Merge pull request #2666 from yuriks/gl-cleanups

PicaShaderConfig cleanups
This commit is contained in:
Yuri Kunde Schlesner 2017-04-19 22:09:51 -07:00 committed by GitHub
commit 5d852467a2
5 changed files with 215 additions and 214 deletions

View File

@ -20,7 +20,6 @@
#include "video_core/regs_texturing.h" #include "video_core/regs_texturing.h"
#include "video_core/renderer_opengl/gl_rasterizer.h" #include "video_core/renderer_opengl/gl_rasterizer.h"
#include "video_core/renderer_opengl/gl_shader_gen.h" #include "video_core/renderer_opengl/gl_shader_gen.h"
#include "video_core/renderer_opengl/gl_shader_util.h"
#include "video_core/renderer_opengl/pica_to_gl.h" #include "video_core/renderer_opengl/pica_to_gl.h"
#include "video_core/renderer_opengl/renderer_opengl.h" #include "video_core/renderer_opengl/renderer_opengl.h"
@ -1005,7 +1004,7 @@ void RasterizerOpenGL::SamplerInfo::SyncWithConfig(
} }
void RasterizerOpenGL::SetShader() { void RasterizerOpenGL::SetShader() {
PicaShaderConfig config = PicaShaderConfig::CurrentConfig(); auto config = GLShader::PicaShaderConfig::BuildFromRegs(Pica::g_state.regs);
std::unique_ptr<PicaShader> shader = std::make_unique<PicaShader>(); std::unique_ptr<PicaShader> shader = std::make_unique<PicaShader>();
// Find (or generate) the GLSL shader for the current TEV state // Find (or generate) the GLSL shader for the current TEV state

