forked from OSchip/llvm-project
29 lines
980 B
LLVM
29 lines
980 B
LLVM
; Test several VSX FMA mutation opportunities. The first one isn't a
|
|
; reasonable transformation because the killed product register is the
|
|
; same as the FMA target register. The second one is legal. The third
|
|
; one doesn't fit the feeding-copy pattern.
|
|
|
|
; RUN: llc -verify-machineinstrs < %s -mtriple=powerpc64-unknown-linux-gnu -mcpu=pwr7 -mattr=+vsx -disable-ppc-vsx-fma-mutation=false | FileCheck %s
|
|
target datalayout = "E-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-f128:128:128-v128:128:128-n32:64"
|
|
target triple = "powerpc64-unknown-linux-gnu"
|
|
|
|
declare double @llvm.sqrt.f64(double)
|
|
|
|
define double @foo3_fmf(double %a) nounwind {
|
|
; CHECK: @foo3_fmf
|
|
; CHECK-NOT: fmr
|
|
; CHECK: xsmaddmdp
|
|
; CHECK: xsmaddadp
|
|
%r = call fast double @llvm.sqrt.f64(double %a)
|
|
ret double %r
|
|
}
|
|
|
|
define double @foo3_safe(double %a) nounwind {
|
|
; CHECK: @foo3_safe
|
|
; CHECK-NOT: fmr
|
|
; CHECK: xssqrtdp
|
|
%r = call double @llvm.sqrt.f64(double %a)
|
|
ret double %r
|
|
}
|
|
|