bool PixelShaderCache::SetShader(DSTALPHA_MODE dstAlphaMode, u32 components)
{
PixelShaderUid uid;
GetPixelShaderUid(uid, dstAlphaMode, API_D3D, components);
if (g_ActiveConfig.bEnableShaderDebugging)
{
PixelShaderCode code;
GeneratePixelShaderCode(code, dstAlphaMode, API_D3D, components);
pixel_uid_checker.AddToIndexAndCheck(code, uid, "Pixel", "p");
}
// Check if the shader is already set
if (last_entry)
{
if (uid == last_uid)
{
GFX_DEBUGGER_PAUSE_AT(NEXT_PIXEL_SHADER_CHANGE,true);
return (last_entry->shader != nullptr);
}
}
last_uid = uid;
// Check if the shader is already in the cache
PSCache::iterator iter;
iter = PixelShaders.find(uid);
if (iter != PixelShaders.end())
{
const PSCacheEntry &entry = iter->second;
last_entry = &entry;
GFX_DEBUGGER_PAUSE_AT(NEXT_PIXEL_SHADER_CHANGE,true);
return (entry.shader != nullptr);
}
// Need to compile a new shader
PixelShaderCode code;
GeneratePixelShaderCode(code, dstAlphaMode, API_D3D, components);
D3DBlob* pbytecode;
if (!D3D::CompilePixelShader(code.GetBuffer(), &pbytecode))
{
GFX_DEBUGGER_PAUSE_AT(NEXT_ERROR, true);
return false;
}
// Insert the bytecode into the caches
g_ps_disk_cache.Append(uid, pbytecode->Data(), pbytecode->Size());
bool success = InsertByteCode(uid, pbytecode->Data(), pbytecode->Size());
pbytecode->Release();
if (g_ActiveConfig.bEnableShaderDebugging && success)
{
PixelShaders[uid].code = code.GetBuffer();
}
GFX_DEBUGGER_PAUSE_AT(NEXT_PIXEL_SHADER_CHANGE, true);
return success;
}
FRAGMENTSHADER* PixelShaderCache::SetShader(DSTALPHA_MODE dstAlphaMode, u32 components)
{
PIXELSHADERUID uid;
GetPixelShaderId(&uid, dstAlphaMode, components);
// Check if the shader is already set
if (last_entry)
{
if (uid == last_uid)
{
GFX_DEBUGGER_PAUSE_AT(NEXT_PIXEL_SHADER_CHANGE, true);
ValidatePixelShaderIDs(API_OPENGL, last_entry->safe_uid, last_entry->shader.strprog, dstAlphaMode, components);
return &last_entry->shader;
}
}
last_uid = uid;
PSCache::iterator iter = PixelShaders.find(uid);
if (iter != PixelShaders.end())
{
PSCacheEntry &entry = iter->second;
last_entry = &entry;
GFX_DEBUGGER_PAUSE_AT(NEXT_PIXEL_SHADER_CHANGE, true);
ValidatePixelShaderIDs(API_OPENGL, entry.safe_uid, entry.shader.strprog, dstAlphaMode, components);
return &last_entry->shader;
}
// Make an entry in the table
PSCacheEntry& newentry = PixelShaders[uid];
last_entry = &newentry;
const char *code = GeneratePixelShaderCode(dstAlphaMode, API_OPENGL, components);
if (g_ActiveConfig.bEnableShaderDebugging && code)
{
GetSafePixelShaderId(&newentry.safe_uid, dstAlphaMode, components);
newentry.shader.strprog = code;
}
#if defined(_DEBUG) || defined(DEBUGFAST)
if (g_ActiveConfig.iLog & CONF_SAVESHADERS && code) {
static int counter = 0;
char szTemp[MAX_PATH];
sprintf(szTemp, "%sps_%04i.txt", File::GetUserPath(D_DUMP_IDX).c_str(), counter++);
SaveData(szTemp, code);
}
#endif
if (!code || !CompilePixelShader(newentry.shader, code)) {
GFX_DEBUGGER_PAUSE_AT(NEXT_ERROR, true);
return NULL;
}
INCSTAT(stats.numPixelShadersCreated);
SETSTAT(stats.numPixelShadersAlive, PixelShaders.size());
GFX_DEBUGGER_PAUSE_AT(NEXT_PIXEL_SHADER_CHANGE, true);
return &last_entry->shader;
}
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 ShaderCache::HandlePSUIDChange(PixelShaderUid ps_uid, DSTALPHA_MODE ps_dst_alpha_mode)
{
s_last_pixel_shader_uid = ps_uid;
auto ps_iterator = s_ps_bytecode_cache.find(ps_uid);
if (ps_iterator != s_ps_bytecode_cache.end())
{
s_last_pixel_shader_bytecode = ps_iterator->second;
GFX_DEBUGGER_PAUSE_AT(NEXT_PIXEL_SHADER_CHANGE, true);
}
else
{
ShaderCode ps_code = GeneratePixelShaderCode(APIType::D3D, ps_uid.GetUidData());
ID3DBlob* ps_bytecode = nullptr;
if (!D3D::CompilePixelShader(ps_code.GetBuffer(), &ps_bytecode))
{
GFX_DEBUGGER_PAUSE_AT(NEXT_ERROR, true);
return;
}
s_last_pixel_shader_bytecode = InsertByteCode(ps_uid, &s_ps_bytecode_cache, ps_bytecode);
s_ps_disk_cache.Append(ps_uid, reinterpret_cast<u8*>(ps_bytecode->GetBufferPointer()),
static_cast<u32>(ps_bytecode->GetBufferSize()));
SETSTAT(stats.numPixelShadersAlive, static_cast<int>(s_ps_bytecode_cache.size()));
INCSTAT(stats.numPixelShadersCreated);
}
}
bool PixelShaderCache::SetShader(DSTALPHA_MODE dstAlphaMode)
{
PixelShaderUid uid = GetPixelShaderUid(dstAlphaMode, API_OPENGL);
// Check if the shader is already set
if (last_entry)
{
if (uid == last_uid)
{
return true;
}
}
last_uid = uid;
// Check if the shader is already in the cache
PSCache::iterator iter;
iter = PixelShaders.find(uid);
if (iter != PixelShaders.end())
{
const std::string &entry = iter->second;
last_entry = &entry;
GFX_DEBUGGER_PAUSE_AT(NEXT_PIXEL_SHADER_CHANGE,true);
return true;
}
// Need to compile a new shader
ShaderCode code = GeneratePixelShaderCode(dstAlphaMode, API_OPENGL);
PixelShaders.insert(std::make_pair(uid, code.GetBuffer()));
GFX_DEBUGGER_PAUSE_AT(NEXT_PIXEL_SHADER_CHANGE, true);
return true;
}
bool VertexShaderCache::SetShader(u32 components)
{
VertexShaderUid uid;
GetVertexShaderUid(uid, components, API_D3D);
if (g_ActiveConfig.bEnableShaderDebugging)
{
ShaderCode code;
GenerateVertexShaderCode(code, components, API_D3D);
vertex_uid_checker.AddToIndexAndCheck(code, uid, "Vertex", "v");
}
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(code, components, API_D3D);
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();
if (g_ActiveConfig.bEnableShaderDebugging && success)
{
vshaders[uid].code = code.GetBuffer();
}
GFX_DEBUGGER_PAUSE_AT(NEXT_VERTEX_SHADER_CHANGE, true);
return success;
}
bool GeometryShaderCache::SetShader(u32 primitive_type)
{
GeometryShaderUid uid = GetGeometryShaderUid(primitive_type);
// Check if the shader is already set
if (last_entry)
{
if (uid == last_uid)
{
GFX_DEBUGGER_PAUSE_AT(NEXT_PIXEL_SHADER_CHANGE, true);
return true;
}
}
last_uid = uid;
// Check if the shader is a pass-through shader
if (uid.GetUidData()->IsPassthrough())
{
// Return the default pass-through shader
last_entry = &pass_entry;
return true;
}
// Check if the shader is already in the cache
GSCache::iterator iter;
iter = GeometryShaders.find(uid);
if (iter != GeometryShaders.end())
{
const GSCacheEntry& entry = iter->second;
last_entry = &entry;