View File

@ -25,210 +25,13 @@
#include "video_core/regs_texturing.h" #include "video_core/regs_texturing.h"
#include "video_core/renderer_opengl/gl_rasterizer_cache.h" #include "video_core/renderer_opengl/gl_rasterizer_cache.h"
#include "video_core/renderer_opengl/gl_resource_manager.h" #include "video_core/renderer_opengl/gl_resource_manager.h"
#include "video_core/renderer_opengl/gl_shader_gen.h"
#include "video_core/renderer_opengl/gl_state.h" #include "video_core/renderer_opengl/gl_state.h"
#include "video_core/renderer_opengl/pica_to_gl.h" #include "video_core/renderer_opengl/pica_to_gl.h"
#include "video_core/shader/shader.h" #include "video_core/shader/shader.h"
struct ScreenInfo; struct ScreenInfo;
/**
* This struct contains all state used to generate the GLSL shader program that emulates the current
* Pica register configuration. This struct is used as a cache key for generated GLSL shader
* programs. The functions in gl_shader_gen.cpp should retrieve state from this struct only, not by
* directly accessing Pica registers. This should reduce the risk of bugs in shader generation where
* Pica state is not being captured in the shader cache key, thereby resulting in (what should be)
* two separate shaders sharing the same key.
*
* We use a union because "implicitly-defined copy/move constructor for a union X copies the object
* representation of X." and "implicitly-defined copy assignment operator for a union X copies the
* object representation (3.9) of X." = Bytewise copy instead of memberwise copy. This is important
* because the padding bytes are included in the hash and comparison between objects.
*/
union PicaShaderConfig {
/// Construct a PicaShaderConfig with the current Pica register configuration.
static PicaShaderConfig CurrentConfig() {
PicaShaderConfig res;
auto& state = res.state;
std::memset(&state, 0, sizeof(PicaShaderConfig::State));
const auto& regs = Pica::g_state.regs;
state.scissor_test_mode = regs.rasterizer.scissor_test.mode;
state.depthmap_enable = regs.rasterizer.depthmap_enable;
state.alpha_test_func = regs.framebuffer.output_merger.alpha_test.enable
? regs.framebuffer.output_merger.alpha_test.func.Value()
: Pica::FramebufferRegs::CompareFunc::Always;
state.texture0_type = regs.texturing.texture0.type;
// Copy relevant tev stages fields.
// We don't sync const_color here because of the high variance, it is a
// shader uniform instead.
const auto& tev_stages = regs.texturing.GetTevStages();
DEBUG_ASSERT(state.tev_stages.size() == tev_stages.size());
for (size_t i = 0; i < tev_stages.size(); i++) {
const auto& tev_stage = tev_stages[i];
state.tev_stages[i].sources_raw = tev_stage.sources_raw;
state.tev_stages[i].modifiers_raw = tev_stage.modifiers_raw;
state.tev_stages[i].ops_raw = tev_stage.ops_raw;
state.tev_stages[i].scales_raw = tev_stage.scales_raw;
}
state.fog_mode = regs.texturing.fog_mode;
state.fog_flip = regs.texturing.fog_flip != 0;
state.combiner_buffer_input =
regs.texturing.tev_combiner_buffer_input.update_mask_rgb.Value() |
