[AMDGPU] Split v32i32 loads
Differential Revision: https://reviews.llvm.org/D51555
llvm-svn: 341266
diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
index f89b741..0028fe1 100644
--- a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -156,12 +156,14 @@
setOperationAction(ISD::LOAD, MVT::v8i32, Custom);
setOperationAction(ISD::LOAD, MVT::v16i32, Custom);
setOperationAction(ISD::LOAD, MVT::i1, Custom);
+ setOperationAction(ISD::LOAD, MVT::v32i32, Custom);
setOperationAction(ISD::STORE, MVT::v2i32, Custom);
setOperationAction(ISD::STORE, MVT::v4i32, Custom);
setOperationAction(ISD::STORE, MVT::v8i32, Custom);
setOperationAction(ISD::STORE, MVT::v16i32, Custom);
setOperationAction(ISD::STORE, MVT::i1, Custom);
+ setOperationAction(ISD::STORE, MVT::v32i32, Custom);
setTruncStoreAction(MVT::v2i32, MVT::v2i16, Expand);
setTruncStoreAction(MVT::v4i32, MVT::v4i16, Expand);
@@ -246,7 +248,7 @@
// We only support LOAD/STORE and vector manipulation ops for vectors
// with > 4 elements.
for (MVT VT : {MVT::v8i32, MVT::v8f32, MVT::v16i32, MVT::v16f32,
- MVT::v2i64, MVT::v2f64, MVT::v4i16, MVT::v4f16 }) {
+ MVT::v2i64, MVT::v2f64, MVT::v4i16, MVT::v4f16, MVT::v32i32 }) {
for (unsigned Op = 0; Op < ISD::BUILTIN_OP_END; ++Op) {
switch (Op) {
case ISD::LOAD:
@@ -6196,7 +6198,7 @@
if (AS == AMDGPUAS::CONSTANT_ADDRESS ||
AS == AMDGPUAS::CONSTANT_ADDRESS_32BIT) {
- if (!Op->isDivergent() && Alignment >= 4)
+ if (!Op->isDivergent() && Alignment >= 4 && NumElements < 32)
return SDValue();
// Non-uniform loads will be selected to MUBUF instructions, so they
// have the same legalization requirements as global and private
@@ -6209,7 +6211,7 @@
AS == AMDGPUAS::GLOBAL_ADDRESS) {
if (Subtarget->getScalarizeGlobalBehavior() && !Op->isDivergent() &&
!Load->isVolatile() && isMemOpHasNoClobberedMemOperand(Load) &&
- Alignment >= 4)
+ Alignment >= 4 && NumElements < 32)
return SDValue();
// Non-uniform loads will be selected to MUBUF instructions, so they
// have the same legalization requirements as global and private