ShuriZma
/
xenia-canary
mirror of https://github.com/xenia-canary/xenia-canary.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Caching pipelines.

This commit is contained in:
Ben Vanik 2014-12-26 23:28:50 -08:00
parent 14ee211ea9
commit ea403b2164
2 changed files with 69 additions and 34 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

91
src/xenia/gpu/gl4/command_processor.cc
View File

@ -38,6 +38,12 @@ const GLuint kAnyTarget = UINT_MAX;
// frame. // frame.
const size_t kScratchBufferCapacity = 64 * 1024 * 1024; const size_t kScratchBufferCapacity = 64 * 1024 * 1024;
CommandProcessor::CachedPipeline::CachedPipeline() = default;
CommandProcessor::CachedPipeline::~CachedPipeline() {
glDeleteProgramPipelines(1, &handles.default_pipeline);
}
CommandProcessor::CommandProcessor(GL4GraphicsSystem* graphics_system) CommandProcessor::CommandProcessor(GL4GraphicsSystem* graphics_system)
: memory_(graphics_system->memory()), : memory_(graphics_system->memory()),
membase_(graphics_system->memory()->membase()), membase_(graphics_system->memory()->membase()),
@ -91,10 +97,12 @@ void CommandProcessor::Shutdown() {
worker_running_ = false; worker_running_ = false;
SetEvent(write_ptr_index_event_); SetEvent(write_ptr_index_event_);
worker_thread_.join(); worker_thread_.join();
context_.reset();
all_pipelines_.clear();
all_shaders_.clear(); all_shaders_.clear();
shader_cache_.clear(); shader_cache_.clear();
context_.reset();
} }
void CommandProcessor::WorkerMain() { void CommandProcessor::WorkerMain() {
@ -1271,10 +1279,10 @@ bool CommandProcessor::UpdateState(DrawCommand* draw_command) {
}; };
}; };
struct UniformDataBlock { struct UniformDataBlock {
float4 window_offset; // tx,ty,rt_w,rt_h float4 window_offset; // tx,ty,rt_w,rt_h
float4 window_scissor; // x0,y0,x1,y1 float4 window_scissor; // x0,y0,x1,y1
float4 viewport_offset; // tx,ty,tz,? float4 viewport_offset; // tx,ty,tz,?
float4 viewport_scale; // sx,sy,sz,? float4 viewport_scale; // sx,sy,sz,?
// TODO(benvanik): vertex format xyzw? // TODO(benvanik): vertex format xyzw?
float4 alpha_test; // alpha test enable, func, ref, ? float4 alpha_test; // alpha test enable, func, ref, ?
@ -1697,38 +1705,53 @@ bool CommandProcessor::UpdateShaders(DrawCommand* draw_command) {
GLuint geometry_program = 0; GLuint geometry_program = 0;
GLuint fragment_program = active_pixel_shader_->program(); GLuint fragment_program = active_pixel_shader_->program();
GLuint pipeline; uint64_t key = (uint64_t(vertex_program) << 32) | fragment_program;
glCreateProgramPipelines(1, &pipeline); CachedPipeline* cached_pipeline = nullptr;
glUseProgramStages(pipeline, GL_VERTEX_SHADER_BIT, vertex_program); auto it = cached_pipelines_.find(key);
glUseProgramStages(pipeline, GL_GEOMETRY_SHADER_BIT, geometry_program); if (it == cached_pipelines_.end()) {
glUseProgramStages(pipeline, GL_FRAGMENT_SHADER_BIT, fragment_program); // Existing pipeline for these programs not found - create it.
auto new_pipeline = std::make_unique<CachedPipeline>();
new_pipeline->vertex_program = vertex_program;
new_pipeline->fragment_program = fragment_program;
new_pipeline->handles.default_pipeline = 0;
cached_pipeline = new_pipeline.get();
all_pipelines_.emplace_back(std::move(new_pipeline));
cached_pipelines_.insert({key, cached_pipeline});
} else {
// Found a pipeline container - it may or may not have what we want.
cached_pipeline = it->second;
}
if (!cached_pipeline->handles.default_pipeline) {
GLuint pipeline;
glCreateProgramPipelines(1, &pipeline);
glUseProgramStages(pipeline, GL_VERTEX_SHADER_BIT, vertex_program);
glUseProgramStages(pipeline, GL_GEOMETRY_SHADER_BIT, geometry_program);
glUseProgramStages(pipeline, GL_FRAGMENT_SHADER_BIT, fragment_program);
// HACK: layout(location=0) on a bindless uniform crashes nvidia driver. // HACK: layout(location=0) on a bindless uniform crashes nvidia driver.
GLint vertex_state_loc = glGetUniformLocation(vertex_program, "state"); GLint vertex_state_loc = glGetUniformLocation(vertex_program, "state");
assert_true(vertex_state_loc == -1 || vertex_state_loc == 0); assert_true(vertex_state_loc == 0);
GLint geometry_state_loc = GLint geometry_state_loc =
geometry_program ? glGetUniformLocation(geometry_program, "state") : -1; geometry_program ? glGetUniformLocation(geometry_program, "state") : -1;
assert_true(geometry_state_loc == -1 || geometry_state_loc == 0); assert_true(geometry_state_loc == -1 || geometry_state_loc == 0);
GLint fragment_state_loc = glGetUniformLocation(fragment_program, "state"); GLint fragment_state_loc = glGetUniformLocation(fragment_program, "state");
assert_true(fragment_state_loc == -1 || fragment_state_loc == 0); assert_true(fragment_state_loc == -1 || fragment_state_loc == 0);
cached_pipeline->handles.default_pipeline = pipeline;
}
// TODO(benvanik): do we need to do this for all stages if the locations // TODO(benvanik): do we need to do this for all stages if the locations
// match? // match?
if (vertex_state_loc != -1) { glProgramUniformHandleui64ARB(vertex_program, 0, cmd.state_data_gpu_ptr);
glProgramUniformHandleui64ARB(vertex_program, vertex_state_loc, /*if (geometry_program && geometry_state_loc != -1) {
glProgramUniformHandleui64ARB(geometry_program, 0, cmd.state_data_gpu_ptr);
}*/
/*if (fragment_state_loc != -1) {
glProgramUniformHandleui64ARB(fragment_program, 0,
cmd.state_data_gpu_ptr); cmd.state_data_gpu_ptr);
} }*/
if (geometry_program && geometry_state_loc != -1) {
glProgramUniformHandleui64ARB(geometry_program, geometry_state_loc,
cmd.state_data_gpu_ptr);
}
if (fragment_state_loc != -1) {
glProgramUniformHandleui64ARB(fragment_program, fragment_state_loc,
cmd.state_data_gpu_ptr);
}
glBindProgramPipeline(pipeline); glBindProgramPipeline(cached_pipeline->handles.default_pipeline);
// glDeleteProgramPipelines(1, &pipeline);
return true; return true;
} }
@ -2051,10 +2074,10 @@ bool CommandProcessor::IssueCopy(DrawCommand* draw_command) {
// glBindBuffer(GL_READ_FRAMEBUFFER, framebuffer) // glBindBuffer(GL_READ_FRAMEBUFFER, framebuffer)
glNamedFramebufferReadBuffer(source_framebuffer->framebuffer, glNamedFramebufferReadBuffer(source_framebuffer->framebuffer,
GL_COLOR_ATTACHMENT0 + copy_src_select); GL_COLOR_ATTACHMENT0 + copy_src_select);
//glReadPixels(x, y, w, h, read_format, read_type, ptr); // glReadPixels(x, y, w, h, read_format, read_type, ptr);
} else { } else {
// Source from the bound depth/stencil target. // Source from the bound depth/stencil target.
//glReadPixels(x, y, w, h, GL_DEPTH_STENCIL, read_type, ptr); // glReadPixels(x, y, w, h, GL_DEPTH_STENCIL, read_type, ptr);
} }
break; break;
case CopyCommand::kRaw: case CopyCommand::kRaw:
@ -2095,7 +2118,7 @@ bool CommandProcessor::IssueCopy(DrawCommand* draw_command) {
glClearNamedFramebufferfi(source_framebuffer->framebuffer, GL_DEPTH_STENCIL, glClearNamedFramebufferfi(source_framebuffer->framebuffer, GL_DEPTH_STENCIL,
depth.float_value, stencil); depth.float_value, stencil);
} }
return true; return true;
} }

