mirror of
https://github.com/cemu-project/Cemu.git
synced 2025-07-05 06:21:19 +12:00
Add all the files
This commit is contained in:
Exzap
parent
e3db07a16a
commit
d60742f52b
1445 changed files with 430238 additions and 0 deletions
540
src/Cafe/HW/Latte/Core/FetchShader.cpp
Normal file
540
src/Cafe/HW/Latte/Core/FetchShader.cpp
Normal file
|
|
@ -0,0 +1,540 @@
|
|||
#include "Cafe/HW/Latte/Core/LatteConst.h"
|
||||
#include "Cafe/HW/Latte/Core/LatteShaderAssembly.h"
|
||||
#include "Cafe/HW/Latte/ISA/RegDefines.h"
|
||||
#include "Cafe/OS/libs/gx2/GX2.h"
|
||||
#include "Cafe/HW/Latte/Core/Latte.h"
|
||||
#include "Cafe/HW/Latte/Core/LatteDraw.h"
|
||||
#include "Cafe/HW/Latte/LegacyShaderDecompiler/LatteDecompiler.h"
|
||||
#include "Cafe/HW/Latte/LegacyShaderDecompiler/LatteDecompilerInstructions.h"
|
||||
#include "Cafe/HW/Latte/Core/FetchShader.h"
|
||||
#include "Cafe/HW/Latte/ISA/LatteInstructions.h"
|
||||
#include "util/containers/LookupTableL3.h"
|
||||
#include "util/helpers/fspinlock.h"
|
||||
#include <openssl/sha.h>
|
||||
|
||||
uint32 LatteShaderRecompiler_getAttributeSize(LatteParsedFetchShaderAttribute_t* attrib)
|
||||
{
|
||||
if (attrib->format == FMT_32_32_32_32 || attrib->format == FMT_32_32_32_32_FLOAT)
|
||||
return 4 * 4;
|
||||
else if (attrib->format == FMT_32_32_32 || attrib->format == FMT_32_32_32_FLOAT)
|
||||
return 3 * 4;
|
||||
else if (attrib->format == FMT_32_32 || attrib->format == FMT_32_32_FLOAT)
|
||||
return 2 * 4;
|
||||
else if (attrib->format == FMT_32 || attrib->format == FMT_32_FLOAT)
|
||||
return 1 * 4;
|
||||
else if (attrib->format == FMT_16_16_16_16 || attrib->format == FMT_16_16_16_16_FLOAT)
|
||||
return 4 * 2;
|
||||
else if (attrib->format == FMT_16_16 || attrib->format == FMT_16_16_FLOAT)
|
||||
return 2 * 2;
|
||||
else if (attrib->format == FMT_16 || attrib->format == FMT_16_FLOAT)
|
||||
return 1 * 2;
|
||||
else if (attrib->format == FMT_8_8_8_8)
|
||||
return 4 * 1;
|
||||
else if (attrib->format == FMT_8_8)
|
||||
return 2 * 1;
|
||||
else if (attrib->format == FMT_8)
|
||||
return 1 * 1;
|
||||
else if (attrib->format == FMT_2_10_10_10)
|
||||
return 4;
|
||||
else
|
||||
cemu_assert_unimplemented();
|
||||
return 0;
|
||||
}
|
||||
|
||||
uint32 LatteShaderRecompiler_getAttributeAlignment(LatteParsedFetchShaderAttribute_t* attrib)
|
||||
{
|
||||
if (attrib->format == FMT_32_32_32_32 || attrib->format == FMT_32_32_32_32_FLOAT)
|
||||
return 4;
|
||||
else if (attrib->format == FMT_32_32_32 || attrib->format == FMT_32_32_32_FLOAT)
|
||||
return 4;
|
||||
else if (attrib->format == FMT_32_32 || attrib->format == FMT_32_32_FLOAT)
|
||||
return 4;
|
||||
else if (attrib->format == FMT_32 || attrib->format == FMT_32_FLOAT)
|
||||
return 4;
|
||||
else if (attrib->format == FMT_16_16_16_16 || attrib->format == FMT_16_16_16_16_FLOAT)
|
||||
return 2;
|
||||
else if (attrib->format == FMT_16_16 || attrib->format == FMT_16_16_FLOAT)
|
||||
return 2;
|
||||
else if (attrib->format == FMT_16 || attrib->format == FMT_16_FLOAT)
|
||||
return 2;
|
||||
else if (attrib->format == FMT_8_8_8_8)
|
||||
return 1;
|
||||
else if (attrib->format == FMT_8_8)
|
||||
return 1;