return (entry.shader != nullptr);
}
// Need to compile a new shader
ShaderCode code =
GenerateGeometryShaderCode(APIType::D3D, ShaderHostConfig::GetCurrent(), uid.GetUidData());
D3DBlob* pbytecode;
if (!D3D::CompileGeometryShader(code.GetBuffer(), &pbytecode))
{
GFX_DEBUGGER_PAUSE_AT(NEXT_ERROR, true);
return false;
}
// Insert the bytecode into the caches
g_gs_disk_cache.Append(uid, pbytecode->Data(), pbytecode->Size());
bool success = InsertByteCode(uid, pbytecode->Data(), pbytecode->Size());
pbytecode->Release();
return success;
}
bool VertexShaderCache::SetShader(u32 components)
{
VertexShaderUid uid;
GetVertexShaderUid(uid, components, API_D3D9);
if (g_ActiveConfig.bEnableShaderDebugging)
{
VertexShaderCode code;
GenerateVertexShaderCode(code, components, API_D3D9);
vertex_uid_checker.AddToIndexAndCheck(code, uid, "Vertex", "v");
}
if (last_entry)
{
if (uid == last_uid)
{
GFX_DEBUGGER_PAUSE_AT(NEXT_VERTEX_SHADER_CHANGE, true);
return (last_entry->shader != NULL);
}
}
last_uid = uid;
VSCache::iterator iter = vshaders.find(uid);
if (iter != vshaders.end())
{
const VSCacheEntry &entry = iter->second;
last_entry = &entry;
if (entry.shader) D3D::SetVertexShader(entry.shader);
GFX_DEBUGGER_PAUSE_AT(NEXT_VERTEX_SHADER_CHANGE, true);
return (entry.shader != NULL);
}
VertexShaderCode code;
GenerateVertexShaderCode(code, components, API_D3D9);
u8 *bytecode;
int bytecodelen;
if (!D3D::CompileVertexShader(code.GetBuffer(), (int)strlen(code.GetBuffer()), &bytecode, &bytecodelen))
{
GFX_DEBUGGER_PAUSE_AT(NEXT_ERROR, true);
return false;
}
g_vs_disk_cache.Append(uid, bytecode, bytecodelen);
bool success = InsertByteCode(uid, bytecode, bytecodelen, true);
if (g_ActiveConfig.bEnableShaderDebugging && success)
{
vshaders[uid].code = code.GetBuffer();
}
delete [] bytecode;
GFX_DEBUGGER_PAUSE_AT(NEXT_VERTEX_SHADER_CHANGE, true);
return success;
}
bool VertexShaderCache::SetShader(u32 components)
{
VERTEXSHADERUID uid;
GetVertexShaderId(&uid, components);
if (last_entry)
{
if (uid == last_uid)
{
GFX_DEBUGGER_PAUSE_AT(NEXT_VERTEX_SHADER_CHANGE, true);
ValidateVertexShaderIDs(API_D3D11, last_entry->safe_uid, last_entry->code, components);
return (last_entry->shader != NULL);
}
}
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);
ValidateVertexShaderIDs(API_D3D11, entry.safe_uid, entry.code, components);
return (entry.shader != NULL);
}
const char *code = GenerateVertexShaderCode(components, API_D3D11);
D3DBlob* pbytecode = NULL;
D3D::CompileVertexShader(code, (int)strlen(code), &pbytecode);
if (pbytecode == NULL)
{
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();
if (g_ActiveConfig.bEnableShaderDebugging && success)
{
vshaders[uid].code = code;
GetSafeVertexShaderId(&vshaders[uid].safe_uid, components);
}
GFX_DEBUGGER_PAUSE_AT(NEXT_VERTEX_SHADER_CHANGE, true);
return success;
}
VERTEXSHADER* VertexShaderCache::SetShader(u32 components)
{
VERTEXSHADERUID uid;
GetVertexShaderId(&uid, components);
if (uid == last_vertex_shader_uid && vshaders[uid].frameCount == frameCount)
{
GFX_DEBUGGER_PAUSE_AT(NEXT_VERTEX_SHADER_CHANGE, true);
return pShaderLast;
}
memcpy(&last_vertex_shader_uid, &uid, sizeof(VERTEXSHADERUID));
VSCache::iterator iter = vshaders.find(uid);
if (iter != vshaders.end())
{
iter->second.frameCount = frameCount;
VSCacheEntry &entry = iter->second;
if (&entry.shader != pShaderLast) {
pShaderLast = &entry.shader;
}
GFX_DEBUGGER_PAUSE_AT(NEXT_VERTEX_SHADER_CHANGE, true);
return pShaderLast;
}
// Make an entry in the table
VSCacheEntry& entry = vshaders[uid];
entry.frameCount = frameCount;
pShaderLast = &entry.shader;
const char *code = GenerateVertexShaderCode(components, API_OPENGL);
#if defined(_DEBUG) || defined(DEBUGFAST)
if (g_ActiveConfig.iLog & CONF_SAVESHADERS && code) {
static int counter = 0;
char szTemp[MAX_PATH];
sprintf(szTemp, "%svs_%04i.txt", File::GetUserPath(D_DUMP_IDX).c_str(), counter++);
SaveData(szTemp, code);
}
#endif
if (!code || !VertexShaderCache::CompileVertexShader(entry.shader, code)) {
ERROR_LOG(VIDEO, "failed to create vertex shader");
GFX_DEBUGGER_PAUSE_AT(NEXT_ERROR, true);
return NULL;
}
INCSTAT(stats.numVertexShadersCreated);
SETSTAT(stats.numVertexShadersAlive, vshaders.size());
GFX_DEBUGGER_PAUSE_AT(NEXT_VERTEX_SHADER_CHANGE, true);
return pShaderLast;
}
void ShaderCache::HandleGSUIDChange(GeometryShaderUid gs_uid, u32 gs_primitive_type)
{
s_last_geometry_shader_uid = gs_uid;
if (gs_uid.GetUidData()->IsPassthrough())
{
s_last_geometry_shader_bytecode = {};
return;
}
auto gs_iterator = s_gs_bytecode_cache.find(gs_uid);
if (gs_iterator != s_gs_bytecode_cache.end())
{
s_last_geometry_shader_bytecode = gs_iterator->second;
}
else
{
ShaderCode gs_code = GenerateGeometryShaderCode(APIType::D3D, gs_uid.GetUidData());
ID3DBlob* gs_bytecode = nullptr;
if (!D3D::CompileGeometryShader(gs_code.GetBuffer(), &gs_bytecode))
{
GFX_DEBUGGER_PAUSE_AT(NEXT_ERROR, true);
return;
}
s_last_geometry_shader_bytecode = InsertByteCode(gs_uid, &s_gs_bytecode_cache, gs_bytecode);
s_gs_disk_cache.Append(gs_uid, reinterpret_cast<u8*>(gs_bytecode->GetBufferPointer()),
static_cast<u32>(gs_bytecode->GetBufferSize()));
}
}
void PixelShaderCache::PrepareShader(
PIXEL_SHADER_RENDER_MODE render_mode,
u32 components,
const XFMemory &xfr,
const BPMemory &bpm,
bool ongputhread)
{
PixelShaderUid uid;
GetPixelShaderUID(uid, render_mode, components, xfr, bpm);
if (ongputhread)
{
s_compiler->ProcCompilationResults();
// Check if the shader is already set
if (s_last_entry[render_mode])
{
if (uid == s_last_uid[render_mode])
{
return;
}
}
s_last_uid[render_mode] = uid;
GFX_DEBUGGER_PAUSE_AT(NEXT_PIXEL_SHADER_CHANGE, true);
}
else
{
if (s_external_last_uid[render_mode] == uid)
{
return;
}
s_external_last_uid[render_mode] = uid;
}
CompilePShader(uid, render_mode, ongputhread);
}
void VertexShaderCache::PrepareShader(
u32 components,
const XFMemory
&xfr,
const BPMemory &bpm,
bool ongputhread)
{
VertexShaderUid uid;
GetVertexShaderUID(uid, components, xfr, bpm);
if (ongputhread)
{
s_compiler->ProcCompilationResults();
if (s_last_entry)
{
if (uid == s_last_uid)
{
return;
}
}
s_last_uid = uid;
GFX_DEBUGGER_PAUSE_AT(NEXT_VERTEX_SHADER_CHANGE, true);
