// See the License for the specific language governing permissions and
// limitations under the License.
-#include "Nucleus.hpp"
-
#include "Reactor.hpp"
-#include "Routine.hpp"
+
+#include "Optimizer.hpp"
+#include "ExecutableMemory.hpp"
#include "src/IceTypes.h"
#include "src/IceCfg.h"
#include "llvm/Support/FileSystem.h"
#include "llvm/Support/raw_os_ostream.h"
+#include "llvm/Support/Compiler.h"
+#if __has_feature(memory_sanitizer)
+#include <sanitizer/msan_interface.h>
+#endif
+
+#if defined(_WIN32)
+#ifndef WIN32_LEAN_AND_MEAN
#define WIN32_LEAN_AND_MEAN
+#endif // !WIN32_LEAN_AND_MEAN
+#ifndef NOMINMAX
#define NOMINMAX
+#endif // !NOMINMAX
#include <Windows.h>
+#else
+#include <sys/mman.h>
+#if !defined(MAP_ANONYMOUS)
+#define MAP_ANONYMOUS MAP_ANON
+#endif
+#endif
#include <mutex>
#include <limits>
Ice::Cfg *function = nullptr;
Ice::CfgNode *basicBlock = nullptr;
Ice::CfgLocalAllocatorScope *allocator = nullptr;
- sw::Routine *routine = nullptr;
+ rr::Routine *routine = nullptr;
std::mutex codegenMutex;
- sw::BasicBlock *falseBB = nullptr;
-
Ice::ELFFileStreamer *elfFile = nullptr;
Ice::Fdstream *out = nullptr;
}
-namespace sw
+namespace
+{
+ #if !defined(__i386__) && defined(_M_IX86)
+ #define __i386__ 1
+ #endif
+
+ #if !defined(__x86_64__) && (defined(_M_AMD64) || defined (_M_X64))
+ #define __x86_64__ 1
+ #endif
+
+ class CPUID
+ {
+ public:
+ const static bool ARM;
+ const static bool SSE4_1;
+
+ private:
+ static void cpuid(int registers[4], int info)
+ {
+ #if defined(__i386__) || defined(__x86_64__)
+ #if defined(_WIN32)
+ __cpuid(registers, info);
+ #else
+ __asm volatile("cpuid": "=a" (registers[0]), "=b" (registers[1]), "=c" (registers[2]), "=d" (registers[3]): "a" (info));
+ #endif
+ #else
+ registers[0] = 0;
+ registers[1] = 0;
+ registers[2] = 0;
+ registers[3] = 0;
+ #endif
+ }
+
+ static bool detectARM()
+ {
+ #if defined(__arm__) || defined(__aarch64__)
+ return true;
+ #elif defined(__i386__) || defined(__x86_64__)
+ return false;
+ #elif defined(__mips__)
+ return false;
+ #else
+ #error "Unknown architecture"
+ #endif
+ }
+
+ static bool detectSSE4_1()
+ {
+ #if defined(__i386__) || defined(__x86_64__)
+ int registers[4];
+ cpuid(registers, 1);
+ return (registers[2] & 0x00080000) != 0;
+ #else
+ return false;
+ #endif
+ }
+ };
+
+ const bool CPUID::ARM = CPUID::detectARM();
+ const bool CPUID::SSE4_1 = CPUID::detectSSE4_1();
+ const bool emulateIntrinsics = false;
+ const bool emulateMismatchedBitCast = CPUID::ARM;
+}
+
+namespace rr
{
enum EmulatedType
{
Type_v2f32 = Ice::IceType_v4f32 | EmulatedV2,
};
- class Value : public Ice::Variable {};
+ class Value : public Ice::Operand {};
+ class SwitchCases : public Ice::InstSwitch {};
class BasicBlock : public Ice::CfgNode {};
Ice::Type T(Type *t)
{
- static_assert(Ice::IceType_NUM < EmulatedBits, "Ice::Type overlaps with our emulated types!");
+ static_assert(static_cast<unsigned int>(Ice::IceType_NUM) < static_cast<unsigned int>(EmulatedBits), "Ice::Type overlaps with our emulated types!");
return (Ice::Type)(reinterpret_cast<std::intptr_t>(t) & ~EmulatedBits);
}
return reinterpret_cast<Type*>(t);
}
- Value *V(Ice::Variable *v)
+ Value *V(Ice::Operand *v)
{
return reinterpret_cast<Value*>(v);
}
- Value *C(Ice::Constant *c) // Only safe for casting right-hand side operand
+ BasicBlock *B(Ice::CfgNode *b)
{
- return reinterpret_cast<Value*>(c);
+ return reinterpret_cast<BasicBlock*>(b);
}
- BasicBlock *B(Ice::CfgNode *b)
+ static size_t typeSize(Type *type)
{
- return reinterpret_cast<BasicBlock*>(b);
+ if(reinterpret_cast<std::intptr_t>(type) & EmulatedBits)
+ {
+ switch(reinterpret_cast<std::intptr_t>(type))
+ {
+ case Type_v2i32: return 8;
+ case Type_v4i16: return 8;
+ case Type_v2i16: return 4;
+ case Type_v8i8: return 8;
+ case Type_v4i8: return 4;
+ case Type_v2f32: return 8;
+ default: assert(false);
+ }
+ }
+
+ return Ice::typeWidthInBytes(T(type));
}
Optimization optimization[10] = {InstructionCombining, Disabled};
{
return reinterpret_cast<const SectionHeader*>((intptr_t)elfHeader + elfHeader->e_shoff);
}
-
+
inline const SectionHeader *elfSection(const ElfHeader *elfHeader, int index)
{
return §ionHeader(elfHeader)[index];
static void *relocateSymbol(const ElfHeader *elfHeader, const Elf32_Rel &relocation, const SectionHeader &relocationTable)
{
const SectionHeader *target = elfSection(elfHeader, relocationTable.sh_info);
-
- intptr_t address = (intptr_t)elfHeader + target->sh_offset;
- int32_t *patchSite = (int*)(address + relocation.r_offset);
+
uint32_t index = relocation.getSymbol();
int table = relocationTable.sh_link;
void *symbolValue = nullptr;
-
+
if(index != SHN_UNDEF)
{
if(table == SHN_UNDEF) return nullptr;
const SectionHeader *symbolTable = elfSection(elfHeader, table);
-
+
uint32_t symtab_entries = symbolTable->sh_size / symbolTable->sh_entsize;
if(index >= symtab_entries)
{
assert(index < symtab_entries && "Symbol Index out of range");
return nullptr;
}
-
+
intptr_t symbolAddress = (intptr_t)elfHeader + symbolTable->sh_offset;
Elf32_Sym &symbol = ((Elf32_Sym*)symbolAddress)[index];
uint16_t section = symbol.st_shndx;
}
}
- switch(relocation.getType())
+ intptr_t address = (intptr_t)elfHeader + target->sh_offset;
+ unaligned_ptr<int32_t> patchSite = (int32_t*)(address + relocation.r_offset);
+
+ if(CPUID::ARM)
{
- case R_386_NONE:
- // No relocation
- break;
- case R_386_32:
- *patchSite = (int32_t)((intptr_t)symbolValue + *patchSite);
- break;
- // case R_386_PC32:
- // *patchSite = (int32_t)((intptr_t)symbolValue + *patchSite - (intptr_t)patchSite);
- // break;
- default:
- assert(false && "Unsupported relocation type");
- return nullptr;
+ switch(relocation.getType())
+ {
+ case R_ARM_NONE:
+ // No relocation
+ break;
+ case R_ARM_MOVW_ABS_NC:
+ {
+ uint32_t thumb = 0; // Calls to Thumb code not supported.
+ uint32_t lo = (uint32_t)(intptr_t)symbolValue | thumb;
+ *patchSite = (*patchSite & 0xFFF0F000) | ((lo & 0xF000) << 4) | (lo & 0x0FFF);
+ }
+ break;
+ case R_ARM_MOVT_ABS:
+ {
+ uint32_t hi = (uint32_t)(intptr_t)(symbolValue) >> 16;
+ *patchSite = (*patchSite & 0xFFF0F000) | ((hi & 0xF000) << 4) | (hi & 0x0FFF);
+ }
+ break;
+ default:
+ assert(false && "Unsupported relocation type");
+ return nullptr;
+ }
+ }
+ else
+ {
+ switch(relocation.getType())
+ {
+ case R_386_NONE:
+ // No relocation
+ break;
+ case R_386_32:
+ *patchSite = (int32_t)((intptr_t)symbolValue + *patchSite);
+ break;
+ // case R_386_PC32:
+ // *patchSite = (int32_t)((intptr_t)symbolValue + *patchSite - (intptr_t)patchSite);
+ // break;
+ default:
+ assert(false && "Unsupported relocation type");
+ return nullptr;
+ }
}
return symbolValue;
static void *relocateSymbol(const ElfHeader *elfHeader, const Elf64_Rela &relocation, const SectionHeader &relocationTable)
{
const SectionHeader *target = elfSection(elfHeader, relocationTable.sh_info);
-
- intptr_t address = (intptr_t)elfHeader + target->sh_offset;
- int32_t *patchSite = (int*)(address + relocation.r_offset);
+
uint32_t index = relocation.getSymbol();
int table = relocationTable.sh_link;
void *symbolValue = nullptr;
{
if(table == SHN_UNDEF) return nullptr;
const SectionHeader *symbolTable = elfSection(elfHeader, table);
-
+
uint32_t symtab_entries = symbolTable->sh_size / symbolTable->sh_entsize;
if(index >= symtab_entries)
{
assert(index < symtab_entries && "Symbol Index out of range");
return nullptr;
}
-
+
intptr_t symbolAddress = (intptr_t)elfHeader + symbolTable->sh_offset;
Elf64_Sym &symbol = ((Elf64_Sym*)symbolAddress)[index];
uint16_t section = symbol.st_shndx;
}
}
+ intptr_t address = (intptr_t)elfHeader + target->sh_offset;
+ unaligned_ptr<int32_t> patchSite32 = (int32_t*)(address + relocation.r_offset);
+ unaligned_ptr<int64_t> patchSite64 = (int64_t*)(address + relocation.r_offset);
+
switch(relocation.getType())
{
case R_X86_64_NONE:
// No relocation
break;
- // case R_X86_64_64:
- // *patchSite = (int32_t)((intptr_t)symbolValue + *patchSite) + relocation->r_addend;
- // break;
+ case R_X86_64_64:
+ *patchSite64 = (int64_t)((intptr_t)symbolValue + *patchSite64 + relocation.r_addend);
+ break;
case R_X86_64_PC32:
- *patchSite = (int32_t)((intptr_t)symbolValue + *patchSite - (intptr_t)patchSite) + relocation.r_addend;
+ *patchSite32 = (int32_t)((intptr_t)symbolValue + *patchSite32 - (intptr_t)patchSite32 + relocation.r_addend);
+ break;
+ case R_X86_64_32S:
+ *patchSite32 = (int32_t)((intptr_t)symbolValue + *patchSite32 + relocation.r_addend);
break;
- // case R_X86_64_32S:
- // *patchSite = (int32_t)((intptr_t)symbolValue + *patchSite) + relocation.r_addend;
- // break;
default:
assert(false && "Unsupported relocation type");
return nullptr;
return symbolValue;
}
- void *loadImage(uint8_t *const elfImage)
+ void *loadImage(uint8_t *const elfImage, size_t &codeSize)
{
ElfHeader *elfHeader = (ElfHeader*)elfImage;
// Expect ELF bitness to match platform
assert(sizeof(void*) == 8 ? elfHeader->getFileClass() == ELFCLASS64 : elfHeader->getFileClass() == ELFCLASS32);
- assert(sizeof(void*) == 8 ? elfHeader->e_machine == EM_X86_64 : elfHeader->e_machine == EM_386);
+ #if defined(__i386__)
+ assert(sizeof(void*) == 4 && elfHeader->e_machine == EM_386);
+ #elif defined(__x86_64__)
+ assert(sizeof(void*) == 8 && elfHeader->e_machine == EM_X86_64);
+ #elif defined(__arm__)
+ assert(sizeof(void*) == 4 && elfHeader->e_machine == EM_ARM);
+ #elif defined(__aarch64__)
+ assert(sizeof(void*) == 8 && elfHeader->e_machine == EM_AARCH64);
+ #elif defined(__mips__)
+ assert(sizeof(void*) == 4 && elfHeader->e_machine == EM_MIPS);
+ #else
+ #error "Unsupported platform"
+ #endif
SectionHeader *sectionHeader = (SectionHeader*)(elfImage + elfHeader->e_shoff);
void *entry = nullptr;
if(sectionHeader[i].sh_flags & SHF_EXECINSTR)
{
entry = elfImage + sectionHeader[i].sh_offset;
+ codeSize = sectionHeader[i].sh_size;
}
}
else if(sectionHeader[i].sh_type == SHT_REL)
{
assert(sizeof(void*) == 4 && "UNIMPLEMENTED"); // Only expected/implemented for 32-bit code
- for(int index = 0; index < sectionHeader[i].sh_size / sectionHeader[i].sh_entsize; index++)
+ for(Elf32_Word index = 0; index < sectionHeader[i].sh_size / sectionHeader[i].sh_entsize; index++)
{
const Elf32_Rel &relocation = ((const Elf32_Rel*)(elfImage + sectionHeader[i].sh_offset))[index];
- void *symbol = relocateSymbol(elfHeader, relocation, sectionHeader[i]);
+ relocateSymbol(elfHeader, relocation, sectionHeader[i]);
}
}
else if(sectionHeader[i].sh_type == SHT_RELA)
{
assert(sizeof(void*) == 8 && "UNIMPLEMENTED"); // Only expected/implemented for 64-bit code
- for(int index = 0; index < sectionHeader[i].sh_size / sectionHeader[i].sh_entsize; index++)
+ for(Elf32_Word index = 0; index < sectionHeader[i].sh_size / sectionHeader[i].sh_entsize; index++)
{
const Elf64_Rela &relocation = ((const Elf64_Rela*)(elfImage + sectionHeader[i].sh_offset))[index];
- void *symbol = relocateSymbol(elfHeader, relocation, sectionHeader[i]);
+ relocateSymbol(elfHeader, relocation, sectionHeader[i]);
}
}
}
T *allocate(size_type n)
{
- return (T*)VirtualAlloc(NULL, sizeof(T) * n, MEM_COMMIT | MEM_RESERVE, PAGE_READWRITE);
+ return (T*)allocateExecutable(sizeof(T) * n);
}
void deallocate(T *p, size_type n)
{
- VirtualFree(p, 0, MEM_RELEASE);
+ deallocateExecutable(p, sizeof(T) * n);
}
};
buffer.reserve(0x1000);
}
- virtual ~ELFMemoryStreamer()
+ ~ELFMemoryStreamer() override
{
- if(buffer.size() != 0)
- {
- DWORD exeProtection;
- VirtualProtect(&buffer[0], buffer.size(), oldProtection, &exeProtection);
- }
+ #if defined(_WIN32)
+ if(buffer.size() != 0)
+ {
+ DWORD exeProtection;
+ VirtualProtect(&buffer[0], buffer.size(), oldProtection, &exeProtection);
+ }
+ #endif
}
void write8(uint8_t Value) override
{
if(!entry)
{
- VirtualProtect(&buffer[0], buffer.size(), PAGE_EXECUTE_READWRITE, &oldProtection);
- position = std::numeric_limits<std::size_t>::max(); // Can't stream more data after this
-
- entry = loadImage(&buffer[0]);
+ position = std::numeric_limits<std::size_t>::max(); // Can't stream more data after this
+
+ size_t codeSize = 0;
+ entry = loadImage(&buffer[0], codeSize);
+
+ #if defined(_WIN32)
+ VirtualProtect(&buffer[0], buffer.size(), PAGE_EXECUTE_READ, &oldProtection);
+ FlushInstructionCache(GetCurrentProcess(), NULL, 0);
+ #else
+ mprotect(&buffer[0], buffer.size(), PROT_READ | PROT_EXEC);
+ __builtin___clear_cache((char*)entry, (char*)entry + codeSize);
+ #endif
}
return entry;
void *entry;
std::vector<uint8_t, ExecutableAllocator<uint8_t>> buffer;
std::size_t position;
+
+ #if defined(_WIN32)
DWORD oldProtection;
+ #endif
};
Nucleus::Nucleus()
Ice::ClFlags &Flags = Ice::ClFlags::Flags;
Ice::ClFlags::getParsedClFlags(Flags);
- Flags.setTargetArch(sizeof(void*) == 8 ? Ice::Target_X8664 : Ice::Target_X8632);
+ #if defined(__arm__)
+ Flags.setTargetArch(Ice::Target_ARM32);
+ Flags.setTargetInstructionSet(Ice::ARM32InstructionSet_HWDivArm);
+ #elif defined(__mips__)
+ Flags.setTargetArch(Ice::Target_MIPS32);
+ Flags.setTargetInstructionSet(Ice::BaseInstructionSet);
+ #else // x86
+ Flags.setTargetArch(sizeof(void*) == 8 ? Ice::Target_X8664 : Ice::Target_X8632);
+ Flags.setTargetInstructionSet(CPUID::SSE4_1 ? Ice::X86InstructionSet_SSE4_1 : Ice::X86InstructionSet_SSE2);
+ #endif
Flags.setOutFileType(Ice::FT_Elf);
Flags.setOptLevel(Ice::Opt_2);
Flags.setApplicationBinaryInterface(Ice::ABI_Platform);
- Flags.setVerbose(false ? Ice::IceV_All : Ice::IceV_None);
+ Flags.setVerbose(false ? Ice::IceV_Most : Ice::IceV_None);
+ Flags.setDisableHybridAssembly(true);
static llvm::raw_os_ostream cout(std::cout);
static llvm::raw_os_ostream cerr(std::cerr);
Nucleus::~Nucleus()
{
+ delete ::routine;
+
delete ::allocator;
delete ::function;
delete ::context;
::codegenMutex.unlock();
}
- Routine *Nucleus::acquireRoutine(const wchar_t *name, bool runOptimizations)
+ Routine *Nucleus::acquireRoutine(const char *name, bool runOptimizations)
{
if(basicBlock->getInsts().empty() || basicBlock->getInsts().back().getKind() != Ice::Inst::Ret)
{
createRetVoid();
}
- std::wstring wideName(name);
- std::string asciiName(wideName.begin(), wideName.end());
- ::function->setFunctionName(Ice::GlobalString::createWithString(::context, asciiName));
+ ::function->setFunctionName(Ice::GlobalString::createWithString(::context, name));
+
+ optimize();
::function->translate();
assert(!::function->hasError());
- auto *globals = ::function->getGlobalInits().release();
+ auto globals = ::function->getGlobalInits();
if(globals && !globals->empty())
{
- ::context->getGlobals()->merge(globals);
+ ::context->getGlobals()->merge(globals.get());
}
::context->emitFileHeader();
objectWriter->writeFunctionCode(::function->getFunctionName(), false, assembler.get());
::context->lowerGlobals("last");
::context->lowerConstants();
+ ::context->lowerJumpTables();
objectWriter->setUndefinedSyms(::context->getConstantExternSyms());
objectWriter->writeNonUserSections();
- return ::routine;
+ Routine *handoffRoutine = ::routine;
+ ::routine = nullptr;
+
+ return handoffRoutine;
}
void Nucleus::optimize()
{
+ rr::optimize(::function);
}
Value *Nucleus::allocateStackVariable(Type *t, int arraySize)
void Nucleus::setInsertBlock(BasicBlock *basicBlock)
{
// assert(::basicBlock->getInsts().back().getTerminatorEdges().size() >= 0 && "Previous basic block must have a terminator");
+
+ Variable::materializeAll();
+
::basicBlock = basicBlock;
}
void Nucleus::createRetVoid()
{
+ // Code generated after this point is unreachable, so any variables
+ // being read can safely return an undefined value. We have to avoid
+ // materializing variables after the terminator ret instruction.
+ Variable::killUnmaterialized();
+
Ice::InstRet *ret = Ice::InstRet::create(::function);
::basicBlock->appendInst(ret);
}
void Nucleus::createRet(Value *v)
{
+ // Code generated after this point is unreachable, so any variables
+ // being read can safely return an undefined value. We have to avoid
+ // materializing variables after the terminator ret instruction.
+ Variable::killUnmaterialized();
+
Ice::InstRet *ret = Ice::InstRet::create(::function, v);
::basicBlock->appendInst(ret);
}
void Nucleus::createBr(BasicBlock *dest)
{
+ Variable::materializeAll();
+
auto br = Ice::InstBr::create(::function, dest);
::basicBlock->appendInst(br);
}
void Nucleus::createCondBr(Value *cond, BasicBlock *ifTrue, BasicBlock *ifFalse)
{
+ Variable::materializeAll();
+
auto br = Ice::InstBr::create(::function, cond, ifTrue, ifFalse);
::basicBlock->appendInst(br);
}
+ static bool isCommutative(Ice::InstArithmetic::OpKind op)
+ {
+ switch(op)
+ {
+ case Ice::InstArithmetic::Add:
+ case Ice::InstArithmetic::Fadd:
+ case Ice::InstArithmetic::Mul:
+ case Ice::InstArithmetic::Fmul:
+ case Ice::InstArithmetic::And:
+ case Ice::InstArithmetic::Or:
+ case Ice::InstArithmetic::Xor:
+ return true;
+ default:
+ return false;
+ }
+ }
+
static Value *createArithmetic(Ice::InstArithmetic::OpKind op, Value *lhs, Value *rhs)
{
- assert(lhs->getType() == rhs->getType() || (llvm::isa<Ice::Constant>(rhs) && (op == Ice::InstArithmetic::Shl || Ice::InstArithmetic::Lshr || Ice::InstArithmetic::Ashr)));
+ assert(lhs->getType() == rhs->getType() || llvm::isa<Ice::Constant>(rhs));
+
+ bool swapOperands = llvm::isa<Ice::Constant>(lhs) && isCommutative(op);
Ice::Variable *result = ::function->makeVariable(lhs->getType());
- Ice::InstArithmetic *arithmetic = Ice::InstArithmetic::create(::function, op, result, lhs, rhs);
+ Ice::InstArithmetic *arithmetic = Ice::InstArithmetic::create(::function, op, result, swapOperands ? rhs : lhs, swapOperands ? lhs : rhs);
::basicBlock->appendInst(arithmetic);
return V(result);
return createArithmetic(Ice::InstArithmetic::Xor, lhs, rhs);
}
- static Value *createAssign(Ice::Constant *constant)
- {
- Ice::Variable *value = ::function->makeVariable(constant->getType());
- auto assign = Ice::InstAssign::create(::function, value, constant);
- ::basicBlock->appendInst(assign);
-
- return V(value);
- }
-
Value *Nucleus::createNeg(Value *v)
{
- assert(false && "UNIMPLEMENTED"); return nullptr;
+ return createSub(createNullValue(T(v->getType())), v);
}
Value *Nucleus::createFNeg(Value *v)
{
- assert(false && "UNIMPLEMENTED"); return nullptr;
+ double c[4] = {-0.0, -0.0, -0.0, -0.0};
+ Value *negativeZero = Ice::isVectorType(v->getType()) ?
+ createConstantVector(c, T(v->getType())) :
+ V(::context->getConstantFloat(-0.0f));
+
+ return createFSub(negativeZero, v);
}
Value *Nucleus::createNot(Value *v)
{
- assert(false && "UNIMPLEMENTED"); return nullptr;
+ if(Ice::isScalarIntegerType(v->getType()))
+ {
+ return createXor(v, V(::context->getConstantInt(v->getType(), -1)));
+ }
+ else // Vector
+ {
+ int64_t c[16] = {-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1};
+ return createXor(v, createConstantVector(c, T(v->getType())));
+ }
}
- Value *Nucleus::createLoad(Value *ptr, Type *type, bool isVolatile, unsigned int align)
+ Value *Nucleus::createLoad(Value *ptr, Type *type, bool isVolatile, unsigned int align, bool atomic, std::memory_order memoryOrder)
{
+ assert(!atomic); // Unimplemented
+ assert(memoryOrder == std::memory_order_relaxed); // Unimplemented
+
int valueType = (int)reinterpret_cast<intptr_t>(type);
Ice::Variable *result = ::function->makeVariable(T(type));
- if(valueType & EmulatedBits)
+ if((valueType & EmulatedBits) && (align != 0)) // Narrow vector not stored on stack.
{
- switch(valueType)
+ if(emulateIntrinsics)
{
- case Type_v4i8:
- case Type_v2i16:
+ if(typeSize(type) == 4)
{
- const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::LoadSubVector, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
- auto target = ::context->getConstantUndef(Ice::IceType_i32);
- auto load = Ice::InstIntrinsicCall::create(::function, 2, result, target, intrinsic);
- load->addArg(::context->getConstantInt32(4));
- load->addArg(ptr);
- ::basicBlock->appendInst(load);
+ auto pointer = RValue<Pointer<Byte>>(ptr);
+ Int x = *Pointer<Int>(pointer);
+
+ Int4 vector;
+ vector = Insert(vector, x, 0);
+
+ auto bitcast = Ice::InstCast::create(::function, Ice::InstCast::Bitcast, result, vector.loadValue());
+ ::basicBlock->appendInst(bitcast);
}
- break;
- case Type_v2i32:
- case Type_v8i8:
- case Type_v4i16:
- case Type_v2f32:
+ else if(typeSize(type) == 8)
{
- const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::LoadSubVector, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
- auto target = ::context->getConstantUndef(Ice::IceType_i32);
- auto load = Ice::InstIntrinsicCall::create(::function, 2, result, target, intrinsic);
- load->addArg(::context->getConstantInt32(8));
- load->addArg(ptr);
- ::basicBlock->appendInst(load);
+ auto pointer = RValue<Pointer<Byte>>(ptr);
+ Int x = *Pointer<Int>(pointer);
+ Int y = *Pointer<Int>(pointer + 4);
+
+ Int4 vector;
+ vector = Insert(vector, x, 0);
+ vector = Insert(vector, y, 1);
+
+ auto bitcast = Ice::InstCast::create(::function, Ice::InstCast::Bitcast, result, vector.loadValue());
+ ::basicBlock->appendInst(bitcast);
}
- break;
- default: assert(false && "UNIMPLEMENTED");
+ else assert(false);
+ }
+ else
+ {
+ const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::LoadSubVector, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
+ auto target = ::context->getConstantUndef(Ice::IceType_i32);
+ auto load = Ice::InstIntrinsicCall::create(::function, 2, result, target, intrinsic);
+ load->addArg(ptr);
+ load->addArg(::context->getConstantInt32(typeSize(type)));
+ ::basicBlock->appendInst(load);
}
}
else
return V(result);
}
- Value *Nucleus::createStore(Value *value, Value *ptr, Type *type, bool isVolatile, unsigned int align)
+ Value *Nucleus::createStore(Value *value, Value *ptr, Type *type, bool isVolatile, unsigned int align, bool atomic, std::memory_order memoryOrder)
{
+ assert(!atomic); // Unimplemented
+ assert(memoryOrder == std::memory_order_relaxed); // Unimplemented
+
+ #if __has_feature(memory_sanitizer)
+ // Mark all (non-stack) memory writes as initialized by calling __msan_unpoison
+ if(align != 0)
+ {
+ auto call = Ice::InstCall::create(::function, 2, nullptr, ::context->getConstantInt64(reinterpret_cast<intptr_t>(__msan_unpoison)), false);
+ call->addArg(ptr);
+ call->addArg(::context->getConstantInt64(typeSize(type)));
+ ::basicBlock->appendInst(call);
+ }
+ #endif
+
int valueType = (int)reinterpret_cast<intptr_t>(type);
- if(valueType & EmulatedBits)
+ if((valueType & EmulatedBits) && (align != 0)) // Narrow vector not stored on stack.
{
- switch(valueType)
+ if(emulateIntrinsics)
{
- case Type_v4i8:
- case Type_v2i16:
+ if(typeSize(type) == 4)
{
- const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::StoreSubVector, Ice::Intrinsics::SideEffects_T, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_T};
- auto target = ::context->getConstantUndef(Ice::IceType_i32);
- auto store = Ice::InstIntrinsicCall::create(::function, 3, nullptr, target, intrinsic);
- store->addArg(::context->getConstantInt32(4));
- store->addArg(value);
- store->addArg(ptr);
- ::basicBlock->appendInst(store);
+ Ice::Variable *vector = ::function->makeVariable(Ice::IceType_v4i32);
+ auto bitcast = Ice::InstCast::create(::function, Ice::InstCast::Bitcast, vector, value);
+ ::basicBlock->appendInst(bitcast);
+
+ RValue<Int4> v(V(vector));
+
+ auto pointer = RValue<Pointer<Byte>>(ptr);
+ Int x = Extract(v, 0);
+ *Pointer<Int>(pointer) = x;
}
- break;
- case Type_v2i32:
- case Type_v8i8:
- case Type_v4i16:
- case Type_v2f32:
+ else if(typeSize(type) == 8)
{
- const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::StoreSubVector, Ice::Intrinsics::SideEffects_T, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_T};
- auto target = ::context->getConstantUndef(Ice::IceType_i32);
- auto store = Ice::InstIntrinsicCall::create(::function, 3, nullptr, target, intrinsic);
- store->addArg(::context->getConstantInt32(8));
- store->addArg(value);
- store->addArg(ptr);
- ::basicBlock->appendInst(store);
+ Ice::Variable *vector = ::function->makeVariable(Ice::IceType_v4i32);
+ auto bitcast = Ice::InstCast::create(::function, Ice::InstCast::Bitcast, vector, value);
+ ::basicBlock->appendInst(bitcast);
+
+ RValue<Int4> v(V(vector));
+
+ auto pointer = RValue<Pointer<Byte>>(ptr);
+ Int x = Extract(v, 0);
+ *Pointer<Int>(pointer) = x;
+ Int y = Extract(v, 1);
+ *Pointer<Int>(pointer + 4) = y;
}
- break;
- default: assert(false && "UNIMPLEMENTED");
+ else assert(false);
+ }
+ else
+ {
+ const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::StoreSubVector, Ice::Intrinsics::SideEffects_T, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_T};
+ auto target = ::context->getConstantUndef(Ice::IceType_i32);
+ auto store = Ice::InstIntrinsicCall::create(::function, 3, nullptr, target, intrinsic);
+ store->addArg(value);
+ store->addArg(ptr);
+ store->addArg(::context->getConstantInt32(typeSize(type)));
+ ::basicBlock->appendInst(store);
}
}
else
{
- assert(T(value->getType()) == type);
+ assert(value->getType() == T(type));
auto store = Ice::InstStore::create(::function, value, ptr, align);
::basicBlock->appendInst(store);
return value;
}
- Value *Nucleus::createGEP(Value *ptr, Type *type, Value *index)
+ Value *Nucleus::createGEP(Value *ptr, Type *type, Value *index, bool unsignedIndex)
{
assert(index->getType() == Ice::IceType_i32);
+ if(auto *constant = llvm::dyn_cast<Ice::ConstantInteger32>(index))
+ {
+ int32_t offset = constant->getValue() * (int)typeSize(type);
+
+ if(offset == 0)
+ {
+ return ptr;
+ }
+
+ return createAdd(ptr, createConstantInt(offset));
+ }
+
if(!Ice::isByteSizedType(T(type)))
{
- index = createMul(index, createConstantInt((int)Ice::typeWidthInBytes(T(type))));
+ index = createMul(index, createConstantInt((int)typeSize(type)));
}
if(sizeof(void*) == 8)
{
- index = createSExt(index, T(Ice::IceType_i64));
+ if(unsignedIndex)
+ {
+ index = createZExt(index, T(Ice::IceType_i64));
+ }
+ else
+ {
+ index = createSExt(index, T(Ice::IceType_i64));
+ }
}
return createAdd(ptr, index);
return createCast(Ice::InstCast::Fptosi, v, destType);
}
- Value *Nucleus::createUIToFP(Value *v, Type *destType)
- {
- return createCast(Ice::InstCast::Uitofp, v, destType);
- }
-
Value *Nucleus::createSIToFP(Value *v, Type *destType)
{
return createCast(Ice::InstCast::Sitofp, v, destType);
Value *Nucleus::createBitCast(Value *v, Type *destType)
{
- return createCast(Ice::InstCast::Bitcast, v, destType);
- }
+ // Bitcasts must be between types of the same logical size. But with emulated narrow vectors we need
+ // support for casting between scalars and wide vectors. For platforms where this is not supported,
+ // emulate them by writing to the stack and reading back as the destination type.