|
||||
else if (attrib->format == FMT_8)
|
||||
return 1;
|
||||
else if (attrib->format == FMT_2_10_10_10)
|
||||
return 4;
|
||||
else
|
||||
cemu_assert_unimplemented();
|
||||
return 4;
|
||||
}
|
||||
|
||||
void LatteShader_calculateFSKey(LatteFetchShader* fetchShader)
|
||||
{
|
||||
uint64 key = 0;
|
||||
for (sint32 g = 0; g < fetchShader->bufferGroups.size(); g++)
|
||||
{
|
||||
LatteParsedFetchShaderBufferGroup_t& group = fetchShader->bufferGroups[g];
|
||||
for (sint32 f = 0; f < group.attribCount; f++)
|
||||
{
|
||||
LatteParsedFetchShaderAttribute_t* attrib = group.attrib + f;
|
||||
key += (uint64)attrib->endianSwap;
|
||||
key = _rotl64(key, 3);
|
||||
key += (uint64)attrib->nfa;
|
||||
key = _rotl64(key, 3);
|
||||
key += (uint64)(attrib->isSigned?1:0);
|
||||
key = _rotl64(key, 1);
|
||||
key += (uint64)attrib->format;
|
||||
key = _rotl64(key, 7);
|
||||
key += (uint64)attrib->fetchType;
|
||||
key = _rotl64(key, 8);
|
||||
key += (uint64)attrib->ds[0];
|
||||
key = _rotl64(key, 2);
|
||||
key += (uint64)attrib->ds[1];
|
||||
key = _rotl64(key, 2);
|
||||
key += (uint64)attrib->ds[2];
|
||||
key = _rotl64(key, 2);
|
||||
key += (uint64)attrib->ds[3];
|
||||
key = _rotl64(key, 2);
|
||||
key += (uint64)(attrib->aluDivisor+1);
|
||||
key = _rotl64(key, 2);
|
||||
key += (uint64)attrib->attributeBufferIndex;
|
||||
key = _rotl64(key, 8);
|
||||
key += (uint64)attrib->semanticId;
|
||||
key = _rotl64(key, 8);
|
||||
key += (uint64)(attrib->offset & 3);
|
||||
key = _rotl64(key, 2);
|
||||
}
|
||||
}
|
||||
// todo - also hash invalid buffer groups?
|
||||
fetchShader->key = key;
|
||||
}
|
||||
|
||||
uint32 LatteParsedFetchShaderBufferGroup_t::getCurrentBufferStride(uint32* contextRegister) const
|
||||
{
|
||||
uint32 bufferIndex = this->attributeBufferIndex;
|
||||
uint32 bufferBaseRegisterIndex = mmSQ_VTX_ATTRIBUTE_BLOCK_START + bufferIndex * 7;
|
||||
uint32 bufferStride = (contextRegister[bufferBaseRegisterIndex + 2] >> 11) & 0xFFFF;
|
||||
return bufferStride;
|
||||
}
|
||||
|
||||
void LatteFetchShader::CalculateFetchShaderVkHash()
|
||||
{
|
||||
// calculate SHA1 of all states that are part of the Vulkan graphics pipeline
|
||||
SHA_CTX ctx;
|
||||
SHA1_Init(&ctx);
|
||||
for(auto& group : bufferGroups)
|
||||
{
|
||||
// offsets
|
||||
for (sint32 t = 0; t < group.attribCount; t++)
|
||||
{
|
||||
uint32 offset = group.attrib[t].offset;
|
||||
SHA1_Update(&ctx, &t, sizeof(t));
|
||||
SHA1_Update(&ctx, &offset, sizeof(offset));
|
||||
}
|
||||
}
|
||||
uint8 shaDigest[20];
|
||||
SHA1_Final(shaDigest, &ctx);
|
||||
// fold SHA1 hash into a 64bit value
|
||||
uint64 h = *(uint64*)(shaDigest + 0);
|
||||
h += *(uint64*)(shaDigest + 8);
|
||||
h += (uint64)*(uint32*)(shaDigest + 16);
|
||||
this->vkPipelineHashFragment = h;
|
||||
}
|
||||
|
||||
void _fetchShaderDecompiler_parseInstruction_VTX_SEMANTIC(LatteFetchShader* parsedFetchShader, uint32* contextRegister, const LatteClauseInstruction_VTX* instr)
|
||||
{
|
||||
uint32 semanticId = instr->getFieldSEM_SEMANTIC_ID(); // location (attribute index inside shader)
|
||||
uint32 bufferId = instr->getField_BUFFER_ID(); // the index used for GX2SetAttribBuffer (+0xA0)