regs.texturing.tev_combiner_buffer_input.update_mask_a.Value() << 4;
// Fragment lighting
state.lighting.enable = !regs.lighting.disable;
state.lighting.src_num = regs.lighting.max_light_index + 1;
for (unsigned light_index = 0; light_index < state.lighting.src_num; ++light_index) {
unsigned num = regs.lighting.light_enable.GetNum(light_index);
const auto& light = regs.lighting.light[num];
state.lighting.light[light_index].num = num;
state.lighting.light[light_index].directional = light.config.directional != 0;
state.lighting.light[light_index].two_sided_diffuse =
light.config.two_sided_diffuse != 0;
state.lighting.light[light_index].dist_atten_enable =
!regs.lighting.IsDistAttenDisabled(num);
}
state.lighting.lut_d0.enable = regs.lighting.config1.disable_lut_d0 == 0;
state.lighting.lut_d0.abs_input = regs.lighting.abs_lut_input.disable_d0 == 0;
state.lighting.lut_d0.type = regs.lighting.lut_input.d0.Value();
state.lighting.lut_d0.scale = regs.lighting.lut_scale.GetScale(regs.lighting.lut_scale.d0);
state.lighting.lut_d1.enable = regs.lighting.config1.disable_lut_d1 == 0;
state.lighting.lut_d1.abs_input = regs.lighting.abs_lut_input.disable_d1 == 0;
state.lighting.lut_d1.type = regs.lighting.lut_input.d1.Value();
state.lighting.lut_d1.scale = regs.lighting.lut_scale.GetScale(regs.lighting.lut_scale.d1);
state.lighting.lut_fr.enable = regs.lighting.config1.disable_lut_fr == 0;
state.lighting.lut_fr.abs_input = regs.lighting.abs_lut_input.disable_fr == 0;
state.lighting.lut_fr.type = regs.lighting.lut_input.fr.Value();
state.lighting.lut_fr.scale = regs.lighting.lut_scale.GetScale(regs.lighting.lut_scale.fr);
state.lighting.lut_rr.enable = regs.lighting.config1.disable_lut_rr == 0;
state.lighting.lut_rr.abs_input = regs.lighting.abs_lut_input.disable_rr == 0;
state.lighting.lut_rr.type = regs.lighting.lut_input.rr.Value();
state.lighting.lut_rr.scale = regs.lighting.lut_scale.GetScale(regs.lighting.lut_scale.rr);
state.lighting.lut_rg.enable = regs.lighting.config1.disable_lut_rg == 0;
state.lighting.lut_rg.abs_input = regs.lighting.abs_lut_input.disable_rg == 0;
state.lighting.lut_rg.type = regs.lighting.lut_input.rg.Value();
state.lighting.lut_rg.scale = regs.lighting.lut_scale.GetScale(regs.lighting.lut_scale.rg);
state.lighting.lut_rb.enable = regs.lighting.config1.disable_lut_rb == 0;
state.lighting.lut_rb.abs_input = regs.lighting.abs_lut_input.disable_rb == 0;
state.lighting.lut_rb.type = regs.lighting.lut_input.rb.Value();
state.lighting.lut_rb.scale = regs.lighting.lut_scale.GetScale(regs.lighting.lut_scale.rb);
state.lighting.config = regs.lighting.config0.config;
state.lighting.fresnel_selector = regs.lighting.config0.fresnel_selector;
state.lighting.bump_mode = regs.lighting.config0.bump_mode;
state.lighting.bump_selector = regs.lighting.config0.bump_selector;
state.lighting.bump_renorm = regs.lighting.config0.disable_bump_renorm == 0;
state.lighting.clamp_highlights = regs.lighting.config0.clamp_highlights != 0;
return res;
}
