void ShaderCache::HandleVSUIDChange(VertexShaderUid vs_uid)
{
s_last_vertex_shader_uid = vs_uid;
auto vs_iterator = s_vs_bytecode_cache.find(vs_uid);
if (vs_iterator != s_vs_bytecode_cache.end())
{
s_last_vertex_shader_bytecode = vs_iterator->second;
GFX_DEBUGGER_PAUSE_AT(NEXT_VERTEX_SHADER_CHANGE, true);
}
else
{
ShaderCode vs_code = GenerateVertexShaderCode(APIType::D3D, vs_uid.GetUidData());
ID3DBlob* vs_bytecode = nullptr;
if (!D3D::CompileVertexShader(vs_code.GetBuffer(), &vs_bytecode))
{
GFX_DEBUGGER_PAUSE_AT(NEXT_ERROR, true);
return;
}
s_last_vertex_shader_bytecode = InsertByteCode(vs_uid, &s_vs_bytecode_cache, vs_bytecode);
s_vs_disk_cache.Append(vs_uid, reinterpret_cast<u8*>(vs_bytecode->GetBufferPointer()),
static_cast<u32>(vs_bytecode->GetBufferSize()));
SETSTAT(stats.numVertexShadersAlive, static_cast<int>(s_vs_bytecode_cache.size()));
INCSTAT(stats.numVertexShadersCreated);
}
}
void VertexShaderCache::CompileVShader(const VertexShaderUid& uid, bool ongputhread)
{
s_vshaders_lock.lock();
VSCacheEntry* entry = &s_vshaders->GetOrAdd(uid);
s_vshaders_lock.unlock();
if (ongputhread)
{
s_last_entry = entry;
}
// Compile only when we have a new instance
if (entry->initialized.test_and_set())
{
return;
}
ShaderCompilerWorkUnit *wunit = s_compiler->NewUnit(VERTEXSHADERGEN_BUFFERSIZE);
wunit->GenerateCodeHandler = [uid](ShaderCompilerWorkUnit* wunit)
{
ShaderCode code;
code.SetBuffer(wunit->code.data());
GenerateVertexShaderCodeD3D11(code, uid.GetUidData());
wunit->codesize = (u32)code.BufferSize();
};
wunit->entrypoint = "main";
#if defined(_DEBUG) || defined(DEBUGFAST)
wunit->flags = D3DCOMPILE_DEBUG | D3DCOMPILE_SKIP_OPTIMIZATION;
#else
wunit->flags = D3DCOMPILE_SKIP_VALIDATION | D3DCOMPILE_OPTIMIZATION_LEVEL3 | D3DCOMPILE_ENABLE_BACKWARDS_COMPATIBILITY;
#endif
wunit->target = D3D::VertexShaderVersionString();
wunit->ResultHandler = [uid, entry](ShaderCompilerWorkUnit* wunit)
{
if (SUCCEEDED(wunit->cresult))
{
g_vs_disk_cache.Append(uid, (const u8*)wunit->shaderbytecode->GetBufferPointer(), (u32)wunit->shaderbytecode->GetBufferSize());
PushByteCode(D3DBlob(D3D::UniquePtr<ID3D10Blob>(wunit->shaderbytecode)), entry);
wunit->shaderbytecode = nullptr;
}
else
{
static int num_failures = 0;
std::string filename = StringFromFormat("%sbad_vs_%04i.txt", File::GetUserPath(D_DUMP_IDX).c_str(), num_failures++);
std::ofstream file;
OpenFStream(file, filename, std::ios_base::out);
file << ((const char*)wunit->code.data());
file.close();
PanicAlert("Failed to compile vertex shader!\nThis usually happens when trying to use Dolphin with an outdated GPU or integrated GPU like the Intel GMA series.\n\nIf you're sure this is Dolphin's error anyway, post the contents of %s along with this error message at the forums.\n\nDebug info (%s):\n%s",
filename,
D3D::VertexShaderVersionString(),