}
else
{
if (s_external_last_uid == uid)
{
return;
}
s_external_last_uid = uid;
}
CompileVShader(uid, ongputhread);
}
bool PixelShaderCache::SetShader()
{
PixelShaderUid uid = GetPixelShaderUid();
// Check if the shader is already set
if (last_entry)
{
if (uid == last_uid)
{
GFX_DEBUGGER_PAUSE_AT(NEXT_PIXEL_SHADER_CHANGE, true);
return (last_entry->shader != nullptr);
}
}
last_uid = uid;
// Check if the shader is already in the cache
PSCache::iterator iter;
iter = PixelShaders.find(uid);
if (iter != PixelShaders.end())
{
const PSCacheEntry& entry = iter->second;
last_entry = &entry;
GFX_DEBUGGER_PAUSE_AT(NEXT_PIXEL_SHADER_CHANGE, true);
return (entry.shader != nullptr);
}
// Need to compile a new shader
ShaderCode code = GeneratePixelShaderCode(APIType::D3D, uid.GetUidData());
D3DBlob* pbytecode;
if (!D3D::CompilePixelShader(code.GetBuffer(), &pbytecode))
{
GFX_DEBUGGER_PAUSE_AT(NEXT_ERROR, true);
return false;
}
// Insert the bytecode into the caches
g_ps_disk_cache.Append(uid, pbytecode->Data(), pbytecode->Size());
bool success = InsertByteCode(uid, pbytecode->Data(), pbytecode->Size());
pbytecode->Release();
GFX_DEBUGGER_PAUSE_AT(NEXT_PIXEL_SHADER_CHANGE, true);
return success;
}
// Used by TextureCacheBase::Load
TextureCacheBase::TCacheEntryBase* TextureCacheBase::ReturnEntry(u32 stage, TCacheEntryBase* entry)
{
entry->frameCount = FRAMECOUNT_INVALID;
bound_textures[stage] = entry;
s_last_texture = std::max(s_last_texture, stage);
GFX_DEBUGGER_PAUSE_AT(NEXT_TEXTURE_CHANGE, true);
return entry;
}
// Used by TextureCache::Load
static TextureCache::TCacheEntryBase* ReturnEntry(unsigned int stage, TextureCache::TCacheEntryBase* entry)
{
entry->frameCount = frameCount;
entry->Bind(stage);
GFX_DEBUGGER_PAUSE_AT(NEXT_TEXTURE_CHANGE, true);
return entry;
}
// Used by TextureCache::Load
TextureCache::TCacheEntryBase* TextureCache::ReturnEntry(unsigned int stage, TCacheEntryBase* entry)
{
entry->frameCount = FRAMECOUNT_INVALID;
bound_textures[stage] = entry;
GFX_DEBUGGER_PAUSE_AT(NEXT_TEXTURE_CHANGE, true);
return entry;
}
bool VertexShaderCache::SetShader(u32 components)
{
VERTEXSHADERUID uid;
GetVertexShaderId(&uid, components);
if (uid == last_vertex_shader_uid && vshaders[uid].frameCount == frameCount)
{
GFX_DEBUGGER_PAUSE_AT(NEXT_VERTEX_SHADER_CHANGE, true);
return (vshaders[uid].shader != NULL);
}
memcpy(&last_vertex_shader_uid, &uid, sizeof(VERTEXSHADERUID));
VSCache::iterator iter = vshaders.find(uid);
if (iter != vshaders.end())
{
iter->second.frameCount = frameCount;
const VSCacheEntry &entry = iter->second;
last_entry = &entry;
if (entry.shader) D3D::SetVertexShader(entry.shader);
GFX_DEBUGGER_PAUSE_AT(NEXT_VERTEX_SHADER_CHANGE, true);
return (entry.shader != NULL);
}
const char *code = GenerateVertexShaderCode(components, API_D3D9);
u8 *bytecode;
int bytecodelen;
if (!D3D::CompileVertexShader(code, (int)strlen(code), &bytecode, &bytecodelen))
{
if (g_ActiveConfig.bShowShaderErrors)
{
PanicAlert("Failed to compile Vertex Shader:\n\n%s", code);
}
GFX_DEBUGGER_PAUSE_AT(NEXT_ERROR, true);
return false;
}
g_vs_disk_cache.Append(uid, bytecode, bytecodelen);
g_vs_disk_cache.Sync();
bool result = InsertByteCode(uid, bytecode, bytecodelen, true);
delete [] bytecode;
GFX_DEBUGGER_PAUSE_AT(NEXT_VERTEX_SHADER_CHANGE, true);
return result;
}
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 Renderer::Swap(u32 xfbAddr, u32 fbWidth, u32 fbHeight, const EFBRectangle& rc, float Gamma)
{
// TODO: merge more generic parts into VideoCommon
g_renderer->SwapImpl(xfbAddr, fbWidth, fbHeight, rc, Gamma);
frameCount++;
GFX_DEBUGGER_PAUSE_AT(NEXT_FRAME, true);
// Begin new frame
// Set default viewport and scissor, for the clear to work correctly
// New frame
stats.ResetFrame();
Core::Callback_VideoCopiedToXFB(XFBWrited || (g_ActiveConfig.bUseXFB && g_ActiveConfig.bUseRealXFB));
XFBWrited = false;
}
bool VertexShaderCache::SetUberShader(D3DVertexFormat* vertex_format)
{
D3DVertexFormat* uber_vertex_format = static_cast<D3DVertexFormat*>(
VertexLoaderManager::GetUberVertexFormat(vertex_format->GetVertexDeclaration()));
UberShader::VertexShaderUid uid = UberShader::GetVertexShaderUid();
if (last_uber_entry && last_uber_uid == uid)
{
if (!last_uber_entry->shader)
return false;
uber_vertex_format->SetInputLayout(last_uber_entry->bytecode);
D3D::stateman->SetVertexShader(last_uber_entry->shader);
return true;
}
auto iter = ubervshaders.find(uid);
if (iter != ubervshaders.end())
{
const VSCacheEntry& entry = iter->second;
last_uber_uid = uid;
last_uber_entry = &entry;
GFX_DEBUGGER_PAUSE_AT(NEXT_VERTEX_SHADER_CHANGE, true);
if (!last_uber_entry->shader)
return false;
uber_vertex_format->SetInputLayout(last_uber_entry->bytecode);
D3D::stateman->SetVertexShader(last_uber_entry->shader);
return true;
}
// Need to compile a new shader
D3DBlob* bytecode = nullptr;
ShaderCode code =
UberShader::GenVertexShader(APIType::D3D, ShaderHostConfig::GetCurrent(), uid.GetUidData());
D3D::CompileVertexShader(code.GetBuffer(), &bytecode);
if (!InsertByteCode(uid, bytecode))
{
SAFE_RELEASE(bytecode);
return false;
}
g_uber_vs_disk_cache.Append(uid, bytecode->Data(), bytecode->Size());
bytecode->Release();
return SetUberShader(vertex_format);
}
void GeometryShaderCache::PrepareShader(
u32 primitive_type,
const XFMemory &xfr,
const u32 components,
bool ongputhread)
{
GeometryShaderUid uid;
GetGeometryShaderUid(uid, primitive_type, xfr, components);
if (ongputhread)
{
s_compiler->ProcCompilationResults();
// Check if the shader is already set
if (s_last_entry)
{
if (uid == s_last_uid)
{
return;
}
}
s_last_uid = uid;
// Check if the shader is a pass-through shader
if (uid.GetUidData().IsPassthrough())
{
// Return the default pass-through shader
s_last_entry = &s_pass_entry;
return;
}
GFX_DEBUGGER_PAUSE_AT(NEXT_PIXEL_SHADER_CHANGE, true);
}
else
{
if (s_external_last_uid == uid)
{
return;
}
s_external_last_uid = uid;
}
CompileGShader(uid, ongputhread);
}
void Renderer::Swap(u32 xfbAddr, u32 fbWidth, u32 fbStride, u32 fbHeight, const EFBRectangle& rc, u64 ticks, float Gamma)
{
// Heuristic to detect if a GameCube game is in 16:9 anamorphic widescreen mode.
