[FuncAttrs][LTO] Relax norecurse attribute inference during postlink LTO #158608
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
@llvm/pr-subscribers-llvm-transforms Author: Usha Gupta (usha1830) ChangesThis PR, which supersedes #139943, extends the scenarios where the 'norecurse' attribute can be inferred. Currently, the 'norecurse' attribute is only inferred if all called functions also have this attribute. This change introduces a new pass in the LTO pipeline, run after Whole Program Devirtualization, to broaden the inference criteria. The new pass inspects all functions in the module and sets a flag if any functions are external or have their addresses taken (while ignoring those already marked norecurse). This flag is then used with the existing conditions to enable inference in more cases. This enhancement allows 'norecurse' to be applied in situations where a function calls a recursive function, but is not part of the same recursion chain. For example, foo can now be marked 'norecurse' in the following scenarios:
Similarly, foo can be marked 'norecurse' here:
Patch is 22.16 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/158608.diff 10 Files Affected:
diff --git a/llvm/include/llvm/Transforms/IPO/FunctionAttrs.h b/llvm/include/llvm/Transforms/IPO/FunctionAttrs.h
index 754714dceb7a6..2724687bb09de 100644
--- a/llvm/include/llvm/Transforms/IPO/FunctionAttrs.h
+++ b/llvm/include/llvm/Transforms/IPO/FunctionAttrs.h
@@ -79,6 +79,18 @@ class ReversePostOrderFunctionAttrsPass
LLVM_ABI PreservedAnalyses run(Module &M, ModuleAnalysisManager &AM);
};
+/// Additional 'norecurse' attribute deduction during postlink LTO phase.
+///
+/// This is a module pass that infers 'norecurse' attribute on functions.
+/// It runs during LTO and analyzes the module's call graph to find functions
+/// that are guaranteed not to call themselves, either directly or indirectly.
+/// The pass uses a module-wide flag which checks if any function's address is
+/// taken or any function in the module has external linkage, to safely handle
+/// indirect and library function calls from current function.
+class NoRecurseLTOInferencePass : public PassInfoMixin<NoRecurseLTOInferencePass> {
+public:
+ LLVM_ABI PreservedAnalyses run(Module &M, ModuleAnalysisManager &MAM);
+};
} // end namespace llvm
#endif // LLVM_TRANSFORMS_IPO_FUNCTIONATTRS_H
diff --git a/llvm/lib/Passes/PassBuilderPipelines.cpp b/llvm/lib/Passes/PassBuilderPipelines.cpp
index 98821bb1408a7..ca16f2e580e85 100644
--- a/llvm/lib/Passes/PassBuilderPipelines.cpp
+++ b/llvm/lib/Passes/PassBuilderPipelines.cpp
@@ -1944,6 +1944,7 @@ PassBuilder::buildLTODefaultPipeline(OptimizationLevel Level,
// is fixed.
MPM.addPass(WholeProgramDevirtPass(ExportSummary, nullptr));
+ MPM.addPass(NoRecurseLTOInferencePass());
// Stop here at -O1.
if (Level == OptimizationLevel::O1) {
// The LowerTypeTestsPass needs to run to lower type metadata and the
@@ -2355,4 +2356,4 @@ AAManager PassBuilder::buildDefaultAAPipeline() {
bool PassBuilder::isInstrumentedPGOUse() const {
return (PGOOpt && PGOOpt->Action == PGOOptions::IRUse) ||
!UseCtxProfile.empty();
-}
\ No newline at end of file
+}
diff --git a/llvm/lib/Passes/PassRegistry.def b/llvm/lib/Passes/PassRegistry.def
index 1d015971dfbdf..2acfd3c1789c8 100644
--- a/llvm/lib/Passes/PassRegistry.def
+++ b/llvm/lib/Passes/PassRegistry.def
@@ -119,6 +119,7 @@ MODULE_PASS("metarenamer", MetaRenamerPass())
MODULE_PASS("module-inline", ModuleInlinerPass())
MODULE_PASS("name-anon-globals", NameAnonGlobalPass())
MODULE_PASS("no-op-module", NoOpModulePass())
+MODULE_PASS("norecurse-lto-inference", NoRecurseLTOInferencePass())
MODULE_PASS("nsan", NumericalStabilitySanitizerPass())
MODULE_PASS("openmp-opt", OpenMPOptPass())
MODULE_PASS("openmp-opt-postlink",
diff --git a/llvm/lib/Transforms/IPO/FunctionAttrs.cpp b/llvm/lib/Transforms/IPO/FunctionAttrs.cpp
index 8d9a0e7eaef63..87fbf911ae713 100644
--- a/llvm/lib/Transforms/IPO/FunctionAttrs.cpp
+++ b/llvm/lib/Transforms/IPO/FunctionAttrs.cpp
@@ -2067,6 +2067,36 @@ static void inferAttrsFromFunctionBodies(const SCCNodeSet &SCCNodes,
AI.run(SCCNodes, Changed);
}
+// Determines if the function 'F' can be marked 'norecurse'.
