#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/MathExtras.h"
#include "llvm/Support/raw_ostream.h"
#include <sstream>
"nvptx-sched4reg",
cl::desc("NVPTX Specific: schedule for register pressue"), cl::init(false));
+static cl::opt<unsigned>
+FMAContractLevelOpt("nvptx-fma-level", cl::ZeroOrMore, cl::Hidden,
+ cl::desc("NVPTX Specific: FMA contraction (0: don't do it"
+ " 1: do it 2: do it aggressively"),
+ cl::init(2));
+
static bool IsPTXVectorType(MVT VT) {
switch (VT.SimpleTy) {
default:
}
}
-static uint64_t GCD( int a, int b)
-{
- if (a < b) std::swap(a,b);
- while (b > 0) {
- uint64_t c = b;
- b = a % b;
- a = c;
- }
- return a;
-}
-
/// ComputePTXValueVTs - For the given Type \p Ty, returns the set of primitive
/// EVTs that compose it. Unlike ComputeValueVTs, this will break apart vectors
/// into their primitive components.
/// NOTE: This is a band-aid for code that expects ComputeValueVTs to return the
/// same number of types as the Ins/Outs arrays in LowerFormalArguments,
/// LowerCall, and LowerReturn.
-static void ComputePTXValueVTs(const TargetLowering &TLI, Type *Ty,
- SmallVectorImpl<EVT> &ValueVTs,
+static void ComputePTXValueVTs(const TargetLowering &TLI, const DataLayout &DL,
+ Type *Ty, SmallVectorImpl<EVT> &ValueVTs,
SmallVectorImpl<uint64_t> *Offsets = nullptr,
uint64_t StartingOffset = 0) {
SmallVector<EVT, 16> TempVTs;
SmallVector<uint64_t, 16> TempOffsets;
- ComputeValueVTs(TLI, Ty, TempVTs, &TempOffsets, StartingOffset);
+ ComputeValueVTs(TLI, DL, Ty, TempVTs, &TempOffsets, StartingOffset);
for (unsigned i = 0, e = TempVTs.size(); i != e; ++i) {
EVT VT = TempVTs[i];
uint64_t Off = TempOffsets[i];
}
// NVPTXTargetLowering Constructor.
-NVPTXTargetLowering::NVPTXTargetLowering(NVPTXTargetMachine &TM)
- : TargetLowering(TM, new NVPTXTargetObjectFile()), nvTM(&TM),
- nvptxSubtarget(TM.getSubtarget<NVPTXSubtarget>()) {
+NVPTXTargetLowering::NVPTXTargetLowering(const NVPTXTargetMachine &TM,
+ const NVPTXSubtarget &STI)
+ : TargetLowering(TM), nvTM(&TM), STI(STI) {
// always lower memset, memcpy, and memmove intrinsics to load/store
// instructions, rather
// condition branches.
setJumpIsExpensive(true);
+ // Wide divides are _very_ slow. Try to reduce the width of the divide if
+ // possible.
+ addBypassSlowDiv(64, 32);
+
// By default, use the Source scheduling
if (sched4reg)
setSchedulingPreference(Sched::RegPressure);
setOperationAction(ISD::SRA_PARTS, MVT::i64 , Custom);
setOperationAction(ISD::SRL_PARTS, MVT::i64 , Custom);
- if (nvptxSubtarget.hasROT64()) {
+ if (STI.hasROT64()) {
setOperationAction(ISD::ROTL, MVT::i64, Legal);
setOperationAction(ISD::ROTR, MVT::i64, Legal);
} else {
setOperationAction(ISD::ROTL, MVT::i64, Expand);
setOperationAction(ISD::ROTR, MVT::i64, Expand);
}
- if (nvptxSubtarget.hasROT32()) {
+ if (STI.hasROT32()) {
setOperationAction(ISD::ROTL, MVT::i32, Legal);
setOperationAction(ISD::ROTR, MVT::i32, Legal);
} else {
setOperationAction(ISD::INTRINSIC_W_CHAIN, MVT::Other, Custom);
// Turn FP extload into load/fextend
- setLoadExtAction(ISD::EXTLOAD, MVT::f32, Expand);
+ setLoadExtAction(ISD::EXTLOAD, MVT::f32, MVT::f16, Expand);
+ setLoadExtAction(ISD::EXTLOAD, MVT::f64, MVT::f16, Expand);
+ setLoadExtAction(ISD::EXTLOAD, MVT::f64, MVT::f32, Expand);
+ setLoadExtAction(ISD::EXTLOAD, MVT::v2f32, MVT::v2f16, Expand);
+ setLoadExtAction(ISD::EXTLOAD, MVT::v2f64, MVT::v2f16, Expand);
+ setLoadExtAction(ISD::EXTLOAD, MVT::v2f64, MVT::v2f32, Expand);
+ setLoadExtAction(ISD::EXTLOAD, MVT::v4f32, MVT::v4f16, Expand);
+ setLoadExtAction(ISD::EXTLOAD, MVT::v4f64, MVT::v4f16, Expand);
+ setLoadExtAction(ISD::EXTLOAD, MVT::v4f64, MVT::v4f32, Expand);
// Turn FP truncstore into trunc + store.
+ // FIXME: vector types should also be expanded
+ setTruncStoreAction(MVT::f32, MVT::f16, Expand);
+ setTruncStoreAction(MVT::f64, MVT::f16, Expand);
setTruncStoreAction(MVT::f64, MVT::f32, Expand);
// PTX does not support load / store predicate registers
setOperationAction(ISD::LOAD, MVT::i1, Custom);
setOperationAction(ISD::STORE, MVT::i1, Custom);
- setLoadExtAction(ISD::SEXTLOAD, MVT::i1, Promote);
- setLoadExtAction(ISD::ZEXTLOAD, MVT::i1, Promote);
- setTruncStoreAction(MVT::i64, MVT::i1, Expand);
- setTruncStoreAction(MVT::i32, MVT::i1, Expand);
- setTruncStoreAction(MVT::i16, MVT::i1, Expand);
- setTruncStoreAction(MVT::i8, MVT::i1, Expand);
+ for (MVT VT : MVT::integer_valuetypes()) {
+ setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i1, Promote);
+ setLoadExtAction(ISD::ZEXTLOAD, VT, MVT::i1, Promote);
+ setTruncStoreAction(VT, MVT::i1, Expand);
+ }
// This is legal in NVPTX
setOperationAction(ISD::ConstantFP, MVT::f64, Legal);
setOperationAction(ISD::ADDE, MVT::i64, Expand);
// Register custom handling for vector loads/stores
- for (int i = MVT::FIRST_VECTOR_VALUETYPE; i <= MVT::LAST_VECTOR_VALUETYPE;
- ++i) {
- MVT VT = (MVT::SimpleValueType) i;
+ for (MVT VT : MVT::vector_valuetypes()) {
if (IsPTXVectorType(VT)) {
setOperationAction(ISD::LOAD, VT, Custom);
setOperationAction(ISD::STORE, VT, Custom);
setOperationAction(ISD::CTPOP, MVT::i32, Legal);
setOperationAction(ISD::CTPOP, MVT::i64, Legal);
+ // PTX does not directly support SELP of i1, so promote to i32 first
+ setOperationAction(ISD::SELECT, MVT::i1, Custom);
+
// We have some custom DAG combine patterns for these nodes
setTargetDAGCombine(ISD::ADD);
setTargetDAGCombine(ISD::AND);
setTargetDAGCombine(ISD::FADD);
setTargetDAGCombine(ISD::MUL);
setTargetDAGCombine(ISD::SHL);
+ setTargetDAGCombine(ISD::SELECT);
// Now deduce the information based on the above mentioned
// actions
- computeRegisterProperties();
+ computeRegisterProperties(STI.getRegisterInfo());
}
const char *NVPTXTargetLowering::getTargetNodeName(unsigned Opcode) const {
- switch (Opcode) {
- default:
- return nullptr;
+ switch ((NVPTXISD::NodeType)Opcode) {
+ case NVPTXISD::FIRST_NUMBER:
+ break;
case NVPTXISD::CALL:
return "NVPTXISD::CALL";
case NVPTXISD::RET_FLAG:
return "NVPTXISD::RET_FLAG";
+ case NVPTXISD::LOAD_PARAM:
+ return "NVPTXISD::LOAD_PARAM";
case NVPTXISD::Wrapper:
return "NVPTXISD::Wrapper";
case NVPTXISD::DeclareParam:
return "NVPTXISD::DeclareScalarParam";
case NVPTXISD::DeclareRet:
return "NVPTXISD::DeclareRet";
+ case NVPTXISD::DeclareScalarRet:
+ return "NVPTXISD::DeclareScalarRet";
case NVPTXISD::DeclareRetParam:
return "NVPTXISD::DeclareRetParam";
case NVPTXISD::PrintCall:
return "NVPTXISD::PrintCall";
+ case NVPTXISD::PrintCallUni:
+ return "NVPTXISD::PrintCallUni";
case NVPTXISD::LoadParam:
return "NVPTXISD::LoadParam";
case NVPTXISD::LoadParamV2:
return "NVPTXISD::FUN_SHFR_CLAMP";
case NVPTXISD::IMAD:
return "NVPTXISD::IMAD";
+ case NVPTXISD::Dummy:
+ return "NVPTXISD::Dummy";
case NVPTXISD::MUL_WIDE_SIGNED:
return "NVPTXISD::MUL_WIDE_SIGNED";
case NVPTXISD::MUL_WIDE_UNSIGNED:
return "NVPTXISD::MUL_WIDE_UNSIGNED";
- case NVPTXISD::Tex1DFloatI32: return "NVPTXISD::Tex1DFloatI32";
+ case NVPTXISD::Tex1DFloatS32: return "NVPTXISD::Tex1DFloatS32";
case NVPTXISD::Tex1DFloatFloat: return "NVPTXISD::Tex1DFloatFloat";
case NVPTXISD::Tex1DFloatFloatLevel:
return "NVPTXISD::Tex1DFloatFloatLevel";
case NVPTXISD::Tex1DFloatFloatGrad:
return "NVPTXISD::Tex1DFloatFloatGrad";
- case NVPTXISD::Tex1DI32I32: return "NVPTXISD::Tex1DI32I32";
- case NVPTXISD::Tex1DI32Float: return "NVPTXISD::Tex1DI32Float";
- case NVPTXISD::Tex1DI32FloatLevel:
- return "NVPTXISD::Tex1DI32FloatLevel";
- case NVPTXISD::Tex1DI32FloatGrad:
- return "NVPTXISD::Tex1DI32FloatGrad";
- case NVPTXISD::Tex1DArrayFloatI32: return "NVPTXISD::Tex2DArrayFloatI32";
- case NVPTXISD::Tex1DArrayFloatFloat: return "NVPTXISD::Tex2DArrayFloatFloat";
+ case NVPTXISD::Tex1DS32S32: return "NVPTXISD::Tex1DS32S32";
+ case NVPTXISD::Tex1DS32Float: return "NVPTXISD::Tex1DS32Float";
+ case NVPTXISD::Tex1DS32FloatLevel:
+ return "NVPTXISD::Tex1DS32FloatLevel";
+ case NVPTXISD::Tex1DS32FloatGrad:
+ return "NVPTXISD::Tex1DS32FloatGrad";
+ case NVPTXISD::Tex1DU32S32: return "NVPTXISD::Tex1DU32S32";
+ case NVPTXISD::Tex1DU32Float: return "NVPTXISD::Tex1DU32Float";
+ case NVPTXISD::Tex1DU32FloatLevel:
+ return "NVPTXISD::Tex1DU32FloatLevel";
+ case NVPTXISD::Tex1DU32FloatGrad:
+ return "NVPTXISD::Tex1DU32FloatGrad";
+ case NVPTXISD::Tex1DArrayFloatS32: return "NVPTXISD::Tex1DArrayFloatS32";
+ case NVPTXISD::Tex1DArrayFloatFloat: return "NVPTXISD::Tex1DArrayFloatFloat";
case NVPTXISD::Tex1DArrayFloatFloatLevel:
- return "NVPTXISD::Tex2DArrayFloatFloatLevel";
+ return "NVPTXISD::Tex1DArrayFloatFloatLevel";
case NVPTXISD::Tex1DArrayFloatFloatGrad:
- return "NVPTXISD::Tex2DArrayFloatFloatGrad";
- case NVPTXISD::Tex1DArrayI32I32: return "NVPTXISD::Tex2DArrayI32I32";
- case NVPTXISD::Tex1DArrayI32Float: return "NVPTXISD::Tex2DArrayI32Float";
- case NVPTXISD::Tex1DArrayI32FloatLevel:
- return "NVPTXISD::Tex2DArrayI32FloatLevel";
- case NVPTXISD::Tex1DArrayI32FloatGrad:
- return "NVPTXISD::Tex2DArrayI32FloatGrad";
- case NVPTXISD::Tex2DFloatI32: return "NVPTXISD::Tex2DFloatI32";
+ return "NVPTXISD::Tex1DArrayFloatFloatGrad";
+ case NVPTXISD::Tex1DArrayS32S32: return "NVPTXISD::Tex1DArrayS32S32";
+ case NVPTXISD::Tex1DArrayS32Float: return "NVPTXISD::Tex1DArrayS32Float";
+ case NVPTXISD::Tex1DArrayS32FloatLevel:
+ return "NVPTXISD::Tex1DArrayS32FloatLevel";
+ case NVPTXISD::Tex1DArrayS32FloatGrad:
+ return "NVPTXISD::Tex1DArrayS32FloatGrad";
+ case NVPTXISD::Tex1DArrayU32S32: return "NVPTXISD::Tex1DArrayU32S32";
+ case NVPTXISD::Tex1DArrayU32Float: return "NVPTXISD::Tex1DArrayU32Float";
+ case NVPTXISD::Tex1DArrayU32FloatLevel:
+ return "NVPTXISD::Tex1DArrayU32FloatLevel";
+ case NVPTXISD::Tex1DArrayU32FloatGrad:
+ return "NVPTXISD::Tex1DArrayU32FloatGrad";
+ case NVPTXISD::Tex2DFloatS32: return "NVPTXISD::Tex2DFloatS32";
case NVPTXISD::Tex2DFloatFloat: return "NVPTXISD::Tex2DFloatFloat";
case NVPTXISD::Tex2DFloatFloatLevel:
return "NVPTXISD::Tex2DFloatFloatLevel";
case NVPTXISD::Tex2DFloatFloatGrad:
return "NVPTXISD::Tex2DFloatFloatGrad";
- case NVPTXISD::Tex2DI32I32: return "NVPTXISD::Tex2DI32I32";
- case NVPTXISD::Tex2DI32Float: return "NVPTXISD::Tex2DI32Float";
- case NVPTXISD::Tex2DI32FloatLevel:
- return "NVPTXISD::Tex2DI32FloatLevel";
- case NVPTXISD::Tex2DI32FloatGrad:
- return "NVPTXISD::Tex2DI32FloatGrad";
- case NVPTXISD::Tex2DArrayFloatI32: return "NVPTXISD::Tex2DArrayFloatI32";
+ case NVPTXISD::Tex2DS32S32: return "NVPTXISD::Tex2DS32S32";
+ case NVPTXISD::Tex2DS32Float: return "NVPTXISD::Tex2DS32Float";
+ case NVPTXISD::Tex2DS32FloatLevel:
+ return "NVPTXISD::Tex2DS32FloatLevel";
+ case NVPTXISD::Tex2DS32FloatGrad:
+ return "NVPTXISD::Tex2DS32FloatGrad";
+ case NVPTXISD::Tex2DU32S32: return "NVPTXISD::Tex2DU32S32";
+ case NVPTXISD::Tex2DU32Float: return "NVPTXISD::Tex2DU32Float";
+ case NVPTXISD::Tex2DU32FloatLevel:
+ return "NVPTXISD::Tex2DU32FloatLevel";
+ case NVPTXISD::Tex2DU32FloatGrad:
+ return "NVPTXISD::Tex2DU32FloatGrad";
+ case NVPTXISD::Tex2DArrayFloatS32: return "NVPTXISD::Tex2DArrayFloatS32";
case NVPTXISD::Tex2DArrayFloatFloat: return "NVPTXISD::Tex2DArrayFloatFloat";
case NVPTXISD::Tex2DArrayFloatFloatLevel:
return "NVPTXISD::Tex2DArrayFloatFloatLevel";
case NVPTXISD::Tex2DArrayFloatFloatGrad:
return "NVPTXISD::Tex2DArrayFloatFloatGrad";
- case NVPTXISD::Tex2DArrayI32I32: return "NVPTXISD::Tex2DArrayI32I32";
- case NVPTXISD::Tex2DArrayI32Float: return "NVPTXISD::Tex2DArrayI32Float";
- case NVPTXISD::Tex2DArrayI32FloatLevel:
- return "NVPTXISD::Tex2DArrayI32FloatLevel";
- case NVPTXISD::Tex2DArrayI32FloatGrad:
- return "NVPTXISD::Tex2DArrayI32FloatGrad";
- case NVPTXISD::Tex3DFloatI32: return "NVPTXISD::Tex3DFloatI32";
+ case NVPTXISD::Tex2DArrayS32S32: return "NVPTXISD::Tex2DArrayS32S32";
+ case NVPTXISD::Tex2DArrayS32Float: return "NVPTXISD::Tex2DArrayS32Float";
+ case NVPTXISD::Tex2DArrayS32FloatLevel:
+ return "NVPTXISD::Tex2DArrayS32FloatLevel";
+ case NVPTXISD::Tex2DArrayS32FloatGrad:
+ return "NVPTXISD::Tex2DArrayS32FloatGrad";
+ case NVPTXISD::Tex2DArrayU32S32: return "NVPTXISD::Tex2DArrayU32S32";
+ case NVPTXISD::Tex2DArrayU32Float: return "NVPTXISD::Tex2DArrayU32Float";
+ case NVPTXISD::Tex2DArrayU32FloatLevel:
+ return "NVPTXISD::Tex2DArrayU32FloatLevel";
+ case NVPTXISD::Tex2DArrayU32FloatGrad:
+ return "NVPTXISD::Tex2DArrayU32FloatGrad";
+ case NVPTXISD::Tex3DFloatS32: return "NVPTXISD::Tex3DFloatS32";
case NVPTXISD::Tex3DFloatFloat: return "NVPTXISD::Tex3DFloatFloat";
case NVPTXISD::Tex3DFloatFloatLevel:
return "NVPTXISD::Tex3DFloatFloatLevel";
case NVPTXISD::Tex3DFloatFloatGrad:
return "NVPTXISD::Tex3DFloatFloatGrad";
- case NVPTXISD::Tex3DI32I32: return "NVPTXISD::Tex3DI32I32";
- case NVPTXISD::Tex3DI32Float: return "NVPTXISD::Tex3DI32Float";
- case NVPTXISD::Tex3DI32FloatLevel:
- return "NVPTXISD::Tex3DI32FloatLevel";
- case NVPTXISD::Tex3DI32FloatGrad:
- return "NVPTXISD::Tex3DI32FloatGrad";
+ case NVPTXISD::Tex3DS32S32: return "NVPTXISD::Tex3DS32S32";
+ case NVPTXISD::Tex3DS32Float: return "NVPTXISD::Tex3DS32Float";
+ case NVPTXISD::Tex3DS32FloatLevel:
+ return "NVPTXISD::Tex3DS32FloatLevel";
+ case NVPTXISD::Tex3DS32FloatGrad:
+ return "NVPTXISD::Tex3DS32FloatGrad";
+ case NVPTXISD::Tex3DU32S32: return "NVPTXISD::Tex3DU32S32";
+ case NVPTXISD::Tex3DU32Float: return "NVPTXISD::Tex3DU32Float";
+ case NVPTXISD::Tex3DU32FloatLevel:
+ return "NVPTXISD::Tex3DU32FloatLevel";
+ case NVPTXISD::Tex3DU32FloatGrad:
+ return "NVPTXISD::Tex3DU32FloatGrad";
+ case NVPTXISD::TexCubeFloatFloat: return "NVPTXISD::TexCubeFloatFloat";
+ case NVPTXISD::TexCubeFloatFloatLevel:
+ return "NVPTXISD::TexCubeFloatFloatLevel";
+ case NVPTXISD::TexCubeS32Float: return "NVPTXISD::TexCubeS32Float";
+ case NVPTXISD::TexCubeS32FloatLevel:
+ return "NVPTXISD::TexCubeS32FloatLevel";
+ case NVPTXISD::TexCubeU32Float: return "NVPTXISD::TexCubeU32Float";
+ case NVPTXISD::TexCubeU32FloatLevel:
+ return "NVPTXISD::TexCubeU32FloatLevel";
+ case NVPTXISD::TexCubeArrayFloatFloat:
+ return "NVPTXISD::TexCubeArrayFloatFloat";
+ case NVPTXISD::TexCubeArrayFloatFloatLevel:
+ return "NVPTXISD::TexCubeArrayFloatFloatLevel";
+ case NVPTXISD::TexCubeArrayS32Float:
+ return "NVPTXISD::TexCubeArrayS32Float";
+ case NVPTXISD::TexCubeArrayS32FloatLevel:
+ return "NVPTXISD::TexCubeArrayS32FloatLevel";
+ case NVPTXISD::TexCubeArrayU32Float:
+ return "NVPTXISD::TexCubeArrayU32Float";
+ case NVPTXISD::TexCubeArrayU32FloatLevel:
+ return "NVPTXISD::TexCubeArrayU32FloatLevel";
+ case NVPTXISD::Tld4R2DFloatFloat:
+ return "NVPTXISD::Tld4R2DFloatFloat";
+ case NVPTXISD::Tld4G2DFloatFloat:
+ return "NVPTXISD::Tld4G2DFloatFloat";
+ case NVPTXISD::Tld4B2DFloatFloat:
+ return "NVPTXISD::Tld4B2DFloatFloat";
+ case NVPTXISD::Tld4A2DFloatFloat:
+ return "NVPTXISD::Tld4A2DFloatFloat";
+ case NVPTXISD::Tld4R2DS64Float:
+ return "NVPTXISD::Tld4R2DS64Float";
+ case NVPTXISD::Tld4G2DS64Float:
+ return "NVPTXISD::Tld4G2DS64Float";
+ case NVPTXISD::Tld4B2DS64Float:
+ return "NVPTXISD::Tld4B2DS64Float";
+ case NVPTXISD::Tld4A2DS64Float:
+ return "NVPTXISD::Tld4A2DS64Float";
+ case NVPTXISD::Tld4R2DU64Float:
+ return "NVPTXISD::Tld4R2DU64Float";
+ case NVPTXISD::Tld4G2DU64Float:
+ return "NVPTXISD::Tld4G2DU64Float";
+ case NVPTXISD::Tld4B2DU64Float:
+ return "NVPTXISD::Tld4B2DU64Float";
+ case NVPTXISD::Tld4A2DU64Float:
+ return "NVPTXISD::Tld4A2DU64Float";
+
+ case NVPTXISD::TexUnified1DFloatS32:
+ return "NVPTXISD::TexUnified1DFloatS32";
+ case NVPTXISD::TexUnified1DFloatFloat:
+ return "NVPTXISD::TexUnified1DFloatFloat";
+ case NVPTXISD::TexUnified1DFloatFloatLevel:
+ return "NVPTXISD::TexUnified1DFloatFloatLevel";
+ case NVPTXISD::TexUnified1DFloatFloatGrad:
+ return "NVPTXISD::TexUnified1DFloatFloatGrad";
+ case NVPTXISD::TexUnified1DS32S32:
+ return "NVPTXISD::TexUnified1DS32S32";
+ case NVPTXISD::TexUnified1DS32Float:
+ return "NVPTXISD::TexUnified1DS32Float";
+ case NVPTXISD::TexUnified1DS32FloatLevel:
+ return "NVPTXISD::TexUnified1DS32FloatLevel";
+ case NVPTXISD::TexUnified1DS32FloatGrad:
+ return "NVPTXISD::TexUnified1DS32FloatGrad";
+ case NVPTXISD::TexUnified1DU32S32:
+ return "NVPTXISD::TexUnified1DU32S32";
+ case NVPTXISD::TexUnified1DU32Float:
+ return "NVPTXISD::TexUnified1DU32Float";
+ case NVPTXISD::TexUnified1DU32FloatLevel:
+ return "NVPTXISD::TexUnified1DU32FloatLevel";
+ case NVPTXISD::TexUnified1DU32FloatGrad:
+ return "NVPTXISD::TexUnified1DU32FloatGrad";
+ case NVPTXISD::TexUnified1DArrayFloatS32:
+ return "NVPTXISD::TexUnified1DArrayFloatS32";
+ case NVPTXISD::TexUnified1DArrayFloatFloat:
+ return "NVPTXISD::TexUnified1DArrayFloatFloat";
+ case NVPTXISD::TexUnified1DArrayFloatFloatLevel:
+ return "NVPTXISD::TexUnified1DArrayFloatFloatLevel";
+ case NVPTXISD::TexUnified1DArrayFloatFloatGrad:
+ return "NVPTXISD::TexUnified1DArrayFloatFloatGrad";
+ case NVPTXISD::TexUnified1DArrayS32S32:
+ return "NVPTXISD::TexUnified1DArrayS32S32";
+ case NVPTXISD::TexUnified1DArrayS32Float:
+ return "NVPTXISD::TexUnified1DArrayS32Float";
+ case NVPTXISD::TexUnified1DArrayS32FloatLevel:
+ return "NVPTXISD::TexUnified1DArrayS32FloatLevel";
+ case NVPTXISD::TexUnified1DArrayS32FloatGrad:
+ return "NVPTXISD::TexUnified1DArrayS32FloatGrad";
+ case NVPTXISD::TexUnified1DArrayU32S32:
+ return "NVPTXISD::TexUnified1DArrayU32S32";
+ case NVPTXISD::TexUnified1DArrayU32Float:
+ return "NVPTXISD::TexUnified1DArrayU32Float";
+ case NVPTXISD::TexUnified1DArrayU32FloatLevel:
+ return "NVPTXISD::TexUnified1DArrayU32FloatLevel";
+ case NVPTXISD::TexUnified1DArrayU32FloatGrad:
+ return "NVPTXISD::TexUnified1DArrayU32FloatGrad";
+ case NVPTXISD::TexUnified2DFloatS32:
+ return "NVPTXISD::TexUnified2DFloatS32";
+ case NVPTXISD::TexUnified2DFloatFloat:
+ return "NVPTXISD::TexUnified2DFloatFloat";
+ case NVPTXISD::TexUnified2DFloatFloatLevel:
+ return "NVPTXISD::TexUnified2DFloatFloatLevel";
+ case NVPTXISD::TexUnified2DFloatFloatGrad:
+ return "NVPTXISD::TexUnified2DFloatFloatGrad";
+ case NVPTXISD::TexUnified2DS32S32:
+ return "NVPTXISD::TexUnified2DS32S32";
+ case NVPTXISD::TexUnified2DS32Float:
+ return "NVPTXISD::TexUnified2DS32Float";
+ case NVPTXISD::TexUnified2DS32FloatLevel:
+ return "NVPTXISD::TexUnified2DS32FloatLevel";
+ case NVPTXISD::TexUnified2DS32FloatGrad:
+ return "NVPTXISD::TexUnified2DS32FloatGrad";
+ case NVPTXISD::TexUnified2DU32S32:
+ return "NVPTXISD::TexUnified2DU32S32";
+ case NVPTXISD::TexUnified2DU32Float:
+ return "NVPTXISD::TexUnified2DU32Float";
+ case NVPTXISD::TexUnified2DU32FloatLevel:
+ return "NVPTXISD::TexUnified2DU32FloatLevel";
+ case NVPTXISD::TexUnified2DU32FloatGrad:
+ return "NVPTXISD::TexUnified2DU32FloatGrad";
+ case NVPTXISD::TexUnified2DArrayFloatS32:
+ return "NVPTXISD::TexUnified2DArrayFloatS32";
+ case NVPTXISD::TexUnified2DArrayFloatFloat:
+ return "NVPTXISD::TexUnified2DArrayFloatFloat";
+ case NVPTXISD::TexUnified2DArrayFloatFloatLevel:
+ return "NVPTXISD::TexUnified2DArrayFloatFloatLevel";
+ case NVPTXISD::TexUnified2DArrayFloatFloatGrad:
+ return "NVPTXISD::TexUnified2DArrayFloatFloatGrad";
+ case NVPTXISD::TexUnified2DArrayS32S32:
+ return "NVPTXISD::TexUnified2DArrayS32S32";
+ case NVPTXISD::TexUnified2DArrayS32Float:
+ return "NVPTXISD::TexUnified2DArrayS32Float";
+ case NVPTXISD::TexUnified2DArrayS32FloatLevel:
+ return "NVPTXISD::TexUnified2DArrayS32FloatLevel";
+ case NVPTXISD::TexUnified2DArrayS32FloatGrad:
+ return "NVPTXISD::TexUnified2DArrayS32FloatGrad";
+ case NVPTXISD::TexUnified2DArrayU32S32:
+ return "NVPTXISD::TexUnified2DArrayU32S32";
+ case NVPTXISD::TexUnified2DArrayU32Float:
+ return "NVPTXISD::TexUnified2DArrayU32Float";
+ case NVPTXISD::TexUnified2DArrayU32FloatLevel:
+ return "NVPTXISD::TexUnified2DArrayU32FloatLevel";
+ case NVPTXISD::TexUnified2DArrayU32FloatGrad:
+ return "NVPTXISD::TexUnified2DArrayU32FloatGrad";
+ case NVPTXISD::TexUnified3DFloatS32:
+ return "NVPTXISD::TexUnified3DFloatS32";
+ case NVPTXISD::TexUnified3DFloatFloat:
+ return "NVPTXISD::TexUnified3DFloatFloat";
+ case NVPTXISD::TexUnified3DFloatFloatLevel:
+ return "NVPTXISD::TexUnified3DFloatFloatLevel";
+ case NVPTXISD::TexUnified3DFloatFloatGrad:
+ return "NVPTXISD::TexUnified3DFloatFloatGrad";
+ case NVPTXISD::TexUnified3DS32S32:
+ return "NVPTXISD::TexUnified3DS32S32";
+ case NVPTXISD::TexUnified3DS32Float:
+ return "NVPTXISD::TexUnified3DS32Float";
+ case NVPTXISD::TexUnified3DS32FloatLevel:
+ return "NVPTXISD::TexUnified3DS32FloatLevel";
+ case NVPTXISD::TexUnified3DS32FloatGrad:
+ return "NVPTXISD::TexUnified3DS32FloatGrad";
+ case NVPTXISD::TexUnified3DU32S32:
+ return "NVPTXISD::TexUnified3DU32S32";
+ case NVPTXISD::TexUnified3DU32Float:
+ return "NVPTXISD::TexUnified3DU32Float";
+ case NVPTXISD::TexUnified3DU32FloatLevel:
+ return "NVPTXISD::TexUnified3DU32FloatLevel";
+ case NVPTXISD::TexUnified3DU32FloatGrad:
+ return "NVPTXISD::TexUnified3DU32FloatGrad";
+ case NVPTXISD::TexUnifiedCubeFloatFloat:
+ return "NVPTXISD::TexUnifiedCubeFloatFloat";
+ case NVPTXISD::TexUnifiedCubeFloatFloatLevel:
+ return "NVPTXISD::TexUnifiedCubeFloatFloatLevel";
+ case NVPTXISD::TexUnifiedCubeS32Float:
+ return "NVPTXISD::TexUnifiedCubeS32Float";
+ case NVPTXISD::TexUnifiedCubeS32FloatLevel:
+ return "NVPTXISD::TexUnifiedCubeS32FloatLevel";
+ case NVPTXISD::TexUnifiedCubeU32Float:
+ return "NVPTXISD::TexUnifiedCubeU32Float";
+ case NVPTXISD::TexUnifiedCubeU32FloatLevel:
+ return "NVPTXISD::TexUnifiedCubeU32FloatLevel";
+ case NVPTXISD::TexUnifiedCubeArrayFloatFloat:
+ return "NVPTXISD::TexUnifiedCubeArrayFloatFloat";
+ case NVPTXISD::TexUnifiedCubeArrayFloatFloatLevel:
+ return "NVPTXISD::TexUnifiedCubeArrayFloatFloatLevel";
+ case NVPTXISD::TexUnifiedCubeArrayS32Float:
+ return "NVPTXISD::TexUnifiedCubeArrayS32Float";
+ case NVPTXISD::TexUnifiedCubeArrayS32FloatLevel:
+ return "NVPTXISD::TexUnifiedCubeArrayS32FloatLevel";
+ case NVPTXISD::TexUnifiedCubeArrayU32Float:
+ return "NVPTXISD::TexUnifiedCubeArrayU32Float";
+ case NVPTXISD::TexUnifiedCubeArrayU32FloatLevel:
+ return "NVPTXISD::TexUnifiedCubeArrayU32FloatLevel";
+ case NVPTXISD::Tld4UnifiedR2DFloatFloat:
+ return "NVPTXISD::Tld4UnifiedR2DFloatFloat";
+ case NVPTXISD::Tld4UnifiedG2DFloatFloat:
+ return "NVPTXISD::Tld4UnifiedG2DFloatFloat";
+ case NVPTXISD::Tld4UnifiedB2DFloatFloat:
+ return "NVPTXISD::Tld4UnifiedB2DFloatFloat";
+ case NVPTXISD::Tld4UnifiedA2DFloatFloat:
+ return "NVPTXISD::Tld4UnifiedA2DFloatFloat";
+ case NVPTXISD::Tld4UnifiedR2DS64Float:
+ return "NVPTXISD::Tld4UnifiedR2DS64Float";
+ case NVPTXISD::Tld4UnifiedG2DS64Float:
+ return "NVPTXISD::Tld4UnifiedG2DS64Float";
+ case NVPTXISD::Tld4UnifiedB2DS64Float:
+ return "NVPTXISD::Tld4UnifiedB2DS64Float";
+ case NVPTXISD::Tld4UnifiedA2DS64Float:
+ return "NVPTXISD::Tld4UnifiedA2DS64Float";
+ case NVPTXISD::Tld4UnifiedR2DU64Float:
+ return "NVPTXISD::Tld4UnifiedR2DU64Float";
+ case NVPTXISD::Tld4UnifiedG2DU64Float:
+ return "NVPTXISD::Tld4UnifiedG2DU64Float";
+ case NVPTXISD::Tld4UnifiedB2DU64Float:
+ return "NVPTXISD::Tld4UnifiedB2DU64Float";
+ case NVPTXISD::Tld4UnifiedA2DU64Float:
+ return "NVPTXISD::Tld4UnifiedA2DU64Float";
+
+ case NVPTXISD::Suld1DI8Clamp: return "NVPTXISD::Suld1DI8Clamp";
+ case NVPTXISD::Suld1DI16Clamp: return "NVPTXISD::Suld1DI16Clamp";
+ case NVPTXISD::Suld1DI32Clamp: return "NVPTXISD::Suld1DI32Clamp";
+ case NVPTXISD::Suld1DI64Clamp: return "NVPTXISD::Suld1DI64Clamp";
+ case NVPTXISD::Suld1DV2I8Clamp: return "NVPTXISD::Suld1DV2I8Clamp";
+ case NVPTXISD::Suld1DV2I16Clamp: return "NVPTXISD::Suld1DV2I16Clamp";
+ case NVPTXISD::Suld1DV2I32Clamp: return "NVPTXISD::Suld1DV2I32Clamp";
+ case NVPTXISD::Suld1DV2I64Clamp: return "NVPTXISD::Suld1DV2I64Clamp";
+ case NVPTXISD::Suld1DV4I8Clamp: return "NVPTXISD::Suld1DV4I8Clamp";
+ case NVPTXISD::Suld1DV4I16Clamp: return "NVPTXISD::Suld1DV4I16Clamp";
+ case NVPTXISD::Suld1DV4I32Clamp: return "NVPTXISD::Suld1DV4I32Clamp";
+
+ case NVPTXISD::Suld1DArrayI8Clamp: return "NVPTXISD::Suld1DArrayI8Clamp";
+ case NVPTXISD::Suld1DArrayI16Clamp: return "NVPTXISD::Suld1DArrayI16Clamp";
+ case NVPTXISD::Suld1DArrayI32Clamp: return "NVPTXISD::Suld1DArrayI32Clamp";
+ case NVPTXISD::Suld1DArrayI64Clamp: return "NVPTXISD::Suld1DArrayI64Clamp";
+ case NVPTXISD::Suld1DArrayV2I8Clamp: return "NVPTXISD::Suld1DArrayV2I8Clamp";
+ case NVPTXISD::Suld1DArrayV2I16Clamp:return "NVPTXISD::Suld1DArrayV2I16Clamp";
+ case NVPTXISD::Suld1DArrayV2I32Clamp:return "NVPTXISD::Suld1DArrayV2I32Clamp";
+ case NVPTXISD::Suld1DArrayV2I64Clamp:return "NVPTXISD::Suld1DArrayV2I64Clamp";
+ case NVPTXISD::Suld1DArrayV4I8Clamp: return "NVPTXISD::Suld1DArrayV4I8Clamp";
+ case NVPTXISD::Suld1DArrayV4I16Clamp:return "NVPTXISD::Suld1DArrayV4I16Clamp";
+ case NVPTXISD::Suld1DArrayV4I32Clamp:return "NVPTXISD::Suld1DArrayV4I32Clamp";
+
+ case NVPTXISD::Suld2DI8Clamp: return "NVPTXISD::Suld2DI8Clamp";
+ case NVPTXISD::Suld2DI16Clamp: return "NVPTXISD::Suld2DI16Clamp";
+ case NVPTXISD::Suld2DI32Clamp: return "NVPTXISD::Suld2DI32Clamp";
+ case NVPTXISD::Suld2DI64Clamp: return "NVPTXISD::Suld2DI64Clamp";
+ case NVPTXISD::Suld2DV2I8Clamp: return "NVPTXISD::Suld2DV2I8Clamp";
+ case NVPTXISD::Suld2DV2I16Clamp: return "NVPTXISD::Suld2DV2I16Clamp";
+ case NVPTXISD::Suld2DV2I32Clamp: return "NVPTXISD::Suld2DV2I32Clamp";
+ case NVPTXISD::Suld2DV2I64Clamp: return "NVPTXISD::Suld2DV2I64Clamp";
+ case NVPTXISD::Suld2DV4I8Clamp: return "NVPTXISD::Suld2DV4I8Clamp";
+ case NVPTXISD::Suld2DV4I16Clamp: return "NVPTXISD::Suld2DV4I16Clamp";
+ case NVPTXISD::Suld2DV4I32Clamp: return "NVPTXISD::Suld2DV4I32Clamp";
+
+ case NVPTXISD::Suld2DArrayI8Clamp: return "NVPTXISD::Suld2DArrayI8Clamp";
+ case NVPTXISD::Suld2DArrayI16Clamp: return "NVPTXISD::Suld2DArrayI16Clamp";
+ case NVPTXISD::Suld2DArrayI32Clamp: return "NVPTXISD::Suld2DArrayI32Clamp";
+ case NVPTXISD::Suld2DArrayI64Clamp: return "NVPTXISD::Suld2DArrayI64Clamp";
+ case NVPTXISD::Suld2DArrayV2I8Clamp: return "NVPTXISD::Suld2DArrayV2I8Clamp";
+ case NVPTXISD::Suld2DArrayV2I16Clamp:return "NVPTXISD::Suld2DArrayV2I16Clamp";
+ case NVPTXISD::Suld2DArrayV2I32Clamp:return "NVPTXISD::Suld2DArrayV2I32Clamp";
+ case NVPTXISD::Suld2DArrayV2I64Clamp:return "NVPTXISD::Suld2DArrayV2I64Clamp";
+ case NVPTXISD::Suld2DArrayV4I8Clamp: return "NVPTXISD::Suld2DArrayV4I8Clamp";
+ case NVPTXISD::Suld2DArrayV4I16Clamp:return "NVPTXISD::Suld2DArrayV4I16Clamp";
+ case NVPTXISD::Suld2DArrayV4I32Clamp:return "NVPTXISD::Suld2DArrayV4I32Clamp";
+
+ case NVPTXISD::Suld3DI8Clamp: return "NVPTXISD::Suld3DI8Clamp";
+ case NVPTXISD::Suld3DI16Clamp: return "NVPTXISD::Suld3DI16Clamp";
+ case NVPTXISD::Suld3DI32Clamp: return "NVPTXISD::Suld3DI32Clamp";
+ case NVPTXISD::Suld3DI64Clamp: return "NVPTXISD::Suld3DI64Clamp";
+ case NVPTXISD::Suld3DV2I8Clamp: return "NVPTXISD::Suld3DV2I8Clamp";
+ case NVPTXISD::Suld3DV2I16Clamp: return "NVPTXISD::Suld3DV2I16Clamp";
+ case NVPTXISD::Suld3DV2I32Clamp: return "NVPTXISD::Suld3DV2I32Clamp";
+ case NVPTXISD::Suld3DV2I64Clamp: return "NVPTXISD::Suld3DV2I64Clamp";
+ case NVPTXISD::Suld3DV4I8Clamp: return "NVPTXISD::Suld3DV4I8Clamp";
+ case NVPTXISD::Suld3DV4I16Clamp: return "NVPTXISD::Suld3DV4I16Clamp";
+ case NVPTXISD::Suld3DV4I32Clamp: return "NVPTXISD::Suld3DV4I32Clamp";
case NVPTXISD::Suld1DI8Trap: return "NVPTXISD::Suld1DI8Trap";
case NVPTXISD::Suld1DI16Trap: return "NVPTXISD::Suld1DI16Trap";
case NVPTXISD::Suld1DI32Trap: return "NVPTXISD::Suld1DI32Trap";
+ case NVPTXISD::Suld1DI64Trap: return "NVPTXISD::Suld1DI64Trap";
case NVPTXISD::Suld1DV2I8Trap: return "NVPTXISD::Suld1DV2I8Trap";
case NVPTXISD::Suld1DV2I16Trap: return "NVPTXISD::Suld1DV2I16Trap";
case NVPTXISD::Suld1DV2I32Trap: return "NVPTXISD::Suld1DV2I32Trap";
+ case NVPTXISD::Suld1DV2I64Trap: return "NVPTXISD::Suld1DV2I64Trap";
case NVPTXISD::Suld1DV4I8Trap: return "NVPTXISD::Suld1DV4I8Trap";
case NVPTXISD::Suld1DV4I16Trap: return "NVPTXISD::Suld1DV4I16Trap";
case NVPTXISD::Suld1DV4I32Trap: return "NVPTXISD::Suld1DV4I32Trap";
case NVPTXISD::Suld1DArrayI8Trap: return "NVPTXISD::Suld1DArrayI8Trap";
case NVPTXISD::Suld1DArrayI16Trap: return "NVPTXISD::Suld1DArrayI16Trap";
case NVPTXISD::Suld1DArrayI32Trap: return "NVPTXISD::Suld1DArrayI32Trap";
+ case NVPTXISD::Suld1DArrayI64Trap: return "NVPTXISD::Suld1DArrayI64Trap";
case NVPTXISD::Suld1DArrayV2I8Trap: return "NVPTXISD::Suld1DArrayV2I8Trap";
case NVPTXISD::Suld1DArrayV2I16Trap: return "NVPTXISD::Suld1DArrayV2I16Trap";
case NVPTXISD::Suld1DArrayV2I32Trap: return "NVPTXISD::Suld1DArrayV2I32Trap";
+ case NVPTXISD::Suld1DArrayV2I64Trap: return "NVPTXISD::Suld1DArrayV2I64Trap";
case NVPTXISD::Suld1DArrayV4I8Trap: return "NVPTXISD::Suld1DArrayV4I8Trap";
case NVPTXISD::Suld1DArrayV4I16Trap: return "NVPTXISD::Suld1DArrayV4I16Trap";
case NVPTXISD::Suld1DArrayV4I32Trap: return "NVPTXISD::Suld1DArrayV4I32Trap";
case NVPTXISD::Suld2DI8Trap: return "NVPTXISD::Suld2DI8Trap";
case NVPTXISD::Suld2DI16Trap: return "NVPTXISD::Suld2DI16Trap";
case NVPTXISD::Suld2DI32Trap: return "NVPTXISD::Suld2DI32Trap";
+ case NVPTXISD::Suld2DI64Trap: return "NVPTXISD::Suld2DI64Trap";
case NVPTXISD::Suld2DV2I8Trap: return "NVPTXISD::Suld2DV2I8Trap";
case NVPTXISD::Suld2DV2I16Trap: return "NVPTXISD::Suld2DV2I16Trap";
case NVPTXISD::Suld2DV2I32Trap: return "NVPTXISD::Suld2DV2I32Trap";
+ case NVPTXISD::Suld2DV2I64Trap: return "NVPTXISD::Suld2DV2I64Trap";
case NVPTXISD::Suld2DV4I8Trap: return "NVPTXISD::Suld2DV4I8Trap";
case NVPTXISD::Suld2DV4I16Trap: return "NVPTXISD::Suld2DV4I16Trap";
case NVPTXISD::Suld2DV4I32Trap: return "NVPTXISD::Suld2DV4I32Trap";
case NVPTXISD::Suld2DArrayI8Trap: return "NVPTXISD::Suld2DArrayI8Trap";
case NVPTXISD::Suld2DArrayI16Trap: return "NVPTXISD::Suld2DArrayI16Trap";
case NVPTXISD::Suld2DArrayI32Trap: return "NVPTXISD::Suld2DArrayI32Trap";
+ case NVPTXISD::Suld2DArrayI64Trap: return "NVPTXISD::Suld2DArrayI64Trap";
case NVPTXISD::Suld2DArrayV2I8Trap: return "NVPTXISD::Suld2DArrayV2I8Trap";
case NVPTXISD::Suld2DArrayV2I16Trap: return "NVPTXISD::Suld2DArrayV2I16Trap";
case NVPTXISD::Suld2DArrayV2I32Trap: return "NVPTXISD::Suld2DArrayV2I32Trap";
+ case NVPTXISD::Suld2DArrayV2I64Trap: return "NVPTXISD::Suld2DArrayV2I64Trap";
case NVPTXISD::Suld2DArrayV4I8Trap: return "NVPTXISD::Suld2DArrayV4I8Trap";
case NVPTXISD::Suld2DArrayV4I16Trap: return "NVPTXISD::Suld2DArrayV4I16Trap";
case NVPTXISD::Suld2DArrayV4I32Trap: return "NVPTXISD::Suld2DArrayV4I32Trap";
case NVPTXISD::Suld3DI8Trap: return "NVPTXISD::Suld3DI8Trap";
case NVPTXISD::Suld3DI16Trap: return "NVPTXISD::Suld3DI16Trap";
case NVPTXISD::Suld3DI32Trap: return "NVPTXISD::Suld3DI32Trap";
+ case NVPTXISD::Suld3DI64Trap: return "NVPTXISD::Suld3DI64Trap";
case NVPTXISD::Suld3DV2I8Trap: return "NVPTXISD::Suld3DV2I8Trap";
case NVPTXISD::Suld3DV2I16Trap: return "NVPTXISD::Suld3DV2I16Trap";
case NVPTXISD::Suld3DV2I32Trap: return "NVPTXISD::Suld3DV2I32Trap";
+ case NVPTXISD::Suld3DV2I64Trap: return "NVPTXISD::Suld3DV2I64Trap";
case NVPTXISD::Suld3DV4I8Trap: return "NVPTXISD::Suld3DV4I8Trap";
case NVPTXISD::Suld3DV4I16Trap: return "NVPTXISD::Suld3DV4I16Trap";
case NVPTXISD::Suld3DV4I32Trap: return "NVPTXISD::Suld3DV4I32Trap";
+
+ case NVPTXISD::Suld1DI8Zero: return "NVPTXISD::Suld1DI8Zero";
+ case NVPTXISD::Suld1DI16Zero: return "NVPTXISD::Suld1DI16Zero";
+ case NVPTXISD::Suld1DI32Zero: return "NVPTXISD::Suld1DI32Zero";
+ case NVPTXISD::Suld1DI64Zero: return "NVPTXISD::Suld1DI64Zero";
+ case NVPTXISD::Suld1DV2I8Zero: return "NVPTXISD::Suld1DV2I8Zero";
+ case NVPTXISD::Suld1DV2I16Zero: return "NVPTXISD::Suld1DV2I16Zero";
+ case NVPTXISD::Suld1DV2I32Zero: return "NVPTXISD::Suld1DV2I32Zero";
+ case NVPTXISD::Suld1DV2I64Zero: return "NVPTXISD::Suld1DV2I64Zero";
+ case NVPTXISD::Suld1DV4I8Zero: return "NVPTXISD::Suld1DV4I8Zero";
+ case NVPTXISD::Suld1DV4I16Zero: return "NVPTXISD::Suld1DV4I16Zero";
+ case NVPTXISD::Suld1DV4I32Zero: return "NVPTXISD::Suld1DV4I32Zero";
+
+ case NVPTXISD::Suld1DArrayI8Zero: return "NVPTXISD::Suld1DArrayI8Zero";
+ case NVPTXISD::Suld1DArrayI16Zero: return "NVPTXISD::Suld1DArrayI16Zero";
+ case NVPTXISD::Suld1DArrayI32Zero: return "NVPTXISD::Suld1DArrayI32Zero";
+ case NVPTXISD::Suld1DArrayI64Zero: return "NVPTXISD::Suld1DArrayI64Zero";
+ case NVPTXISD::Suld1DArrayV2I8Zero: return "NVPTXISD::Suld1DArrayV2I8Zero";
+ case NVPTXISD::Suld1DArrayV2I16Zero: return "NVPTXISD::Suld1DArrayV2I16Zero";
+ case NVPTXISD::Suld1DArrayV2I32Zero: return "NVPTXISD::Suld1DArrayV2I32Zero";
+ case NVPTXISD::Suld1DArrayV2I64Zero: return "NVPTXISD::Suld1DArrayV2I64Zero";
+ case NVPTXISD::Suld1DArrayV4I8Zero: return "NVPTXISD::Suld1DArrayV4I8Zero";
+ case NVPTXISD::Suld1DArrayV4I16Zero: return "NVPTXISD::Suld1DArrayV4I16Zero";
+ case NVPTXISD::Suld1DArrayV4I32Zero: return "NVPTXISD::Suld1DArrayV4I32Zero";
+
+ case NVPTXISD::Suld2DI8Zero: return "NVPTXISD::Suld2DI8Zero";
+ case NVPTXISD::Suld2DI16Zero: return "NVPTXISD::Suld2DI16Zero";
+ case NVPTXISD::Suld2DI32Zero: return "NVPTXISD::Suld2DI32Zero";
+ case NVPTXISD::Suld2DI64Zero: return "NVPTXISD::Suld2DI64Zero";
+ case NVPTXISD::Suld2DV2I8Zero: return "NVPTXISD::Suld2DV2I8Zero";
