diff options
| author | Hal Finkel <hfinkel@anl.gov> | 2012-03-31 03:38:40 +0000 |
|---|---|---|
| committer | Hal Finkel <hfinkel@anl.gov> | 2012-03-31 03:38:40 +0000 |
| commit | 6173ed95daf2f209fe3883faee45967e4800ae75 (patch) | |
| tree | 323330792dd78ca9fb9dd6b3ecf38e3184b46e5b /test/Transforms/BBVectorize | |
| parent | 9f2a9d741f58f0a2a591ec16e5e038c905142dbc (diff) | |
| download | llvm-6173ed95daf2f209fe3883faee45967e4800ae75.tar.gz llvm-6173ed95daf2f209fe3883faee45967e4800ae75.tar.bz2 llvm-6173ed95daf2f209fe3883faee45967e4800ae75.tar.xz | |
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.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@153797 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'test/Transforms/BBVectorize')
| -rw-r--r-- | test/Transforms/BBVectorize/simple-int.ll | 44 |
1 files changed, 44 insertions, 0 deletions
diff --git a/test/Transforms/BBVectorize/simple-int.ll b/test/Transforms/BBVectorize/simple-int.ll index b2ef27b2b8..6844977143 100644 --- a/test/Transforms/BBVectorize/simple-int.ll +++ b/test/Transforms/BBVectorize/simple-int.ll @@ -3,6 +3,7 @@ target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f3 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 @@ define double @test2(double %A1, double %A2, double %B1, double %B2) { ; 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 |