llvm-6502/test/CodeGen/X86/sse_partial_update.ll
Sanjay Patel 9ccbf1a260 Model sqrtss as a binary operation with one source operand tied to the destination (PR14221)
This is a continuation of r167064 ( http://llvm.org/viewvc/llvm-project?view=revision&revision=167064 ).
That patch started to fix PR14221 ( http://llvm.org/bugs/show_bug.cgi?id=14221 ), but it was not completed. 

Differential Revision: http://reviews.llvm.org/D6330



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@224624 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-19 22:16:28 +00:00

70 lines
2.4 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 each scalar unary op are the same.
define void @rsqrtss(<4 x float> %a) nounwind uwtable ssp {
entry:
; CHECK-LABEL: rsqrtss:
; CHECK: rsqrtss %xmm0, %xmm0
; CHECK-NEXT: cvtss2sd %xmm0
; CHECK-NEXT: shufps
; CHECK-NEXT: cvtss2sd %xmm0
; CHECK-NEXT: movap
; CHECK-NEXT: jmp
%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 @rcpss(<4 x float> %a) nounwind uwtable ssp {
entry:
; CHECK-LABEL: rcpss:
; CHECK: rcpss %xmm0, %xmm0
; CHECK-NEXT: cvtss2sd %xmm0
; CHECK-NEXT: shufps
; CHECK-NEXT: cvtss2sd %xmm0
; CHECK-NEXT: movap
; CHECK-NEXT: jmp
%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
define void @sqrtss(<4 x float> %a) nounwind uwtable ssp {
entry:
; CHECK-LABEL: sqrtss:
; CHECK: sqrtss %xmm0, %xmm0
; CHECK-NEXT: cvtss2sd %xmm0
; CHECK-NEXT: shufps
; CHECK-NEXT: cvtss2sd %xmm0
; CHECK-NEXT: movap
; CHECK-NEXT: jmp
%0 = tail call <4 x float> @llvm.x86.sse.sqrt.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.sqrt.ss(<4 x float>) nounwind readnone