RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-26 12:43:36 +01:00
Code Issues Projects Releases Wiki Activity

become more strict about when it's safe to use X86ISD::MOVLPS

Browse Source 
llvm-svn: 112799
This commit is contained in:
Bruno Cardoso Lopes 2010-09-02 02:35:51 +00:00
parent 6c3fa8c78f
commit dcdab94661
2 changed files with 12 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
lib/Target/X86/X86ISelLowering.cpp
View File

@ -5082,10 +5082,10 @@ SDValue getMOVLP(SDValue &Op, DebugLoc &dl, SelectionDAG &DAG, bool HasSSE2) {
SDValue TmpV1 = V1;
SDValue TmpV2 = V2;
// Trivial case, when V2 is a load.
if (TmpV2.getOpcode() == ISD::BIT_CONVERT)
// Trivial case, when V2 comes from a load.
if (TmpV2.hasOneUse() && TmpV2.getOpcode() == ISD::BIT_CONVERT)
TmpV2 = TmpV2.getOperand(0);
if (TmpV2.getOpcode() == ISD::SCALAR_TO_VECTOR)
if (TmpV2.hasOneUse() && TmpV2.getOpcode() == ISD::SCALAR_TO_VECTOR)
TmpV2 = TmpV2.getOperand(0);
if (MayFoldLoad(TmpV2))
CanFoldLoad = true;
@ -5095,9 +5095,9 @@ SDValue getMOVLP(SDValue &Op, DebugLoc &dl, SelectionDAG &DAG, bool HasSSE2) {
// turns into:
// (MOVLPSmr addr:$src1, VR128:$src2)
// So, recognize this potential and also use MOVLPS or MOVLPD
if (TmpV1.getOpcode() == ISD::BIT_CONVERT)
if (TmpV1.hasOneUse() && TmpV1.getOpcode() == ISD::BIT_CONVERT)
TmpV1 = TmpV1.getOperand(0);
if (MayFoldLoad(TmpV1))
if (MayFoldLoad(TmpV1) && MayFoldIntoStore(Op))
CanFoldLoad = true;
if (CanFoldLoad) {

10
lib/Target/X86/X86InstrSSE.td
View File

@ -5829,6 +5829,13 @@ def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
(MOVSSrr (v4f32 VR128:$src1),
(EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
// FIXME: Instead of a X86Movss there should be a X86Movlps here, the problem
// is during lowering, where it's not possible to recognize the load fold cause
// it has two uses through a bitcast. One use disappears at isel time and the
// fold opportunity reappears.
def : Pat<(X86Movss VR128:$src1,
(bc_v4i32 (v2i64 (load addr:$src2)))),
(MOVLPSrm VR128:$src1, addr:$src2)>;
// Shuffle with MOVSD
def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
@ -5909,9 +5916,6 @@ def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
def : Pat<(X86Movlps VR128:$src1,
(bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
(MOVLPSrm VR128:$src1, addr:$src2)>;
def : Pat<(X86Movlps VR128:$src1,
(bc_v4i32 (v2i64 (load addr:$src2)))),
(MOVLPSrm VR128:$src1, addr:$src2)>;
// Shuffle with MOVLPD
def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),