+// It returns true if any call within 'F' could lead to a recursive
+// call back to 'F', and false otherwise.
+// The 'AnyFunctionsAddressIsTaken' parameter is a module-wide flag
+// that is true if any function's address is taken, or if any function
+// has external linkage. This is used to determine the safety of
+// external/library calls.
+static bool hasRecursiveCallee(Function &F,
+ bool AnyFunctionsAddressIsTaken = true) {
+ for (const auto &BB : F) {
+ for (const auto &I : BB.instructionsWithoutDebug()) {
+ if (const auto *CB = dyn_cast<CallBase>(&I)) {
+ const Function *Callee = CB->getCalledFunction();
+ if (!Callee || Callee == &F)
+ return true;
+
+ if (Callee->doesNotRecurse())
+ continue;
+
+ if (!AnyFunctionsAddressIsTaken ||
+ (Callee->isDeclaration() &&
+ Callee->hasFnAttribute(Attribute::NoCallback)))
+ continue;
+ return true;
+ }
+ }
+ }
+ return false;
+}
+
static void addNoRecurseAttrs(const SCCNodeSet &SCCNodes,
SmallPtrSet<Function *, 8> &Changed) {
// Try and identify functions that do not recurse.
@@ -2078,28 +2108,14 @@ static void addNoRecurseAttrs(const SCCNodeSet &SCCNodes,
Function *F = *SCCNodes.begin();
if (!F || !F->hasExactDefinition() || F->doesNotRecurse())
return;
-
- // If all of the calls in F are identifiable and are to norecurse functions, F
- // is norecurse. This check also detects self-recursion as F is not currently
- // marked norecurse, so any called from F to F will not be marked norecurse.
- for (auto &BB : *F)
- for (auto &I : BB.instructionsWithoutDebug())
- if (auto *CB = dyn_cast<CallBase>(&I)) {
- Function *Callee = CB->getCalledFunction();
- if (!Callee || Callee == F ||
- (!Callee->doesNotRecurse() &&
- !(Callee->isDeclaration() &&
- Callee->hasFnAttribute(Attribute::NoCallback))))
- // Function calls a potentially recursive function.
- return;
- }
-
- // Every call was to a non-recursive function other than this function, and
- // we have no indirect recursion as the SCC size is one. This function cannot
- // recurse.
- F->setDoesNotRecurse();
- ++NumNoRecurse;
- Changed.insert(F);
+ if (!hasRecursiveCallee(*F)) {
+ // Every call was to a non-recursive function other than this function, and
+ // we have no indirect recursion as the SCC size is one. This function
+ // cannot recurse.
+ F->setDoesNotRecurse();
+ ++NumNoRecurse;
+ Changed.insert(F);
+ }
}
// Set the noreturn function attribute if possible.
@@ -2429,3 +2445,66 @@ ReversePostOrderFunctionAttrsPass::run(Module &M, ModuleAnalysisManager &AM) {
PA.preserve<LazyCallGraphAnalysis>();
return PA;
}
+
+PreservedAnalyses NoRecurseLTOInferencePass::run(Module &M,
+ ModuleAnalysisManager &MAM) {
+
+ // Check if any function in the whole program has its address taken or has
+ // potentially external linkage.
+ // We use this information when inferring norecurse attribute: If there is
+ // no function whose address is taken and all functions have internal
+ // linkage, there is no path for a callback to any user function.
+ bool AnyFunctionsAddressIsTaken = false;
+ for (Function &F : M) {
+ if (F.isDeclaration() || F.doesNotRecurse()) {
+ continue;
+ }
+ if (!F.hasLocalLinkage() || F.hasAddressTaken()) {
+ AnyFunctionsAddressIsTaken = true;
+ break;
+ }
+ }
+
+ // Run norecurse inference on all RefSCCs in the LazyCallGraph for this
+ // module.