(char*)wunit->error->GetBufferPointer());
}
};
s_compiler->CompileShaderAsync(wunit);
}
bool VertexShaderCache::SetShader()
{
VertexShaderUid uid = GetVertexShaderUid();
if (last_entry)
{
if (uid == last_uid)
{
GFX_DEBUGGER_PAUSE_AT(NEXT_VERTEX_SHADER_CHANGE, true);
return (last_entry->shader != nullptr);
}
}
last_uid = uid;
VSCache::iterator iter = vshaders.find(uid);
if (iter != vshaders.end())
{
const VSCacheEntry& entry = iter->second;
last_entry = &entry;
GFX_DEBUGGER_PAUSE_AT(NEXT_VERTEX_SHADER_CHANGE, true);
return (entry.shader != nullptr);
}
ShaderCode code = GenerateVertexShaderCode(API_D3D, uid.GetUidData());
D3DBlob* pbytecode = nullptr;
D3D::CompileVertexShader(code.GetBuffer(), &pbytecode);
if (pbytecode == nullptr)
{
GFX_DEBUGGER_PAUSE_AT(NEXT_ERROR, true);
return false;
}
g_vs_disk_cache.Append(uid, pbytecode->Data(), pbytecode->Size());
bool success = InsertByteCode(uid, pbytecode);
pbytecode->Release();
GFX_DEBUGGER_PAUSE_AT(NEXT_VERTEX_SHADER_CHANGE, true);
return success;
}
void ShaderCache::Init()
{
s_compiler = &HLSLAsyncCompiler::getInstance();
s_shaders_lock.unlock();
s_pass_entry.m_compiled = true;
s_pass_entry.m_initialized.test_and_set();
// This class intentionally shares its shader cache files with DX11, as the shaders are (right now) identical.
// Reduces unnecessary compilation when switching between APIs.
s_last_domain_shader_bytecode = &s_pass_entry;
s_last_hull_shader_bytecode = &s_pass_entry;
s_last_geometry_shader_bytecode = &s_pass_entry;
s_last_pixel_shader_bytecode = nullptr;
s_last_vertex_shader_bytecode = nullptr;
s_last_geometry_shader_uid = {};
s_last_pixel_shader_uid = {};
s_last_vertex_shader_uid = {};
s_last_tessellation_shader_uid = {};
s_last_cpu_geometry_shader_uid = {};
s_last_cpu_pixel_shader_uid = {};
s_last_cpu_vertex_shader_uid = {};
s_last_cpu_tessellation_shader_uid = {};
// Ensure shader cache directory exists..
std::string shader_cache_path = File::GetUserPath(D_SHADERCACHE_IDX);
if (!File::Exists(shader_cache_path))
File::CreateDir(File::GetUserPath(D_SHADERCACHE_IDX));
std::string title_unique_id = SConfig::GetInstance().m_strGameID;
std::string ds_cache_filename = StringFromFormat("%sIDX11-%s-ds.cache", shader_cache_path.c_str(), title_unique_id.c_str());
std::string hs_cache_filename = StringFromFormat("%sIDX11-%s-hs.cache", shader_cache_path.c_str(), title_unique_id.c_str());
std::string gs_cache_filename = StringFromFormat("%sIDX11-%s-gs.cache", shader_cache_path.c_str(), title_unique_id.c_str());
std::string ps_cache_filename = StringFromFormat("%sIDX11-%s-ps.cache", shader_cache_path.c_str(), title_unique_id.c_str());
std::string vs_cache_filename = StringFromFormat("%sIDX11-%s-vs.cache", shader_cache_path.c_str(), title_unique_id.c_str());
pKey_t gameid = (pKey_t)GetMurmurHash3(reinterpret_cast<const u8*>(SConfig::GetInstance().m_strGameID.data()), (u32)SConfig::GetInstance().m_strGameID.size(), 0);