12
src/xenia/gpu/gl4/command_processor.h
View File

@ -112,6 +112,16 @@ class CommandProcessor {
xenos::DepthRenderTargetFormat format; xenos::DepthRenderTargetFormat format;
GLuint texture; GLuint texture;
}; };
struct CachedPipeline {
CachedPipeline();
~CachedPipeline();
GLuint vertex_program;
GLuint fragment_program;
struct {
GLuint default_pipeline;
// TODO(benvanik): others with geometry shaders.
} handles;
};
void WorkerMain(); void WorkerMain();
bool SetupGL(); bool SetupGL();
@ -239,6 +249,8 @@ class CommandProcessor {
std::vector<CachedFramebuffer> cached_framebuffers_; std::vector<CachedFramebuffer> cached_framebuffers_;
std::vector<CachedColorRenderTarget> cached_color_render_targets_; std::vector<CachedColorRenderTarget> cached_color_render_targets_;
std::vector<CachedDepthRenderTarget> cached_depth_render_targets_; std::vector<CachedDepthRenderTarget> cached_depth_render_targets_;
std::vector<std::unique_ptr<CachedPipeline>> all_pipelines_;
std::unordered_map<uint64_t, CachedPipeline*> cached_pipelines_;
CircularBuffer scratch_buffer_; CircularBuffer scratch_buffer_;