steam-frame/FEX-Emu--FEX
1
0 Fork 0
mirror of https://github.com/FEX-Emu/FEX.git synced 2026-10-29 04:01:27 +01:00
Code Issues Packages Projects Releases 67 Wiki Activity

Merge pull request #2944 from lioncash/interparray

Browse Source 
Interpreter: Use alias for temporary vector data
This commit is contained in:
Ryan Houdek authored and GitHub committed 2023-08-21 11:28:24 -07:00
2 parents 084d102c9c a9a7cbce21
commit 8b9ee997b2
4 files changed
+334 -324
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

FEXCore/Source/Interface/Core/Interpreter/ConversionOps.cpp
+13 -13
View File
@@ -69,14 +69,14 @@ DEF_OP(VDupFromGPR) {
const auto ElementSize = IROp->ElementSize;
const auto NumElements = OpSize / IROp->ElementSize;
uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};
TempVectorDataArray Tmp{};
const auto *Src = GetSrc<void*>(Data->SSAData, Op->Src);
for (size_t i = 0; i < NumElements; i++) {
memcpy(Tmp + (i * ElementSize), Src, ElementSize);
memcpy(&Tmp[i * ElementSize], Src, ElementSize);
}
memcpy(GDP, Tmp, sizeof(Tmp));
memcpy(GDP, Tmp.data(), sizeof(Tmp));
}
DEF_OP(Float_FromGPR_S) {
@@ -130,7 +130,7 @@ DEF_OP(Vector_SToF) {
const uint8_t OpSize = IROp->Size;
void *Src = GetSrc<void*>(Data->SSAData, Op->Vector);
uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};
TempVectorDataArray Tmp{};
const uint8_t ElementSize = Op->Header.ElementSize;
const uint8_t Elements = OpSize / ElementSize;
@@ -143,7 +143,7 @@ DEF_OP(Vector_SToF) {
LOGMAN_MSG_A_FMT("Unknown Element Size: {}", ElementSize);
break;
}
memcpy(GDP, Tmp, OpSize);
memcpy(GDP, Tmp.data(), OpSize);
}
DEF_OP(Vector_FToZS) {
@@ -151,7 +151,7 @@ DEF_OP(Vector_FToZS) {
const uint8_t OpSize = IROp->Size;
void *Src = GetSrc<void*>(Data->SSAData, Op->Vector);
uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};
TempVectorDataArray Tmp{};
const uint8_t ElementSize = Op->Header.ElementSize;
const uint8_t Elements = OpSize / ElementSize;
@@ -164,7 +164,7 @@ DEF_OP(Vector_FToZS) {
LOGMAN_MSG_A_FMT("Unknown Element Size: {}", ElementSize);
break;
}
memcpy(GDP, Tmp, OpSize);
memcpy(GDP, Tmp.data(), OpSize);
}
DEF_OP(Vector_FToS) {
@@ -172,7 +172,7 @@ DEF_OP(Vector_FToS) {
const uint8_t OpSize = IROp->Size;
void *Src = GetSrc<void*>(Data->SSAData, Op->Vector);
uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};
TempVectorDataArray Tmp{};
const uint8_t ElementSize = Op->Header.ElementSize;
const uint8_t Elements = OpSize / ElementSize;
@@ -185,7 +185,7 @@ DEF_OP(Vector_FToS) {
LOGMAN_MSG_A_FMT("Unknown Element Size: {}", ElementSize);
break;
}
memcpy(GDP, Tmp, OpSize);
memcpy(GDP, Tmp.data(), OpSize);
}
DEF_OP(Vector_FToF) {
@@ -193,7 +193,7 @@ DEF_OP(Vector_FToF) {
const uint8_t OpSize = IROp->Size;
void *Src = GetSrc<void*>(Data->SSAData, Op->Vector);
uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};
TempVectorDataArray Tmp{};
const uint16_t ElementSize = Op->Header.ElementSize;
const uint16_t Conv = (ElementSize << 8) | Op->SrcElementSize;