+ if(emulateMismatchedBitCast)
+ {
+ if(!Ice::isVectorType(v->getType()) && Ice::isVectorType(T(destType)))
+ {
+ Value *address = allocateStackVariable(destType);
+ createStore(v, address, T(v->getType()));
+ return createLoad(address, destType);
+ }
+ else if(Ice::isVectorType(v->getType()) && !Ice::isVectorType(T(destType)))
+ {
+ Value *address = allocateStackVariable(T(v->getType()));
+ createStore(v, address, T(v->getType()));
+ return createLoad(address, destType);
+ }
+ }
- Value *Nucleus::createIntCast(Value *v, Type *destType, bool isSigned)
- {
- assert(false && "UNIMPLEMENTED"); return nullptr;
+ return createCast(Ice::InstCast::Bitcast, v, destType);
}
static Value *createIntCompare(Ice::InstIcmp::ICond condition, Value *lhs, Value *rhs)
return V(result);
}
- Value *Nucleus::createSwitch(Value *v, BasicBlock *Dest, unsigned NumCases)
+ SwitchCases *Nucleus::createSwitch(Value *control, BasicBlock *defaultBranch, unsigned numCases)
{
- assert(false && "UNIMPLEMENTED"); return nullptr;
+ auto switchInst = Ice::InstSwitch::create(::function, numCases, control, defaultBranch);
+ ::basicBlock->appendInst(switchInst);
+
+ return reinterpret_cast<SwitchCases*>(switchInst);
}
- void Nucleus::addSwitchCase(Value *Switch, int Case, BasicBlock *Branch)
+ void Nucleus::addSwitchCase(SwitchCases *switchCases, int label, BasicBlock *branch)
{
- assert(false && "UNIMPLEMENTED"); return;
+ switchCases->addBranch(label, label, branch);
}
void Nucleus::createUnreachable()
::basicBlock->appendInst(unreachable);
}
- static Value *createSwizzle4(Value *val, unsigned char select)
- {
- int swizzle[4] =
- {
- (select >> 0) & 0x03,
- (select >> 2) & 0x03,
- (select >> 4) & 0x03,
- (select >> 6) & 0x03,
- };
-
- return Nucleus::createShuffleVector(val, val, swizzle);
- }
-
- static Value *createMask4(Value *lhs, Value *rhs, unsigned char select)
- {
- assert(false && "UNIMPLEMENTED"); return nullptr;
- }
-
- Value *Nucleus::createConstantPointer(const void *address, Type *Ty, unsigned int align)
- {
- if(sizeof(void*) == 8)
- {
- return createAssign(::context->getConstantInt64(reinterpret_cast<intptr_t>(address)));
- }
- else
- {
- return createAssign(::context->getConstantInt32(reinterpret_cast<intptr_t>(address)));
- }
- }
-
Type *Nucleus::getPointerType(Type *ElementType)
{
if(sizeof(void*) == 8)
{
if(Ice::isVectorType(T(Ty)))
{
- int64_t c[4] = {0, 0, 0, 0};
+ assert(Ice::typeNumElements(T(Ty)) <= 16);
+ int64_t c[16] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
return createConstantVector(c, Ty);
}
else
{
- return createAssign(::context->getConstantZero(T(Ty)));
+ return V(::context->getConstantZero(T(Ty)));
}
}
Value *Nucleus::createConstantLong(int64_t i)
{
- return createAssign(::context->getConstantInt64(i));
+ return V(::context->getConstantInt64(i));
}
Value *Nucleus::createConstantInt(int i)
{
- return createAssign(::context->getConstantInt32(i));
+ return V(::context->getConstantInt32(i));
}
Value *Nucleus::createConstantInt(unsigned int i)
{
- return createAssign(::context->getConstantInt32(i));
+ return V(::context->getConstantInt32(i));
}
Value *Nucleus::createConstantBool(bool b)
{
- return createAssign(::context->getConstantInt1(b));
+ return V(::context->getConstantInt1(b));
}
Value *Nucleus::createConstantByte(signed char i)
{
- return createAssign(::context->getConstantInt8(i));
+ return V(::context->getConstantInt8(i));
}
Value *Nucleus::createConstantByte(unsigned char i)
{
- return createAssign(::context->getConstantInt8(i));
+ return V(::context->getConstantInt8(i));
}
Value *Nucleus::createConstantShort(short i)
{
- return createAssign(::context->getConstantInt16(i));
+ return V(::context->getConstantInt16(i));
}
Value *Nucleus::createConstantShort(unsigned short i)
{
- return createAssign(::context->getConstantInt16(i));
+ return V(::context->getConstantInt16(i));
}
Value *Nucleus::createConstantFloat(float x)
{
- return createAssign(::context->getConstantFloat(x));
+ return V(::context->getConstantFloat(x));
}
Value *Nucleus::createNullPointer(Type *Ty)
{
- if(true)
- {
- return createNullValue(T(sizeof(void*) == 8 ? Ice::IceType_i64 : Ice::IceType_i32));
- }
- else
- {
- return createConstantPointer(nullptr, Ty);
- }
+ return createNullValue(T(sizeof(void*) == 8 ? Ice::IceType_i64 : Ice::IceType_i32));
}
Value *Nucleus::createConstantVector(const int64_t *constants, Type *type)
switch((int)reinterpret_cast<intptr_t>(type))
{
case Ice::IceType_v4i32:
+ case Ice::IceType_v4i1:
{
const int initializer[4] = {(int)i[0], (int)i[1], (int)i[2], (int)i[3]};
static_assert(sizeof(initializer) == vectorSize, "!");
}
break;
case Ice::IceType_v8i16:
+ case Ice::IceType_v8i1:
{
const short initializer[8] = {(short)i[0], (short)i[1], (short)i[2], (short)i[3], (short)i[4], (short)i[5], (short)i[6], (short)i[7]};
static_assert(sizeof(initializer) == vectorSize, "!");
}
break;
case Ice::IceType_v16i8:
+ case Ice::IceType_v16i1:
{
const char initializer[16] = {(char)i[0], (char)i[1], (char)i[2], (char)i[3], (char)i[4], (char)i[5], (char)i[6], (char)i[7], (char)i[8], (char)i[9], (char)i[10], (char)i[11], (char)i[12], (char)i[13], (char)i[14], (char)i[15]};
static_assert(sizeof(initializer) == vectorSize, "!");
variableDeclaration->setAlignment(alignment);
variableDeclaration->setIsConstant(true);
variableDeclaration->addInitializer(dataInitializer);
-
+
::function->addGlobal(variableDeclaration);
constexpr int32_t offset = 0;
return T(Ice::IceType_void);
}
- Bool::Bool(Argument<Bool> argument)
+ Type *Bool::getType()
{
- storeValue(argument.value);
+ return T(Ice::IceType_i1);
}
- Bool::Bool()
+ Type *Byte::getType()
{
+ return T(Ice::IceType_i8);
}
- Bool::Bool(bool x)
+ Type *SByte::getType()
{
- storeValue(Nucleus::createConstantBool(x));
+ return T(Ice::IceType_i8);
}
- Bool::Bool(RValue<Bool> rhs)
+ Type *Short::getType()
{
- storeValue(rhs.value);
+ return T(Ice::IceType_i16);
}
- Bool::Bool(const Bool &rhs)
+ Type *UShort::getType()
{
- Value *value = rhs.loadValue();
- storeValue(value);
+ return T(Ice::IceType_i16);
}
- Bool::Bool(const Reference<Bool> &rhs)
+ Type *Byte4::getType()
{
- Value *value = rhs.loadValue();
- storeValue(value);
+ return T(Type_v4i8);
}
- RValue<Bool> Bool::operator=(RValue<Bool> rhs) const
+ Type *SByte4::getType()
{
- storeValue(rhs.value);
-
- return rhs;
+ return T(Type_v4i8);
}
- RValue<Bool> Bool::operator=(const Bool &rhs) const
+ namespace
{
- Value *value = rhs.loadValue();
- storeValue(value);
+ RValue<Byte> SaturateUnsigned(RValue<Short> x)
+ {
+ return Byte(IfThenElse(Int(x) > 0xFF, Int(0xFF), IfThenElse(Int(x) < 0, Int(0), Int(x))));
+ }
- return RValue<Bool>(value);
+ RValue<Byte> Extract(RValue<Byte8> val, int i)
+ {
+ return RValue<Byte>(Nucleus::createExtractElement(val.value, Byte::getType(), i));
+ }
+
+ RValue<Byte8> Insert(RValue<Byte8> val, RValue<Byte> element, int i)
+ {
+ return RValue<Byte8>(Nucleus::createInsertElement(val.value, element.value, i));
+ }
}
- RValue<Bool> Bool::operator=(const Reference<Bool> &rhs) const
+ RValue<Byte8> AddSat(RValue<Byte8> x, RValue<Byte8> y)
{
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<Bool>(value);
+ if(emulateIntrinsics)
+ {
+ Byte8 result;
+ result = Insert(result, SaturateUnsigned(Short(Int(Extract(x, 0)) + Int(Extract(y, 0)))), 0);
+ result = Insert(result, SaturateUnsigned(Short(Int(Extract(x, 1)) + Int(Extract(y, 1)))), 1);
+ result = Insert(result, SaturateUnsigned(Short(Int(Extract(x, 2)) + Int(Extract(y, 2)))), 2);
+ result = Insert(result, SaturateUnsigned(Short(Int(Extract(x, 3)) + Int(Extract(y, 3)))), 3);
+ result = Insert(result, SaturateUnsigned(Short(Int(Extract(x, 4)) + Int(Extract(y, 4)))), 4);
+ result = Insert(result, SaturateUnsigned(Short(Int(Extract(x, 5)) + Int(Extract(y, 5)))), 5);
+ result = Insert(result, SaturateUnsigned(Short(Int(Extract(x, 6)) + Int(Extract(y, 6)))), 6);
+ result = Insert(result, SaturateUnsigned(Short(Int(Extract(x, 7)) + Int(Extract(y, 7)))), 7);
+
+ return result;
+ }
+ else
+ {
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_v16i8);
+ const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::AddSaturateUnsigned, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
+ auto target = ::context->getConstantUndef(Ice::IceType_i32);
+ auto paddusb = Ice::InstIntrinsicCall::create(::function, 2, result, target, intrinsic);
+ paddusb->addArg(x.value);
+ paddusb->addArg(y.value);
+ ::basicBlock->appendInst(paddusb);
+
+ return RValue<Byte8>(V(result));
+ }
}
- RValue<Bool> operator!(RValue<Bool> val)
+ RValue<Byte8> SubSat(RValue<Byte8> x, RValue<Byte8> y)
{
- return RValue<Bool>(Nucleus::createNot(val.value));
+ if(emulateIntrinsics)
+ {
+ Byte8 result;
+ result = Insert(result, SaturateUnsigned(Short(Int(Extract(x, 0)) - Int(Extract(y, 0)))), 0);
+ result = Insert(result, SaturateUnsigned(Short(Int(Extract(x, 1)) - Int(Extract(y, 1)))), 1);
+ result = Insert(result, SaturateUnsigned(Short(Int(Extract(x, 2)) - Int(Extract(y, 2)))), 2);
+ result = Insert(result, SaturateUnsigned(Short(Int(Extract(x, 3)) - Int(Extract(y, 3)))), 3);
+ result = Insert(result, SaturateUnsigned(Short(Int(Extract(x, 4)) - Int(Extract(y, 4)))), 4);
+ result = Insert(result, SaturateUnsigned(Short(Int(Extract(x, 5)) - Int(Extract(y, 5)))), 5);
+ result = Insert(result, SaturateUnsigned(Short(Int(Extract(x, 6)) - Int(Extract(y, 6)))), 6);
+ result = Insert(result, SaturateUnsigned(Short(Int(Extract(x, 7)) - Int(Extract(y, 7)))), 7);
+
+ return result;
+ }
+ else
+ {
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_v16i8);
+ const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::SubtractSaturateUnsigned, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
+ auto target = ::context->getConstantUndef(Ice::IceType_i32);
+ auto psubusw = Ice::InstIntrinsicCall::create(::function, 2, result, target, intrinsic);
+ psubusw->addArg(x.value);
+ psubusw->addArg(y.value);
+ ::basicBlock->appendInst(psubusw);
+
+ return RValue<Byte8>(V(result));
+ }
}
- RValue<Bool> operator&&(RValue<Bool> lhs, RValue<Bool> rhs)
+ RValue<SByte> Extract(RValue<SByte8> val, int i)
{
- return RValue<Bool>(Nucleus::createAnd(lhs.value, rhs.value));
+ return RValue<SByte>(Nucleus::createExtractElement(val.value, SByte::getType(), i));
}
- RValue<Bool> operator||(RValue<Bool> lhs, RValue<Bool> rhs)
+ RValue<SByte8> Insert(RValue<SByte8> val, RValue<SByte> element, int i)
{
- return RValue<Bool>(Nucleus::createOr(lhs.value, rhs.value));
+ return RValue<SByte8>(Nucleus::createInsertElement(val.value, element.value, i));
}
- Type *Bool::getType()
+ RValue<SByte8> operator>>(RValue<SByte8> lhs, unsigned char rhs)
{
- return T(Ice::IceType_i1);
+ if(emulateIntrinsics)
+ {
+ SByte8 result;
+ result = Insert(result, Extract(lhs, 0) >> SByte(rhs), 0);
+ result = Insert(result, Extract(lhs, 1) >> SByte(rhs), 1);
+ result = Insert(result, Extract(lhs, 2) >> SByte(rhs), 2);
+ result = Insert(result, Extract(lhs, 3) >> SByte(rhs), 3);
+ result = Insert(result, Extract(lhs, 4) >> SByte(rhs), 4);
+ result = Insert(result, Extract(lhs, 5) >> SByte(rhs), 5);
+ result = Insert(result, Extract(lhs, 6) >> SByte(rhs), 6);
+ result = Insert(result, Extract(lhs, 7) >> SByte(rhs), 7);
+
+ return result;
+ }
+ else
+ {
+ #if defined(__i386__) || defined(__x86_64__)
+ // SSE2 doesn't support byte vector shifts, so shift as shorts and recombine.
+ RValue<Short4> hi = (As<Short4>(lhs) >> rhs) & Short4(0xFF00u);
+ RValue<Short4> lo = As<Short4>(As<UShort4>((As<Short4>(lhs) << 8) >> rhs) >> 8);
+
+ return As<SByte8>(hi | lo);
+ #else
+ return RValue<SByte8>(Nucleus::createAShr(lhs.value, V(::context->getConstantInt32(rhs))));
+ #endif
+ }
}
- Byte::Byte(Argument<Byte> argument)
+ RValue<Int> SignMask(RValue<Byte8> x)
{
- storeValue(argument.value);
+ if(emulateIntrinsics || CPUID::ARM)
+ {
+ Byte8 xx = As<Byte8>(As<SByte8>(x) >> 7) & Byte8(0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80);
+ return Int(Extract(xx, 0)) | Int(Extract(xx, 1)) | Int(Extract(xx, 2)) | Int(Extract(xx, 3)) | Int(Extract(xx, 4)) | Int(Extract(xx, 5)) | Int(Extract(xx, 6)) | Int(Extract(xx, 7));
+ }
+ else
+ {
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_i32);
+ const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::SignMask, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
+ auto target = ::context->getConstantUndef(Ice::IceType_i32);
+ auto movmsk = Ice::InstIntrinsicCall::create(::function, 1, result, target, intrinsic);
+ movmsk->addArg(x.value);
+ ::basicBlock->appendInst(movmsk);
+
+ return RValue<Int>(V(result)) & 0xFF;
+ }
}
- Byte::Byte(RValue<Int> cast)
- {
- Value *integer = Nucleus::createTrunc(cast.value, Byte::getType());
+// RValue<Byte8> CmpGT(RValue<Byte8> x, RValue<Byte8> y)
+// {
+// return RValue<Byte8>(createIntCompare(Ice::InstIcmp::Ugt, x.value, y.value));
+// }
- storeValue(integer);
+ RValue<Byte8> CmpEQ(RValue<Byte8> x, RValue<Byte8> y)
+ {
+ return RValue<Byte8>(Nucleus::createICmpEQ(x.value, y.value));
}
- Byte::Byte(RValue<UInt> cast)
+ Type *Byte8::getType()
{
- Value *integer = Nucleus::createTrunc(cast.value, Byte::getType());
-
- storeValue(integer);
+ return T(Type_v8i8);
}
- Byte::Byte(RValue<UShort> cast)
- {
- Value *integer = Nucleus::createTrunc(cast.value, Byte::getType());
+// RValue<SByte8> operator<<(RValue<SByte8> lhs, unsigned char rhs)
+// {
+// return RValue<SByte8>(Nucleus::createShl(lhs.value, V(::context->getConstantInt32(rhs))));
+// }
- storeValue(integer);
- }
+// RValue<SByte8> operator>>(RValue<SByte8> lhs, unsigned char rhs)
+// {
+// return RValue<SByte8>(Nucleus::createAShr(lhs.value, V(::context->getConstantInt32(rhs))));
+// }
- Byte::Byte()
+ RValue<SByte> SaturateSigned(RValue<Short> x)
{
+ return SByte(IfThenElse(Int(x) > 0x7F, Int(0x7F), IfThenElse(Int(x) < -0x80, Int(0x80), Int(x))));
}
- Byte::Byte(int x)
+ RValue<SByte8> AddSat(RValue<SByte8> x, RValue<SByte8> y)
{
- storeValue(Nucleus::createConstantByte((unsigned char)x));
+ if(emulateIntrinsics)
+ {
+ SByte8 result;
+ result = Insert(result, SaturateSigned(Short(Int(Extract(x, 0)) + Int(Extract(y, 0)))), 0);
+ result = Insert(result, SaturateSigned(Short(Int(Extract(x, 1)) + Int(Extract(y, 1)))), 1);
+ result = Insert(result, SaturateSigned(Short(Int(Extract(x, 2)) + Int(Extract(y, 2)))), 2);
+ result = Insert(result, SaturateSigned(Short(Int(Extract(x, 3)) + Int(Extract(y, 3)))), 3);
+ result = Insert(result, SaturateSigned(Short(Int(Extract(x, 4)) + Int(Extract(y, 4)))), 4);
+ result = Insert(result, SaturateSigned(Short(Int(Extract(x, 5)) + Int(Extract(y, 5)))), 5);
+ result = Insert(result, SaturateSigned(Short(Int(Extract(x, 6)) + Int(Extract(y, 6)))), 6);
+ result = Insert(result, SaturateSigned(Short(Int(Extract(x, 7)) + Int(Extract(y, 7)))), 7);
+
+ return result;
+ }
+ else
+ {
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_v16i8);
+ const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::AddSaturateSigned, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
+ auto target = ::context->getConstantUndef(Ice::IceType_i32);
+ auto paddsb = Ice::InstIntrinsicCall::create(::function, 2, result, target, intrinsic);
+ paddsb->addArg(x.value);
+ paddsb->addArg(y.value);
+ ::basicBlock->appendInst(paddsb);
+
+ return RValue<SByte8>(V(result));
+ }
}
- Byte::Byte(unsigned char x)
+ RValue<SByte8> SubSat(RValue<SByte8> x, RValue<SByte8> y)
{
- storeValue(Nucleus::createConstantByte(x));
- }
-
- Byte::Byte(RValue<Byte> rhs)
- {
- storeValue(rhs.value);
+ if(emulateIntrinsics)
+ {
+ SByte8 result;
+ result = Insert(result, SaturateSigned(Short(Int(Extract(x, 0)) - Int(Extract(y, 0)))), 0);
+ result = Insert(result, SaturateSigned(Short(Int(Extract(x, 1)) - Int(Extract(y, 1)))), 1);
+ result = Insert(result, SaturateSigned(Short(Int(Extract(x, 2)) - Int(Extract(y, 2)))), 2);
+ result = Insert(result, SaturateSigned(Short(Int(Extract(x, 3)) - Int(Extract(y, 3)))), 3);
+ result = Insert(result, SaturateSigned(Short(Int(Extract(x, 4)) - Int(Extract(y, 4)))), 4);
+ result = Insert(result, SaturateSigned(Short(Int(Extract(x, 5)) - Int(Extract(y, 5)))), 5);
+ result = Insert(result, SaturateSigned(Short(Int(Extract(x, 6)) - Int(Extract(y, 6)))), 6);
+ result = Insert(result, SaturateSigned(Short(Int(Extract(x, 7)) - Int(Extract(y, 7)))), 7);
+
+ return result;
+ }
+ else
+ {
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_v16i8);
+ const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::SubtractSaturateSigned, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
+ auto target = ::context->getConstantUndef(Ice::IceType_i32);
+ auto psubsb = Ice::InstIntrinsicCall::create(::function, 2, result, target, intrinsic);
+ psubsb->addArg(x.value);
+ psubsb->addArg(y.value);
+ ::basicBlock->appendInst(psubsb);
+
+ return RValue<SByte8>(V(result));
+ }
}
- Byte::Byte(const Byte &rhs)
+ RValue<Int> SignMask(RValue<SByte8> x)
{
- Value *value = rhs.loadValue();
- storeValue(value);
+ if(emulateIntrinsics || CPUID::ARM)
+ {
+ SByte8 xx = (x >> 7) & SByte8(0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80);
+ return Int(Extract(xx, 0)) | Int(Extract(xx, 1)) | Int(Extract(xx, 2)) | Int(Extract(xx, 3)) | Int(Extract(xx, 4)) | Int(Extract(xx, 5)) | Int(Extract(xx, 6)) | Int(Extract(xx, 7));
+ }
+ else
+ {
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_i32);
+ const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::SignMask, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
+ auto target = ::context->getConstantUndef(Ice::IceType_i32);
+ auto movmsk = Ice::InstIntrinsicCall::create(::function, 1, result, target, intrinsic);
+ movmsk->addArg(x.value);
+ ::basicBlock->appendInst(movmsk);
+
+ return RValue<Int>(V(result)) & 0xFF;
+ }
}
- Byte::Byte(const Reference<Byte> &rhs)
+ RValue<Byte8> CmpGT(RValue<SByte8> x, RValue<SByte8> y)
{
- Value *value = rhs.loadValue();
- storeValue(value);
+ return RValue<Byte8>(createIntCompare(Ice::InstIcmp::Sgt, x.value, y.value));
}
- RValue<Byte> Byte::operator=(RValue<Byte> rhs) const
+ RValue<Byte8> CmpEQ(RValue<SByte8> x, RValue<SByte8> y)
{
- storeValue(rhs.value);
-
- return rhs;
+ return RValue<Byte8>(Nucleus::createICmpEQ(x.value, y.value));
}
- RValue<Byte> Byte::operator=(const Byte &rhs) const
+ Type *SByte8::getType()
{
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<Byte>(value);
+ return T(Type_v8i8);
}
- RValue<Byte> Byte::operator=(const Reference<Byte> &rhs) const
+ Type *Byte16::getType()
{
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<Byte>(value);
+ return T(Ice::IceType_v16i8);
}
- RValue<Byte> operator+(RValue<Byte> lhs, RValue<Byte> rhs)
+ Type *SByte16::getType()
{
- return RValue<Byte>(Nucleus::createAdd(lhs.value, rhs.value));
+ return T(Ice::IceType_v16i8);
}
- RValue<Byte> operator-(RValue<Byte> lhs, RValue<Byte> rhs)
+ Type *Short2::getType()
{
- return RValue<Byte>(Nucleus::createSub(lhs.value, rhs.value));
+ return T(Type_v2i16);
}
- RValue<Byte> operator*(RValue<Byte> lhs, RValue<Byte> rhs)
+ Type *UShort2::getType()
{
- return RValue<Byte>(Nucleus::createMul(lhs.value, rhs.value));
+ return T(Type_v2i16);
}
- RValue<Byte> operator/(RValue<Byte> lhs, RValue<Byte> rhs)
+ Short4::Short4(RValue<Int4> cast)
{
- return RValue<Byte>(Nucleus::createUDiv(lhs.value, rhs.value));
- }
+ int select[8] = {0, 2, 4, 6, 0, 2, 4, 6};
+ Value *short8 = Nucleus::createBitCast(cast.value, Short8::getType());
+ Value *packed = Nucleus::createShuffleVector(short8, short8, select);
- RValue<Byte> operator%(RValue<Byte> lhs, RValue<Byte> rhs)
- {
- return RValue<Byte>(Nucleus::createURem(lhs.value, rhs.value));
+ Value *int2 = RValue<Int2>(Int2(As<Int4>(packed))).value;
+ Value *short4 = Nucleus::createBitCast(int2, Short4::getType());
+
+ storeValue(short4);
}
- RValue<Byte> operator&(RValue<Byte> lhs, RValue<Byte> rhs)
+// Short4::Short4(RValue<Float> cast)
+// {
+// }
+
+ Short4::Short4(RValue<Float4> cast)
{
- return RValue<Byte>(Nucleus::createAnd(lhs.value, rhs.value));
+ assert(false && "UNIMPLEMENTED");
}
- RValue<Byte> operator|(RValue<Byte> lhs, RValue<Byte> rhs)
+ RValue<Short4> operator<<(RValue<Short4> lhs, unsigned char rhs)
{
- return RValue<Byte>(Nucleus::createOr(lhs.value, rhs.value));
+ if(emulateIntrinsics)
+ {
+ Short4 result;
+ result = Insert(result, Extract(lhs, 0) << Short(rhs), 0);
+ result = Insert(result, Extract(lhs, 1) << Short(rhs), 1);
+ result = Insert(result, Extract(lhs, 2) << Short(rhs), 2);
+ result = Insert(result, Extract(lhs, 3) << Short(rhs), 3);
+
+ return result;
+ }
+ else
+ {
+ return RValue<Short4>(Nucleus::createShl(lhs.value, V(::context->getConstantInt32(rhs))));
+ }
}
- RValue<Byte> operator^(RValue<Byte> lhs, RValue<Byte> rhs)
+ RValue<Short4> operator>>(RValue<Short4> lhs, unsigned char rhs)
{
- return RValue<Byte>(Nucleus::createXor(lhs.value, rhs.value));
+ if(emulateIntrinsics)
+ {
+ Short4 result;
+ result = Insert(result, Extract(lhs, 0) >> Short(rhs), 0);
+ result = Insert(result, Extract(lhs, 1) >> Short(rhs), 1);
+ result = Insert(result, Extract(lhs, 2) >> Short(rhs), 2);
+ result = Insert(result, Extract(lhs, 3) >> Short(rhs), 3);
+
+ return result;
+ }
+ else
+ {
+ return RValue<Short4>(Nucleus::createAShr(lhs.value, V(::context->getConstantInt32(rhs))));
+ }
}
- RValue<Byte> operator<<(RValue<Byte> lhs, RValue<Byte> rhs)
+ RValue<Short4> Max(RValue<Short4> x, RValue<Short4> y)
{
- return RValue<Byte>(Nucleus::createShl(lhs.value, rhs.value));
+ Ice::Variable *condition = ::function->makeVariable(Ice::IceType_v8i1);
+ auto cmp = Ice::InstIcmp::create(::function, Ice::InstIcmp::Sle, condition, x.value, y.value);
+ ::basicBlock->appendInst(cmp);
+
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_v8i16);
+ auto select = Ice::InstSelect::create(::function, result, condition, y.value, x.value);
+ ::basicBlock->appendInst(select);
+
+ return RValue<Short4>(V(result));
}
- RValue<Byte> operator>>(RValue<Byte> lhs, RValue<Byte> rhs)
+ RValue<Short4> Min(RValue<Short4> x, RValue<Short4> y)
{
- return RValue<Byte>(Nucleus::createLShr(lhs.value, rhs.value));
+ Ice::Variable *condition = ::function->makeVariable(Ice::IceType_v8i1);
+ auto cmp = Ice::InstIcmp::create(::function, Ice::InstIcmp::Sgt, condition, x.value, y.value);
+ ::basicBlock->appendInst(cmp);
+
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_v8i16);
+ auto select = Ice::InstSelect::create(::function, result, condition, y.value, x.value);
+ ::basicBlock->appendInst(select);
+
+ return RValue<Short4>(V(result));
}
- RValue<Byte> operator+=(const Byte &lhs, RValue<Byte> rhs)
+ RValue<Short> SaturateSigned(RValue<Int> x)
{
- return lhs = lhs + rhs;
+ return Short(IfThenElse(x > 0x7FFF, Int(0x7FFF), IfThenElse(x < -0x8000, Int(0x8000), x)));
}
- RValue<Byte> operator-=(const Byte &lhs, RValue<Byte> rhs)
+ RValue<Short4> AddSat(RValue<Short4> x, RValue<Short4> y)
{
- return lhs = lhs - rhs;
+ if(emulateIntrinsics)
+ {
+ Short4 result;
+ result = Insert(result, SaturateSigned(Int(Extract(x, 0)) + Int(Extract(y, 0))), 0);
+ result = Insert(result, SaturateSigned(Int(Extract(x, 1)) + Int(Extract(y, 1))), 1);
+ result = Insert(result, SaturateSigned(Int(Extract(x, 2)) + Int(Extract(y, 2))), 2);
+ result = Insert(result, SaturateSigned(Int(Extract(x, 3)) + Int(Extract(y, 3))), 3);
+
+ return result;
+ }
+ else
+ {
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_v8i16);
+ const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::AddSaturateSigned, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
+ auto target = ::context->getConstantUndef(Ice::IceType_i32);
+ auto paddsw = Ice::InstIntrinsicCall::create(::function, 2, result, target, intrinsic);
+ paddsw->addArg(x.value);
+ paddsw->addArg(y.value);
+ ::basicBlock->appendInst(paddsw);
+
+ return RValue<Short4>(V(result));
+ }
}
- RValue<Byte> operator*=(const Byte &lhs, RValue<Byte> rhs)
+ RValue<Short4> SubSat(RValue<Short4> x, RValue<Short4> y)
{
- return lhs = lhs * rhs;
+ if(emulateIntrinsics)
+ {
+ Short4 result;
+ result = Insert(result, SaturateSigned(Int(Extract(x, 0)) - Int(Extract(y, 0))), 0);
+ result = Insert(result, SaturateSigned(Int(Extract(x, 1)) - Int(Extract(y, 1))), 1);
+ result = Insert(result, SaturateSigned(Int(Extract(x, 2)) - Int(Extract(y, 2))), 2);
+ result = Insert(result, SaturateSigned(Int(Extract(x, 3)) - Int(Extract(y, 3))), 3);
+
+ return result;
+ }
+ else
+ {
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_v8i16);
+ const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::SubtractSaturateSigned, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
+ auto target = ::context->getConstantUndef(Ice::IceType_i32);
+ auto psubsw = Ice::InstIntrinsicCall::create(::function, 2, result, target, intrinsic);
+ psubsw->addArg(x.value);
+ psubsw->addArg(y.value);
+ ::basicBlock->appendInst(psubsw);
+
+ return RValue<Short4>(V(result));
+ }
}
- RValue<Byte> operator/=(const Byte &lhs, RValue<Byte> rhs)
+ RValue<Short4> MulHigh(RValue<Short4> x, RValue<Short4> y)
{
- return lhs = lhs / rhs;
+ if(emulateIntrinsics)
+ {
+ Short4 result;
+ result = Insert(result, Short((Int(Extract(x, 0)) * Int(Extract(y, 0))) >> 16), 0);
+ result = Insert(result, Short((Int(Extract(x, 1)) * Int(Extract(y, 1))) >> 16), 1);
+ result = Insert(result, Short((Int(Extract(x, 2)) * Int(Extract(y, 2))) >> 16), 2);
+ result = Insert(result, Short((Int(Extract(x, 3)) * Int(Extract(y, 3))) >> 16), 3);
+
+ return result;
+ }
+ else
+ {
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_v8i16);
+ const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::MultiplyHighSigned, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
+ auto target = ::context->getConstantUndef(Ice::IceType_i32);
+ auto pmulhw = Ice::InstIntrinsicCall::create(::function, 2, result, target, intrinsic);
+ pmulhw->addArg(x.value);
+ pmulhw->addArg(y.value);
+ ::basicBlock->appendInst(pmulhw);
+
+ return RValue<Short4>(V(result));
+ }
}
- RValue<Byte> operator%=(const Byte &lhs, RValue<Byte> rhs)
+ RValue<Int2> MulAdd(RValue<Short4> x, RValue<Short4> y)
{
- return lhs = lhs % rhs;
+ if(emulateIntrinsics)
+ {
+ Int2 result;
+ result = Insert(result, Int(Extract(x, 0)) * Int(Extract(y, 0)) + Int(Extract(x, 1)) * Int(Extract(y, 1)), 0);
+ result = Insert(result, Int(Extract(x, 2)) * Int(Extract(y, 2)) + Int(Extract(x, 3)) * Int(Extract(y, 3)), 1);
+
+ return result;
+ }
+ else
+ {
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_v8i16);
+ const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::MultiplyAddPairs, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
+ auto target = ::context->getConstantUndef(Ice::IceType_i32);
+ auto pmaddwd = Ice::InstIntrinsicCall::create(::function, 2, result, target, intrinsic);
+ pmaddwd->addArg(x.value);
+ pmaddwd->addArg(y.value);
+ ::basicBlock->appendInst(pmaddwd);
+
+ return As<Int2>(V(result));
+ }
}
- RValue<Byte> operator&=(const Byte &lhs, RValue<Byte> rhs)
+ RValue<SByte8> PackSigned(RValue<Short4> x, RValue<Short4> y)
{
- return lhs = lhs & rhs;
+ if(emulateIntrinsics)
+ {
+ SByte8 result;
+ result = Insert(result, SaturateSigned(Extract(x, 0)), 0);
+ result = Insert(result, SaturateSigned(Extract(x, 1)), 1);
+ result = Insert(result, SaturateSigned(Extract(x, 2)), 2);
+ result = Insert(result, SaturateSigned(Extract(x, 3)), 3);
+ result = Insert(result, SaturateSigned(Extract(y, 0)), 4);
+ result = Insert(result, SaturateSigned(Extract(y, 1)), 5);
+ result = Insert(result, SaturateSigned(Extract(y, 2)), 6);
+ result = Insert(result, SaturateSigned(Extract(y, 3)), 7);
+
+ return result;
+ }
+ else
+ {
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_v16i8);
+ const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::VectorPackSigned, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
+ auto target = ::context->getConstantUndef(Ice::IceType_i32);
+ auto pack = Ice::InstIntrinsicCall::create(::function, 2, result, target, intrinsic);
+ pack->addArg(x.value);
+ pack->addArg(y.value);
+ ::basicBlock->appendInst(pack);
+
+ return As<SByte8>(Swizzle(As<Int4>(V(result)), 0x88));
+ }
}
- RValue<Byte> operator|=(const Byte &lhs, RValue<Byte> rhs)
+ RValue<Byte8> PackUnsigned(RValue<Short4> x, RValue<Short4> y)
{
- return lhs = lhs | rhs;
+ if(emulateIntrinsics)
+ {
+ Byte8 result;
+ result = Insert(result, SaturateUnsigned(Extract(x, 0)), 0);
+ result = Insert(result, SaturateUnsigned(Extract(x, 1)), 1);
+ result = Insert(result, SaturateUnsigned(Extract(x, 2)), 2);
+ result = Insert(result, SaturateUnsigned(Extract(x, 3)), 3);
+ result = Insert(result, SaturateUnsigned(Extract(y, 0)), 4);
+ result = Insert(result, SaturateUnsigned(Extract(y, 1)), 5);
+ result = Insert(result, SaturateUnsigned(Extract(y, 2)), 6);
+ result = Insert(result, SaturateUnsigned(Extract(y, 3)), 7);
+
+ return result;
+ }
+ else
+ {
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_v16i8);
+ const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::VectorPackUnsigned, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
+ auto target = ::context->getConstantUndef(Ice::IceType_i32);
+ auto pack = Ice::InstIntrinsicCall::create(::function, 2, result, target, intrinsic);
+ pack->addArg(x.value);
+ pack->addArg(y.value);
+ ::basicBlock->appendInst(pack);
+
+ return As<Byte8>(Swizzle(As<Int4>(V(result)), 0x88));
+ }
}
- RValue<Byte> operator^=(const Byte &lhs, RValue<Byte> rhs)
+ RValue<Short4> CmpGT(RValue<Short4> x, RValue<Short4> y)
{
- return lhs = lhs ^ rhs;
+ return RValue<Short4>(createIntCompare(Ice::InstIcmp::Sgt, x.value, y.value));
}
- RValue<Byte> operator<<=(const Byte &lhs, RValue<Byte> rhs)
+ RValue<Short4> CmpEQ(RValue<Short4> x, RValue<Short4> y)
{
- return lhs = lhs << rhs;
+ return RValue<Short4>(Nucleus::createICmpEQ(x.value, y.value));
}
- RValue<Byte> operator>>=(const Byte &lhs, RValue<Byte> rhs)
+ Type *Short4::getType()
{
- return lhs = lhs >> rhs;
+ return T(Type_v4i16);
}
- RValue<Byte> operator+(RValue<Byte> val)
+ UShort4::UShort4(RValue<Float4> cast, bool saturate)
{
- return val;
+ if(saturate)
+ {
+ if(CPUID::SSE4_1)
+ {
+ // x86 produces 0x80000000 on 32-bit integer overflow/underflow.