|
||||
LatteConst::VertexFetchType2 fetchType = instr->getField_FETCH_TYPE();
|
||||
auto srcSelX = instr->getField_SRC_SEL_X();
|
||||
auto dsx = instr->getField_DST_SEL(0);
|
||||
auto dsy = instr->getField_DST_SEL(1);
|
||||
auto dsz = instr->getField_DST_SEL(2);
|
||||
auto dsw = instr->getField_DST_SEL(3);
|
||||
auto dataFormat = instr->getField_DATA_FORMAT();
|
||||
uint32 offset = instr->getField_OFFSET();
|
||||
auto nfa = instr->getField_NUM_FORMAT_ALL();
|
||||
bool isSigned = instr->getField_FORMAT_COMP_ALL() == LatteClauseInstruction_VTX::FORMAT_COMP::COMP_SIGNED;
|
||||
auto endianSwap = instr->getField_ENDIAN_SWAP();
|
||||
|
||||
// get buffer
|
||||
cemu_assert_debug(bufferId >= 0xA0 && bufferId < 0xB0);
|
||||
uint32 bufferIndex = (bufferId - 0xA0);
|
||||
|
||||
// get or add new attribute group (by buffer index)
|
||||
LatteParsedFetchShaderBufferGroup_t* attribGroup = nullptr;
|
||||
if (LatteFetchShader::isValidBufferIndex(bufferIndex))
|
||||
{
|
||||
auto bufferGroupItr = std::find_if(parsedFetchShader->bufferGroups.begin(), parsedFetchShader->bufferGroups.end(), [bufferIndex](LatteParsedFetchShaderBufferGroup_t& bufferGroup) {return bufferGroup.attributeBufferIndex == bufferIndex; });
|
||||
if (bufferGroupItr != parsedFetchShader->bufferGroups.end())
|
||||
attribGroup = &(*bufferGroupItr);
|
||||
}
|
||||
else
|
||||
{
|
||||
auto bufferGroupItr = std::find_if(parsedFetchShader->bufferGroupsInvalid.begin(), parsedFetchShader->bufferGroupsInvalid.end(), [bufferIndex](LatteParsedFetchShaderBufferGroup_t& bufferGroup) {return bufferGroup.attributeBufferIndex == bufferIndex; });
|
||||
if (bufferGroupItr != parsedFetchShader->bufferGroupsInvalid.end())
|
||||
attribGroup = &(*bufferGroupItr);
|
||||
}
|
||||
// create new group if none found
|
||||
if (attribGroup == nullptr)
|
||||
{
|
||||
if (LatteFetchShader::isValidBufferIndex(bufferIndex))
|
||||
attribGroup = &parsedFetchShader->bufferGroups.emplace_back();
|
||||
else
|
||||
attribGroup = &parsedFetchShader->bufferGroupsInvalid.emplace_back();
|
||||
|
||||
attribGroup->attributeBufferIndex = bufferIndex;
|
||||
attribGroup->minOffset = offset;
|
||||
attribGroup->maxOffset = offset;
|
||||
}
|
||||
// add attribute
|
||||
sint32 groupAttribIndex = attribGroup->attribCount;
|
||||
if (attribGroup->attribCount < (groupAttribIndex + 1))
|
||||
{
|
||||
attribGroup->attribCount = (groupAttribIndex + 1);
|
||||
attribGroup->attrib = (LatteParsedFetchShaderAttribute_t*)realloc(attribGroup->attrib, sizeof(LatteParsedFetchShaderAttribute_t) * attribGroup->attribCount);
|
||||
}
|
||||
attribGroup->attrib[groupAttribIndex].semanticId = semanticId;
|
||||
attribGroup->attrib[groupAttribIndex].format = (uint8)dataFormat;
|
||||
attribGroup->attrib[groupAttribIndex].fetchType = fetchType;
|
||||
attribGroup->attrib[groupAttribIndex].nfa = (uint8)nfa;
|
||||
attribGroup->attrib[groupAttribIndex].isSigned = isSigned;
|
||||
attribGroup->attrib[groupAttribIndex].offset = offset;
|
||||
attribGroup->attrib[groupAttribIndex].ds[0] = (uint8)dsx;
|
||||
attribGroup->attrib[groupAttribIndex].ds[1] = (uint8)dsy;
|