+ case NVPTXISD::Suld2DV2I16Zero: return "NVPTXISD::Suld2DV2I16Zero";
+ case NVPTXISD::Suld2DV2I32Zero: return "NVPTXISD::Suld2DV2I32Zero";
+ case NVPTXISD::Suld2DV2I64Zero: return "NVPTXISD::Suld2DV2I64Zero";
+ case NVPTXISD::Suld2DV4I8Zero: return "NVPTXISD::Suld2DV4I8Zero";
+ case NVPTXISD::Suld2DV4I16Zero: return "NVPTXISD::Suld2DV4I16Zero";
+ case NVPTXISD::Suld2DV4I32Zero: return "NVPTXISD::Suld2DV4I32Zero";
+
+ case NVPTXISD::Suld2DArrayI8Zero: return "NVPTXISD::Suld2DArrayI8Zero";
+ case NVPTXISD::Suld2DArrayI16Zero: return "NVPTXISD::Suld2DArrayI16Zero";
+ case NVPTXISD::Suld2DArrayI32Zero: return "NVPTXISD::Suld2DArrayI32Zero";
+ case NVPTXISD::Suld2DArrayI64Zero: return "NVPTXISD::Suld2DArrayI64Zero";
+ case NVPTXISD::Suld2DArrayV2I8Zero: return "NVPTXISD::Suld2DArrayV2I8Zero";
+ case NVPTXISD::Suld2DArrayV2I16Zero: return "NVPTXISD::Suld2DArrayV2I16Zero";
+ case NVPTXISD::Suld2DArrayV2I32Zero: return "NVPTXISD::Suld2DArrayV2I32Zero";
+ case NVPTXISD::Suld2DArrayV2I64Zero: return "NVPTXISD::Suld2DArrayV2I64Zero";
+ case NVPTXISD::Suld2DArrayV4I8Zero: return "NVPTXISD::Suld2DArrayV4I8Zero";
+ case NVPTXISD::Suld2DArrayV4I16Zero: return "NVPTXISD::Suld2DArrayV4I16Zero";
+ case NVPTXISD::Suld2DArrayV4I32Zero: return "NVPTXISD::Suld2DArrayV4I32Zero";
+
+ case NVPTXISD::Suld3DI8Zero: return "NVPTXISD::Suld3DI8Zero";
+ case NVPTXISD::Suld3DI16Zero: return "NVPTXISD::Suld3DI16Zero";
+ case NVPTXISD::Suld3DI32Zero: return "NVPTXISD::Suld3DI32Zero";
+ case NVPTXISD::Suld3DI64Zero: return "NVPTXISD::Suld3DI64Zero";
+ case NVPTXISD::Suld3DV2I8Zero: return "NVPTXISD::Suld3DV2I8Zero";
+ case NVPTXISD::Suld3DV2I16Zero: return "NVPTXISD::Suld3DV2I16Zero";
+ case NVPTXISD::Suld3DV2I32Zero: return "NVPTXISD::Suld3DV2I32Zero";
+ case NVPTXISD::Suld3DV2I64Zero: return "NVPTXISD::Suld3DV2I64Zero";
+ case NVPTXISD::Suld3DV4I8Zero: return "NVPTXISD::Suld3DV4I8Zero";
+ case NVPTXISD::Suld3DV4I16Zero: return "NVPTXISD::Suld3DV4I16Zero";
+ case NVPTXISD::Suld3DV4I32Zero: return "NVPTXISD::Suld3DV4I32Zero";
}
+ return nullptr;
}
-bool NVPTXTargetLowering::shouldSplitVectorType(EVT VT) const {
- return VT.getScalarType() == MVT::i1;
+TargetLoweringBase::LegalizeTypeAction
+NVPTXTargetLowering::getPreferredVectorAction(EVT VT) const {
+ if (VT.getVectorNumElements() != 1 && VT.getScalarType() == MVT::i1)
+ return TypeSplitVector;
+
+ return TargetLoweringBase::getPreferredVectorAction(VT);
}
SDValue
NVPTXTargetLowering::LowerGlobalAddress(SDValue Op, SelectionDAG &DAG) const {
SDLoc dl(Op);
const GlobalValue *GV = cast<GlobalAddressSDNode>(Op)->getGlobal();
- Op = DAG.getTargetGlobalAddress(GV, dl, getPointerTy());
- return DAG.getNode(NVPTXISD::Wrapper, dl, getPointerTy(), Op);
+ auto PtrVT = getPointerTy(DAG.getDataLayout());
+ Op = DAG.getTargetGlobalAddress(GV, dl, PtrVT);
+ return DAG.getNode(NVPTXISD::Wrapper, dl, PtrVT, Op);
}
-std::string
-NVPTXTargetLowering::getPrototype(Type *retTy, const ArgListTy &Args,
- const SmallVectorImpl<ISD::OutputArg> &Outs,
- unsigned retAlignment,
- const ImmutableCallSite *CS) const {
+std::string NVPTXTargetLowering::getPrototype(
+ const DataLayout &DL, Type *retTy, const ArgListTy &Args,
+ const SmallVectorImpl<ISD::OutputArg> &Outs, unsigned retAlignment,
+ const ImmutableCallSite *CS) const {
+ auto PtrVT = getPointerTy(DL);
- bool isABI = (nvptxSubtarget.getSmVersion() >= 20);
+ bool isABI = (STI.getSmVersion() >= 20);
assert(isABI && "Non-ABI compilation is not supported");
if (!isABI)
return "";
O << "(";
if (retTy->isFloatingPointTy() || retTy->isIntegerTy()) {
unsigned size = 0;
- if (const IntegerType *ITy = dyn_cast<IntegerType>(retTy)) {
+ if (auto *ITy = dyn_cast<IntegerType>(retTy)) {
size = ITy->getBitWidth();
if (size < 32)
size = 32;
O << ".param .b" << size << " _";
} else if (isa<PointerType>(retTy)) {
- O << ".param .b" << getPointerTy().getSizeInBits() << " _";
+ O << ".param .b" << PtrVT.getSizeInBits() << " _";
+ } else if ((retTy->getTypeID() == Type::StructTyID) ||
+ isa<VectorType>(retTy)) {
+ auto &DL = CS->getCalledFunction()->getParent()->getDataLayout();
+ O << ".param .align " << retAlignment << " .b8 _["
+ << DL.getTypeAllocSize(retTy) << "]";
} else {
- if((retTy->getTypeID() == Type::StructTyID) ||
- isa<VectorType>(retTy)) {
- O << ".param .align "
- << retAlignment
- << " .b8 _["
- << getDataLayout()->getTypeAllocSize(retTy) << "]";
- } else {
- assert(false && "Unknown return type");
- }
+ llvm_unreachable("Unknown return type");
}
O << ") ";
}
O << "_ (";
bool first = true;
- MVT thePointerTy = getPointerTy();
unsigned OIdx = 0;
for (unsigned i = 0, e = Args.size(); i != e; ++i, ++OIdx) {
}
first = false;
- if (Outs[OIdx].Flags.isByVal() == false) {
+ if (!Outs[OIdx].Flags.isByVal()) {
if (Ty->isAggregateType() || Ty->isVectorTy()) {
unsigned align = 0;
const CallInst *CallI = cast<CallInst>(CS->getInstruction());
- const DataLayout *TD = getDataLayout();
// +1 because index 0 is reserved for return type alignment
if (!llvm::getAlign(*CallI, i + 1, align))
- align = TD->getABITypeAlignment(Ty);
- unsigned sz = TD->getTypeAllocSize(Ty);
+ align = DL.getABITypeAlignment(Ty);
+ unsigned sz = DL.getTypeAllocSize(Ty);
O << ".param .align " << align << " .b8 ";
O << "_";
O << "[" << sz << "]";
// update the index for Outs
SmallVector<EVT, 16> vtparts;
- ComputeValueVTs(*this, Ty, vtparts);
+ ComputeValueVTs(*this, DL, Ty, vtparts);
if (unsigned len = vtparts.size())
OIdx += len - 1;
continue;
}
// i8 types in IR will be i16 types in SDAG
- assert((getValueType(Ty) == Outs[OIdx].VT ||
- (getValueType(Ty) == MVT::i8 && Outs[OIdx].VT == MVT::i16)) &&
+ assert((getValueType(DL, Ty) == Outs[OIdx].VT ||
+ (getValueType(DL, Ty) == MVT::i8 && Outs[OIdx].VT == MVT::i16)) &&
"type mismatch between callee prototype and arguments");
// scalar type
unsigned sz = 0;
if (sz < 32)
sz = 32;
} else if (isa<PointerType>(Ty))
- sz = thePointerTy.getSizeInBits();
+ sz = PtrVT.getSizeInBits();
else
sz = Ty->getPrimitiveSizeInBits();
O << ".param .b" << sz << " ";
O << "_";
continue;
}
-
const PointerType *PTy = dyn_cast<PointerType>(Ty);
+
auto *PTy = dyn_cast<PointerType>(Ty);
assert(PTy && "Param with byval attribute should be a pointer type");
Type *ETy = PTy->getElementType();
unsigned align = Outs[OIdx].Flags.getByValAlign();
- unsigned sz = getDataLayout()->getTypeAllocSize(ETy);
+ unsigned sz = DL.getTypeAllocSize(ETy);
O << ".param .align " << align << " .b8 ";
O << "_";
O << "[" << sz << "]";
const ImmutableCallSite *CS,
Type *Ty,
unsigned Idx) const {
- const DataLayout *TD = getDataLayout();
unsigned Align = 0;
const Value *DirectCallee = CS->getCalledFunction();
// Call is indirect or alignment information is not available, fall back to
// the ABI type alignment
- return TD->getABITypeAlignment(Ty);
+ auto &DL = CS->getCaller()->getParent()->getDataLayout();
+ return DL.getABITypeAlignment(Ty);
}
SDValue NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
Type *retTy = CLI.RetTy;
ImmutableCallSite *CS = CLI.CS;
- bool isABI = (nvptxSubtarget.getSmVersion() >= 20);
+ bool isABI = (STI.getSmVersion() >= 20);
assert(isABI && "Non-ABI compilation is not supported");
if (!isABI)
return Chain;
- const DataLayout *TD = getDataLayout();
MachineFunction &MF = DAG.getMachineFunction();
const Function *F = MF.getFunction();
+ auto &DL = MF.getDataLayout();
SDValue tempChain = Chain;
- Chain =
- DAG.getCALLSEQ_START(Chain, DAG.getIntPtrConstant(uniqueCallSite, true),
- dl);
+ Chain = DAG.getCALLSEQ_START(Chain,
+ DAG.getIntPtrConstant(uniqueCallSite, dl, true),
+ dl);
SDValue InFlag = Chain.getValue(1);
unsigned paramCount = 0;
EVT VT = Outs[OIdx].VT;
Type *Ty = Args[i].Ty;
- if (Outs[OIdx].Flags.isByVal() == false) {
+ if (!Outs[OIdx].Flags.isByVal()) {
if (Ty->isAggregateType()) {
// aggregate
SmallVector<EVT, 16> vtparts;
SmallVector<uint64_t, 16> Offsets;
- ComputePTXValueVTs(*this, Ty, vtparts, &Offsets, 0);
+ ComputePTXValueVTs(*this, DAG.getDataLayout(), Ty, vtparts, &Offsets,
+ 0);
unsigned align = getArgumentAlignment(Callee, CS, Ty, paramCount + 1);
// declare .param .align <align> .b8 .param<n>[<size>];
- unsigned sz = TD->getTypeAllocSize(Ty);
+ unsigned sz = DL.getTypeAllocSize(Ty);
SDVTList DeclareParamVTs = DAG.getVTList(MVT::Other, MVT::Glue);
- SDValue DeclareParamOps[] = { Chain, DAG.getConstant(align, MVT::i32),
- DAG.getConstant(paramCount, MVT::i32),
- DAG.getConstant(sz, MVT::i32), InFlag };
+ SDValue DeclareParamOps[] = { Chain, DAG.getConstant(align, dl,
+ MVT::i32),
+ DAG.getConstant(paramCount, dl, MVT::i32),
+ DAG.getConstant(sz, dl, MVT::i32),
+ InFlag };
Chain = DAG.getNode(NVPTXISD::DeclareParam, dl, DeclareParamVTs,
DeclareParamOps);
InFlag = Chain.getValue(1);
for (unsigned j = 0, je = vtparts.size(); j != je; ++j) {
EVT elemtype = vtparts[j];
- unsigned ArgAlign = GCD(align, Offsets[j]);
+ unsigned ArgAlign = GreatestCommonDivisor64(align, Offsets[j]);
if (elemtype.isInteger() && (sz < 8))
sz = 8;
SDValue StVal = OutVals[OIdx];
}
SDVTList CopyParamVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue CopyParamOps[] = { Chain,
- DAG.getConstant(paramCount, MVT::i32),
- DAG.getConstant(Offsets[j], MVT::i32),
+ DAG.getConstant(paramCount, dl, MVT::i32),
+ DAG.getConstant(Offsets[j], dl, MVT::i32),
StVal, InFlag };
Chain = DAG.getMemIntrinsicNode(NVPTXISD::StoreParam, dl,
CopyParamVTs, CopyParamOps,
continue;
}
if (Ty->isVectorTy()) {
- EVT ObjectVT = getValueType(Ty);
+ EVT ObjectVT = getValueType(DL, Ty);
unsigned align = getArgumentAlignment(Callee, CS, Ty, paramCount + 1);
// declare .param .align <align> .b8 .param<n>[<size>];
- unsigned sz = TD->getTypeAllocSize(Ty);
+ unsigned sz = DL.getTypeAllocSize(Ty);
SDVTList DeclareParamVTs = DAG.getVTList(MVT::Other, MVT::Glue);
- SDValue DeclareParamOps[] = { Chain, DAG.getConstant(align, MVT::i32),
- DAG.getConstant(paramCount, MVT::i32),
- DAG.getConstant(sz, MVT::i32), InFlag };
+ SDValue DeclareParamOps[] = { Chain,
+ DAG.getConstant(align, dl, MVT::i32),
+ DAG.getConstant(paramCount, dl, MVT::i32),
+ DAG.getConstant(sz, dl, MVT::i32),
+ InFlag };
Chain = DAG.getNode(NVPTXISD::DeclareParam, dl, DeclareParamVTs,
DeclareParamOps);
InFlag = Chain.getValue(1);
SDVTList CopyParamVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue CopyParamOps[] = { Chain,
- DAG.getConstant(paramCount, MVT::i32),
- DAG.getConstant(0, MVT::i32), Elt,
+ DAG.getConstant(paramCount, dl, MVT::i32),
+ DAG.getConstant(0, dl, MVT::i32), Elt,
InFlag };
Chain = DAG.getMemIntrinsicNode(NVPTXISD::StoreParam, dl,
CopyParamVTs, CopyParamOps,
SDVTList CopyParamVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue CopyParamOps[] = { Chain,
- DAG.getConstant(paramCount, MVT::i32),
- DAG.getConstant(0, MVT::i32), Elt0, Elt1,
- InFlag };
+ DAG.getConstant(paramCount, dl, MVT::i32),
+ DAG.getConstant(0, dl, MVT::i32), Elt0,
+ Elt1, InFlag };
Chain = DAG.getMemIntrinsicNode(NVPTXISD::StoreParamV2, dl,
CopyParamVTs, CopyParamOps,
MemVT, MachinePointerInfo());
SDValue StoreVal;
SmallVector<SDValue, 8> Ops;
Ops.push_back(Chain);
- Ops.push_back(DAG.getConstant(paramCount, MVT::i32));
- Ops.push_back(DAG.getConstant(curOffset, MVT::i32));
+ Ops.push_back(DAG.getConstant(paramCount, dl, MVT::i32));
+ Ops.push_back(DAG.getConstant(curOffset, dl, MVT::i32));
unsigned Opc = NVPTXISD::StoreParamV2;
}
SDVTList DeclareParamVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue DeclareParamOps[] = { Chain,
- DAG.getConstant(paramCount, MVT::i32),
- DAG.getConstant(sz, MVT::i32),
- DAG.getConstant(0, MVT::i32), InFlag };
+ DAG.getConstant(paramCount, dl, MVT::i32),
+ DAG.getConstant(sz, dl, MVT::i32),
+ DAG.getConstant(0, dl, MVT::i32), InFlag };
Chain = DAG.getNode(NVPTXISD::DeclareScalarParam, dl, DeclareParamVTs,
DeclareParamOps);
InFlag = Chain.getValue(1);
OutV = DAG.getNode(opc, dl, MVT::i16, OutV);
}
SDVTList CopyParamVTs = DAG.getVTList(MVT::Other, MVT::Glue);
- SDValue CopyParamOps[] = { Chain, DAG.getConstant(paramCount, MVT::i32),
- DAG.getConstant(0, MVT::i32), OutV, InFlag };
+ SDValue CopyParamOps[] = { Chain,
+ DAG.getConstant(paramCount, dl, MVT::i32),
+ DAG.getConstant(0, dl, MVT::i32), OutV,
+ InFlag };
unsigned opcode = NVPTXISD::StoreParam;
if (Outs[OIdx].Flags.isZExt())
// struct or vector
SmallVector<EVT, 16> vtparts;
SmallVector<uint64_t, 16> Offsets;
-
const PointerType *PTy = dyn_cast<PointerType>(Args[i].Ty);
+
auto *PTy = dyn_cast<PointerType>(Args[i].Ty);
assert(PTy && "Type of a byval parameter should be pointer");
- ComputePTXValueVTs(*this, PTy->getElementType(), vtparts, &Offsets, 0);
+ ComputePTXValueVTs(*this, DAG.getDataLayout(), PTy->getElementType(),
+ vtparts, &Offsets, 0);
// declare .param .align <align> .b8 .param<n>[<size>];
unsigned sz = Outs[OIdx].Flags.getByValSize();
// so we don't need to worry about natural alignment or not.
// See TargetLowering::LowerCallTo().
SDValue DeclareParamOps[] = {
- Chain, DAG.getConstant(Outs[OIdx].Flags.getByValAlign(), MVT::i32),
- DAG.getConstant(paramCount, MVT::i32), DAG.getConstant(sz, MVT::i32),
- InFlag
+ Chain, DAG.getConstant(Outs[OIdx].Flags.getByValAlign(), dl, MVT::i32),
+ DAG.getConstant(paramCount, dl, MVT::i32),
+ DAG.getConstant(sz, dl, MVT::i32), InFlag
};
Chain = DAG.getNode(NVPTXISD::DeclareParam, dl, DeclareParamVTs,
DeclareParamOps);
for (unsigned j = 0, je = vtparts.size(); j != je; ++j) {
EVT elemtype = vtparts[j];
int curOffset = Offsets[j];
- unsigned PartAlign = GCD(ArgAlign, curOffset);
- SDValue srcAddr =
- DAG.getNode(ISD::ADD, dl, getPointerTy(), OutVals[OIdx],
- DAG.getConstant(curOffset, getPointerTy()));
+ unsigned PartAlign = GreatestCommonDivisor64(ArgAlign, curOffset);
+ auto PtrVT = getPointerTy(DAG.getDataLayout());
+ SDValue srcAddr = DAG.getNode(ISD::ADD, dl, PtrVT, OutVals[OIdx],
+ DAG.getConstant(curOffset, dl, PtrVT));
SDValue theVal = DAG.getLoad(elemtype, dl, tempChain, srcAddr,
MachinePointerInfo(), false, false, false,
PartAlign);
theVal = DAG.getNode(ISD::ANY_EXTEND, dl, MVT::i16, theVal);
}
SDVTList CopyParamVTs = DAG.getVTList(MVT::Other, MVT::Glue);
- SDValue CopyParamOps[] = { Chain, DAG.getConstant(paramCount, MVT::i32),
- DAG.getConstant(curOffset, MVT::i32), theVal,
- InFlag };
+ SDValue CopyParamOps[] = { Chain,
+ DAG.getConstant(paramCount, dl, MVT::i32),
+ DAG.getConstant(curOffset, dl, MVT::i32),
+ theVal, InFlag };
Chain = DAG.getMemIntrinsicNode(NVPTXISD::StoreParam, dl, CopyParamVTs,
CopyParamOps, elemtype,
MachinePointerInfo());
// Handle Result
if (Ins.size() > 0) {
SmallVector<EVT, 16> resvtparts;
- ComputeValueVTs(*this, retTy, resvtparts);
+ ComputeValueVTs(*this, DL, retTy, resvtparts);
// Declare
// .param .align 16 .b8 retval0[<size-in-bytes>], or
// .param .b<size-in-bits> retval0
- unsigned resultsz = TD->getTypeAllocSizeInBits(retTy);
- if (retTy->isSingleValueType()) {
+ unsigned resultsz = DL.getTypeAllocSizeInBits(retTy);
+ // Emit ".param .b<size-in-bits> retval0" instead of byte arrays only for
+ // these three types to match the logic in
+ // NVPTXAsmPrinter::printReturnValStr and NVPTXTargetLowering::getPrototype.
+ // Plus, this behavior is consistent with nvcc's.
+ if (retTy->isFloatingPointTy() || retTy->isIntegerTy() ||
+ retTy->isPointerTy()) {
// Scalar needs to be at least 32bit wide
if (resultsz < 32)
resultsz = 32;
SDVTList DeclareRetVTs = DAG.getVTList(MVT::Other, MVT::Glue);
- SDValue DeclareRetOps[] = { Chain, DAG.getConstant(1, MVT::i32),
- DAG.getConstant(resultsz, MVT::i32),
- DAG.getConstant(0, MVT::i32), InFlag };
+ SDValue DeclareRetOps[] = { Chain, DAG.getConstant(1, dl, MVT::i32),
+ DAG.getConstant(resultsz, dl, MVT::i32),
+ DAG.getConstant(0, dl, MVT::i32), InFlag };
Chain = DAG.getNode(NVPTXISD::DeclareRet, dl, DeclareRetVTs,
DeclareRetOps);
InFlag = Chain.getValue(1);
retAlignment = getArgumentAlignment(Callee, CS, retTy, 0);
SDVTList DeclareRetVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue DeclareRetOps[] = { Chain,
- DAG.getConstant(retAlignment, MVT::i32),
- DAG.getConstant(resultsz / 8, MVT::i32),
- DAG.getConstant(0, MVT::i32), InFlag };
+ DAG.getConstant(retAlignment, dl, MVT::i32),
+ DAG.getConstant(resultsz / 8, dl, MVT::i32),
+ DAG.getConstant(0, dl, MVT::i32), InFlag };
Chain = DAG.getNode(NVPTXISD::DeclareRetParam, dl, DeclareRetVTs,
DeclareRetOps);
InFlag = Chain.getValue(1);
// The prototype is embedded in a string and put as the operand for a
// CallPrototype SDNode which will print out to the value of the string.