vs_bytecode_cache = VsBytecodeCache::Create(
gameid,
VERTEXSHADERGEN_UID_VERSION,
"Ishiiruka.vs",
StringFromFormat("%s.vs", title_unique_id.c_str())
);
ps_bytecode_cache = PsBytecodeCache::Create(
gameid,
PIXELSHADERGEN_UID_VERSION,
"Ishiiruka.ps",
StringFromFormat("%s.ps", title_unique_id.c_str())
);
gs_bytecode_cache = GsBytecodeCache::Create(
gameid,
GEOMETRYSHADERGEN_UID_VERSION,
"Ishiiruka.gs",
StringFromFormat("%s.gs", title_unique_id.c_str())
);
ts_bytecode_cache = TsBytecodeCache::Create(
gameid,
TESSELLATIONSHADERGEN_UID_VERSION,
"Ishiiruka.ts",
StringFromFormat("%s.ts", title_unique_id.c_str())
);
DShaderCacheInserter ds_inserter;
s_ds_disk_cache.OpenAndRead(ds_cache_filename, ds_inserter);
HShaderCacheInserter hs_inserter;
s_hs_disk_cache.OpenAndRead(hs_cache_filename, hs_inserter);
ShaderCacheInserter<GeometryShaderUid, GsBytecodeCache, &gs_bytecode_cache> gs_inserter;
s_gs_disk_cache.OpenAndRead(gs_cache_filename, gs_inserter);
ShaderCacheInserter<PixelShaderUid, PsBytecodeCache, &ps_bytecode_cache> ps_inserter;
s_ps_disk_cache.OpenAndRead(ps_cache_filename, ps_inserter);
ShaderCacheInserter<VertexShaderUid, VsBytecodeCache, &vs_bytecode_cache> vs_inserter;
s_vs_disk_cache.OpenAndRead(vs_cache_filename, vs_inserter);
// Clear out disk cache when debugging shaders to ensure stale ones don't stick around..
SETSTAT(stats.numGeometryShadersAlive, static_cast<int>(gs_bytecode_cache->size()));
SETSTAT(stats.numGeometryShadersCreated, 0);
SETSTAT(stats.numPixelShadersAlive, static_cast<int>(ps_bytecode_cache->size()));
SETSTAT(stats.numPixelShadersCreated, 0);
SETSTAT(stats.numVertexShadersAlive, static_cast<int>(vs_bytecode_cache->size()));
SETSTAT(stats.numVertexShadersCreated, 0);
if (g_ActiveConfig.bCompileShaderOnStartup)
{
size_t shader_count = 0;
ps_bytecode_cache->ForEachMostUsedByCategory(gameid,
[&](const PixelShaderUid& it, size_t total)
{
PixelShaderUid item = it;
item.ClearHASH();
item.CalculateUIDHash();
HandlePSUIDChange(item, true);
shader_count++;
if ((shader_count & 7) == 0)
{
Host_UpdateTitle(StringFromFormat("Compiling Pixel Shaders %i %% (%i/%i)", (shader_count * 100) / total, shader_count, total));
s_compiler->WaitForFinish();
}
},
[](ByteCodeCacheEntry& entry)
{
return !entry.m_shader_bytecode.pShaderBytecode;
}
, true);
shader_count = 0;
vs_bytecode_cache->ForEachMostUsedByCategory(gameid,
[&](const VertexShaderUid& it, size_t total)
{
VertexShaderUid item = it;
item.ClearHASH();
item.CalculateUIDHash();
HandleVSUIDChange(item, true);
shader_count++;
if ((shader_count & 31) == 0)
{
Host_UpdateTitle(StringFromFormat("Compiling Vertex Shaders %i %% (%i/%i)", (shader_count * 100) / total, shader_count, total));
s_compiler->WaitForFinish();
}
},
[](ByteCodeCacheEntry& entry)
{
return !entry.m_shader_bytecode.pShaderBytecode;