||||
attribGroup->attrib[groupAttribIndex].ds[2] = (uint8)dsz;
|
||||
attribGroup->attrib[groupAttribIndex].ds[3] = (uint8)dsw;
|
||||
attribGroup->attrib[groupAttribIndex].attributeBufferIndex = bufferIndex;
|
||||
attribGroup->attrib[groupAttribIndex].endianSwap = endianSwap;
|
||||
attribGroup->minOffset = (std::min)(attribGroup->minOffset, offset);
|
||||
attribGroup->maxOffset = (std::max)(attribGroup->maxOffset, offset);
|
||||
// get alu divisor
|
||||
if (srcSelX == LatteClauseInstruction_VTX::SRC_SEL::SEL_X)
|
||||
{
|
||||
cemu_assert_debug(fetchType != LatteConst::VertexFetchType2::INSTANCE_DATA); // aluDivisor 0 in combination with instanced data is not allowed?
|
||||
attribGroup->attrib[groupAttribIndex].aluDivisor = -1;
|
||||
}
|
||||
else if (srcSelX == LatteClauseInstruction_VTX::SRC_SEL::SEL_W)
|
||||
{
|
||||
cemu_assert_debug(fetchType == LatteConst::VertexFetchType2::INSTANCE_DATA); // using constant divisor 1 with per-vertex data seems strange? (divisor is instance-only)
|
||||
// aluDivisor is constant 1
|
||||
attribGroup->attrib[groupAttribIndex].aluDivisor = 1;
|
||||
}
|
||||
else if (srcSelX == LatteClauseInstruction_VTX::SRC_SEL::SEL_Y)
|
||||
{
|
||||
// use alu divisor 1
|
||||
attribGroup->attrib[groupAttribIndex].aluDivisor = (sint32)contextRegister[mmVGT_INSTANCE_STEP_RATE_0 + 0];
|
||||
cemu_assert_debug(attribGroup->attrib[groupAttribIndex].aluDivisor > 0);
|
||||
}
|
||||
else if (srcSelX == LatteClauseInstruction_VTX::SRC_SEL::SEL_Z)
|
||||
{
|
||||
// use alu divisor 2
|
||||
attribGroup->attrib[groupAttribIndex].aluDivisor = (sint32)contextRegister[mmVGT_INSTANCE_STEP_RATE_0 + 1];
|
||||
cemu_assert_debug(attribGroup->attrib[groupAttribIndex].aluDivisor > 0);
|
||||
}
|
||||
}
|
||||
|
||||
void _fetchShaderDecompiler_parseVTXClause(LatteFetchShader* parsedFetchShader, uint32* contextRegister, std::span<uint8> clauseCode, size_t numInstructions)
|
||||
{
|
||||
const LatteClauseInstruction_VTX* instr = (LatteClauseInstruction_VTX*)clauseCode.data();
|
||||
const LatteClauseInstruction_VTX* end = instr + numInstructions;
|
||||
while (instr < end)
|
||||
{
|
||||
if (instr->getField_VTX_INST() == LatteClauseInstruction_VTX::VTX_INST::_VTX_INST_SEMANTIC)
|
||||
{
|
||||
_fetchShaderDecompiler_parseInstruction_VTX_SEMANTIC(parsedFetchShader, contextRegister, instr);
|
||||
}
|
||||
else
|
||||
{
|
||||
assert_dbg();
|
||||
}
|
||||
instr++;
|
||||
}
|
||||
}
|
||||
|
||||
void _fetchShaderDecompiler_parseCF(LatteFetchShader* parsedFetchShader, uint32* contextRegister, std::span<uint8> programCode)
|
||||
{
|
||||
size_t maxCountCFInstructions = programCode.size_bytes() / sizeof(LatteCFInstruction);
|
||||
const LatteCFInstruction* cfInstruction = (LatteCFInstruction*)programCode.data();
|
||||
const LatteCFInstruction* end = cfInstruction + maxCountCFInstructions;
|
||||
while (cfInstruction < end)
|
||||
{
|
||||
if (cfInstruction->getField_Opcode() == LatteCFInstruction::INST_VTX_TC)
|
||||
{
|
||||
auto vtxInstruction = cfInstruction->getParserIfOpcodeMatch<LatteCFInstruction_DEFAULT>();
|
||||
cemu_assert_debug(vtxInstruction->getField_COND() == LatteCFInstruction::CF_COND::CF_COND_ACTIVE);