SDVTList ProtoVTs = DAG.getVTList(MVT::Other, MVT::Glue);
- std::string Proto = getPrototype(retTy, Args, Outs, retAlignment, CS);
+ std::string Proto =
+ getPrototype(DAG.getDataLayout(), retTy, Args, Outs, retAlignment, CS);
const char *ProtoStr =
nvTM->getManagedStrPool()->getManagedString(Proto.c_str())->c_str();
SDValue ProtoOps[] = {
// Op to just print "call"
SDVTList PrintCallVTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue PrintCallOps[] = {
- Chain, DAG.getConstant((Ins.size() == 0) ? 0 : 1, MVT::i32), InFlag
+ Chain, DAG.getConstant((Ins.size() == 0) ? 0 : 1, dl, MVT::i32), InFlag
};
Chain = DAG.getNode(Func ? (NVPTXISD::PrintCallUni) : (NVPTXISD::PrintCall),
dl, PrintCallVTs, PrintCallOps);
else
opcode = NVPTXISD::CallArg;
SDVTList CallArgVTs = DAG.getVTList(MVT::Other, MVT::Glue);
- SDValue CallArgOps[] = { Chain, DAG.getConstant(1, MVT::i32),
- DAG.getConstant(i, MVT::i32), InFlag };
+ SDValue CallArgOps[] = { Chain, DAG.getConstant(1, dl, MVT::i32),
+ DAG.getConstant(i, dl, MVT::i32), InFlag };
Chain = DAG.getNode(opcode, dl, CallArgVTs, CallArgOps);
InFlag = Chain.getValue(1);
}
SDVTList CallArgEndVTs = DAG.getVTList(MVT::Other, MVT::Glue);
- SDValue CallArgEndOps[] = { Chain, DAG.getConstant(Func ? 1 : 0, MVT::i32),
+ SDValue CallArgEndOps[] = { Chain,
+ DAG.getConstant(Func ? 1 : 0, dl, MVT::i32),
InFlag };
Chain = DAG.getNode(NVPTXISD::CallArgEnd, dl, CallArgEndVTs, CallArgEndOps);
InFlag = Chain.getValue(1);
if (!Func) {
SDVTList PrototypeVTs = DAG.getVTList(MVT::Other, MVT::Glue);
- SDValue PrototypeOps[] = { Chain, DAG.getConstant(uniqueCallSite, MVT::i32),
+ SDValue PrototypeOps[] = { Chain,
+ DAG.getConstant(uniqueCallSite, dl, MVT::i32),
InFlag };
Chain = DAG.getNode(NVPTXISD::Prototype, dl, PrototypeVTs, PrototypeOps);
InFlag = Chain.getValue(1);
// Generate loads from param memory/moves from registers for result
if (Ins.size() > 0) {
if (retTy && retTy->isVectorTy()) {
- EVT ObjectVT = getValueType(retTy);
+ EVT ObjectVT = getValueType(DL, retTy);
unsigned NumElts = ObjectVT.getVectorNumElements();
EVT EltVT = ObjectVT.getVectorElementType();
- assert(nvTM->getTargetLowering()->getNumRegisters(F->getContext(),
- ObjectVT) == NumElts &&
+ assert(STI.getTargetLowering()->getNumRegisters(F->getContext(),
+ ObjectVT) == NumElts &&
"Vector was not scalarized");
unsigned sz = EltVT.getSizeInBits();
- bool needTruncate = sz < 8 ? true : false;
+ bool needTruncate = sz < 8;
if (NumElts == 1) {
// Just a simple load
LoadRetVTs.push_back(EltVT);
LoadRetVTs.push_back(MVT::Other);
LoadRetVTs.push_back(MVT::Glue);
- SmallVector<SDValue, 4> LoadRetOps;
- LoadRetOps.push_back(Chain);
- LoadRetOps.push_back(DAG.getConstant(1, MVT::i32));
- LoadRetOps.push_back(DAG.getConstant(0, MVT::i32));
- LoadRetOps.push_back(InFlag);
+ SDValue LoadRetOps[] = {Chain, DAG.getConstant(1, dl, MVT::i32),
+ DAG.getConstant(0, dl, MVT::i32), InFlag};
SDValue retval = DAG.getMemIntrinsicNode(
NVPTXISD::LoadParam, dl,
DAG.getVTList(LoadRetVTs), LoadRetOps, EltVT, MachinePointerInfo());
}
LoadRetVTs.push_back(MVT::Other);
LoadRetVTs.push_back(MVT::Glue);
- SmallVector<SDValue, 4> LoadRetOps;
- LoadRetOps.push_back(Chain);
- LoadRetOps.push_back(DAG.getConstant(1, MVT::i32));
- LoadRetOps.push_back(DAG.getConstant(0, MVT::i32));
- LoadRetOps.push_back(InFlag);
+ SDValue LoadRetOps[] = {Chain, DAG.getConstant(1, dl, MVT::i32),
+ DAG.getConstant(0, dl, MVT::i32), InFlag};
SDValue retval = DAG.getMemIntrinsicNode(
NVPTXISD::LoadParamV2, dl,
DAG.getVTList(LoadRetVTs), LoadRetOps, EltVT, MachinePointerInfo());
}
LoadRetVTs.push_back(MVT::Other);
LoadRetVTs.push_back(MVT::Glue);
- SmallVector<SDValue, 4> LoadRetOps;
- LoadRetOps.push_back(Chain);
- LoadRetOps.push_back(DAG.getConstant(1, MVT::i32));
- LoadRetOps.push_back(DAG.getConstant(Ofst, MVT::i32));
- LoadRetOps.push_back(InFlag);
+ SDValue LoadRetOps[] = {Chain, DAG.getConstant(1, dl, MVT::i32),
+ DAG.getConstant(Ofst, dl, MVT::i32), InFlag};
SDValue retval = DAG.getMemIntrinsicNode(
Opc, dl, DAG.getVTList(LoadRetVTs),
LoadRetOps, EltVT, MachinePointerInfo());
Elt = DAG.getNode(ISD::TRUNCATE, dl, EltVT, Elt);
InVals.push_back(Elt);
}
- Ofst += TD->getTypeAllocSize(VecVT.getTypeForEVT(F->getContext()));
+ Ofst += DL.getTypeAllocSize(VecVT.getTypeForEVT(F->getContext()));
}
}
} else {
SmallVector<EVT, 16> VTs;
SmallVector<uint64_t, 16> Offsets;
- ComputePTXValueVTs(*this, retTy, VTs, &Offsets, 0);
+ ComputePTXValueVTs(*this, DAG.getDataLayout(), retTy, VTs, &Offsets, 0);
assert(VTs.size() == Ins.size() && "Bad value decomposition");
unsigned RetAlign = getArgumentAlignment(Callee, CS, retTy, 0);
for (unsigned i = 0, e = Ins.size(); i != e; ++i) {
unsigned sz = VTs[i].getSizeInBits();
- unsigned AlignI = GCD(RetAlign, Offsets[i]);
- bool needTruncate = sz < 8 ? true : false;
+ unsigned AlignI = GreatestCommonDivisor64(RetAlign, Offsets[i]);
+ bool needTruncate = sz < 8;
if (VTs[i].isInteger() && (sz < 8))
sz = 8;
SmallVector<EVT, 4> LoadRetVTs;
EVT TheLoadType = VTs[i];
- if (retTy->isIntegerTy() &&
- TD->getTypeAllocSizeInBits(retTy) < 32) {
+ if (retTy->isIntegerTy() && DL.getTypeAllocSizeInBits(retTy) < 32) {
// This is for integer types only, and specifically not for
// aggregates.
LoadRetVTs.push_back(MVT::i32);
LoadRetVTs.push_back(MVT::Other);
LoadRetVTs.push_back(MVT::Glue);
- SmallVector<SDValue, 4> LoadRetOps;
- LoadRetOps.push_back(Chain);
- LoadRetOps.push_back(DAG.getConstant(1, MVT::i32));
- LoadRetOps.push_back(DAG.getConstant(Offsets[i], MVT::i32));
- LoadRetOps.push_back(InFlag);
+ SDValue LoadRetOps[] = {Chain, DAG.getConstant(1, dl, MVT::i32),
+ DAG.getConstant(Offsets[i], dl, MVT::i32),
+ InFlag};
SDValue retval = DAG.getMemIntrinsicNode(
NVPTXISD::LoadParam, dl,
DAG.getVTList(LoadRetVTs), LoadRetOps,
}
}
- Chain = DAG.getCALLSEQ_END(Chain, DAG.getIntPtrConstant(uniqueCallSite, true),
- DAG.getIntPtrConstant(uniqueCallSite + 1, true),
+ Chain = DAG.getCALLSEQ_END(Chain,
+ DAG.getIntPtrConstant(uniqueCallSite, dl, true),
+ DAG.getIntPtrConstant(uniqueCallSite + 1, dl,
+ true),
InFlag, dl);
uniqueCallSite++;
unsigned NumSubElem = VVT.getVectorNumElements();
for (unsigned j = 0; j < NumSubElem; ++j) {
Ops.push_back(DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, EltVT, SubOp,
- DAG.getIntPtrConstant(j)));
+ DAG.getIntPtrConstant(j, dl)));
}
}
return DAG.getNode(ISD::BUILD_VECTOR, dl, Node->getValueType(0), Ops);
SDValue ShAmt = Op.getOperand(2);
unsigned Opc = (Op.getOpcode() == ISD::SRA_PARTS) ? ISD::SRA : ISD::SRL;
- if (VTBits == 32 && nvptxSubtarget.getSmVersion() >= 35) {
+ if (VTBits == 32 && STI.getSmVersion() >= 35) {
// For 32bit and sm35, we can use the funnel shift 'shf' instruction.
// {dHi, dLo} = {aHi, aLo} >> Amt
// dHi = aHi >> Amt
SDValue RevShAmt = DAG.getNode(ISD::SUB, dl, MVT::i32,
- DAG.getConstant(VTBits, MVT::i32), ShAmt);
+ DAG.getConstant(VTBits, dl, MVT::i32),
+ ShAmt);
SDValue Tmp1 = DAG.getNode(ISD::SRL, dl, VT, ShOpLo, ShAmt);
SDValue ExtraShAmt = DAG.getNode(ISD::SUB, dl, MVT::i32, ShAmt,
- DAG.getConstant(VTBits, MVT::i32));
+ DAG.getConstant(VTBits, dl, MVT::i32));
SDValue Tmp2 = DAG.getNode(ISD::SHL, dl, VT, ShOpHi, RevShAmt);
SDValue FalseVal = DAG.getNode(ISD::OR, dl, VT, Tmp1, Tmp2);
SDValue TrueVal = DAG.getNode(Opc, dl, VT, ShOpHi, ExtraShAmt);
SDValue Cmp = DAG.getSetCC(dl, MVT::i1, ShAmt,
- DAG.getConstant(VTBits, MVT::i32), ISD::SETGE);
+ DAG.getConstant(VTBits, dl, MVT::i32),
+ ISD::SETGE);
SDValue Hi = DAG.getNode(Opc, dl, VT, ShOpHi, ShAmt);
SDValue Lo = DAG.getNode(ISD::SELECT, dl, VT, Cmp, TrueVal, FalseVal);
SDValue ShOpHi = Op.getOperand(1);
SDValue ShAmt = Op.getOperand(2);
- if (VTBits == 32 && nvptxSubtarget.getSmVersion() >= 35) {
+ if (VTBits == 32 && STI.getSmVersion() >= 35) {
// For 32bit and sm35, we can use the funnel shift 'shf' instruction.
// {dHi, dLo} = {aHi, aLo} << Amt
// dHi = (aHi << Amt) | (aLo >> (size-Amt))
SDValue RevShAmt = DAG.getNode(ISD::SUB, dl, MVT::i32,
- DAG.getConstant(VTBits, MVT::i32), ShAmt);
+ DAG.getConstant(VTBits, dl, MVT::i32),
+ ShAmt);
SDValue Tmp1 = DAG.getNode(ISD::SHL, dl, VT, ShOpHi, ShAmt);
SDValue ExtraShAmt = DAG.getNode(ISD::SUB, dl, MVT::i32, ShAmt,
- DAG.getConstant(VTBits, MVT::i32));
+ DAG.getConstant(VTBits, dl, MVT::i32));
SDValue Tmp2 = DAG.getNode(ISD::SRL, dl, VT, ShOpLo, RevShAmt);
SDValue FalseVal = DAG.getNode(ISD::OR, dl, VT, Tmp1, Tmp2);
SDValue TrueVal = DAG.getNode(ISD::SHL, dl, VT, ShOpLo, ExtraShAmt);
SDValue Cmp = DAG.getSetCC(dl, MVT::i1, ShAmt,
- DAG.getConstant(VTBits, MVT::i32), ISD::SETGE);
+ DAG.getConstant(VTBits, dl, MVT::i32),
+ ISD::SETGE);
SDValue Lo = DAG.getNode(ISD::SHL, dl, VT, ShOpLo, ShAmt);
SDValue Hi = DAG.getNode(ISD::SELECT, dl, VT, Cmp, TrueVal, FalseVal);
case ISD::SRA_PARTS:
case ISD::SRL_PARTS:
return LowerShiftRightParts(Op, DAG);
+ case ISD::SELECT:
+ return LowerSelect(Op, DAG);
default:
llvm_unreachable("Custom lowering not defined for operation");
}
}
+SDValue NVPTXTargetLowering::LowerSelect(SDValue Op, SelectionDAG &DAG) const {
+ SDValue Op0 = Op->getOperand(0);
+ SDValue Op1 = Op->getOperand(1);
+ SDValue Op2 = Op->getOperand(2);
+ SDLoc DL(Op.getNode());
+
+ assert(Op.getValueType() == MVT::i1 && "Custom lowering enabled only for i1");
+
+ Op1 = DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i32, Op1);
+ Op2 = DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i32, Op2);
+ SDValue Select = DAG.getNode(ISD::SELECT, DL, MVT::i32, Op0, Op1, Op2);
+ SDValue Trunc = DAG.getNode(ISD::TRUNCATE, DL, MVT::i1, Select);
+
+ return Trunc;
+}
+
SDValue NVPTXTargetLowering::LowerLOAD(SDValue Op, SelectionDAG &DAG) const {
if (Op.getValueType() == MVT::i1)
return LowerLOADi1(Op, DAG);
break;
}
+ MemSDNode *MemSD = cast<MemSDNode>(N);
+ const DataLayout &TD = DAG.getDataLayout();
+
+ unsigned Align = MemSD->getAlignment();
+ unsigned PrefAlign =
+ TD.getPrefTypeAlignment(ValVT.getTypeForEVT(*DAG.getContext()));
+ if (Align < PrefAlign) {
+ // This store is not sufficiently aligned, so bail out and let this vector
+ // store be scalarized. Note that we may still be able to emit smaller
+ // vector stores. For example, if we are storing a <4 x float> with an
+ // alignment of 8, this check will fail but the legalizer will try again
+ // with 2 x <2 x float>, which will succeed with an alignment of 8.
+ return SDValue();
+ }
+
unsigned Opcode = 0;
EVT EltVT = ValVT.getVectorElementType();
unsigned NumElts = ValVT.getVectorNumElements();
// Then the split values
for (unsigned i = 0; i < NumElts; ++i) {
SDValue ExtVal = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, EltVT, Val,
- DAG.getIntPtrConstant(i));
+ DAG.getIntPtrConstant(i, DL));
if (NeedExt)
ExtVal = DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i16, ExtVal);
Ops.push_back(ExtVal);
}
// Then any remaining arguments
- for (unsigned i = 2, e = N->getNumOperands(); i != e; ++i) {
- Ops.push_back(N->getOperand(i));
- }
-
- MemSDNode *MemSD = cast<MemSDNode>(N);
+ Ops.append(N->op_begin() + 2, N->op_end());
SDValue NewSt = DAG.getMemIntrinsicNode(
Opcode, DL, DAG.getVTList(MVT::Other), Ops,
"struct._image3d_t",
"struct._sampler_t" };
- const Type *Ty = arg->getType();
-
const PointerType *PTy = dyn_cast<PointerType>(Ty);
+ Type *Ty = arg->getType();
+
auto *PTy = dyn_cast<PointerType>(Ty);
if (!PTy)
return false;
if (!context)
return false;
- const StructType *STy = dyn_cast<StructType>(PTy->getElementType());
+ auto *STy = dyn_cast<StructType>(PTy->getElementType());
const std::string TypeName = STy && !STy->isLiteral() ? STy->getName() : "";
for (int i = 0, e = array_lengthof(specialTypes); i != e; ++i)
const SmallVectorImpl<ISD::InputArg> &Ins, SDLoc dl, SelectionDAG &DAG,
SmallVectorImpl<SDValue> &InVals) const {
MachineFunction &MF = DAG.getMachineFunction();
- const DataLayout *TD = getDataLayout();
+ const DataLayout &DL = DAG.getDataLayout();
+ auto PtrVT = getPointerTy(DAG.getDataLayout());
const Function *F = MF.getFunction();
const AttributeSet &PAL = F->getAttributes();
- const TargetLowering *TLI = DAG.getTarget().getTargetLowering();
+ const TargetLowering *TLI = STI.getTargetLowering();
SDValue Root = DAG.getRoot();
std::vector<SDValue> OutChains;
bool isKernel = llvm::isKernelFunction(*F);
- bool isABI = (nvptxSubtarget.getSmVersion() >= 20);
+ bool isABI = (STI.getSmVersion() >= 20);
assert(isABI && "Non-ABI compilation is not supported");
if (!isABI)
return Chain;
(theArgs[i]->getParent() ? theArgs[i]->getParent()->getParent()
: nullptr))) {
assert(isKernel && "Only kernels can have image/sampler params");
- InVals.push_back(DAG.getConstant(i + 1, MVT::i32));
+ InVals.push_back(DAG.getConstant(i + 1, dl, MVT::i32));
continue;
}
if (Ty->isAggregateType()) {
SmallVector<EVT, 16> vtparts;
- ComputePTXValueVTs(*this, Ty, vtparts);
+ ComputePTXValueVTs(*this, DAG.getDataLayout(), Ty, vtparts);
assert(vtparts.size() > 0 && "empty aggregate type not expected");
for (unsigned parti = 0, parte = vtparts.size(); parti != parte;
++parti) {
- EVT partVT = vtparts[parti];
- InVals.push_back(DAG.getNode(ISD::UNDEF, dl, partVT));
+ InVals.push_back(DAG.getNode(ISD::UNDEF, dl, Ins[InsIdx].VT));
++InsIdx;
}
if (vtparts.size() > 0)
continue;
}
if (Ty->isVectorTy()) {
- EVT ObjectVT = getValueType(Ty);
+ EVT ObjectVT = getValueType(DL, Ty);
unsigned NumRegs = TLI->getNumRegisters(F->getContext(), ObjectVT);
for (unsigned parti = 0; parti < NumRegs; ++parti) {
InVals.push_back(DAG.getNode(ISD::UNDEF, dl, Ins[InsIdx].VT));
// to newly created nodes. The SDNodes for params have to
// appear in the same order as their order of appearance
// in the original function. "idx+1" holds that order.
- if (PAL.hasAttribute(i + 1, Attribute::ByVal) == false) {
+ if (!PAL.hasAttribute(i + 1, Attribute::ByVal)) {
if (Ty->isAggregateType()) {
SmallVector<EVT, 16> vtparts;
SmallVector<uint64_t, 16> offsets;
// NOTE: Here, we lose the ability to issue vector loads for vectors
// that are a part of a struct. This should be investigated in the
// future.
- ComputePTXValueVTs(*this, Ty, vtparts, &offsets, 0);
+ ComputePTXValueVTs(*this, DAG.getDataLayout(), Ty, vtparts, &offsets,
+ 0);
assert(vtparts.size() > 0 && "empty aggregate type not expected");
bool aggregateIsPacked = false;
if (StructType *STy = llvm::dyn_cast<StructType>(Ty))
aggregateIsPacked = STy->isPacked();
- SDValue Arg = getParamSymbol(DAG, idx, getPointerTy());
+ SDValue Arg = getParamSymbol(DAG, idx, PtrVT);
for (unsigned parti = 0, parte = vtparts.size(); parti != parte;
++parti) {
EVT partVT = vtparts[parti];
PointerType::get(partVT.getTypeForEVT(F->getContext()),
llvm::ADDRESS_SPACE_PARAM));
SDValue srcAddr =
- DAG.getNode(ISD::ADD, dl, getPointerTy(), Arg,
- DAG.getConstant(offsets[parti], getPointerTy()));
- unsigned partAlign =
- aggregateIsPacked ? 1
- : TD->getABITypeAlignment(
- partVT.getTypeForEVT(F->getContext()));
+ DAG.getNode(ISD::ADD, dl, PtrVT, Arg,
+ DAG.getConstant(offsets[parti], dl, PtrVT));
+ unsigned partAlign = aggregateIsPacked
+ ? 1
+ : DL.getABITypeAlignment(
+ partVT.getTypeForEVT(F->getContext()));
SDValue p;
if (Ins[InsIdx].VT.getSizeInBits() > partVT.getSizeInBits()) {
ISD::LoadExtType ExtOp = Ins[InsIdx].Flags.isSExt() ?
ISD::SEXTLOAD : ISD::ZEXTLOAD;
p = DAG.getExtLoad(ExtOp, dl, Ins[InsIdx].VT, Root, srcAddr,
MachinePointerInfo(srcValue), partVT, false,
- false, partAlign);
+ false, false, partAlign);
} else {
p = DAG.getLoad(partVT, dl, Root, srcAddr,
MachinePointerInfo(srcValue), false, false, false,
continue;
}
if (Ty->isVectorTy()) {
- EVT ObjectVT = getValueType(Ty);
- SDValue Arg = getParamSymbol(DAG, idx, getPointerTy());
+ EVT ObjectVT = getValueType(DL, Ty);
+ SDValue Arg = getParamSymbol(DAG, idx, PtrVT);
unsigned NumElts = ObjectVT.getVectorNumElements();
assert(TLI->getNumRegisters(F->getContext(), ObjectVT) == NumElts &&
"Vector was not scalarized");
- unsigned Ofst = 0;
EVT EltVT = ObjectVT.getVectorElementType();
// V1 load
// We only have one element, so just directly load it
Value *SrcValue = Constant::getNullValue(PointerType::get(
EltVT.getTypeForEVT(F->getContext()), llvm::ADDRESS_SPACE_PARAM));
- SDValue SrcAddr = DAG.getNode(ISD::ADD, dl, getPointerTy(), Arg,
- DAG.getConstant(Ofst, getPointerTy()));
SDValue P = DAG.getLoad(
- EltVT, dl, Root, SrcAddr, MachinePointerInfo(SrcValue), false,
- false, true,
- TD->getABITypeAlignment(EltVT.getTypeForEVT(F->getContext())));
+ EltVT, dl, Root, Arg, MachinePointerInfo(SrcValue), false, false,
+ true,
+ DL.getABITypeAlignment(EltVT.getTypeForEVT(F->getContext())));
if (P.getNode())
P.getNode()->setIROrder(idx + 1);
if (Ins[InsIdx].VT.getSizeInBits() > EltVT.getSizeInBits())
P = DAG.getNode(ISD::ANY_EXTEND, dl, Ins[InsIdx].VT, P);
InVals.push_back(P);
- Ofst += TD->getTypeAllocSize(EltVT.getTypeForEVT(F->getContext()));
++InsIdx;
} else if (NumElts == 2) {
// V2 load
EVT VecVT = EVT::getVectorVT(F->getContext(), EltVT, 2);
Value *SrcValue = Constant::getNullValue(PointerType::get(
VecVT.getTypeForEVT(F->getContext()), llvm::ADDRESS_SPACE_PARAM));
- SDValue SrcAddr = DAG.getNode(ISD::ADD, dl, getPointerTy(), Arg,
- DAG.getConstant(Ofst, getPointerTy()));
SDValue P = DAG.getLoad(
- VecVT, dl, Root, SrcAddr, MachinePointerInfo(SrcValue), false,
- false, true,
- TD->getABITypeAlignment(VecVT.getTypeForEVT(F->getContext())));
+ VecVT, dl, Root, Arg, MachinePointerInfo(SrcValue), false, false,
+ true,
+ DL.getABITypeAlignment(VecVT.getTypeForEVT(F->getContext())));
if (P.getNode())
P.getNode()->setIROrder(idx + 1);
SDValue Elt0 = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, EltVT, P,
- DAG.getIntPtrConstant(0));
+ DAG.getIntPtrConstant(0, dl));
SDValue Elt1 = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, EltVT, P,
- DAG.getIntPtrConstant(1));
+ DAG.getIntPtrConstant(1, dl));
if (Ins[InsIdx].VT.getSizeInBits() > EltVT.getSizeInBits()) {
Elt0 = DAG.getNode(ISD::ANY_EXTEND, dl, Ins[InsIdx].VT, Elt0);
InVals.push_back(Elt0);
InVals.push_back(Elt1);
- Ofst += TD->getTypeAllocSize(VecVT.getTypeForEVT(F->getContext()));
InsIdx += 2;
} else {
// V4 loads
VecSize = 2;
}
EVT VecVT = EVT::getVectorVT(F->getContext(), EltVT, VecSize);
+ unsigned Ofst = 0;
for (unsigned i = 0; i < NumElts; i += VecSize) {
Value *SrcValue = Constant::getNullValue(
PointerType::get(VecVT.getTypeForEVT(F->getContext()),
llvm::ADDRESS_SPACE_PARAM));
- SDValue SrcAddr =
- DAG.getNode(ISD::ADD, dl, getPointerTy(), Arg,
- DAG.getConstant(Ofst, getPointerTy()));
+ SDValue SrcAddr = DAG.getNode(ISD::ADD, dl, PtrVT, Arg,
+ DAG.getConstant(Ofst, dl, PtrVT));
SDValue P = DAG.getLoad(
VecVT, dl, Root, SrcAddr, MachinePointerInfo(SrcValue), false,
false, true,
- TD->getABITypeAlignment(VecVT.getTypeForEVT(F->getContext())));
+ DL.getABITypeAlignment(VecVT.getTypeForEVT(F->getContext())));
if (P.getNode())
P.getNode()->setIROrder(idx + 1);
if (i + j >= NumElts)
break;
SDValue Elt = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, EltVT, P,
- DAG.getIntPtrConstant(j));
+ DAG.getIntPtrConstant(j, dl));
if (Ins[InsIdx].VT.getSizeInBits() > EltVT.getSizeInBits())
Elt = DAG.getNode(ISD::ANY_EXTEND, dl, Ins[InsIdx].VT, Elt);
InVals.push_back(Elt);
}
- Ofst += TD->getTypeAllocSize(VecVT.getTypeForEVT(F->getContext()));
+ Ofst += DL.getTypeAllocSize(VecVT.getTypeForEVT(F->getContext()));
}
InsIdx += NumElts;
}
continue;
}
// A plain scalar.