@@ -220,7 +220,7 @@ DEF_OP(Vector_FToF) {
LOGMAN_MSG_A_FMT("Unknown Conversion Type : 0x{:04x}", Conv);
break;
}
memcpy(GDP, Tmp, OpSize);
memcpy(GDP, Tmp.data(), OpSize);
}
DEF_OP(Vector_FToI) {
@@ -228,7 +228,7 @@ DEF_OP(Vector_FToI) {
const uint8_t OpSize = IROp->Size;
void *Src = GetSrc<void*>(Data->SSAData, Op->Vector);
uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};
TempVectorDataArray Tmp{};
const uint8_t ElementSize = Op->Header.ElementSize;
const uint8_t Elements = OpSize / ElementSize;
@@ -270,7 +270,7 @@ DEF_OP(Vector_FToI) {
}
break;
}
memcpy(GDP, Tmp, OpSize);
memcpy(GDP, Tmp.data(), OpSize);
}
#undef DEF_OP
FEXCore/Source/Interface/Core/Interpreter/InterpreterDefines.h
+138 -137
View File
@@ -1,172 +1,173 @@
#pragma once
#include <array>
#include <FEXCore/IR/IR.h>
#define GD *GetDest<uint64_t*>(Data->SSAData, Node)
#define GDP GetDest<void*>(Data->SSAData, Node)
#define DO_OP(size, type, func) \
case size: { \
#define DO_OP(size, type, func) \
case size: { \
auto *Dst_d = reinterpret_cast<type*>(GDP); \
auto *Src1_d = reinterpret_cast<type*>(Src1); \
auto *Src2_d = reinterpret_cast<type*>(Src2); \
*Dst_d = func(*Src1_d, *Src2_d); \
break; \
*Dst_d = func(*Src1_d, *Src2_d); \
break; \
}
#define DO_SCALAR_COMPARE_OP(size, type, type2, func) \
case size: { \
auto *Dst_d = reinterpret_cast<type2*>(Tmp); \
auto *Src1_d = reinterpret_cast<type*>(Src1); \
auto *Src2_d = reinterpret_cast<type*>(Src2); \
Dst_d[0] = func(Src1_d[0], Src2_d[0]); \
break; \
#define DO_SCALAR_COMPARE_OP(size, type, type2, func) \
case size: { \
auto *Dst_d = reinterpret_cast<type2*>(std::data(Tmp)); \
auto *Src1_d = reinterpret_cast<type*>(Src1); \
auto *Src2_d = reinterpret_cast<type*>(Src2); \
Dst_d[0] = func(Src1_d[0], Src2_d[0]); \
break; \
}
#define DO_VECTOR_COMPARE_OP(size, type, type2, func) \
case size: { \
auto *Dst_d = reinterpret_cast<type2*>(Tmp); \
auto *Src1_d = reinterpret_cast<type*>(Src1); \
auto *Src2_d = reinterpret_cast<type*>(Src2); \
for (uint8_t i = 0; i < Elements; ++i) { \
Dst_d[i] = func(Src1_d[i], Src2_d[i]); \
} \
break; \
#define DO_VECTOR_COMPARE_OP(size, type, type2, func) \
case size: { \
auto *Dst_d = reinterpret_cast<type2*>(std::data(Tmp)); \
auto *Src1_d = reinterpret_cast<type*>(Src1); \
auto *Src2_d = reinterpret_cast<type*>(Src2); \
for (uint8_t i = 0; i < Elements; ++i) { \
Dst_d[i] = func(Src1_d[i], Src2_d[i]); \
} \
break; \
}
#define DO_VECTOR_OP(size, type, func) \
case size: { \
auto *Dst_d = reinterpret_cast<type*>(Tmp); \
auto *Src1_d = reinterpret_cast<type*>(Src1); \
auto *Src2_d = reinterpret_cast<type*>(Src2); \
for (uint8_t i = 0; i < Elements; ++i) { \
Dst_d[i] = func(Src1_d[i], Src2_d[i]); \
} \
break; \
#define DO_VECTOR_OP(size, type, func) \
case size: { \
auto *Dst_d = reinterpret_cast<type*>(std::data(Tmp)); \
auto *Src1_d = reinterpret_cast<type*>(Src1); \
auto *Src2_d = reinterpret_cast<type*>(Src2); \
for (uint8_t i = 0; i < Elements; ++i) { \
