[llvm-commits] [llvm] r153797 - in /llvm/trunk: lib/Transforms/Vectorize/BBVectorize.cpp test/Transforms/BBVectorize/simple-int.ll
Hal Finkel
hfinkel at anl.gov
Fri Mar 30 20:38:40 PDT 2012
Author: hfinkel
Date: Fri Mar 30 22:38:40 2012
New Revision: 153797
URL: http://llvm.org/viewvc/llvm-project?rev=153797&view=rev
Log:
Correctly vectorize powi.
The powi intrinsic requires special handling because it always takes a single
integer power regardless of the result type. As a result, we can vectorize
only if the powers are equal. Fixes PR12364.
Modified:
llvm/trunk/lib/Transforms/Vectorize/BBVectorize.cpp
llvm/trunk/test/Transforms/BBVectorize/simple-int.ll
Modified: llvm/trunk/lib/Transforms/Vectorize/BBVectorize.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Vectorize/BBVectorize.cpp?rev=153797&r1=153796&r2=153797&view=diff
==============================================================================
--- llvm/trunk/lib/Transforms/Vectorize/BBVectorize.cpp (original)
+++ llvm/trunk/lib/Transforms/Vectorize/BBVectorize.cpp Fri Mar 30 22:38:40 2012
@@ -647,6 +647,20 @@
// FIXME: We may want to vectorize non-constant shuffles also.
}
+ // The powi intrinsic is special because only the first argument is
+ // vectorized, the second arguments must be equal.
+ CallInst *CI = dyn_cast<CallInst>(I);
+ Function *FI;
+ if (CI && (FI = CI->getCalledFunction()) &&
+ FI->getIntrinsicID() == Intrinsic::powi) {
+
+ Value *A1I = CI->getArgOperand(1),
+ *A1J = cast<CallInst>(J)->getArgOperand(1);
+ const SCEV *A1ISCEV = SE->getSCEV(A1I),
+ *A1JSCEV = SE->getSCEV(A1J);
+ return (A1ISCEV == A1JSCEV);
+ }
+
return true;
}
@@ -1518,19 +1532,27 @@
ReplacedOperands[o] = getReplacementPointerInput(Context, I, J, o,
FlipMemInputs);
continue;
- } else if (isa<CallInst>(I) && o == NumOperands-1) {
+ } else if (isa<CallInst>(I)) {
Function *F = cast<CallInst>(I)->getCalledFunction();
unsigned IID = F->getIntrinsicID();
- BasicBlock &BB = *I->getParent();
-
- Module *M = BB.getParent()->getParent();
- Type *ArgType = I->getType();
- Type *VArgType = getVecTypeForPair(ArgType);
-
- // FIXME: is it safe to do this here?
- ReplacedOperands[o] = Intrinsic::getDeclaration(M,
- (Intrinsic::ID) IID, VArgType);
- continue;
+ if (o == NumOperands-1) {
+ BasicBlock &BB = *I->getParent();
+
+ Module *M = BB.getParent()->getParent();
+ Type *ArgType = I->getType();
+ Type *VArgType = getVecTypeForPair(ArgType);
+
+ // FIXME: is it safe to do this here?
+ ReplacedOperands[o] = Intrinsic::getDeclaration(M,
+ (Intrinsic::ID) IID, VArgType);
+ continue;
+ } else if (IID == Intrinsic::powi && o == 1) {
+ // The second argument of powi is a single integer and we've already
+ // checked that both arguments are equal. As a result, we just keep
+ // I's second argument.
+ ReplacedOperands[o] = I->getOperand(o);
+ continue;
+ }
} else if (isa<ShuffleVectorInst>(I) && o == NumOperands-1) {
ReplacedOperands[o] = getReplacementShuffleMask(Context, I, J);
continue;
Modified: llvm/trunk/test/Transforms/BBVectorize/simple-int.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/BBVectorize/simple-int.ll?rev=153797&r1=153796&r2=153797&view=diff
==============================================================================
--- llvm/trunk/test/Transforms/BBVectorize/simple-int.ll (original)
+++ llvm/trunk/test/Transforms/BBVectorize/simple-int.ll Fri Mar 30 22:38:40 2012
@@ -3,6 +3,7 @@
declare double @llvm.fma.f64(double, double, double)
declare double @llvm.cos.f64(double)
+declare double @llvm.powi.f64(double, i32)
; Basic depth-3 chain with fma
define double @test1(double %A1, double %A2, double %B1, double %B2, double %C1, double %C2) {
@@ -54,6 +55,49 @@
; CHECK: ret double %R
}
+; Basic depth-3 chain with powi
+define double @test3(double %A1, double %A2, double %B1, double %B2, i32 %P) {
+
+ %X1 = fsub double %A1, %B1
+ %X2 = fsub double %A2, %B2
+ %Y1 = call double @llvm.powi.f64(double %X1, i32 %P)
+ %Y2 = call double @llvm.powi.f64(double %X2, i32 %P)
+ %Z1 = fadd double %Y1, %B1
+ %Z2 = fadd double %Y2, %B2
+ %R = fmul double %Z1, %Z2
+ ret double %R
+; CHECK: @test3
+; CHECK: %X1.v.i1.1 = insertelement <2 x double> undef, double %B1, i32 0
+; CHECK: %X1.v.i0.1 = insertelement <2 x double> undef, double %A1, i32 0
+; CHECK: %X1.v.i1.2 = insertelement <2 x double> %X1.v.i1.1, double %B2, i32 1
+; CHECK: %X1.v.i0.2 = insertelement <2 x double> %X1.v.i0.1, double %A2, i32 1
+; CHECK: %X1 = fsub <2 x double> %X1.v.i0.2, %X1.v.i1.2
+; CHECK: %Y1 = call <2 x double> @llvm.powi.v2f64(<2 x double> %X1, i32 %P)
+; CHECK: %Z1 = fadd <2 x double> %Y1, %X1.v.i1.2
+; CHECK: %Z1.v.r1 = extractelement <2 x double> %Z1, i32 0
+; CHECK: %Z1.v.r2 = extractelement <2 x double> %Z1, i32 1
+; CHECK: %R = fmul double %Z1.v.r1, %Z1.v.r2
+; CHECK: ret double %R
+}
+
+; Basic depth-3 chain with powi (different powers: should not vectorize)
+define double @test4(double %A1, double %A2, double %B1, double %B2, i32 %P) {
+
+ %X1 = fsub double %A1, %B1
+ %X2 = fsub double %A2, %B2
+ %P2 = add i32 %P, 1
+ %Y1 = call double @llvm.powi.f64(double %X1, i32 %P)
+ %Y2 = call double @llvm.powi.f64(double %X2, i32 %P2)
+ %Z1 = fadd double %Y1, %B1
+ %Z2 = fadd double %Y2, %B2
+ %R = fmul double %Z1, %Z2
+ ret double %R
+; CHECK: @test4
+; CHECK-NOT: <2 x double>
+; CHECK: ret double %R
+}
+
; CHECK: declare <2 x double> @llvm.fma.v2f64(<2 x double>, <2 x double>, <2 x double>) nounwind readnone
; CHECK: declare <2 x double> @llvm.cos.v2f64(<2 x double>) nounwind readonly
+; CHECK: declare <2 x double> @llvm.powi.v2f64(<2 x double>, i32) nounwind readonly
More information about the llvm-commits
mailing list