mirror of
https://github.com/RPCS3/llvm.git
synced 2024-12-14 07:31:53 +00:00
3d89ab8299
This is a follow-up to the FIXME that was added with D7474 ( http://reviews.llvm.org/rL229531 ). I thought this load folding bug had been made hard-to-hit, but it turns out to be very easy when targeting 32-bit x86 and causes a miscompile/crash in Wine: https://bugs.winehq.org/show_bug.cgi?id=38826 https://llvm.org/bugs/show_bug.cgi?id=22371#c25 The quick fix is to simply remove the scalar FP logical instructions from the load folding table in X86InstrInfo, but that causes us to miss load folds that should be possible when lowering fabs, fneg, fcopysign. So the majority of this patch is altering those lowerings to use *vector* FP logical instructions (because that's all x86 gives us anyway). That lets us do the load folding legally. Differential Revision: http://reviews.llvm.org/D11477 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243361 91177308-0d34-0410-b5e6-96231b3b80d8
53 lines
2.3 KiB
LLVM
53 lines
2.3 KiB
LLVM
; RUN: llc < %s -march=x86 -mattr=+sse2 | FileCheck %s
|
|
; PR2656
|
|
|
|
target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128"
|
|
target triple = "i686-apple-darwin9.4.0"
|
|
%struct.anon = type <{ float, float }>
|
|
@.str = internal constant [17 x i8] c"pt: %.0f, %.0f\0A\00\00" ; <[17 x i8]*> [#uses=1]
|
|
|
|
; We can not fold either stack load into an 'xor' instruction because that
|
|
; would change what should be a 4-byte load into a 16-byte load.
|
|
; We can fold the 16-byte constant load into either 'xor' instruction,
|
|
; but we do not. It has more than one use, so it gets loaded into a register.
|
|
|
|
define void @foo(%struct.anon* byval %p) nounwind {
|
|
; CHECK-LABEL: foo:
|
|
; CHECK: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
|
|
; CHECK-NEXT: movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
|
|
; CHECK-NEXT: movaps {{.*#+}} xmm2 = [2147483648,2147483648,2147483648,2147483648]
|
|
; CHECK-NEXT: xorps %xmm2, %xmm0
|
|
; CHECK-NEXT: cvtss2sd %xmm0, %xmm0
|
|
; CHECK-NEXT: xorps %xmm2, %xmm1
|
|
entry:
|
|
%tmp = getelementptr %struct.anon, %struct.anon* %p, i32 0, i32 0 ; <float*> [#uses=1]
|
|
%tmp1 = load float, float* %tmp ; <float> [#uses=1]
|
|
%tmp2 = getelementptr %struct.anon, %struct.anon* %p, i32 0, i32 1 ; <float*> [#uses=1]
|
|
%tmp3 = load float, float* %tmp2 ; <float> [#uses=1]
|
|
%neg = fsub float -0.000000e+00, %tmp1 ; <float> [#uses=1]
|
|
%conv = fpext float %neg to double ; <double> [#uses=1]
|
|
%neg4 = fsub float -0.000000e+00, %tmp3 ; <float> [#uses=1]
|
|
%conv5 = fpext float %neg4 to double ; <double> [#uses=1]
|
|
%call = call i32 (...) @printf( i8* getelementptr ([17 x i8], [17 x i8]* @.str, i32 0, i32 0), double %conv, double %conv5 ) ; <i32> [#uses=0]
|
|
ret void
|
|
}
|
|
|
|
declare i32 @printf(...)
|
|
|
|
; We can not fold the load from the stack into the 'and' instruction because
|
|
; that changes an 8-byte load into a 16-byte load (illegal memory access).
|
|
; We can fold the load of the constant because it is a 16-byte vector constant.
|
|
|
|
define double @PR22371(double %x) {
|
|
; CHECK-LABEL: PR22371:
|
|
; CHECK: movsd 16(%esp), %xmm0
|
|
; CHECK-NEXT: andpd LCPI1_0, %xmm0
|
|
; CHECK-NEXT: movlpd %xmm0, (%esp)
|
|
%call = tail call double @fabs(double %x) #0
|
|
ret double %call
|
|
}
|
|
|
|
declare double @fabs(double) #0
|
|
attributes #0 = { readnone }
|
|
|