if (!SConfig::GetInstance().bWii)
{
size_t flush_count_4_3, flush_count_anamorphic;
std::tie(flush_count_4_3, flush_count_anamorphic) =
g_vertex_manager->ResetFlushAspectRatioCount();
size_t flush_total = flush_count_4_3 + flush_count_anamorphic;
// Modify the threshold based on which aspect ratio we're already using: if
// the game's in 4:3, it probably won't switch to anamorphic, and vice-versa.
if (m_aspect_wide)
m_aspect_wide = !(flush_count_4_3 > 0.75 * flush_total);
else
m_aspect_wide = flush_count_anamorphic > 0.75 * flush_total;
}
// TODO: merge more generic parts into VideoCommon
SwapImpl(xfbAddr, fbWidth, fbStride, fbHeight, rc, ticks, Gamma);
if (m_xfb_written || (g_ActiveConfig.bUseXFB && g_ActiveConfig.bUseRealXFB))
m_fps_counter.Update();
frameCount++;
GFX_DEBUGGER_PAUSE_AT(NEXT_FRAME, true);
if (g_ActiveConfig.bBlackFrameInsertion)
{
InsertBlackFrame();
frameCount++;
}
// Begin new frame
// Set default viewport and scissor, for the clear to work correctly
// New frame
stats.ResetFrame();
Core::Callback_VideoCopiedToXFB(m_xfb_written || (g_ActiveConfig.bUseXFB && g_ActiveConfig.bUseRealXFB));
m_xfb_written = false;
}
SHADER* ProgramShaderCache::SetShader ( DSTALPHA_MODE dstAlphaMode, u32 components )
{
SHADERUID uid;
GetShaderId(&uid, dstAlphaMode, components);
// Check if the shader is already set
if (last_entry)
{
if (uid == last_uid)
{
GFX_DEBUGGER_PAUSE_AT(NEXT_PIXEL_SHADER_CHANGE, true);
last_entry->shader.Bind();
return &last_entry->shader;
}
}
last_uid = uid;
// Check if shader is already in cache
PCache::iterator iter = pshaders.find(uid);
if (iter != pshaders.end())
{
PCacheEntry *entry = &iter->second;
last_entry = entry;
GFX_DEBUGGER_PAUSE_AT(NEXT_PIXEL_SHADER_CHANGE, true);
last_entry->shader.Bind();
return &last_entry->shader;
}
// Make an entry in the table
PCacheEntry& newentry = pshaders[uid];
last_entry = &newentry;
newentry.in_cache = 0;
VertexShaderCode vcode;
PixelShaderCode pcode;
GenerateVertexShaderCode(vcode, components, API_OPENGL);
GeneratePixelShaderCode(pcode, dstAlphaMode, API_OPENGL, components);
if (g_ActiveConfig.bEnableShaderDebugging)
{
newentry.shader.strvprog = vcode.GetBuffer();
newentry.shader.strpprog = pcode.GetBuffer();
}
#if defined(_DEBUG) || defined(DEBUGFAST)
if (g_ActiveConfig.iLog & CONF_SAVESHADERS) {
static int counter = 0;
char szTemp[MAX_PATH];
sprintf(szTemp, "%svs_%04i.txt", File::GetUserPath(D_DUMP_IDX).c_str(), counter++);
SaveData(szTemp, vcode.GetBuffer());
sprintf(szTemp, "%sps_%04i.txt", File::GetUserPath(D_DUMP_IDX).c_str(), counter++);
SaveData(szTemp, pcode.GetBuffer());
}
#endif
if (!CompileShader(newentry.shader, vcode.GetBuffer(), pcode.GetBuffer())) {
GFX_DEBUGGER_PAUSE_AT(NEXT_ERROR, true);
return NULL;
}
INCSTAT(stats.numPixelShadersCreated);
SETSTAT(stats.numPixelShadersAlive, pshaders.size());
GFX_DEBUGGER_PAUSE_AT(NEXT_PIXEL_SHADER_CHANGE, true);
last_entry->shader.Bind();
return &last_entry->shader;
}
TextureCache::TCacheEntryBase* TextureCache::Load(unsigned int const stage,
u32 const address, unsigned int width, unsigned int height, int const texformat,
unsigned int const tlutaddr, int const tlutfmt, bool const use_mipmaps, unsigned int maxlevel, bool const from_tmem)
{
if (0 == address)
return nullptr;
// TexelSizeInNibbles(format) * width * height / 16;
const unsigned int bsw = TexDecoder_GetBlockWidthInTexels(texformat) - 1;
const unsigned int bsh = TexDecoder_GetBlockHeightInTexels(texformat) - 1;
unsigned int expandedWidth = (width + bsw) & (~bsw);
unsigned int expandedHeight = (height + bsh) & (~bsh);
const unsigned int nativeW = width;
const unsigned int nativeH = height;
u32 texID = address;
// Hash assigned to texcache entry (also used to generate filenames used for texture dumping and custom texture lookup)
u64 tex_hash = TEXHASH_INVALID;
u64 tlut_hash = TEXHASH_INVALID;
u32 full_format = texformat;
PC_TexFormat pcfmt = PC_TEX_FMT_NONE;
const bool isPaletteTexture = (texformat == GX_TF_C4 || texformat == GX_TF_C8 || texformat == GX_TF_C14X2);
if (isPaletteTexture)
full_format = texformat | (tlutfmt << 16);
const u32 texture_size = TexDecoder_GetTextureSizeInBytes(expandedWidth, expandedHeight, texformat);
const u8* src_data;
if (from_tmem)
src_data = &texMem[bpmem.tex[stage / 4].texImage1[stage % 4].tmem_even * TMEM_LINE_SIZE];
else
src_data = Memory::GetPointer(address);
// TODO: This doesn't hash GB tiles for preloaded RGBA8 textures (instead, it's hashing more data from the low tmem bank than it should)
tex_hash = GetHash64(src_data, texture_size, g_ActiveConfig.iSafeTextureCache_ColorSamples);
if (isPaletteTexture)
{
const u32 palette_size = TexDecoder_GetPaletteSize(texformat);
tlut_hash = GetHash64(&texMem[tlutaddr], palette_size, g_ActiveConfig.iSafeTextureCache_ColorSamples);
// NOTE: For non-paletted textures, texID is equal to the texture address.
// A paletted texture, however, may have multiple texIDs assigned though depending on the currently used tlut.
// This (changing texID depending on the tlut_hash) is a trick to get around
// an issue with Metroid Prime's fonts (it has multiple sets of fonts on each other
// stored in a single texture and uses the palette to make different characters
// visible or invisible. Thus, unless we want to recreate the textures for every drawn character,
// we must make sure that a paletted texture gets assigned multiple IDs for each tlut used.