bool TevStageUpdatesCombinerBufferColor(unsigned stage_index) const {
return (stage_index < 4) && (state.combiner_buffer_input & (1 << stage_index));
}
bool TevStageUpdatesCombinerBufferAlpha(unsigned stage_index) const {
return (stage_index < 4) && ((state.combiner_buffer_input >> 4) & (1 << stage_index));
}
bool operator==(const PicaShaderConfig& o) const {
return std::memcmp(&state, &o.state, sizeof(PicaShaderConfig::State)) == 0;
};
// NOTE: MSVC15 (Update 2) doesn't think `delete`'d constructors and operators are TC.
// This makes BitField not TC when used in a union or struct so we have to resort
// to this ugly hack.
// Once that bug is fixed we can use Pica::Regs::TevStageConfig here.
// Doesn't include const_color because we don't sync it, see comment in CurrentConfig()
struct TevStageConfigRaw {
u32 sources_raw;
u32 modifiers_raw;
u32 ops_raw;
u32 scales_raw;
explicit operator Pica::TexturingRegs::TevStageConfig() const noexcept {
Pica::TexturingRegs::TevStageConfig stage;
stage.sources_raw = sources_raw;
stage.modifiers_raw = modifiers_raw;
stage.ops_raw = ops_raw;
stage.const_color = 0;
stage.scales_raw = scales_raw;
return stage;
}
};
struct State {
Pica::FramebufferRegs::CompareFunc alpha_test_func;
Pica::RasterizerRegs::ScissorMode scissor_test_mode;
Pica::TexturingRegs::TextureConfig::TextureType texture0_type;
std::array<TevStageConfigRaw, 6> tev_stages;
u8 combiner_buffer_input;
Pica::RasterizerRegs::DepthBuffering depthmap_enable;
Pica::TexturingRegs::FogMode fog_mode;
bool fog_flip;
struct {
struct {
unsigned num;
bool directional;
bool two_sided_diffuse;
bool dist_atten_enable;
} light[8];
bool enable;
unsigned src_num;
Pica::LightingRegs::LightingBumpMode bump_mode;
unsigned bump_selector;
bool bump_renorm;
bool clamp_highlights;
Pica::LightingRegs::LightingConfig config;
Pica::LightingRegs::LightingFresnelSelector fresnel_selector;
struct {
bool enable;
bool abs_input;
Pica::LightingRegs::LightingLutInput type;
float scale;
} lut_d0, lut_d1, lut_fr, lut_rr, lut_rg, lut_rb;
} lighting;
} state;
};
#if (__GNUC__ >= 5) || defined(__clang__) || defined(_MSC_VER)
static_assert(std::is_trivially_copyable<PicaShaderConfig::State>::value,
"PicaShaderConfig::State must be trivially copyable");
#endif
namespace std {
template <>
struct hash<PicaShaderConfig> {
size_t operator()(const PicaShaderConfig& k) const {
return Common::ComputeHash64(&k.state, sizeof(PicaShaderConfig::State));
}
};
} // namespace std
class RasterizerOpenGL : public VideoCore::RasterizerInterface { class RasterizerOpenGL : public VideoCore::RasterizerInterface {
public: public:
RasterizerOpenGL(); RasterizerOpenGL();
@ -437,7 +240,7 @@ private:
std::vector<HardwareVertex> vertex_batch; std::vector<HardwareVertex> vertex_batch;
std::unordered_map<PicaShaderConfig, std::unique_ptr<PicaShader>> shader_cache; std::unordered_map<GLShader::PicaShaderConfig, std::unique_ptr<PicaShader>> shader_cache;
const PicaShader* current_shader = nullptr; const PicaShader* current_shader = nullptr;
bool shader_dirty; bool shader_dirty;