+ // PackUnsigned takes care of 0x0000 saturation.
+ Int4 int4(Min(cast, Float4(0xFFFF)));
+ *this = As<UShort4>(PackUnsigned(int4, int4));
+ }
+ else if(CPUID::ARM)
+ {
+ // ARM saturates the 32-bit integer result on overflow/undeflow.
+ Int4 int4(cast);
+ *this = As<UShort4>(PackUnsigned(int4, int4));
+ }
+ else
+ {
+ *this = Short4(Int4(Max(Min(cast, Float4(0xFFFF)), Float4(0x0000))));
+ }
+ }
+ else
+ {
+ *this = Short4(Int4(cast));
+ }
}
- RValue<Byte> operator-(RValue<Byte> val)
+ RValue<UShort> Extract(RValue<UShort4> val, int i)
{
- return RValue<Byte>(Nucleus::createNeg(val.value));
+ return RValue<UShort>(Nucleus::createExtractElement(val.value, UShort::getType(), i));
}
- RValue<Byte> operator~(RValue<Byte> val)
+ RValue<UShort4> Insert(RValue<UShort4> val, RValue<UShort> element, int i)
{
- return RValue<Byte>(Nucleus::createNot(val.value));
+ return RValue<UShort4>(Nucleus::createInsertElement(val.value, element.value, i));
}
- RValue<Byte> operator++(const Byte &val, int) // Post-increment
+ RValue<UShort4> operator<<(RValue<UShort4> lhs, unsigned char rhs)
{
- RValue<Byte> res = val;
-
- assert(false && "UNIMPLEMENTED");
+ if(emulateIntrinsics)
+ {
+ UShort4 result;
+ result = Insert(result, Extract(lhs, 0) << UShort(rhs), 0);
+ result = Insert(result, Extract(lhs, 1) << UShort(rhs), 1);
+ result = Insert(result, Extract(lhs, 2) << UShort(rhs), 2);
+ result = Insert(result, Extract(lhs, 3) << UShort(rhs), 3);
- return res;
+ return result;
+ }
+ else
+ {
+ return RValue<UShort4>(Nucleus::createShl(lhs.value, V(::context->getConstantInt32(rhs))));
+ }
}
- const Byte &operator++(const Byte &val) // Pre-increment
+ RValue<UShort4> operator>>(RValue<UShort4> lhs, unsigned char rhs)
{
- assert(false && "UNIMPLEMENTED");
+ if(emulateIntrinsics)
+ {
+ UShort4 result;
+ result = Insert(result, Extract(lhs, 0) >> UShort(rhs), 0);
+ result = Insert(result, Extract(lhs, 1) >> UShort(rhs), 1);
+ result = Insert(result, Extract(lhs, 2) >> UShort(rhs), 2);
+ result = Insert(result, Extract(lhs, 3) >> UShort(rhs), 3);
- return val;
+ return result;
+ }
+ else
+ {
+ return RValue<UShort4>(Nucleus::createLShr(lhs.value, V(::context->getConstantInt32(rhs))));
+ }
}
- RValue<Byte> operator--(const Byte &val, int) // Post-decrement
+ RValue<UShort4> Max(RValue<UShort4> x, RValue<UShort4> y)
{
- RValue<Byte> res = val;
+ Ice::Variable *condition = ::function->makeVariable(Ice::IceType_v8i1);
+ auto cmp = Ice::InstIcmp::create(::function, Ice::InstIcmp::Ule, condition, x.value, y.value);
+ ::basicBlock->appendInst(cmp);
- assert(false && "UNIMPLEMENTED");
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_v8i16);
+ auto select = Ice::InstSelect::create(::function, result, condition, y.value, x.value);
+ ::basicBlock->appendInst(select);
- return res;
+ return RValue<UShort4>(V(result));
}
- const Byte &operator--(const Byte &val) // Pre-decrement
+ RValue<UShort4> Min(RValue<UShort4> x, RValue<UShort4> y)
{
- assert(false && "UNIMPLEMENTED");
+ Ice::Variable *condition = ::function->makeVariable(Ice::IceType_v8i1);
+ auto cmp = Ice::InstIcmp::create(::function, Ice::InstIcmp::Ugt, condition, x.value, y.value);
+ ::basicBlock->appendInst(cmp);
- return val;
- }
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_v8i16);
+ auto select = Ice::InstSelect::create(::function, result, condition, y.value, x.value);
+ ::basicBlock->appendInst(select);
- RValue<Bool> operator<(RValue<Byte> lhs, RValue<Byte> rhs)
- {
- return RValue<Bool>(Nucleus::createICmpULT(lhs.value, rhs.value));
+ return RValue<UShort4>(V(result));
}
- RValue<Bool> operator<=(RValue<Byte> lhs, RValue<Byte> rhs)
+ RValue<UShort> SaturateUnsigned(RValue<Int> x)
{
- return RValue<Bool>(Nucleus::createICmpULE(lhs.value, rhs.value));
+ return UShort(IfThenElse(x > 0xFFFF, Int(0xFFFF), IfThenElse(x < 0, Int(0), x)));
}
- RValue<Bool> operator>(RValue<Byte> lhs, RValue<Byte> rhs)
+ RValue<UShort4> AddSat(RValue<UShort4> x, RValue<UShort4> y)
{
- return RValue<Bool>(Nucleus::createICmpUGT(lhs.value, rhs.value));
- }
+ if(emulateIntrinsics)
+ {
+ UShort4 result;
+ result = Insert(result, SaturateUnsigned(Int(Extract(x, 0)) + Int(Extract(y, 0))), 0);
+ result = Insert(result, SaturateUnsigned(Int(Extract(x, 1)) + Int(Extract(y, 1))), 1);
+ result = Insert(result, SaturateUnsigned(Int(Extract(x, 2)) + Int(Extract(y, 2))), 2);
+ result = Insert(result, SaturateUnsigned(Int(Extract(x, 3)) + Int(Extract(y, 3))), 3);
- RValue<Bool> operator>=(RValue<Byte> lhs, RValue<Byte> rhs)
- {
- return RValue<Bool>(Nucleus::createICmpUGE(lhs.value, rhs.value));
+ return result;
+ }
+ else
+ {
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_v8i16);
+ const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::AddSaturateUnsigned, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
+ auto target = ::context->getConstantUndef(Ice::IceType_i32);
+ auto paddusw = Ice::InstIntrinsicCall::create(::function, 2, result, target, intrinsic);
+ paddusw->addArg(x.value);
+ paddusw->addArg(y.value);
+ ::basicBlock->appendInst(paddusw);
+
+ return RValue<UShort4>(V(result));
+ }
}
- RValue<Bool> operator!=(RValue<Byte> lhs, RValue<Byte> rhs)
+ RValue<UShort4> SubSat(RValue<UShort4> x, RValue<UShort4> y)
{
- return RValue<Bool>(Nucleus::createICmpNE(lhs.value, rhs.value));
- }
+ if(emulateIntrinsics)
+ {
+ UShort4 result;
+ result = Insert(result, SaturateUnsigned(Int(Extract(x, 0)) - Int(Extract(y, 0))), 0);
+ result = Insert(result, SaturateUnsigned(Int(Extract(x, 1)) - Int(Extract(y, 1))), 1);
+ result = Insert(result, SaturateUnsigned(Int(Extract(x, 2)) - Int(Extract(y, 2))), 2);
+ result = Insert(result, SaturateUnsigned(Int(Extract(x, 3)) - Int(Extract(y, 3))), 3);
- RValue<Bool> operator==(RValue<Byte> lhs, RValue<Byte> rhs)
- {
- return RValue<Bool>(Nucleus::createICmpEQ(lhs.value, rhs.value));
+ return result;
+ }
+ else
+ {
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_v8i16);
+ const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::SubtractSaturateUnsigned, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
+ auto target = ::context->getConstantUndef(Ice::IceType_i32);
+ auto psubusw = Ice::InstIntrinsicCall::create(::function, 2, result, target, intrinsic);
+ psubusw->addArg(x.value);
+ psubusw->addArg(y.value);
+ ::basicBlock->appendInst(psubusw);
+
+ return RValue<UShort4>(V(result));
+ }
}
- Type *Byte::getType()
+ RValue<UShort4> MulHigh(RValue<UShort4> x, RValue<UShort4> y)
{
- return T(Ice::IceType_i8);
- }
+ if(emulateIntrinsics)
+ {
+ UShort4 result;
+ result = Insert(result, UShort((UInt(Extract(x, 0)) * UInt(Extract(y, 0))) >> 16), 0);
+ result = Insert(result, UShort((UInt(Extract(x, 1)) * UInt(Extract(y, 1))) >> 16), 1);
+ result = Insert(result, UShort((UInt(Extract(x, 2)) * UInt(Extract(y, 2))) >> 16), 2);
+ result = Insert(result, UShort((UInt(Extract(x, 3)) * UInt(Extract(y, 3))) >> 16), 3);
- SByte::SByte(Argument<SByte> argument)
- {
- storeValue(argument.value);
+ return result;
+ }
+ else
+ {
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_v8i16);
+ const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::MultiplyHighUnsigned, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
+ auto target = ::context->getConstantUndef(Ice::IceType_i32);
+ auto pmulhuw = Ice::InstIntrinsicCall::create(::function, 2, result, target, intrinsic);
+ pmulhuw->addArg(x.value);
+ pmulhuw->addArg(y.value);
+ ::basicBlock->appendInst(pmulhuw);
+
+ return RValue<UShort4>(V(result));
+ }
}
- SByte::SByte(RValue<Int> cast)
+ RValue<Int4> MulHigh(RValue<Int4> x, RValue<Int4> y)
{
- Value *integer = Nucleus::createTrunc(cast.value, SByte::getType());
+ // TODO: For x86, build an intrinsics version of this which uses shuffles + pmuludq.
+
+ // Scalarized implementation.
+ Int4 result;
+ result = Insert(result, Int((Long(Extract(x, 0)) * Long(Extract(y, 0))) >> Long(Int(32))), 0);
+ result = Insert(result, Int((Long(Extract(x, 1)) * Long(Extract(y, 1))) >> Long(Int(32))), 1);
+ result = Insert(result, Int((Long(Extract(x, 2)) * Long(Extract(y, 2))) >> Long(Int(32))), 2);
+ result = Insert(result, Int((Long(Extract(x, 3)) * Long(Extract(y, 3))) >> Long(Int(32))), 3);
- storeValue(integer);
+ return result;
}
- SByte::SByte(RValue<Short> cast)
+ RValue<UInt4> MulHigh(RValue<UInt4> x, RValue<UInt4> y)
{
- Value *integer = Nucleus::createTrunc(cast.value, SByte::getType());
+ // TODO: For x86, build an intrinsics version of this which uses shuffles + pmuludq.
- storeValue(integer);
- }
+ if(false) // Partial product based implementation.
+ {
+ auto xh = x >> 16;
+ auto yh = y >> 16;
+ auto xl = x & UInt4(0x0000FFFF);
+ auto yl = y & UInt4(0x0000FFFF);
+ auto xlyh = xl * yh;
+ auto xhyl = xh * yl;
+ auto xlyhh = xlyh >> 16;
+ auto xhylh = xhyl >> 16;
+ auto xlyhl = xlyh & UInt4(0x0000FFFF);
+ auto xhyll = xhyl & UInt4(0x0000FFFF);
+ auto xlylh = (xl * yl) >> 16;
+ auto oflow = (xlyhl + xhyll + xlylh) >> 16;
+
+ return (xh * yh) + (xlyhh + xhylh) + oflow;
+ }
- SByte::SByte()
- {
- }
+ // Scalarized implementation.
+ Int4 result;
+ result = Insert(result, Int((Long(UInt(Extract(As<Int4>(x), 0))) * Long(UInt(Extract(As<Int4>(y), 0)))) >> Long(Int(32))), 0);
+ result = Insert(result, Int((Long(UInt(Extract(As<Int4>(x), 1))) * Long(UInt(Extract(As<Int4>(y), 1)))) >> Long(Int(32))), 1);
+ result = Insert(result, Int((Long(UInt(Extract(As<Int4>(x), 2))) * Long(UInt(Extract(As<Int4>(y), 2)))) >> Long(Int(32))), 2);
+ result = Insert(result, Int((Long(UInt(Extract(As<Int4>(x), 3))) * Long(UInt(Extract(As<Int4>(y), 3)))) >> Long(Int(32))), 3);
- SByte::SByte(signed char x)
- {
- storeValue(Nucleus::createConstantByte(x));
+ return As<UInt4>(result);
}
- SByte::SByte(RValue<SByte> rhs)
+ RValue<UShort4> Average(RValue<UShort4> x, RValue<UShort4> y)
{
- storeValue(rhs.value);
+ assert(false && "UNIMPLEMENTED"); return RValue<UShort4>(V(nullptr));
}
- SByte::SByte(const SByte &rhs)
+ Type *UShort4::getType()
{
- Value *value = rhs.loadValue();
- storeValue(value);
+ return T(Type_v4i16);
}
- SByte::SByte(const Reference<SByte> &rhs)
+ RValue<Short> Extract(RValue<Short8> val, int i)
{
- Value *value = rhs.loadValue();
- storeValue(value);
+ return RValue<Short>(Nucleus::createExtractElement(val.value, Short::getType(), i));
}
- RValue<SByte> SByte::operator=(RValue<SByte> rhs) const
+ RValue<Short8> Insert(RValue<Short8> val, RValue<Short> element, int i)
{
- storeValue(rhs.value);
-
- return rhs;
+ return RValue<Short8>(Nucleus::createInsertElement(val.value, element.value, i));
}
- RValue<SByte> SByte::operator=(const SByte &rhs) const
+ RValue<Short8> operator<<(RValue<Short8> lhs, unsigned char rhs)
{
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<SByte>(value);
+ if(emulateIntrinsics)
+ {
+ Short8 result;
+ result = Insert(result, Extract(lhs, 0) << Short(rhs), 0);
+ result = Insert(result, Extract(lhs, 1) << Short(rhs), 1);
+ result = Insert(result, Extract(lhs, 2) << Short(rhs), 2);
+ result = Insert(result, Extract(lhs, 3) << Short(rhs), 3);
+ result = Insert(result, Extract(lhs, 4) << Short(rhs), 4);
+ result = Insert(result, Extract(lhs, 5) << Short(rhs), 5);
+ result = Insert(result, Extract(lhs, 6) << Short(rhs), 6);
+ result = Insert(result, Extract(lhs, 7) << Short(rhs), 7);
+
+ return result;
+ }
+ else
+ {
+ return RValue<Short8>(Nucleus::createShl(lhs.value, V(::context->getConstantInt32(rhs))));
+ }
}
- RValue<SByte> SByte::operator=(const Reference<SByte> &rhs) const
+ RValue<Short8> operator>>(RValue<Short8> lhs, unsigned char rhs)
{
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<SByte>(value);
+ if(emulateIntrinsics)
+ {
+ Short8 result;
+ result = Insert(result, Extract(lhs, 0) >> Short(rhs), 0);
+ result = Insert(result, Extract(lhs, 1) >> Short(rhs), 1);
+ result = Insert(result, Extract(lhs, 2) >> Short(rhs), 2);
+ result = Insert(result, Extract(lhs, 3) >> Short(rhs), 3);
+ result = Insert(result, Extract(lhs, 4) >> Short(rhs), 4);
+ result = Insert(result, Extract(lhs, 5) >> Short(rhs), 5);
+ result = Insert(result, Extract(lhs, 6) >> Short(rhs), 6);
+ result = Insert(result, Extract(lhs, 7) >> Short(rhs), 7);
+
+ return result;
+ }
+ else
+ {
+ return RValue<Short8>(Nucleus::createAShr(lhs.value, V(::context->getConstantInt32(rhs))));
+ }
}
- RValue<SByte> operator+(RValue<SByte> lhs, RValue<SByte> rhs)
+ RValue<Int4> MulAdd(RValue<Short8> x, RValue<Short8> y)
{
- return RValue<SByte>(Nucleus::createAdd(lhs.value, rhs.value));
+ assert(false && "UNIMPLEMENTED"); return RValue<Int4>(V(nullptr));
}
- RValue<SByte> operator-(RValue<SByte> lhs, RValue<SByte> rhs)
+ RValue<Short8> MulHigh(RValue<Short8> x, RValue<Short8> y)
{
- return RValue<SByte>(Nucleus::createSub(lhs.value, rhs.value));
+ assert(false && "UNIMPLEMENTED"); return RValue<Short8>(V(nullptr));
}
- RValue<SByte> operator*(RValue<SByte> lhs, RValue<SByte> rhs)
+ Type *Short8::getType()
{
- return RValue<SByte>(Nucleus::createMul(lhs.value, rhs.value));
+ return T(Ice::IceType_v8i16);
}
- RValue<SByte> operator/(RValue<SByte> lhs, RValue<SByte> rhs)
+ RValue<UShort> Extract(RValue<UShort8> val, int i)
{
- return RValue<SByte>(Nucleus::createSDiv(lhs.value, rhs.value));
+ return RValue<UShort>(Nucleus::createExtractElement(val.value, UShort::getType(), i));
}
- RValue<SByte> operator%(RValue<SByte> lhs, RValue<SByte> rhs)
+ RValue<UShort8> Insert(RValue<UShort8> val, RValue<UShort> element, int i)
{
- return RValue<SByte>(Nucleus::createSRem(lhs.value, rhs.value));
+ return RValue<UShort8>(Nucleus::createInsertElement(val.value, element.value, i));
}
- RValue<SByte> operator&(RValue<SByte> lhs, RValue<SByte> rhs)
+ RValue<UShort8> operator<<(RValue<UShort8> lhs, unsigned char rhs)
{
- return RValue<SByte>(Nucleus::createAnd(lhs.value, rhs.value));
+ if(emulateIntrinsics)
+ {
+ UShort8 result;
+ result = Insert(result, Extract(lhs, 0) << UShort(rhs), 0);
+ result = Insert(result, Extract(lhs, 1) << UShort(rhs), 1);
+ result = Insert(result, Extract(lhs, 2) << UShort(rhs), 2);
+ result = Insert(result, Extract(lhs, 3) << UShort(rhs), 3);
+ result = Insert(result, Extract(lhs, 4) << UShort(rhs), 4);
+ result = Insert(result, Extract(lhs, 5) << UShort(rhs), 5);
+ result = Insert(result, Extract(lhs, 6) << UShort(rhs), 6);
+ result = Insert(result, Extract(lhs, 7) << UShort(rhs), 7);
+
+ return result;
+ }
+ else
+ {
+ return RValue<UShort8>(Nucleus::createShl(lhs.value, V(::context->getConstantInt32(rhs))));
+ }
}
- RValue<SByte> operator|(RValue<SByte> lhs, RValue<SByte> rhs)
+ RValue<UShort8> operator>>(RValue<UShort8> lhs, unsigned char rhs)
{
- return RValue<SByte>(Nucleus::createOr(lhs.value, rhs.value));
+ if(emulateIntrinsics)
+ {
+ UShort8 result;
+ result = Insert(result, Extract(lhs, 0) >> UShort(rhs), 0);
+ result = Insert(result, Extract(lhs, 1) >> UShort(rhs), 1);
+ result = Insert(result, Extract(lhs, 2) >> UShort(rhs), 2);
+ result = Insert(result, Extract(lhs, 3) >> UShort(rhs), 3);
+ result = Insert(result, Extract(lhs, 4) >> UShort(rhs), 4);
+ result = Insert(result, Extract(lhs, 5) >> UShort(rhs), 5);
+ result = Insert(result, Extract(lhs, 6) >> UShort(rhs), 6);
+ result = Insert(result, Extract(lhs, 7) >> UShort(rhs), 7);
+
+ return result;
+ }
+ else
+ {
+ return RValue<UShort8>(Nucleus::createLShr(lhs.value, V(::context->getConstantInt32(rhs))));
+ }
}
- RValue<SByte> operator^(RValue<SByte> lhs, RValue<SByte> rhs)
+ RValue<UShort8> Swizzle(RValue<UShort8> x, char select0, char select1, char select2, char select3, char select4, char select5, char select6, char select7)
{
- return RValue<SByte>(Nucleus::createXor(lhs.value, rhs.value));
+ assert(false && "UNIMPLEMENTED"); return RValue<UShort8>(V(nullptr));
}
- RValue<SByte> operator<<(RValue<SByte> lhs, RValue<SByte> rhs)
+ RValue<UShort8> MulHigh(RValue<UShort8> x, RValue<UShort8> y)
{
- return RValue<SByte>(Nucleus::createShl(lhs.value, rhs.value));
+ assert(false && "UNIMPLEMENTED"); return RValue<UShort8>(V(nullptr));
}
- RValue<SByte> operator>>(RValue<SByte> lhs, RValue<SByte> rhs)
- {
- return RValue<SByte>(Nucleus::createAShr(lhs.value, rhs.value));
- }
+ // FIXME: Implement as Shuffle(x, y, Select(i0, ..., i16)) and Shuffle(x, y, SELECT_PACK_REPEAT(element))
+// RValue<UShort8> PackRepeat(RValue<Byte16> x, RValue<Byte16> y, int element)
+// {
+// assert(false && "UNIMPLEMENTED"); return RValue<UShort8>(V(nullptr));
+// }
- RValue<SByte> operator+=(const SByte &lhs, RValue<SByte> rhs)
+ Type *UShort8::getType()
{
- return lhs = lhs + rhs;
+ return T(Ice::IceType_v8i16);
}
- RValue<SByte> operator-=(const SByte &lhs, RValue<SByte> rhs)
+ RValue<Int> operator++(Int &val, int) // Post-increment
{
- return lhs = lhs - rhs;
+ RValue<Int> res = val;
+ val += 1;
+ return res;
}
- RValue<SByte> operator*=(const SByte &lhs, RValue<SByte> rhs)
+ const Int &operator++(Int &val) // Pre-increment
{
- return lhs = lhs * rhs;
+ val += 1;
+ return val;
}
- RValue<SByte> operator/=(const SByte &lhs, RValue<SByte> rhs)
+ RValue<Int> operator--(Int &val, int) // Post-decrement
{
- return lhs = lhs / rhs;
+ RValue<Int> res = val;
+ val -= 1;
+ return res;
}
- RValue<SByte> operator%=(const SByte &lhs, RValue<SByte> rhs)
+ const Int &operator--(Int &val) // Pre-decrement
{
- return lhs = lhs % rhs;
+ val -= 1;
+ return val;
}
- RValue<SByte> operator&=(const SByte &lhs, RValue<SByte> rhs)
+ RValue<Int> RoundInt(RValue<Float> cast)
{
- return lhs = lhs & rhs;
- }
-
- RValue<SByte> operator|=(const SByte &lhs, RValue<SByte> rhs)
- {
- return lhs = lhs | rhs;
- }
-
- RValue<SByte> operator^=(const SByte &lhs, RValue<SByte> rhs)
- {
- return lhs = lhs ^ rhs;
- }
-
- RValue<SByte> operator<<=(const SByte &lhs, RValue<SByte> rhs)
- {
- return lhs = lhs << rhs;
- }
-
- RValue<SByte> operator>>=(const SByte &lhs, RValue<SByte> rhs)
- {
- return lhs = lhs >> rhs;
- }
-
- RValue<SByte> operator+(RValue<SByte> val)
- {
- return val;
- }
-
- RValue<SByte> operator-(RValue<SByte> val)
- {
- return RValue<SByte>(Nucleus::createNeg(val.value));
- }
-
- RValue<SByte> operator~(RValue<SByte> val)
- {
- return RValue<SByte>(Nucleus::createNot(val.value));
- }
-
- RValue<SByte> operator++(const SByte &val, int) // Post-increment
- {
- RValue<SByte> res = val;
-
- assert(false && "UNIMPLEMENTED");
-
- return res;
- }
-
- const SByte &operator++(const SByte &val) // Pre-increment
- {
- assert(false && "UNIMPLEMENTED");
- assert(false && "UNIMPLEMENTED");
-
- return val;
- }
-
- RValue<SByte> operator--(const SByte &val, int) // Post-decrement
- {
- RValue<SByte> res = val;
-
- assert(false && "UNIMPLEMENTED");
- assert(false && "UNIMPLEMENTED");
-
- return res;
- }
-
- const SByte &operator--(const SByte &val) // Pre-decrement
- {
- assert(false && "UNIMPLEMENTED");
- assert(false && "UNIMPLEMENTED");
-
- return val;
- }
-
- RValue<Bool> operator<(RValue<SByte> lhs, RValue<SByte> rhs)
- {
- return RValue<Bool>(Nucleus::createICmpSLT(lhs.value, rhs.value));
- }
-
- RValue<Bool> operator<=(RValue<SByte> lhs, RValue<SByte> rhs)
- {
- return RValue<Bool>(Nucleus::createICmpSLE(lhs.value, rhs.value));
- }
-
- RValue<Bool> operator>(RValue<SByte> lhs, RValue<SByte> rhs)
- {
- return RValue<Bool>(Nucleus::createICmpSGT(lhs.value, rhs.value));
- }
-
- RValue<Bool> operator>=(RValue<SByte> lhs, RValue<SByte> rhs)
- {
- return RValue<Bool>(Nucleus::createICmpSGE(lhs.value, rhs.value));
- }
-
- RValue<Bool> operator!=(RValue<SByte> lhs, RValue<SByte> rhs)
- {
- return RValue<Bool>(Nucleus::createICmpNE(lhs.value, rhs.value));
- }
-
- RValue<Bool> operator==(RValue<SByte> lhs, RValue<SByte> rhs)
- {
- return RValue<Bool>(Nucleus::createICmpEQ(lhs.value, rhs.value));
- }
-
- Type *SByte::getType()
- {
- return T(Ice::IceType_i8);
- }
-
- Short::Short(Argument<Short> argument)
- {
- storeValue(argument.value);
- }
-
- Short::Short(RValue<Int> cast)
- {
- Value *integer = Nucleus::createTrunc(cast.value, Short::getType());
-
- storeValue(integer);
- }
-
- Short::Short()
- {
- }
-
- Short::Short(short x)
- {
- storeValue(Nucleus::createConstantShort(x));
- }
-
- Short::Short(RValue<Short> rhs)
- {
- storeValue(rhs.value);
- }
-
- Short::Short(const Short &rhs)
- {
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- Short::Short(const Reference<Short> &rhs)
- {
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- RValue<Short> Short::operator=(RValue<Short> rhs) const
- {
- storeValue(rhs.value);
-
- return rhs;
- }
-
- RValue<Short> Short::operator=(const Short &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<Short>(value);
- }
-
- RValue<Short> Short::operator=(const Reference<Short> &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<Short>(value);
- }
-
- RValue<Short> operator+(RValue<Short> lhs, RValue<Short> rhs)
- {
- return RValue<Short>(Nucleus::createAdd(lhs.value, rhs.value));
- }
-
- RValue<Short> operator-(RValue<Short> lhs, RValue<Short> rhs)
- {
- return RValue<Short>(Nucleus::createSub(lhs.value, rhs.value));
- }
-
- RValue<Short> operator*(RValue<Short> lhs, RValue<Short> rhs)
- {
- return RValue<Short>(Nucleus::createMul(lhs.value, rhs.value));
- }
-
- RValue<Short> operator/(RValue<Short> lhs, RValue<Short> rhs)
- {
- return RValue<Short>(Nucleus::createSDiv(lhs.value, rhs.value));
- }
-
- RValue<Short> operator%(RValue<Short> lhs, RValue<Short> rhs)
- {
- return RValue<Short>(Nucleus::createSRem(lhs.value, rhs.value));
- }
-
- RValue<Short> operator&(RValue<Short> lhs, RValue<Short> rhs)
- {
- return RValue<Short>(Nucleus::createAnd(lhs.value, rhs.value));
- }
-
- RValue<Short> operator|(RValue<Short> lhs, RValue<Short> rhs)
- {
- return RValue<Short>(Nucleus::createOr(lhs.value, rhs.value));
- }
-
- RValue<Short> operator^(RValue<Short> lhs, RValue<Short> rhs)
- {
- return RValue<Short>(Nucleus::createXor(lhs.value, rhs.value));
- }
-
- RValue<Short> operator<<(RValue<Short> lhs, RValue<Short> rhs)
- {
- return RValue<Short>(Nucleus::createShl(lhs.value, rhs.value));
- }
-
- RValue<Short> operator>>(RValue<Short> lhs, RValue<Short> rhs)
- {
- return RValue<Short>(Nucleus::createAShr(lhs.value, rhs.value));
- }
-
- RValue<Short> operator+=(const Short &lhs, RValue<Short> rhs)
- {
- return lhs = lhs + rhs;
- }
-
- RValue<Short> operator-=(const Short &lhs, RValue<Short> rhs)
- {
- return lhs = lhs - rhs;
- }
-
- RValue<Short> operator*=(const Short &lhs, RValue<Short> rhs)
- {
- return lhs = lhs * rhs;
- }
-
- RValue<Short> operator/=(const Short &lhs, RValue<Short> rhs)
- {
- return lhs = lhs / rhs;
- }
-
- RValue<Short> operator%=(const Short &lhs, RValue<Short> rhs)
- {
- return lhs = lhs % rhs;
- }
-
- RValue<Short> operator&=(const Short &lhs, RValue<Short> rhs)
- {
- return lhs = lhs & rhs;
- }
-
- RValue<Short> operator|=(const Short &lhs, RValue<Short> rhs)
- {
- return lhs = lhs | rhs;
- }
-
- RValue<Short> operator^=(const Short &lhs, RValue<Short> rhs)
- {
- return lhs = lhs ^ rhs;
- }
-
- RValue<Short> operator<<=(const Short &lhs, RValue<Short> rhs)
- {
- return lhs = lhs << rhs;
+ if(emulateIntrinsics || CPUID::ARM)
+ {
+ // Push the fractional part off the mantissa. Accurate up to +/-2^22.