}
, true);
shader_count = 0;
gs_bytecode_cache->ForEachMostUsedByCategory(gameid,
[&](const GeometryShaderUid& it, size_t total)
{
GeometryShaderUid item = it;
item.ClearHASH();
item.CalculateUIDHash();
HandleGSUIDChange(item, true);
shader_count++;
if ((shader_count & 7) == 0)
{
Host_UpdateTitle(StringFromFormat("Compiling Geometry Shaders %i %% (%i/%i)", (shader_count * 100) / total, shader_count, total));
s_compiler->WaitForFinish();
}
},
[](ByteCodeCacheEntry& entry)
{
return !entry.m_shader_bytecode.pShaderBytecode;
}
, true);
shader_count = 0;
ts_bytecode_cache->ForEachMostUsedByCategory(gameid,
[&](const TessellationShaderUid& it, size_t total)
{
TessellationShaderUid item = it;
item.ClearHASH();
item.CalculateUIDHash();
HandleTSUIDChange(item, true);
shader_count++;
if ((shader_count & 31) == 0)
{
Host_UpdateTitle(StringFromFormat("Compiling Tessellation Shaders %i %% (%i/%i)", (shader_count * 100) / total, shader_count, total));
s_compiler->WaitForFinish();
}
},
[](std::pair<ByteCodeCacheEntry, ByteCodeCacheEntry>& entry)
{
return !entry.first.m_shader_bytecode.pShaderBytecode;
}
, true);
s_compiler->WaitForFinish();
}
}
})hlsl";
void VertexShaderCache::Init()
{
s_vshaders_lock.unlock();
s_compiler = &HLSLAsyncCompiler::getInstance();
const D3D11_INPUT_ELEMENT_DESC simpleelems[3] =
{
{ "POSITION", 0, DXGI_FORMAT_R32G32B32_FLOAT, 0, 0, D3D11_INPUT_PER_VERTEX_DATA, 0 },
{ "TEXCOORD", 0, DXGI_FORMAT_R32G32B32_FLOAT, 0, 12, D3D11_INPUT_PER_VERTEX_DATA, 0 },
{ "TEXCOORD", 1, DXGI_FORMAT_R32G32B32_FLOAT, 0, 24, D3D11_INPUT_PER_VERTEX_DATA, 0 }
};
const D3D11_INPUT_ELEMENT_DESC clearelems[2] =
{
{ "POSITION", 0, DXGI_FORMAT_R32G32B32_FLOAT, 0, 0, D3D11_INPUT_PER_VERTEX_DATA, 0 },
{ "COLOR", 0, DXGI_FORMAT_R8G8B8A8_UNORM, 0, 12, D3D11_INPUT_PER_VERTEX_DATA, 0 },
};
bool use_partial_buffer_update = D3D::SupportPartialContantBufferUpdate();
u32 cbsize = VertexShaderManager::ConstantBufferSize * sizeof(float) * (use_partial_buffer_update ? 1024 : 1); // is always multiple of 16
vscbuf = new D3D::ConstantStreamBuffer(cbsize);
ID3D11Buffer* buf = vscbuf->GetBuffer();
CHECK(buf != nullptr, "Create vertex shader constant buffer (size=%u)", cbsize);
D3D::SetDebugObjectName(buf, "vertex shader constant buffer used to emulate the GX pipeline");
D3DBlob blob;
D3D::CompileShader(D3D::ShaderType::Vertex, simple_shader_code, blob);
D3D::device->CreateInputLayout(simpleelems, 3, blob.Data(), blob.Size(), D3D::ToAddr(s_simple_layout));
s_simple_vertex_shader = D3D::CreateVertexShaderFromByteCode(blob);
if (s_simple_layout == nullptr ||
s_simple_vertex_shader == nullptr)
PanicAlert("Failed to create simple vertex shader or input layout at %s %d\n", __FILE__, __LINE__);
D3D::SetDebugObjectName(s_simple_layout.get(), "simple input layout");
D3D::SetDebugObjectName(s_simple_vertex_shader.get(), "simple vertex shader");
