Fixed the lost FastMathFlags for CALL operations in SLPVectorizer.

Reviewer: Michael Zolotukhin.
Differential Revision: https://reviews.llvm.org/D26575

llvm-svn: 287064

diff --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index 6b27ac6..aac5aae 100644 (file)
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -2641,6 +2641,7 @@ Value *BoUpSLP::vectorizeTree(TreeEntry *E) {
         ExternalUses.push_back(ExternalUser(ScalarArg, cast<User>(V), 0));
 
       E->VectorizedValue = V;
+      propagateIRFlags(E->VectorizedValue, E->Scalars);
       ++NumVectorInstructions;
       return V;
     }

diff --git a/llvm/test/Transforms/SLPVectorizer/X86/call.ll b/llvm/test/Transforms/SLPVectorizer/X86/call.ll
index d6c0ebd..923cbe7 100644 (file)
--- a/llvm/test/Transforms/SLPVectorizer/X86/call.ll
+++ b/llvm/test/Transforms/SLPVectorizer/X86/call.ll
@@ -98,7 +98,7 @@ entry:
 
 
 ; CHECK: sqrt_libm
-; CHECK: call <2 x double> @llvm.sqrt.v2f64
+; CHECK: call nnan <2 x double> @llvm.sqrt.v2f64
 ; CHECK: ret void
 define void @sqrt_libm(double* %a, double* %b, double* %c) {
 entry:

diff --git a/llvm/test/Transforms/SLPVectorizer/X86/propagate_ir_flags.ll b/llvm/test/Transforms/SLPVectorizer/X86/propagate_ir_flags.ll
index 3f26c81..28217fc 100644 (file)
--- a/llvm/test/Transforms/SLPVectorizer/X86/propagate_ir_flags.ll
+++ b/llvm/test/Transforms/SLPVectorizer/X86/propagate_ir_flags.ll
@@ -398,5 +398,42 @@ define void @fcmp_no_fast(double* %x) #1 {
   ret void
 }
 
-attributes #1 = { "target-features"="+avx" }
+declare double @llvm.fabs.f64(double) nounwind readnone
+
+;CHECK-LABEL: @call_fast(
+;CHECK: call fast <2 x double> @llvm.fabs.v2f64
+define void @call_fast(double* %x) {
+  %idx1 = getelementptr inbounds double, double* %x, i64 0
+  %idx2 = getelementptr inbounds double, double* %x, i64 1
+
+  %load1 = load double, double* %idx1, align 8
+  %load2 = load double, double* %idx2, align 8
+
+  %call1 = tail call fast double @llvm.fabs.f64(double %load1) nounwind readnone
+  %call2 = tail call fast double @llvm.fabs.f64(double %load2) nounwind readnone
+
+  store double %call1, double* %idx1, align 8
+  store double %call2, double* %idx2, align 8
+
+  ret void
+}
 
+;CHECK-LABEL: @call_no_fast(
+;CHECK: call <2 x double> @llvm.fabs.v2f64
+define void @call_no_fast(double* %x) {
+  %idx1 = getelementptr inbounds double, double* %x, i64 0
+  %idx2 = getelementptr inbounds double, double* %x, i64 1
+
+  %load1 = load double, double* %idx1, align 8
+  %load2 = load double, double* %idx2, align 8
+
+  %call1 = tail call fast double @llvm.fabs.f64(double %load1) nounwind readnone
+  %call2 = tail call double @llvm.fabs.f64(double %load2) nounwind readnone
+
+  store double %call1, double* %idx1, align 8
+  store double %call2, double* %idx2, align 8
+
+  ret void
+}
+
+attributes #1 = { "target-features"="+avx" }