mirror of
https://github.com/c64scene-ar/llvm-6502.git
synced 2024-07-05 13:29:22 +00:00
dfd0b9b460
the first source operand is tied to the destination operand. This is to accurately model the corresponding instructions where the upper bits are unmodified. rdar://12558838 PR14221 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@167064 91177308-0d34-0410-b5e6-96231b3b80d8
37 lines
1.5 KiB
LLVM
37 lines
1.5 KiB
LLVM
; RUN: llc < %s -mtriple=x86_64-apple-macosx -mattr=+sse2 -mcpu=nehalem | FileCheck %s
|
|
|
|
; rdar: 12558838
|
|
; PR14221
|
|
; There is a mismatch between the intrinsic and the actual instruction.
|
|
; The actual instruction has a partial update of dest, while the intrinsic
|
|
; passes through the upper FP values. Here, we make sure the source and
|
|
; destination of rsqrtss are the same.
|
|
define void @t1(<4 x float> %a) nounwind uwtable ssp {
|
|
entry:
|
|
; CHECK: t1:
|
|
; CHECK: rsqrtss %xmm0, %xmm0
|
|
%0 = tail call <4 x float> @llvm.x86.sse.rsqrt.ss(<4 x float> %a) nounwind
|
|
%a.addr.0.extract = extractelement <4 x float> %0, i32 0
|
|
%conv = fpext float %a.addr.0.extract to double
|
|
%a.addr.4.extract = extractelement <4 x float> %0, i32 1
|
|
%conv3 = fpext float %a.addr.4.extract to double
|
|
tail call void @callee(double %conv, double %conv3) nounwind
|
|
ret void
|
|
}
|
|
declare void @callee(double, double)
|
|
declare <4 x float> @llvm.x86.sse.rsqrt.ss(<4 x float>) nounwind readnone
|
|
|
|
define void @t2(<4 x float> %a) nounwind uwtable ssp {
|
|
entry:
|
|
; CHECK: t2:
|
|
; CHECK: rcpss %xmm0, %xmm0
|
|
%0 = tail call <4 x float> @llvm.x86.sse.rcp.ss(<4 x float> %a) nounwind
|
|
%a.addr.0.extract = extractelement <4 x float> %0, i32 0
|
|
%conv = fpext float %a.addr.0.extract to double
|
|
%a.addr.4.extract = extractelement <4 x float> %0, i32 1
|
|
%conv3 = fpext float %a.addr.4.extract to double
|
|
tail call void @callee(double %conv, double %conv3) nounwind
|
|
ret void
|
|
}
|
|
declare <4 x float> @llvm.x86.sse.rcp.ss(<4 x float>) nounwind readnone
|