fabi/Xenia-Canary
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[GPU] Non-ROV f24 trunc/round, host shader modifications, cache dir

Browse Source
This commit is contained in:
Triang3l
2020-12-07 22:23:54 +03:00
parent a86609e93a
commit 9a4643d0f2
71 changed files with 3656 additions and 1633 deletions
Download Patch File Download Diff File Expand all files Collapse all files

91
src/xenia/gpu/d3d12/pipeline_cache.h
View File

@@ -27,6 +27,7 @@
#include "xenia/gpu/d3d12/d3d12_shader.h"
#include "xenia/gpu/d3d12/render_target_cache.h"
#include "xenia/gpu/dxbc_shader_translator.h"
#include "xenia/gpu/gpu_flags.h"
#include "xenia/gpu/register_file.h"
#include "xenia/gpu/xenos.h"
#include "xenia/ui/d3d12/d3d12_api.h"
@@ -43,36 +44,39 @@ class PipelineCache {
PipelineCache(D3D12CommandProcessor& command_processor,
const RegisterFile& register_file, bool bindless_resources_used,
bool edram_rov_used, uint32_t resolution_scale);
bool edram_rov_used,
flags::DepthFloat24Conversion depth_float24_conversion,
uint32_t resolution_scale);
~PipelineCache();
bool Initialize();
void Shutdown();
void ClearCache(bool shutting_down = false);
void InitializeShaderStorage(const std::filesystem::path& storage_root,
void InitializeShaderStorage(const std::filesystem::path& cache_root,
uint32_t title_id, bool blocking);
void ShutdownShaderStorage();
void EndSubmission();
bool IsCreatingPipelines();
D3D12Shader* LoadShader(xenos::ShaderType shader_type, uint32_t guest_address,
D3D12Shader* LoadShader(xenos::ShaderType shader_type,
const uint32_t* host_address, uint32_t dword_count);
// Returns the host vertex shader type for the current draw if it's valid and
// supported, or Shader::HostVertexShaderType(-1) if not.
Shader::HostVertexShaderType GetHostVertexShaderTypeIfValid() const;
// Retrieves the shader modifications for the current state, and returns
// whether they are valid.
bool GetCurrentShaderModifications(
DxbcShaderTranslator::Modification& vertex_shader_modification_out,
DxbcShaderTranslator::Modification& pixel_shader_modification_out) const;
// Translates shaders if needed, also making shader info up to date.
bool EnsureShadersTranslated(
D3D12Shader* vertex_shader, D3D12Shader* pixel_shader,
Shader::HostVertexShaderType host_vertex_shader_type);
bool EnsureShadersTranslated(D3D12Shader::D3D12Translation* vertex_shader,
D3D12Shader::D3D12Translation* pixel_shader);
bool ConfigurePipeline(
D3D12Shader* vertex_shader, D3D12Shader* pixel_shader,
D3D12Shader::D3D12Translation* vertex_shader,
D3D12Shader::D3D12Translation* pixel_shader,
xenos::PrimitiveType primitive_type, xenos::IndexFormat index_format,
bool early_z,
const RenderTargetCache::PipelineRenderTarget render_targets[5],
void** pipeline_handle_out, ID3D12RootSignature** root_signature_out);
@@ -86,13 +90,12 @@ class PipelineCache {
XEPACKEDSTRUCT(ShaderStoredHeader, {
uint64_t ucode_data_hash;
uint32_t ucode_dword_count : 16;
uint32_t ucode_dword_count : 31;
xenos::ShaderType type : 1;
Shader::HostVertexShaderType host_vertex_shader_type : 3;
reg::SQ_PROGRAM_CNTL sq_program_cntl;
static constexpr uint32_t kVersion = 0x20200405;
static constexpr uint32_t kVersion = 0x20201129;
});
// Update PipelineDescription::kVersion if any of the Pipeline* enums are
@@ -170,28 +173,28 @@ class PipelineCache {
uint64_t vertex_shader_hash;
// 0 if drawing without a pixel shader.
uint64_t pixel_shader_hash;
uint32_t vertex_shader_modification;
uint32_t pixel_shader_modification;
int32_t depth_bias;
float depth_bias_slope_scaled;
PipelineStripCutIndex strip_cut_index : 2; // 2
Shader::HostVertexShaderType host_vertex_shader_type : 3; // 5
// PipelinePrimitiveTopologyType for a vertex shader.
// xenos::TessellationMode for a domain shader.
uint32_t primitive_topology_type_or_tessellation_mode : 2; // 7
