From 6b825e50f7f41f40cb473587bfe65bcb0069da47 Mon Sep 17 00:00:00 2001 From: Pavel Samolysov Date: Thu, 28 Apr 2022 09:37:35 -0700 Subject: [PATCH] [ArgPromotion] Change the condition to check the promotion limit The condition should be 'ArgParts.size() > MaxElements', so that if we have exactly 3 elements in the 'ArgParts' vector, the promotion should be allowed because the 'MaxElement' threshold is not exceeded yet. The default value for 'MaxElement' has been decreased to 2 in order to avoid an actual change in argument promoting behavior. However, this changes byval argument transformation behavior by allowing adding not more than 2 arguments to the function instead of 3 allowed before. Reviewed By: aeubanks Differential Revision: https://reviews.llvm.org/D124178 --- .../llvm/Transforms/IPO/ArgumentPromotion.h | 2 +- llvm/lib/Transforms/IPO/ArgumentPromotion.cpp | 2 +- .../ArgumentPromotion/max-elements-limit.ll | 90 ++++++++++++++++++++++ 3 files changed, 92 insertions(+), 2 deletions(-) create mode 100644 llvm/test/Transforms/ArgumentPromotion/max-elements-limit.ll diff --git a/llvm/include/llvm/Transforms/IPO/ArgumentPromotion.h b/llvm/include/llvm/Transforms/IPO/ArgumentPromotion.h index 3ea5f8e..3548184 100644 --- a/llvm/include/llvm/Transforms/IPO/ArgumentPromotion.h +++ b/llvm/include/llvm/Transforms/IPO/ArgumentPromotion.h @@ -25,7 +25,7 @@ class ArgumentPromotionPass : public PassInfoMixin { unsigned MaxElements; public: - ArgumentPromotionPass(unsigned MaxElements = 3u) : MaxElements(MaxElements) {} + ArgumentPromotionPass(unsigned MaxElements = 2u) : MaxElements(MaxElements) {} /// Checks if a type could have padding bytes. static bool isDenselyPacked(Type *Ty, const DataLayout &DL); diff --git a/llvm/lib/Transforms/IPO/ArgumentPromotion.cpp b/llvm/lib/Transforms/IPO/ArgumentPromotion.cpp index f8d94af..81faf2a 100644 --- a/llvm/lib/Transforms/IPO/ArgumentPromotion.cpp +++ b/llvm/lib/Transforms/IPO/ArgumentPromotion.cpp @@ -509,7 +509,7 @@ static bool findArgParts(Argument *Arg, const DataLayout &DL, AAResults &AAR, // We limit promotion to only promoting up to a fixed number of elements of // the aggregate. - if (MaxElements > 0 && ArgParts.size() >= MaxElements) { + if (MaxElements > 0 && ArgParts.size() > MaxElements) { LLVM_DEBUG(dbgs() << "ArgPromotion of " << *Arg << " failed: " << "more than " << MaxElements << " parts\n"); return false; diff --git a/llvm/test/Transforms/ArgumentPromotion/max-elements-limit.ll b/llvm/test/Transforms/ArgumentPromotion/max-elements-limit.ll new file mode 100644 index 0000000..7e59c6d --- /dev/null +++ b/llvm/test/Transforms/ArgumentPromotion/max-elements-limit.ll @@ -0,0 +1,90 @@ +; RUN: opt -passes=argpromotion -S %s | FileCheck %s + +define internal i32 @callee2(i32* noundef %0) { +; CHECK-LABEL: define {{[^@]+}}@callee2 +; CHECK-SAME: (i32 [[P_0:%.*]], i32 [[P_1:%.*]]) { +; CHECK-NEXT: [[SUM:%.*]] = add nsw i32 [[P_0]], [[P_1]] +; CHECK-NEXT: ret i32 [[SUM]] +; + %2 = getelementptr inbounds i32, i32* %0, i64 0 + %3 = load i32, i32* %2, align 4 + %4 = getelementptr inbounds i32, i32* %0, i64 1 + %5 = load i32, i32* %4, align 4 + %6 = add nsw i32 %3, %5 + ret i32 %6 +} + +define i32 @caller2(i32 %0, i32 %1) { +; CHECK-LABEL: define {{[^@]+}}@caller2 +; CHECK-SAME: (i32 [[P_0:%.*]], i32 [[P_1:%.*]]) { +; CHECK-NEXT: [[TMP1:%.