View File

@ -4,6 +4,7 @@
#include <array> #include <array>
#include <cstddef> #include <cstddef>
#include <cstring>
#include "common/assert.h" #include "common/assert.h"
#include "common/bit_field.h" #include "common/bit_field.h"
#include "common/logging/log.h" #include "common/logging/log.h"
@ -23,6 +24,97 @@ using TevStageConfig = TexturingRegs::TevStageConfig;
namespace GLShader { namespace GLShader {
PicaShaderConfig PicaShaderConfig::BuildFromRegs(const Pica::Regs& regs) {
PicaShaderConfig res;
auto& state = res.state;
std::memset(&state, 0, sizeof(PicaShaderConfig::State));
state.scissor_test_mode = regs.rasterizer.scissor_test.mode;
state.depthmap_enable = regs.rasterizer.depthmap_enable;
state.alpha_test_func = regs.framebuffer.output_merger.alpha_test.enable
? regs.framebuffer.output_merger.alpha_test.func.Value()
: Pica::FramebufferRegs::CompareFunc::Always;
state.texture0_type = regs.texturing.texture0.type;
// Copy relevant tev stages fields.
// We don't sync const_color here because of the high variance, it is a
// shader uniform instead.
const auto& tev_stages = regs.texturing.GetTevStages();
DEBUG_ASSERT(state.tev_stages.size() == tev_stages.size());
for (size_t i = 0; i < tev_stages.size(); i++) {
const auto& tev_stage = tev_stages[i];
state.tev_stages[i].sources_raw = tev_stage.sources_raw;
state.tev_stages[i].modifiers_raw = tev_stage.modifiers_raw;
state.tev_stages[i].ops_raw = tev_stage.ops_raw;
state.tev_stages[i].scales_raw = tev_stage.scales_raw;
}
state.fog_mode = regs.texturing.fog_mode;
state.fog_flip = regs.texturing.fog_flip != 0;
state.combiner_buffer_input = regs.texturing.tev_combiner_buffer_input.update_mask_rgb.Value() |
regs.texturing.tev_combiner_buffer_input.update_mask_a.Value()
<< 4;
// Fragment lighting
state.lighting.enable = !regs.lighting.disable;
state.lighting.src_num = regs.lighting.max_light_index + 1;
for (unsigned light_index = 0; light_index < state.lighting.src_num; ++light_index) {
unsigned num = regs.lighting.light_enable.GetNum(light_index);
const auto& light = regs.lighting.light[num];
state.lighting.light[light_index].num = num;
state.lighting.light[light_index].directional = light.config.directional != 0;
state.lighting.light[light_index].two_sided_diffuse = light.config.two_sided_diffuse != 0;
state.lighting.light[light_index].dist_atten_enable =
!regs.lighting.IsDistAttenDisabled(num);
}
state.lighting.lut_d0.enable = regs.lighting.config1.disable_lut_d0 == 0;
state.lighting.lut_d0.abs_input = regs.lighting.abs_lut_input.disable_d0 == 0;
state.lighting.lut_d0.type = regs.lighting.lut_input.d0.Value();
state.lighting.lut_d0.scale = regs.lighting.lut_scale.GetScale(regs.lighting.lut_scale.d0);
state.lighting.lut_d1.enable = regs.lighting.config1.disable_lut_d1 == 0;
state.lighting.lut_d1.abs_input = regs.lighting.abs_lut_input.disable_d1 == 0;
state.lighting.lut_d1.type = regs.lighting.lut_input.d1.Value();
state.lighting.lut_d1.scale = regs.lighting.lut_scale.GetScale(regs.lighting.lut_scale.d1);
state.lighting.lut_fr.enable = regs.lighting.config1.disable_lut_fr == 0;
state.lighting.lut_fr.abs_input = regs.lighting.abs_lut_input.disable_fr == 0;
state.lighting.lut_fr.type = regs.lighting.lut_input.fr.Value();
state.lighting.lut_fr.scale = regs.lighting.lut_scale.GetScale(regs.lighting.lut_scale.fr);
state.lighting.lut_rr.enable = regs.lighting.config1.disable_lut_rr == 0;
state.lighting.lut_rr.abs_input = regs.lighting.abs_lut_input.disable_rr == 0;
state.lighting.lut_rr.type = regs.lighting.lut_input.rr.Value();
state.lighting.lut_rr.scale = regs.lighting.lut_scale.GetScale(regs.lighting.lut_scale.rr);
state.lighting.lut_rg.enable = regs.lighting.config1.disable_lut_rg == 0;
state.lighting.lut_rg.abs_input = regs.lighting.abs_lut_input.disable_rg == 0;
state.lighting.lut_rg.type = regs.lighting.lut_input.rg.Value();
state.lighting.lut_rg.scale = regs.lighting.lut_scale.GetScale(regs.lighting.lut_scale.rg);
state.lighting.lut_rb.enable = regs.lighting.config1.disable_lut_rb == 0;
state.lighting.lut_rb.abs_input = regs.lighting.abs_lut_input.disable_rb == 0;
state.lighting.lut_rb.type = regs.lighting.lut_input.rb.Value();
state.lighting.lut_rb.scale = regs.lighting.lut_scale.GetScale(regs.lighting.lut_scale.rb);
state.lighting.config = regs.lighting.config0.config;
state.lighting.fresnel_selector = regs.lighting.config0.fresnel_selector;
state.lighting.bump_mode = regs.lighting.config0.bump_mode;
state.lighting.bump_selector = regs.lighting.config0.bump_selector;
state.lighting.bump_renorm = regs.lighting.config0.disable_bump_renorm == 0;
state.lighting.clamp_highlights = regs.lighting.config0.clamp_highlights != 0;
return res;
}
/// Detects if a TEV stage is configured to be skipped (to avoid generating unnecessary code) /// Detects if a TEV stage is configured to be skipped (to avoid generating unnecessary code)
static bool IsPassThroughTevStage(const TevStageConfig& stage) { static bool IsPassThroughTevStage(const TevStageConfig& stage) {
return (stage.color_op == TevStageConfig::Operation::Replace && return (stage.color_op == TevStageConfig::Operation::Replace &&