+ return Int((cast + Float(0x00C00000)) - Float(0x00C00000));
+ }
+ else
+ {
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_i32);
+ const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::Nearbyint, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
+ auto target = ::context->getConstantUndef(Ice::IceType_i32);
+ auto nearbyint = Ice::InstIntrinsicCall::create(::function, 1, result, target, intrinsic);
+ nearbyint->addArg(cast.value);
+ ::basicBlock->appendInst(nearbyint);
+
+ return RValue<Int>(V(result));
+ }
}
- RValue<Short> operator>>=(const Short &lhs, RValue<Short> rhs)
+ Type *Int::getType()
{
- return lhs = lhs >> rhs;
+ return T(Ice::IceType_i32);
}
- RValue<Short> operator+(RValue<Short> val)
+ Type *Long::getType()
{
- return val;
+ return T(Ice::IceType_i64);
}
- RValue<Short> operator-(RValue<Short> val)
+ UInt::UInt(RValue<Float> cast)
{
- return RValue<Short>(Nucleus::createNeg(val.value));
- }
+ // Smallest positive value representable in UInt, but not in Int
+ const unsigned int ustart = 0x80000000u;
+ const float ustartf = float(ustart);
- RValue<Short> operator~(RValue<Short> val)
- {
- return RValue<Short>(Nucleus::createNot(val.value));
+ // If the value is negative, store 0, otherwise store the result of the conversion
+ storeValue((~(As<Int>(cast) >> 31) &
+ // Check if the value can be represented as an Int
+ IfThenElse(cast >= ustartf,
+ // If the value is too large, subtract ustart and re-add it after conversion.
+ As<Int>(As<UInt>(Int(cast - Float(ustartf))) + UInt(ustart)),
+ // Otherwise, just convert normally
+ Int(cast))).value);
}
- RValue<Short> operator++(const Short &val, int) // Post-increment
+ RValue<UInt> operator++(UInt &val, int) // Post-increment
{
- RValue<Short> res = val;
-
- assert(false && "UNIMPLEMENTED");
- assert(false && "UNIMPLEMENTED");
-
+ RValue<UInt> res = val;
+ val += 1;
return res;
}
- const Short &operator++(const Short &val) // Pre-increment
+ const UInt &operator++(UInt &val) // Pre-increment
{
- assert(false && "UNIMPLEMENTED");
- assert(false && "UNIMPLEMENTED");
-
+ val += 1;
return val;
}
- RValue<Short> operator--(const Short &val, int) // Post-decrement
+ RValue<UInt> operator--(UInt &val, int) // Post-decrement
{
- RValue<Short> res = val;
-
- assert(false && "UNIMPLEMENTED");
- assert(false && "UNIMPLEMENTED");
-
+ RValue<UInt> res = val;
+ val -= 1;
return res;
}
- const Short &operator--(const Short &val) // Pre-decrement
+ const UInt &operator--(UInt &val) // Pre-decrement
{
- assert(false && "UNIMPLEMENTED");
- assert(false && "UNIMPLEMENTED");
-
+ val -= 1;
return val;
}
- RValue<Bool> operator<(RValue<Short> lhs, RValue<Short> rhs)
- {
- return RValue<Bool>(Nucleus::createICmpSLT(lhs.value, rhs.value));
- }
-
- RValue<Bool> operator<=(RValue<Short> lhs, RValue<Short> rhs)
- {
- return RValue<Bool>(Nucleus::createICmpSLE(lhs.value, rhs.value));
- }
-
- RValue<Bool> operator>(RValue<Short> lhs, RValue<Short> rhs)
- {
- return RValue<Bool>(Nucleus::createICmpSGT(lhs.value, rhs.value));
- }
-
- RValue<Bool> operator>=(RValue<Short> lhs, RValue<Short> rhs)
- {
- return RValue<Bool>(Nucleus::createICmpSGE(lhs.value, rhs.value));
- }
-
- RValue<Bool> operator!=(RValue<Short> lhs, RValue<Short> rhs)
- {
- return RValue<Bool>(Nucleus::createICmpNE(lhs.value, rhs.value));
- }
-
- RValue<Bool> operator==(RValue<Short> lhs, RValue<Short> rhs)
- {
- return RValue<Bool>(Nucleus::createICmpEQ(lhs.value, rhs.value));
- }
+// RValue<UInt> RoundUInt(RValue<Float> cast)
+// {
+// assert(false && "UNIMPLEMENTED"); return RValue<UInt>(V(nullptr));
+// }
- Type *Short::getType()
+ Type *UInt::getType()
{
- return T(Ice::IceType_i16);
+ return T(Ice::IceType_i32);
}
- UShort::UShort(Argument<UShort> argument)
- {
- storeValue(argument.value);
- }
+// Int2::Int2(RValue<Int> cast)
+// {
+// Value *extend = Nucleus::createZExt(cast.value, Long::getType());
+// Value *vector = Nucleus::createBitCast(extend, Int2::getType());
+//
+// Constant *shuffle[2];
+// shuffle[0] = Nucleus::createConstantInt(0);
+// shuffle[1] = Nucleus::createConstantInt(0);
+//
+// Value *replicate = Nucleus::createShuffleVector(vector, UndefValue::get(Int2::getType()), Nucleus::createConstantVector(shuffle, 2));
+//
+// storeValue(replicate);
+// }
- UShort::UShort(RValue<UInt> cast)
+ RValue<Int2> operator<<(RValue<Int2> lhs, unsigned char rhs)
{
- Value *integer = Nucleus::createTrunc(cast.value, UShort::getType());
+ if(emulateIntrinsics)
+ {
+ Int2 result;
+ result = Insert(result, Extract(lhs, 0) << Int(rhs), 0);
+ result = Insert(result, Extract(lhs, 1) << Int(rhs), 1);
- storeValue(integer);
+ return result;
+ }
+ else
+ {
+ return RValue<Int2>(Nucleus::createShl(lhs.value, V(::context->getConstantInt32(rhs))));
+ }
}
- UShort::UShort(RValue<Int> cast)
+ RValue<Int2> operator>>(RValue<Int2> lhs, unsigned char rhs)
{
- Value *integer = Nucleus::createTrunc(cast.value, UShort::getType());
+ if(emulateIntrinsics)
+ {
+ Int2 result;
+ result = Insert(result, Extract(lhs, 0) >> Int(rhs), 0);
+ result = Insert(result, Extract(lhs, 1) >> Int(rhs), 1);
- storeValue(integer);
+ return result;
+ }
+ else
+ {
+ return RValue<Int2>(Nucleus::createAShr(lhs.value, V(::context->getConstantInt32(rhs))));
+ }
}
- UShort::UShort()
+ Type *Int2::getType()
{
+ return T(Type_v2i32);
}
- UShort::UShort(unsigned short x)
+ RValue<UInt> Extract(RValue<UInt2> val, int i)
{
- storeValue(Nucleus::createConstantShort(x));
+ return RValue<UInt>(Nucleus::createExtractElement(val.value, UInt::getType(), i));
}
- UShort::UShort(RValue<UShort> rhs)
+ RValue<UInt2> Insert(RValue<UInt2> val, RValue<UInt> element, int i)
{
- storeValue(rhs.value);
+ return RValue<UInt2>(Nucleus::createInsertElement(val.value, element.value, i));
}
- UShort::UShort(const UShort &rhs)
+ RValue<UInt2> operator<<(RValue<UInt2> lhs, unsigned char rhs)
{
- Value *value = rhs.loadValue();
- storeValue(value);
- }
+ if(emulateIntrinsics)
+ {
+ UInt2 result;
+ result = Insert(result, Extract(lhs, 0) << UInt(rhs), 0);
+ result = Insert(result, Extract(lhs, 1) << UInt(rhs), 1);
- UShort::UShort(const Reference<UShort> &rhs)
- {
- Value *value = rhs.loadValue();
- storeValue(value);
+ return result;
+ }
+ else
+ {
+ return RValue<UInt2>(Nucleus::createShl(lhs.value, V(::context->getConstantInt32(rhs))));
+ }
}
- RValue<UShort> UShort::operator=(RValue<UShort> rhs) const
- {
- storeValue(rhs.value);
-
- return rhs;
- }
-
- RValue<UShort> UShort::operator=(const UShort &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<UShort>(value);
- }
-
- RValue<UShort> UShort::operator=(const Reference<UShort> &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<UShort>(value);
- }
-
- RValue<UShort> operator+(RValue<UShort> lhs, RValue<UShort> rhs)
- {
- return RValue<UShort>(Nucleus::createAdd(lhs.value, rhs.value));
- }
-
- RValue<UShort> operator-(RValue<UShort> lhs, RValue<UShort> rhs)
- {
- return RValue<UShort>(Nucleus::createSub(lhs.value, rhs.value));
- }
-
- RValue<UShort> operator*(RValue<UShort> lhs, RValue<UShort> rhs)
- {
- return RValue<UShort>(Nucleus::createMul(lhs.value, rhs.value));
- }
-
- RValue<UShort> operator/(RValue<UShort> lhs, RValue<UShort> rhs)
- {
- return RValue<UShort>(Nucleus::createUDiv(lhs.value, rhs.value));
- }
-
- RValue<UShort> operator%(RValue<UShort> lhs, RValue<UShort> rhs)
- {
- return RValue<UShort>(Nucleus::createURem(lhs.value, rhs.value));
- }
-
- RValue<UShort> operator&(RValue<UShort> lhs, RValue<UShort> rhs)
- {
- return RValue<UShort>(Nucleus::createAnd(lhs.value, rhs.value));
- }
-
- RValue<UShort> operator|(RValue<UShort> lhs, RValue<UShort> rhs)
- {
- return RValue<UShort>(Nucleus::createOr(lhs.value, rhs.value));
- }
-
- RValue<UShort> operator^(RValue<UShort> lhs, RValue<UShort> rhs)
- {
- return RValue<UShort>(Nucleus::createXor(lhs.value, rhs.value));
- }
-
- RValue<UShort> operator<<(RValue<UShort> lhs, RValue<UShort> rhs)
- {
- return RValue<UShort>(Nucleus::createShl(lhs.value, rhs.value));
- }
-
- RValue<UShort> operator>>(RValue<UShort> lhs, RValue<UShort> rhs)
- {
- return RValue<UShort>(Nucleus::createLShr(lhs.value, rhs.value));
- }
-
- RValue<UShort> operator+=(const UShort &lhs, RValue<UShort> rhs)
- {
- return lhs = lhs + rhs;
- }
-
- RValue<UShort> operator-=(const UShort &lhs, RValue<UShort> rhs)
- {
- return lhs = lhs - rhs;
- }
-
- RValue<UShort> operator*=(const UShort &lhs, RValue<UShort> rhs)
- {
- return lhs = lhs * rhs;
- }
-
- RValue<UShort> operator/=(const UShort &lhs, RValue<UShort> rhs)
- {
- return lhs = lhs / rhs;
- }
-
- RValue<UShort> operator%=(const UShort &lhs, RValue<UShort> rhs)
- {
- return lhs = lhs % rhs;
- }
-
- RValue<UShort> operator&=(const UShort &lhs, RValue<UShort> rhs)
- {
- return lhs = lhs & rhs;
- }
-
- RValue<UShort> operator|=(const UShort &lhs, RValue<UShort> rhs)
- {
- return lhs = lhs | rhs;
- }
-
- RValue<UShort> operator^=(const UShort &lhs, RValue<UShort> rhs)
- {
- return lhs = lhs ^ rhs;
- }
-
- RValue<UShort> operator<<=(const UShort &lhs, RValue<UShort> rhs)
- {
- return lhs = lhs << rhs;
- }
-
- RValue<UShort> operator>>=(const UShort &lhs, RValue<UShort> rhs)
- {
- return lhs = lhs >> rhs;
- }
-
- RValue<UShort> operator+(RValue<UShort> val)
- {
- return val;
- }
-
- RValue<UShort> operator-(RValue<UShort> val)
- {
- return RValue<UShort>(Nucleus::createNeg(val.value));
- }
-
- RValue<UShort> operator~(RValue<UShort> val)
- {
- return RValue<UShort>(Nucleus::createNot(val.value));
- }
-
- RValue<UShort> operator++(const UShort &val, int) // Post-increment
- {
- RValue<UShort> res = val;
-
- assert(false && "UNIMPLEMENTED");
- assert(false && "UNIMPLEMENTED");
-
- return res;
- }
-
- const UShort &operator++(const UShort &val) // Pre-increment
- {
- assert(false && "UNIMPLEMENTED");
- assert(false && "UNIMPLEMENTED");
-
- return val;
- }
-
- RValue<UShort> operator--(const UShort &val, int) // Post-decrement
- {
- RValue<UShort> res = val;
-
- assert(false && "UNIMPLEMENTED");
- assert(false && "UNIMPLEMENTED");
-
- return res;
- }
-
- const UShort &operator--(const UShort &val) // Pre-decrement
- {
- assert(false && "UNIMPLEMENTED");
- assert(false && "UNIMPLEMENTED");
-
- return val;
- }
-
- RValue<Bool> operator<(RValue<UShort> lhs, RValue<UShort> rhs)
- {
- return RValue<Bool>(Nucleus::createICmpULT(lhs.value, rhs.value));
- }
-
- RValue<Bool> operator<=(RValue<UShort> lhs, RValue<UShort> rhs)
- {
- return RValue<Bool>(Nucleus::createICmpULE(lhs.value, rhs.value));
- }
-
- RValue<Bool> operator>(RValue<UShort> lhs, RValue<UShort> rhs)
- {
- return RValue<Bool>(Nucleus::createICmpUGT(lhs.value, rhs.value));
- }
-
- RValue<Bool> operator>=(RValue<UShort> lhs, RValue<UShort> rhs)
- {
- return RValue<Bool>(Nucleus::createICmpUGE(lhs.value, rhs.value));
- }
-
- RValue<Bool> operator!=(RValue<UShort> lhs, RValue<UShort> rhs)
- {
- return RValue<Bool>(Nucleus::createICmpNE(lhs.value, rhs.value));
- }
-
- RValue<Bool> operator==(RValue<UShort> lhs, RValue<UShort> rhs)
- {
- return RValue<Bool>(Nucleus::createICmpEQ(lhs.value, rhs.value));
- }
-
- Type *UShort::getType()
- {
- return T(Ice::IceType_i16);
- }
-
- Byte4::Byte4(RValue<Byte8> cast)
- {
- // xyzw.parent = this;
-
- storeValue(Nucleus::createBitCast(cast.value, getType()));
- }
-
- Byte4::Byte4(const Reference<Byte4> &rhs)
- {
- // xyzw.parent = this;
-
- assert(false && "UNIMPLEMENTED");
- }
-
- Type *Byte4::getType()
- {
- return T(Type_v4i8);
- }
-
- Type *SByte4::getType()
- {
- return T(Type_v4i8);
- }
-
- Byte8::Byte8()
- {
- }
-
- Byte8::Byte8(uint8_t x0, uint8_t x1, uint8_t x2, uint8_t x3, uint8_t x4, uint8_t x5, uint8_t x6, uint8_t x7)
- {
- int64_t constantVector[8] = {x0, x1, x2, x3, x4, x5, x6, x7};
- storeValue(Nucleus::createConstantVector(constantVector, getType()));
- }
-
- Byte8::Byte8(RValue<Byte8> rhs)
- {
- storeValue(rhs.value);
- }
-
- Byte8::Byte8(const Byte8 &rhs)
- {
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- Byte8::Byte8(const Reference<Byte8> &rhs)
- {
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- RValue<Byte8> Byte8::operator=(RValue<Byte8> rhs) const
- {
- storeValue(rhs.value);
-
- return rhs;
- }
-
- RValue<Byte8> Byte8::operator=(const Byte8 &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<Byte8>(value);
- }
-
- RValue<Byte8> Byte8::operator=(const Reference<Byte8> &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<Byte8>(value);
- }
-
- RValue<Byte8> operator+(RValue<Byte8> lhs, RValue<Byte8> rhs)
- {
- return RValue<Byte8>(Nucleus::createAdd(lhs.value, rhs.value));
- }
-
- RValue<Byte8> operator-(RValue<Byte8> lhs, RValue<Byte8> rhs)
- {
- return RValue<Byte8>(Nucleus::createSub(lhs.value, rhs.value));
- }
-
-// RValue<Byte8> operator*(RValue<Byte8> lhs, RValue<Byte8> rhs)
-// {
-// return RValue<Byte8>(Nucleus::createMul(lhs.value, rhs.value));
-// }
-
-// RValue<Byte8> operator/(RValue<Byte8> lhs, RValue<Byte8> rhs)
-// {
-// return RValue<Byte8>(Nucleus::createUDiv(lhs.value, rhs.value));
-// }
-
-// RValue<Byte8> operator%(RValue<Byte8> lhs, RValue<Byte8> rhs)
-// {
-// return RValue<Byte8>(Nucleus::createURem(lhs.value, rhs.value));
-// }
-
- RValue<Byte8> operator&(RValue<Byte8> lhs, RValue<Byte8> rhs)
- {
- return RValue<Byte8>(Nucleus::createAnd(lhs.value, rhs.value));
- }
-
- RValue<Byte8> operator|(RValue<Byte8> lhs, RValue<Byte8> rhs)
- {
- return RValue<Byte8>(Nucleus::createOr(lhs.value, rhs.value));
- }
-
- RValue<Byte8> operator^(RValue<Byte8> lhs, RValue<Byte8> rhs)
- {
- return RValue<Byte8>(Nucleus::createXor(lhs.value, rhs.value));
- }
-
-// RValue<Byte8> operator<<(RValue<Byte8> lhs, unsigned char rhs)
-// {
-// return RValue<Byte8>(Nucleus::createShl(lhs.value, C(::context->getConstantInt32(rhs))));
-// }
-
-// RValue<Byte8> operator>>(RValue<Byte8> lhs, unsigned char rhs)
-// {
-// return RValue<Byte8>(Nucleus::createLShr(lhs.value, C(::context->getConstantInt32(rhs))));
-// }
-
- RValue<Byte8> operator+=(const Byte8 &lhs, RValue<Byte8> rhs)
- {
- return lhs = lhs + rhs;
- }
-
- RValue<Byte8> operator-=(const Byte8 &lhs, RValue<Byte8> rhs)
- {
- return lhs = lhs - rhs;
- }
-
-// RValue<Byte8> operator*=(const Byte8 &lhs, RValue<Byte8> rhs)
-// {
-// return lhs = lhs * rhs;
-// }
-
-// RValue<Byte8> operator/=(const Byte8 &lhs, RValue<Byte8> rhs)
-// {
-// return lhs = lhs / rhs;
-// }
-
-// RValue<Byte8> operator%=(const Byte8 &lhs, RValue<Byte8> rhs)
-// {
-// return lhs = lhs % rhs;
-// }
-
- RValue<Byte8> operator&=(const Byte8 &lhs, RValue<Byte8> rhs)
- {
- return lhs = lhs & rhs;
- }
-
- RValue<Byte8> operator|=(const Byte8 &lhs, RValue<Byte8> rhs)
- {
- return lhs = lhs | rhs;
- }
-
- RValue<Byte8> operator^=(const Byte8 &lhs, RValue<Byte8> rhs)
- {
- return lhs = lhs ^ rhs;
- }
-
-// RValue<Byte8> operator<<=(const Byte8 &lhs, RValue<Byte8> rhs)
-// {
-// return lhs = lhs << rhs;
-// }
-
-// RValue<Byte8> operator>>=(const Byte8 &lhs, RValue<Byte8> rhs)
-// {
-// return lhs = lhs >> rhs;
-// }
-
-// RValue<Byte8> operator+(RValue<Byte8> val)
-// {
-// return val;
-// }
-
-// RValue<Byte8> operator-(RValue<Byte8> val)
-// {
-// return RValue<Byte8>(Nucleus::createNeg(val.value));
-// }
-
- RValue<Byte8> operator~(RValue<Byte8> val)
- {
- return RValue<Byte8>(Nucleus::createNot(val.value));
- }
-
- RValue<Byte8> AddSat(RValue<Byte8> x, RValue<Byte8> y)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Byte8>(V(nullptr));
- }
-
- RValue<Byte8> SubSat(RValue<Byte8> x, RValue<Byte8> y)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Byte8>(V(nullptr));
- }
-
- RValue<Short4> Unpack(RValue<Byte4> x)
- {
- int shuffle[16] = {0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7}; // Real type is v16i8
- return RValue<Short4>(Nucleus::createShuffleVector(x.value, x.value, shuffle));
- }
-
- RValue<Short4> UnpackLow(RValue<Byte8> x, RValue<Byte8> y)
- {
- int shuffle[16] = {0, 16, 1, 17, 2, 18, 3, 19, 4, 20, 5, 21, 6, 22, 7, 23}; // Real type is v16i8
- return RValue<Short4>(Nucleus::createShuffleVector(x.value, y.value, shuffle));
- }
-
- RValue<Short4> UnpackHigh(RValue<Byte8> x, RValue<Byte8> y)
- {
- int shuffle[16] = {0, 16, 1, 17, 2, 18, 3, 19, 4, 20, 5, 21, 6, 22, 7, 23}; // Real type is v16i8
- auto lowHigh = RValue<Byte16>(Nucleus::createShuffleVector(x.value, y.value, shuffle));
- return As<Short4>(Swizzle(As<Int4>(lowHigh), 0xEE));
- }
-
- RValue<Int> SignMask(RValue<Byte8> x)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Int>(V(nullptr));
- }
-
-// RValue<Byte8> CmpGT(RValue<Byte8> x, RValue<Byte8> y)
-// {
-// assert(false && "UNIMPLEMENTED"); return RValue<Byte8>(V(nullptr));
-// }
-
- RValue<Byte8> CmpEQ(RValue<Byte8> x, RValue<Byte8> y)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Byte8>(V(nullptr));
- }
-
- Type *Byte8::getType()
- {
- return T(Type_v8i8);
- }
-
- SByte8::SByte8()
- {
- // xyzw.parent = this;
- }
-
- SByte8::SByte8(uint8_t x0, uint8_t x1, uint8_t x2, uint8_t x3, uint8_t x4, uint8_t x5, uint8_t x6, uint8_t x7)
- {
- // xyzw.parent = this;
-
- int64_t constantVector[8] = { x0, x1, x2, x3, x4, x5, x6, x7 };
- Value *vector = V(Nucleus::createConstantVector(constantVector, getType()));
-
- storeValue(Nucleus::createBitCast(vector, getType()));
- }
-
- SByte8::SByte8(RValue<SByte8> rhs)
- {
- // xyzw.parent = this;
-
- storeValue(rhs.value);
- }
-
- SByte8::SByte8(const SByte8 &rhs)
- {
- // xyzw.parent = this;
-
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- SByte8::SByte8(const Reference<SByte8> &rhs)
- {
- // xyzw.parent = this;
-
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- RValue<SByte8> SByte8::operator=(RValue<SByte8> rhs) const
- {
- storeValue(rhs.value);
-
- return rhs;
- }
-
- RValue<SByte8> SByte8::operator=(const SByte8 &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<SByte8>(value);
- }
-
- RValue<SByte8> SByte8::operator=(const Reference<SByte8> &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<SByte8>(value);
- }
-
- RValue<SByte8> operator+(RValue<SByte8> lhs, RValue<SByte8> rhs)
- {
- return RValue<SByte8>(Nucleus::createAdd(lhs.value, rhs.value));
- }
-
- RValue<SByte8> operator-(RValue<SByte8> lhs, RValue<SByte8> rhs)
- {
- return RValue<SByte8>(Nucleus::createSub(lhs.value, rhs.value));
- }
-
-// RValue<SByte8> operator*(RValue<SByte8> lhs, RValue<SByte8> rhs)
-// {
-// return RValue<SByte8>(Nucleus::createMul(lhs.value, rhs.value));
-// }
-
-// RValue<SByte8> operator/(RValue<SByte8> lhs, RValue<SByte8> rhs)
-// {
-// return RValue<SByte8>(Nucleus::createSDiv(lhs.value, rhs.value));
-// }
-
-// RValue<SByte8> operator%(RValue<SByte8> lhs, RValue<SByte8> rhs)
-// {
-// return RValue<SByte8>(Nucleus::createSRem(lhs.value, rhs.value));
-// }
-
- RValue<SByte8> operator&(RValue<SByte8> lhs, RValue<SByte8> rhs)
- {
- return RValue<SByte8>(Nucleus::createAnd(lhs.value, rhs.value));
- }
-
- RValue<SByte8> operator|(RValue<SByte8> lhs, RValue<SByte8> rhs)
- {
- return RValue<SByte8>(Nucleus::createOr(lhs.value, rhs.value));
- }
-
- RValue<SByte8> operator^(RValue<SByte8> lhs, RValue<SByte8> rhs)
- {
- return RValue<SByte8>(Nucleus::createXor(lhs.value, rhs.value));
- }
-
-// RValue<SByte8> operator<<(RValue<SByte8> lhs, unsigned char rhs)
-// {
-// return RValue<SByte8>(Nucleus::createShl(lhs.value, C(::context->getConstantInt32(rhs))));
-// }
-
-// RValue<SByte8> operator>>(RValue<SByte8> lhs, unsigned char rhs)
-// {
-// return RValue<SByte8>(Nucleus::createAShr(lhs.value, C(::context->getConstantInt32(rhs))));
-// }
-
- RValue<SByte8> operator+=(const SByte8 &lhs, RValue<SByte8> rhs)
- {
- return lhs = lhs + rhs;
- }
-
- RValue<SByte8> operator-=(const SByte8 &lhs, RValue<SByte8> rhs)
- {
- return lhs = lhs - rhs;
- }
-
-// RValue<SByte8> operator*=(const SByte8 &lhs, RValue<SByte8> rhs)
-// {
-// return lhs = lhs * rhs;
-// }
-
-// RValue<SByte8> operator/=(const SByte8 &lhs, RValue<SByte8> rhs)
-// {
-// return lhs = lhs / rhs;
-// }
-
-// RValue<SByte8> operator%=(const SByte8 &lhs, RValue<SByte8> rhs)
-// {
-// return lhs = lhs % rhs;
-// }
-
- RValue<SByte8> operator&=(const SByte8 &lhs, RValue<SByte8> rhs)
- {
- return lhs = lhs & rhs;
- }
-
- RValue<SByte8> operator|=(const SByte8 &lhs, RValue<SByte8> rhs)
- {
- return lhs = lhs | rhs;
- }
-
- RValue<SByte8> operator^=(const SByte8 &lhs, RValue<SByte8> rhs)
- {
- return lhs = lhs ^ rhs;
- }
-
-// RValue<SByte8> operator<<=(const SByte8 &lhs, RValue<SByte8> rhs)
-// {
-// return lhs = lhs << rhs;
-// }
-
-// RValue<SByte8> operator>>=(const SByte8 &lhs, RValue<SByte8> rhs)
-// {
-// return lhs = lhs >> rhs;
-// }
-
-// RValue<SByte8> operator+(RValue<SByte8> val)
-// {
-// return val;
-// }
-
-// RValue<SByte8> operator-(RValue<SByte8> val)
-// {
-// return RValue<SByte8>(Nucleus::createNeg(val.value));
-// }
-
- RValue<SByte8> operator~(RValue<SByte8> val)
- {
- return RValue<SByte8>(Nucleus::createNot(val.value));
- }
-
- RValue<SByte8> AddSat(RValue<SByte8> x, RValue<SByte8> y)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<SByte8>(V(nullptr));
- }
-
- RValue<SByte8> SubSat(RValue<SByte8> x, RValue<SByte8> y)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<SByte8>(V(nullptr));
- }
-
- RValue<Short4> UnpackLow(RValue<SByte8> x, RValue<SByte8> y)
- {
- int shuffle[16] = {0, 16, 1, 17, 2, 18, 3, 19, 4, 20, 5, 21, 6, 22, 7, 23}; // Real type is v16i8
- return RValue<Short4>(Nucleus::createShuffleVector(x.value, y.value, shuffle));
- }
-
- RValue<Short4> UnpackHigh(RValue<SByte8> x, RValue<SByte8> y)
- {
- int shuffle[16] = {0, 16, 1, 17, 2, 18, 3, 19, 4, 20, 5, 21, 6, 22, 7, 23}; // Real type is v16i8
- auto lowHigh = RValue<Byte16>(Nucleus::createShuffleVector(x.value, y.value, shuffle));
- return As<Short4>(Swizzle(As<Int4>(lowHigh), 0xEE));
- }
-
- RValue<Int> SignMask(RValue<SByte8> x)
- {
- Ice::Variable *result = ::function->makeVariable(Ice::IceType_i32);
- const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::SignMask, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
- auto target = ::context->getConstantUndef(Ice::IceType_i32);
- auto movmsk = Ice::InstIntrinsicCall::create(::function, 1, result, target, intrinsic);
- movmsk->addArg(x.value);
- ::basicBlock->appendInst(movmsk);
-
- return RValue<Int>(V(result));
- }
-
- RValue<Byte8> CmpGT(RValue<SByte8> x, RValue<SByte8> y)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Byte8>(V(nullptr));
- }
-
- RValue<Byte8> CmpEQ(RValue<SByte8> x, RValue<SByte8> y)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Byte8>(V(nullptr));
- }
-
- Type *SByte8::getType()
- {
- return T(Type_v8i8);
- }
-
- Byte16::Byte16(RValue<Byte16> rhs)
- {
- // xyzw.parent = this;
-
- storeValue(rhs.value);
- }
-
- Byte16::Byte16(const Byte16 &rhs)
- {
- // xyzw.parent = this;
-
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- Byte16::Byte16(const Reference<Byte16> &rhs)
- {
- // xyzw.parent = this;
-
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- RValue<Byte16> Byte16::operator=(RValue<Byte16> rhs) const
- {
- storeValue(rhs.value);
-
- return rhs;
- }
-
- RValue<Byte16> Byte16::operator=(const Byte16 &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<Byte16>(value);
- }
-
- RValue<Byte16> Byte16::operator=(const Reference<Byte16> &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<Byte16>(value);
- }
-
- Type *Byte16::getType()
- {
- return T(Ice::IceType_v16i8);
- }
-
- Type *SByte16::getType()
- {
- return T(Ice::IceType_v16i8);
- }
-
- Short2::Short2(RValue<Short4> cast)
- {
- assert(false && "UNIMPLEMENTED");
- }
-
- Type *Short2::getType()
- {
- return T(Type_v2i16);
- }
-
- UShort2::UShort2(RValue<UShort4> cast)
- {
- assert(false && "UNIMPLEMENTED");
- }
-
- Type *UShort2::getType()
- {
- return T(Type_v2i16);
- }
-
- Short4::Short4(RValue<Int> cast)
- {
- Value *extend = Nucleus::createZExt(cast.value, Long::getType());
- Value *swizzle = Swizzle(RValue<Short4>(extend), 0x00).value;
-
- storeValue(swizzle);
- }
-
- Short4::Short4(RValue<Int4> cast)
- {
- assert(false && "UNIMPLEMENTED");
- }
-
-// Short4::Short4(RValue<Float> cast)
-// {
-// }
-
- Short4::Short4(RValue<Float4> cast)
- {
- assert(false && "UNIMPLEMENTED");
- }
-
- Short4::Short4()
- {
- // xyzw.parent = this;
- }
-
- Short4::Short4(short xyzw)
- {
- // xyzw.parent = this;
-
- int64_t constantVector[4] = {xyzw, xyzw, xyzw, xyzw};
- storeValue(Nucleus::createConstantVector(constantVector, getType()));
- }
-
- Short4::Short4(short x, short y, short z, short w)
- {
- // xyzw.parent = this;
-
- int64_t constantVector[4] = {x, y, z, w};
- storeValue(Nucleus::createConstantVector(constantVector, getType()));
- }
-