|
||||
_fetchShaderDecompiler_parseVTXClause(parsedFetchShader, contextRegister, vtxInstruction->getClauseCode(programCode), vtxInstruction->getField_COUNT());
|
||||
}
|
||||
else if (cfInstruction->getField_Opcode() == LatteCFInstruction::INST_RETURN)
|
||||
{
|
||||
cemu_assert_debug(!cfInstruction->getField_END_OF_PROGRAM());
|
||||
return;
|
||||
}
|
||||
else
|
||||
{
|
||||
cemu_assert_debug(false); // unhandled / unexpected CF instruction
|
||||
}
|
||||
if (cfInstruction->getField_END_OF_PROGRAM())
|
||||
{
|
||||
cemu_assert_debug(false); // unusual for fetch shader? They should end with a return instruction
|
||||
break;
|
||||
}
|
||||
cfInstruction++;
|
||||
}
|
||||
cemu_assert_debug(false); // program must be terminated with an instruction that has EOP set?
|
||||
}
|
||||
|
||||
// parse fetch shader and create LatteFetchShader object
|
||||
// also registers the fs in the cache (s_fetchShaderByHash)
|
||||
// can be assumed to be thread-safe, if called simultaneously on the same fetch shader only one shader will become registered. The others will be destroyed
|
||||
LatteFetchShader* LatteShaderRecompiler_createFetchShader(LatteFetchShader::CacheHash fsHash, uint32* contextRegister, uint32* fsProgramCode, uint32 fsProgramSize)
|
||||
{
|
||||
LatteFetchShader* newFetchShader = new LatteFetchShader();
|
||||
newFetchShader->m_cacheHash = fsHash;
|
||||
if( (fsProgramSize&0xF) != 0 )
|
||||
debugBreakpoint();
|
||||
uint32 index = 0;
|
||||
|
||||
// if the first instruction is a CF instruction then parse shader properly
|
||||
// otherwise fall back to our broken legacy method (where we assumed fetch shaders had no CF program)
|
||||
// this workaround is required to make sure old shader caches dont break
|
||||
|
||||
// from old fetch shader gen (CF part missing):
|
||||
// {0x0000a001, 0x27961000, 0x00020000, 0x00000000}
|
||||
// {0x0000a001, 0x2c151002, 0x00020000, 0x00000000, 0x0000a001, 0x068d1000, 0x0000000c, ...}
|
||||
// {0x0000a001, 0x2c151000, 0x00020000, 0x00000000}
|
||||
// {0x0300aa21, 0x28cd1006, 0x00000000, 0x00000000, 0x0300ab21, 0x28cd1007, 0x00000000, ...}
|
||||
|
||||
// shaders shipped with games (e.g. BotW):
|
||||
// {0x00000002, 0x01800400, 0x00000000, 0x8a000000, 0x1c00a001, 0x280d1000, 0x00090000, ...}
|
||||
// {0x00000002, 0x01800000, 0x00000000, 0x8a000000, 0x1c00a001, 0x27961000, 0x000a0000, ...}
|
||||
// {0x00000002, 0x01800c00, 0x00000000, 0x8a000000, 0x2c00a001, 0x2c151000, 0x000a0000, ...} // size 0x50
|
||||
// {0x00000002, 0x01801000, 0x00000000, 0x8a000000, 0x1c00a001, 0x280d1000, 0x00090000, ...} // size 0x60
|
||||
// {0x00000002, 0x01801c00, 0x00000000, 0x8a000000, 0x1c00a001, 0x280d1000, 0x00090000, ...} // size 0x90
|
||||
|
||||
// our new implementation:
|
||||
// {0x00000002, 0x01800400, 0x00000000, 0x8a000000, 0x0000a001, 0x2c151000, 0x00020000, ...}
|
||||
|
||||
// for ALU instructions everything except the 01 is dynamic
|
||||
newFetchShader->bufferGroups.reserve(16);
|
||||
if (fsProgramSize == 0)
|
||||
{
|
||||
// empty fetch shader, seen in Minecraft
|
||||
// these only make sense when vertex shader does not call FS?