+ bool Changed = false;
+ LazyCallGraph &CG = MAM.getResult<LazyCallGraphAnalysis>(M);
+ CG.buildRefSCCs();
+
+ for (LazyCallGraph::RefSCC &RC : CG.postorder_ref_sccs()) {
+ // Skip any RefSCC that is part of a call cycle. A RefSCC containing more
+ // than one SCC indicates a recursive relationship, which could involve
+ // direct or indirect calls.
+ if (RC.size() > 1) {
+ continue;
+ }
+
+ // A single-SCC RefSCC could still be a self-loop.
+ LazyCallGraph::SCC &S = *RC.begin();
+ if (S.size() > 1) {
+ continue;
+ }
+
+ // Get the single function from this SCC.
+ Function &F = S.begin()->getFunction();
+ if (!F.hasExactDefinition() || F.doesNotRecurse()) {
+ continue;
+ }
+
+ // If the analysis confirms that this function has no recursive calls
+ // (either direct, indirect, or through external linkages),
+ // we can safely apply the norecurse attribute.
+ if (!hasRecursiveCallee(F, AnyFunctionsAddressIsTaken)) {
+ F.setDoesNotRecurse();
+ ++NumNoRecurse;
+ Changed = true;
+ }
+ }
+
+ PreservedAnalyses PA;
+ if (Changed)
+ PA.preserve<LazyCallGraphAnalysis>();
+ else
+ PA = PreservedAnalyses::all();
+ return PA;
+}
diff --git a/llvm/test/Other/new-pm-lto-defaults.ll b/llvm/test/Other/new-pm-lto-defaults.ll
index 3aea0f2061f3e..f595dfe1d6845 100644
--- a/llvm/test/Other/new-pm-lto-defaults.ll
+++ b/llvm/test/Other/new-pm-lto-defaults.ll
@@ -67,6 +67,7 @@
; CHECK-O1-NEXT: Running analysis: TargetLibraryAnalysis
; CHECK-O-NEXT: Running pass: GlobalSplitPass
; CHECK-O-NEXT: Running pass: WholeProgramDevirtPass
+; CHECK-O-NEXT: Running pass: NoRecurseLTOInferencePass
; CHECK-O23SZ-NEXT: Running pass: CoroEarlyPass
; CHECK-O1-NEXT: Running pass: LowerTypeTestsPass
; CHECK-O23SZ-NEXT: Running pass: GlobalOptPass
diff --git a/llvm/test/Transforms/FunctionAttrs/norecurse_libfunc_address_taken.ll b/llvm/test/Transforms/FunctionAttrs/norecurse_libfunc_address_taken.ll
new file mode 100644
index 0000000000000..0ec36f8147872
--- /dev/null
+++ b/llvm/test/Transforms/FunctionAttrs/norecurse_libfunc_address_taken.ll
@@ -0,0 +1,34 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-attributes --version 5
+; RUN: opt < %s -passes=norecurse-lto-inference -S | FileCheck %s
+
+; This test includes a call to a library function which is not marked as
+; NoCallback. Function bob() does not have internal linkage and hence prevents
+; norecurse to be added.
+
+@.str = private unnamed_addr constant [12 x i8] c"Hello World\00", align 1
+
+define dso_local void @bob() {
+; CHECK-LABEL: define dso_local void @bob() {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[CALL:%.*]] = tail call i32 (ptr, ...) @printf(ptr noundef nonnull dereferenceable(1) @.str)
+; CHECK-NEXT: ret void
+;
+entry:
+ %call = tail call i32 (ptr, ...) @printf(ptr noundef nonnull dereferenceable(1) @.str)
+ ret void
+}
+
+declare noundef i32 @printf(ptr noundef readonly captures(none), ...)
