Don't unroll loops containing function calls.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@95454 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/Scalar/LoopUnrollPass.cpp b/lib/Transforms/Scalar/LoopUnrollPass.cpp
index ee8cb4f9a732db0c7ab3cdadb7c2b34a947816bb..a355ec3a7e68ff0245db1407887cf86b2d00dde6 100644 (file)
--- a/lib/Transforms/Scalar/LoopUnrollPass.cpp
+++ b/lib/Transforms/Scalar/LoopUnrollPass.cpp
@@ -76,11 +76,12 @@ static RegisterPass<LoopUnroll> X("loop-unroll", "Unroll loops");
 Pass *llvm::createLoopUnrollPass() { return new LoopUnroll(); }
 
 /// ApproximateLoopSize - Approximate the size of the loop.
-static unsigned ApproximateLoopSize(const Loop *L) {
+static unsigned ApproximateLoopSize(const Loop *L, unsigned &NumCalls) {
   CodeMetrics Metrics;
   for (Loop::block_iterator I = L->block_begin(), E = L->block_end();
        I != E; ++I)
     Metrics.analyzeBasicBlock(*I);
+  NumCalls = Metrics.NumCalls;
   return Metrics.NumInsts;
 }
 
@@ -110,8 +111,13 @@ bool LoopUnroll::runOnLoop(Loop *L, LPPassManager &LPM) {
 
   // Enforce the threshold.
   if (UnrollThreshold != NoThreshold) {
-    unsigned LoopSize = ApproximateLoopSize(L);
+    unsigned NumCalls;
+    unsigned LoopSize = ApproximateLoopSize(L, NumCalls);
     DEBUG(dbgs() << "  Loop Size = " << LoopSize << "\n");
+    if (NumCalls != 0) {
+      DEBUG(dbgs() << "  Not unrolling loop with function calls.\n");
+      return false;
+    }
     uint64_t Size = (uint64_t)LoopSize*Count;
     if (TripCount != 1 && Size > UnrollThreshold) {
       DEBUG(dbgs() << "  Too large to fully unroll with count: " << Count

diff --git a/test/CodeGen/X86/nancvt.ll b/test/CodeGen/X86/nancvt.ll
index 0b56644f125a453c1bb4d88205feb6cf16ad16e8..82b73319ad149e13d139f58ca1063dac1edf8d42 100644 (file)
--- a/test/CodeGen/X86/nancvt.ll
+++ b/test/CodeGen/X86/nancvt.ll
@@ -16,6 +16,8 @@ target triple = "i686-apple-darwin8"
 @.str = internal constant [10 x i8] c"%08x%08x\0A\00"          ; <[10 x i8]*> [#uses=2]
 @.str1 = internal constant [6 x i8] c"%08x\0A\00"              ; <[6 x i8]*> [#uses=2]
 
+@var = external global i32
+
 define i32 @main() {
 entry:
        %retval = alloca i32, align 4           ; <i32*> [#uses=1]
@@ -50,7 +52,8 @@ bb:           ; preds = %bb23
        %tmp17 = ashr i64 %tmp16, %.cast                ; <i64> [#uses=1]
        %tmp1718 = trunc i64 %tmp17 to i32              ; <i32> [#uses=1]
        %tmp19 = getelementptr [10 x i8]* @.str, i32 0, i32 0           ; <i8*> [#uses=1]
-       %tmp20 = call i32 (i8*, ...)* @printf( i8* %tmp19, i32 %tmp1718, i32 %tmp13 )           ; <i32> [#uses=0]
+       volatile store i32 %tmp1718, i32* @var
+       volatile store i32 %tmp13, i32* @var
        %tmp21 = load i32* %i, align 4          ; <i32> [#uses=1]
        %tmp22 = add i32 %tmp21, 1              ; <i32> [#uses=1]
        store i32 %tmp22, i32* %i, align 4
@@ -83,7 +86,7 @@ bb28:         ; preds = %bb46
        %tmp3940 = bitcast float* %tmp39 to i32*                ; <i32*> [#uses=1]
        %tmp41 = load i32* %tmp3940, align 4            ; <i32> [#uses=1]
        %tmp42 = getelementptr [6 x i8]* @.str1, i32 0, i32 0           ; <i8*> [#uses=1]
-       %tmp43 = call i32 (i8*, ...)* @printf( i8* %tmp42, i32 %tmp41 )         ; <i32> [#uses=0]
+       volatile store i32 %tmp41, i32* @var
        %tmp44 = load i32* %i, align 4          ; <i32> [#uses=1]
        %tmp45 = add i32 %tmp44, 1              ; <i32> [#uses=1]
        store i32 %tmp45, i32* %i, align 4
@@ -124,7 +127,8 @@ bb52:               ; preds = %bb78
        %tmp72 = ashr i64 %tmp70, %.cast71              ; <i64> [#uses=1]
        %tmp7273 = trunc i64 %tmp72 to i32              ; <i32> [#uses=1]
        %tmp74 = getelementptr [10 x i8]* @.str, i32 0, i32 0           ; <i8*> [#uses=1]
-       %tmp75 = call i32 (i8*, ...)* @printf( i8* %tmp74, i32 %tmp7273, i32 %tmp66 )           ; <i32> [#uses=0]
+       volatile store i32 %tmp7273, i32* @var
+       volatile store i32 %tmp66, i32* @var
        %tmp76 = load i32* %i, align 4          ; <i32> [#uses=1]
        %tmp77 = add i32 %tmp76, 1              ; <i32> [#uses=1]
        store i32 %tmp77, i32* %i, align 4
@@ -157,7 +161,7 @@ bb84:               ; preds = %bb101
        %tmp9495 = bitcast float* %tmp94 to i32*                ; <i32*> [#uses=1]
        %tmp96 = load i32* %tmp9495, align 4            ; <i32> [#uses=1]
        %tmp97 = getelementptr [6 x i8]* @.str1, i32 0, i32 0           ; <i8*> [#uses=1]
-       %tmp98 = call i32 (i8*, ...)* @printf( i8* %tmp97, i32 %tmp96 )         ; <i32> [#uses=0]
+       volatile store i32 %tmp96, i32* @var
        %tmp99 = load i32* %i, align 4          ; <i32> [#uses=1]
        %tmp100 = add i32 %tmp99, 1             ; <i32> [#uses=1]
        store i32 %tmp100, i32* %i, align 4
@@ -177,5 +181,3 @@ return:             ; preds = %bb106
        %retval107 = load i32* %retval          ; <i32> [#uses=1]
        ret i32 %retval107
 }
-
-declare i32 @printf(i8*, ...)