|
||||
LatteShader_calculateFSKey(newFetchShader);
|
||||
newFetchShader->CalculateFetchShaderVkHash();
|
||||
return newFetchShader;
|
||||
}
|
||||
|
||||
if ((fsProgramCode[0] & 1) == 0 && fsProgramCode[0] <= 0x30 && (fsProgramCode[1]&~((3 << 10)| (1 << 19))) == 0x01800000)
|
||||
{
|
||||
// very likely a CF instruction
|
||||
_fetchShaderDecompiler_parseCF(newFetchShader, contextRegister, { (uint8*)fsProgramCode, fsProgramSize });
|
||||
}
|
||||
else
|
||||
{
|
||||
while (index < (fsProgramSize / 4))
|
||||
{
|
||||
uint32 dword0 = fsProgramCode[index];
|
||||
uint32 opcode = dword0 & 0x1F;
|
||||
index++;
|
||||
if (opcode == VTX_INST_MEM)
|
||||
{
|
||||
// this might be the clause initialization instruction? (Seems to be the first instruction always)
|
||||
// todo - upon further investigation, it seems like fetch shaders also start with a CF program. Our implementation doesnt emit one right now
|
||||
uint32 opcode2 = (dword0 >> 8) & 7;
|
||||
|
||||
index += 3;
|
||||
}
|
||||
else if (opcode == VTX_INST_SEMANTIC)
|
||||
{
|
||||
_fetchShaderDecompiler_parseInstruction_VTX_SEMANTIC(newFetchShader, contextRegister, (const LatteClauseInstruction_VTX*)(fsProgramCode + index - 1));
|
||||
index += 3;
|
||||
}
|
||||
}
|
||||
}
|
||||
newFetchShader->bufferGroups.shrink_to_fit();
|
||||
// calculate group information
|
||||
// VBO offsets and stride
|
||||
uint32 vboOffset = 0;
|
||||
for (auto& bufferGroup : newFetchShader->bufferGroups)
|
||||
{
|
||||
for(sint32 i=0; i< bufferGroup.attribCount; i++)
|
||||
{
|
||||
uint32 attribSize = LatteShaderRecompiler_getAttributeSize(bufferGroup.attrib+i);
|
||||
uint32 attribAlignment = LatteShaderRecompiler_getAttributeAlignment(bufferGroup.attrib+i);
|
||||
// fix alignment
|
||||
vboOffset = (vboOffset+attribAlignment-1)&~(attribAlignment-1);
|
||||
vboOffset += attribSize;
|
||||
// index type
|
||||
if(bufferGroup.attrib[i].fetchType == LatteConst::VERTEX_DATA)
|
||||
bufferGroup.hasVtxIndexAccess = true;
|
||||
else if (bufferGroup.attrib[i].fetchType == LatteConst::INSTANCE_DATA)
|
||||
bufferGroup.hasInstanceIndexAccess = true;
|
||||
}
|
||||
// fix alignment of whole vertex
|
||||
if(bufferGroup.attribCount > 0 )
|
||||
{
|
||||
uint32 attribAlignment = LatteShaderRecompiler_getAttributeAlignment(bufferGroup.attrib+0);
|
||||
vboOffset = (vboOffset+attribAlignment-1)&~(attribAlignment-1);
|
||||
}
|
||||
bufferGroup.vboStride = vboOffset;
|
||||
}
|
||||
LatteShader_calculateFSKey(newFetchShader);
|
||||
newFetchShader->CalculateFetchShaderVkHash();
|
||||
|
||||
// register in cache
|
||||
// its possible that during multi-threaded shader cache loading, two identical (same hash) fetch shaders get created simultaneously
|
||||
// we catch and handle this case here. RegisterInCache() is atomic and if another fetch shader is already registered, we abandon the local instance
|
||||