Dst_d[i] = func(Src1_d[i], Src2_d[i]); \
} \
break; \
}
#define DO_VECTOR_PAIR_OP(size, type, func) \
case size: { \
auto *Dst_d = reinterpret_cast<type*>(Tmp); \
auto *Src1_d = reinterpret_cast<type*>(Src1); \
auto *Src2_d = reinterpret_cast<type*>(Src2); \
for (uint8_t i = 0; i < Elements; ++i) { \
#define DO_VECTOR_PAIR_OP(size, type, func) \
case size: { \
auto *Dst_d = reinterpret_cast<type*>(std::data(Tmp)); \
auto *Src1_d = reinterpret_cast<type*>(Src1); \
auto *Src2_d = reinterpret_cast<type*>(Src2); \
for (uint8_t i = 0; i < Elements; ++i) { \
Dst_d[i] = func(Src1_d[i*2], Src1_d[i*2 + 1]); \
Dst_d[i+Elements] = func(Src2_d[i*2], Src2_d[i*2 + 1]); \
} \
break; \
Dst_d[i+Elements] = func(Src2_d[i*2], Src2_d[i*2 + 1]); \
} \
break; \
}
#define DO_VECTOR_SCALAR_OP(size, type, func)\
case size: { \
auto *Dst_d = reinterpret_cast<type*>(Tmp); \
auto *Src1_d = reinterpret_cast<type*>(Src1); \
auto *Src2_d = reinterpret_cast<type*>(Src2); \
for (uint8_t i = 0; i < Elements; ++i) { \
Dst_d[i] = func(Src1_d[i], *Src2_d); \
} \
break; \
#define DO_VECTOR_SCALAR_OP(size, type, func) \
case size: { \
auto *Dst_d = reinterpret_cast<type*>(std::data(Tmp)); \
auto *Src1_d = reinterpret_cast<type*>(Src1); \
auto *Src2_d = reinterpret_cast<type*>(Src2); \
for (uint8_t i = 0; i < Elements; ++i) { \
Dst_d[i] = func(Src1_d[i], *Src2_d); \
} \
break; \
}
#define DO_VECTOR_SCALAR_WIDE_OP(size, type, func)\
case size: { \
auto *Dst_d = reinterpret_cast<type*>(Tmp); \
auto *Src1_d = reinterpret_cast<type*>(Src1); \
for (uint8_t i = 0; i < Elements; ++i) { \
Dst_d[i] = func(Src1_d[i], Src2); \
} \
break; \
#define DO_VECTOR_SCALAR_WIDE_OP(size, type, func) \
case size: { \
auto *Dst_d = reinterpret_cast<type*>(std::data(Tmp)); \
auto *Src1_d = reinterpret_cast<type*>(Src1); \
for (uint8_t i = 0; i < Elements; ++i) { \
Dst_d[i] = func(Src1_d[i], Src2); \
} \
break; \
}
#define DO_VECTOR_0SRC_OP(size, type, func) \
case size: { \
auto *Dst_d = reinterpret_cast<type*>(Tmp); \
for (uint8_t i = 0; i < Elements; ++i) { \
Dst_d[i] = func(); \
} \
break; \
#define DO_VECTOR_0SRC_OP(size, type, func) \
case size: { \
auto *Dst_d = reinterpret_cast<type*>(std::data(Tmp)); \
for (uint8_t i = 0; i < Elements; ++i) { \
Dst_d[i] = func(); \
} \
break; \
}
#define DO_VECTOR_1SRC_OP(size, type, func) \
case size: { \
auto *Dst_d = reinterpret_cast<type*>(Tmp); \
auto *Src_d = reinterpret_cast<type*>(Src); \
for (uint8_t i = 0; i < Elements; ++i) { \
Dst_d[i] = func(Src_d[i]); \
} \
break; \
#define DO_VECTOR_1SRC_OP(size, type, func) \
case size: { \
auto *Dst_d = reinterpret_cast<type*>(std::data(Tmp)); \
auto *Src_d = reinterpret_cast<type*>(Src); \
for (uint8_t i = 0; i < Elements; ++i) { \
Dst_d[i] = func(Src_d[i]); \
} \
break; \
}
#define DO_VECTOR_REDUCE_1SRC_OP(size, type, func, start_val) \
case size: { \
auto *Dst_d = reinterpret_cast<type*>(Tmp); \
auto *Src_d = reinterpret_cast<type*>(Src); \
type begin = start_val; \