View File

@ -4,12 +4,121 @@
#pragma once #pragma once
#include <array>
#include <cstring>
#include <functional>
#include <string> #include <string>
#include <type_traits>
union PicaShaderConfig; #include "video_core/regs.h"
namespace GLShader { namespace GLShader {
enum Attributes {
ATTRIBUTE_POSITION,
ATTRIBUTE_COLOR,
ATTRIBUTE_TEXCOORD0,
ATTRIBUTE_TEXCOORD1,
ATTRIBUTE_TEXCOORD2,
ATTRIBUTE_TEXCOORD0_W,
ATTRIBUTE_NORMQUAT,
ATTRIBUTE_VIEW,
};
/**
* This struct contains all state used to generate the GLSL shader program that emulates the current
* Pica register configuration. This struct is used as a cache key for generated GLSL shader
* programs. The functions in gl_shader_gen.cpp should retrieve state from this struct only, not by
* directly accessing Pica registers. This should reduce the risk of bugs in shader generation where
* Pica state is not being captured in the shader cache key, thereby resulting in (what should be)
* two separate shaders sharing the same key.
*
* We use a union because "implicitly-defined copy/move constructor for a union X copies the object
* representation of X." and "implicitly-defined copy assignment operator for a union X copies the
* object representation (3.9) of X." = Bytewise copy instead of memberwise copy. This is important
* because the padding bytes are included in the hash and comparison between objects.
*/
union PicaShaderConfig {
/// Construct a PicaShaderConfig with the given Pica register configuration.
static PicaShaderConfig BuildFromRegs(const Pica::Regs& regs);
bool TevStageUpdatesCombinerBufferColor(unsigned stage_index) const {
return (stage_index < 4) && (state.combiner_buffer_input & (1 << stage_index));
}
bool TevStageUpdatesCombinerBufferAlpha(unsigned stage_index) const {
return (stage_index < 4) && ((state.combiner_buffer_input >> 4) & (1 << stage_index));
}
bool operator==(const PicaShaderConfig& o) const {
return std::memcmp(&state, &o.state, sizeof(PicaShaderConfig::State)) == 0;
};
// NOTE: MSVC15 (Update 2) doesn't think `delete`'d constructors and operators are TC.
// This makes BitField not TC when used in a union or struct so we have to resort
// to this ugly hack.
// Once that bug is fixed we can use Pica::Regs::TevStageConfig here.
// Doesn't include const_color because we don't sync it, see comment in BuildFromRegs()
struct TevStageConfigRaw {
u32 sources_raw;
u32 modifiers_raw;
u32 ops_raw;
u32 scales_raw;
explicit operator Pica::TexturingRegs::TevStageConfig() const noexcept {
Pica::TexturingRegs::TevStageConfig stage;
stage.sources_raw = sources_raw;
stage.modifiers_raw = modifiers_raw;
stage.ops_raw = ops_raw;
stage.const_color = 0;
stage.scales_raw = scales_raw;
return stage;
}
};
struct State {
Pica::FramebufferRegs::CompareFunc alpha_test_func;
Pica::RasterizerRegs::ScissorMode scissor_test_mode;
Pica::TexturingRegs::TextureConfig::TextureType texture0_type;
std::array<TevStageConfigRaw, 6> tev_stages;
u8 combiner_buffer_input;
Pica::RasterizerRegs::DepthBuffering depthmap_enable;
Pica::TexturingRegs::FogMode fog_mode;
bool fog_flip;
struct {
struct {
unsigned num;
bool directional;
bool two_sided_diffuse;
bool dist_atten_enable;
} light[8];
bool enable;
unsigned src_num;
Pica::LightingRegs::LightingBumpMode bump_mode;
unsigned bump_selector;
bool bump_renorm;
bool clamp_highlights;
Pica::LightingRegs::LightingConfig config;
Pica::LightingRegs::LightingFresnelSelector fresnel_selector;
struct {
bool enable;
bool abs_input;
Pica::LightingRegs::LightingLutInput type;
float scale;
} lut_d0, lut_d1, lut_fr, lut_rr, lut_rg, lut_rb;
} lighting;
} state;
};
#if (__GNUC__ >= 5) || defined(__clang__) || defined(_MSC_VER)
static_assert(std::is_trivially_copyable<PicaShaderConfig::State>::value,
"PicaShaderConfig::State must be trivially copyable");
#endif
/** /**
* Generates the GLSL vertex shader program source code for the current Pica state * Generates the GLSL vertex shader program source code for the current Pica state
* @returns String of the shader source code * @returns String of the shader source code
@ -25,3 +134,12 @@ std::string GenerateVertexShader();
std::string GenerateFragmentShader(const PicaShaderConfig& config); std::string GenerateFragmentShader(const PicaShaderConfig& config);
} // namespace GLShader } // namespace GLShader
namespace std {
template <>
struct hash<GLShader::PicaShaderConfig> {
size_t operator()(const GLShader::PicaShaderConfig& k) const {
return Common::ComputeHash64(&k.state, sizeof(GLShader::PicaShaderConfig::State));
}
};
} // namespace std

View File

@ -8,17 +8,6 @@
namespace GLShader { namespace GLShader {
enum Attributes {
ATTRIBUTE_POSITION,
ATTRIBUTE_COLOR,
ATTRIBUTE_TEXCOORD0,
ATTRIBUTE_TEXCOORD1,
ATTRIBUTE_TEXCOORD2,
ATTRIBUTE_TEXCOORD0_W,
ATTRIBUTE_NORMQUAT,
ATTRIBUTE_VIEW,
};
/** /**
* Utility function to create and compile an OpenGL GLSL shader program (vertex + fragment shader) * Utility function to create and compile an OpenGL GLSL shader program (vertex + fragment shader)
* @param vertex_shader String of the GLSL vertex shader program * @param vertex_shader String of the GLSL vertex shader program