighetto approved these changes Nov 7, 2025

View reviewed changes

fhahn added 2 commits November 10, 2025 09:12

Merge branch 'main' into perf/drop-deref-assumes

0aebaab

!fixup update test

7c7dc16

fhahn merged commit 80fa6e1 into llvm:main Nov 10, 2025
8 of 9 checks passed

fhahn deleted the perf/drop-deref-assumes branch November 10, 2025 10:52

artagnon reviewed Nov 11, 2025

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[DropAssumes] Drop dereferenceable assumptions after vectorization. #166947

[DropAssumes] Drop dereferenceable assumptions after vectorization. #166947

Uh oh!

fhahn commented Nov 7, 2025 •

edited by dtcxzyw

Loading

Uh oh!

llvmbot commented Nov 7, 2025

Uh oh!

dtcxzyw left a comment

Uh oh!

antoniofrighetto left a comment

Uh oh!

Uh oh!

artagnon left a comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

[DropAssumes] Drop dereferenceable assumptions after vectorization. #166947

[DropAssumes] Drop dereferenceable assumptions after vectorization. #166947

Uh oh!

Conversation

fhahn commented Nov 7, 2025 • edited by dtcxzyw Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

llvmbot commented Nov 7, 2025

Uh oh!

dtcxzyw left a comment

Choose a reason for hiding this comment

Uh oh!

antoniofrighetto left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

artagnon left a comment

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

fhahn commented Nov 7, 2025 •

edited by dtcxzyw

Loading