maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

SelectionDAG: when constructing VZEXT_LOAD from other loads, make sure its 

output chain is correctly setup.

As an example, if the original load must happen before later stores, we need
to make sure the constructed VZEXT_LOAD is constrained to be before the stores.

rdar://11457792

llvm-svn: 163036
This commit is contained in:
Manman Ren 2012-08-31 23:16:57 +00:00
parent 908e685102
commit 26c5d0f607
2 changed files with 43 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
llvm/lib/Target/X86/X86ISelLowering.cpp
View File

@ -4977,6 +4977,18 @@ static SDValue EltsFromConsecutiveLoads(EVT VT, SmallVectorImpl<SDValue> &Elts,
LDBase->getAlignment(),
false/*isVolatile*/, true/*ReadMem*/,
false/*WriteMem*/);
// Make sure the newly-created LOAD is in the same position as LDBase in
// terms of dependency. We create a TokenFactor for LDBase and ResNode, and
// update uses of LDBase's output chain to use the TokenFactor.
if (LDBase->hasAnyUseOfValue(1)) {
SDValue NewChain = DAG.getNode(ISD::TokenFactor, DL, MVT::Other,
SDValue(LDBase, 1), SDValue(ResNode.getNode(), 1));
DAG.ReplaceAllUsesOfValueWith(SDValue(LDBase, 1), NewChain);
DAG.UpdateNodeOperands(NewChain.getNode(), SDValue(LDBase, 1),
SDValue(ResNode.getNode(), 1));
}
return DAG.getNode(ISD::BITCAST, DL, VT, ResNode);
}
return SDValue();

34
llvm/test/CodeGen/X86/vec_shuffle-26.ll
View File

@ -1,6 +1,4 @@
; RUN: llc < %s -march=x86 -mattr=sse41 -o %t
; RUN: grep unpcklps %t | count 1
; RUN: grep unpckhps %t | count 3
; RUN: llc < %s -march=x86 -mattr=sse41 | FileCheck %s
; Transpose example using the more generic vector shuffle. Return float8
; instead of float16
@ -14,6 +12,11 @@ target triple = "i386-apple-cl.1.0"
define <8 x float> @__transpose2(<4 x float> %p0, <4 x float> %p1, <4 x float> %p2, <4 x float> %p3) nounwind {
entry:
; CHECK: transpose2
; CHECK: unpckhps
; CHECK: unpckhps
; CHECK: unpcklps
; CHECK: unpckhps
%unpcklps = shufflevector <4 x float> %p0, <4 x float> %p2, <4 x i32> < i32 0, i32 4, i32 1, i32 5 > ; <<4 x float>> [#uses=2]
%unpckhps = shufflevector <4 x float> %p0, <4 x float> %p2, <4 x i32> < i32 2, i32 6, i32 3, i32 7 > ; <<4 x float>> [#uses=2]
%unpcklps8 = shufflevector <4 x float> %p1, <4 x float> %p3, <4 x i32> < i32 0, i32 4, i32 1, i32 5 > ; <<4 x float>> [#uses=2]
@ -27,3 +30,28 @@ entry:
; %r3 = shufflevector <8 x float> %r1, <8 x float> %r2, <16 x i32> < i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15 >;
ret <8 x float> %r2
}
define <2 x i64> @lo_hi_shift(float* nocapture %x, float* nocapture %y) nounwind {
entry:
; movhps should happen before extractps to assure it gets the correct value.
; CHECK: lo_hi_shift
; CHECK: movhps ([[BASEREG:%[a-z]+]]),
; CHECK: extractps ${{[0-9]+}}, %xmm{{[0-9]+}}, {{[0-9]*}}([[BASEREG]])
; CHECK: extractps ${{[0-9]+}}, %xmm{{[0-9]+}}, {{[0-9]*}}([[BASEREG]])
%v.i = bitcast float* %y to <4 x float>*
%0 = load <4 x float>* %v.i, align 1
%1 = bitcast float* %x to <1 x i64>*
%.val = load <1 x i64>* %1, align 1
%2 = bitcast <1 x i64> %.val to <2 x float>
%shuffle.i = shufflevector <2 x float> %2, <2 x float> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>
%shuffle1.i = shufflevector <4 x float> %0, <4 x float> %shuffle.i, <4 x i32> <i32 0, i32 1, i32 4, i32 5>
%cast.i = bitcast <4 x float> %0 to <2 x i64>
%extract.i = extractelement <2 x i64> %cast.i, i32 1
%3 = bitcast float* %x to i64*
store i64 %extract.i, i64* %3, align 4
%4 = bitcast <4 x float> %0 to <16 x i8>
%5 = bitcast <4 x float> %shuffle1.i to <16 x i8>
%palignr = shufflevector <16 x i8> %5, <16 x i8> %4, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23>
%6 = bitcast <16 x i8> %palignr to <2 x i64>
ret <2 x i64> %6
}