//
// TODO: Because texID isn't always the same as the address now, CopyRenderTargetToTexture might be broken now
texID ^= ((u32)tlut_hash) ^(u32)(tlut_hash >> 32);
tex_hash ^= tlut_hash;
}
// D3D doesn't like when the specified mipmap count would require more than one 1x1-sized LOD in the mipmap chain
// e.g. 64x64 with 7 LODs would have the mipmap chain 64x64,32x32,16x16,8x8,4x4,2x2,1x1,1x1, so we limit the mipmap count to 6 there
while (g_ActiveConfig.backend_info.bUseMinimalMipCount && std::max(expandedWidth, expandedHeight) >> maxlevel == 0)
--maxlevel;
TCacheEntryBase *entry = textures[texID];
if (entry)
{
// 1. Calculate reference hash:
// calculated from RAM texture data for normal textures. Hashes for paletted textures are modified by tlut_hash. 0 for virtual EFB copies.
if (g_ActiveConfig.bCopyEFBToTexture && entry->IsEfbCopy())
tex_hash = TEXHASH_INVALID;
// 2. a) For EFB copies, only the hash and the texture address need to match
if (entry->IsEfbCopy() && tex_hash == entry->hash && address == entry->addr)
{
entry->type = TCET_EC_VRAM;
// TODO: Print a warning if the format changes! In this case,
// we could reinterpret the internal texture object data to the new pixel format
// (similar to what is already being done in Renderer::ReinterpretPixelFormat())
return ReturnEntry(stage, entry);
}
// 2. b) For normal textures, all texture parameters need to match
if (address == entry->addr && tex_hash == entry->hash && full_format == entry->format &&
entry->num_mipmaps > maxlevel && entry->native_width == nativeW && entry->native_height == nativeH)
{
return ReturnEntry(stage, entry);
}
// 3. If we reach this line, we'll have to upload the new texture data to VRAM.
// If we're lucky, the texture parameters didn't change and we can reuse the internal texture object instead of destroying and recreating it.
//
// TODO: Don't we need to force texture decoding to RGBA8 for dynamic EFB copies?
// TODO: Actually, it should be enough if the internal texture format matches...
if ((entry->type == TCET_NORMAL &&
width == entry->virtual_width &&
height == entry->virtual_height &&
full_format == entry->format &&
entry->num_mipmaps > maxlevel) ||
(entry->type == TCET_EC_DYNAMIC &&
entry->native_width == width &&
entry->native_height == height))
{
// reuse the texture
}
else
{
// delete the texture and make a new one
delete entry;
entry = nullptr;
}
}
bool using_custom_texture = false;
if (g_ActiveConfig.bHiresTextures)
{
// This function may modify width/height.
pcfmt = LoadCustomTexture(tex_hash, texformat, 0, width, height);
if (pcfmt != PC_TEX_FMT_NONE)
{
if (expandedWidth != width || expandedHeight != height)
{
expandedWidth = width;
expandedHeight = height;
// If we thought we could reuse the texture before, make sure to pool it now!
if (entry)
{
delete entry;
entry = nullptr;
}
}
using_custom_texture = true;
}
}
if (!using_custom_texture)
{
if (!(texformat == GX_TF_RGBA8 && from_tmem))
{
pcfmt = TexDecoder_Decode(temp, src_data, expandedWidth,
expandedHeight, texformat, tlutaddr, tlutfmt, g_ActiveConfig.backend_info.bUseRGBATextures);
}
else
{
u8* src_data_gb = &texMem[bpmem.tex[stage/4].texImage2[stage%4].tmem_odd * TMEM_LINE_SIZE];
pcfmt = TexDecoder_DecodeRGBA8FromTmem(temp, src_data, src_data_gb, expandedWidth, expandedHeight);
}
}
u32 texLevels = use_mipmaps ? (maxlevel + 1) : 1;
const bool using_custom_lods = using_custom_texture && CheckForCustomTextureLODs(tex_hash, texformat, texLevels);
// Only load native mips if their dimensions fit to our virtual texture dimensions
const bool use_native_mips = use_mipmaps && !using_custom_lods && (width == nativeW && height == nativeH);
texLevels = (use_native_mips || using_custom_lods) ? texLevels : 1; // TODO: Should be forced to 1 for non-pow2 textures (e.g. efb copies with automatically adjusted IR)
// create the entry/texture
if (nullptr == entry)
{
textures[texID] = entry = g_texture_cache->CreateTexture(width, height, expandedWidth, texLevels, pcfmt);
// Sometimes, we can get around recreating a texture if only the number of mip levels changes
// e.g. if our texture cache entry got too many mipmap levels we can limit the number of used levels by setting the appropriate render states
// Thus, we don't update this member for every Load, but just whenever the texture gets recreated
// TODO: This is the wrong value. We should be storing the number of levels our actual texture has.
// But that will currently make the above "existing entry" tests fail as "texLevels" is not calculated until after.
// Currently, we might try to reuse a texture which appears to have more levels than actual, maybe..
entry->num_mipmaps = maxlevel + 1;
entry->type = TCET_NORMAL;
GFX_DEBUGGER_PAUSE_AT(NEXT_NEW_TEXTURE, true);
}
else
{
// load texture (CreateTexture also loads level 0)
entry->Load(width, height, expandedWidth, 0);
}
entry->SetGeneralParameters(address, texture_size, full_format, entry->num_mipmaps);
entry->SetDimensions(nativeW, nativeH, width, height);
entry->hash = tex_hash;
if (entry->IsEfbCopy() && !g_ActiveConfig.bCopyEFBToTexture)
entry->type = TCET_EC_DYNAMIC;
else
entry->type = TCET_NORMAL;
if (g_ActiveConfig.bDumpTextures && !using_custom_texture)
DumpTexture(entry, 0);
u32 level = 1;
// load mips - TODO: Loading mipmaps from tmem is untested!
if (pcfmt != PC_TEX_FMT_NONE)
{
if (use_native_mips)
{
src_data += texture_size;
const u8* ptr_even = nullptr;
const u8* ptr_odd = nullptr;
if (from_tmem)
{
ptr_even = &texMem[bpmem.tex[stage/4].texImage1[stage%4].tmem_even * TMEM_LINE_SIZE + texture_size];
ptr_odd = &texMem[bpmem.tex[stage/4].texImage2[stage%4].tmem_odd * TMEM_LINE_SIZE];
}
for (; level != texLevels; ++level)
{
const u32 mip_width = CalculateLevelSize(width, level);
const u32 mip_height = CalculateLevelSize(height, level);
const u32 expanded_mip_width = (mip_width + bsw) & (~bsw);
const u32 expanded_mip_height = (mip_height + bsh) & (~bsh);
const u8*& mip_src_data = from_tmem
? ((level % 2) ? ptr_odd : ptr_even)
: src_data;