+
+define dso_local noundef i32 @main() norecurse {
+; CHECK: Function Attrs: norecurse
+; CHECK-LABEL: define dso_local noundef i32 @main(
+; CHECK-SAME: ) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: tail call void @bob()
+; CHECK-NEXT: ret i32 0
+;
+entry:
+ tail call void @bob()
+ ret i32 0
+}
diff --git a/llvm/test/Transforms/FunctionAttrs/norecurse_libfunc_no_address_taken.ll b/llvm/test/Transforms/FunctionAttrs/norecurse_libfunc_no_address_taken.ll
new file mode 100644
index 0000000000000..6d13d5262f9f7
--- /dev/null
+++ b/llvm/test/Transforms/FunctionAttrs/norecurse_libfunc_no_address_taken.ll
@@ -0,0 +1,39 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-attributes --version 5
+; RUN: opt < %s -passes=norecurse-lto-inference -S | FileCheck %s
+
+; This test includes a call to a library function which is not marked as
+; NoCallback. All functions except main() are internal and main is marked
+; norecurse, so as to not block norecurse to be added to bob().
+
+@.str = private unnamed_addr constant [12 x i8] c"Hello World\00", align 1
+
+; Function Attrs: nofree noinline nounwind uwtable
+define internal void @bob() {
+; CHECK: Function Attrs: norecurse
+; CHECK-LABEL: define internal void @bob(
+; CHECK-SAME: ) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[CALL:%.*]] = tail call i32 (ptr, ...) @printf(ptr noundef nonnull dereferenceable(1) @.str)
+; CHECK-NEXT: ret void
+;
+entry:
+ %call = tail call i32 (ptr, ...) @printf(ptr noundef nonnull dereferenceable(1) @.str)
+ ret void
+}
+
+; Function Attrs: nofree nounwind
+declare noundef i32 @printf(ptr noundef readonly captures(none), ...)
+
+; Function Attrs: nofree norecurse nounwind uwtable
+define dso_local noundef i32 @main() norecurse {
+; CHECK: Function Attrs: norecurse
+; CHECK-LABEL: define dso_local noundef i32 @main(
+; CHECK-SAME: ) #[[ATTR0]] {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: tail call void @bob()
+; CHECK-NEXT: ret i32 0
+;
+entry:
+ tail call void @bob()
+ ret i32 0
+}
diff --git a/llvm/test/Transforms/FunctionAttrs/norecurse_multi_scc_indirect_recursion.ll b/llvm/test/Transforms/FunctionAttrs/norecurse_multi_scc_indirect_recursion.ll
new file mode 100644
index 0000000000000..8264cf33df4eb
--- /dev/null
+++ b/llvm/test/Transforms/FunctionAttrs/norecurse_multi_scc_indirect_recursion.ll
@@ -0,0 +1,138 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-attributes --version 5
+; RUN: opt < %s -passes=norecurse-lto-inference -S | FileCheck %s
+
+; This test includes a call graph with multiple SCCs. The purpose of this is
+; to check that norecurse is not added when a function is part of non-singular
+; SCC.
+; There are three different SCCs in this test:
+; SCC#1: f1, foo, bar, foo1, bar1
+; SCC#2: bar2, bar3, bar4
+; SCC#3: baz, fun
+; None of these functions should be marked as norecurse
+
+define internal void @bar1() {
+; CHECK-LABEL: define internal void @bar1() {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: tail call void @f1()
+; CHECK-NEXT: ret void
+;
+entry:
+ tail call void @f1()
+ ret void
+}
+
+define internal void @f1() {
+; CHECK-LABEL: define internal void @f1() {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: tail call void @foo()
+; CHECK-NEXT: tail call void @bar2()
+; CHECK-NEXT: tail call void @baz()
+; CHECK-NEXT: ret void
+;
+entry:
+ tail call void @foo()
+ tail call void @bar2()
+ tail call void @baz()
+ ret void
+}
+
+define dso_local noundef i32 @main() norecurse {
+; CHECK: Function Attrs: norecurse
+; CHECK-LABEL: define dso_local noundef i32 @main(
+; CHECK-SAME: ) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: tail call void @f1()
+; CHECK-NEXT: ret i32 0
+;
+entry:
+ tail call void @f1()
+ ret i32 0
+}
+
+define internal void @foo1() {
+; CHECK-LABEL: define internal void @foo1() {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: tail call void @bar1()
+; CHECK-NEXT: ret void
+;
+entry:
+ tail call void @bar1()
+ ret void
+}
+
+define internal void @bar() {
+; CHECK-LABEL: define internal void @bar() {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: tail call void @foo1()
+; CHECK-NEXT: ret void
+;
+entry:
+ tail call void @foo1()
+ ret void
+}
+
+define internal void @foo() {
+; CHECK-LABEL: define internal void @foo() {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: tail call void @bar()
+; CHECK-NEXT: ret void
+;
+entry:
+ tail call void @bar()
+ ret void
+}
+