- EVT ObjectVT = getValueType(Ty);
+ EVT ObjectVT = getValueType(DL, Ty);
// If ABI, load from the param symbol
- SDValue Arg = getParamSymbol(DAG, idx, getPointerTy());
+ SDValue Arg = getParamSymbol(DAG, idx, PtrVT);
Value *srcValue = Constant::getNullValue(PointerType::get(
ObjectVT.getTypeForEVT(F->getContext()), llvm::ADDRESS_SPACE_PARAM));
SDValue p;
if (ObjectVT.getSizeInBits() < Ins[InsIdx].VT.getSizeInBits()) {
ISD::LoadExtType ExtOp = Ins[InsIdx].Flags.isSExt() ?
ISD::SEXTLOAD : ISD::ZEXTLOAD;
- p = DAG.getExtLoad(ExtOp, dl, Ins[InsIdx].VT, Root, Arg,
- MachinePointerInfo(srcValue), ObjectVT, false, false,
- TD->getABITypeAlignment(ObjectVT.getTypeForEVT(F->getContext())));
+ p = DAG.getExtLoad(
+ ExtOp, dl, Ins[InsIdx].VT, Root, Arg, MachinePointerInfo(srcValue),
+ ObjectVT, false, false, false,
+ DL.getABITypeAlignment(ObjectVT.getTypeForEVT(F->getContext())));
} else {
- p = DAG.getLoad(Ins[InsIdx].VT, dl, Root, Arg,
- MachinePointerInfo(srcValue), false, false, false,
- TD->getABITypeAlignment(ObjectVT.getTypeForEVT(F->getContext())));
+ p = DAG.getLoad(
+ Ins[InsIdx].VT, dl, Root, Arg, MachinePointerInfo(srcValue), false,
+ false, false,
+ DL.getABITypeAlignment(ObjectVT.getTypeForEVT(F->getContext())));
}
if (p.getNode())
p.getNode()->setIROrder(idx + 1);
// machine instruction fails because TargetExternalSymbol
// (not lowered) is target dependent, and CopyToReg assumes
// the source is lowered.
- EVT ObjectVT = getValueType(Ty);
+ EVT ObjectVT = getValueType(DL, Ty);
assert(ObjectVT == Ins[InsIdx].VT &&
"Ins type did not match function type");
- SDValue Arg = getParamSymbol(DAG, idx, getPointerTy());
+ SDValue Arg = getParamSymbol(DAG, idx, PtrVT);
SDValue p = DAG.getNode(NVPTXISD::MoveParam, dl, ObjectVT, Arg);
if (p.getNode())
p.getNode()->setIROrder(idx + 1);
else {
SDValue p2 = DAG.getNode(
ISD::INTRINSIC_WO_CHAIN, dl, ObjectVT,
- DAG.getConstant(Intrinsic::nvvm_ptr_local_to_gen, MVT::i32), p);
+ DAG.getConstant(Intrinsic::nvvm_ptr_local_to_gen, dl, MVT::i32), p);
InVals.push_back(p2);
}
}
MachineFunction &MF = DAG.getMachineFunction();
const Function *F = MF.getFunction();
Type *RetTy = F->getReturnType();
- const DataLayout *TD = getDataLayout();
+ const DataLayout &TD = DAG.getDataLayout();
- bool isABI = (nvptxSubtarget.getSmVersion() >= 20);
+ bool isABI = (STI.getSmVersion() >= 20);
assert(isABI && "Non-ABI compilation is not supported");
if (!isABI)
return Chain;
assert(NumElts == Outs.size() && "Bad scalarization of return value");
// const_cast can be removed in later LLVM versions
- EVT EltVT = getValueType(RetTy).getVectorElementType();
+ EVT EltVT = getValueType(TD, RetTy).getVectorElementType();
bool NeedExtend = false;
if (EltVT.getSizeInBits() < 16)
NeedExtend = true;
// We only have one element, so just directly store it
if (NeedExtend)
StoreVal = DAG.getNode(ISD::ZERO_EXTEND, dl, MVT::i16, StoreVal);
- SDValue Ops[] = { Chain, DAG.getConstant(0, MVT::i32), StoreVal };
+ SDValue Ops[] = { Chain, DAG.getConstant(0, dl, MVT::i32), StoreVal };
Chain = DAG.getMemIntrinsicNode(NVPTXISD::StoreRetval, dl,
DAG.getVTList(MVT::Other), Ops,
EltVT, MachinePointerInfo());
StoreVal1 = DAG.getNode(ISD::ZERO_EXTEND, dl, MVT::i16, StoreVal1);
}
- SDValue Ops[] = { Chain, DAG.getConstant(0, MVT::i32), StoreVal0,
+ SDValue Ops[] = { Chain, DAG.getConstant(0, dl, MVT::i32), StoreVal0,
StoreVal1 };
Chain = DAG.getMemIntrinsicNode(NVPTXISD::StoreRetvalV2, dl,
DAG.getVTList(MVT::Other), Ops,
unsigned Offset = 0;
EVT VecVT =
- EVT::getVectorVT(F->getContext(), OutVals[0].getValueType(), VecSize);
+ EVT::getVectorVT(F->getContext(), EltVT, VecSize);
unsigned PerStoreOffset =
- TD->getTypeAllocSize(VecVT.getTypeForEVT(F->getContext()));
+ TD.getTypeAllocSize(VecVT.getTypeForEVT(F->getContext()));
for (unsigned i = 0; i < NumElts; i += VecSize) {
// Get values
SDValue StoreVal;
SmallVector<SDValue, 8> Ops;
Ops.push_back(Chain);
- Ops.push_back(DAG.getConstant(Offset, MVT::i32));
+ Ops.push_back(DAG.getConstant(Offset, dl, MVT::i32));
unsigned Opc = NVPTXISD::StoreRetvalV2;
EVT ExtendedVT = (NeedExtend) ? MVT::i16 : OutVals[0].getValueType();
}
} else {
SmallVector<EVT, 16> ValVTs;
- // const_cast is necessary since we are still using an LLVM version from
- // before the type system re-write.
- ComputePTXValueVTs(*this, RetTy, ValVTs);
+ SmallVector<uint64_t, 16> Offsets;
+ ComputePTXValueVTs(*this, DAG.getDataLayout(), RetTy, ValVTs, &Offsets, 0);
assert(ValVTs.size() == OutVals.size() && "Bad return value decomposition");
- unsigned SizeSoFar = 0;
for (unsigned i = 0, e = Outs.size(); i != e; ++i) {
SDValue theVal = OutVals[i];
EVT TheValType = theVal.getValueType();
if (TheValType.isVector())
TmpVal = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl,
TheValType.getVectorElementType(), TmpVal,
- DAG.getIntPtrConstant(j));
+ DAG.getIntPtrConstant(j, dl));
EVT TheStoreType = ValVTs[i];
- if (RetTy->isIntegerTy() &&
- TD->getTypeAllocSizeInBits(RetTy) < 32) {
+ if (RetTy->isIntegerTy() && TD.getTypeAllocSizeInBits(RetTy) < 32) {
// The following zero-extension is for integer types only, and
// specifically not for aggregates.
TmpVal = DAG.getNode(ISD::ZERO_EXTEND, dl, MVT::i32, TmpVal);
else if (TmpVal.getValueType().getSizeInBits() < 16)
TmpVal = DAG.getNode(ISD::ANY_EXTEND, dl, MVT::i16, TmpVal);
- SDValue Ops[] = { Chain, DAG.getConstant(SizeSoFar, MVT::i32), TmpVal };
+ SDValue Ops[] = {
+ Chain,
+ DAG.getConstant(Offsets[i], dl, MVT::i32),
+ TmpVal };
Chain = DAG.getMemIntrinsicNode(NVPTXISD::StoreRetval, dl,
DAG.getVTList(MVT::Other), Ops,
TheStoreType,
MachinePointerInfo());
- if(TheValType.isVector())
- SizeSoFar +=
- TheStoreType.getVectorElementType().getStoreSizeInBits() / 8;
- else
- SizeSoFar += TheStoreType.getStoreSizeInBits()/8;
}
}
}
default:
return 0;
- case Intrinsic::nvvm_tex_1d_v4f32_i32:
- return NVPTXISD::Tex1DFloatI32;
+ case Intrinsic::nvvm_tex_1d_v4f32_s32:
+ return NVPTXISD::Tex1DFloatS32;
case Intrinsic::nvvm_tex_1d_v4f32_f32:
return NVPTXISD::Tex1DFloatFloat;
case Intrinsic::nvvm_tex_1d_level_v4f32_f32:
return NVPTXISD::Tex1DFloatFloatLevel;
case Intrinsic::nvvm_tex_1d_grad_v4f32_f32:
return NVPTXISD::Tex1DFloatFloatGrad;
- case Intrinsic::nvvm_tex_1d_v4i32_i32:
- return NVPTXISD::Tex1DI32I32;
- case Intrinsic::nvvm_tex_1d_v4i32_f32:
- return NVPTXISD::Tex1DI32Float;
- case Intrinsic::nvvm_tex_1d_level_v4i32_f32:
- return NVPTXISD::Tex1DI32FloatLevel;
- case Intrinsic::nvvm_tex_1d_grad_v4i32_f32:
- return NVPTXISD::Tex1DI32FloatGrad;
-
- case Intrinsic::nvvm_tex_1d_array_v4f32_i32:
- return NVPTXISD::Tex1DArrayFloatI32;
+ case Intrinsic::nvvm_tex_1d_v4s32_s32:
+ return NVPTXISD::Tex1DS32S32;
+ case Intrinsic::nvvm_tex_1d_v4s32_f32:
+ return NVPTXISD::Tex1DS32Float;
+ case Intrinsic::nvvm_tex_1d_level_v4s32_f32:
+ return NVPTXISD::Tex1DS32FloatLevel;
+ case Intrinsic::nvvm_tex_1d_grad_v4s32_f32:
+ return NVPTXISD::Tex1DS32FloatGrad;
+ case Intrinsic::nvvm_tex_1d_v4u32_s32:
+ return NVPTXISD::Tex1DU32S32;
+ case Intrinsic::nvvm_tex_1d_v4u32_f32:
+ return NVPTXISD::Tex1DU32Float;
+ case Intrinsic::nvvm_tex_1d_level_v4u32_f32:
+ return NVPTXISD::Tex1DU32FloatLevel;
+ case Intrinsic::nvvm_tex_1d_grad_v4u32_f32:
+ return NVPTXISD::Tex1DU32FloatGrad;
+
+ case Intrinsic::nvvm_tex_1d_array_v4f32_s32:
+ return NVPTXISD::Tex1DArrayFloatS32;
case Intrinsic::nvvm_tex_1d_array_v4f32_f32:
return NVPTXISD::Tex1DArrayFloatFloat;
case Intrinsic::nvvm_tex_1d_array_level_v4f32_f32:
return NVPTXISD::Tex1DArrayFloatFloatLevel;
case Intrinsic::nvvm_tex_1d_array_grad_v4f32_f32:
return NVPTXISD::Tex1DArrayFloatFloatGrad;
- case Intrinsic::nvvm_tex_1d_array_v4i32_i32:
- return NVPTXISD::Tex1DArrayI32I32;
- case Intrinsic::nvvm_tex_1d_array_v4i32_f32:
- return NVPTXISD::Tex1DArrayI32Float;
- case Intrinsic::nvvm_tex_1d_array_level_v4i32_f32:
- return NVPTXISD::Tex1DArrayI32FloatLevel;
- case Intrinsic::nvvm_tex_1d_array_grad_v4i32_f32:
- return NVPTXISD::Tex1DArrayI32FloatGrad;
-
- case Intrinsic::nvvm_tex_2d_v4f32_i32:
- return NVPTXISD::Tex2DFloatI32;
+ case Intrinsic::nvvm_tex_1d_array_v4s32_s32:
+ return NVPTXISD::Tex1DArrayS32S32;
+ case Intrinsic::nvvm_tex_1d_array_v4s32_f32:
+ return NVPTXISD::Tex1DArrayS32Float;
+ case Intrinsic::nvvm_tex_1d_array_level_v4s32_f32:
+ return NVPTXISD::Tex1DArrayS32FloatLevel;
+ case Intrinsic::nvvm_tex_1d_array_grad_v4s32_f32:
+ return NVPTXISD::Tex1DArrayS32FloatGrad;
+ case Intrinsic::nvvm_tex_1d_array_v4u32_s32:
+ return NVPTXISD::Tex1DArrayU32S32;
+ case Intrinsic::nvvm_tex_1d_array_v4u32_f32:
+ return NVPTXISD::Tex1DArrayU32Float;
+ case Intrinsic::nvvm_tex_1d_array_level_v4u32_f32:
+ return NVPTXISD::Tex1DArrayU32FloatLevel;
+ case Intrinsic::nvvm_tex_1d_array_grad_v4u32_f32:
+ return NVPTXISD::Tex1DArrayU32FloatGrad;
+
+ case Intrinsic::nvvm_tex_2d_v4f32_s32:
+ return NVPTXISD::Tex2DFloatS32;
case Intrinsic::nvvm_tex_2d_v4f32_f32:
return NVPTXISD::Tex2DFloatFloat;
case Intrinsic::nvvm_tex_2d_level_v4f32_f32:
return NVPTXISD::Tex2DFloatFloatLevel;
case Intrinsic::nvvm_tex_2d_grad_v4f32_f32:
return NVPTXISD::Tex2DFloatFloatGrad;
- case Intrinsic::nvvm_tex_2d_v4i32_i32:
- return NVPTXISD::Tex2DI32I32;
- case Intrinsic::nvvm_tex_2d_v4i32_f32:
- return NVPTXISD::Tex2DI32Float;
- case Intrinsic::nvvm_tex_2d_level_v4i32_f32:
- return NVPTXISD::Tex2DI32FloatLevel;
- case Intrinsic::nvvm_tex_2d_grad_v4i32_f32:
- return NVPTXISD::Tex2DI32FloatGrad;
-
- case Intrinsic::nvvm_tex_2d_array_v4f32_i32:
- return NVPTXISD::Tex2DArrayFloatI32;
+ case Intrinsic::nvvm_tex_2d_v4s32_s32:
+ return NVPTXISD::Tex2DS32S32;
+ case Intrinsic::nvvm_tex_2d_v4s32_f32:
+ return NVPTXISD::Tex2DS32Float;
+ case Intrinsic::nvvm_tex_2d_level_v4s32_f32:
+ return NVPTXISD::Tex2DS32FloatLevel;
+ case Intrinsic::nvvm_tex_2d_grad_v4s32_f32:
+ return NVPTXISD::Tex2DS32FloatGrad;
+ case Intrinsic::nvvm_tex_2d_v4u32_s32:
+ return NVPTXISD::Tex2DU32S32;
+ case Intrinsic::nvvm_tex_2d_v4u32_f32:
+ return NVPTXISD::Tex2DU32Float;
+ case Intrinsic::nvvm_tex_2d_level_v4u32_f32:
+ return NVPTXISD::Tex2DU32FloatLevel;
+ case Intrinsic::nvvm_tex_2d_grad_v4u32_f32:
+ return NVPTXISD::Tex2DU32FloatGrad;
+
+ case Intrinsic::nvvm_tex_2d_array_v4f32_s32:
+ return NVPTXISD::Tex2DArrayFloatS32;
case Intrinsic::nvvm_tex_2d_array_v4f32_f32:
return NVPTXISD::Tex2DArrayFloatFloat;
case Intrinsic::nvvm_tex_2d_array_level_v4f32_f32:
return NVPTXISD::Tex2DArrayFloatFloatLevel;
case Intrinsic::nvvm_tex_2d_array_grad_v4f32_f32:
return NVPTXISD::Tex2DArrayFloatFloatGrad;
- case Intrinsic::nvvm_tex_2d_array_v4i32_i32:
- return NVPTXISD::Tex2DArrayI32I32;
- case Intrinsic::nvvm_tex_2d_array_v4i32_f32:
- return NVPTXISD::Tex2DArrayI32Float;
- case Intrinsic::nvvm_tex_2d_array_level_v4i32_f32:
- return NVPTXISD::Tex2DArrayI32FloatLevel;
- case Intrinsic::nvvm_tex_2d_array_grad_v4i32_f32:
- return NVPTXISD::Tex2DArrayI32FloatGrad;
-
- case Intrinsic::nvvm_tex_3d_v4f32_i32:
- return NVPTXISD::Tex3DFloatI32;
+ case Intrinsic::nvvm_tex_2d_array_v4s32_s32:
+ return NVPTXISD::Tex2DArrayS32S32;
+ case Intrinsic::nvvm_tex_2d_array_v4s32_f32:
+ return NVPTXISD::Tex2DArrayS32Float;
+ case Intrinsic::nvvm_tex_2d_array_level_v4s32_f32:
+ return NVPTXISD::Tex2DArrayS32FloatLevel;
+ case Intrinsic::nvvm_tex_2d_array_grad_v4s32_f32:
+ return NVPTXISD::Tex2DArrayS32FloatGrad;
+ case Intrinsic::nvvm_tex_2d_array_v4u32_s32:
+ return NVPTXISD::Tex2DArrayU32S32;
+ case Intrinsic::nvvm_tex_2d_array_v4u32_f32:
+ return NVPTXISD::Tex2DArrayU32Float;
+ case Intrinsic::nvvm_tex_2d_array_level_v4u32_f32:
+ return NVPTXISD::Tex2DArrayU32FloatLevel;
+ case Intrinsic::nvvm_tex_2d_array_grad_v4u32_f32:
+ return NVPTXISD::Tex2DArrayU32FloatGrad;
+
+ case Intrinsic::nvvm_tex_3d_v4f32_s32:
+ return NVPTXISD::Tex3DFloatS32;
case Intrinsic::nvvm_tex_3d_v4f32_f32:
return NVPTXISD::Tex3DFloatFloat;
case Intrinsic::nvvm_tex_3d_level_v4f32_f32:
return NVPTXISD::Tex3DFloatFloatLevel;
case Intrinsic::nvvm_tex_3d_grad_v4f32_f32:
return NVPTXISD::Tex3DFloatFloatGrad;
- case Intrinsic::nvvm_tex_3d_v4i32_i32:
- return NVPTXISD::Tex3DI32I32;
- case Intrinsic::nvvm_tex_3d_v4i32_f32:
- return NVPTXISD::Tex3DI32Float;
- case Intrinsic::nvvm_tex_3d_level_v4i32_f32:
- return NVPTXISD::Tex3DI32FloatLevel;
- case Intrinsic::nvvm_tex_3d_grad_v4i32_f32:
- return NVPTXISD::Tex3DI32FloatGrad;
+ case Intrinsic::nvvm_tex_3d_v4s32_s32:
+ return NVPTXISD::Tex3DS32S32;
+ case Intrinsic::nvvm_tex_3d_v4s32_f32:
+ return NVPTXISD::Tex3DS32Float;
+ case Intrinsic::nvvm_tex_3d_level_v4s32_f32:
+ return NVPTXISD::Tex3DS32FloatLevel;
+ case Intrinsic::nvvm_tex_3d_grad_v4s32_f32:
+ return NVPTXISD::Tex3DS32FloatGrad;
+ case Intrinsic::nvvm_tex_3d_v4u32_s32:
+ return NVPTXISD::Tex3DU32S32;
+ case Intrinsic::nvvm_tex_3d_v4u32_f32:
+ return NVPTXISD::Tex3DU32Float;
+ case Intrinsic::nvvm_tex_3d_level_v4u32_f32:
+ return NVPTXISD::Tex3DU32FloatLevel;
+ case Intrinsic::nvvm_tex_3d_grad_v4u32_f32:
+ return NVPTXISD::Tex3DU32FloatGrad;
+
+ case Intrinsic::nvvm_tex_cube_v4f32_f32:
+ return NVPTXISD::TexCubeFloatFloat;
+ case Intrinsic::nvvm_tex_cube_level_v4f32_f32:
+ return NVPTXISD::TexCubeFloatFloatLevel;
+ case Intrinsic::nvvm_tex_cube_v4s32_f32:
+ return NVPTXISD::TexCubeS32Float;
+ case Intrinsic::nvvm_tex_cube_level_v4s32_f32:
+ return NVPTXISD::TexCubeS32FloatLevel;
+ case Intrinsic::nvvm_tex_cube_v4u32_f32:
+ return NVPTXISD::TexCubeU32Float;
+ case Intrinsic::nvvm_tex_cube_level_v4u32_f32:
+ return NVPTXISD::TexCubeU32FloatLevel;
+
+ case Intrinsic::nvvm_tex_cube_array_v4f32_f32:
+ return NVPTXISD::TexCubeArrayFloatFloat;
+ case Intrinsic::nvvm_tex_cube_array_level_v4f32_f32:
+ return NVPTXISD::TexCubeArrayFloatFloatLevel;
+ case Intrinsic::nvvm_tex_cube_array_v4s32_f32:
+ return NVPTXISD::TexCubeArrayS32Float;
+ case Intrinsic::nvvm_tex_cube_array_level_v4s32_f32:
+ return NVPTXISD::TexCubeArrayS32FloatLevel;
+ case Intrinsic::nvvm_tex_cube_array_v4u32_f32:
+ return NVPTXISD::TexCubeArrayU32Float;
+ case Intrinsic::nvvm_tex_cube_array_level_v4u32_f32:
+ return NVPTXISD::TexCubeArrayU32FloatLevel;
+
+ case Intrinsic::nvvm_tld4_r_2d_v4f32_f32:
+ return NVPTXISD::Tld4R2DFloatFloat;
+ case Intrinsic::nvvm_tld4_g_2d_v4f32_f32:
+ return NVPTXISD::Tld4G2DFloatFloat;
+ case Intrinsic::nvvm_tld4_b_2d_v4f32_f32:
+ return NVPTXISD::Tld4B2DFloatFloat;
+ case Intrinsic::nvvm_tld4_a_2d_v4f32_f32:
+ return NVPTXISD::Tld4A2DFloatFloat;
+ case Intrinsic::nvvm_tld4_r_2d_v4s32_f32:
+ return NVPTXISD::Tld4R2DS64Float;
+ case Intrinsic::nvvm_tld4_g_2d_v4s32_f32:
+ return NVPTXISD::Tld4G2DS64Float;
+ case Intrinsic::nvvm_tld4_b_2d_v4s32_f32:
+ return NVPTXISD::Tld4B2DS64Float;
+ case Intrinsic::nvvm_tld4_a_2d_v4s32_f32:
+ return NVPTXISD::Tld4A2DS64Float;
+ case Intrinsic::nvvm_tld4_r_2d_v4u32_f32:
+ return NVPTXISD::Tld4R2DU64Float;
+ case Intrinsic::nvvm_tld4_g_2d_v4u32_f32:
+ return NVPTXISD::Tld4G2DU64Float;
+ case Intrinsic::nvvm_tld4_b_2d_v4u32_f32:
+ return NVPTXISD::Tld4B2DU64Float;
+ case Intrinsic::nvvm_tld4_a_2d_v4u32_f32:
+ return NVPTXISD::Tld4A2DU64Float;
+
+ case Intrinsic::nvvm_tex_unified_1d_v4f32_s32:
+ return NVPTXISD::TexUnified1DFloatS32;
+ case Intrinsic::nvvm_tex_unified_1d_v4f32_f32:
+ return NVPTXISD::TexUnified1DFloatFloat;
+ case Intrinsic::nvvm_tex_unified_1d_level_v4f32_f32:
+ return NVPTXISD::TexUnified1DFloatFloatLevel;
+ case Intrinsic::nvvm_tex_unified_1d_grad_v4f32_f32:
+ return NVPTXISD::TexUnified1DFloatFloatGrad;
+ case Intrinsic::nvvm_tex_unified_1d_v4s32_s32:
+ return NVPTXISD::TexUnified1DS32S32;
+ case Intrinsic::nvvm_tex_unified_1d_v4s32_f32:
+ return NVPTXISD::TexUnified1DS32Float;
+ case Intrinsic::nvvm_tex_unified_1d_level_v4s32_f32:
+ return NVPTXISD::TexUnified1DS32FloatLevel;
+ case Intrinsic::nvvm_tex_unified_1d_grad_v4s32_f32:
+ return NVPTXISD::TexUnified1DS32FloatGrad;
+ case Intrinsic::nvvm_tex_unified_1d_v4u32_s32:
+ return NVPTXISD::TexUnified1DU32S32;
+ case Intrinsic::nvvm_tex_unified_1d_v4u32_f32:
+ return NVPTXISD::TexUnified1DU32Float;
+ case Intrinsic::nvvm_tex_unified_1d_level_v4u32_f32:
+ return NVPTXISD::TexUnified1DU32FloatLevel;
+ case Intrinsic::nvvm_tex_unified_1d_grad_v4u32_f32:
+ return NVPTXISD::TexUnified1DU32FloatGrad;
+
+ case Intrinsic::nvvm_tex_unified_1d_array_v4f32_s32:
+ return NVPTXISD::TexUnified1DArrayFloatS32;