D3D::CompileShader(D3D::ShaderType::Vertex, clear_shader_code, blob);
D3D::device->CreateInputLayout(clearelems, 2, blob.Data(), blob.Size(), D3D::ToAddr(s_clear_layout));
s_clear_vertex_shader = D3D::CreateVertexShaderFromByteCode(blob);
if (s_clear_layout == nullptr ||
s_clear_vertex_shader == nullptr)
PanicAlert("Failed to create clear vertex shader or input layout at %s %d\n", __FILE__, __LINE__);
D3D::SetDebugObjectName(s_clear_vertex_shader.get(), "clear vertex shader");
D3D::SetDebugObjectName(s_clear_layout.get(), "clear input layout");
Clear();
if (!File::Exists(File::GetUserPath(D_SHADERCACHE_IDX)))
File::CreateDir(File::GetUserPath(D_SHADERCACHE_IDX).c_str());
SETSTAT(stats.numVertexShadersCreated, 0);
SETSTAT(stats.numVertexShadersAlive, 0);
pKey_t gameid = (pKey_t)GetMurmurHash3(reinterpret_cast<const u8*>(SConfig::GetInstance().m_strGameID.data()), (u32)SConfig::GetInstance().m_strGameID.size(), 0);
s_vshaders = VSCache::Create(
gameid,
VERTEXSHADERGEN_UID_VERSION,
"Ishiiruka.vs",
StringFromFormat("%s.vs", SConfig::GetInstance().m_strGameID.c_str())
);
std::string cache_filename = StringFromFormat("%sIDX11-%s-vs.cache", File::GetUserPath(D_SHADERCACHE_IDX).c_str(),
SConfig::GetInstance().m_strGameID.c_str());
VertexShaderCacheInserter inserter;
g_vs_disk_cache.OpenAndRead(cache_filename, inserter);
if (g_ActiveConfig.bCompileShaderOnStartup)
{
size_t shader_count = 0;
s_vshaders->ForEachMostUsedByCategory(gameid,
[&](const VertexShaderUid& it, size_t total)
{
VertexShaderUid item = it;
item.ClearHASH();
item.CalculateUIDHash();
CompileVShader(item, true);
shader_count++;
if ((shader_count & 7) == 0)
{
Host_UpdateTitle(StringFromFormat("Compiling Vertex Shaders %i %% (%i/%i)", (shader_count * 100) / total, shader_count, total));
s_compiler->WaitForFinish();
}
},
[](VSCacheEntry& entry)
{
return !entry.shader;
}
, true);
s_compiler->WaitForFinish();
}
s_last_entry = nullptr;
}
bool VertexShaderCache::SetShader(D3DVertexFormat* vertex_format)
{
if (g_ActiveConfig.bDisableSpecializedShaders)
return SetUberShader(vertex_format);
VertexShaderUid uid = GetVertexShaderUid();
if (last_entry && uid == last_uid)
{
if (last_entry->pending)
return SetUberShader(vertex_format);
if (!last_entry->shader)
return false;
vertex_format->SetInputLayout(last_entry->bytecode);
D3D::stateman->SetVertexShader(last_entry->shader);
return true;
}
auto iter = vshaders.find(uid);
if (iter != vshaders.end())
{
const VSCacheEntry& entry = iter->second;
if (entry.pending)
return SetUberShader(vertex_format);
last_uid = uid;
last_entry = &entry;
GFX_DEBUGGER_PAUSE_AT(NEXT_VERTEX_SHADER_CHANGE, true);
if (!last_entry->shader)
return false;
vertex_format->SetInputLayout(last_entry->bytecode);
D3D::stateman->SetVertexShader(last_entry->shader);
return true;
}
// Background compiling?