- Short4::Short4(RValue<Short4> rhs)
- {
- // xyzw.parent = this;
-
- storeValue(rhs.value);
- }
-
- Short4::Short4(const Short4 &rhs)
- {
- // xyzw.parent = this;
-
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- Short4::Short4(const Reference<Short4> &rhs)
- {
- // xyzw.parent = this;
-
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- Short4::Short4(RValue<UShort4> rhs)
- {
- // xyzw.parent = this;
-
- storeValue(rhs.value);
- }
-
- Short4::Short4(const UShort4 &rhs)
- {
- // xyzw.parent = this;
-
- storeValue(rhs.loadValue());
- }
-
- Short4::Short4(const Reference<UShort4> &rhs)
- {
- // xyzw.parent = this;
-
- storeValue(rhs.loadValue());
- }
-
- RValue<Short4> Short4::operator=(RValue<Short4> rhs) const
- {
- storeValue(rhs.value);
-
- return rhs;
- }
-
- RValue<Short4> Short4::operator=(const Short4 &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<Short4>(value);
- }
-
- RValue<Short4> Short4::operator=(const Reference<Short4> &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<Short4>(value);
- }
-
- RValue<Short4> Short4::operator=(RValue<UShort4> rhs) const
- {
- storeValue(rhs.value);
-
- return RValue<Short4>(rhs);
- }
-
- RValue<Short4> Short4::operator=(const UShort4 &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<Short4>(value);
- }
-
- RValue<Short4> Short4::operator=(const Reference<UShort4> &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<Short4>(value);
- }
-
- RValue<Short4> operator+(RValue<Short4> lhs, RValue<Short4> rhs)
- {
- return RValue<Short4>(Nucleus::createAdd(lhs.value, rhs.value));
- }
-
- RValue<Short4> operator-(RValue<Short4> lhs, RValue<Short4> rhs)
- {
- return RValue<Short4>(Nucleus::createSub(lhs.value, rhs.value));
- }
-
- RValue<Short4> operator*(RValue<Short4> lhs, RValue<Short4> rhs)
- {
- return RValue<Short4>(Nucleus::createMul(lhs.value, rhs.value));
- }
-
-// RValue<Short4> operator/(RValue<Short4> lhs, RValue<Short4> rhs)
-// {
-// return RValue<Short4>(Nucleus::createSDiv(lhs.value, rhs.value));
-// }
-
-// RValue<Short4> operator%(RValue<Short4> lhs, RValue<Short4> rhs)
-// {
-// return RValue<Short4>(Nucleus::createSRem(lhs.value, rhs.value));
-// }
-
- RValue<Short4> operator&(RValue<Short4> lhs, RValue<Short4> rhs)
- {
- return RValue<Short4>(Nucleus::createAnd(lhs.value, rhs.value));
- }
-
- RValue<Short4> operator|(RValue<Short4> lhs, RValue<Short4> rhs)
- {
- return RValue<Short4>(Nucleus::createOr(lhs.value, rhs.value));
- }
-
- RValue<Short4> operator^(RValue<Short4> lhs, RValue<Short4> rhs)
- {
- return RValue<Short4>(Nucleus::createXor(lhs.value, rhs.value));
- }
-
- RValue<Short4> operator<<(RValue<Short4> lhs, unsigned char rhs)
- {
- return RValue<Short4>(Nucleus::createShl(lhs.value, C(::context->getConstantInt32(rhs))));
- }
-
- RValue<Short4> operator>>(RValue<Short4> lhs, unsigned char rhs)
- {
- return RValue<Short4>(Nucleus::createAShr(lhs.value, C(::context->getConstantInt32(rhs))));
- }
-
- RValue<Short4> operator<<(RValue<Short4> lhs, RValue<Long1> rhs)
- {
- // return RValue<Short4>(Nucleus::createShl(lhs.value, rhs.value));
-
- assert(false && "UNIMPLEMENTED"); return RValue<Short4>(V(nullptr));
- }
-
- RValue<Short4> operator>>(RValue<Short4> lhs, RValue<Long1> rhs)
- {
- // return RValue<Short4>(Nucleus::createAShr(lhs.value, rhs.value));
-
- assert(false && "UNIMPLEMENTED"); return RValue<Short4>(V(nullptr));
- }
-
- RValue<Short4> operator+=(const Short4 &lhs, RValue<Short4> rhs)
- {
- return lhs = lhs + rhs;
- }
-
- RValue<Short4> operator-=(const Short4 &lhs, RValue<Short4> rhs)
- {
- return lhs = lhs - rhs;
- }
-
- RValue<Short4> operator*=(const Short4 &lhs, RValue<Short4> rhs)
- {
- return lhs = lhs * rhs;
- }
-
-// RValue<Short4> operator/=(const Short4 &lhs, RValue<Short4> rhs)
-// {
-// return lhs = lhs / rhs;
-// }
-
-// RValue<Short4> operator%=(const Short4 &lhs, RValue<Short4> rhs)
-// {
-// return lhs = lhs % rhs;
-// }
-
- RValue<Short4> operator&=(const Short4 &lhs, RValue<Short4> rhs)
- {
- return lhs = lhs & rhs;
- }
-
- RValue<Short4> operator|=(const Short4 &lhs, RValue<Short4> rhs)
- {
- return lhs = lhs | rhs;
- }
-
- RValue<Short4> operator^=(const Short4 &lhs, RValue<Short4> rhs)
- {
- return lhs = lhs ^ rhs;
- }
-
- RValue<Short4> operator<<=(const Short4 &lhs, unsigned char rhs)
- {
- return lhs = lhs << rhs;
- }
-
- RValue<Short4> operator>>=(const Short4 &lhs, unsigned char rhs)
- {
- return lhs = lhs >> rhs;
- }
-
- RValue<Short4> operator<<=(const Short4 &lhs, RValue<Long1> rhs)
- {
- return lhs = lhs << rhs;
- }
-
- RValue<Short4> operator>>=(const Short4 &lhs, RValue<Long1> rhs)
- {
- return lhs = lhs >> rhs;
- }
-
-// RValue<Short4> operator+(RValue<Short4> val)
-// {
-// return val;
-// }
-
- RValue<Short4> operator-(RValue<Short4> val)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Short4>(V(nullptr));
- }
-
- RValue<Short4> operator~(RValue<Short4> val)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Short4>(V(nullptr));
- }
-
- RValue<Short4> RoundShort4(RValue<Float4> cast)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Short4>(V(nullptr));
- }
-
- RValue<Short4> Max(RValue<Short4> x, RValue<Short4> y)
- {
- Ice::Variable *condition = ::function->makeVariable(Ice::IceType_v8i1);
- auto cmp = Ice::InstIcmp::create(::function, Ice::InstIcmp::Sle, condition, x.value, y.value);
- ::basicBlock->appendInst(cmp);
-
- Ice::Variable *result = ::function->makeVariable(Ice::IceType_v8i16);
- auto select = Ice::InstSelect::create(::function, result, condition, y.value, x.value);
- ::basicBlock->appendInst(select);
-
- return RValue<Short4>(V(result));
- }
-
- RValue<Short4> Min(RValue<Short4> x, RValue<Short4> y)
- {
- Ice::Variable *condition = ::function->makeVariable(Ice::IceType_v8i1);
- auto cmp = Ice::InstIcmp::create(::function, Ice::InstIcmp::Sgt, condition, x.value, y.value);
- ::basicBlock->appendInst(cmp);
-
- Ice::Variable *result = ::function->makeVariable(Ice::IceType_v8i16);
- auto select = Ice::InstSelect::create(::function, result, condition, y.value, x.value);
- ::basicBlock->appendInst(select);
-
- return RValue<Short4>(V(result));
- }
-
- RValue<Short4> AddSat(RValue<Short4> x, RValue<Short4> y)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Short4>(V(nullptr));
- }
-
- RValue<Short4> SubSat(RValue<Short4> x, RValue<Short4> y)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Short4>(V(nullptr));
- }
-
- RValue<Short4> MulHigh(RValue<Short4> x, RValue<Short4> y)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Short4>(V(nullptr));
- }
-
- RValue<Int2> MulAdd(RValue<Short4> x, RValue<Short4> y)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Int2>(V(nullptr));
- }
-
- RValue<SByte8> Pack(RValue<Short4> x, RValue<Short4> y)
- {
- Ice::Variable *result = ::function->makeVariable(Ice::IceType_v16i8);
- const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::VectorPackSigned, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
- auto target = ::context->getConstantUndef(Ice::IceType_i32);
- auto pack = Ice::InstIntrinsicCall::create(::function, 2, result, target, intrinsic);
- pack->addArg(x.value);
- pack->addArg(y.value);
- ::basicBlock->appendInst(pack);
-
- return As<SByte8>(Swizzle(As<Int4>(V(result)), 0x88));
- }
-
- RValue<Int2> UnpackLow(RValue<Short4> x, RValue<Short4> y)
- {
- int shuffle[8] = {0, 8, 1, 9, 2, 10, 3, 11}; // Real type is v8i16
- return RValue<Int2>(Nucleus::createShuffleVector(x.value, y.value, shuffle));
- }
-
- RValue<Int2> UnpackHigh(RValue<Short4> x, RValue<Short4> y)
- {
- int shuffle[8] = {0, 8, 1, 9, 2, 10, 3, 11}; // Real type is v8i16
- auto lowHigh = RValue<Short8>(Nucleus::createShuffleVector(x.value, y.value, shuffle));
- return As<Int2>(Swizzle(As<Int4>(lowHigh), 0xEE));
- }
-
- RValue<Short4> Swizzle(RValue<Short4> x, unsigned char select)
- {
- // Real type is v8i16
- int shuffle[8] =
- {
- (select >> 0) & 0x03,
- (select >> 2) & 0x03,
- (select >> 4) & 0x03,
- (select >> 6) & 0x03,
- (select >> 0) & 0x03,
- (select >> 2) & 0x03,
- (select >> 4) & 0x03,
- (select >> 6) & 0x03,
- };
-
- return RValue<Short4>(Nucleus::createShuffleVector(x.value, x.value, shuffle));
- }
-
- RValue<Short4> Insert(RValue<Short4> val, RValue<Short> element, int i)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Short4>(V(nullptr));
- }
-
- RValue<Short> Extract(RValue<Short4> val, int i)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Short>(V(nullptr));
- }
-
- RValue<Short4> CmpGT(RValue<Short4> x, RValue<Short4> y)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Short4>(V(nullptr));
- }
-
- RValue<Short4> CmpEQ(RValue<Short4> x, RValue<Short4> y)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Short4>(V(nullptr));
- }
-
- Type *Short4::getType()
- {
- return T(Type_v4i16);
- }
-
- UShort4::UShort4(RValue<Int4> cast)
- {
- *this = Short4(cast);
- }
-
- UShort4::UShort4(RValue<Float4> cast, bool saturate)
- {
- assert(false && "UNIMPLEMENTED");
- }
-
- UShort4::UShort4()
- {
- // xyzw.parent = this;
- }
-
- UShort4::UShort4(unsigned short xyzw)
- {
- // xyzw.parent = this;
-
- int64_t constantVector[4] = {xyzw, xyzw, xyzw, xyzw};
- storeValue(Nucleus::createConstantVector(constantVector, getType()));
- }
-
- UShort4::UShort4(unsigned short x, unsigned short y, unsigned short z, unsigned short w)
- {
- // xyzw.parent = this;
-
- int64_t constantVector[4] = {x, y, z, w};
- storeValue(Nucleus::createConstantVector(constantVector, getType()));
- }
-
- UShort4::UShort4(RValue<UShort4> rhs)
- {
- // xyzw.parent = this;
-
- storeValue(rhs.value);
- }
-
- UShort4::UShort4(const UShort4 &rhs)
- {
- // xyzw.parent = this;
-
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- UShort4::UShort4(const Reference<UShort4> &rhs)
- {
- // xyzw.parent = this;
-
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- UShort4::UShort4(RValue<Short4> rhs)
- {
- // xyzw.parent = this;
-
- storeValue(rhs.value);
- }
-
- UShort4::UShort4(const Short4 &rhs)
- {
- // xyzw.parent = this;
-
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- UShort4::UShort4(const Reference<Short4> &rhs)
- {
- // xyzw.parent = this;
-
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- RValue<UShort4> UShort4::operator=(RValue<UShort4> rhs) const
- {
- storeValue(rhs.value);
-
- return rhs;
- }
-
- RValue<UShort4> UShort4::operator=(const UShort4 &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<UShort4>(value);
- }
-
- RValue<UShort4> UShort4::operator=(const Reference<UShort4> &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<UShort4>(value);
- }
-
- RValue<UShort4> UShort4::operator=(RValue<Short4> rhs) const
- {
- storeValue(rhs.value);
-
- return RValue<UShort4>(rhs);
- }
-
- RValue<UShort4> UShort4::operator=(const Short4 &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<UShort4>(value);
- }
-
- RValue<UShort4> UShort4::operator=(const Reference<Short4> &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<UShort4>(value);
- }
-
- RValue<UShort4> operator+(RValue<UShort4> lhs, RValue<UShort4> rhs)
- {
- return RValue<Short4>(Nucleus::createAdd(lhs.value, rhs.value));
- }
-
- RValue<UShort4> operator-(RValue<UShort4> lhs, RValue<UShort4> rhs)
- {
- return RValue<UShort4>(Nucleus::createSub(lhs.value, rhs.value));
- }
-
- RValue<UShort4> operator*(RValue<UShort4> lhs, RValue<UShort4> rhs)
- {
- return RValue<UShort4>(Nucleus::createMul(lhs.value, rhs.value));
- }
-
- RValue<UShort4> operator&(RValue<UShort4> lhs, RValue<UShort4> rhs)
- {
- return RValue<UShort4>(Nucleus::createAnd(lhs.value, rhs.value));
- }
-
- RValue<UShort4> operator|(RValue<UShort4> lhs, RValue<UShort4> rhs)
- {
- return RValue<UShort4>(Nucleus::createOr(lhs.value, rhs.value));
- }
-
- RValue<UShort4> operator^(RValue<UShort4> lhs, RValue<UShort4> rhs)
- {
- return RValue<UShort4>(Nucleus::createXor(lhs.value, rhs.value));
- }
-
- RValue<UShort4> operator<<(RValue<UShort4> lhs, unsigned char rhs)
- {
- return RValue<UShort4>(Nucleus::createShl(lhs.value, C(::context->getConstantInt32(rhs))));
- }
-
- RValue<UShort4> operator>>(RValue<UShort4> lhs, unsigned char rhs)
- {
- return RValue<UShort4>(Nucleus::createLShr(lhs.value, C(::context->getConstantInt32(rhs))));
- }
-
- RValue<UShort4> operator<<(RValue<UShort4> lhs, RValue<Long1> rhs)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<UShort4>(V(nullptr));
- }
-
- RValue<UShort4> operator>>(RValue<UShort4> lhs, RValue<Long1> rhs)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<UShort4>(V(nullptr));
- }
-
- RValue<UShort4> operator<<=(const UShort4 &lhs, unsigned char rhs)
- {
- return lhs = lhs << rhs;
- }
-
- RValue<UShort4> operator>>=(const UShort4 &lhs, unsigned char rhs)
- {
- return lhs = lhs >> rhs;
- }
-
- RValue<UShort4> operator<<=(const UShort4 &lhs, RValue<Long1> rhs)
- {
- return lhs = lhs << rhs;
- }
-
- RValue<UShort4> operator>>=(const UShort4 &lhs, RValue<Long1> rhs)
- {
- return lhs = lhs >> rhs;
- }
-
- RValue<UShort4> operator~(RValue<UShort4> val)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<UShort4>(V(nullptr));
- }
-
- RValue<UShort4> Max(RValue<UShort4> x, RValue<UShort4> y)
- {
- Ice::Variable *condition = ::function->makeVariable(Ice::IceType_v8i1);
- auto cmp = Ice::InstIcmp::create(::function, Ice::InstIcmp::Ule, condition, x.value, y.value);
- ::basicBlock->appendInst(cmp);
-
- Ice::Variable *result = ::function->makeVariable(Ice::IceType_v8i16);
- auto select = Ice::InstSelect::create(::function, result, condition, y.value, x.value);
- ::basicBlock->appendInst(select);
-
- return RValue<UShort4>(V(result));
- }
-
- RValue<UShort4> Min(RValue<UShort4> x, RValue<UShort4> y)
- {
- Ice::Variable *condition = ::function->makeVariable(Ice::IceType_v8i1);
- auto cmp = Ice::InstIcmp::create(::function, Ice::InstIcmp::Ugt, condition, x.value, y.value);
- ::basicBlock->appendInst(cmp);
-
- Ice::Variable *result = ::function->makeVariable(Ice::IceType_v8i16);
- auto select = Ice::InstSelect::create(::function, result, condition, y.value, x.value);
- ::basicBlock->appendInst(select);
-
- return RValue<UShort4>(V(result));
- }
-
- RValue<UShort4> AddSat(RValue<UShort4> x, RValue<UShort4> y)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<UShort4>(V(nullptr));
- }
-
- RValue<UShort4> SubSat(RValue<UShort4> x, RValue<UShort4> y)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<UShort4>(V(nullptr));
- }
-
- RValue<UShort4> MulHigh(RValue<UShort4> x, RValue<UShort4> y)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<UShort4>(V(nullptr));
- }
-
- RValue<UShort4> Average(RValue<UShort4> x, RValue<UShort4> y)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<UShort4>(V(nullptr));
- }
-
- RValue<Byte8> Pack(RValue<UShort4> x, RValue<UShort4> y)
- {
- Ice::Variable *result = ::function->makeVariable(Ice::IceType_v16i8);
- const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::VectorPackUnsigned, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
- auto target = ::context->getConstantUndef(Ice::IceType_i32);
- auto pack = Ice::InstIntrinsicCall::create(::function, 2, result, target, intrinsic);
- pack->addArg(x.value);
- pack->addArg(y.value);
- ::basicBlock->appendInst(pack);
-
- return As<Byte8>(Swizzle(As<Int4>(V(result)), 0x88));
- }
-
- Type *UShort4::getType()
- {
- return T(Type_v4i16);
- }
-
- Short8::Short8(short c0, short c1, short c2, short c3, short c4, short c5, short c6, short c7)
- {
- // xyzw.parent = this;
-
- int64_t constantVector[8] = {c0, c1, c2, c3, c4, c5, c6, c7};
- storeValue(Nucleus::createConstantVector(constantVector, getType()));
- }
-
- Short8::Short8(RValue<Short8> rhs)
- {
- // xyzw.parent = this;
-
- storeValue(rhs.value);
- }
-
- Short8::Short8(const Reference<Short8> &rhs)
- {
- // xyzw.parent = this;
-
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- Short8::Short8(RValue<Short4> lo, RValue<Short4> hi)
- {
- assert(false && "UNIMPLEMENTED");
- }
-
- RValue<Short8> operator+(RValue<Short8> lhs, RValue<Short8> rhs)
- {
- return RValue<Short8>(Nucleus::createAdd(lhs.value, rhs.value));
- }
-
- RValue<Short8> operator&(RValue<Short8> lhs, RValue<Short8> rhs)
- {
- return RValue<Short8>(Nucleus::createAnd(lhs.value, rhs.value));
- }
-
- RValue<Short8> operator<<(RValue<Short8> lhs, unsigned char rhs)
- {
- return RValue<Short8>(Nucleus::createShl(lhs.value, C(::context->getConstantInt32(rhs))));
- }
-
- RValue<Short8> operator>>(RValue<Short8> lhs, unsigned char rhs)
- {
- return RValue<Short8>(Nucleus::createAShr(lhs.value, C(::context->getConstantInt32(rhs))));
- }
-
- RValue<Int4> MulAdd(RValue<Short8> x, RValue<Short8> y)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Int4>(V(nullptr));
- }
-
- RValue<Int4> Abs(RValue<Int4> x)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Int4>(V(nullptr));
- }
-
- RValue<Short8> MulHigh(RValue<Short8> x, RValue<Short8> y)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Short8>(V(nullptr));
- }
-
- Type *Short8::getType()
- {
- return T(Ice::IceType_v8i16);
- }
-
- UShort8::UShort8(unsigned short c0, unsigned short c1, unsigned short c2, unsigned short c3, unsigned short c4, unsigned short c5, unsigned short c6, unsigned short c7)
- {
- int64_t constantVector[8] = {c0, c1, c2, c3, c4, c5, c6, c7};
- storeValue(Nucleus::createConstantVector(constantVector, getType()));
- }
-
- UShort8::UShort8(RValue<UShort8> rhs)
- {
- storeValue(rhs.value);
- }
-
- UShort8::UShort8(const Reference<UShort8> &rhs)
- {
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- UShort8::UShort8(RValue<UShort4> lo, RValue<UShort4> hi)
- {
- assert(false && "UNIMPLEMENTED");
- }
-
- RValue<UShort8> UShort8::operator=(RValue<UShort8> rhs) const
- {
- storeValue(rhs.value);
-
- return rhs;
- }
-
- RValue<UShort8> UShort8::operator=(const UShort8 &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<UShort8>(value);
- }
-
- RValue<UShort8> UShort8::operator=(const Reference<UShort8> &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<UShort8>(value);
- }
-
- RValue<UShort8> operator&(RValue<UShort8> lhs, RValue<UShort8> rhs)
- {
- return RValue<UShort8>(Nucleus::createAnd(lhs.value, rhs.value));
- }
-
- RValue<UShort8> operator<<(RValue<UShort8> lhs, unsigned char rhs)
- {
- return RValue<UShort8>(Nucleus::createShl(lhs.value, C(::context->getConstantInt32(rhs))));
- }
-
- RValue<UShort8> operator>>(RValue<UShort8> lhs, unsigned char rhs)
- {
- return RValue<UShort8>(Nucleus::createLShr(lhs.value, C(::context->getConstantInt32(rhs))));
- }
-
- RValue<UShort8> operator+(RValue<UShort8> lhs, RValue<UShort8> rhs)
- {
- return RValue<UShort8>(Nucleus::createAdd(lhs.value, rhs.value));
- }
-
- RValue<UShort8> operator*(RValue<UShort8> lhs, RValue<UShort8> rhs)
- {
- return RValue<UShort8>(Nucleus::createMul(lhs.value, rhs.value));
- }
-
- RValue<UShort8> operator+=(const UShort8 &lhs, RValue<UShort8> rhs)
- {
- return lhs = lhs + rhs;
- }
-
- RValue<UShort8> operator~(RValue<UShort8> val)
- {
- return RValue<UShort8>(Nucleus::createNot(val.value));
- }
-
- RValue<UShort8> Swizzle(RValue<UShort8> x, char select0, char select1, char select2, char select3, char select4, char select5, char select6, char select7)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<UShort8>(V(nullptr));
- }
-
- RValue<UShort8> MulHigh(RValue<UShort8> x, RValue<UShort8> y)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<UShort8>(V(nullptr));
- }
-
- // FIXME: Implement as Shuffle(x, y, Select(i0, ..., i16)) and Shuffle(x, y, SELECT_PACK_REPEAT(element))
-// RValue<UShort8> PackRepeat(RValue<Byte16> x, RValue<Byte16> y, int element)
-// {
-// assert(false && "UNIMPLEMENTED"); return RValue<UShort8>(V(nullptr));
-// }
-
- Type *UShort8::getType()
- {
- return T(Ice::IceType_v8i16);
- }
-
- Int::Int(Argument<Int> argument)
- {
- storeValue(argument.value);
- }
-
- Int::Int(RValue<Byte> cast)
- {
- Value *integer = Nucleus::createZExt(cast.value, Int::getType());
-
- storeValue(integer);
- }
-
- Int::Int(RValue<SByte> cast)
- {
- Value *integer = Nucleus::createSExt(cast.value, Int::getType());
-
- storeValue(integer);
- }
-
- Int::Int(RValue<Short> cast)
- {
- Value *integer = Nucleus::createSExt(cast.value, Int::getType());
-
- storeValue(integer);
- }
-
- Int::Int(RValue<UShort> cast)
- {
- Value *integer = Nucleus::createZExt(cast.value, Int::getType());
-
- storeValue(integer);
- }
-
- Int::Int(RValue<Int2> cast)
- {
- *this = Extract(cast, 0);
- }
-
- Int::Int(RValue<Long> cast)
- {
- Value *integer = Nucleus::createTrunc(cast.value, Int::getType());
-
- storeValue(integer);
- }
-
- Int::Int(RValue<Float> cast)
- {
- Value *integer = Nucleus::createFPToSI(cast.value, Int::getType());
-
- storeValue(integer);
- }
-
- Int::Int()
- {
- }
-
- Int::Int(int x)
- {
- storeValue(Nucleus::createConstantInt(x));
- }
-
- Int::Int(RValue<Int> rhs)
- {
- storeValue(rhs.value);
- }
-
- Int::Int(RValue<UInt> rhs)
- {
- storeValue(rhs.value);
- }
-
- Int::Int(const Int &rhs)
- {
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- Int::Int(const Reference<Int> &rhs)
- {
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- Int::Int(const UInt &rhs)
- {
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- Int::Int(const Reference<UInt> &rhs)
- {
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- RValue<Int> Int::operator=(int rhs) const
- {
- return RValue<Int>(storeValue(Nucleus::createConstantInt(rhs)));
- }
-
- RValue<Int> Int::operator=(RValue<Int> rhs) const
- {
- storeValue(rhs.value);
-
- return rhs;
- }
-
- RValue<Int> Int::operator=(RValue<UInt> rhs) const
- {
- storeValue(rhs.value);
-
- return RValue<Int>(rhs);
- }
-
- RValue<Int> Int::operator=(const Int &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<Int>(value);
- }
-
- RValue<Int> Int::operator=(const Reference<Int> &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<Int>(value);
- }
-
- RValue<Int> Int::operator=(const UInt &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<Int>(value);
- }
-
- RValue<Int> Int::operator=(const Reference<UInt> &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<Int>(value);
- }
-
- RValue<Int> operator+(RValue<Int> lhs, RValue<Int> rhs)
- {
- return RValue<Int>(Nucleus::createAdd(lhs.value, rhs.value));
- }
-
- RValue<Int> operator-(RValue<Int> lhs, RValue<Int> rhs)
- {
- return RValue<Int>(Nucleus::createSub(lhs.value, rhs.value));
- }
-
- RValue<Int> operator*(RValue<Int> lhs, RValue<Int> rhs)
- {
- return RValue<Int>(Nucleus::createMul(lhs.value, rhs.value));
- }
-
- RValue<Int> operator/(RValue<Int> lhs, RValue<Int> rhs)
- {
- return RValue<Int>(Nucleus::createSDiv(lhs.value, rhs.value));
- }
-
- RValue<Int> operator%(RValue<Int> lhs, RValue<Int> rhs)
- {
- return RValue<Int>(Nucleus::createSRem(lhs.value, rhs.value));
- }
-
- RValue<Int> operator&(RValue<Int> lhs, RValue<Int> rhs)
- {
- return RValue<Int>(Nucleus::createAnd(lhs.value, rhs.value));
- }
-
- RValue<Int> operator|(RValue<Int> lhs, RValue<Int> rhs)
- {
- return RValue<Int>(Nucleus::createOr(lhs.value, rhs.value));
- }
-
- RValue<Int> operator^(RValue<Int> lhs, RValue<Int> rhs)
- {
- return RValue<Int>(Nucleus::createXor(lhs.value, rhs.value));
- }
-
- RValue<Int> operator<<(RValue<Int> lhs, RValue<Int> rhs)
- {
- return RValue<Int>(Nucleus::createShl(lhs.value, rhs.value));
- }
-
- RValue<Int> operator>>(RValue<Int> lhs, RValue<Int> rhs)
- {
- return RValue<Int>(Nucleus::createAShr(lhs.value, rhs.value));
- }
-
- RValue<Int> operator+=(const Int &lhs, RValue<Int> rhs)
- {
- return lhs = lhs + rhs;
- }
-
- RValue<Int> operator-=(const Int &lhs, RValue<Int> rhs)
- {
- return lhs = lhs - rhs;
- }
-
- RValue<Int> operator*=(const Int &lhs, RValue<Int> rhs)
- {
- return lhs = lhs * rhs;
- }
-
- RValue<Int> operator/=(const Int &lhs, RValue<Int> rhs)
- {
- return lhs = lhs / rhs;
- }
-
- RValue<Int> operator%=(const Int &lhs, RValue<Int> rhs)
- {
- return lhs = lhs % rhs;
- }
-
- RValue<Int> operator&=(const Int &lhs, RValue<Int> rhs)
- {
- return lhs = lhs & rhs;
- }
-
- RValue<Int> operator|=(const Int &lhs, RValue<Int> rhs)
- {
- return lhs = lhs | rhs;
- }
-
- RValue<Int> operator^=(const Int &lhs, RValue<Int> rhs)
- {
- return lhs = lhs ^ rhs;
- }
-
- RValue<Int> operator<<=(const Int &lhs, RValue<Int> rhs)
- {
- return lhs = lhs << rhs;
- }
-
- RValue<Int> operator>>=(const Int &lhs, RValue<Int> rhs)
- {
- return lhs = lhs >> rhs;
- }
-
- RValue<Int> operator+(RValue<Int> val)
- {
- return val;
- }
-
- RValue<Int> operator-(RValue<Int> val)
- {
- return RValue<Int>(Nucleus::createNeg(val.value));
- }
-
- RValue<Int> operator~(RValue<Int> val)
- {
- return RValue<Int>(Nucleus::createNot(val.value));