for (uint8_t i = 0; i < Elements; ++i) { \
begin = func(begin, Src_d[i]); \
} \
Dst_d[0] = begin; \
break; \
#define DO_VECTOR_REDUCE_1SRC_OP(size, type, func, start_val) \
case size: { \
auto *Dst_d = reinterpret_cast<type*>(std::data(Tmp)); \
auto *Src_d = reinterpret_cast<type*>(Src); \
type begin = start_val; \
for (uint8_t i = 0; i < Elements; ++i) { \
begin = func(begin, Src_d[i]); \
} \
Dst_d[0] = begin; \
break; \
}
#define DO_VECTOR_SAT_OP(size, type, func, min, max) \
case size: { \
auto *Dst_d = reinterpret_cast<type*>(Tmp); \
auto *Src1_d = reinterpret_cast<type*>(Src1); \
auto *Src2_d = reinterpret_cast<type*>(Src2); \
for (uint8_t i = 0; i < Elements; ++i) { \
Dst_d[i] = func(Src1_d[i], Src2_d[i], min, max); \
} \
break; \
#define DO_VECTOR_SAT_OP(size, type, func, min, max) \
case size: { \
auto *Dst_d = reinterpret_cast<type*>(std::data(Tmp)); \
auto *Src1_d = reinterpret_cast<type*>(Src1); \
auto *Src2_d = reinterpret_cast<type*>(Src2); \
for (uint8_t i = 0; i < Elements; ++i) { \
Dst_d[i] = func(Src1_d[i], Src2_d[i], min, max); \
} \
break; \
}
#define DO_VECTOR_1SRC_2TYPE_OP(size, type, type2, func, min, max) \
case size: { \
auto *Dst_d = reinterpret_cast<type*>(Tmp); \
auto *Src_d = reinterpret_cast<type2*>(Src); \
for (uint8_t i = 0; i < Elements; ++i) { \
Dst_d[i] = (type)func(Src_d[i], min, max); \
} \
break; \
#define DO_VECTOR_1SRC_2TYPE_OP(size, type, type2, func, min, max) \
case size: { \
auto *Dst_d = reinterpret_cast<type*>(std::data(Tmp)); \
auto *Src_d = reinterpret_cast<type2*>(Src); \
for (uint8_t i = 0; i < Elements; ++i) { \
Dst_d[i] = (type)func(Src_d[i], min, max); \
} \
break; \
}
#define DO_VECTOR_1SRC_2TYPE_OP_NOSIZE(type, type2, func, min, max) \
auto *Dst_d = reinterpret_cast<type*>(Tmp); \
auto *Src_d = reinterpret_cast<type2*>(Src); \
for (uint8_t i = 0; i < Elements; ++i) { \
Dst_d[i] = (type)func(Src_d[i], min, max); \
#define DO_VECTOR_1SRC_2TYPE_OP_NOSIZE(type, type2, func, min, max) \
auto *Dst_d = reinterpret_cast<type*>(std::data(Tmp)); \
auto *Src_d = reinterpret_cast<type2*>(Src); \
for (uint8_t i = 0; i < Elements; ++i) { \
Dst_d[i] = (type)func(Src_d[i], min, max); \
}
#define DO_VECTOR_1SRC_2TYPE_OP_TOP(size, type, type2, func, min, max) \
case size: { \
auto *Dst_d = reinterpret_cast<type*>(Tmp); \
auto *Src_d = reinterpret_cast<type2*>(Src2); \
memcpy(Dst_d, Src1, Elements * sizeof(type2));\
for (uint8_t i = 0; i < Elements; ++i) { \
Dst_d[i+Elements] = (type)func(Src_d[i], min, max); \
} \
break; \
#define DO_VECTOR_1SRC_2TYPE_OP_TOP(size, type, type2, func, min, max) \
case size: { \
auto *Dst_d = reinterpret_cast<type*>(std::data(Tmp)); \
auto *Src_d = reinterpret_cast<type2*>(Src2); \
memcpy(Dst_d, Src1, Elements * sizeof(type2)); \
for (uint8_t i = 0; i < Elements; ++i) { \
Dst_d[i+Elements] = (type)func(Src_d[i], min, max); \
} \
break; \
}
#define DO_VECTOR_1SRC_2TYPE_OP_TOP_SRC(size, type, type2, func, min, max) \
case size: { \