TexDecoder_Decode(temp, mip_src_data, expanded_mip_width, expanded_mip_height, texformat, tlutaddr, tlutfmt, g_ActiveConfig.backend_info.bUseRGBATextures);
mip_src_data += TexDecoder_GetTextureSizeInBytes(expanded_mip_width, expanded_mip_height, texformat);
entry->Load(mip_width, mip_height, expanded_mip_width, level);
if (g_ActiveConfig.bDumpTextures)
DumpTexture(entry, level);
}
}
else if (using_custom_lods)
{
for (; level != texLevels; ++level)
{
unsigned int mip_width = CalculateLevelSize(width, level);
unsigned int mip_height = CalculateLevelSize(height, level);
LoadCustomTexture(tex_hash, texformat, level, mip_width, mip_height);
entry->Load(mip_width, mip_height, mip_width, level);
}
}
}
INCSTAT(stats.numTexturesCreated);
SETSTAT(stats.numTexturesAlive, textures.size());
return ReturnEntry(stage, entry);
}
bool GeometryShaderCache::SetShader(u32 primitive_type)
{
switch (primitive_type)
{
case PRIMITIVE_TRIANGLES:
currentPrimitiveTopology = D3D12_PRIMITIVE_TOPOLOGY_TYPE_TRIANGLE;
break;
case PRIMITIVE_LINES:
currentPrimitiveTopology = D3D12_PRIMITIVE_TOPOLOGY_TYPE_LINE;
break;
case PRIMITIVE_POINTS:
currentPrimitiveTopology = D3D12_PRIMITIVE_TOPOLOGY_TYPE_POINT;
break;
default:
CHECK(0, "Invalid primitive type.");
break;
}
GeometryShaderUid uid = GetGeometryShaderUid(primitive_type, API_D3D);
// Check if the shader is already set
if (uid == last_uid)
{
GFX_DEBUGGER_PAUSE_AT(NEXT_PIXEL_SHADER_CHANGE,true);
return true;
}
last_uid = uid;
D3D::commandListMgr->dirtyPso = true;
if (g_ActiveConfig.bEnableShaderDebugging)
{
ShaderCode code = GenerateGeometryShaderCode(primitive_type, API_D3D);
geometry_uid_checker.AddToIndexAndCheck(code, uid, "Geometry", "g");
}
// Check if the shader is a pass-through shader
if (uid.GetUidData()->IsPassthrough())
{
// Return the default pass-through shader
last_entry = &pass_entry;
return true;
}
// Check if the shader is already in the cache
GSCache::iterator iter;
iter = GeometryShaders.find(uid);
if (iter != GeometryShaders.end())
{
const GSCacheEntry &entry = iter->second;
last_entry = &entry;
return (entry.shader12.pShaderBytecode != nullptr);
}
// Need to compile a new shader
ShaderCode code = GenerateGeometryShaderCode(primitive_type, API_D3D);
D3DBlob* pbytecode;
if (!D3D::CompileGeometryShader(code.GetBuffer(), &pbytecode))
{
GFX_DEBUGGER_PAUSE_AT(NEXT_ERROR, true);
return false;
}
// Insert the bytecode into the caches
g_gs_disk_cache.Append(uid, pbytecode->Data(), pbytecode->Size());
bool success = InsertByteCode(uid, pbytecode->Data(), pbytecode->Size());
pbytecode->Release();
if (g_ActiveConfig.bEnableShaderDebugging && success)
{
GeometryShaders[uid].code = code.GetBuffer();
}
return success;
}
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 PixelShaderCache::PrepareShader(
PIXEL_SHADER_RENDER_MODE render_mode,
u32 components,
const XFMemory &xfr,
const BPMemory &bpm,
bool ongputhread)
{
const API_TYPE api = ((D3D::GetCaps().PixelShaderVersion >> 8) & 0xFF) < 3 ? API_D3D9_SM20 : API_D3D9_SM30;
PixelShaderUid uid;
GetPixelShaderUID(uid, render_mode, components, xfr, bpm);
if (ongputhread)
{
Compiler->ProcCompilationResults();
#if defined(_DEBUG) || defined(DEBUGFAST)
if (g_ActiveConfig.bEnableShaderDebugging)
{
ShaderCode code;
GeneratePixelShaderCodeD3D9(code, uid.GetUidData());
}
#endif
// Check if the shader is already set
if (last_entry[render_mode])
{
if (uid == last_uid[render_mode])
{
return;
}
}
last_uid[render_mode] = uid;
GFX_DEBUGGER_PAUSE_AT(NEXT_PIXEL_SHADER_CHANGE, true);
}
else
{
if (external_last_uid[render_mode] == uid)
{
return;
}
external_last_uid[render_mode] = uid;
}
PixelShadersLock.lock();
PSCacheEntry* entry = &PixelShaders[uid];
PixelShadersLock.unlock();
if (ongputhread)
{
last_entry[render_mode] = entry;
}
// Compile only when we have a new instance
if (entry->initialized.test_and_set())
{
return;
}
// Need to compile a new shader
ShaderCompilerWorkUnit *wunit = Compiler->NewUnit(PIXELSHADERGEN_BUFFERSIZE);
wunit->GenerateCodeHandler = [uid, api](ShaderCompilerWorkUnit* wunit)
{
ShaderCode code;
code.SetBuffer(wunit->code.data());
if (api == API_D3D9_SM20)
{
GeneratePixelShaderCodeD3D9SM2(code, uid.GetUidData());
}
else
{
GeneratePixelShaderCodeD3D9(code, uid.GetUidData());
}
wunit->codesize = (u32)code.BufferSize();
};
wunit->entrypoint = "main";
wunit->flags = D3DCOMPILE_SKIP_VALIDATION | D3DCOMPILE_OPTIMIZATION_LEVEL3;
wunit->target = D3D::PixelShaderVersionString();
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_ps_disk_cache.Append(uid, bytecode, bytecodelen);
PushByteCode(uid, bytecode, bytecodelen, entry);
#if defined(_DEBUG) || defined(DEBUGFAST)
if (g_ActiveConfig.bEnableShaderDebugging)
{
u32 code_hash = HashAdler32((const u8 *)wunit->code.data(), wunit->codesize);
unique_shaders.insert(code_hash);
entry->code = wunit->code.data();
}
if (g_ActiveConfig.iLog & CONF_SAVESHADERS) {
static int counter = 0;
char szTemp[MAX_PATH];
sprintf(szTemp, "%sps_%04i.txt", File::GetUserPath(D_DUMP_IDX).c_str(), counter++);
SaveData(szTemp, wunit->code.data());
}
#endif
}
else
{
static int num_failures = 0;
std::string filename = StringFromFormat("%sbad_ps_%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 pixel 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(),
(const char*)wunit->error->GetBufferPointer());
}
};
Compiler->CompileShaderAsync(wunit);
}
void VertexManager::vFlush()
{
GLVertexFormat *nativeVertexFmt = (GLVertexFormat*)g_nativeVertexFmt;
u32 stride = nativeVertexFmt->GetVertexStride();
if(m_last_vao != nativeVertexFmt->VAO) {
glBindVertexArray(nativeVertexFmt->VAO);
m_last_vao = nativeVertexFmt->VAO;
}
PrepareDrawBuffers(stride);
GL_REPORT_ERRORD();
bool useDstAlpha = !g_ActiveConfig.bDstAlphaPass && bpmem.dstalpha.enable && bpmem.blendmode.alphaupdate
&& bpmem.zcontrol.pixel_format == PIXELFMT_RGBA6_Z24;
// Makes sure we can actually do Dual source blending
bool dualSourcePossible = g_ActiveConfig.backend_info.bSupportsDualSourceBlend;
// finally bind
if (dualSourcePossible)
{
if (useDstAlpha)
{
// If host supports GL_ARB_blend_func_extended, we can do dst alpha in
// the same pass as regular rendering.