- }
-
- RValue<Int> operator++(const Int &val, int) // Post-increment
- {
- auto oldValue = val.loadValue();
- auto newValue = ::function->makeVariable(Ice::IceType_i32);
- auto inc = Ice::InstArithmetic::create(::function, Ice::InstArithmetic::Add, newValue, oldValue, ::context->getConstantInt32(1));
- ::basicBlock->appendInst(inc);
- val.storeValue(V(newValue));
-
- return RValue<Int>(oldValue);
- }
-
- const Int &operator++(const Int &val) // Pre-increment
- {
- assert(false && "UNIMPLEMENTED"); return val;
- }
-
- RValue<Int> operator--(const Int &val, int) // Post-decrement
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Int>(V(nullptr));
- }
-
- const Int &operator--(const Int &val) // Pre-decrement
- {
- assert(false && "UNIMPLEMENTED"); return val;
- }
-
- RValue<Bool> operator<(RValue<Int> lhs, RValue<Int> rhs)
- {
- return RValue<Bool>(Nucleus::createICmpSLT(lhs.value, rhs.value));
- }
-
- RValue<Bool> operator<=(RValue<Int> lhs, RValue<Int> rhs)
- {
- return RValue<Bool>(Nucleus::createICmpSLE(lhs.value, rhs.value));
- }
-
- RValue<Bool> operator>(RValue<Int> lhs, RValue<Int> rhs)
- {
- return RValue<Bool>(Nucleus::createICmpSGT(lhs.value, rhs.value));
- }
-
- RValue<Bool> operator>=(RValue<Int> lhs, RValue<Int> rhs)
- {
- return RValue<Bool>(Nucleus::createICmpSGE(lhs.value, rhs.value));
- }
-
- RValue<Bool> operator!=(RValue<Int> lhs, RValue<Int> rhs)
- {
- return RValue<Bool>(Nucleus::createICmpNE(lhs.value, rhs.value));
- }
-
- RValue<Bool> operator==(RValue<Int> lhs, RValue<Int> rhs)
- {
- return RValue<Bool>(Nucleus::createICmpEQ(lhs.value, rhs.value));
- }
-
- RValue<Int> Max(RValue<Int> x, RValue<Int> y)
- {
- return IfThenElse(x > y, x, y);
- }
-
- RValue<Int> Min(RValue<Int> x, RValue<Int> y)
- {
- return IfThenElse(x < y, x, y);
- }
-
- RValue<Int> Clamp(RValue<Int> x, RValue<Int> min, RValue<Int> max)
- {
- return Min(Max(x, min), max);
- }
-
- RValue<Int> RoundInt(RValue<Float> cast)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Int>(V(nullptr));
- }
-
- Type *Int::getType()
- {
- return T(Ice::IceType_i32);
- }
-
- Long::Long(RValue<Int> cast)
- {
- Value *integer = Nucleus::createSExt(cast.value, Long::getType());
-
- storeValue(integer);
- }
-
- Long::Long(RValue<UInt> cast)
- {
- Value *integer = Nucleus::createZExt(cast.value, Long::getType());
-
- storeValue(integer);
- }
-
- Long::Long()
- {
- }
-
- Long::Long(RValue<Long> rhs)
- {
- storeValue(rhs.value);
- }
-
- RValue<Long> Long::operator=(int64_t rhs) const
- {
- return RValue<Long>(storeValue(Nucleus::createConstantLong(rhs)));
- }
-
- RValue<Long> Long::operator=(RValue<Long> rhs) const
- {
- storeValue(rhs.value);
-
- return rhs;
- }
-
- RValue<Long> Long::operator=(const Long &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<Long>(value);
- }
-
- RValue<Long> Long::operator=(const Reference<Long> &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<Long>(value);
- }
-
- RValue<Long> operator+(RValue<Long> lhs, RValue<Long> rhs)
- {
- return RValue<Long>(Nucleus::createAdd(lhs.value, rhs.value));
- }
-
- RValue<Long> operator-(RValue<Long> lhs, RValue<Long> rhs)
- {
- return RValue<Long>(Nucleus::createSub(lhs.value, rhs.value));
- }
-
- RValue<Long> operator+=(const Long &lhs, RValue<Long> rhs)
- {
- return lhs = lhs + rhs;
- }
-
- RValue<Long> operator-=(const Long &lhs, RValue<Long> rhs)
- {
- return lhs = lhs - rhs;
- }
-
- RValue<Long> AddAtomic(RValue<Pointer<Long> > x, RValue<Long> y)
- {
- return RValue<Long>(Nucleus::createAtomicAdd(x.value, y.value));
- }
-
- Type *Long::getType()
- {
- return T(Ice::IceType_i64);
- }
-
- Long1::Long1(const RValue<UInt> cast)
- {
- assert(false && "UNIMPLEMENTED");
- }
-
- Long1::Long1(RValue<Long1> rhs)
- {
- storeValue(rhs.value);
- }
-
- Type *Long1::getType()
- {
- assert(false && "UNIMPLEMENTED"); return nullptr;
- }
-
- UInt::UInt(Argument<UInt> argument)
- {
- storeValue(argument.value);
- }
-
- UInt::UInt(RValue<UShort> cast)
- {
- Value *integer = Nucleus::createZExt(cast.value, UInt::getType());
-
- storeValue(integer);
- }
-
- UInt::UInt(RValue<Long> cast)
- {
- Value *integer = Nucleus::createTrunc(cast.value, UInt::getType());
-
- storeValue(integer);
- }
-
- UInt::UInt(RValue<Float> cast)
- {
- assert(false && "UNIMPLEMENTED");
- }
-
- UInt::UInt()
- {
- }
-
- UInt::UInt(int x)
- {
- storeValue(Nucleus::createConstantInt(x));
- }
-
- UInt::UInt(unsigned int x)
- {
- storeValue(Nucleus::createConstantInt(x));
- }
-
- UInt::UInt(RValue<UInt> rhs)
- {
- storeValue(rhs.value);
- }
-
- UInt::UInt(RValue<Int> rhs)
- {
- storeValue(rhs.value);
- }
-
- UInt::UInt(const UInt &rhs)
- {
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- UInt::UInt(const Reference<UInt> &rhs)
- {
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- UInt::UInt(const Int &rhs)
- {
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- UInt::UInt(const Reference<Int> &rhs)
- {
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- RValue<UInt> UInt::operator=(unsigned int rhs) const
- {
- return RValue<UInt>(storeValue(Nucleus::createConstantInt(rhs)));
- }
-
- RValue<UInt> UInt::operator=(RValue<UInt> rhs) const
- {
- storeValue(rhs.value);
-
- return rhs;
- }
-
- RValue<UInt> UInt::operator=(RValue<Int> rhs) const
- {
- storeValue(rhs.value);
-
- return RValue<UInt>(rhs);
- }
-
- RValue<UInt> UInt::operator=(const UInt &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<UInt>(value);
- }
-
- RValue<UInt> UInt::operator=(const Reference<UInt> &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<UInt>(value);
- }
-
- RValue<UInt> UInt::operator=(const Int &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<UInt>(value);
- }
-
- RValue<UInt> UInt::operator=(const Reference<Int> &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<UInt>(value);
- }
-
- RValue<UInt> operator+(RValue<UInt> lhs, RValue<UInt> rhs)
- {
- return RValue<UInt>(Nucleus::createAdd(lhs.value, rhs.value));
- }
-
- RValue<UInt> operator-(RValue<UInt> lhs, RValue<UInt> rhs)
- {
- return RValue<UInt>(Nucleus::createSub(lhs.value, rhs.value));
- }
-
- RValue<UInt> operator*(RValue<UInt> lhs, RValue<UInt> rhs)
- {
- return RValue<UInt>(Nucleus::createMul(lhs.value, rhs.value));
- }
-
- RValue<UInt> operator/(RValue<UInt> lhs, RValue<UInt> rhs)
- {
- return RValue<UInt>(Nucleus::createUDiv(lhs.value, rhs.value));
- }
-
- RValue<UInt> operator%(RValue<UInt> lhs, RValue<UInt> rhs)
- {
- return RValue<UInt>(Nucleus::createURem(lhs.value, rhs.value));
- }
-
- RValue<UInt> operator&(RValue<UInt> lhs, RValue<UInt> rhs)
- {
- return RValue<UInt>(Nucleus::createAnd(lhs.value, rhs.value));
- }
-
- RValue<UInt> operator|(RValue<UInt> lhs, RValue<UInt> rhs)
- {
- return RValue<UInt>(Nucleus::createOr(lhs.value, rhs.value));
- }
-
- RValue<UInt> operator^(RValue<UInt> lhs, RValue<UInt> rhs)
- {
- return RValue<UInt>(Nucleus::createXor(lhs.value, rhs.value));
- }
-
- RValue<UInt> operator<<(RValue<UInt> lhs, RValue<UInt> rhs)
- {
- return RValue<UInt>(Nucleus::createShl(lhs.value, rhs.value));
- }
-
- RValue<UInt> operator>>(RValue<UInt> lhs, RValue<UInt> rhs)
- {
- return RValue<UInt>(Nucleus::createLShr(lhs.value, rhs.value));
- }
-
- RValue<UInt> operator+=(const UInt &lhs, RValue<UInt> rhs)
- {
- return lhs = lhs + rhs;
- }
-
- RValue<UInt> operator-=(const UInt &lhs, RValue<UInt> rhs)
- {
- return lhs = lhs - rhs;
- }
-
- RValue<UInt> operator*=(const UInt &lhs, RValue<UInt> rhs)
- {
- return lhs = lhs * rhs;
- }
-
- RValue<UInt> operator/=(const UInt &lhs, RValue<UInt> rhs)
- {
- return lhs = lhs / rhs;
- }
-
- RValue<UInt> operator%=(const UInt &lhs, RValue<UInt> rhs)
- {
- return lhs = lhs % rhs;
- }
-
- RValue<UInt> operator&=(const UInt &lhs, RValue<UInt> rhs)
- {
- return lhs = lhs & rhs;
- }
-
- RValue<UInt> operator|=(const UInt &lhs, RValue<UInt> rhs)
- {
- return lhs = lhs | rhs;
- }
-
- RValue<UInt> operator^=(const UInt &lhs, RValue<UInt> rhs)
- {
- return lhs = lhs ^ rhs;
- }
-
- RValue<UInt> operator<<=(const UInt &lhs, RValue<UInt> rhs)
- {
- return lhs = lhs << rhs;
- }
-
- RValue<UInt> operator>>=(const UInt &lhs, RValue<UInt> rhs)
- {
- return lhs = lhs >> rhs;
- }
-
- RValue<UInt> operator+(RValue<UInt> val)
- {
- return val;
- }
-
- RValue<UInt> operator-(RValue<UInt> val)
- {
- return RValue<UInt>(Nucleus::createNeg(val.value));
- }
-
- RValue<UInt> operator~(RValue<UInt> val)
- {
- return RValue<UInt>(Nucleus::createNot(val.value));
- }
-
- RValue<UInt> operator++(const UInt &val, int) // Post-increment
- {
- assert(false && "UNIMPLEMENTED"); return RValue<UInt>(V(nullptr));
- }
-
- const UInt &operator++(const UInt &val) // Pre-increment
- {
- assert(false && "UNIMPLEMENTED"); return val;
- }
-
- RValue<UInt> operator--(const UInt &val, int) // Post-decrement
- {
- assert(false && "UNIMPLEMENTED"); return RValue<UInt>(V(nullptr));
- }
-
- const UInt &operator--(const UInt &val) // Pre-decrement
- {
- assert(false && "UNIMPLEMENTED"); return val;
- }
-
- RValue<UInt> Max(RValue<UInt> x, RValue<UInt> y)
- {
- return IfThenElse(x > y, x, y);
- }
-
- RValue<UInt> Min(RValue<UInt> x, RValue<UInt> y)
- {
- return IfThenElse(x < y, x, y);
- }
-
- RValue<UInt> Clamp(RValue<UInt> x, RValue<UInt> min, RValue<UInt> max)
- {
- return Min(Max(x, min), max);
- }
-
- RValue<Bool> operator<(RValue<UInt> lhs, RValue<UInt> rhs)
- {
- return RValue<Bool>(Nucleus::createICmpULT(lhs.value, rhs.value));
- }
-
- RValue<Bool> operator<=(RValue<UInt> lhs, RValue<UInt> rhs)
- {
- return RValue<Bool>(Nucleus::createICmpULE(lhs.value, rhs.value));
- }
-
- RValue<Bool> operator>(RValue<UInt> lhs, RValue<UInt> rhs)
- {
- return RValue<Bool>(Nucleus::createICmpUGT(lhs.value, rhs.value));
- }
-
- RValue<Bool> operator>=(RValue<UInt> lhs, RValue<UInt> rhs)
- {
- return RValue<Bool>(Nucleus::createICmpUGE(lhs.value, rhs.value));
- }
-
- RValue<Bool> operator!=(RValue<UInt> lhs, RValue<UInt> rhs)
- {
- return RValue<Bool>(Nucleus::createICmpNE(lhs.value, rhs.value));
- }
-
- RValue<Bool> operator==(RValue<UInt> lhs, RValue<UInt> rhs)
- {
- return RValue<Bool>(Nucleus::createICmpEQ(lhs.value, rhs.value));
- }
-
-// RValue<UInt> RoundUInt(RValue<Float> cast)
-// {
-// assert(false && "UNIMPLEMENTED"); return RValue<UInt>(V(nullptr));
-// }
-
- Type *UInt::getType()
- {
- return T(Ice::IceType_i32);
- }
-
-// Int2::Int2(RValue<Int> cast)
-// {
-// Value *extend = Nucleus::createZExt(cast.value, Long::getType());
-// Value *vector = Nucleus::createBitCast(extend, Int2::getType());
-//
-// Constant *shuffle[2];
-// shuffle[0] = Nucleus::createConstantInt(0);
-// shuffle[1] = Nucleus::createConstantInt(0);
-//
-// Value *replicate = Nucleus::createShuffleVector(vector, UndefValue::get(Int2::getType()), Nucleus::createConstantVector(shuffle, 2));
-//
-// storeValue(replicate);
-// }
-
- Int2::Int2(RValue<Int4> cast)
- {
- storeValue(Nucleus::createBitCast(cast.value, getType()));
- }
-
- Int2::Int2()
- {
- // xy.parent = this;
- }
-
- Int2::Int2(int x, int y)
- {
- // xy.parent = this;
-
- int64_t constantVector[2] = {x, y};
- storeValue(Nucleus::createConstantVector(constantVector, getType()));
- }
-
- Int2::Int2(RValue<Int2> rhs)
- {
- // xy.parent = this;
-
- storeValue(rhs.value);
- }
-
- Int2::Int2(const Int2 &rhs)
- {
- // xy.parent = this;
-
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- Int2::Int2(const Reference<Int2> &rhs)
- {
- // xy.parent = this;
-
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- Int2::Int2(RValue<Int> lo, RValue<Int> hi)
- {
- assert(false && "UNIMPLEMENTED");
- }
-
- RValue<Int2> Int2::operator=(RValue<Int2> rhs) const
- {
- storeValue(rhs.value);
-
- return rhs;
- }
-
- RValue<Int2> Int2::operator=(const Int2 &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<Int2>(value);
- }
-
- RValue<Int2> Int2::operator=(const Reference<Int2> &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<Int2>(value);
- }
-
- RValue<Int2> operator+(RValue<Int2> lhs, RValue<Int2> rhs)
- {
- return RValue<Int2>(Nucleus::createAdd(lhs.value, rhs.value));
- }
-
- RValue<Int2> operator-(RValue<Int2> lhs, RValue<Int2> rhs)
- {
- return RValue<Int2>(Nucleus::createSub(lhs.value, rhs.value));
- }
-
-// RValue<Int2> operator*(RValue<Int2> lhs, RValue<Int2> rhs)
-// {
-// return RValue<Int2>(Nucleus::createMul(lhs.value, rhs.value));
-// }
-
-// RValue<Int2> operator/(RValue<Int2> lhs, RValue<Int2> rhs)
-// {
-// return RValue<Int2>(Nucleus::createSDiv(lhs.value, rhs.value));
-// }
-
-// RValue<Int2> operator%(RValue<Int2> lhs, RValue<Int2> rhs)
-// {
-// return RValue<Int2>(Nucleus::createSRem(lhs.value, rhs.value));
-// }
-
- RValue<Int2> operator&(RValue<Int2> lhs, RValue<Int2> rhs)
- {
- return RValue<Int2>(Nucleus::createAnd(lhs.value, rhs.value));
- }
-
- RValue<Int2> operator|(RValue<Int2> lhs, RValue<Int2> rhs)
- {
- return RValue<Int2>(Nucleus::createOr(lhs.value, rhs.value));
- }
-
- RValue<Int2> operator^(RValue<Int2> lhs, RValue<Int2> rhs)
- {
- return RValue<Int2>(Nucleus::createXor(lhs.value, rhs.value));
- }
-
- RValue<Int2> operator<<(RValue<Int2> lhs, unsigned char rhs)
- {
- return RValue<Int2>(Nucleus::createShl(lhs.value, C(::context->getConstantInt32(rhs))));
- }
-
- RValue<Int2> operator>>(RValue<Int2> lhs, unsigned char rhs)
- {
- return RValue<Int2>(Nucleus::createAShr(lhs.value, C(::context->getConstantInt32(rhs))));
- }
-
- RValue<Int2> operator<<(RValue<Int2> lhs, RValue<Long1> rhs)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Int2>(V(nullptr));
- }
-
- RValue<Int2> operator>>(RValue<Int2> lhs, RValue<Long1> rhs)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Int2>(V(nullptr));
- }
-
- RValue<Int2> operator+=(const Int2 &lhs, RValue<Int2> rhs)
- {
- return lhs = lhs + rhs;
- }
-
- RValue<Int2> operator-=(const Int2 &lhs, RValue<Int2> rhs)
- {
- return lhs = lhs - rhs;
- }
-
-// RValue<Int2> operator*=(const Int2 &lhs, RValue<Int2> rhs)
-// {
-// return lhs = lhs * rhs;
-// }
-
-// RValue<Int2> operator/=(const Int2 &lhs, RValue<Int2> rhs)
-// {
-// return lhs = lhs / rhs;
-// }
-
-// RValue<Int2> operator%=(const Int2 &lhs, RValue<Int2> rhs)
-// {
-// return lhs = lhs % rhs;
-// }
-
- RValue<Int2> operator&=(const Int2 &lhs, RValue<Int2> rhs)
- {
- return lhs = lhs & rhs;
- }
-
- RValue<Int2> operator|=(const Int2 &lhs, RValue<Int2> rhs)
- {
- return lhs = lhs | rhs;
- }
-
- RValue<Int2> operator^=(const Int2 &lhs, RValue<Int2> rhs)
- {
- return lhs = lhs ^ rhs;
- }
-
- RValue<Int2> operator<<=(const Int2 &lhs, unsigned char rhs)
- {
- return lhs = lhs << rhs;
- }
-
- RValue<Int2> operator>>=(const Int2 &lhs, unsigned char rhs)
- {
- return lhs = lhs >> rhs;
- }
-
- RValue<Int2> operator<<=(const Int2 &lhs, RValue<Long1> rhs)
- {
- return lhs = lhs << rhs;
- }
-
- RValue<Int2> operator>>=(const Int2 &lhs, RValue<Long1> rhs)
- {
- return lhs = lhs >> rhs;
- }
-
-// RValue<Int2> operator+(RValue<Int2> val)
-// {
-// return val;
-// }
-
-// RValue<Int2> operator-(RValue<Int2> val)
-// {
-// return RValue<Int2>(Nucleus::createNeg(val.value));
-// }
-
- RValue<Int2> operator~(RValue<Int2> val)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Int2>(V(nullptr));
- }
-
- RValue<Long1> UnpackLow(RValue<Int2> x, RValue<Int2> y)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Long1>(V(nullptr));
- }
-
- RValue<Long1> UnpackHigh(RValue<Int2> x, RValue<Int2> y)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Long1>(V(nullptr));
- }
-
- RValue<Int> Extract(RValue<Int2> val, int i)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Int>(V(nullptr));
- }
-
- RValue<Int2> Insert(RValue<Int2> val, RValue<Int> element, int i)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Int2>(V(nullptr));
- }
-
- Type *Int2::getType()
- {
- return T(Type_v2i32);
- }
-
- UInt2::UInt2()
- {
- // xy.parent = this;
- }
-
- UInt2::UInt2(unsigned int x, unsigned int y)
- {
- // xy.parent = this;
-
- int64_t constantVector[2] = {x, y};
- storeValue(Nucleus::createConstantVector(constantVector, getType()));
- }
-
- UInt2::UInt2(RValue<UInt2> rhs)
- {
- // xy.parent = this;
-
- storeValue(rhs.value);
- }
-
- UInt2::UInt2(const UInt2 &rhs)
- {
- // xy.parent = this;
-
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- UInt2::UInt2(const Reference<UInt2> &rhs)
- {
- // xy.parent = this;
-
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- RValue<UInt2> UInt2::operator=(RValue<UInt2> rhs) const
- {
- storeValue(rhs.value);
-
- return rhs;
- }
-
- RValue<UInt2> UInt2::operator=(const UInt2 &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<UInt2>(value);
- }
-
- RValue<UInt2> UInt2::operator=(const Reference<UInt2> &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<UInt2>(value);
- }
-
- RValue<UInt2> operator+(RValue<UInt2> lhs, RValue<UInt2> rhs)
- {
- return RValue<UInt2>(Nucleus::createAdd(lhs.value, rhs.value));
- }
-
- RValue<UInt2> operator-(RValue<UInt2> lhs, RValue<UInt2> rhs)
- {
- return RValue<UInt2>(Nucleus::createSub(lhs.value, rhs.value));
- }
-
-// RValue<UInt2> operator*(RValue<UInt2> lhs, RValue<UInt2> rhs)
-// {
-// return RValue<UInt2>(Nucleus::createMul(lhs.value, rhs.value));
-// }
-
-// RValue<UInt2> operator/(RValue<UInt2> lhs, RValue<UInt2> rhs)
-// {
-// return RValue<UInt2>(Nucleus::createUDiv(lhs.value, rhs.value));
-// }
-
-// RValue<UInt2> operator%(RValue<UInt2> lhs, RValue<UInt2> rhs)
-// {
-// return RValue<UInt2>(Nucleus::createURem(lhs.value, rhs.value));
-// }
-
- RValue<UInt2> operator&(RValue<UInt2> lhs, RValue<UInt2> rhs)
- {
- return RValue<UInt2>(Nucleus::createAnd(lhs.value, rhs.value));
- }
-
- RValue<UInt2> operator|(RValue<UInt2> lhs, RValue<UInt2> rhs)
- {
- return RValue<UInt2>(Nucleus::createOr(lhs.value, rhs.value));
- }
-
- RValue<UInt2> operator^(RValue<UInt2> lhs, RValue<UInt2> rhs)
- {
- return RValue<UInt2>(Nucleus::createXor(lhs.value, rhs.value));
- }
-
- RValue<UInt2> operator<<(RValue<UInt2> lhs, unsigned char rhs)
- {
- return RValue<UInt2>(Nucleus::createShl(lhs.value, C(::context->getConstantInt32(rhs))));
- }
-
- RValue<UInt2> operator>>(RValue<UInt2> lhs, unsigned char rhs)
- {
- return RValue<UInt2>(Nucleus::createLShr(lhs.value, C(::context->getConstantInt32(rhs))));
- }
-
- RValue<UInt2> operator<<(RValue<UInt2> lhs, RValue<Long1> rhs)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<UInt2>(V(nullptr));
- }
-
- RValue<UInt2> operator>>(RValue<UInt2> lhs, RValue<Long1> rhs)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<UInt2>(V(nullptr));
- }
-
- RValue<UInt2> operator+=(const UInt2 &lhs, RValue<UInt2> rhs)
- {
- return lhs = lhs + rhs;
- }
-
- RValue<UInt2> operator-=(const UInt2 &lhs, RValue<UInt2> rhs)
- {
- return lhs = lhs - rhs;
- }
-
-// RValue<UInt2> operator*=(const UInt2 &lhs, RValue<UInt2> rhs)
-// {
-// return lhs = lhs * rhs;
-// }
-
-// RValue<UInt2> operator/=(const UInt2 &lhs, RValue<UInt2> rhs)
-// {
-// return lhs = lhs / rhs;
-// }
-
-// RValue<UInt2> operator%=(const UInt2 &lhs, RValue<UInt2> rhs)
-// {
-// return lhs = lhs % rhs;
-// }
-
- RValue<UInt2> operator&=(const UInt2 &lhs, RValue<UInt2> rhs)
- {
- return lhs = lhs & rhs;
- }
-
- RValue<UInt2> operator|=(const UInt2 &lhs, RValue<UInt2> rhs)
- {
- return lhs = lhs | rhs;
- }
-
- RValue<UInt2> operator^=(const UInt2 &lhs, RValue<UInt2> rhs)
- {
- return lhs = lhs ^ rhs;
- }
-
- RValue<UInt2> operator<<=(const UInt2 &lhs, unsigned char rhs)
- {
- return lhs = lhs << rhs;
- }
-
- RValue<UInt2> operator>>=(const UInt2 &lhs, unsigned char rhs)
- {
- return lhs = lhs >> rhs;
- }
-
- RValue<UInt2> operator<<=(const UInt2 &lhs, RValue<Long1> rhs)
- {
- return lhs = lhs << rhs;
- }
-
- RValue<UInt2> operator>>=(const UInt2 &lhs, RValue<Long1> rhs)
- {
- return lhs = lhs >> rhs;
- }
-
-// RValue<UInt2> operator+(RValue<UInt2> val)
-// {
-// return val;
-// }
-
-// RValue<UInt2> operator-(RValue<UInt2> val)
-// {
-// return RValue<UInt2>(Nucleus::createNeg(val.value));
-// }
-
- RValue<UInt2> operator~(RValue<UInt2> val)
- {
- return RValue<UInt2>(Nucleus::createNot(val.value));
- }
-
- Type *UInt2::getType()
- {
- return T(Type_v2i32);
- }
-
- Int4::Int4(RValue<Byte4> cast)
- {
- assert(false && "UNIMPLEMENTED");
- }
-
- Int4::Int4(RValue<SByte4> cast)
- {
- assert(false && "UNIMPLEMENTED");
- }
-
- Int4::Int4(RValue<Float4> cast)
- {
- // xyzw.parent = this;
-
- Value *xyzw = Nucleus::createFPToSI(cast.value, Int4::getType());
-
- storeValue(xyzw);
- }
-
- Int4::Int4(RValue<Short4> cast)
- {
- assert(false && "UNIMPLEMENTED");
- }
-
- Int4::Int4(RValue<UShort4> cast)
- {
- assert(false && "UNIMPLEMENTED");
- }
-
- Int4::Int4()
- {
- // xyzw.parent = this;
- }
-
- Int4::Int4(int xyzw)
- {
- constant(xyzw, xyzw, xyzw, xyzw);
- }
-
- Int4::Int4(int x, int yzw)
- {
- constant(x, yzw, yzw, yzw);
- }
-
- Int4::Int4(int x, int y, int zw)
- {
- constant(x, y, zw, zw);
- }
-
- Int4::Int4(int x, int y, int z, int w)
- {
- constant(x, y, z, w);
- }
-
- void Int4::constant(int x, int y, int z, int w)
- {
- // xyzw.parent = this;
-
- int64_t constantVector[4] = {x, y, z, w};
- storeValue(Nucleus::createConstantVector(constantVector, getType()));
- }
-
- Int4::Int4(RValue<Int4> rhs)
- {
- // xyzw.parent = this;
-
- storeValue(rhs.value);
- }
-
- Int4::Int4(const Int4 &rhs)
- {
- // xyzw.parent = this;
-
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- Int4::Int4(const Reference<Int4> &rhs)
- {
- // xyzw.parent = this;
-
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- Int4::Int4(RValue<UInt4> rhs)
- {
- // xyzw.parent = this;
-
- storeValue(rhs.value);
- }
-
- Int4::Int4(const UInt4 &rhs)
- {
- // xyzw.parent = this;
-
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- Int4::Int4(const Reference<UInt4> &rhs)
- {
- // xyzw.parent = this;
-
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- Int4::Int4(RValue<Int2> lo, RValue<Int2> hi)
- {
- assert(false && "UNIMPLEMENTED");
- }
-
- Int4::Int4(RValue<Int> rhs)
- {
- // xyzw.parent = this;
-
- assert(false && "UNIMPLEMENTED");
- }
-
- Int4::Int4(const Int &rhs)
- {
- // xyzw.parent = this;
-
- *this = RValue<Int>(rhs.loadValue());
- }
-
- Int4::Int4(const Reference<Int> &rhs)
- {
- // xyzw.parent = this;
-
- *this = RValue<Int>(rhs.loadValue());
- }
-
- RValue<Int4> Int4::operator=(RValue<Int4> rhs) const
- {
- storeValue(rhs.value);
-
- return rhs;
- }
-
- RValue<Int4> Int4::operator=(const Int4 &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<Int4>(value);
- }
-
- RValue<Int4> Int4::operator=(const Reference<Int4> &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<Int4>(value);
- }
-
- RValue<Int4> operator+(RValue<Int4> lhs, RValue<Int4> rhs)
- {
- return RValue<Int4>(Nucleus::createAdd(lhs.value, rhs.value));
- }
-
- RValue<Int4> operator-(RValue<Int4> lhs, RValue<Int4> rhs)
- {
- return RValue<Int4>(Nucleus::createSub(lhs.value, rhs.value));
- }
-
- RValue<Int4> operator*(RValue<Int4> lhs, RValue<Int4> rhs)
- {
- return RValue<Int4>(Nucleus::createMul(lhs.value, rhs.value));
- }
-
- RValue<Int4> operator/(RValue<Int4> lhs, RValue<Int4> rhs)
- {
- return RValue<Int4>(Nucleus::createSDiv(lhs.value, rhs.value));
- }
-
- RValue<Int4> operator%(RValue<Int4> lhs, RValue<Int4> rhs)
- {
- return RValue<Int4>(Nucleus::createSRem(lhs.value, rhs.value));
- }
-
- RValue<Int4> operator&(RValue<Int4> lhs, RValue<Int4> rhs)
- {
- return RValue<Int4>(Nucleus::createAnd(lhs.value, rhs.value));
- }
-
- RValue<Int4> operator|(RValue<Int4> lhs, RValue<Int4> rhs)
- {
- return RValue<Int4>(Nucleus::createOr(lhs.value, rhs.value));
- }
-
- RValue<Int4> operator^(RValue<Int4> lhs, RValue<Int4> rhs)
- {
- return RValue<Int4>(Nucleus::createXor(lhs.value, rhs.value));
- }
-
- RValue<Int4> operator<<(RValue<Int4> lhs, unsigned char rhs)
- {
- return RValue<Int4>(Nucleus::createShl(lhs.value, C(::context->getConstantInt32(rhs))));
- }
-
- RValue<Int4> operator>>(RValue<Int4> lhs, unsigned char rhs)
- {
- return RValue<Int4>(Nucleus::createAShr(lhs.value, C(::context->getConstantInt32(rhs))));
- }
-
- RValue<Int4> operator<<(RValue<Int4> lhs, RValue<Int4> rhs)
- {
- return RValue<Int4>(Nucleus::createShl(lhs.value, rhs.value));
- }
-
- RValue<Int4> operator>>(RValue<Int4> lhs, RValue<Int4> rhs)
- {
- return RValue<Int4>(Nucleus::createAShr(lhs.value, rhs.value));
- }
-
- RValue<Int4> operator+=(const Int4 &lhs, RValue<Int4> rhs)
- {
- return lhs = lhs + rhs;
- }
-
- RValue<Int4> operator-=(const Int4 &lhs, RValue<Int4> rhs)
- {
- return lhs = lhs - rhs;
- }
-
- RValue<Int4> operator*=(const Int4 &lhs, RValue<Int4> rhs)
- {
- return lhs = lhs * rhs;
- }
-
-// RValue<Int4> operator/=(const Int4 &lhs, RValue<Int4> rhs)