uint32_t primitive_topology_type_or_tessellation_mode : 2; // 4
// Zero for non-kVertex host_vertex_shader_type.
PipelineGeometryShader geometry_shader : 2; // 9
uint32_t fill_mode_wireframe : 1; // 10
PipelineCullMode cull_mode : 2; // 12
uint32_t front_counter_clockwise : 1; // 13
uint32_t depth_clip : 1; // 14
uint32_t rov_msaa : 1; // 15
xenos::DepthRenderTargetFormat depth_format : 1; // 16
xenos::CompareFunction depth_func : 3; // 19
uint32_t depth_write : 1; // 20
uint32_t stencil_enable : 1; // 21
uint32_t stencil_read_mask : 8; // 29
uint32_t force_early_z : 1; // 30
PipelineGeometryShader geometry_shader : 2; // 6
uint32_t fill_mode_wireframe : 1; // 7
PipelineCullMode cull_mode : 2; // 9
uint32_t front_counter_clockwise : 1; // 10
uint32_t depth_clip : 1; // 11
uint32_t rov_msaa : 1; // 12
xenos::DepthRenderTargetFormat depth_format : 1; // 13
xenos::CompareFunction depth_func : 3; // 16
uint32_t depth_write : 1; // 17
uint32_t stencil_enable : 1; // 18
uint32_t stencil_read_mask : 8; // 26
uint32_t stencil_write_mask : 8; // 8
xenos::StencilOp stencil_front_fail_op : 3; // 11
@@ -205,7 +208,7 @@ class PipelineCache {
PipelineRenderTarget render_targets[4];
static constexpr uint32_t kVersion = 0x20200405;
static constexpr uint32_t kVersion = 0x20201202;
});
XEPACKEDSTRUCT(PipelineStoredDescription, {
@@ -215,24 +218,31 @@ class PipelineCache {
struct PipelineRuntimeDescription {
ID3D12RootSignature* root_signature;
D3D12Shader* vertex_shader;
D3D12Shader* pixel_shader;
D3D12Shader::D3D12Translation* vertex_shader;
D3D12Shader::D3D12Translation* pixel_shader;
PipelineDescription description;
};
// Returns the host vertex shader type for the current draw if it's valid and
// supported, or Shader::HostVertexShaderType(-1) if not.
Shader::HostVertexShaderType GetCurrentHostVertexShaderTypeIfValid() const;
D3D12Shader* LoadShader(xenos::ShaderType shader_type,
const uint32_t* host_address, uint32_t dword_count,
uint64_t data_hash);
// Can be called from multiple threads.
bool TranslateShader(DxbcShaderTranslator& translator, D3D12Shader& shader,
bool TranslateShader(DxbcShaderTranslator& translator,
D3D12Shader::D3D12Translation& translation,
reg::SQ_PROGRAM_CNTL cntl,
IDxbcConverter* dxbc_converter = nullptr,
IDxcUtils* dxc_utils = nullptr,
IDxcCompiler* dxc_compiler = nullptr,
Shader::HostVertexShaderType host_vertex_shader_type =
Shader::HostVertexShaderType::kVertex);
IDxcCompiler* dxc_compiler = nullptr);
bool GetCurrentStateDescription(
D3D12Shader* vertex_shader, D3D12Shader* pixel_shader,
D3D12Shader::D3D12Translation* vertex_shader,
D3D12Shader::D3D12Translation* pixel_shader,
xenos::PrimitiveType primitive_type, xenos::IndexFormat index_format,
bool early_z,
const RenderTargetCache::PipelineRenderTarget render_targets[5],
PipelineRuntimeDescription& runtime_description_out);
@@ -243,6 +253,8 @@ class PipelineCache {
const RegisterFile& register_file_;
bool bindless_resources_used_;
bool edram_rov_used_;
// 20e4 depth conversion mode to use for non-ROV output.
flags::DepthFloat24Conversion depth_float24_conversion_;
uint32_t resolution_scale_;
// Reusable shader translator.
@@ -300,11 +312,14 @@ class PipelineCache {
Pipeline* current_pipeline_ = nullptr;
// Currently open shader storage path.
std::filesystem::path shader_storage_root_;
std::filesystem::path shader_storage_cache_root_;
uint32_t shader_storage_title_id_ = 0;
// Shader storage output stream, for preload in the next emulator runs.
FILE* shader_storage_file_ = nullptr;
// For only writing shaders to the currently open storage once, incremented
// when switching the storage.
uint32_t shader_storage_index_ = 0;
bool shader_storage_file_flush_needed_ = false;
// Pipeline storage output stream, for preload in the next emulator runs.