+ case Intrinsic::nvvm_tex_unified_1d_array_v4f32_f32:
+ return NVPTXISD::TexUnified1DArrayFloatFloat;
+ case Intrinsic::nvvm_tex_unified_1d_array_level_v4f32_f32:
+ return NVPTXISD::TexUnified1DArrayFloatFloatLevel;
+ case Intrinsic::nvvm_tex_unified_1d_array_grad_v4f32_f32:
+ return NVPTXISD::TexUnified1DArrayFloatFloatGrad;
+ case Intrinsic::nvvm_tex_unified_1d_array_v4s32_s32:
+ return NVPTXISD::TexUnified1DArrayS32S32;
+ case Intrinsic::nvvm_tex_unified_1d_array_v4s32_f32:
+ return NVPTXISD::TexUnified1DArrayS32Float;
+ case Intrinsic::nvvm_tex_unified_1d_array_level_v4s32_f32:
+ return NVPTXISD::TexUnified1DArrayS32FloatLevel;
+ case Intrinsic::nvvm_tex_unified_1d_array_grad_v4s32_f32:
+ return NVPTXISD::TexUnified1DArrayS32FloatGrad;
+ case Intrinsic::nvvm_tex_unified_1d_array_v4u32_s32:
+ return NVPTXISD::TexUnified1DArrayU32S32;
+ case Intrinsic::nvvm_tex_unified_1d_array_v4u32_f32:
+ return NVPTXISD::TexUnified1DArrayU32Float;
+ case Intrinsic::nvvm_tex_unified_1d_array_level_v4u32_f32:
+ return NVPTXISD::TexUnified1DArrayU32FloatLevel;
+ case Intrinsic::nvvm_tex_unified_1d_array_grad_v4u32_f32:
+ return NVPTXISD::TexUnified1DArrayU32FloatGrad;
+
+ case Intrinsic::nvvm_tex_unified_2d_v4f32_s32:
+ return NVPTXISD::TexUnified2DFloatS32;
+ case Intrinsic::nvvm_tex_unified_2d_v4f32_f32:
+ return NVPTXISD::TexUnified2DFloatFloat;
+ case Intrinsic::nvvm_tex_unified_2d_level_v4f32_f32:
+ return NVPTXISD::TexUnified2DFloatFloatLevel;
+ case Intrinsic::nvvm_tex_unified_2d_grad_v4f32_f32:
+ return NVPTXISD::TexUnified2DFloatFloatGrad;
+ case Intrinsic::nvvm_tex_unified_2d_v4s32_s32:
+ return NVPTXISD::TexUnified2DS32S32;
+ case Intrinsic::nvvm_tex_unified_2d_v4s32_f32:
+ return NVPTXISD::TexUnified2DS32Float;
+ case Intrinsic::nvvm_tex_unified_2d_level_v4s32_f32:
+ return NVPTXISD::TexUnified2DS32FloatLevel;
+ case Intrinsic::nvvm_tex_unified_2d_grad_v4s32_f32:
+ return NVPTXISD::TexUnified2DS32FloatGrad;
+ case Intrinsic::nvvm_tex_unified_2d_v4u32_s32:
+ return NVPTXISD::TexUnified2DU32S32;
+ case Intrinsic::nvvm_tex_unified_2d_v4u32_f32:
+ return NVPTXISD::TexUnified2DU32Float;
+ case Intrinsic::nvvm_tex_unified_2d_level_v4u32_f32:
+ return NVPTXISD::TexUnified2DU32FloatLevel;
+ case Intrinsic::nvvm_tex_unified_2d_grad_v4u32_f32:
+ return NVPTXISD::TexUnified2DU32FloatGrad;
+
+ case Intrinsic::nvvm_tex_unified_2d_array_v4f32_s32:
+ return NVPTXISD::TexUnified2DArrayFloatS32;
+ case Intrinsic::nvvm_tex_unified_2d_array_v4f32_f32:
+ return NVPTXISD::TexUnified2DArrayFloatFloat;
+ case Intrinsic::nvvm_tex_unified_2d_array_level_v4f32_f32:
+ return NVPTXISD::TexUnified2DArrayFloatFloatLevel;
+ case Intrinsic::nvvm_tex_unified_2d_array_grad_v4f32_f32:
+ return NVPTXISD::TexUnified2DArrayFloatFloatGrad;
+ case Intrinsic::nvvm_tex_unified_2d_array_v4s32_s32:
+ return NVPTXISD::TexUnified2DArrayS32S32;
+ case Intrinsic::nvvm_tex_unified_2d_array_v4s32_f32:
+ return NVPTXISD::TexUnified2DArrayS32Float;
+ case Intrinsic::nvvm_tex_unified_2d_array_level_v4s32_f32:
+ return NVPTXISD::TexUnified2DArrayS32FloatLevel;
+ case Intrinsic::nvvm_tex_unified_2d_array_grad_v4s32_f32:
+ return NVPTXISD::TexUnified2DArrayS32FloatGrad;
+ case Intrinsic::nvvm_tex_unified_2d_array_v4u32_s32:
+ return NVPTXISD::TexUnified2DArrayU32S32;
+ case Intrinsic::nvvm_tex_unified_2d_array_v4u32_f32:
+ return NVPTXISD::TexUnified2DArrayU32Float;
+ case Intrinsic::nvvm_tex_unified_2d_array_level_v4u32_f32:
+ return NVPTXISD::TexUnified2DArrayU32FloatLevel;
+ case Intrinsic::nvvm_tex_unified_2d_array_grad_v4u32_f32:
+ return NVPTXISD::TexUnified2DArrayU32FloatGrad;
+
+ case Intrinsic::nvvm_tex_unified_3d_v4f32_s32:
+ return NVPTXISD::TexUnified3DFloatS32;
+ case Intrinsic::nvvm_tex_unified_3d_v4f32_f32:
+ return NVPTXISD::TexUnified3DFloatFloat;
+ case Intrinsic::nvvm_tex_unified_3d_level_v4f32_f32:
+ return NVPTXISD::TexUnified3DFloatFloatLevel;
+ case Intrinsic::nvvm_tex_unified_3d_grad_v4f32_f32:
+ return NVPTXISD::TexUnified3DFloatFloatGrad;
+ case Intrinsic::nvvm_tex_unified_3d_v4s32_s32:
+ return NVPTXISD::TexUnified3DS32S32;
+ case Intrinsic::nvvm_tex_unified_3d_v4s32_f32:
+ return NVPTXISD::TexUnified3DS32Float;
+ case Intrinsic::nvvm_tex_unified_3d_level_v4s32_f32:
+ return NVPTXISD::TexUnified3DS32FloatLevel;
+ case Intrinsic::nvvm_tex_unified_3d_grad_v4s32_f32:
+ return NVPTXISD::TexUnified3DS32FloatGrad;
+ case Intrinsic::nvvm_tex_unified_3d_v4u32_s32:
+ return NVPTXISD::TexUnified3DU32S32;
+ case Intrinsic::nvvm_tex_unified_3d_v4u32_f32:
+ return NVPTXISD::TexUnified3DU32Float;
+ case Intrinsic::nvvm_tex_unified_3d_level_v4u32_f32:
+ return NVPTXISD::TexUnified3DU32FloatLevel;
+ case Intrinsic::nvvm_tex_unified_3d_grad_v4u32_f32:
+ return NVPTXISD::TexUnified3DU32FloatGrad;
+
+ case Intrinsic::nvvm_tex_unified_cube_v4f32_f32:
+ return NVPTXISD::TexUnifiedCubeFloatFloat;
+ case Intrinsic::nvvm_tex_unified_cube_level_v4f32_f32:
+ return NVPTXISD::TexUnifiedCubeFloatFloatLevel;
+ case Intrinsic::nvvm_tex_unified_cube_v4s32_f32:
+ return NVPTXISD::TexUnifiedCubeS32Float;
+ case Intrinsic::nvvm_tex_unified_cube_level_v4s32_f32:
+ return NVPTXISD::TexUnifiedCubeS32FloatLevel;
+ case Intrinsic::nvvm_tex_unified_cube_v4u32_f32:
+ return NVPTXISD::TexUnifiedCubeU32Float;
+ case Intrinsic::nvvm_tex_unified_cube_level_v4u32_f32:
+ return NVPTXISD::TexUnifiedCubeU32FloatLevel;
+
+ case Intrinsic::nvvm_tex_unified_cube_array_v4f32_f32:
+ return NVPTXISD::TexUnifiedCubeArrayFloatFloat;
+ case Intrinsic::nvvm_tex_unified_cube_array_level_v4f32_f32:
+ return NVPTXISD::TexUnifiedCubeArrayFloatFloatLevel;
+ case Intrinsic::nvvm_tex_unified_cube_array_v4s32_f32:
+ return NVPTXISD::TexUnifiedCubeArrayS32Float;
+ case Intrinsic::nvvm_tex_unified_cube_array_level_v4s32_f32:
+ return NVPTXISD::TexUnifiedCubeArrayS32FloatLevel;
+ case Intrinsic::nvvm_tex_unified_cube_array_v4u32_f32:
+ return NVPTXISD::TexUnifiedCubeArrayU32Float;
+ case Intrinsic::nvvm_tex_unified_cube_array_level_v4u32_f32:
+ return NVPTXISD::TexUnifiedCubeArrayU32FloatLevel;
+
+ case Intrinsic::nvvm_tld4_unified_r_2d_v4f32_f32:
+ return NVPTXISD::Tld4UnifiedR2DFloatFloat;
+ case Intrinsic::nvvm_tld4_unified_g_2d_v4f32_f32:
+ return NVPTXISD::Tld4UnifiedG2DFloatFloat;
+ case Intrinsic::nvvm_tld4_unified_b_2d_v4f32_f32:
+ return NVPTXISD::Tld4UnifiedB2DFloatFloat;
+ case Intrinsic::nvvm_tld4_unified_a_2d_v4f32_f32:
+ return NVPTXISD::Tld4UnifiedA2DFloatFloat;
+ case Intrinsic::nvvm_tld4_unified_r_2d_v4s32_f32:
+ return NVPTXISD::Tld4UnifiedR2DS64Float;
+ case Intrinsic::nvvm_tld4_unified_g_2d_v4s32_f32:
+ return NVPTXISD::Tld4UnifiedG2DS64Float;
+ case Intrinsic::nvvm_tld4_unified_b_2d_v4s32_f32:
+ return NVPTXISD::Tld4UnifiedB2DS64Float;
+ case Intrinsic::nvvm_tld4_unified_a_2d_v4s32_f32:
+ return NVPTXISD::Tld4UnifiedA2DS64Float;
+ case Intrinsic::nvvm_tld4_unified_r_2d_v4u32_f32:
+ return NVPTXISD::Tld4UnifiedR2DU64Float;
+ case Intrinsic::nvvm_tld4_unified_g_2d_v4u32_f32:
+ return NVPTXISD::Tld4UnifiedG2DU64Float;
+ case Intrinsic::nvvm_tld4_unified_b_2d_v4u32_f32:
+ return NVPTXISD::Tld4UnifiedB2DU64Float;
+ case Intrinsic::nvvm_tld4_unified_a_2d_v4u32_f32:
+ return NVPTXISD::Tld4UnifiedA2DU64Float;
}
}
switch (Intrinsic) {
default:
return 0;
+ case Intrinsic::nvvm_suld_1d_i8_clamp:
+ return NVPTXISD::Suld1DI8Clamp;
+ case Intrinsic::nvvm_suld_1d_i16_clamp:
+ return NVPTXISD::Suld1DI16Clamp;
+ case Intrinsic::nvvm_suld_1d_i32_clamp:
+ return NVPTXISD::Suld1DI32Clamp;
+ case Intrinsic::nvvm_suld_1d_i64_clamp:
+ return NVPTXISD::Suld1DI64Clamp;
+ case Intrinsic::nvvm_suld_1d_v2i8_clamp:
+ return NVPTXISD::Suld1DV2I8Clamp;
+ case Intrinsic::nvvm_suld_1d_v2i16_clamp:
+ return NVPTXISD::Suld1DV2I16Clamp;
+ case Intrinsic::nvvm_suld_1d_v2i32_clamp:
+ return NVPTXISD::Suld1DV2I32Clamp;
+ case Intrinsic::nvvm_suld_1d_v2i64_clamp:
+ return NVPTXISD::Suld1DV2I64Clamp;
+ case Intrinsic::nvvm_suld_1d_v4i8_clamp:
+ return NVPTXISD::Suld1DV4I8Clamp;
+ case Intrinsic::nvvm_suld_1d_v4i16_clamp:
+ return NVPTXISD::Suld1DV4I16Clamp;
+ case Intrinsic::nvvm_suld_1d_v4i32_clamp:
+ return NVPTXISD::Suld1DV4I32Clamp;
+ case Intrinsic::nvvm_suld_1d_array_i8_clamp:
+ return NVPTXISD::Suld1DArrayI8Clamp;
+ case Intrinsic::nvvm_suld_1d_array_i16_clamp:
+ return NVPTXISD::Suld1DArrayI16Clamp;
+ case Intrinsic::nvvm_suld_1d_array_i32_clamp:
+ return NVPTXISD::Suld1DArrayI32Clamp;
+ case Intrinsic::nvvm_suld_1d_array_i64_clamp:
+ return NVPTXISD::Suld1DArrayI64Clamp;
+ case Intrinsic::nvvm_suld_1d_array_v2i8_clamp:
+ return NVPTXISD::Suld1DArrayV2I8Clamp;
+ case Intrinsic::nvvm_suld_1d_array_v2i16_clamp:
+ return NVPTXISD::Suld1DArrayV2I16Clamp;
+ case Intrinsic::nvvm_suld_1d_array_v2i32_clamp:
+ return NVPTXISD::Suld1DArrayV2I32Clamp;
+ case Intrinsic::nvvm_suld_1d_array_v2i64_clamp:
+ return NVPTXISD::Suld1DArrayV2I64Clamp;
+ case Intrinsic::nvvm_suld_1d_array_v4i8_clamp:
+ return NVPTXISD::Suld1DArrayV4I8Clamp;
+ case Intrinsic::nvvm_suld_1d_array_v4i16_clamp:
+ return NVPTXISD::Suld1DArrayV4I16Clamp;
+ case Intrinsic::nvvm_suld_1d_array_v4i32_clamp:
+ return NVPTXISD::Suld1DArrayV4I32Clamp;
+ case Intrinsic::nvvm_suld_2d_i8_clamp:
+ return NVPTXISD::Suld2DI8Clamp;
+ case Intrinsic::nvvm_suld_2d_i16_clamp:
+ return NVPTXISD::Suld2DI16Clamp;
+ case Intrinsic::nvvm_suld_2d_i32_clamp:
+ return NVPTXISD::Suld2DI32Clamp;
+ case Intrinsic::nvvm_suld_2d_i64_clamp:
+ return NVPTXISD::Suld2DI64Clamp;
+ case Intrinsic::nvvm_suld_2d_v2i8_clamp:
+ return NVPTXISD::Suld2DV2I8Clamp;
+ case Intrinsic::nvvm_suld_2d_v2i16_clamp:
+ return NVPTXISD::Suld2DV2I16Clamp;
+ case Intrinsic::nvvm_suld_2d_v2i32_clamp:
+ return NVPTXISD::Suld2DV2I32Clamp;
+ case Intrinsic::nvvm_suld_2d_v2i64_clamp:
+ return NVPTXISD::Suld2DV2I64Clamp;
+ case Intrinsic::nvvm_suld_2d_v4i8_clamp:
+ return NVPTXISD::Suld2DV4I8Clamp;
+ case Intrinsic::nvvm_suld_2d_v4i16_clamp:
+ return NVPTXISD::Suld2DV4I16Clamp;
+ case Intrinsic::nvvm_suld_2d_v4i32_clamp:
+ return NVPTXISD::Suld2DV4I32Clamp;
+ case Intrinsic::nvvm_suld_2d_array_i8_clamp:
+ return NVPTXISD::Suld2DArrayI8Clamp;
+ case Intrinsic::nvvm_suld_2d_array_i16_clamp:
+ return NVPTXISD::Suld2DArrayI16Clamp;
+ case Intrinsic::nvvm_suld_2d_array_i32_clamp:
+ return NVPTXISD::Suld2DArrayI32Clamp;
+ case Intrinsic::nvvm_suld_2d_array_i64_clamp:
+ return NVPTXISD::Suld2DArrayI64Clamp;
+ case Intrinsic::nvvm_suld_2d_array_v2i8_clamp:
+ return NVPTXISD::Suld2DArrayV2I8Clamp;
+ case Intrinsic::nvvm_suld_2d_array_v2i16_clamp:
+ return NVPTXISD::Suld2DArrayV2I16Clamp;
+ case Intrinsic::nvvm_suld_2d_array_v2i32_clamp:
+ return NVPTXISD::Suld2DArrayV2I32Clamp;
+ case Intrinsic::nvvm_suld_2d_array_v2i64_clamp:
+ return NVPTXISD::Suld2DArrayV2I64Clamp;
+ case Intrinsic::nvvm_suld_2d_array_v4i8_clamp:
+ return NVPTXISD::Suld2DArrayV4I8Clamp;
+ case Intrinsic::nvvm_suld_2d_array_v4i16_clamp:
+ return NVPTXISD::Suld2DArrayV4I16Clamp;
+ case Intrinsic::nvvm_suld_2d_array_v4i32_clamp:
+ return NVPTXISD::Suld2DArrayV4I32Clamp;
+ case Intrinsic::nvvm_suld_3d_i8_clamp:
+ return NVPTXISD::Suld3DI8Clamp;
+ case Intrinsic::nvvm_suld_3d_i16_clamp:
+ return NVPTXISD::Suld3DI16Clamp;
+ case Intrinsic::nvvm_suld_3d_i32_clamp:
+ return NVPTXISD::Suld3DI32Clamp;
+ case Intrinsic::nvvm_suld_3d_i64_clamp:
+ return NVPTXISD::Suld3DI64Clamp;
+ case Intrinsic::nvvm_suld_3d_v2i8_clamp:
+ return NVPTXISD::Suld3DV2I8Clamp;
+ case Intrinsic::nvvm_suld_3d_v2i16_clamp:
+ return NVPTXISD::Suld3DV2I16Clamp;
+ case Intrinsic::nvvm_suld_3d_v2i32_clamp:
+ return NVPTXISD::Suld3DV2I32Clamp;
+ case Intrinsic::nvvm_suld_3d_v2i64_clamp:
+ return NVPTXISD::Suld3DV2I64Clamp;
+ case Intrinsic::nvvm_suld_3d_v4i8_clamp:
+ return NVPTXISD::Suld3DV4I8Clamp;
+ case Intrinsic::nvvm_suld_3d_v4i16_clamp:
+ return NVPTXISD::Suld3DV4I16Clamp;
+ case Intrinsic::nvvm_suld_3d_v4i32_clamp:
+ return NVPTXISD::Suld3DV4I32Clamp;
case Intrinsic::nvvm_suld_1d_i8_trap:
return NVPTXISD::Suld1DI8Trap;
case Intrinsic::nvvm_suld_1d_i16_trap:
return NVPTXISD::Suld1DI16Trap;
case Intrinsic::nvvm_suld_1d_i32_trap:
return NVPTXISD::Suld1DI32Trap;
+ case Intrinsic::nvvm_suld_1d_i64_trap:
+ return NVPTXISD::Suld1DI64Trap;
case Intrinsic::nvvm_suld_1d_v2i8_trap:
return NVPTXISD::Suld1DV2I8Trap;
case Intrinsic::nvvm_suld_1d_v2i16_trap:
return NVPTXISD::Suld1DV2I16Trap;
case Intrinsic::nvvm_suld_1d_v2i32_trap:
return NVPTXISD::Suld1DV2I32Trap;
+ case Intrinsic::nvvm_suld_1d_v2i64_trap:
+ return NVPTXISD::Suld1DV2I64Trap;
case Intrinsic::nvvm_suld_1d_v4i8_trap:
return NVPTXISD::Suld1DV4I8Trap;
case Intrinsic::nvvm_suld_1d_v4i16_trap:
return NVPTXISD::Suld1DArrayI16Trap;
case Intrinsic::nvvm_suld_1d_array_i32_trap:
return NVPTXISD::Suld1DArrayI32Trap;
+ case Intrinsic::nvvm_suld_1d_array_i64_trap:
+ return NVPTXISD::Suld1DArrayI64Trap;
case Intrinsic::nvvm_suld_1d_array_v2i8_trap:
return NVPTXISD::Suld1DArrayV2I8Trap;
case Intrinsic::nvvm_suld_1d_array_v2i16_trap:
return NVPTXISD::Suld1DArrayV2I16Trap;
case Intrinsic::nvvm_suld_1d_array_v2i32_trap:
return NVPTXISD::Suld1DArrayV2I32Trap;
+ case Intrinsic::nvvm_suld_1d_array_v2i64_trap:
+ return NVPTXISD::Suld1DArrayV2I64Trap;
case Intrinsic::nvvm_suld_1d_array_v4i8_trap:
return NVPTXISD::Suld1DArrayV4I8Trap;
case Intrinsic::nvvm_suld_1d_array_v4i16_trap:
return NVPTXISD::Suld2DI16Trap;
case Intrinsic::nvvm_suld_2d_i32_trap:
return NVPTXISD::Suld2DI32Trap;
+ case Intrinsic::nvvm_suld_2d_i64_trap:
+ return NVPTXISD::Suld2DI64Trap;
case Intrinsic::nvvm_suld_2d_v2i8_trap:
return NVPTXISD::Suld2DV2I8Trap;
case Intrinsic::nvvm_suld_2d_v2i16_trap:
return NVPTXISD::Suld2DV2I16Trap;
case Intrinsic::nvvm_suld_2d_v2i32_trap:
return NVPTXISD::Suld2DV2I32Trap;
+ case Intrinsic::nvvm_suld_2d_v2i64_trap:
+ return NVPTXISD::Suld2DV2I64Trap;
case Intrinsic::nvvm_suld_2d_v4i8_trap:
return NVPTXISD::Suld2DV4I8Trap;
case Intrinsic::nvvm_suld_2d_v4i16_trap:
return NVPTXISD::Suld2DArrayI16Trap;
case Intrinsic::nvvm_suld_2d_array_i32_trap:
return NVPTXISD::Suld2DArrayI32Trap;
+ case Intrinsic::nvvm_suld_2d_array_i64_trap:
+ return NVPTXISD::Suld2DArrayI64Trap;
case Intrinsic::nvvm_suld_2d_array_v2i8_trap:
return NVPTXISD::Suld2DArrayV2I8Trap;
case Intrinsic::nvvm_suld_2d_array_v2i16_trap:
return NVPTXISD::Suld2DArrayV2I16Trap;
case Intrinsic::nvvm_suld_2d_array_v2i32_trap:
return NVPTXISD::Suld2DArrayV2I32Trap;
+ case Intrinsic::nvvm_suld_2d_array_v2i64_trap:
+ return NVPTXISD::Suld2DArrayV2I64Trap;
case Intrinsic::nvvm_suld_2d_array_v4i8_trap:
return NVPTXISD::Suld2DArrayV4I8Trap;
case Intrinsic::nvvm_suld_2d_array_v4i16_trap:
return NVPTXISD::Suld3DI16Trap;
case Intrinsic::nvvm_suld_3d_i32_trap:
return NVPTXISD::Suld3DI32Trap;
+ case Intrinsic::nvvm_suld_3d_i64_trap:
+ return NVPTXISD::Suld3DI64Trap;
case Intrinsic::nvvm_suld_3d_v2i8_trap:
return NVPTXISD::Suld3DV2I8Trap;
case Intrinsic::nvvm_suld_3d_v2i16_trap:
return NVPTXISD::Suld3DV2I16Trap;
case Intrinsic::nvvm_suld_3d_v2i32_trap:
return NVPTXISD::Suld3DV2I32Trap;
+ case Intrinsic::nvvm_suld_3d_v2i64_trap:
+ return NVPTXISD::Suld3DV2I64Trap;
case Intrinsic::nvvm_suld_3d_v4i8_trap:
return NVPTXISD::Suld3DV4I8Trap;
case Intrinsic::nvvm_suld_3d_v4i16_trap:
return NVPTXISD::Suld3DV4I16Trap;
case Intrinsic::nvvm_suld_3d_v4i32_trap:
return NVPTXISD::Suld3DV4I32Trap;
+ case Intrinsic::nvvm_suld_1d_i8_zero:
+ return NVPTXISD::Suld1DI8Zero;
+ case Intrinsic::nvvm_suld_1d_i16_zero:
+ return NVPTXISD::Suld1DI16Zero;
+ case Intrinsic::nvvm_suld_1d_i32_zero:
+ return NVPTXISD::Suld1DI32Zero;
+ case Intrinsic::nvvm_suld_1d_i64_zero:
+ return NVPTXISD::Suld1DI64Zero;
+ case Intrinsic::nvvm_suld_1d_v2i8_zero:
+ return NVPTXISD::Suld1DV2I8Zero;
+ case Intrinsic::nvvm_suld_1d_v2i16_zero:
+ return NVPTXISD::Suld1DV2I16Zero;
+ case Intrinsic::nvvm_suld_1d_v2i32_zero:
+ return NVPTXISD::Suld1DV2I32Zero;
+ case Intrinsic::nvvm_suld_1d_v2i64_zero:
+ return NVPTXISD::Suld1DV2I64Zero;
+ case Intrinsic::nvvm_suld_1d_v4i8_zero:
+ return NVPTXISD::Suld1DV4I8Zero;
+ case Intrinsic::nvvm_suld_1d_v4i16_zero:
+ return NVPTXISD::Suld1DV4I16Zero;
+ case Intrinsic::nvvm_suld_1d_v4i32_zero:
+ return NVPTXISD::Suld1DV4I32Zero;
+ case Intrinsic::nvvm_suld_1d_array_i8_zero:
+ return NVPTXISD::Suld1DArrayI8Zero;
+ case Intrinsic::nvvm_suld_1d_array_i16_zero:
+ return NVPTXISD::Suld1DArrayI16Zero;