-// {
-// return lhs = lhs / rhs;
-// }
-
-// RValue<Int4> operator%=(const Int4 &lhs, RValue<Int4> rhs)
-// {
-// return lhs = lhs % rhs;
-// }
-
- RValue<Int4> operator&=(const Int4 &lhs, RValue<Int4> rhs)
- {
- return lhs = lhs & rhs;
- }
-
- RValue<Int4> operator|=(const Int4 &lhs, RValue<Int4> rhs)
- {
- return lhs = lhs | rhs;
- }
-
- RValue<Int4> operator^=(const Int4 &lhs, RValue<Int4> rhs)
- {
- return lhs = lhs ^ rhs;
- }
-
- RValue<Int4> operator<<=(const Int4 &lhs, unsigned char rhs)
- {
- return lhs = lhs << rhs;
- }
-
- RValue<Int4> operator>>=(const Int4 &lhs, unsigned char rhs)
- {
- return lhs = lhs >> rhs;
- }
-
- RValue<Int4> operator+(RValue<Int4> val)
- {
- return val;
- }
-
- RValue<Int4> operator-(RValue<Int4> val)
- {
- return RValue<Int4>(Nucleus::createNeg(val.value));
- }
-
- RValue<Int4> operator~(RValue<Int4> val)
- {
- return RValue<Int4>(Nucleus::createNot(val.value));
- }
-
- RValue<Int4> CmpEQ(RValue<Int4> x, RValue<Int4> y)
- {
- return RValue<Int4>(Nucleus::createSExt(Nucleus::createICmpEQ(x.value, y.value), Int4::getType()));
- }
-
- RValue<Int4> CmpLT(RValue<Int4> x, RValue<Int4> y)
- {
- return RValue<Int4>(Nucleus::createSExt(Nucleus::createICmpSLT(x.value, y.value), Int4::getType()));
- }
-
- RValue<Int4> CmpLE(RValue<Int4> x, RValue<Int4> y)
- {
- return RValue<Int4>(Nucleus::createSExt(Nucleus::createICmpSLE(x.value, y.value), Int4::getType()));
- }
-
- RValue<Int4> CmpNEQ(RValue<Int4> x, RValue<Int4> y)
- {
- return RValue<Int4>(Nucleus::createSExt(Nucleus::createICmpNE(x.value, y.value), Int4::getType()));
- }
-
- RValue<Int4> CmpNLT(RValue<Int4> x, RValue<Int4> y)
- {
- return RValue<Int4>(Nucleus::createSExt(Nucleus::createICmpSGE(x.value, y.value), Int4::getType()));
- }
-
- RValue<Int4> CmpNLE(RValue<Int4> x, RValue<Int4> y)
- {
- return RValue<Int4>(Nucleus::createSExt(Nucleus::createICmpSGT(x.value, y.value), Int4::getType()));
- }
-
- RValue<Int4> Max(RValue<Int4> x, RValue<Int4> y)
- {
- Ice::Variable *condition = ::function->makeVariable(Ice::IceType_v4i1);
- auto cmp = Ice::InstIcmp::create(::function, Ice::InstIcmp::Sle, condition, x.value, y.value);
- ::basicBlock->appendInst(cmp);
-
- Ice::Variable *result = ::function->makeVariable(Ice::IceType_v4i32);
- auto select = Ice::InstSelect::create(::function, result, condition, y.value, x.value);
- ::basicBlock->appendInst(select);
-
- return RValue<Int4>(V(result));
- }
-
- RValue<Int4> Min(RValue<Int4> x, RValue<Int4> y)
- {
- Ice::Variable *condition = ::function->makeVariable(Ice::IceType_v4i1);
- auto cmp = Ice::InstIcmp::create(::function, Ice::InstIcmp::Sgt, condition, x.value, y.value);
- ::basicBlock->appendInst(cmp);
-
- Ice::Variable *result = ::function->makeVariable(Ice::IceType_v4i32);
- auto select = Ice::InstSelect::create(::function, result, condition, y.value, x.value);
- ::basicBlock->appendInst(select);
-
- return RValue<Int4>(V(result));
- }
-
- RValue<Int4> RoundInt(RValue<Float4> cast)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Int4>(V(nullptr));
- }
-
- RValue<Short8> Pack(RValue<Int4> x, RValue<Int4> y)
- {
- Ice::Variable *result = ::function->makeVariable(Ice::IceType_v8i16);
- const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::VectorPackSigned, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
- auto target = ::context->getConstantUndef(Ice::IceType_i32);
- auto pack = Ice::InstIntrinsicCall::create(::function, 2, result, target, intrinsic);
- pack->addArg(x.value);
- pack->addArg(y.value);
- ::basicBlock->appendInst(pack);
-
- return RValue<Short8>(V(result));
- }
-
- RValue<Int> Extract(RValue<Int4> x, int i)
- {
- return RValue<Int>(Nucleus::createExtractElement(x.value, Int::getType(), i));
- }
-
- RValue<Int4> Insert(RValue<Int4> x, RValue<Int> element, int i)
- {
- return RValue<Int4>(Nucleus::createInsertElement(x.value, element.value, i));
- }
-
- RValue<Int> SignMask(RValue<Int4> x)
- {
- Ice::Variable *result = ::function->makeVariable(Ice::IceType_i32);
- const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::SignMask, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
- auto target = ::context->getConstantUndef(Ice::IceType_i32);
- auto movmsk = Ice::InstIntrinsicCall::create(::function, 1, result, target, intrinsic);
- movmsk->addArg(x.value);
- ::basicBlock->appendInst(movmsk);
-
- return RValue<Int>(V(result));
- }
-
- RValue<Int4> Swizzle(RValue<Int4> x, unsigned char select)
- {
- return RValue<Int4>(createSwizzle4(x.value, select));
- }
-
- Type *Int4::getType()
- {
- return T(Ice::IceType_v4i32);
- }
-
- UInt4::UInt4(RValue<Float4> cast)
- {
- // xyzw.parent = this;
-
- assert(false && "UNIMPLEMENTED");
- }
-
- UInt4::UInt4()
- {
- // xyzw.parent = this;
- }
-
- UInt4::UInt4(int xyzw)
- {
- constant(xyzw, xyzw, xyzw, xyzw);
- }
-
- UInt4::UInt4(int x, int yzw)
- {
- constant(x, yzw, yzw, yzw);
- }
-
- UInt4::UInt4(int x, int y, int zw)
- {
- constant(x, y, zw, zw);
- }
-
- UInt4::UInt4(int x, int y, int z, int w)
- {
- constant(x, y, z, w);
- }
-
- void UInt4::constant(int x, int y, int z, int w)
- {
- // xyzw.parent = this;
-
- int64_t constantVector[4] = {x, y, z, w};
- storeValue(Nucleus::createConstantVector(constantVector, getType()));
- }
-
- UInt4::UInt4(RValue<UInt4> rhs)
- {
- // xyzw.parent = this;
-
- storeValue(rhs.value);
- }
-
- UInt4::UInt4(const UInt4 &rhs)
- {
- // xyzw.parent = this;
-
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- UInt4::UInt4(const Reference<UInt4> &rhs)
- {
- // xyzw.parent = this;
-
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- UInt4::UInt4(RValue<Int4> rhs)
- {
- // xyzw.parent = this;
-
- storeValue(rhs.value);
- }
-
- UInt4::UInt4(const Int4 &rhs)
- {
- // xyzw.parent = this;
-
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- UInt4::UInt4(const Reference<Int4> &rhs)
- {
- // xyzw.parent = this;
-
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- UInt4::UInt4(RValue<UInt2> lo, RValue<UInt2> hi)
- {
- assert(false && "UNIMPLEMENTED");
- }
-
- RValue<UInt4> UInt4::operator=(RValue<UInt4> rhs) const
- {
- storeValue(rhs.value);
-
- return rhs;
- }
-
- RValue<UInt4> UInt4::operator=(const UInt4 &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<UInt4>(value);
- }
-
- RValue<UInt4> UInt4::operator=(const Reference<UInt4> &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<UInt4>(value);
- }
-
- RValue<UInt4> operator+(RValue<UInt4> lhs, RValue<UInt4> rhs)
- {
- return RValue<UInt4>(Nucleus::createAdd(lhs.value, rhs.value));
- }
-
- RValue<UInt4> operator-(RValue<UInt4> lhs, RValue<UInt4> rhs)
- {
- return RValue<UInt4>(Nucleus::createSub(lhs.value, rhs.value));
- }
-
- RValue<UInt4> operator*(RValue<UInt4> lhs, RValue<UInt4> rhs)
- {
- return RValue<UInt4>(Nucleus::createMul(lhs.value, rhs.value));
- }
-
- RValue<UInt4> operator/(RValue<UInt4> lhs, RValue<UInt4> rhs)
- {
- return RValue<UInt4>(Nucleus::createUDiv(lhs.value, rhs.value));
- }
-
- RValue<UInt4> operator%(RValue<UInt4> lhs, RValue<UInt4> rhs)
- {
- return RValue<UInt4>(Nucleus::createURem(lhs.value, rhs.value));
- }
-
- RValue<UInt4> operator&(RValue<UInt4> lhs, RValue<UInt4> rhs)
- {
- return RValue<UInt4>(Nucleus::createAnd(lhs.value, rhs.value));
- }
-
- RValue<UInt4> operator|(RValue<UInt4> lhs, RValue<UInt4> rhs)
- {
- return RValue<UInt4>(Nucleus::createOr(lhs.value, rhs.value));
- }
-
- RValue<UInt4> operator^(RValue<UInt4> lhs, RValue<UInt4> rhs)
- {
- return RValue<UInt4>(Nucleus::createXor(lhs.value, rhs.value));
- }
-
- RValue<UInt4> operator<<(RValue<UInt4> lhs, unsigned char rhs)
- {
- return RValue<UInt4>(Nucleus::createShl(lhs.value, C(::context->getConstantInt32(rhs))));
- }
-
- RValue<UInt4> operator>>(RValue<UInt4> lhs, unsigned char rhs)
- {
- return RValue<UInt4>(Nucleus::createLShr(lhs.value, C(::context->getConstantInt32(rhs))));
- }
-
- RValue<UInt4> operator<<(RValue<UInt4> lhs, RValue<UInt4> rhs)
- {
- return RValue<UInt4>(Nucleus::createShl(lhs.value, rhs.value));
- }
-
- RValue<UInt4> operator>>(RValue<UInt4> lhs, RValue<UInt4> rhs)
- {
- return RValue<UInt4>(Nucleus::createLShr(lhs.value, rhs.value));
- }
-
- RValue<UInt4> operator+=(const UInt4 &lhs, RValue<UInt4> rhs)
- {
- return lhs = lhs + rhs;
- }
-
- RValue<UInt4> operator-=(const UInt4 &lhs, RValue<UInt4> rhs)
- {
- return lhs = lhs - rhs;
- }
-
- RValue<UInt4> operator*=(const UInt4 &lhs, RValue<UInt4> rhs)
- {
- return lhs = lhs * rhs;
- }
-
-// RValue<UInt4> operator/=(const UInt4 &lhs, RValue<UInt4> rhs)
-// {
-// return lhs = lhs / rhs;
-// }
-
-// RValue<UInt4> operator%=(const UInt4 &lhs, RValue<UInt4> rhs)
-// {
-// return lhs = lhs % rhs;
-// }
-
- RValue<UInt4> operator&=(const UInt4 &lhs, RValue<UInt4> rhs)
- {
- return lhs = lhs & rhs;
- }
-
- RValue<UInt4> operator|=(const UInt4 &lhs, RValue<UInt4> rhs)
- {
- return lhs = lhs | rhs;
- }
-
- RValue<UInt4> operator^=(const UInt4 &lhs, RValue<UInt4> rhs)
- {
- return lhs = lhs ^ rhs;
- }
-
- RValue<UInt4> operator<<=(const UInt4 &lhs, unsigned char rhs)
- {
- return lhs = lhs << rhs;
- }
-
- RValue<UInt4> operator>>=(const UInt4 &lhs, unsigned char rhs)
- {
- return lhs = lhs >> rhs;
- }
-
- RValue<UInt4> operator+(RValue<UInt4> val)
- {
- return val;
- }
-
- RValue<UInt4> operator-(RValue<UInt4> val)
- {
- return RValue<UInt4>(Nucleus::createNeg(val.value));
- }
-
- RValue<UInt4> operator~(RValue<UInt4> val)
- {
- return RValue<UInt4>(Nucleus::createNot(val.value));
- }
-
- RValue<UInt4> CmpEQ(RValue<UInt4> x, RValue<UInt4> y)
- {
- return RValue<UInt4>(Nucleus::createSExt(Nucleus::createICmpEQ(x.value, y.value), Int4::getType()));
- }
-
- RValue<UInt4> CmpLT(RValue<UInt4> x, RValue<UInt4> y)
- {
- return RValue<UInt4>(Nucleus::createSExt(Nucleus::createICmpULT(x.value, y.value), Int4::getType()));
- }
-
- RValue<UInt4> CmpLE(RValue<UInt4> x, RValue<UInt4> y)
- {
- return RValue<UInt4>(Nucleus::createSExt(Nucleus::createICmpULE(x.value, y.value), Int4::getType()));
- }
-
- RValue<UInt4> CmpNEQ(RValue<UInt4> x, RValue<UInt4> y)
- {
- return RValue<UInt4>(Nucleus::createSExt(Nucleus::createICmpNE(x.value, y.value), Int4::getType()));
- }
-
- RValue<UInt4> CmpNLT(RValue<UInt4> x, RValue<UInt4> y)
- {
- return RValue<UInt4>(Nucleus::createSExt(Nucleus::createICmpUGE(x.value, y.value), Int4::getType()));
- }
-
- RValue<UInt4> CmpNLE(RValue<UInt4> x, RValue<UInt4> y)
- {
- return RValue<UInt4>(Nucleus::createSExt(Nucleus::createICmpUGT(x.value, y.value), Int4::getType()));
- }
-
- RValue<UInt4> Max(RValue<UInt4> x, RValue<UInt4> y)
- {
- Ice::Variable *condition = ::function->makeVariable(Ice::IceType_v4i1);
- auto cmp = Ice::InstIcmp::create(::function, Ice::InstIcmp::Ule, condition, x.value, y.value);
- ::basicBlock->appendInst(cmp);
-
- Ice::Variable *result = ::function->makeVariable(Ice::IceType_v4i32);
- auto select = Ice::InstSelect::create(::function, result, condition, y.value, x.value);
- ::basicBlock->appendInst(select);
-
- return RValue<UInt4>(V(result));
- }
-
- RValue<UInt4> Min(RValue<UInt4> x, RValue<UInt4> y)
- {
- Ice::Variable *condition = ::function->makeVariable(Ice::IceType_v4i1);
- auto cmp = Ice::InstIcmp::create(::function, Ice::InstIcmp::Ugt, condition, x.value, y.value);
- ::basicBlock->appendInst(cmp);
-
- Ice::Variable *result = ::function->makeVariable(Ice::IceType_v4i32);
- auto select = Ice::InstSelect::create(::function, result, condition, y.value, x.value);
- ::basicBlock->appendInst(select);
-
- return RValue<UInt4>(V(result));
- }
-
- RValue<UShort8> Pack(RValue<UInt4> x, RValue<UInt4> y)
- {
- Ice::Variable *result = ::function->makeVariable(Ice::IceType_v8i16);
- const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::VectorPackUnsigned, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
- auto target = ::context->getConstantUndef(Ice::IceType_i32);
- auto pack = Ice::InstIntrinsicCall::create(::function, 2, result, target, intrinsic);
- pack->addArg(x.value);
- pack->addArg(y.value);
- ::basicBlock->appendInst(pack);
-
- return RValue<UShort8>(V(result));
- }
-
- Type *UInt4::getType()
- {
- return T(Ice::IceType_v4i32);
- }
-
- Float::Float(RValue<Int> cast)
- {
- Value *integer = Nucleus::createSIToFP(cast.value, Float::getType());
-
- storeValue(integer);
- }
-
- Float::Float()
- {
- }
-
- Float::Float(float x)
- {
- storeValue(Nucleus::createConstantFloat(x));
- }
-
- Float::Float(RValue<Float> rhs)
- {
- storeValue(rhs.value);
- }
-
- Float::Float(const Float &rhs)
- {
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- Float::Float(const Reference<Float> &rhs)
- {
- Value *value = rhs.loadValue();
- storeValue(value);
- }
-
- RValue<Float> Float::operator=(RValue<Float> rhs) const
- {
- storeValue(rhs.value);
-
- return rhs;
- }
-
- RValue<Float> Float::operator=(const Float &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<Float>(value);
- }
-
- RValue<Float> Float::operator=(const Reference<Float> &rhs) const
- {
- Value *value = rhs.loadValue();
- storeValue(value);
-
- return RValue<Float>(value);
- }
-
- RValue<Float> operator+(RValue<Float> lhs, RValue<Float> rhs)
- {
- return RValue<Float>(Nucleus::createFAdd(lhs.value, rhs.value));
- }
-
- RValue<Float> operator-(RValue<Float> lhs, RValue<Float> rhs)
- {
- return RValue<Float>(Nucleus::createFSub(lhs.value, rhs.value));
- }
-
- RValue<Float> operator*(RValue<Float> lhs, RValue<Float> rhs)
- {
- return RValue<Float>(Nucleus::createFMul(lhs.value, rhs.value));
- }
-
- RValue<Float> operator/(RValue<Float> lhs, RValue<Float> rhs)
- {
- return RValue<Float>(Nucleus::createFDiv(lhs.value, rhs.value));
- }
-
- RValue<Float> operator+=(const Float &lhs, RValue<Float> rhs)
- {
- return lhs = lhs + rhs;
- }
-
- RValue<Float> operator-=(const Float &lhs, RValue<Float> rhs)
- {
- return lhs = lhs - rhs;
- }
-
- RValue<Float> operator*=(const Float &lhs, RValue<Float> rhs)
- {
- return lhs = lhs * rhs;
- }
-
- RValue<Float> operator/=(const Float &lhs, RValue<Float> rhs)
- {
- return lhs = lhs / rhs;
- }
-
- RValue<Float> operator+(RValue<Float> val)
- {
- return val;
- }
-
- RValue<Float> operator-(RValue<Float> val)
- {
- return RValue<Float>(Nucleus::createFNeg(val.value));
- }
-
- RValue<Bool> operator<(RValue<Float> lhs, RValue<Float> rhs)
- {
- return RValue<Bool>(Nucleus::createFCmpOLT(lhs.value, rhs.value));
- }
-
- RValue<Bool> operator<=(RValue<Float> lhs, RValue<Float> rhs)
- {
- return RValue<Bool>(Nucleus::createFCmpOLE(lhs.value, rhs.value));
- }
-
- RValue<Bool> operator>(RValue<Float> lhs, RValue<Float> rhs)
- {
- return RValue<Bool>(Nucleus::createFCmpOGT(lhs.value, rhs.value));
- }
-
- RValue<Bool> operator>=(RValue<Float> lhs, RValue<Float> rhs)
- {
- return RValue<Bool>(Nucleus::createFCmpOGE(lhs.value, rhs.value));
- }
-
- RValue<Bool> operator!=(RValue<Float> lhs, RValue<Float> rhs)
- {
- return RValue<Bool>(Nucleus::createFCmpONE(lhs.value, rhs.value));
- }
-
- RValue<Bool> operator==(RValue<Float> lhs, RValue<Float> rhs)
- {
- return RValue<Bool>(Nucleus::createFCmpOEQ(lhs.value, rhs.value));
- }
-
- RValue<Float> Abs(RValue<Float> x)
- {
- return IfThenElse(x > 0.0f, x, -x);
- }
-
- RValue<Float> Max(RValue<Float> x, RValue<Float> y)
- {
- return IfThenElse(x > y, x, y);
- }
-
- RValue<Float> Min(RValue<Float> x, RValue<Float> y)
- {
- return IfThenElse(x < y, x, y);
- }
-
- RValue<Float> Rcp_pp(RValue<Float> x, bool exactAtPow2)
- {
- return 1.0f / x;
- }
-
- RValue<Float> RcpSqrt_pp(RValue<Float> x)
- {
- return Rcp_pp(Sqrt(x));
- }
-
- RValue<Float> Sqrt(RValue<Float> x)
- {
- Ice::Variable *result = ::function->makeVariable(Ice::IceType_f32);
- const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::Sqrt, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
- auto target = ::context->getConstantUndef(Ice::IceType_i32);
- auto sqrt = Ice::InstIntrinsicCall::create(::function, 1, result, target, intrinsic);
- sqrt->addArg(x.value);
- ::basicBlock->appendInst(sqrt);
-
- return RValue<Float>(V(result));
- }
-
- RValue<Float> Round(RValue<Float> x)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Float>(V(nullptr));
- }
-
- RValue<Float> Trunc(RValue<Float> x)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Float>(V(nullptr));
- }
-
- RValue<Float> Frac(RValue<Float> x)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Float>(V(nullptr));
- }
-
- RValue<Float> Floor(RValue<Float> x)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Float>(V(nullptr));
- }
-
- RValue<Float> Ceil(RValue<Float> x)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Float>(V(nullptr));
- }
-
- Type *Float::getType()
- {
- return T(Ice::IceType_f32);
- }
-
- Float2::Float2(RValue<Float4> cast)
- {
- storeValue(Nucleus::createBitCast(cast.value, getType()));
- }
-
- Type *Float2::getType()
- {
- return T(Type_v2f32);
- }
-
- Float4::Float4(RValue<Byte4> cast)
- {
- xyzw.parent = this;
-
- assert(false && "UNIMPLEMENTED");
- }
-
- Float4::Float4(RValue<SByte4> cast)
- {
- xyzw.parent = this;
-
- assert(false && "UNIMPLEMENTED");
- }
-
- Float4::Float4(RValue<Short4> cast)
+ RValue<UInt2> operator>>(RValue<UInt2> lhs, unsigned char rhs)
{
- xyzw.parent = this;
+ if(emulateIntrinsics)
+ {
+ UInt2 result;
+ result = Insert(result, Extract(lhs, 0) >> UInt(rhs), 0);
+ result = Insert(result, Extract(lhs, 1) >> UInt(rhs), 1);
- Int4 c(cast);
- storeValue(Nucleus::createSIToFP(RValue<Int4>(c).value, Float4::getType()));
+ return result;
+ }
+ else
+ {
+ return RValue<UInt2>(Nucleus::createLShr(lhs.value, V(::context->getConstantInt32(rhs))));
+ }
}
- Float4::Float4(RValue<UShort4> cast)
+ Type *UInt2::getType()
{
- xyzw.parent = this;
-
- Int4 c(cast);
- storeValue(Nucleus::createSIToFP(RValue<Int4>(c).value, Float4::getType()));
+ return T(Type_v2i32);
}
- Float4::Float4(RValue<Int4> cast)
+ Int4::Int4(RValue<Byte4> cast) : XYZW(this)
{
- xyzw.parent = this;
+ Value *x = Nucleus::createBitCast(cast.value, Int::getType());
+ Value *a = Nucleus::createInsertElement(loadValue(), x, 0);
- Value *xyzw = Nucleus::createSIToFP(cast.value, Float4::getType());
+ Value *e;
+ int swizzle[16] = {0, 16, 1, 17, 2, 18, 3, 19, 4, 20, 5, 21, 6, 22, 7, 23};
+ Value *b = Nucleus::createBitCast(a, Byte16::getType());
+ Value *c = Nucleus::createShuffleVector(b, V(Nucleus::createNullValue(Byte16::getType())), swizzle);
- storeValue(xyzw);
+ int swizzle2[8] = {0, 8, 1, 9, 2, 10, 3, 11};
+ Value *d = Nucleus::createBitCast(c, Short8::getType());
+ e = Nucleus::createShuffleVector(d, V(Nucleus::createNullValue(Short8::getType())), swizzle2);
+
+ Value *f = Nucleus::createBitCast(e, Int4::getType());
+ storeValue(f);
}
- Float4::Float4(RValue<UInt4> cast)
+ Int4::Int4(RValue<SByte4> cast) : XYZW(this)
{
- xyzw.parent = this;
+ Value *x = Nucleus::createBitCast(cast.value, Int::getType());
+ Value *a = Nucleus::createInsertElement(loadValue(), x, 0);
- Value *xyzw = Nucleus::createUIToFP(cast.value, Float4::getType());
+ int swizzle[16] = {0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7};
+ Value *b = Nucleus::createBitCast(a, Byte16::getType());
+ Value *c = Nucleus::createShuffleVector(b, b, swizzle);
- storeValue(xyzw);
- }
+ int swizzle2[8] = {0, 0, 1, 1, 2, 2, 3, 3};
+ Value *d = Nucleus::createBitCast(c, Short8::getType());
+ Value *e = Nucleus::createShuffleVector(d, d, swizzle2);
- Float4::Float4()
- {
- xyzw.parent = this;
+ *this = As<Int4>(e) >> 24;
}
- Float4::Float4(float xyzw)
+ Int4::Int4(RValue<Short4> cast) : XYZW(this)
{
- constant(xyzw, xyzw, xyzw, xyzw);
- }
+ int swizzle[8] = {0, 0, 1, 1, 2, 2, 3, 3};
+ Value *c = Nucleus::createShuffleVector(cast.value, cast.value, swizzle);
- Float4::Float4(float x, float yzw)
- {
- constant(x, yzw, yzw, yzw);
+ *this = As<Int4>(c) >> 16;
}
- Float4::Float4(float x, float y, float zw)
+ Int4::Int4(RValue<UShort4> cast) : XYZW(this)
{
- constant(x, y, zw, zw);
+ int swizzle[8] = {0, 8, 1, 9, 2, 10, 3, 11};
+ Value *c = Nucleus::createShuffleVector(cast.value, Short8(0, 0, 0, 0, 0, 0, 0, 0).loadValue(), swizzle);
+ Value *d = Nucleus::createBitCast(c, Int4::getType());
+ storeValue(d);
}
- Float4::Float4(float x, float y, float z, float w)
+ Int4::Int4(RValue<Int> rhs) : XYZW(this)
{
- constant(x, y, z, w);
- }
+ Value *vector = Nucleus::createBitCast(rhs.value, Int4::getType());
- void Float4::constant(float x, float y, float z, float w)
- {
- xyzw.parent = this;
+ int swizzle[4] = {0, 0, 0, 0};
+ Value *replicate = Nucleus::createShuffleVector(vector, vector, swizzle);
- double constantVector[4] = {x, y, z, w};
- storeValue(Nucleus::createConstantVector(constantVector, getType()));
+ storeValue(replicate);
}
- Float4::Float4(RValue<Float4> rhs)
+ RValue<Int4> operator<<(RValue<Int4> lhs, unsigned char rhs)
{
- xyzw.parent = this;
+ if(emulateIntrinsics)
+ {
+ Int4 result;
+ result = Insert(result, Extract(lhs, 0) << Int(rhs), 0);
+ result = Insert(result, Extract(lhs, 1) << Int(rhs), 1);
+ result = Insert(result, Extract(lhs, 2) << Int(rhs), 2);
+ result = Insert(result, Extract(lhs, 3) << Int(rhs), 3);
- storeValue(rhs.value);
+ return result;
+ }
+ else
+ {
+ return RValue<Int4>(Nucleus::createShl(lhs.value, V(::context->getConstantInt32(rhs))));
+ }
}
- Float4::Float4(const Float4 &rhs)
+ RValue<Int4> operator>>(RValue<Int4> lhs, unsigned char rhs)
{
- xyzw.parent = this;
+ if(emulateIntrinsics)
+ {
+ Int4 result;
+ result = Insert(result, Extract(lhs, 0) >> Int(rhs), 0);
+ result = Insert(result, Extract(lhs, 1) >> Int(rhs), 1);
+ result = Insert(result, Extract(lhs, 2) >> Int(rhs), 2);
+ result = Insert(result, Extract(lhs, 3) >> Int(rhs), 3);
- Value *value = rhs.loadValue();
- storeValue(value);
+ return result;
+ }
+ else
+ {
+ return RValue<Int4>(Nucleus::createAShr(lhs.value, V(::context->getConstantInt32(rhs))));
+ }
}
- Float4::Float4(const Reference<Float4> &rhs)
+ RValue<Int4> CmpEQ(RValue<Int4> x, RValue<Int4> y)
{
- xyzw.parent = this;
-
- Value *value = rhs.loadValue();
- storeValue(value);
+ return RValue<Int4>(Nucleus::createICmpEQ(x.value, y.value));
}
- Float4::Float4(RValue<Float> rhs)
+ RValue<Int4> CmpLT(RValue<Int4> x, RValue<Int4> y)
{
- xyzw.parent = this;
-
- assert(false && "UNIMPLEMENTED");
+ return RValue<Int4>(Nucleus::createICmpSLT(x.value, y.value));
}
- Float4::Float4(const Float &rhs)
+ RValue<Int4> CmpLE(RValue<Int4> x, RValue<Int4> y)
{
- xyzw.parent = this;
-
- *this = RValue<Float>(rhs.loadValue());
+ return RValue<Int4>(Nucleus::createICmpSLE(x.value, y.value));
}
- Float4::Float4(const Reference<Float> &rhs)
+ RValue<Int4> CmpNEQ(RValue<Int4> x, RValue<Int4> y)
{
- xyzw.parent = this;
-
- *this = RValue<Float>(rhs.loadValue());
+ return RValue<Int4>(Nucleus::createICmpNE(x.value, y.value));
}
- RValue<Float4> Float4::operator=(float x) const
+ RValue<Int4> CmpNLT(RValue<Int4> x, RValue<Int4> y)
{
- return *this = Float4(x, x, x, x);
+ return RValue<Int4>(Nucleus::createICmpSGE(x.value, y.value));
}
- RValue<Float4> Float4::operator=(RValue<Float4> rhs) const
+ RValue<Int4> CmpNLE(RValue<Int4> x, RValue<Int4> y)
{
- storeValue(rhs.value);
-
- return rhs;
+ return RValue<Int4>(Nucleus::createICmpSGT(x.value, y.value));
}
- RValue<Float4> Float4::operator=(const Float4 &rhs) const
+ RValue<Int4> Max(RValue<Int4> x, RValue<Int4> y)
{
- Value *value = rhs.loadValue();
- storeValue(value);
+ Ice::Variable *condition = ::function->makeVariable(Ice::IceType_v4i1);
+ auto cmp = Ice::InstIcmp::create(::function, Ice::InstIcmp::Sle, condition, x.value, y.value);
+ ::basicBlock->appendInst(cmp);
+
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_v4i32);
+ auto select = Ice::InstSelect::create(::function, result, condition, y.value, x.value);
+ ::basicBlock->appendInst(select);
- return RValue<Float4>(value);
+ return RValue<Int4>(V(result));
}
- RValue<Float4> Float4::operator=(const Reference<Float4> &rhs) const
+ RValue<Int4> Min(RValue<Int4> x, RValue<Int4> y)
{
- Value *value = rhs.loadValue();
- storeValue(value);
+ Ice::Variable *condition = ::function->makeVariable(Ice::IceType_v4i1);
+ auto cmp = Ice::InstIcmp::create(::function, Ice::InstIcmp::Sgt, condition, x.value, y.value);
+ ::basicBlock->appendInst(cmp);
+
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_v4i32);
+ auto select = Ice::InstSelect::create(::function, result, condition, y.value, x.value);
+ ::basicBlock->appendInst(select);
- return RValue<Float4>(value);
+ return RValue<Int4>(V(result));
}
- RValue<Float4> Float4::operator=(RValue<Float> rhs) const
+ RValue<Int4> RoundInt(RValue<Float4> cast)
{
- return *this = Float4(rhs);
+ if(emulateIntrinsics || CPUID::ARM)
+ {
+ // Push the fractional part off the mantissa. Accurate up to +/-2^22.
