summaryrefslogtreecommitdiffstats
path: root/src/video_core
diff options
context:
space:
mode:
Diffstat (limited to 'src/video_core')
-rw-r--r--src/video_core/engines/shader_bytecode.h3
-rw-r--r--src/video_core/renderer_opengl/gl_rasterizer.cpp116
-rw-r--r--src/video_core/renderer_opengl/gl_rasterizer.h6
-rw-r--r--src/video_core/renderer_opengl/gl_shader_cache.cpp45
-rw-r--r--src/video_core/renderer_opengl/gl_shader_cache.h2
-rw-r--r--src/video_core/renderer_opengl/gl_shader_decompiler.cpp308
-rw-r--r--src/video_core/renderer_opengl/gl_shader_gen.cpp7
-rw-r--r--src/video_core/renderer_opengl/gl_shader_gen.h3
8 files changed, 327 insertions, 163 deletions
diff --git a/src/video_core/engines/shader_bytecode.h b/src/video_core/engines/shader_bytecode.h
index b9faaf8e0..5ea094e64 100644
--- a/src/video_core/engines/shader_bytecode.h
+++ b/src/video_core/engines/shader_bytecode.h
@@ -1049,6 +1049,7 @@ union Instruction {
BitField<49, 1, u64> nodep_flag;
BitField<50, 3, u64> component_mask_selector;
BitField<53, 4, u64> texture_info;
+ BitField<60, 1, u64> fp32_flag;
TextureType GetTextureType() const {
// The TEXS instruction has a weird encoding for the texture type.
@@ -1549,7 +1550,7 @@ private:
INST("1110111011011---", Id::STG, Type::Memory, "STG"),
INST("110000----111---", Id::TEX, Type::Memory, "TEX"),
INST("1101111101001---", Id::TXQ, Type::Memory, "TXQ"),
- INST("1101100---------", Id::TEXS, Type::Memory, "TEXS"),
+ INST("1101-00---------", Id::TEXS, Type::Memory, "TEXS"),
INST("1101101---------", Id::TLDS, Type::Memory, "TLDS"),
INST("110010----111---", Id::TLD4, Type::Memory, "TLD4"),
INST("1101111100------", Id::TLD4S, Type::Memory, "TLD4S"),
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp
index 9e93bd609..2b29fc45f 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.cpp
+++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp
@@ -79,6 +79,26 @@ struct DrawParameters {
}
};
+struct FramebufferCacheKey {
+ bool is_single_buffer = false;
+ bool stencil_enable = false;
+
+ std::array<GLenum, Maxwell::NumRenderTargets> color_attachments{};
+ std::array<GLuint, Tegra::Engines::Maxwell3D::Regs::NumRenderTargets> colors{};
+ u32 colors_count = 0;
+
+ GLuint zeta = 0;
+
+ auto Tie() const {
+ return std::tie(is_single_buffer, stencil_enable, color_attachments, colors, colors_count,
+ zeta);
+ }
+
+ bool operator<(const FramebufferCacheKey& rhs) const {
+ return Tie() < rhs.Tie();
+ }
+};
+
RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& window, ScreenInfo& info)
: res_cache{*this}, shader_cache{*this}, emu_window{window}, screen_info{info},
buffer_cache(*this, STREAM_BUFFER_SIZE) {
@@ -90,9 +110,6 @@ RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& window, ScreenInfo
OpenGLState::ApplyDefaultState();
- // Create render framebuffer
- framebuffer.Create();
-
shader_program_manager = std::make_unique<GLShader::ProgramManager>();
state.draw.shader_program = 0;
state.Apply();
@@ -361,6 +378,44 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) {
SyncClipEnabled(clip_distances);
}
+void RasterizerOpenGL::SetupCachedFramebuffer(const FramebufferCacheKey& fbkey,
+ OpenGLState& current_state) {
+ const auto [entry, is_cache_miss] = framebuffer_cache.try_emplace(fbkey);
+ auto& framebuffer = entry->second;
+
+ if (is_cache_miss)
+ framebuffer.Create();
+
+ current_state.draw.draw_framebuffer = framebuffer.handle;
+ current_state.ApplyFramebufferState();
+
+ if (!is_cache_miss)
+ return;
+
+ if (fbkey.is_single_buffer) {
+ if (fbkey.color_attachments[0] != GL_NONE) {
+ glFramebufferTexture(GL_DRAW_FRAMEBUFFER, fbkey.color_attachments[0], fbkey.colors[0],
+ 0);
+ }
+ glDrawBuffer(fbkey.color_attachments[0]);
+ } else {
+ for (std::size_t index = 0; index < Maxwell::NumRenderTargets; ++index) {
+ if (fbkey.colors[index]) {
+ glFramebufferTexture(GL_DRAW_FRAMEBUFFER,
+ GL_COLOR_ATTACHMENT0 + static_cast<GLenum>(index),
+ fbkey.colors[index], 0);
+ }
+ }
+ glDrawBuffers(fbkey.colors_count, fbkey.color_attachments.data());
+ }
+
+ if (fbkey.zeta) {
+ GLenum zeta_attachment =