+ case Intrinsic::nvvm_suld_1d_array_i32_zero:
+ return NVPTXISD::Suld1DArrayI32Zero;
+ case Intrinsic::nvvm_suld_1d_array_i64_zero:
+ return NVPTXISD::Suld1DArrayI64Zero;
+ case Intrinsic::nvvm_suld_1d_array_v2i8_zero:
+ return NVPTXISD::Suld1DArrayV2I8Zero;
+ case Intrinsic::nvvm_suld_1d_array_v2i16_zero:
+ return NVPTXISD::Suld1DArrayV2I16Zero;
+ case Intrinsic::nvvm_suld_1d_array_v2i32_zero:
+ return NVPTXISD::Suld1DArrayV2I32Zero;
+ case Intrinsic::nvvm_suld_1d_array_v2i64_zero:
+ return NVPTXISD::Suld1DArrayV2I64Zero;
+ case Intrinsic::nvvm_suld_1d_array_v4i8_zero:
+ return NVPTXISD::Suld1DArrayV4I8Zero;
+ case Intrinsic::nvvm_suld_1d_array_v4i16_zero:
+ return NVPTXISD::Suld1DArrayV4I16Zero;
+ case Intrinsic::nvvm_suld_1d_array_v4i32_zero:
+ return NVPTXISD::Suld1DArrayV4I32Zero;
+ case Intrinsic::nvvm_suld_2d_i8_zero:
+ return NVPTXISD::Suld2DI8Zero;
+ case Intrinsic::nvvm_suld_2d_i16_zero:
+ return NVPTXISD::Suld2DI16Zero;
+ case Intrinsic::nvvm_suld_2d_i32_zero:
+ return NVPTXISD::Suld2DI32Zero;
+ case Intrinsic::nvvm_suld_2d_i64_zero:
+ return NVPTXISD::Suld2DI64Zero;
+ case Intrinsic::nvvm_suld_2d_v2i8_zero:
+ return NVPTXISD::Suld2DV2I8Zero;
+ case Intrinsic::nvvm_suld_2d_v2i16_zero:
+ return NVPTXISD::Suld2DV2I16Zero;
+ case Intrinsic::nvvm_suld_2d_v2i32_zero:
+ return NVPTXISD::Suld2DV2I32Zero;
+ case Intrinsic::nvvm_suld_2d_v2i64_zero:
+ return NVPTXISD::Suld2DV2I64Zero;
+ case Intrinsic::nvvm_suld_2d_v4i8_zero:
+ return NVPTXISD::Suld2DV4I8Zero;
+ case Intrinsic::nvvm_suld_2d_v4i16_zero:
+ return NVPTXISD::Suld2DV4I16Zero;
+ case Intrinsic::nvvm_suld_2d_v4i32_zero:
+ return NVPTXISD::Suld2DV4I32Zero;
+ case Intrinsic::nvvm_suld_2d_array_i8_zero:
+ return NVPTXISD::Suld2DArrayI8Zero;
+ case Intrinsic::nvvm_suld_2d_array_i16_zero:
+ return NVPTXISD::Suld2DArrayI16Zero;
+ case Intrinsic::nvvm_suld_2d_array_i32_zero:
+ return NVPTXISD::Suld2DArrayI32Zero;
+ case Intrinsic::nvvm_suld_2d_array_i64_zero:
+ return NVPTXISD::Suld2DArrayI64Zero;
+ case Intrinsic::nvvm_suld_2d_array_v2i8_zero:
+ return NVPTXISD::Suld2DArrayV2I8Zero;
+ case Intrinsic::nvvm_suld_2d_array_v2i16_zero:
+ return NVPTXISD::Suld2DArrayV2I16Zero;
+ case Intrinsic::nvvm_suld_2d_array_v2i32_zero:
+ return NVPTXISD::Suld2DArrayV2I32Zero;
+ case Intrinsic::nvvm_suld_2d_array_v2i64_zero:
+ return NVPTXISD::Suld2DArrayV2I64Zero;
+ case Intrinsic::nvvm_suld_2d_array_v4i8_zero:
+ return NVPTXISD::Suld2DArrayV4I8Zero;
+ case Intrinsic::nvvm_suld_2d_array_v4i16_zero:
+ return NVPTXISD::Suld2DArrayV4I16Zero;
+ case Intrinsic::nvvm_suld_2d_array_v4i32_zero:
+ return NVPTXISD::Suld2DArrayV4I32Zero;
+ case Intrinsic::nvvm_suld_3d_i8_zero:
+ return NVPTXISD::Suld3DI8Zero;
+ case Intrinsic::nvvm_suld_3d_i16_zero:
+ return NVPTXISD::Suld3DI16Zero;
+ case Intrinsic::nvvm_suld_3d_i32_zero:
+ return NVPTXISD::Suld3DI32Zero;
+ case Intrinsic::nvvm_suld_3d_i64_zero:
+ return NVPTXISD::Suld3DI64Zero;
+ case Intrinsic::nvvm_suld_3d_v2i8_zero:
+ return NVPTXISD::Suld3DV2I8Zero;
+ case Intrinsic::nvvm_suld_3d_v2i16_zero:
+ return NVPTXISD::Suld3DV2I16Zero;
+ case Intrinsic::nvvm_suld_3d_v2i32_zero:
+ return NVPTXISD::Suld3DV2I32Zero;
+ case Intrinsic::nvvm_suld_3d_v2i64_zero:
+ return NVPTXISD::Suld3DV2I64Zero;
+ case Intrinsic::nvvm_suld_3d_v4i8_zero:
+ return NVPTXISD::Suld3DV4I8Zero;
+ case Intrinsic::nvvm_suld_3d_v4i16_zero:
+ return NVPTXISD::Suld3DV4I16Zero;
+ case Intrinsic::nvvm_suld_3d_v4i32_zero:
+ return NVPTXISD::Suld3DV4I32Zero;
}
}
case Intrinsic::nvvm_ldu_global_i:
case Intrinsic::nvvm_ldu_global_f:
- case Intrinsic::nvvm_ldu_global_p:
-
+ case Intrinsic::nvvm_ldu_global_p: {
+ auto &DL = I.getModule()->getDataLayout();
Info.opc = ISD::INTRINSIC_W_CHAIN;
if (Intrinsic == Intrinsic::nvvm_ldu_global_i)
- Info.memVT = getValueType(I.getType());
- else if (Intrinsic == Intrinsic::nvvm_ldu_global_p)
- Info.memVT = getValueType(I.getType());
+ Info.memVT = getValueType(DL, I.getType());
+ else if(Intrinsic == Intrinsic::nvvm_ldu_global_p)
+ Info.memVT = getPointerTy(DL);
else
- Info.memVT = MVT::f32;
+ Info.memVT = getValueType(DL, I.getType());
Info.ptrVal = I.getArgOperand(0);
Info.offset = 0;
Info.vol = 0;
Info.readMem = true;
Info.writeMem = false;
- Info.align = 0;
+ Info.align = cast<ConstantInt>(I.getArgOperand(1))->getZExtValue();
+
return true;
+ }
+ case Intrinsic::nvvm_ldg_global_i:
+ case Intrinsic::nvvm_ldg_global_f:
+ case Intrinsic::nvvm_ldg_global_p: {
+ auto &DL = I.getModule()->getDataLayout();
- case Intrinsic::nvvm_tex_1d_v4f32_i32:
+ Info.opc = ISD::INTRINSIC_W_CHAIN;
+ if (Intrinsic == Intrinsic::nvvm_ldg_global_i)
+ Info.memVT = getValueType(DL, I.getType());
+ else if(Intrinsic == Intrinsic::nvvm_ldg_global_p)
+ Info.memVT = getPointerTy(DL);
+ else
+ Info.memVT = getValueType(DL, I.getType());
+ Info.ptrVal = I.getArgOperand(0);
+ Info.offset = 0;
+ Info.vol = 0;
+ Info.readMem = true;
+ Info.writeMem = false;
+ Info.align = cast<ConstantInt>(I.getArgOperand(1))->getZExtValue();
+
+ return true;
+ }
+
+ case Intrinsic::nvvm_tex_1d_v4f32_s32:
case Intrinsic::nvvm_tex_1d_v4f32_f32:
case Intrinsic::nvvm_tex_1d_level_v4f32_f32:
case Intrinsic::nvvm_tex_1d_grad_v4f32_f32:
- case Intrinsic::nvvm_tex_1d_array_v4f32_i32:
+ case Intrinsic::nvvm_tex_1d_array_v4f32_s32:
case Intrinsic::nvvm_tex_1d_array_v4f32_f32:
case Intrinsic::nvvm_tex_1d_array_level_v4f32_f32:
case Intrinsic::nvvm_tex_1d_array_grad_v4f32_f32:
- case Intrinsic::nvvm_tex_2d_v4f32_i32:
+ case Intrinsic::nvvm_tex_2d_v4f32_s32:
case Intrinsic::nvvm_tex_2d_v4f32_f32:
case Intrinsic::nvvm_tex_2d_level_v4f32_f32:
case Intrinsic::nvvm_tex_2d_grad_v4f32_f32:
- case Intrinsic::nvvm_tex_2d_array_v4f32_i32:
+ case Intrinsic::nvvm_tex_2d_array_v4f32_s32:
case Intrinsic::nvvm_tex_2d_array_v4f32_f32:
case Intrinsic::nvvm_tex_2d_array_level_v4f32_f32:
case Intrinsic::nvvm_tex_2d_array_grad_v4f32_f32:
- case Intrinsic::nvvm_tex_3d_v4f32_i32:
+ case Intrinsic::nvvm_tex_3d_v4f32_s32:
case Intrinsic::nvvm_tex_3d_v4f32_f32:
case Intrinsic::nvvm_tex_3d_level_v4f32_f32:
- case Intrinsic::nvvm_tex_3d_grad_v4f32_f32: {
+ case Intrinsic::nvvm_tex_3d_grad_v4f32_f32:
+ case Intrinsic::nvvm_tex_cube_v4f32_f32:
+ case Intrinsic::nvvm_tex_cube_level_v4f32_f32:
+ case Intrinsic::nvvm_tex_cube_array_v4f32_f32:
+ case Intrinsic::nvvm_tex_cube_array_level_v4f32_f32:
+ case Intrinsic::nvvm_tld4_r_2d_v4f32_f32:
+ case Intrinsic::nvvm_tld4_g_2d_v4f32_f32:
+ case Intrinsic::nvvm_tld4_b_2d_v4f32_f32:
+ case Intrinsic::nvvm_tld4_a_2d_v4f32_f32:
+ case Intrinsic::nvvm_tex_unified_1d_v4f32_s32:
+ case Intrinsic::nvvm_tex_unified_1d_v4f32_f32:
+ case Intrinsic::nvvm_tex_unified_1d_level_v4f32_f32:
+ case Intrinsic::nvvm_tex_unified_1d_grad_v4f32_f32:
+ case Intrinsic::nvvm_tex_unified_1d_array_v4f32_s32:
+ case Intrinsic::nvvm_tex_unified_1d_array_v4f32_f32:
+ case Intrinsic::nvvm_tex_unified_1d_array_level_v4f32_f32:
+ case Intrinsic::nvvm_tex_unified_1d_array_grad_v4f32_f32:
+ case Intrinsic::nvvm_tex_unified_2d_v4f32_s32:
+ case Intrinsic::nvvm_tex_unified_2d_v4f32_f32:
+ case Intrinsic::nvvm_tex_unified_2d_level_v4f32_f32:
+ case Intrinsic::nvvm_tex_unified_2d_grad_v4f32_f32:
+ case Intrinsic::nvvm_tex_unified_2d_array_v4f32_s32:
+ case Intrinsic::nvvm_tex_unified_2d_array_v4f32_f32:
+ case Intrinsic::nvvm_tex_unified_2d_array_level_v4f32_f32:
+ case Intrinsic::nvvm_tex_unified_2d_array_grad_v4f32_f32:
+ case Intrinsic::nvvm_tex_unified_3d_v4f32_s32:
+ case Intrinsic::nvvm_tex_unified_3d_v4f32_f32:
+ case Intrinsic::nvvm_tex_unified_3d_level_v4f32_f32:
+ case Intrinsic::nvvm_tex_unified_3d_grad_v4f32_f32:
+ case Intrinsic::nvvm_tex_unified_cube_v4f32_f32:
+ case Intrinsic::nvvm_tex_unified_cube_level_v4f32_f32:
+ case Intrinsic::nvvm_tex_unified_cube_array_v4f32_f32:
+ case Intrinsic::nvvm_tex_unified_cube_array_level_v4f32_f32:
+ case Intrinsic::nvvm_tld4_unified_r_2d_v4f32_f32:
+ case Intrinsic::nvvm_tld4_unified_g_2d_v4f32_f32:
+ case Intrinsic::nvvm_tld4_unified_b_2d_v4f32_f32:
+ case Intrinsic::nvvm_tld4_unified_a_2d_v4f32_f32: {
Info.opc = getOpcForTextureInstr(Intrinsic);
- Info.memVT = MVT::f32;
+ Info.memVT = MVT::v4f32;
Info.ptrVal = nullptr;
Info.offset = 0;
Info.vol = 0;
Info.align = 16;
return true;
}
- case Intrinsic::nvvm_tex_1d_v4i32_i32:
- case Intrinsic::nvvm_tex_1d_v4i32_f32:
- case Intrinsic::nvvm_tex_1d_level_v4i32_f32:
- case Intrinsic::nvvm_tex_1d_grad_v4i32_f32:
- case Intrinsic::nvvm_tex_1d_array_v4i32_i32:
- case Intrinsic::nvvm_tex_1d_array_v4i32_f32:
- case Intrinsic::nvvm_tex_1d_array_level_v4i32_f32:
- case Intrinsic::nvvm_tex_1d_array_grad_v4i32_f32:
- case Intrinsic::nvvm_tex_2d_v4i32_i32:
- case Intrinsic::nvvm_tex_2d_v4i32_f32:
- case Intrinsic::nvvm_tex_2d_level_v4i32_f32:
- case Intrinsic::nvvm_tex_2d_grad_v4i32_f32:
- case Intrinsic::nvvm_tex_2d_array_v4i32_i32:
- case Intrinsic::nvvm_tex_2d_array_v4i32_f32:
- case Intrinsic::nvvm_tex_2d_array_level_v4i32_f32:
- case Intrinsic::nvvm_tex_2d_array_grad_v4i32_f32:
- case Intrinsic::nvvm_tex_3d_v4i32_i32:
- case Intrinsic::nvvm_tex_3d_v4i32_f32:
- case Intrinsic::nvvm_tex_3d_level_v4i32_f32:
- case Intrinsic::nvvm_tex_3d_grad_v4i32_f32: {
+ case Intrinsic::nvvm_tex_1d_v4s32_s32:
+ case Intrinsic::nvvm_tex_1d_v4s32_f32:
+ case Intrinsic::nvvm_tex_1d_level_v4s32_f32:
+ case Intrinsic::nvvm_tex_1d_grad_v4s32_f32:
+ case Intrinsic::nvvm_tex_1d_array_v4s32_s32:
+ case Intrinsic::nvvm_tex_1d_array_v4s32_f32:
+ case Intrinsic::nvvm_tex_1d_array_level_v4s32_f32:
+ case Intrinsic::nvvm_tex_1d_array_grad_v4s32_f32:
+ case Intrinsic::nvvm_tex_2d_v4s32_s32:
+ case Intrinsic::nvvm_tex_2d_v4s32_f32:
+ case Intrinsic::nvvm_tex_2d_level_v4s32_f32:
+ case Intrinsic::nvvm_tex_2d_grad_v4s32_f32:
+ case Intrinsic::nvvm_tex_2d_array_v4s32_s32:
+ case Intrinsic::nvvm_tex_2d_array_v4s32_f32:
+ case Intrinsic::nvvm_tex_2d_array_level_v4s32_f32:
+ case Intrinsic::nvvm_tex_2d_array_grad_v4s32_f32:
+ case Intrinsic::nvvm_tex_3d_v4s32_s32:
+ case Intrinsic::nvvm_tex_3d_v4s32_f32:
+ case Intrinsic::nvvm_tex_3d_level_v4s32_f32:
+ case Intrinsic::nvvm_tex_3d_grad_v4s32_f32:
+ case Intrinsic::nvvm_tex_cube_v4s32_f32:
+ case Intrinsic::nvvm_tex_cube_level_v4s32_f32:
+ case Intrinsic::nvvm_tex_cube_array_v4s32_f32:
+ case Intrinsic::nvvm_tex_cube_array_level_v4s32_f32:
+ case Intrinsic::nvvm_tex_cube_v4u32_f32:
+ case Intrinsic::nvvm_tex_cube_level_v4u32_f32:
+ case Intrinsic::nvvm_tex_cube_array_v4u32_f32:
+ case Intrinsic::nvvm_tex_cube_array_level_v4u32_f32:
+ case Intrinsic::nvvm_tex_1d_v4u32_s32:
+ case Intrinsic::nvvm_tex_1d_v4u32_f32:
+ case Intrinsic::nvvm_tex_1d_level_v4u32_f32:
+ case Intrinsic::nvvm_tex_1d_grad_v4u32_f32:
+ case Intrinsic::nvvm_tex_1d_array_v4u32_s32:
+ case Intrinsic::nvvm_tex_1d_array_v4u32_f32:
+ case Intrinsic::nvvm_tex_1d_array_level_v4u32_f32:
+ case Intrinsic::nvvm_tex_1d_array_grad_v4u32_f32:
+ case Intrinsic::nvvm_tex_2d_v4u32_s32:
+ case Intrinsic::nvvm_tex_2d_v4u32_f32:
+ case Intrinsic::nvvm_tex_2d_level_v4u32_f32:
+ case Intrinsic::nvvm_tex_2d_grad_v4u32_f32:
+ case Intrinsic::nvvm_tex_2d_array_v4u32_s32:
+ case Intrinsic::nvvm_tex_2d_array_v4u32_f32:
+ case Intrinsic::nvvm_tex_2d_array_level_v4u32_f32:
+ case Intrinsic::nvvm_tex_2d_array_grad_v4u32_f32:
+ case Intrinsic::nvvm_tex_3d_v4u32_s32:
+ case Intrinsic::nvvm_tex_3d_v4u32_f32:
+ case Intrinsic::nvvm_tex_3d_level_v4u32_f32:
+ case Intrinsic::nvvm_tex_3d_grad_v4u32_f32:
+ case Intrinsic::nvvm_tld4_r_2d_v4s32_f32:
+ case Intrinsic::nvvm_tld4_g_2d_v4s32_f32:
+ case Intrinsic::nvvm_tld4_b_2d_v4s32_f32:
+ case Intrinsic::nvvm_tld4_a_2d_v4s32_f32:
+ case Intrinsic::nvvm_tld4_r_2d_v4u32_f32:
+ case Intrinsic::nvvm_tld4_g_2d_v4u32_f32:
+ case Intrinsic::nvvm_tld4_b_2d_v4u32_f32:
+ case Intrinsic::nvvm_tld4_a_2d_v4u32_f32:
+ case Intrinsic::nvvm_tex_unified_1d_v4s32_s32:
+ case Intrinsic::nvvm_tex_unified_1d_v4s32_f32:
+ case Intrinsic::nvvm_tex_unified_1d_level_v4s32_f32:
+ case Intrinsic::nvvm_tex_unified_1d_grad_v4s32_f32:
+ case Intrinsic::nvvm_tex_unified_1d_array_v4s32_s32:
+ case Intrinsic::nvvm_tex_unified_1d_array_v4s32_f32:
+ case Intrinsic::nvvm_tex_unified_1d_array_level_v4s32_f32:
+ case Intrinsic::nvvm_tex_unified_1d_array_grad_v4s32_f32:
+ case Intrinsic::nvvm_tex_unified_2d_v4s32_s32:
+ case Intrinsic::nvvm_tex_unified_2d_v4s32_f32:
+ case Intrinsic::nvvm_tex_unified_2d_level_v4s32_f32:
+ case Intrinsic::nvvm_tex_unified_2d_grad_v4s32_f32:
+ case Intrinsic::nvvm_tex_unified_2d_array_v4s32_s32:
+ case Intrinsic::nvvm_tex_unified_2d_array_v4s32_f32:
+ case Intrinsic::nvvm_tex_unified_2d_array_level_v4s32_f32:
+ case Intrinsic::nvvm_tex_unified_2d_array_grad_v4s32_f32:
+ case Intrinsic::nvvm_tex_unified_3d_v4s32_s32:
+ case Intrinsic::nvvm_tex_unified_3d_v4s32_f32:
+ case Intrinsic::nvvm_tex_unified_3d_level_v4s32_f32:
+ case Intrinsic::nvvm_tex_unified_3d_grad_v4s32_f32:
+ case Intrinsic::nvvm_tex_unified_1d_v4u32_s32:
+ case Intrinsic::nvvm_tex_unified_1d_v4u32_f32:
+ case Intrinsic::nvvm_tex_unified_1d_level_v4u32_f32:
+ case Intrinsic::nvvm_tex_unified_1d_grad_v4u32_f32:
+ case Intrinsic::nvvm_tex_unified_1d_array_v4u32_s32:
+ case Intrinsic::nvvm_tex_unified_1d_array_v4u32_f32:
+ case Intrinsic::nvvm_tex_unified_1d_array_level_v4u32_f32:
+ case Intrinsic::nvvm_tex_unified_1d_array_grad_v4u32_f32:
+ case Intrinsic::nvvm_tex_unified_2d_v4u32_s32:
+ case Intrinsic::nvvm_tex_unified_2d_v4u32_f32:
+ case Intrinsic::nvvm_tex_unified_2d_level_v4u32_f32:
+ case Intrinsic::nvvm_tex_unified_2d_grad_v4u32_f32:
+ case Intrinsic::nvvm_tex_unified_2d_array_v4u32_s32:
+ case Intrinsic::nvvm_tex_unified_2d_array_v4u32_f32:
+ case Intrinsic::nvvm_tex_unified_2d_array_level_v4u32_f32:
+ case Intrinsic::nvvm_tex_unified_2d_array_grad_v4u32_f32:
+ case Intrinsic::nvvm_tex_unified_3d_v4u32_s32:
+ case Intrinsic::nvvm_tex_unified_3d_v4u32_f32:
+ case Intrinsic::nvvm_tex_unified_3d_level_v4u32_f32:
+ case Intrinsic::nvvm_tex_unified_3d_grad_v4u32_f32:
+ case Intrinsic::nvvm_tex_unified_cube_v4s32_f32:
+ case Intrinsic::nvvm_tex_unified_cube_level_v4s32_f32:
+ case Intrinsic::nvvm_tex_unified_cube_array_v4s32_f32:
+ case Intrinsic::nvvm_tex_unified_cube_array_level_v4s32_f32:
+ case Intrinsic::nvvm_tex_unified_cube_v4u32_f32:
+ case Intrinsic::nvvm_tex_unified_cube_level_v4u32_f32:
+ case Intrinsic::nvvm_tex_unified_cube_array_v4u32_f32:
+ case Intrinsic::nvvm_tex_unified_cube_array_level_v4u32_f32:
+ case Intrinsic::nvvm_tld4_unified_r_2d_v4s32_f32:
+ case Intrinsic::nvvm_tld4_unified_g_2d_v4s32_f32:
+ case Intrinsic::nvvm_tld4_unified_b_2d_v4s32_f32:
+ case Intrinsic::nvvm_tld4_unified_a_2d_v4s32_f32:
+ case Intrinsic::nvvm_tld4_unified_r_2d_v4u32_f32:
+ case Intrinsic::nvvm_tld4_unified_g_2d_v4u32_f32:
+ case Intrinsic::nvvm_tld4_unified_b_2d_v4u32_f32:
+ case Intrinsic::nvvm_tld4_unified_a_2d_v4u32_f32: {
Info.opc = getOpcForTextureInstr(Intrinsic);
- Info.memVT = MVT::i32;
+ Info.memVT = MVT::v4i32;
Info.ptrVal = nullptr;
Info.offset = 0;
Info.vol = 0;
Info.align = 16;
return true;
}
+ case Intrinsic::nvvm_suld_1d_i8_clamp:
+ case Intrinsic::nvvm_suld_1d_v2i8_clamp:
+ case Intrinsic::nvvm_suld_1d_v4i8_clamp:
+ case Intrinsic::nvvm_suld_1d_array_i8_clamp:
+ case Intrinsic::nvvm_suld_1d_array_v2i8_clamp:
+ case Intrinsic::nvvm_suld_1d_array_v4i8_clamp:
+ case Intrinsic::nvvm_suld_2d_i8_clamp:
+ case Intrinsic::nvvm_suld_2d_v2i8_clamp:
+ case Intrinsic::nvvm_suld_2d_v4i8_clamp:
+ case Intrinsic::nvvm_suld_2d_array_i8_clamp:
+ case Intrinsic::nvvm_suld_2d_array_v2i8_clamp:
+ case Intrinsic::nvvm_suld_2d_array_v4i8_clamp:
+ case Intrinsic::nvvm_suld_3d_i8_clamp:
+ case Intrinsic::nvvm_suld_3d_v2i8_clamp:
+ case Intrinsic::nvvm_suld_3d_v4i8_clamp:
case Intrinsic::nvvm_suld_1d_i8_trap:
case Intrinsic::nvvm_suld_1d_v2i8_trap:
case Intrinsic::nvvm_suld_1d_v4i8_trap:
case Intrinsic::nvvm_suld_2d_array_v4i8_trap:
case Intrinsic::nvvm_suld_3d_i8_trap:
case Intrinsic::nvvm_suld_3d_v2i8_trap:
- case Intrinsic::nvvm_suld_3d_v4i8_trap: {
+ case Intrinsic::nvvm_suld_3d_v4i8_trap:
+ case Intrinsic::nvvm_suld_1d_i8_zero:
+ case Intrinsic::nvvm_suld_1d_v2i8_zero:
+ case Intrinsic::nvvm_suld_1d_v4i8_zero:
+ case Intrinsic::nvvm_suld_1d_array_i8_zero:
+ case Intrinsic::nvvm_suld_1d_array_v2i8_zero:
+ case Intrinsic::nvvm_suld_1d_array_v4i8_zero:
+ case Intrinsic::nvvm_suld_2d_i8_zero:
+ case Intrinsic::nvvm_suld_2d_v2i8_zero:
+ case Intrinsic::nvvm_suld_2d_v4i8_zero:
+ case Intrinsic::nvvm_suld_2d_array_i8_zero:
+ case Intrinsic::nvvm_suld_2d_array_v2i8_zero:
+ case Intrinsic::nvvm_suld_2d_array_v4i8_zero:
+ case Intrinsic::nvvm_suld_3d_i8_zero:
+ case Intrinsic::nvvm_suld_3d_v2i8_zero:
+ case Intrinsic::nvvm_suld_3d_v4i8_zero: {
Info.opc = getOpcForSurfaceInstr(Intrinsic);
Info.memVT = MVT::i8;
Info.ptrVal = nullptr;
Info.align = 16;
return true;
}
+ case Intrinsic::nvvm_suld_1d_i16_clamp:
+ case Intrinsic::nvvm_suld_1d_v2i16_clamp:
+ case Intrinsic::nvvm_suld_1d_v4i16_clamp:
+ case Intrinsic::nvvm_suld_1d_array_i16_clamp:
+ case Intrinsic::nvvm_suld_1d_array_v2i16_clamp:
+ case Intrinsic::nvvm_suld_1d_array_v4i16_clamp:
+ case Intrinsic::nvvm_suld_2d_i16_clamp:
+ case Intrinsic::nvvm_suld_2d_v2i16_clamp:
+ case Intrinsic::nvvm_suld_2d_v4i16_clamp:
+ case Intrinsic::nvvm_suld_2d_array_i16_clamp:
+ case Intrinsic::nvvm_suld_2d_array_v2i16_clamp:
+ case Intrinsic::nvvm_suld_2d_array_v4i16_clamp:
+ case Intrinsic::nvvm_suld_3d_i16_clamp:
+ case Intrinsic::nvvm_suld_3d_v2i16_clamp:
+ case Intrinsic::nvvm_suld_3d_v4i16_clamp:
case Intrinsic::nvvm_suld_1d_i16_trap:
case Intrinsic::nvvm_suld_1d_v2i16_trap:
case Intrinsic::nvvm_suld_1d_v4i16_trap:
case Intrinsic::nvvm_suld_2d_array_v4i16_trap:
case Intrinsic::nvvm_suld_3d_i16_trap:
case Intrinsic::nvvm_suld_3d_v2i16_trap:
- case Intrinsic::nvvm_suld_3d_v4i16_trap: {
+ case Intrinsic::nvvm_suld_3d_v4i16_trap:
+ case Intrinsic::nvvm_suld_1d_i16_zero:
+ case Intrinsic::nvvm_suld_1d_v2i16_zero:
+ case Intrinsic::nvvm_suld_1d_v4i16_zero:
+ case Intrinsic::nvvm_suld_1d_array_i16_zero:
+ case Intrinsic::nvvm_suld_1d_array_v2i16_zero:
+ case Intrinsic::nvvm_suld_1d_array_v4i16_zero:
+ case Intrinsic::nvvm_suld_2d_i16_zero:
+ case Intrinsic::nvvm_suld_2d_v2i16_zero:
+ case Intrinsic::nvvm_suld_2d_v4i16_zero:
+ case Intrinsic::nvvm_suld_2d_array_i16_zero:
+ case Intrinsic::nvvm_suld_2d_array_v2i16_zero:
+ case Intrinsic::nvvm_suld_2d_array_v4i16_zero:
+ case Intrinsic::nvvm_suld_3d_i16_zero:
+ case Intrinsic::nvvm_suld_3d_v2i16_zero:
+ case Intrinsic::nvvm_suld_3d_v4i16_zero: {
Info.opc = getOpcForSurfaceInstr(Intrinsic);
Info.memVT = MVT::i16;
Info.ptrVal = nullptr;
Info.align = 16;
return true;
}
+ case Intrinsic::nvvm_suld_1d_i32_clamp:
+ case Intrinsic::nvvm_suld_1d_v2i32_clamp:
+ case Intrinsic::nvvm_suld_1d_v4i32_clamp:
+ case Intrinsic::nvvm_suld_1d_array_i32_clamp:
+ case Intrinsic::nvvm_suld_1d_array_v2i32_clamp:
+ case Intrinsic::nvvm_suld_1d_array_v4i32_clamp:
+ case Intrinsic::nvvm_suld_2d_i32_clamp:
+ case Intrinsic::nvvm_suld_2d_v2i32_clamp:
+ case Intrinsic::nvvm_suld_2d_v4i32_clamp:
+ case Intrinsic::nvvm_suld_2d_array_i32_clamp:
+ case Intrinsic::nvvm_suld_2d_array_v2i32_clamp:
+ case Intrinsic::nvvm_suld_2d_array_v4i32_clamp:
+ case Intrinsic::nvvm_suld_3d_i32_clamp:
+ case Intrinsic::nvvm_suld_3d_v2i32_clamp:
+ case Intrinsic::nvvm_suld_3d_v4i32_clamp:
case Intrinsic::nvvm_suld_1d_i32_trap:
case Intrinsic::nvvm_suld_1d_v2i32_trap:
case Intrinsic::nvvm_suld_1d_v4i32_trap:
case Intrinsic::nvvm_suld_2d_array_v4i32_trap:
case Intrinsic::nvvm_suld_3d_i32_trap:
case Intrinsic::nvvm_suld_3d_v2i32_trap:
- case Intrinsic::nvvm_suld_3d_v4i32_trap: {
+ case Intrinsic::nvvm_suld_3d_v4i32_trap:
+ case Intrinsic::nvvm_suld_1d_i32_zero:
+ case Intrinsic::nvvm_suld_1d_v2i32_zero:
+ case Intrinsic::nvvm_suld_1d_v4i32_zero:
+ case Intrinsic::nvvm_suld_1d_array_i32_zero:
+ case Intrinsic::nvvm_suld_1d_array_v2i32_zero:
+ case Intrinsic::nvvm_suld_1d_array_v4i32_zero:
+ case Intrinsic::nvvm_suld_2d_i32_zero:
+ case Intrinsic::nvvm_suld_2d_v2i32_zero:
+ case Intrinsic::nvvm_suld_2d_v4i32_zero:
+ case Intrinsic::nvvm_suld_2d_array_i32_zero:
+ case Intrinsic::nvvm_suld_2d_array_v2i32_zero:
+ case Intrinsic::nvvm_suld_2d_array_v4i32_zero:
+ case Intrinsic::nvvm_suld_3d_i32_zero:
+ case Intrinsic::nvvm_suld_3d_v2i32_zero:
+ case Intrinsic::nvvm_suld_3d_v4i32_zero: {
Info.opc = getOpcForSurfaceInstr(Intrinsic);
Info.memVT = MVT::i32;
Info.ptrVal = nullptr;
Info.align = 16;
return true;
}
-
+ case Intrinsic::nvvm_suld_1d_i64_clamp:
+ case Intrinsic::nvvm_suld_1d_v2i64_clamp:
+ case Intrinsic::nvvm_suld_1d_array_i64_clamp:
+ case Intrinsic::nvvm_suld_1d_array_v2i64_clamp:
+ case Intrinsic::nvvm_suld_2d_i64_clamp:
+ case Intrinsic::nvvm_suld_2d_v2i64_clamp:
+ case Intrinsic::nvvm_suld_2d_array_i64_clamp:
+ case Intrinsic::nvvm_suld_2d_array_v2i64_clamp:
+ case Intrinsic::nvvm_suld_3d_i64_clamp:
+ case Intrinsic::nvvm_suld_3d_v2i64_clamp:
+ case Intrinsic::nvvm_suld_1d_i64_trap:
+ case Intrinsic::nvvm_suld_1d_v2i64_trap:
+ case Intrinsic::nvvm_suld_1d_array_i64_trap:
+ case Intrinsic::nvvm_suld_1d_array_v2i64_trap:
+ case Intrinsic::nvvm_suld_2d_i64_trap:
+ case Intrinsic::nvvm_suld_2d_v2i64_trap:
+ case Intrinsic::nvvm_suld_2d_array_i64_trap:
+ case Intrinsic::nvvm_suld_2d_array_v2i64_trap:
+ case Intrinsic::nvvm_suld_3d_i64_trap:
+ case Intrinsic::nvvm_suld_3d_v2i64_trap:
+ case Intrinsic::nvvm_suld_1d_i64_zero:
+ case Intrinsic::nvvm_suld_1d_v2i64_zero:
+ case Intrinsic::nvvm_suld_1d_array_i64_zero:
+ case Intrinsic::nvvm_suld_1d_array_v2i64_zero:
+ case Intrinsic::nvvm_suld_2d_i64_zero:
+ case Intrinsic::nvvm_suld_2d_v2i64_zero:
+ case Intrinsic::nvvm_suld_2d_array_i64_zero:
+ case Intrinsic::nvvm_suld_2d_array_v2i64_zero:
+ case Intrinsic::nvvm_suld_3d_i64_zero:
+ case Intrinsic::nvvm_suld_3d_v2i64_zero: {
+ Info.opc = getOpcForSurfaceInstr(Intrinsic);
+ Info.memVT = MVT::i64;
+ Info.ptrVal = nullptr;
+ Info.offset = 0;
+ Info.vol = 0;
+ Info.readMem = true;
+ Info.writeMem = false;
+ Info.align = 16;
+ return true;
+ }
}
return false;
}
/// Used to guide target specific optimizations, like loop strength reduction
/// (LoopStrengthReduce.cpp) and memory optimization for address mode
/// (CodeGenPrepare.cpp)
-bool NVPTXTargetLowering::isLegalAddressingMode(const AddrMode &AM,
- Type *Ty) const {
+bool NVPTXTargetLowering::isLegalAddressingMode(const DataLayout &DL,
+ const AddrMode &AM, Type *Ty,
+ unsigned AS) const {
// AddrMode - This represents an addressing mode of:
// BaseGV + BaseOffs + BaseReg + Scale*ScaleReg
// - [immAddr]
if (AM.BaseGV) {
- if (AM.BaseOffs || AM.HasBaseReg || AM.Scale)
- return false;
- return true;
+ return !AM.BaseOffs && !AM.HasBaseReg && !AM.Scale;
}
switch (AM.Scale) {
/// getConstraintType - Given a constraint letter, return the type of
/// constraint it is for this target.
NVPTXTargetLowering::ConstraintType
-NVPTXTargetLowering::getConstraintType(const std::string &Constraint) const {
+NVPTXTargetLowering::getConstraintType(StringRef Constraint) const {
if (Constraint.size() == 1) {
switch (Constraint[0]) {
default:
break;
+ case 'b':
case 'r':
case 'h':
case 'c':
}
std::pair<unsigned, const TargetRegisterClass *>
-NVPTXTargetLowering::getRegForInlineAsmConstraint(const std::string &Constraint,
+NVPTXTargetLowering::getRegForInlineAsmConstraint(const TargetRegisterInfo *TRI,
+ StringRef Constraint,
MVT VT) const {
if (Constraint.size() == 1) {
switch (Constraint[0]) {
+ case 'b':
+ return std::make_pair(0U, &NVPTX::Int1RegsRegClass);
case 'c':
return std::make_pair(0U, &NVPTX::Int16RegsRegClass);
case 'h':
return std::make_pair(0U, &NVPTX::Float64RegsRegClass);
}
}
- return TargetLowering::getRegForInlineAsmConstraint(Constraint, VT);
+ return TargetLowering::getRegForInlineAsmConstraint(TRI, Constraint, VT);
}
/// getFunctionAlignment - Return the Log2 alignment of this function.
// NVPTX DAG Combining
//===----------------------------------------------------------------------===//
-extern unsigned FMAContractLevel;
+bool NVPTXTargetLowering::allowFMA(MachineFunction &MF,
+ CodeGenOpt::Level OptLevel) const {
+ const Function *F = MF.getFunction();
+ const TargetOptions &TO = MF.getTarget().Options;
+
+ // Always honor command-line argument
+ if (FMAContractLevelOpt.getNumOccurrences() > 0) {
+ return FMAContractLevelOpt > 0;
+ } else if (OptLevel == 0) {
+ // Do not contract if we're not optimizing the code
+ return false;
+ } else if (TO.AllowFPOpFusion == FPOpFusion::Fast || TO.UnsafeFPMath) {
+ // Honor TargetOptions flags that explicitly say fusion is okay
+ return true;
+ } else if (F->hasFnAttribute("unsafe-fp-math")) {
+ // Check for unsafe-fp-math=true coming from Clang
+ Attribute Attr = F->getFnAttribute("unsafe-fp-math");
+ StringRef Val = Attr.getValueAsString();
+ if (Val == "true")
+ return true;
+ }
+
+ // We did not have a clear indication that fusion is allowed, so assume not
+ return false;
+}
/// PerformADDCombineWithOperands - Try DAG combinations for an ADD with
/// operands N0 and N1. This is a helper for PerformADDCombine that is
}
else if (N0.getOpcode() == ISD::FMUL) {
if (VT == MVT::f32 || VT == MVT::f64) {
- if (FMAContractLevel == 0)
+ const auto *TLI = static_cast<const NVPTXTargetLowering *>(
+ &DAG.getTargetLoweringInfo());
+ if (!TLI->allowFMA(DAG.getMachineFunction(), OptLevel))
return SDValue();
// For floating point:
const SDNode *left = N0.getOperand(0).getNode();
const SDNode *right = N0.getOperand(1).getNode();
- if (dyn_cast<ConstantSDNode>(left) || dyn_cast<ConstantSDNode>(right))
+ if (isa<ConstantSDNode>(left) || isa<ConstantSDNode>(right))
opIsLive = true;
if (!opIsLive)
return SDValue();
}
+static SDValue PerformSELECTCombine(SDNode *N,
+ TargetLowering::DAGCombinerInfo &DCI) {
+ // Currently this detects patterns for integer min and max and
+ // lowers them to PTX-specific intrinsics that enable hardware
+ // support.
+
+ const SDValue Cond = N->getOperand(0);
+ if (Cond.getOpcode() != ISD::SETCC) return SDValue();
+
+ const SDValue LHS = Cond.getOperand(0);
+ const SDValue RHS = Cond.getOperand(1);
+ const SDValue True = N->getOperand(1);
+ const SDValue False = N->getOperand(2);
+ if (!(LHS == True && RHS == False) && !(LHS == False && RHS == True))
+ return SDValue();
+
+ const EVT VT = N->getValueType(0);
+ if (VT != MVT::i32 && VT != MVT::i64) return SDValue();
+
+ const ISD::CondCode CC = cast<CondCodeSDNode>(Cond.getOperand(2))->get();
+ SDValue Larger; // The larger of LHS and RHS when condition is true.
+ switch (CC) {
+ case ISD::SETULT:
+ case ISD::SETULE:
+ case ISD::SETLT:
+ case ISD::SETLE:
+ Larger = RHS;
+ break;
+
+ case ISD::SETGT:
+ case ISD::SETGE:
+ case ISD::SETUGT:
+ case ISD::SETUGE:
+ Larger = LHS;
+ break;
+
+ default:
+ return SDValue();
+ }
+ const bool IsMax = (Larger == True);
+ const bool IsSigned = ISD::isSignedIntSetCC(CC);
+
+ unsigned IntrinsicId;
+ if (VT == MVT::i32) {
+ if (IsSigned)
+ IntrinsicId = IsMax ? Intrinsic::nvvm_max_i : Intrinsic::nvvm_min_i;
+ else
+ IntrinsicId = IsMax ? Intrinsic::nvvm_max_ui : Intrinsic::nvvm_min_ui;
+ } else {
+ assert(VT == MVT::i64);
+ if (IsSigned)
+ IntrinsicId = IsMax ? Intrinsic::nvvm_max_ll : Intrinsic::nvvm_min_ll;
+ else
+ IntrinsicId = IsMax ? Intrinsic::nvvm_max_ull : Intrinsic::nvvm_min_ull;
+ }
+
+ SDLoc DL(N);
+ return DCI.DAG.getNode(ISD::INTRINSIC_WO_CHAIN, DL, VT,
+ DCI.DAG.getConstant(IntrinsicId, DL, VT), LHS, RHS);
+}
+
enum OperandSignedness {
Signed = 0,
Unsigned,
if (Op.getOpcode() == ISD::SIGN_EXTEND ||
Op.getOpcode() == ISD::SIGN_EXTEND_INREG) {
EVT OrigVT = Op.getOperand(0).getValueType();
- if (OrigVT.getSizeInBits() == OptSize) {
+ if (OrigVT.getSizeInBits() <= OptSize) {
S = Signed;
return true;
}
} else if (Op.getOpcode() == ISD::ZERO_EXTEND) {
EVT OrigVT = Op.getOperand(0).getValueType();
- if (OrigVT.getSizeInBits() == OptSize) {
+ if (OrigVT.getSizeInBits() <= OptSize) {
S = Unsigned;
return true;
}
if (ConstantSDNode *CI = dyn_cast<ConstantSDNode>(RHS)) {
APInt Val = CI->getAPIntValue();
if (LHSSign == Unsigned) {
- if (Val.isIntN(OptSize)) {
- return true;
- }
- return false;
+ return Val.isIntN(OptSize);
} else {
- if (Val.isSignedIntN(OptSize)) {
- return true;
- }
- return false;
+ return Val.isSignedIntN(OptSize);
}
} else {
OperandSignedness RHSSign;
if (!IsMulWideOperandDemotable(RHS, OptSize, RHSSign))
return false;
- if (LHSSign != RHSSign)
- return false;
-
- return true;
+ return LHSSign == RHSSign;
}
}
return SDValue();
}
+ SDLoc DL(N);
unsigned OptSize = MulType.getSizeInBits() >> 1;
SDValue LHS = N->getOperand(0);
SDValue RHS = N->getOperand(1);
unsigned BitWidth = MulType.getSizeInBits();
if (ShiftAmt.sge(0) && ShiftAmt.slt(BitWidth)) {
APInt MulVal = APInt(BitWidth, 1) << ShiftAmt;
- RHS = DCI.DAG.getConstant(MulVal, MulType);
+ RHS = DCI.DAG.getConstant(MulVal, DL, MulType);
} else {
return SDValue();
}
// Truncate the operands to the correct size. Note that these are just for
// type consistency and will (likely) be eliminated in later phases.
SDValue TruncLHS =
- DCI.DAG.getNode(ISD::TRUNCATE, SDLoc(N), DemotedVT, LHS);
+ DCI.DAG.getNode(ISD::TRUNCATE, DL, DemotedVT, LHS);
SDValue TruncRHS =
- DCI.DAG.getNode(ISD::TRUNCATE, SDLoc(N), DemotedVT, RHS);
+ DCI.DAG.getNode(ISD::TRUNCATE, DL, DemotedVT, RHS);
unsigned Opc;
if (Signed) {
Opc = NVPTXISD::MUL_WIDE_UNSIGNED;
}
- return DCI.DAG.getNode(Opc, SDLoc(N), MulType, TruncLHS, TruncRHS);
+ return DCI.DAG.getNode(Opc, DL, MulType, TruncLHS, TruncRHS);
}
/// PerformMULCombine - Runs PTX-specific DAG combine patterns on MUL nodes.
SDValue NVPTXTargetLowering::PerformDAGCombine(SDNode *N,
DAGCombinerInfo &DCI) const {
- // FIXME: Get this from the DAG somehow
- CodeGenOpt::Level OptLevel = CodeGenOpt::Aggressive;
+ CodeGenOpt::Level OptLevel = getTargetMachine().getOptLevel();
switch (N->getOpcode()) {
default: break;
case ISD::ADD:
case ISD::FADD:
- return PerformADDCombine(N, DCI, nvptxSubtarget, OptLevel);
+ return PerformADDCombine(N, DCI, STI, OptLevel);
case ISD::MUL:
return PerformMULCombine(N, DCI, OptLevel);
case ISD::SHL:
return PerformSHLCombine(N, DCI, OptLevel);
case ISD::AND:
return PerformANDCombine(N, DCI);
+ case ISD::SELECT:
+ return PerformSELECTCombine(N, DCI);
}
return SDValue();
}
break;
}
+ LoadSDNode *LD = cast<LoadSDNode>(N);
+
+ unsigned Align = LD->getAlignment();
+ auto &TD = DAG.getDataLayout();
+ unsigned PrefAlign =
+ TD.getPrefTypeAlignment(ResVT.getTypeForEVT(*DAG.getContext()));
+ if (Align < PrefAlign) {
+ // This load is not sufficiently aligned, so bail out and let this vector
+ // load be scalarized. Note that we may still be able to emit smaller
+ // vector loads. For example, if we are loading a <4 x float> with an
+ // alignment of 8, this check will fail but the legalizer will try again
+ // with 2 x <2 x float>, which will succeed with an alignment of 8.
+ return;
+ }
+
EVT EltVT = ResVT.getVectorElementType();
unsigned NumElts = ResVT.getVectorNumElements();
}
}
- SmallVector<SDValue, 8> OtherOps;
-
// Copy regular operands
- for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i)
- OtherOps.push_back(N->getOperand(i));
-
- LoadSDNode *LD = cast<LoadSDNode>(N);
+ SmallVector<SDValue, 8> OtherOps(N->op_begin(), N->op_end());
// The select routine does not have access to the LoadSDNode instance, so
// pass along the extension information
- OtherOps.push_back(DAG.getIntPtrConstant(LD->getExtensionType()));
+ OtherOps.push_back(DAG.getIntPtrConstant(LD->getExtensionType(), DL));
SDValue NewLD = DAG.getMemIntrinsicNode(Opcode, DL, LdResVTs, OtherOps,
LD->getMemoryVT(),
OtherOps.push_back(Chain); // Chain
// Skip operand 1 (intrinsic ID)
// Others
- for (unsigned i = 2, e = N->getNumOperands(); i != e; ++i)
- OtherOps.push_back(N->getOperand(i));
+ OtherOps.append(N->op_begin() + 2, N->op_end());
MemIntrinsicSDNode *MemSD = cast<MemIntrinsicSDNode>(N);
"Custom handling of non-i8 ldu/ldg?");
// Just copy all operands as-is
- SmallVector<SDValue, 4> Ops;
- for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i)
- Ops.push_back(N->getOperand(i));
+ SmallVector<SDValue, 4> Ops(N->op_begin(), N->op_end());
// Force output to i16
SDVTList LdResVTs = DAG.getVTList(MVT::i16, MVT::Other);
delete DwarfLocSection;
delete DwarfARangesSection;
delete DwarfRangesSection;
- delete DwarfMacroInfoSection;
+}
+
+MCSection *
+NVPTXTargetObjectFile::SelectSectionForGlobal(const GlobalValue *GV,
+ SectionKind Kind, Mangler &Mang,
+ const TargetMachine &TM) const {
+ return getDataSection();
}
projects / oota-llvm.git / blobdiff