ProgramShaderCache::SetShader(DSTALPHA_DUAL_SOURCE_BLEND, g_nativeVertexFmt->m_components);
}
else
{
ProgramShaderCache::SetShader(DSTALPHA_NONE,g_nativeVertexFmt->m_components);
}
}
else
{
ProgramShaderCache::SetShader(DSTALPHA_NONE,g_nativeVertexFmt->m_components);
}
// upload global constants
ProgramShaderCache::UploadConstants();
// setup the pointers
if (g_nativeVertexFmt)
g_nativeVertexFmt->SetupVertexPointers();
GL_REPORT_ERRORD();
g_perf_query->EnableQuery(bpmem.zcontrol.early_ztest ? PQG_ZCOMP_ZCOMPLOC : PQG_ZCOMP);
Draw(stride);
g_perf_query->DisableQuery(bpmem.zcontrol.early_ztest ? PQG_ZCOMP_ZCOMPLOC : PQG_ZCOMP);
//ERROR_LOG(VIDEO, "PerfQuery result: %d", g_perf_query->GetQueryResult(bpmem.zcontrol.early_ztest ? PQ_ZCOMP_OUTPUT_ZCOMPLOC : PQ_ZCOMP_OUTPUT));
// run through vertex groups again to set alpha
if (useDstAlpha && !dualSourcePossible)
{
ProgramShaderCache::SetShader(DSTALPHA_ALPHA_PASS,g_nativeVertexFmt->m_components);
// only update alpha
glColorMask(GL_FALSE, GL_FALSE, GL_FALSE, GL_TRUE);
glDisable(GL_BLEND);
Draw(stride);
// restore color mask
g_renderer->SetColorMask();
if (bpmem.blendmode.blendenable || bpmem.blendmode.subtract)
glEnable(GL_BLEND);
}
GFX_DEBUGGER_PAUSE_AT(NEXT_FLUSH, true);
#if defined(_DEBUG) || defined(DEBUGFAST)
if (g_ActiveConfig.iLog & CONF_SAVESHADERS)
{
// save the shaders
ProgramShaderCache::PCacheEntry prog = ProgramShaderCache::GetShaderProgram();
char strfile[255];
sprintf(strfile, "%sps%.3d.txt", File::GetUserPath(D_DUMPFRAMES_IDX).c_str(), g_ActiveConfig.iSaveTargetId);
std::ofstream fps;
OpenFStream(fps, strfile, std::ios_base::out);
fps << prog.shader.strpprog.c_str();
sprintf(strfile, "%svs%.3d.txt", File::GetUserPath(D_DUMPFRAMES_IDX).c_str(), g_ActiveConfig.iSaveTargetId);
std::ofstream fvs;
OpenFStream(fvs, strfile, std::ios_base::out);
fvs << prog.shader.strvprog.c_str();
}
if (g_ActiveConfig.iLog & CONF_SAVETARGETS)
{
char str[128];
sprintf(str, "%starg%.3d.png", File::GetUserPath(D_DUMPFRAMES_IDX).c_str(), g_ActiveConfig.iSaveTargetId);
TargetRectangle tr;
tr.left = 0;
tr.right = Renderer::GetTargetWidth();
tr.top = 0;
tr.bottom = Renderer::GetTargetHeight();
g_renderer->SaveScreenshot(str, tr);
}
#endif
g_Config.iSaveTargetId++;
ClearEFBCache();
GL_REPORT_ERRORD();
}
TextureCache::TCacheEntryBase* TextureCache::Load(const u32 stage)
{
const FourTexUnits &tex = bpmem.tex[stage >> 2];
const u32 id = stage & 3;
const u32 address = (tex.texImage3[id].image_base/* & 0x1FFFFF*/) << 5;
u32 width = tex.texImage0[id].width + 1;
u32 height = tex.texImage0[id].height + 1;
const int texformat = tex.texImage0[id].format;
const u32 tlutaddr = tex.texTlut[id].tmem_offset << 9;
const u32 tlutfmt = tex.texTlut[id].tlut_format;
const bool use_mipmaps = (tex.texMode0[id].min_filter & 3) != 0;
u32 tex_levels = use_mipmaps ? ((tex.texMode1[id].max_lod + 0xf) / 0x10 + 1) : 1;
const bool from_tmem = tex.texImage1[id].image_type != 0;
if (0 == address)
return nullptr;
// TexelSizeInNibbles(format) * width * height / 16;
const unsigned int bsw = TexDecoder_GetBlockWidthInTexels(texformat);
const unsigned int bsh = TexDecoder_GetBlockHeightInTexels(texformat);
unsigned int expandedWidth = ROUND_UP(width, bsw);
unsigned int expandedHeight = ROUND_UP(height, bsh);
const unsigned int nativeW = width;
const unsigned int nativeH = height;
// Hash assigned to texcache entry (also used to generate filenames used for texture dumping and custom texture lookup)
u64 base_hash = TEXHASH_INVALID;
u64 full_hash = TEXHASH_INVALID;
u32 full_format = texformat;
const bool isPaletteTexture = (texformat == GX_TF_C4 || texformat == GX_TF_C8 || texformat == GX_TF_C14X2);
// Reject invalid tlut format.
if (isPaletteTexture && tlutfmt > GX_TL_RGB5A3)
return nullptr;
if (isPaletteTexture)
full_format = texformat | (tlutfmt << 16);
const u32 texture_size = TexDecoder_GetTextureSizeInBytes(expandedWidth, expandedHeight, texformat);
u32 additional_mips_size = 0; // not including level 0, which is texture_size
// GPUs don't like when the specified mipmap count would require more than one 1x1-sized LOD in the mipmap chain
// e.g. 64x64 with 7 LODs would have the mipmap chain 64x64,32x32,16x16,8x8,4x4,2x2,1x1,0x0, so we limit the mipmap count to 6 there
tex_levels = std::min<u32>(IntLog2(std::max(width, height)) + 1, tex_levels);
for (u32 level = 1; level != tex_levels; ++level)
{
// We still need to calculate the original size of the mips
const u32 expanded_mip_width = ROUND_UP(CalculateLevelSize(width, level), bsw);
const u32 expanded_mip_height = ROUND_UP(CalculateLevelSize(height, level), bsh);
additional_mips_size += TexDecoder_GetTextureSizeInBytes(expanded_mip_width, expanded_mip_height, texformat);
}
// If we are recording a FifoLog, keep track of what memory we read.
// FifiRecorder does it's own memory modification tracking independant of the texture hashing below.
if (g_bRecordFifoData && !from_tmem)
FifoRecorder::GetInstance().UseMemory(address, texture_size + additional_mips_size, MemoryUpdate::TEXTURE_MAP);
const u8* src_data;
if (from_tmem)
src_data = &texMem[bpmem.tex[stage / 4].texImage1[stage % 4].tmem_even * TMEM_LINE_SIZE];
else
src_data = Memory::GetPointer(address);
// TODO: This doesn't hash GB tiles for preloaded RGBA8 textures (instead, it's hashing more data from the low tmem bank than it should)
base_hash = GetHash64(src_data, texture_size, g_ActiveConfig.iSafeTextureCache_ColorSamples);
u32 palette_size = 0;
if (isPaletteTexture)
{
palette_size = TexDecoder_GetPaletteSize(texformat);
full_hash = base_hash ^ GetHash64(&texMem[tlutaddr], palette_size, g_ActiveConfig.iSafeTextureCache_ColorSamples);
}
else
{
full_hash = base_hash;
}
// Search the texture cache for textures by address
//
// Find all texture cache entries for the current texture address, and decide whether to use one of
// them, or to create a new one
//
// In most cases, the fastest way is to use only one texture cache entry for the same address. Usually,
// when a texture changes, the old version of the texture is unlikely to be used again. If there were
// new cache entries created for normal texture updates, there would be a slowdown due to a huge amount
// of unused cache entries. Also thanks to texture pooling, overwriting an existing cache entry is
// faster than creating a new one from scratch.
//
// Some games use the same address for different textures though. If the same cache entry was used in
// this case, it would be constantly overwritten, and effectively there wouldn't be any caching for
// those textures. Examples for this are Metroid Prime and Castlevania 3. Metroid Prime has multiple
// sets of fonts on each other stored in a single texture and uses the palette to make different
// characters visible or invisible. In Castlevania 3 some textures are used for 2 different things or
// at least in 2 different ways(size 1024x1024 vs 1024x256).
//
// To determine whether to use multiple cache entries or a single entry, use the following heuristic:
// If the same texture address is used several times during the same frame, assume the address is used
// for different purposes and allow creating an additional cache entry. If there's at least one entry
// that hasn't been used for the same frame, then overwrite it, in order to keep the cache as small as
// possible. If the current texture is found in the cache, use that entry.
//
// For efb copies, the entry created in CopyRenderTargetToTexture always has to be used, or else it was
// done in vain.
std::pair<TexCache::iterator, TexCache::iterator> iter_range = textures_by_address.equal_range((u64)address);
TexCache::iterator iter = iter_range.first;
TexCache::iterator oldest_entry = iter;
int temp_frameCount = 0x7fffffff;
TexCache::iterator unconverted_copy = textures_by_address.end();
while (iter != iter_range.second)
{
TCacheEntryBase* entry = iter->second;
// Do not load strided EFB copies, they are not meant to be used directly
if (entry->IsEfbCopy() && entry->native_width == nativeW && entry->native_height == nativeH &&
entry->memory_stride == entry->CacheLinesPerRow() * 32)
{
// EFB copies have slightly different rules as EFB copy formats have different
// meanings from texture formats.
if ((base_hash == entry->hash && (!isPaletteTexture || g_Config.backend_info.bSupportsPaletteConversion)) ||
IsPlayingBackFifologWithBrokenEFBCopies)
{
// TODO: We should check format/width/height/levels for EFB copies. Checking
// format is complicated because EFB copy formats don't exactly match
// texture formats. I'm not sure what effect checking width/height/levels
// would have.
if (!isPaletteTexture || !g_Config.backend_info.bSupportsPaletteConversion)
return ReturnEntry(stage, entry);
// Note that we found an unconverted EFB copy, then continue. We'll
// perform the conversion later. Currently, we only convert EFB copies to
// palette textures; we could do other conversions if it proved to be
// beneficial.
unconverted_copy = iter;
}
else
{
// Aggressively prune EFB copies: if it isn't useful here, it will probably
// never be useful again. It's theoretically possible for a game to do
// something weird where the copy could become useful in the future, but in
// practice it doesn't happen.
iter = FreeTexture(iter);
continue;
}
}
else
{
// For normal textures, all texture parameters need to match
if (entry->hash == full_hash && entry->format == full_format && entry->native_levels >= tex_levels &&
entry->native_width == nativeW && entry->native_height == nativeH)
{
entry = DoPartialTextureUpdates(iter);
return ReturnEntry(stage, entry);
}
}
// Find the texture which hasn't been used for the longest time. Count paletted
// textures as the same texture here, when the texture itself is the same. This
// improves the performance a lot in some games that use paletted textures.