+define internal void @bar4() {
+; CHECK-LABEL: define internal void @bar4() {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: tail call void @bar2()
+; CHECK-NEXT: ret void
+;
+entry:
+ tail call void @bar2()
+ ret void
+}
+
+define internal void @bar2() {
+; CHECK-LABEL: define internal void @bar2() {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: tail call void @bar3()
+; CHECK-NEXT: ret void
+;
+entry:
+ tail call void @bar3()
+ ret void
+}
+
+define internal void @bar3() {
+; CHECK-LABEL: define internal void @bar3() {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: tail call void @bar4()
+; CHECK-NEXT: ret void
+;
+entry:
+ tail call void @bar4()
+ ret void
+}
+
+define internal void @fun() {
+; CHECK-LABEL: define internal void @fun() {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: tail call void @baz()
+; CHECK-NEXT: ret void
+;
+entry:
+ tail call void @baz()
+ ret void
+}
+
+define internal void @baz() {
+; CHECK-LABEL: define internal void @baz() {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: tail call void @fun()
+; CHECK-NEXT: ret void
+;
+entry:
+ tail call void @fun()
+ ret void
+}
diff --git a/llvm/test/Transforms/FunctionAttrs/norecurse_multi_scc_indirect_recursion1.ll b/llvm/test/Transforms/FunctionAttrs/norecurse_multi_scc_indirect_recursion1.ll
new file mode 100644
index 0000000000000..af986eff3e13c
--- /dev/null
+++ b/llvm/test/Transforms/FunctionAttrs/norecurse_multi_scc_indirect_recursion1.ll
@@ -0,0 +1,95 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-attributes --version 5
+; RUN: opt < %s -passes=norecurse-lto-inference -S | FileCheck %s
+
+; This test includes a call graph with multiple SCCs. The purpose of this is
+; to check that norecurse is added to a function which calls a function which
+; is indirectly recursive but is not part of the recursive chain.
+; There are two SCCs in this test:
+; SCC#1: bar2, bar3, bar4
+; SCC#2: baz, fun
+; f1() calls bar2 and baz, both of which are part of some indirect recursive
+; chain. but does not call back f1() and hence f1() can be marked as
+; norecurse.
+
+define dso_local noundef i32 @main() norecurse {
+; CHECK: Function Attrs: norecurse
+; CHECK-LABEL: define dso_local noundef i32 @main(
+; CHECK-SAME: ) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: tail call void @f1()
+; CHECK-NEXT: ret i32 0
+;
+entry:
+ tail call void @f1()
+ ret i32 0
+}
+
+define internal void @f1() {
+; CHECK: Function Attrs: norecurse
+; CHECK-LABEL: define internal void @f1(
+; CHECK-SAME: ) #[[ATTR0]] {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: tail call void @bar2()
+; CHECK-NEXT: tail call void @baz()
+; CHECK-NEXT: ret void
+;
+entry:
+ tail call void @bar2()
+ tail call void @baz()
+ ret void
+}
+
+define internal void @bar4() {
+; CHECK-LABEL: define internal void @bar4() {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: tail call void @bar2()
+; CHECK-NEXT: ret void
+;
+entry:
+ tail call void @bar2()
+ ret void
+}
+
+define internal void @bar2() {
+; CHECK-LABEL: define internal void @bar2() {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: tail call void @bar3()
+; CHECK-NEXT: ret void
+;
+entry:
+ tail call void @bar3()
+ ret void
+}
+
+define internal void @bar3() {
+; CHECK-LABEL: define internal void @bar3() {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: tail call void @bar4()
+; CHECK-NEXT: ret void
+;
+entry:
+ tail call void @bar4()
+ ret void
+}
+
+define internal void @fun() {
+; CHECK-LABEL: define internal void @fun() {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: tail call void @baz()
+; CHECK-NEXT: ret void
+;
+entry:
+ tail call void @baz()
+ ret void
+}
+
+define internal void @baz() {
+; CHECK-LABEL: define internal void @baz() {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: tail call void @fun()
+; CHECK-NEXT: ret void
+;
+entry:
+ tail call void @fun()
+ ret void
+}
diff --git a/llvm/test/Transforms/FunctionAttrs/norecurse_self_recursive_callee.ll b/llvm/test/Transforms/FunctionAttrs/norecurse_self_recursive_callee.ll
new file mode 100644
index 0000000000000..554642ff6963c
--- /dev/null
+++ b/llvm/test/Transforms/FunctionAttrs/norecurse_self_recursive_callee.ll
@@ -0,0 +1,135 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-attributes --version 5
+; RUN: opt < %s -passes=norecurse-lto-inference -S | FileCheck %s
+
+; This test includes a call graph with a self recursive function.
