diff options
| author | Evan Cheng <evan.cheng@apple.com> | 2011-02-11 02:28:55 +0000 |
|---|---|---|
| committer | Evan Cheng <evan.cheng@apple.com> | 2011-02-11 02:28:55 +0000 |
| commit | 2da1c9599354c82c1c61fead4d162127bf636f80 (patch) | |
| tree | 7154e156b7ec5dedf4607ca81f0b5777190279de /llvm/test/CodeGen | |
| parent | d5090c16f8b2b3c12bb54ac235af32945357b502 (diff) | |
| download | bcm5719-llvm-2da1c9599354c82c1c61fead4d162127bf636f80.tar.gz bcm5719-llvm-2da1c9599354c82c1c61fead4d162127bf636f80.zip | |
Fix buggy fcopysign lowering.
This
define float @foo(float %x, float %y) nounwind readnone {
entry:
%0 = tail call float @copysignf(float %x, float %y) nounwind readnone
ret float %0
}
Was compiled to:
vmov s0, r1
bic r0, r0, #-2147483648
vmov s1, r0
vcmpe.f32 s0, #0
vmrs apsr_nzcv, fpscr
it lt
vneglt.f32 s1, s1
vmov r0, s1
bx lr
This fails to copy the sign of -0.0f because it's lost during the float to int
conversion. Also, it's sub-optimal when the inputs are in GPR registers.
Now it uses integer and + or operations when it's profitable. And it's correct!
lsrs r1, r1, #31
bfi r0, r1, #31, #1
bx lr
rdar://8984306
llvm-svn: 125357
Diffstat (limited to 'llvm/test/CodeGen')
| -rw-r--r-- | llvm/test/CodeGen/ARM/fcopysign.ll | 53 |
1 files changed, 40 insertions, 13 deletions
diff --git a/llvm/test/CodeGen/ARM/fcopysign.ll b/llvm/test/CodeGen/ARM/fcopysign.ll index a6d741087a8..1050cd26599 100644 --- a/llvm/test/CodeGen/ARM/fcopysign.ll +++ b/llvm/test/CodeGen/ARM/fcopysign.ll @@ -1,18 +1,45 @@ -; RUN: llc < %s -march=arm | grep bic | count 2 -; RUN: llc < %s -march=arm -mattr=+v6,+vfp2 | \ -; RUN: grep vneg | count 2 +; RUN: llc < %s -mtriple=armv7-apple-darwin -mcpu=cortex-a8 | FileCheck %s -check-prefix=SOFT +; RUN: llc < %s -mtriple=armv7-gnueabi -float-abi=hard -mcpu=cortex-a8 | FileCheck %s -check-prefix=HARD -define float @test1(float %x, double %y) { - %tmp = fpext float %x to double - %tmp2 = tail call double @copysign( double %tmp, double %y ) - %tmp3 = fptrunc double %tmp2 to float - ret float %tmp3 +; rdar://8984306 +define float @test1(float %x, float %y) nounwind { +entry: +; SOFT: test1: +; SOFT: lsr r1, r1, #31 +; SOFT: bfi r0, r1, #31, #1 + +; HARD: test1: +; HARD: vabs.f32 d0, d0 +; HARD: cmp r0, #0 +; HARD: vneglt.f32 s0, s0 + %0 = tail call float @copysignf(float %x, float %y) nounwind + ret float %0 +} + +define double @test2(double %x, double %y) nounwind { +entry: +; SOFT: test2: +; SOFT: lsr r2, r3, #31 +; SOFT: bfi r1, r2, #31, #1 + +; HARD: test2: +; HARD: vabs.f64 d0, d0 +; HARD: cmp r1, #0 +; HARD: vneglt.f64 d0, d0 + %0 = tail call double @copysign(double %x, double %y) nounwind + ret double %0 } -define double @test2(double %x, float %y) { - %tmp = fpext float %y to double - %tmp2 = tail call double @copysign( double %x, double %tmp ) - ret double %tmp2 +define double @test3(double %x, double %y, double %z) nounwind { +entry: +; SOFT: test3: +; SOFT: vabs.f64 +; SOFT: cmp {{.*}}, #0 +; SOFT: vneglt.f64 + %0 = fmul double %x, %y + %1 = tail call double @copysign(double %0, double %z) nounwind + ret double %1 } -declare double @copysign(double, double) +declare double @copysign(double, double) nounwind +declare float @copysignf(float, float) nounwind |