LatteFetchShader* registeredFS = newFetchShader->RegisterInCache(fsHash);
|
||||
if (registeredFS)
|
||||
{
|
||||
delete newFetchShader;
|
||||
newFetchShader = registeredFS;
|
||||
}
|
||||
else
|
||||
{
|
||||
newFetchShader->m_isRegistered = true;
|
||||
}
|
||||
|
||||
|
||||
return newFetchShader;
|
||||
}
|
||||
|
||||
LatteFetchShader::~LatteFetchShader()
|
||||
{
|
||||
UnregisterInCache();
|
||||
}
|
||||
|
||||
struct FetchShaderLookupInfo
|
||||
{
|
||||
LatteFetchShader* fetchShader;
|
||||
uint32 programSize;
|
||||
uint32 lastFrameAccessed;
|
||||
};
|
||||
|
||||
LookupTableL3<8, 8, 8, FetchShaderLookupInfo*> g_fetchShaderLookupCache;
|
||||
|
||||
LatteFetchShader::CacheHash LatteFetchShader::CalculateCacheHash(void* programCode, uint32 programSize)
|
||||
{
|
||||
uint32* programCodeU32 = (uint32*)programCode;
|
||||
uint64 progHash1 = 0;
|
||||
uint64 progHash2 = 0;
|
||||
for (uint32 i = 0; i < programSize / 4; i++)
|
||||
{
|
||||
uint32 temp = programCodeU32[i];
|
||||
progHash1 += (uint64)temp;
|
||||
progHash2 ^= (uint64)temp;
|
||||
progHash1 = (progHash1 << 3) | (progHash1 >> 61);
|
||||
progHash2 = (progHash2 >> 7) | (progHash2 << 57);
|
||||
}
|
||||
|
||||
// todo - we should incorporate the value of VGT_INSTANCE_STEP_RATE_0/1 into the hash since it affects the generated LatteFetchShader object
|
||||
// However, this would break compatibility with shader caches and gfx packs due to altering the shader base hashes
|
||||
|
||||
return progHash1 + progHash2;
|
||||
}
|
||||
|
||||
LatteFetchShader* LatteFetchShader::FindInCacheByHash(LatteFetchShader::CacheHash fsHash)
|
||||
{
|
||||
// does not hold s_fetchShaderCache for better performance. Be careful not to call this while another thread invokes RegisterInCache()
|
||||
auto itr = s_fetchShaderByHash.find(fsHash);
|
||||
if (itr == s_fetchShaderByHash.end())
|
||||
return nullptr;
|
||||
return itr->second;
|
||||
}
|
||||
|
||||
void* _getFSProgramPtr()
|
||||
{
|
||||
return memory_getPointerFromPhysicalOffset(LatteGPUState.contextRegister[mmSQ_PGM_START_FS + 0] << 8);
|
||||
}
|
||||
|
||||
uint32 _getFSProgramSize()
|
||||
{
|
||||
return LatteGPUState.contextRegister[mmSQ_PGM_START_FS + 1] << 3;
|
||||
}
|
||||
|
||||
LatteFetchShader* LatteFetchShader::FindByGPUState()
|
||||
{
|
||||
// retrieve fetch shader that matches the currently set GPU context registers
|
||||
uint32 fsPhysAddr24 = LatteGPUState.contextRegister[mmSQ_PGM_START_FS + 0];
|
||||
cemu_assert_debug(fsPhysAddr24 < 0x1000000); // should only contain the upper 24 bit of the address in the lower 24 bit of the register
|
||||
|
||||
FetchShaderLookupInfo* lookupInfo = g_fetchShaderLookupCache.lookup(fsPhysAddr24);
|
||||
if (lookupInfo)
|
||||
{
|
||||
// return fetch shader if still the same
|
||||
uint32 fsSize = _getFSProgramSize();
|
||||
