RPCSX/llvm
1
0
Fork 0
mirror of https://github.com/RPCSX/llvm.git synced 2024-12-25 13:35:39 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

DAGCombiner: Don't narrow volatile vector loads + extract

Browse Source 
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@273909 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Matt Arsenault 2016-06-27 19:31:04 +00:00
parent c01c8ebe61
commit bd288e1778
2 changed files with 42 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
lib/CodeGen/SelectionDAG/DAGCombiner.cpp
View File

@ -12263,6 +12263,8 @@ SDValue DAGCombiner::visitINSERT_VECTOR_ELT(SDNode *N) {
SDValue DAGCombiner::ReplaceExtractVectorEltOfLoadWithNarrowedLoad(
SDNode *EVE, EVT InVecVT, SDValue EltNo, LoadSDNode *OriginalLoad) {
assert(!OriginalLoad->isVolatile());
EVT ResultVT = EVE->getValueType(0);
EVT VecEltVT = InVecVT.getVectorElementType();
unsigned Align = OriginalLoad->getAlignment();
@ -12457,9 +12459,12 @@ SDValue DAGCombiner::visitEXTRACT_VECTOR_ELT(SDNode *N) {
ISD::isNormalLoad(InVec.getNode()) &&
!N->getOperand(1)->hasPredecessor(InVec.getNode())) {
SDValue Index = N->getOperand(1);
if (LoadSDNode *OrigLoad = dyn_cast<LoadSDNode>(InVec))
return ReplaceExtractVectorEltOfLoadWithNarrowedLoad(N, VT, Index,
OrigLoad);
if (LoadSDNode *OrigLoad = dyn_cast<LoadSDNode>(InVec)) {
if (!OrigLoad->isVolatile()) {
return ReplaceExtractVectorEltOfLoadWithNarrowedLoad(N, VT, Index,
OrigLoad);
}
}
}
// Perform only after legalization to ensure build_vector / vector_shuffle

34
test/CodeGen/AMDGPU/extractelt-to-trunc.ll
View File

@ -41,3 +41,37 @@ define void @bitcast_int_to_fpvector_extract_0(float addrspace(1)* %out, i64 add
store float %extract, float addrspace(1)* %out
ret void
}
; GCN-LABEL: {{^}}no_extract_volatile_load_extract0:
; GCN: buffer_load_dwordx4
; GCN: buffer_store_dword v
define void @no_extract_volatile_load_extract0(i32 addrspace(1)* %out, <4 x i32> addrspace(1)* %in) {
entry:
%vec = load volatile <4 x i32>, <4 x i32> addrspace(1)* %in
%elt0 = extractelement <4 x i32> %vec, i32 0
store i32 %elt0, i32 addrspace(1)* %out
ret void
}
; GCN-LABEL: {{^}}no_extract_volatile_load_extract2:
; GCN: buffer_load_dwordx4
; GCN: buffer_store_dword v
define void @no_extract_volatile_load_extract2(i32 addrspace(1)* %out, <4 x i32> addrspace(1)* %in) {
entry:
%vec = load volatile <4 x i32>, <4 x i32> addrspace(1)* %in
%elt2 = extractelement <4 x i32> %vec, i32 2
store i32 %elt2, i32 addrspace(1)* %out
ret void
}
; GCN-LABEL: {{^}}no_extract_volatile_load_dynextract:
; GCN: buffer_load_dwordx4
; GCN: buffer_store_dword v
define void @no_extract_volatile_load_dynextract(i32 addrspace(1)* %out, <4 x i32> addrspace(1)* %in, i32 %idx) {
entry:
%vec = load volatile <4 x i32>, <4 x i32> addrspace(1)* %in
%eltN = extractelement <4 x i32> %vec, i32 %idx
store i32 %eltN, i32 addrspace(1)* %out
ret void
}