AMDGPU/GlobalISel: Increase max legal size to 1024
There are 1024 bit register classes defined for AGPRs. Additionally
OpenCL defines vectors up to 16 x i64, and this helps those tests
legalize.
llvm-svn: 373350
diff --git a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
index 10420d6..e795451 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
@@ -40,7 +40,7 @@
static LegalityPredicate isMultiple32(unsigned TypeIdx,
- unsigned MaxSize = 512) {
+ unsigned MaxSize = 1024) {
return [=](const LegalityQuery &Query) {
const LLT Ty = Query.Types[TypeIdx];
const LLT EltTy = Ty.getScalarType();
@@ -115,7 +115,7 @@
};
}
-// Any combination of 32 or 64-bit elements up to 512 bits, and multiples of
+// Any combination of 32 or 64-bit elements up to 1024 bits, and multiples of
// v2s16.
static LegalityPredicate isRegisterType(unsigned TypeIdx) {
return [=](const LegalityQuery &Query) {
@@ -127,7 +127,7 @@
EltSize == 128 || EltSize == 256;
}
- return Ty.getSizeInBits() % 32 == 0 && Ty.getSizeInBits() <= 512;
+ return Ty.getSizeInBits() % 32 == 0 && Ty.getSizeInBits() <= 1024;
};
}
@@ -162,7 +162,7 @@
const LLT S96 = LLT::scalar(96);
const LLT S128 = LLT::scalar(128);
const LLT S256 = LLT::scalar(256);
- const LLT S512 = LLT::scalar(512);
+ const LLT S1024 = LLT::scalar(1024);
const LLT V2S16 = LLT::vector(2, 16);
const LLT V4S16 = LLT::vector(4, 16);
@@ -293,7 +293,7 @@
.legalFor({S1, S32, S64, S16, V2S32, V4S32, V2S16, V4S16, GlobalPtr,
ConstantPtr, LocalPtr, FlatPtr, PrivatePtr})
.moreElementsIf(isSmallOddVector(0), oneMoreElement(0))
- .clampScalarOrElt(0, S32, S512)
+ .clampScalarOrElt(0, S32, S1024)
.legalIf(isMultiple32(0))
.widenScalarToNextPow2(0, 32)
.clampMaxNumElements(0, S32, 16);
@@ -884,7 +884,7 @@
return (EltTy.getSizeInBits() == 16 ||
EltTy.getSizeInBits() % 32 == 0) &&
VecTy.getSizeInBits() % 32 == 0 &&
- VecTy.getSizeInBits() <= 512 &&
+ VecTy.getSizeInBits() <= 1024 &&
IdxTy.getSizeInBits() == 32;
})
.clampScalar(EltTypeIdx, S32, S64)
@@ -991,7 +991,7 @@
.fewerElementsIf(
[=](const LegalityQuery &Query) { return notValidElt(Query, 1); },
scalarize(1))
- .clampScalar(BigTyIdx, S32, S512)
+ .clampScalar(BigTyIdx, S32, S1024)
.lowerFor({{S16, V2S16}})
.widenScalarIf(
[=](const LegalityQuery &Query) {
@@ -1022,7 +1022,7 @@
return BigTy.getSizeInBits() % 16 == 0 &&
LitTy.getSizeInBits() % 16 == 0 &&
- BigTy.getSizeInBits() <= 512;
+ BigTy.getSizeInBits() <= 1024;
})
// Any vectors left are the wrong size. Scalarize them.
.scalarize(0)