+ return Int4((cast + Float4(0x00C00000)) - Float4(0x00C00000));
+ }
+ else
+ {
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_v4i32);
+ const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::Nearbyint, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
+ auto target = ::context->getConstantUndef(Ice::IceType_i32);
+ auto nearbyint = Ice::InstIntrinsicCall::create(::function, 1, result, target, intrinsic);
+ nearbyint->addArg(cast.value);
+ ::basicBlock->appendInst(nearbyint);
+
+ return RValue<Int4>(V(result));
+ }
}
- RValue<Float4> Float4::operator=(const Float &rhs) const
+ RValue<Short8> PackSigned(RValue<Int4> x, RValue<Int4> y)
{
- return *this = Float4(rhs);
+ if(emulateIntrinsics)
+ {
+ Short8 result;
+ result = Insert(result, SaturateSigned(Extract(x, 0)), 0);
+ result = Insert(result, SaturateSigned(Extract(x, 1)), 1);
+ result = Insert(result, SaturateSigned(Extract(x, 2)), 2);
+ result = Insert(result, SaturateSigned(Extract(x, 3)), 3);
+ result = Insert(result, SaturateSigned(Extract(y, 0)), 4);
+ result = Insert(result, SaturateSigned(Extract(y, 1)), 5);
+ result = Insert(result, SaturateSigned(Extract(y, 2)), 6);
+ result = Insert(result, SaturateSigned(Extract(y, 3)), 7);
+
+ return result;
+ }
+ else
+ {
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_v8i16);
+ const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::VectorPackSigned, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
+ auto target = ::context->getConstantUndef(Ice::IceType_i32);
+ auto pack = Ice::InstIntrinsicCall::create(::function, 2, result, target, intrinsic);
+ pack->addArg(x.value);
+ pack->addArg(y.value);
+ ::basicBlock->appendInst(pack);
+
+ return RValue<Short8>(V(result));
+ }
}
- RValue<Float4> Float4::operator=(const Reference<Float> &rhs) const
+ RValue<UShort8> PackUnsigned(RValue<Int4> x, RValue<Int4> y)
{
- return *this = Float4(rhs);
+ if(emulateIntrinsics || !(CPUID::SSE4_1 || CPUID::ARM))
+ {
+ RValue<Int4> sx = As<Int4>(x);
+ RValue<Int4> bx = (sx & ~(sx >> 31)) - Int4(0x8000);
+
+ RValue<Int4> sy = As<Int4>(y);
+ RValue<Int4> by = (sy & ~(sy >> 31)) - Int4(0x8000);
+
+ return As<UShort8>(PackSigned(bx, by) + Short8(0x8000u));
+ }
+ else
+ {
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_v8i16);
+ const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::VectorPackUnsigned, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
+ auto target = ::context->getConstantUndef(Ice::IceType_i32);
+ auto pack = Ice::InstIntrinsicCall::create(::function, 2, result, target, intrinsic);
+ pack->addArg(x.value);
+ pack->addArg(y.value);
+ ::basicBlock->appendInst(pack);
+
+ return RValue<UShort8>(V(result));
+ }
}
- RValue<Float4> operator+(RValue<Float4> lhs, RValue<Float4> rhs)
+ RValue<Int> SignMask(RValue<Int4> x)
{
- return RValue<Float4>(Nucleus::createFAdd(lhs.value, rhs.value));
+ if(emulateIntrinsics || CPUID::ARM)
+ {
+ Int4 xx = (x >> 31) & Int4(0x00000001, 0x00000002, 0x00000004, 0x00000008);
+ return Extract(xx, 0) | Extract(xx, 1) | Extract(xx, 2) | Extract(xx, 3);
+ }
+ else
+ {
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_i32);
+ const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::SignMask, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
+ auto target = ::context->getConstantUndef(Ice::IceType_i32);
+ auto movmsk = Ice::InstIntrinsicCall::create(::function, 1, result, target, intrinsic);
+ movmsk->addArg(x.value);
+ ::basicBlock->appendInst(movmsk);
+
+ return RValue<Int>(V(result));
+ }
}
- RValue<Float4> operator-(RValue<Float4> lhs, RValue<Float4> rhs)
+ Type *Int4::getType()
{
- return RValue<Float4>(Nucleus::createFSub(lhs.value, rhs.value));
+ return T(Ice::IceType_v4i32);
}
- RValue<Float4> operator*(RValue<Float4> lhs, RValue<Float4> rhs)
+ UInt4::UInt4(RValue<Float4> cast) : XYZW(this)
{
- return RValue<Float4>(Nucleus::createFMul(lhs.value, rhs.value));
+ // Smallest positive value representable in UInt, but not in Int
+ const unsigned int ustart = 0x80000000u;
+ const float ustartf = float(ustart);
+
+ // Check if the value can be represented as an Int
+ Int4 uiValue = CmpNLT(cast, Float4(ustartf));
+ // If the value is too large, subtract ustart and re-add it after conversion.
+ uiValue = (uiValue & As<Int4>(As<UInt4>(Int4(cast - Float4(ustartf))) + UInt4(ustart))) |
+ // Otherwise, just convert normally
+ (~uiValue & Int4(cast));
+ // If the value is negative, store 0, otherwise store the result of the conversion
+ storeValue((~(As<Int4>(cast) >> 31) & uiValue).value);
}
- RValue<Float4> operator/(RValue<Float4> lhs, RValue<Float4> rhs)
+ RValue<UInt> Extract(RValue<UInt4> x, int i)
{
- return RValue<Float4>(Nucleus::createFDiv(lhs.value, rhs.value));
+ return RValue<UInt>(Nucleus::createExtractElement(x.value, UInt::getType(), i));
}
- RValue<Float4> operator%(RValue<Float4> lhs, RValue<Float4> rhs)
+ RValue<UInt4> Insert(RValue<UInt4> x, RValue<UInt> element, int i)
{
- return RValue<Float4>(Nucleus::createFRem(lhs.value, rhs.value));
+ return RValue<UInt4>(Nucleus::createInsertElement(x.value, element.value, i));
}
- RValue<Float4> operator+=(const Float4 &lhs, RValue<Float4> rhs)
+ RValue<UInt4> operator<<(RValue<UInt4> lhs, unsigned char rhs)
{
- return lhs = lhs + rhs;
+ if(emulateIntrinsics)
+ {
+ UInt4 result;
+ result = Insert(result, Extract(lhs, 0) << UInt(rhs), 0);
+ result = Insert(result, Extract(lhs, 1) << UInt(rhs), 1);
+ result = Insert(result, Extract(lhs, 2) << UInt(rhs), 2);
+ result = Insert(result, Extract(lhs, 3) << UInt(rhs), 3);
+
+ return result;
+ }
+ else
+ {
+ return RValue<UInt4>(Nucleus::createShl(lhs.value, V(::context->getConstantInt32(rhs))));
+ }
}
- RValue<Float4> operator-=(const Float4 &lhs, RValue<Float4> rhs)
+ RValue<UInt4> operator>>(RValue<UInt4> lhs, unsigned char rhs)
{
- return lhs = lhs - rhs;
+ if(emulateIntrinsics)
+ {
+ UInt4 result;
+ result = Insert(result, Extract(lhs, 0) >> UInt(rhs), 0);
+ result = Insert(result, Extract(lhs, 1) >> UInt(rhs), 1);
+ result = Insert(result, Extract(lhs, 2) >> UInt(rhs), 2);
+ result = Insert(result, Extract(lhs, 3) >> UInt(rhs), 3);
+
+ return result;
+ }
+ else
+ {
+ return RValue<UInt4>(Nucleus::createLShr(lhs.value, V(::context->getConstantInt32(rhs))));
+ }
}
- RValue<Float4> operator*=(const Float4 &lhs, RValue<Float4> rhs)
+ RValue<UInt4> CmpEQ(RValue<UInt4> x, RValue<UInt4> y)
{
- return lhs = lhs * rhs;
+ return RValue<UInt4>(Nucleus::createICmpEQ(x.value, y.value));
}
- RValue<Float4> operator/=(const Float4 &lhs, RValue<Float4> rhs)
+ RValue<UInt4> CmpLT(RValue<UInt4> x, RValue<UInt4> y)
{
- return lhs = lhs / rhs;
+ return RValue<UInt4>(Nucleus::createICmpULT(x.value, y.value));
}
- RValue<Float4> operator%=(const Float4 &lhs, RValue<Float4> rhs)
+ RValue<UInt4> CmpLE(RValue<UInt4> x, RValue<UInt4> y)
{
- return lhs = lhs % rhs;
+ return RValue<UInt4>(Nucleus::createICmpULE(x.value, y.value));
}
- RValue<Float4> operator+(RValue<Float4> val)
+ RValue<UInt4> CmpNEQ(RValue<UInt4> x, RValue<UInt4> y)
{
- return val;
+ return RValue<UInt4>(Nucleus::createICmpNE(x.value, y.value));
}
- RValue<Float4> operator-(RValue<Float4> val)
+ RValue<UInt4> CmpNLT(RValue<UInt4> x, RValue<UInt4> y)
{
- return RValue<Float4>(Nucleus::createFNeg(val.value));
+ return RValue<UInt4>(Nucleus::createICmpUGE(x.value, y.value));
}
- RValue<Float4> Abs(RValue<Float4> x)
+ RValue<UInt4> CmpNLE(RValue<UInt4> x, RValue<UInt4> y)
{
- assert(false && "UNIMPLEMENTED"); return RValue<Float4>(V(nullptr));
+ return RValue<UInt4>(Nucleus::createICmpUGT(x.value, y.value));
}
- RValue<Float4> Max(RValue<Float4> x, RValue<Float4> y)
+ RValue<UInt4> Max(RValue<UInt4> x, RValue<UInt4> y)
{
Ice::Variable *condition = ::function->makeVariable(Ice::IceType_v4i1);
- auto cmp = Ice::InstFcmp::create(::function, Ice::InstFcmp::Ule, condition, x.value, y.value);
+ auto cmp = Ice::InstIcmp::create(::function, Ice::InstIcmp::Ule, condition, x.value, y.value);
::basicBlock->appendInst(cmp);
- Ice::Variable *result = ::function->makeVariable(Ice::IceType_v4f32);
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_v4i32);
auto select = Ice::InstSelect::create(::function, result, condition, y.value, x.value);
::basicBlock->appendInst(select);
- return RValue<Float4>(V(result));
+ return RValue<UInt4>(V(result));
}
- RValue<Float4> Min(RValue<Float4> x, RValue<Float4> y)
+ RValue<UInt4> Min(RValue<UInt4> x, RValue<UInt4> y)
{
Ice::Variable *condition = ::function->makeVariable(Ice::IceType_v4i1);
- auto cmp = Ice::InstFcmp::create(::function, Ice::InstFcmp::Ugt, condition, x.value, y.value);
+ auto cmp = Ice::InstIcmp::create(::function, Ice::InstIcmp::Ugt, condition, x.value, y.value);
::basicBlock->appendInst(cmp);
- Ice::Variable *result = ::function->makeVariable(Ice::IceType_v4f32);
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_v4i32);
auto select = Ice::InstSelect::create(::function, result, condition, y.value, x.value);
::basicBlock->appendInst(select);
- return RValue<Float4>(V(result));
+ return RValue<UInt4>(V(result));
}
- RValue<Float4> Rcp_pp(RValue<Float4> x, bool exactAtPow2)
+ Type *UInt4::getType()
{
- return Float4(1.0f) / x;
+ return T(Ice::IceType_v4i32);
}
- RValue<Float4> RcpSqrt_pp(RValue<Float4> x)
+ Type *Half::getType()
+ {
+ return T(Ice::IceType_i16);
+ }
+
+ RValue<Float> Rcp_pp(RValue<Float> x, bool exactAtPow2)
+ {
+ return 1.0f / x;
+ }
+
+ RValue<Float> RcpSqrt_pp(RValue<Float> x)
{
return Rcp_pp(Sqrt(x));
}
- RValue<Float4> Sqrt(RValue<Float4> x)
+ RValue<Float> Sqrt(RValue<Float> x)
{
- Ice::Variable *result = ::function->makeVariable(Ice::IceType_v4f32);
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_f32);
const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::Sqrt, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
auto target = ::context->getConstantUndef(Ice::IceType_i32);
auto sqrt = Ice::InstIntrinsicCall::create(::function, 1, result, target, intrinsic);
sqrt->addArg(x.value);
::basicBlock->appendInst(sqrt);
- return RValue<Float4>(V(result));
+ return RValue<Float>(V(result));
}
- RValue<Float4> Insert(const Float4 &val, RValue<Float> element, int i)
+ RValue<Float> Round(RValue<Float> x)
{
- Value *value = val.loadValue();
- Value *insert = Nucleus::createInsertElement(value, element.value, i);
-
- val = RValue<Float4>(insert);
-
- return val;
+ return Float4(Round(Float4(x))).x;
}
- RValue<Float> Extract(RValue<Float4> x, int i)
+ RValue<Float> Trunc(RValue<Float> x)
{
- return RValue<Float>(Nucleus::createExtractElement(x.value, Float::getType(), i));
+ return Float4(Trunc(Float4(x))).x;
}
- RValue<Float4> Swizzle(RValue<Float4> x, unsigned char select)
+ RValue<Float> Frac(RValue<Float> x)
{
- return RValue<Float4>(createSwizzle4(x.value, select));
+ return Float4(Frac(Float4(x))).x;
}
- RValue<Float4> ShuffleLowHigh(RValue<Float4> x, RValue<Float4> y, unsigned char imm)
+ RValue<Float> Floor(RValue<Float> x)
{
- int shuffle[4] =
- {
- ((imm >> 0) & 0x03) + 0,
- ((imm >> 2) & 0x03) + 0,
- ((imm >> 4) & 0x03) + 4,
- ((imm >> 6) & 0x03) + 4,
- };
-
- return RValue<Float4>(Nucleus::createShuffleVector(x.value, y.value, shuffle));
+ return Float4(Floor(Float4(x))).x;
}
- RValue<Float4> UnpackLow(RValue<Float4> x, RValue<Float4> y)
+ RValue<Float> Ceil(RValue<Float> x)
{
- int shuffle[4] = {0, 4, 1, 5};
- return RValue<Float4>(Nucleus::createShuffleVector(x.value, y.value, shuffle));
+ return Float4(Ceil(Float4(x))).x;
}
- RValue<Float4> UnpackHigh(RValue<Float4> x, RValue<Float4> y)
+ Type *Float::getType()
{
- int shuffle[4] = {2, 6, 3, 7};
- return RValue<Float4>(Nucleus::createShuffleVector(x.value, y.value, shuffle));
+ return T(Ice::IceType_f32);
}
- RValue<Float4> Mask(Float4 &lhs, RValue<Float4> rhs, unsigned char select)
+ Type *Float2::getType()
{
- Value *vector = lhs.loadValue();
- Value *shuffle = createMask4(vector, rhs.value, select);
- lhs.storeValue(shuffle);
-
- return RValue<Float4>(shuffle);
+ return T(Type_v2f32);
}
- RValue<Int> SignMask(RValue<Float4> x)
+ Float4::Float4(RValue<Float> rhs) : XYZW(this)
{
- Ice::Variable *result = ::function->makeVariable(Ice::IceType_i32);
- const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::SignMask, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
- auto target = ::context->getConstantUndef(Ice::IceType_i32);
- auto movmsk = Ice::InstIntrinsicCall::create(::function, 1, result, target, intrinsic);
- movmsk->addArg(x.value);
- ::basicBlock->appendInst(movmsk);
+ Value *vector = Nucleus::createBitCast(rhs.value, Float4::getType());
- return RValue<Int>(V(result));
- }
+ int swizzle[4] = {0, 0, 0, 0};
+ Value *replicate = Nucleus::createShuffleVector(vector, vector, swizzle);
- RValue<Int4> CmpEQ(RValue<Float4> x, RValue<Float4> y)
- {
- return RValue<Int4>(Nucleus::createSExt(Nucleus::createFCmpOEQ(x.value, y.value), Int4::getType()));
+ storeValue(replicate);
}
- RValue<Int4> CmpLT(RValue<Float4> x, RValue<Float4> y)
+ RValue<Float4> Max(RValue<Float4> x, RValue<Float4> y)
{
- return RValue<Int4>(Nucleus::createSExt(Nucleus::createFCmpOLT(x.value, y.value), Int4::getType()));
- }
+ Ice::Variable *condition = ::function->makeVariable(Ice::IceType_v4i1);
+ auto cmp = Ice::InstFcmp::create(::function, Ice::InstFcmp::Ogt, condition, x.value, y.value);
+ ::basicBlock->appendInst(cmp);
- RValue<Int4> CmpLE(RValue<Float4> x, RValue<Float4> y)
- {
- return RValue<Int4>(Nucleus::createSExt(Nucleus::createFCmpOLE(x.value, y.value), Int4::getType()));
- }
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_v4f32);
+ auto select = Ice::InstSelect::create(::function, result, condition, x.value, y.value);
+ ::basicBlock->appendInst(select);
- RValue<Int4> CmpNEQ(RValue<Float4> x, RValue<Float4> y)
- {
- return RValue<Int4>(Nucleus::createSExt(Nucleus::createFCmpONE(x.value, y.value), Int4::getType()));
+ return RValue<Float4>(V(result));
}
- RValue<Int4> CmpNLT(RValue<Float4> x, RValue<Float4> y)
+ RValue<Float4> Min(RValue<Float4> x, RValue<Float4> y)
{
- return RValue<Int4>(Nucleus::createSExt(Nucleus::createFCmpOGE(x.value, y.value), Int4::getType()));
- }
+ Ice::Variable *condition = ::function->makeVariable(Ice::IceType_v4i1);
+ auto cmp = Ice::InstFcmp::create(::function, Ice::InstFcmp::Olt, condition, x.value, y.value);
+ ::basicBlock->appendInst(cmp);
- RValue<Int4> CmpNLE(RValue<Float4> x, RValue<Float4> y)
- {
- return RValue<Int4>(Nucleus::createSExt(Nucleus::createFCmpOGT(x.value, y.value), Int4::getType()));
- }
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_v4f32);
+ auto select = Ice::InstSelect::create(::function, result, condition, x.value, y.value);
+ ::basicBlock->appendInst(select);
- RValue<Float4> Round(RValue<Float4> x)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Float4>(V(nullptr));
+ return RValue<Float4>(V(result));
}
- RValue<Float4> Trunc(RValue<Float4> x)
+ RValue<Float4> Rcp_pp(RValue<Float4> x, bool exactAtPow2)
{
- assert(false && "UNIMPLEMENTED"); return RValue<Float4>(V(nullptr));
+ return Float4(1.0f) / x;
}
- RValue<Float4> Frac(RValue<Float4> x)
+ RValue<Float4> RcpSqrt_pp(RValue<Float4> x)
{
- assert(false && "UNIMPLEMENTED"); return RValue<Float4>(V(nullptr));
+ return Rcp_pp(Sqrt(x));
}
- RValue<Float4> Floor(RValue<Float4> x)
+ RValue<Float4> Sqrt(RValue<Float4> x)
{
- assert(false && "UNIMPLEMENTED"); return RValue<Float4>(V(nullptr));
- }
+ if(emulateIntrinsics || CPUID::ARM)
+ {
+ Float4 result;
+ result.x = Sqrt(Float(Float4(x).x));
+ result.y = Sqrt(Float(Float4(x).y));
+ result.z = Sqrt(Float(Float4(x).z));
+ result.w = Sqrt(Float(Float4(x).w));
- RValue<Float4> Ceil(RValue<Float4> x)
- {
- assert(false && "UNIMPLEMENTED"); return RValue<Float4>(V(nullptr));
+ return result;
+ }
+ else
+ {
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_v4f32);
+ const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::Sqrt, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
+ auto target = ::context->getConstantUndef(Ice::IceType_i32);
+ auto sqrt = Ice::InstIntrinsicCall::create(::function, 1, result, target, intrinsic);
+ sqrt->addArg(x.value);
+ ::basicBlock->appendInst(sqrt);
+
+ return RValue<Float4>(V(result));
+ }
}
- Type *Float4::getType()
+ RValue<Int> SignMask(RValue<Float4> x)
{
- return T(Ice::IceType_v4f32);
+ if(emulateIntrinsics || CPUID::ARM)
+ {
+ Int4 xx = (As<Int4>(x) >> 31) & Int4(0x00000001, 0x00000002, 0x00000004, 0x00000008);
+ return Extract(xx, 0) | Extract(xx, 1) | Extract(xx, 2) | Extract(xx, 3);
+ }
+ else
+ {
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_i32);
+ const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::SignMask, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
+ auto target = ::context->getConstantUndef(Ice::IceType_i32);
+ auto movmsk = Ice::InstIntrinsicCall::create(::function, 1, result, target, intrinsic);
+ movmsk->addArg(x.value);
+ ::basicBlock->appendInst(movmsk);
+
+ return RValue<Int>(V(result));
+ }
}
- RValue<Pointer<Byte>> operator+(RValue<Pointer<Byte>> lhs, int offset)
+ RValue<Int4> CmpEQ(RValue<Float4> x, RValue<Float4> y)
{
- return lhs + RValue<Int>(Nucleus::createConstantInt(offset));
+ return RValue<Int4>(Nucleus::createFCmpOEQ(x.value, y.value));
}
- RValue<Pointer<Byte>> operator+(RValue<Pointer<Byte>> lhs, RValue<Int> offset)
+ RValue<Int4> CmpLT(RValue<Float4> x, RValue<Float4> y)
{
- return RValue<Pointer<Byte>>(Nucleus::createGEP(lhs.value, Byte::getType(), offset.value));
+ return RValue<Int4>(Nucleus::createFCmpOLT(x.value, y.value));
}
- RValue<Pointer<Byte>> operator+(RValue<Pointer<Byte>> lhs, RValue<UInt> offset)
+ RValue<Int4> CmpLE(RValue<Float4> x, RValue<Float4> y)
{
- return RValue<Pointer<Byte>>(Nucleus::createGEP(lhs.value, Byte::getType(), offset.value));
+ return RValue<Int4>(Nucleus::createFCmpOLE(x.value, y.value));
}
- RValue<Pointer<Byte>> operator+=(const Pointer<Byte> &lhs, int offset)
+ RValue<Int4> CmpNEQ(RValue<Float4> x, RValue<Float4> y)
{
- return lhs = lhs + offset;
+ return RValue<Int4>(Nucleus::createFCmpONE(x.value, y.value));
}
- RValue<Pointer<Byte>> operator+=(const Pointer<Byte> &lhs, RValue<Int> offset)
+ RValue<Int4> CmpNLT(RValue<Float4> x, RValue<Float4> y)
{
- return lhs = lhs + offset;
+ return RValue<Int4>(Nucleus::createFCmpOGE(x.value, y.value));
}
- RValue<Pointer<Byte>> operator+=(const Pointer<Byte> &lhs, RValue<UInt> offset)
+ RValue<Int4> CmpNLE(RValue<Float4> x, RValue<Float4> y)
{
- return lhs = lhs + offset;
+ return RValue<Int4>(Nucleus::createFCmpOGT(x.value, y.value));
}
- RValue<Pointer<Byte>> operator-(RValue<Pointer<Byte>> lhs, int offset)
+ RValue<Int4> CmpUEQ(RValue<Float4> x, RValue<Float4> y)
{
- return lhs + -offset;
+ return RValue<Int4>(Nucleus::createFCmpUEQ(x.value, y.value));
}
- RValue<Pointer<Byte>> operator-(RValue<Pointer<Byte>> lhs, RValue<Int> offset)
+ RValue<Int4> CmpULT(RValue<Float4> x, RValue<Float4> y)
{
- return lhs + -offset;
+ return RValue<Int4>(Nucleus::createFCmpULT(x.value, y.value));
}
- RValue<Pointer<Byte>> operator-(RValue<Pointer<Byte>> lhs, RValue<UInt> offset)
+ RValue<Int4> CmpULE(RValue<Float4> x, RValue<Float4> y)
{
- return lhs + -offset;
+ return RValue<Int4>(Nucleus::createFCmpULE(x.value, y.value));
}
- RValue<Pointer<Byte>> operator-=(const Pointer<Byte> &lhs, int offset)
+ RValue<Int4> CmpUNEQ(RValue<Float4> x, RValue<Float4> y)
{
- return lhs = lhs - offset;
+ return RValue<Int4>(Nucleus::createFCmpUNE(x.value, y.value));
}
- RValue<Pointer<Byte>> operator-=(const Pointer<Byte> &lhs, RValue<Int> offset)
+ RValue<Int4> CmpUNLT(RValue<Float4> x, RValue<Float4> y)
{
- return lhs = lhs - offset;
+ return RValue<Int4>(Nucleus::createFCmpUGE(x.value, y.value));
}
- RValue<Pointer<Byte>> operator-=(const Pointer<Byte> &lhs, RValue<UInt> offset)
+ RValue<Int4> CmpUNLE(RValue<Float4> x, RValue<Float4> y)
{
- return lhs = lhs - offset;
+ return RValue<Int4>(Nucleus::createFCmpUGT(x.value, y.value));
}
- void Return()
+ RValue<Float4> Round(RValue<Float4> x)
{
- Nucleus::createRetVoid();
- Nucleus::setInsertBlock(Nucleus::createBasicBlock());
- Nucleus::createUnreachable();
+ if(emulateIntrinsics || CPUID::ARM)
+ {
+ // Push the fractional part off the mantissa. Accurate up to +/-2^22.
+ return (x + Float4(0x00C00000)) - Float4(0x00C00000);
+ }
+ else if(CPUID::SSE4_1)
+ {
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_v4f32);
+ const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::Round, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
+ auto target = ::context->getConstantUndef(Ice::IceType_i32);
+ auto round = Ice::InstIntrinsicCall::create(::function, 2, result, target, intrinsic);
+ round->addArg(x.value);
+ round->addArg(::context->getConstantInt32(0));
+ ::basicBlock->appendInst(round);
+
+ return RValue<Float4>(V(result));
+ }
+ else
+ {
+ return Float4(RoundInt(x));
+ }
}
- void Return(bool ret)
+ RValue<Float4> Trunc(RValue<Float4> x)
{
- Nucleus::createRet(Nucleus::createConstantInt(ret));
- Nucleus::setInsertBlock(Nucleus::createBasicBlock());
- Nucleus::createUnreachable();
+ if(CPUID::SSE4_1)
+ {
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_v4f32);
+ const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::Round, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
+ auto target = ::context->getConstantUndef(Ice::IceType_i32);
+ auto round = Ice::InstIntrinsicCall::create(::function, 2, result, target, intrinsic);
+ round->addArg(x.value);
+ round->addArg(::context->getConstantInt32(3));
+ ::basicBlock->appendInst(round);
+
+ return RValue<Float4>(V(result));
+ }
+ else
+ {
+ return Float4(Int4(x));
+ }
}
- void Return(const Int &ret)
+ RValue<Float4> Frac(RValue<Float4> x)
{
- Nucleus::createRet(ret.loadValue());
- Nucleus::setInsertBlock(Nucleus::createBasicBlock());
- Nucleus::createUnreachable();
- }
+ Float4 frc;
- bool branch(RValue<Bool> cmp, BasicBlock *bodyBB, BasicBlock *endBB)
- {
- Nucleus::createCondBr(cmp.value, bodyBB, endBB);
- Nucleus::setInsertBlock(bodyBB);
+ if(CPUID::SSE4_1)
+ {
+ frc = x - Floor(x);
+ }
+ else
+ {
+ frc = x - Float4(Int4(x)); // Signed fractional part.
+
+ frc += As<Float4>(As<Int4>(CmpNLE(Float4(0.0f), frc)) & As<Int4>(Float4(1, 1, 1, 1))); // Add 1.0 if negative.
+ }
- return true;
+ // x - floor(x) can be 1.0 for very small negative x.
+ // Clamp against the value just below 1.0.
+ return Min(frc, As<Float4>(Int4(0x3F7FFFFF)));
}
- void endIf(BasicBlock *falseBB)
+ RValue<Float4> Floor(RValue<Float4> x)
{
- ::falseBB = falseBB;
+ if(CPUID::SSE4_1)
+ {
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_v4f32);
+ const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::Round, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
+ auto target = ::context->getConstantUndef(Ice::IceType_i32);
+ auto round = Ice::InstIntrinsicCall::create(::function, 2, result, target, intrinsic);
+ round->addArg(x.value);
+ round->addArg(::context->getConstantInt32(1));
+ ::basicBlock->appendInst(round);
+
+ return RValue<Float4>(V(result));
+ }
+ else
+ {
+ return x - Frac(x);
+ }
}
- bool elseBlock(BasicBlock *falseBB)
+ RValue<Float4> Ceil(RValue<Float4> x)
{
- assert(falseBB && "Else not preceded by If");
- falseBB->getInsts().back().setDeleted();
- Nucleus::setInsertBlock(falseBB);
-
- return true;
+ if(CPUID::SSE4_1)
+ {
+ Ice::Variable *result = ::function->makeVariable(Ice::IceType_v4f32);
+ const Ice::Intrinsics::IntrinsicInfo intrinsic = {Ice::Intrinsics::Round, Ice::Intrinsics::SideEffects_F, Ice::Intrinsics::ReturnsTwice_F, Ice::Intrinsics::MemoryWrite_F};
+ auto target = ::context->getConstantUndef(Ice::IceType_i32);
+ auto round = Ice::InstIntrinsicCall::create(::function, 2, result, target, intrinsic);
+ round->addArg(x.value);
+ round->addArg(::context->getConstantInt32(2));
+ ::basicBlock->appendInst(round);
+
+ return RValue<Float4>(V(result));
+ }
+ else
+ {
+ return -Floor(-x);
+ }
}
- BasicBlock *beginElse()
+ Type *Float4::getType()
{
- BasicBlock *falseBB = ::falseBB;
- ::falseBB = nullptr;
-
- return falseBB;
+ return T(Ice::IceType_v4f32);
}
RValue<Long> Ticks()
assert(false && "UNIMPLEMENTED"); return RValue<Long>(V(nullptr));
}
}
-
OSDN Git Service
