forked from OSchip/llvm-project
parent
6c8100748f
commit
ca6dcfcf59
|
@ -613,9 +613,9 @@ SDValue AMDGPUTargetLowering::MergeVectorStore(const SDValue &Op,
|
|||
EVT MemVT = Store->getMemoryVT();
|
||||
unsigned MemBits = MemVT.getSizeInBits();
|
||||
|
||||
// Byte stores are really expensive, so if possible, try to pack
|
||||
// 32-bit vector truncatating store into an i32 store.
|
||||
// XXX: We could also handle optimize other vector bitwidths
|
||||
// Byte stores are really expensive, so if possible, try to pack 32-bit vector
|
||||
// truncating store into an i32 store.
|
||||
// XXX: We could also handle optimize other vector bitwidths.
|
||||
if (!MemVT.isVector() || MemBits > 32) {
|
||||
return SDValue();
|
||||
}
|
||||
|
|
Loading…
Reference in New Issue