uint32 framesSinceLastAccess = LatteGPUState.frameCounter - lookupInfo->lastFrameAccessed;
|
||||
if (lookupInfo->programSize == fsSize && framesSinceLastAccess == 0)
|
||||
{
|
||||
lookupInfo->lastFrameAccessed = LatteGPUState.frameCounter;
|
||||
return lookupInfo->fetchShader;
|
||||
}
|
||||
// update lookup info
|
||||
CacheHash fsHash = CalculateCacheHash(_getFSProgramPtr(), _getFSProgramSize());
|
||||
LatteFetchShader* fetchShader = FindInCacheByHash(fsHash);
|
||||
if (!fetchShader)
|
||||
{
|
||||
fetchShader = LatteShaderRecompiler_createFetchShader(fsHash, LatteGPUState.contextNew.GetRawView(), (uint32*)_getFSProgramPtr(), _getFSProgramSize());
|
||||
cemu_assert(fetchShader);
|
||||
}
|
||||
lookupInfo->fetchShader = fetchShader;
|
||||
lookupInfo->programSize = fsSize;
|
||||
lookupInfo->lastFrameAccessed = LatteGPUState.frameCounter;
|
||||
return fetchShader;
|
||||
}
|
||||
else
|
||||
{
|
||||
// try to find fetch shader by data hash
|
||||
CacheHash fsHash = CalculateCacheHash(_getFSProgramPtr(), _getFSProgramSize());
|
||||
LatteFetchShader* fetchShader = FindInCacheByHash(fsHash);
|
||||
if (!fetchShader)
|
||||
{
|
||||
fetchShader = LatteShaderRecompiler_createFetchShader(fsHash, LatteGPUState.contextNew.GetRawView(), (uint32*)_getFSProgramPtr(), _getFSProgramSize());
|
||||
cemu_assert(fetchShader);
|
||||
}
|
||||
// create new lookup entry
|
||||
lookupInfo = new FetchShaderLookupInfo();
|
||||
lookupInfo->fetchShader = fetchShader;
|
||||
lookupInfo->programSize = _getFSProgramSize();
|
||||
lookupInfo->lastFrameAccessed = LatteGPUState.frameCounter;
|
||||
g_fetchShaderLookupCache.store(fsPhysAddr24, lookupInfo);
|
||||
#ifndef PUBLIC_RELEASE
|
||||
cemu_assert_debug(g_fetchShaderLookupCache.lookup(fsPhysAddr24) == lookupInfo);
|
||||
#endif
|
||||
}
|
||||
return lookupInfo->fetchShader;
|
||||
}
|
||||
|
||||
FSpinlock s_spinlockFetchShaderCache;
|
||||
|
||||
LatteFetchShader* LatteFetchShader::RegisterInCache(CacheHash fsHash)
|
||||
{
|
||||
s_spinlockFetchShaderCache.acquire();
|
||||
auto itr = s_fetchShaderByHash.find(fsHash);
|
||||
if (itr != s_fetchShaderByHash.end())
|
||||
{
|
||||
LatteFetchShader* fs = itr->second;
|
||||
s_spinlockFetchShaderCache.release();
|
||||
return fs;
|
||||
}
|
||||
s_fetchShaderByHash.emplace(fsHash, this);
|
||||
s_spinlockFetchShaderCache.release();
|
||||
return nullptr;
|
||||
}
|
||||
|
||||
void LatteFetchShader::UnregisterInCache()
|
||||
{
|
||||
if (!m_isRegistered)
|
||||
return;
|
||||
s_spinlockFetchShaderCache.acquire();
|
||||
auto itr = s_fetchShaderByHash.find(m_cacheHash);
|
||||
cemu_assert(itr == s_fetchShaderByHash.end());
|
||||
s_fetchShaderByHash.erase(itr);
|
||||
s_spinlockFetchShaderCache.release();
|
||||
}
|
||||
|
||||
std::unordered_map<LatteFetchShader::CacheHash, LatteFetchShader*> LatteFetchShader::s_fetchShaderByHash;
|
||||
Loading…
Add table
Add a link
Reference in a new issue