if (g_ActiveConfig.CanBackgroundCompileShaders())
{
// Create a pending entry
VSCacheEntry entry;
entry.pending = true;
vshaders[uid] = entry;
// Queue normal shader compiling and use ubershader
g_async_compiler->QueueWorkItem(
g_async_compiler->CreateWorkItem<VertexShaderCompilerWorkItem>(uid));
return SetUberShader(vertex_format);
}
// Need to compile a new shader
D3DBlob* bytecode = nullptr;
ShaderCode code =
GenerateVertexShaderCode(APIType::D3D, ShaderHostConfig::GetCurrent(), uid.GetUidData());
D3D::CompileVertexShader(code.GetBuffer(), &bytecode);
if (!InsertByteCode(uid, bytecode))
{
SAFE_RELEASE(bytecode);
return false;
}
g_vs_disk_cache.Append(uid, bytecode->Data(), bytecode->Size());
bytecode->Release();
return SetShader(vertex_format);
}
void VertexShaderCache::PrepareShader(u32 components, const XFMemory &xfr, const BPMemory &bpm, bool ongputhread)
{
VertexShaderUid uid;
GetVertexShaderUID(uid, components, xfr, bpm);
if (ongputhread)
{
Compiler->ProcCompilationResults();
#if defined(_DEBUG) || defined(DEBUGFAST)
if (g_ActiveConfig.bEnableShaderDebugging)
{
ShaderCode code;
GenerateVertexShaderCodeD3D9(code, uid.GetUidData());
}
#endif
if (last_entry)
{
if (uid == last_uid)
{
return;
}
}
last_uid = uid;
GFX_DEBUGGER_PAUSE_AT(NEXT_VERTEX_SHADER_CHANGE, true);
}
else
{
if (external_last_uid == uid)
{
return;
}
external_last_uid = uid;
}
vshaderslock.lock();
VSCacheEntry *entry = &vshaders[uid];
vshaderslock.unlock();
if (ongputhread)
{
last_entry = entry;
}
// Compile only when we have a new instance
if (entry->initialized.test_and_set())
{
return;
}
ShaderCompilerWorkUnit *wunit = Compiler->NewUnit(VERTEXSHADERGEN_BUFFERSIZE);
wunit->GenerateCodeHandler = [uid](ShaderCompilerWorkUnit* wunit)
{
ShaderCode code;
code.SetBuffer(wunit->code.data());
GenerateVertexShaderCodeD3D9(code, uid.GetUidData());
wunit->codesize = (u32)code.BufferSize();
};
wunit->entrypoint = "main";
wunit->flags = D3DCOMPILE_SKIP_VALIDATION | D3DCOMPILE_OPTIMIZATION_LEVEL3;
wunit->target = D3D::VertexShaderVersionString();
wunit->ResultHandler = [uid, entry](ShaderCompilerWorkUnit* wunit)
{
if (SUCCEEDED(wunit->cresult))
{
ID3DBlob* shaderBuffer = wunit->shaderbytecode;
const u8* bytecode = (const u8*)shaderBuffer->GetBufferPointer();
u32 bytecodelen = (u32)shaderBuffer->GetBufferSize();
g_vs_disk_cache.Append(uid, bytecode, bytecodelen);
PushByteCode(uid, bytecode, bytecodelen, entry);
#if defined(_DEBUG) || defined(DEBUGFAST)
if (g_ActiveConfig.bEnableShaderDebugging)
{
entry->code = wunit->code.data();
}
#endif
}
else
{
static int num_failures = 0;
std::string filename = StringFromFormat("%sbad_vs_%04i.txt", File::GetUserPath(D_DUMP_IDX).c_str(), num_failures++);
std::ofstream file;
OpenFStream(file, filename, std::ios_base::out);
file << ((const char*)wunit->code.data());
file << ((const char*)wunit->error->GetBufferPointer());
file.close();
PanicAlert("Failed to compile vertex shader!\nThis usually happens when trying to use Dolphin with an outdated GPU or integrated GPU like the Intel GMA series.\n\nIf you're sure this is Dolphin's error anyway, post the contents of %s along with this error message at the forums.\n\nDebug info (%s):\n%s",
filename,
D3D::VertexShaderVersionString(),
(char*)wunit->error->GetBufferPointer());
}
};
Compiler->CompileShaderAsync(wunit);
}