+; The purpose of this is to check that norecurse is added to functions
+; which have a self-recursive function in the call-chain.
+; The call-chain in this test is as follows
+; main -> bob -> callee1 -> callee2 -> callee3 -> callee4 -> callee5
+; where callee5 is self recursive.
+
+@x = dso_local global i32 4, align 4
+@y = dso_local global i32 2, align 4
+
+define internal void @callee6() {
+; CHECK: Function Attrs: norecurse
+; CHECK-LABEL: define internal void @callee6(
+; CHECK-SAME: ) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[TMP0:%.*]] = load volatile i32, ptr @y, align 4
+; CHECK-NEXT: [[INC:%.*]] = add nsw i32 [[TMP0]], 1
+; CHECK-NEXT: store volatile i32 [[INC]], ptr @y, align 4
+; CHECK-NEXT: ret void
+;
+entry:
+ %0 = load volatile i32, ptr @y, align 4
+ %inc = add nsw i32 %0, 1
+ store volatile i32 %inc, ptr @y, align 4
+ ret void
+}
+
+define internal void @callee5(i32 noundef %x) {
+; CHECK-LABEL: define internal void @callee5(
+; CHECK-SAME: i32 noundef [[X:%.*]]) {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp sgt i32 [[X]], 0
+; CHECK-NEXT: br i1 [[CMP]], label %[[IF_THEN:.*]], label %[[IF_END:.*]]
+; CHECK: [[IF_THEN]]:
+; CHECK-NEXT: tail call void @callee5(i32 noundef [[X]])
+; CHECK-NEXT: br label %[[IF_END]]
+; CHECK: [[IF_END]]:
+; CHECK-NEXT: tail call void @callee6()
+; CHECK-NEXT: ret void
+;
+entry:
+ %cmp = icmp sgt i32 %x, 0
+ br i1 %cmp, label %if.then, label %if.end
+
+if.then: ...
[truncated]
|
usha1830
force-pushed
the
relax_norecurse_lto
branch
from
430c3db to
09220c5
Compare
david-arm
reviewed
Sep 17, 2025
llvm/test/Transforms/FunctionAttrs/norecurse_self_recursive_callee.ll
Outdated
Show resolved
Hide resolved
llvm/test/Transforms/FunctionAttrs/norecurse_self_recursive_callee.ll
Outdated
Show resolved
Hide resolved
llvm/test/Transforms/FunctionAttrs/norecurse_self_recursive_callee.ll
Outdated
Show resolved
Hide resolved
|
Thanks for the review @david-arm. |
david-arm
reviewed
Sep 22, 2025
david-arm
approved these changes
Sep 30, 2025
llvm/test/Transforms/FunctionAttrs/norecurse_multinode_refscc.ll
Outdated
Show resolved
Hide resolved
usha1830
force-pushed
the
relax_norecurse_lto
branch
from
69ec338 to
d160bc8
Compare
Thanks @david-arm. I have just merged it. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR, which supersedes #139943, extends the scenarios where the 'norecurse' attribute can be inferred.
Currently, the 'norecurse' attribute is only inferred if all called functions also have this attribute. This change introduces a new pass in the LTO pipeline, run after Whole Program Devirtualization, to broaden the inference criteria. The new pass inspects all functions in the module and sets a flag if any functions are external or have their addresses taken (while ignoring those already marked norecurse). This flag is then used with the existing conditions to enable inference in more cases.
This enhancement allows 'norecurse' to be applied in situations where a function calls a recursive function, but is not part of the same recursion chain.
For example, foo can now be marked 'norecurse' in the following scenarios:
foo -> callee1 -> callee2 -> callee2In this case, foo and callee1 can both be marked 'norecurse' because they're not part of the callee2 recursion.
Similarly, foo can be marked 'norecurse' here:
foo -> callee1 -> callee2 -> callee1Here, foo is not part of the callee1 -> callee2 -> callee1 recursion chain, so it can be marked 'norecurse'.