auto *Dst_d = reinterpret_cast<type*>(Tmp); \
auto *Src_d = reinterpret_cast<type2*>(Src); \
for (uint8_t i = 0; i < Elements; ++i) { \
Dst_d[i] = (type)func(Src_d[i+Elements], min, max); \
} \
break; \
#define DO_VECTOR_1SRC_2TYPE_OP_TOP_SRC(size, type, type2, func, min, max) \
case size: { \
auto *Dst_d = reinterpret_cast<type*>(std::data(Tmp)); \
auto *Src_d = reinterpret_cast<type2*>(Src); \
for (uint8_t i = 0; i < Elements; ++i) { \
Dst_d[i] = (type)func(Src_d[i+Elements], min, max); \
} \
break; \
}
#define DO_VECTOR_2SRC_2TYPE_OP(size, type, type2, func) \
case size: { \
auto *Dst_d = reinterpret_cast<type*>(Tmp); \
auto *Src1_d = reinterpret_cast<type2*>(Src1); \
auto *Src2_d = reinterpret_cast<type2*>(Src2); \
for (uint8_t i = 0; i < Elements; ++i) { \
Dst_d[i] = (type)func((type)Src1_d[i], (type)Src2_d[i]); \
} \
break; \
#define DO_VECTOR_2SRC_2TYPE_OP(size, type, type2, func) \
case size: { \
auto *Dst_d = reinterpret_cast<type*>(std::data(Tmp)); \
auto *Src1_d = reinterpret_cast<type2*>(Src1); \
auto *Src2_d = reinterpret_cast<type2*>(Src2); \
for (uint8_t i = 0; i < Elements; ++i) { \
Dst_d[i] = (type)func((type)Src1_d[i], (type)Src2_d[i]); \
} \
break; \
}
#define DO_VECTOR_2SRC_2TYPE_OP_TOP_SRC(size, type, type2, func) \
case size: { \
auto *Dst_d = reinterpret_cast<type*>(Tmp); \
auto *Src1_d = reinterpret_cast<type2*>(Src1); \
auto *Src2_d = reinterpret_cast<type2*>(Src2); \
for (uint8_t i = 0; i < Elements; ++i) { \
Dst_d[i] = (type)func((type)Src1_d[i+Elements], (type)Src2_d[i+Elements]); \
} \
break; \
#define DO_VECTOR_2SRC_2TYPE_OP_TOP_SRC(size, type, type2, func) \
case size: { \
auto *Dst_d = reinterpret_cast<type*>(std::data(Tmp)); \
auto *Src1_d = reinterpret_cast<type2*>(Src1); \
auto *Src2_d = reinterpret_cast<type2*>(Src2); \
for (uint8_t i = 0; i < Elements; ++i) { \
Dst_d[i] = (type)func((type)Src1_d[i+Elements], (type)Src2_d[i+Elements]); \
} \
break; \
}
struct InterpVector256 {
FEXCore/Source/Interface/Core/Interpreter/InterpreterOps.h
+11 -2
View File
@@ -1,5 +1,7 @@
#pragma once
#include <stdint.h>
#include <array>
#include <cstdint>
#include <FEXCore/Core/CoreState.h>
#include <FEXCore/IR/IR.h>
@@ -47,7 +49,6 @@ namespace FEXCore::CPU {
};
class InterpreterOps {
public:
static void InterpretIR(FEXCore::Core::CpuStateFrame *Frame, FEXCore::IR::IRListView const *IR);
static void FillFallbackIndexPointers(uint64_t *Info);
@@ -432,5 +433,13 @@ namespace FEXCore::CPU {
return IROp->Size;
}
// Alias for specifying temporary data that is operated on
// before storing into a destination.
//
// NOTE: This should be the maximum size a vector can be within
// FEX. e.g. If we ever support AVX-512, this should be changed
// to be 64 bytes in size.
using TempVectorDataArray = std::array<uint8_t, Core::CPUState::XMM_AVX_REG_SIZE>;
};
} // namespace FEXCore::CPU
FEXCore/Source/Interface/Core/Interpreter/VectorOps.cpp
+172 -172
View File
File diff suppressed because it is too large. Load diff