// Example: Sonic the Fighters (inside Sonic Gems Collection)
// Skip EFB copies here, so they can be used for partial texture updates
if (entry->frameCount != FRAMECOUNT_INVALID && entry->frameCount < temp_frameCount &&
!entry->IsEfbCopy() && !(isPaletteTexture && entry->base_hash == base_hash))
{
temp_frameCount = entry->frameCount;
oldest_entry = iter;
}
++iter;
}
if (unconverted_copy != textures_by_address.end())
{
// Perform palette decoding.
TCacheEntryBase *entry = unconverted_copy->second;
TCacheEntryConfig config;
config.rendertarget = true;
config.width = entry->config.width;
config.height = entry->config.height;
config.layers = FramebufferManagerBase::GetEFBLayers();
TCacheEntryBase *decoded_entry = AllocateTexture(config);
decoded_entry->SetGeneralParameters(address, texture_size, full_format);
decoded_entry->SetDimensions(entry->native_width, entry->native_height, 1);
decoded_entry->SetHashes(base_hash, full_hash);
decoded_entry->frameCount = FRAMECOUNT_INVALID;
decoded_entry->is_efb_copy = false;
g_texture_cache->ConvertTexture(decoded_entry, entry, &texMem[tlutaddr], (TlutFormat)tlutfmt);
textures_by_address.emplace((u64)address, decoded_entry);
return ReturnEntry(stage, decoded_entry);
}
// Search the texture cache for normal textures by hash
//
// If the texture was fully hashed, the address does not need to match. Identical duplicate textures cause unnecessary slowdowns
// Example: Tales of Symphonia (GC) uses over 500 small textures in menus, but only around 70 different ones
if (g_ActiveConfig.iSafeTextureCache_ColorSamples == 0 ||
std::max(texture_size, palette_size) <= (u32)g_ActiveConfig.iSafeTextureCache_ColorSamples * 8)
{
iter_range = textures_by_hash.equal_range(full_hash);
iter = iter_range.first;
while (iter != iter_range.second)
{
TCacheEntryBase* entry = iter->second;
// All parameters, except the address, need to match here
if (entry->format == full_format && entry->native_levels >= tex_levels &&
entry->native_width == nativeW && entry->native_height == nativeH)
{
entry = DoPartialTextureUpdates(iter);
return ReturnEntry(stage, entry);
}
++iter;
}
}
// If at least one entry was not used for the same frame, overwrite the oldest one
if (temp_frameCount != 0x7fffffff)
{
// pool this texture and make a new one later
FreeTexture(oldest_entry);
}
std::shared_ptr<HiresTexture> hires_tex;
if (g_ActiveConfig.bHiresTextures)
{
hires_tex = HiresTexture::Search(
src_data, texture_size,
&texMem[tlutaddr], palette_size,
width, height,
texformat, use_mipmaps
);
if (hires_tex)
{
auto& l = hires_tex->m_levels[0];
if (l.width != width || l.height != height)
{
width = l.width;
height = l.height;
}
expandedWidth = l.width;
expandedHeight = l.height;
CheckTempSize(l.data_size);
memcpy(temp, l.data, l.data_size);
}
}
if (!hires_tex)
{
if (!(texformat == GX_TF_RGBA8 && from_tmem))
{
const u8* tlut = &texMem[tlutaddr];
TexDecoder_Decode(temp, src_data, expandedWidth, expandedHeight, texformat, tlut, (TlutFormat)tlutfmt);
}
else
{
u8* src_data_gb = &texMem[bpmem.tex[stage / 4].texImage2[stage % 4].tmem_odd * TMEM_LINE_SIZE];
TexDecoder_DecodeRGBA8FromTmem(temp, src_data, src_data_gb, expandedWidth, expandedHeight);
}
}
// how many levels the allocated texture shall have
const u32 texLevels = hires_tex ? (u32)hires_tex->m_levels.size() : tex_levels;
// create the entry/texture
TCacheEntryConfig config;
config.width = width;
config.height = height;
config.levels = texLevels;
TCacheEntryBase* entry = AllocateTexture(config);
GFX_DEBUGGER_PAUSE_AT(NEXT_NEW_TEXTURE, true);
iter = textures_by_address.emplace((u64)address, entry);
if (g_ActiveConfig.iSafeTextureCache_ColorSamples == 0 ||
std::max(texture_size, palette_size) <= (u32)g_ActiveConfig.iSafeTextureCache_ColorSamples * 8)
{
entry->textures_by_hash_iter = textures_by_hash.emplace(full_hash, entry);
}
entry->SetGeneralParameters(address, texture_size, full_format);
entry->SetDimensions(nativeW, nativeH, tex_levels);
entry->SetHashes(base_hash, full_hash);
entry->is_efb_copy = false;
entry->is_custom_tex = hires_tex != nullptr;
// load texture
entry->Load(width, height, expandedWidth, 0);
std::string basename = "";
if (g_ActiveConfig.bDumpTextures && !hires_tex)
{
basename = HiresTexture::GenBaseName(
src_data, texture_size,
&texMem[tlutaddr], palette_size,
width, height,
texformat, use_mipmaps,
true
);
DumpTexture(entry, basename, 0);
}
if (hires_tex)
{
for (u32 level = 1; level != texLevels; ++level)
{
auto& l = hires_tex->m_levels[level];
CheckTempSize(l.data_size);
memcpy(temp, l.data, l.data_size);
entry->Load(l.width, l.height, l.width, level);
}
}
else
{
// load mips - TODO: Loading mipmaps from tmem is untested!
src_data += texture_size;
const u8* ptr_even = nullptr;
const u8* ptr_odd = nullptr;
if (from_tmem)
{
ptr_even = &texMem[bpmem.tex[stage / 4].texImage1[stage % 4].tmem_even * TMEM_LINE_SIZE + texture_size];
ptr_odd = &texMem[bpmem.tex[stage / 4].texImage2[stage % 4].tmem_odd * TMEM_LINE_SIZE];
}
for (u32 level = 1; level != texLevels; ++level)
{
const u32 mip_width = CalculateLevelSize(width, level);
const u32 mip_height = CalculateLevelSize(height, level);
const u32 expanded_mip_width = ROUND_UP(mip_width, bsw);
const u32 expanded_mip_height = ROUND_UP(mip_height, bsh);
const u8*& mip_src_data = from_tmem
? ((level % 2) ? ptr_odd : ptr_even)
: src_data;
const u8* tlut = &texMem[tlutaddr];
TexDecoder_Decode(temp, mip_src_data, expanded_mip_width, expanded_mip_height, texformat, tlut, (TlutFormat)tlutfmt);
mip_src_data += TexDecoder_GetTextureSizeInBytes(expanded_mip_width, expanded_mip_height, texformat);
entry->Load(mip_width, mip_height, expanded_mip_width, level);
if (g_ActiveConfig.bDumpTextures)
DumpTexture(entry, basename, level);
}
}
INCSTAT(stats.numTexturesUploaded);
SETSTAT(stats.numTexturesAlive, textures_by_address.size());
entry = DoPartialTextureUpdates(iter);
return ReturnEntry(stage, entry);
}