*]] = alloca [2 x i32], align 4 +; CHECK-NEXT: [[PL_0:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[TMP1]], i64 0, i64 0 +; CHECK-NEXT: store i32 [[P_0]], i32* [[PL_0]], align 4 +; CHECK-NEXT: [[PL_1:%.*]] = getelementptr inbounds i32, i32* [[PL_0]], i64 1 +; CHECK-NEXT: store i32 [[P_1]], i32* [[PL_1]], align 4 +; CHECK-NEXT: [[PL_2:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[TMP1]], i64 0, i64 0 +; CHECK-NEXT: [[VAL_0:%.*]] = load i32, i32* [[PL_2]], align 4 +; CHECK-NEXT: [[PL_3:%.*]] = getelementptr i32, i32* [[PL_2]], i64 1 +; CHECK-NEXT: [[VAL_1:%.*]] = load i32, i32* [[PL_3]], align 4 +; CHECK-NEXT: [[RES:%.*]] = call i32 @callee2(i32 [[VAL_0]], i32 [[VAL_1]]) +; CHECK-NEXT: ret i32 [[RES]] +; + %3 = alloca [2 x i32], align 4 + %4 = getelementptr inbounds [2 x i32], [2 x i32]* %3, i64 0, i64 0 + store i32 %0, i32* %4, align 4 + %5 = getelementptr inbounds i32, i32* %4, i64 1 + store i32 %1, i32* %5, align 4 + %6 = getelementptr inbounds [2 x i32], [2 x i32]* %3, i64 0, i64 0 + %7 = call i32 @callee2(i32* noundef %6) + ret i32 %7 +} + +define internal i32 @callee3(i32* noundef %0) { +; CHECK-LABEL: define {{[^@]+}}@callee3 +; CHECK-SAME: (i32* noundef [[P_0:%.*]]) { +; CHECK-NEXT: [[PL_0:%.*]] = getelementptr inbounds i32, i32* [[P_0]], i64 0 +; CHECK-NEXT: [[VAL_0:%.*]] = load i32, i32* [[PL_0]], align 4 +; CHECK-NEXT: [[PL_1:%.*]] = getelementptr inbounds i32, i32* [[P_0]], i64 1 +; CHECK-NEXT: [[VAL_1:%.*]] = load i32, i32* [[PL_1]], align 4 +; CHECK-NEXT: [[SUM_0:%.*]] = add nsw i32 [[VAL_0]], [[VAL_1]] +; CHECK-NEXT: [[PL_2:%.*]] = getelementptr inbounds i32, i32* [[P_0]], i64 2 +; CHECK-NEXT: [[VAL_2:%.*]] = load i32, i32* [[PL_2]], align 4 +; CHECK-NEXT: [[RES:%.*]] = add nsw i32 [[SUM_0]], [[VAL_2]] +; CHECK-NEXT: ret i32 [[RES]] +; + %2 = getelementptr inbounds i32, i32* %0, i64 0 + %3 = load i32, i32* %2, align 4 + %4 = getelementptr inbounds i32, i32* %0, i64 1 + %5 = load i32, i32* %4, align 4 + %6 = add nsw i32 %3, %5 + %7 = getelementptr inbounds i32, i32* %0, i64 2 + %8 = load i32, i32* %7, align 4 + %9 = add nsw i32 %6, %8 + ret i32 %9 +} + +define i32 @caller3(i32 %0, i32 %1, i32 %2) { +; CHECK-LABEL: define {{[^@]+}}@caller3 +; CHECK-SAME: (i32 [[P_0:%.*]], i32 [[P_1:%.*]], i32 [[P_2:%.*]]) { +; CHECK-NEXT: [[TMP1:%.*]] = alloca [3 x i32], align 4 +; CHECK-NEXT: [[PL_0:%.*]] = getelementptr inbounds [3 x i32], [3 x i32]* [[TMP1]], i64 0, i64 0 +; CHECK-NEXT: store i32 [[P_0]], i32* [[PL_0]], align 4 +; CHECK-NEXT: [[PL_1:%.*]] = getelementptr inbounds i32, i32* [[PL_0]], i64 1 +; CHECK-NEXT: store i32 [[P_1]], i32* [[PL_1]], align 4 +; CHECK-NEXT: [[PL_2:%.*]] = getelementptr inbounds i32, i32* [[PL_1]], i64 1 +; CHECK-NEXT: store i32 [[P_2]], i32* [[PL_2]], align 4 +; CHECK-NEXT: [[PL_3:%.*]] = getelementptr inbounds [3 x i32], [3 x i32]* [[TMP1]], i64 0, i64 0 +; CHECK-NEXT: [[RES:%.*]] = call i32 @callee3(i32* noundef [[PL_3]]) +; CHECK-NEXT: ret i32 [[RES]] +; + %4 = alloca [3 x i32], align 4 + %5 = getelementptr inbounds [3 x i32], [3 x i32]* %4, i64 0, i64 0 + store i32 %0, i32* %5, align 4 + %6 = getelementptr inbounds i32, i32* %5, i64 1 + store i32 %1, i32* %6, align 4 + %7 = getelementptr inbounds i32, i32* %6, i64 1 + store i32 %2, i32* %7, align 4 + %8 = getelementptr inbounds [3 x i32], [3 x i32]* %4, i64 0, i64 0 + %9 = call i32 @callee3(i32* noundef %8) + ret i32 %9 +} -- 2.7.4