+ fbkey.stencil_enable ? GL_DEPTH_STENCIL_ATTACHMENT : GL_DEPTH_ATTACHMENT;
+ glFramebufferTexture(GL_DRAW_FRAMEBUFFER, zeta_attachment, fbkey.zeta, 0);
+ }
+}
+
std::size_t RasterizerOpenGL::CalculateVertexArraysSize() const {
const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs;
@@ -444,10 +499,10 @@ void RasterizerOpenGL::ConfigureFramebuffers(OpenGLState& current_state, bool us
UNIMPLEMENTED_IF(regs.rt_separate_frag_data != 0);
// Bind the framebuffer surfaces
- current_state.draw.draw_framebuffer = framebuffer.handle;
- current_state.ApplyFramebufferState();
current_state.framebuffer_srgb.enabled = regs.framebuffer_srgb != 0;
+ FramebufferCacheKey fbkey;
+
if (using_color_fb) {
if (single_color_target) {
// Used when just a single color attachment is enabled, e.g. for clearing a color buffer
@@ -463,14 +518,12 @@ void RasterizerOpenGL::ConfigureFramebuffers(OpenGLState& current_state, bool us
state.framebuffer_srgb.enabled |= color_surface->GetSurfaceParams().srgb_conversion;
}
- glFramebufferTexture2D(
- GL_DRAW_FRAMEBUFFER,
- GL_COLOR_ATTACHMENT0 + static_cast<GLenum>(*single_color_target), GL_TEXTURE_2D,
- color_surface != nullptr ? color_surface->Texture().handle : 0, 0);
- glDrawBuffer(GL_COLOR_ATTACHMENT0 + static_cast<GLenum>(*single_color_target));
+ fbkey.is_single_buffer = true;
+ fbkey.color_attachments[0] =
+ GL_COLOR_ATTACHMENT0 + static_cast<GLenum>(*single_color_target);
+ fbkey.colors[0] = color_surface != nullptr ? color_surface->Texture().handle : 0;
} else {
// Multiple color attachments are enabled
- std::array<GLenum, Maxwell::NumRenderTargets> buffers;
for (std::size_t index = 0; index < Maxwell::NumRenderTargets; ++index) {
Surface color_surface = res_cache.GetColorBufferSurface(index, preserve_contents);
@@ -485,22 +538,17 @@ void RasterizerOpenGL::ConfigureFramebuffers(OpenGLState& current_state, bool us
color_surface->GetSurfaceParams().srgb_conversion;
}
- buffers[index] = GL_COLOR_ATTACHMENT0 + regs.rt_control.GetMap(index);
- glFramebufferTexture2D(
- GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + static_cast<GLenum>(index),
- GL_TEXTURE_2D, color_surface != nullptr ? color_surface->Texture().handle : 0,
- 0);
+ fbkey.color_attachments[index] =
+ GL_COLOR_ATTACHMENT0 + regs.rt_control.GetMap(index);
+ fbkey.colors[index] =
+ color_surface != nullptr ? color_surface->Texture().handle : 0;
}
- glDrawBuffers(regs.rt_control.count, buffers.data());
+ fbkey.is_single_buffer = false;
+ fbkey.colors_count = regs.rt_control.count;
}
} else {
- // No color attachments are enabled - zero out all of them
- for (std::size_t index = 0; index < Maxwell::NumRenderTargets; ++index) {
- glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER,
- GL_COLOR_ATTACHMENT0 + static_cast<GLenum>(index), GL_TEXTURE_2D,
- 0, 0);
- }
- glDrawBuffer(GL_NONE);
+ // No color attachments are enabled - leave them as zero
+ fbkey.is_single_buffer = true;
}
if (depth_surface) {
@@ -508,22 +556,12 @@ void RasterizerOpenGL::ConfigureFramebuffers(OpenGLState& current_state, bool us
// the shader doesn't actually write to it.
depth_surface->MarkAsModified(true, res_cache);
- if (regs.stencil_enable) {
- // Attach both depth and stencil
- glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D,
- depth_surface->Texture().handle, 0);
- } else {
- // Attach depth
- glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, GL_TEXTURE_2D,
- depth_surface->Texture().handle, 0);
- // Clear stencil attachment
- glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0);
- }
- } else {
- // Clear both depth and stencil attachment
- glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0,
- 0);
+ fbkey.zeta = depth_surface->Texture().handle;
+ fbkey.stencil_enable = regs.stencil_enable;
}
+
+ SetupCachedFramebuffer(fbkey, current_state);
+
SyncViewport(current_state);
}
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h
index 988fa3e27..8a891ffc7 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.h
+++ b/src/video_core/renderer_opengl/gl_rasterizer.h
@@ -40,6 +40,7 @@ namespace OpenGL {
struct ScreenInfo;
struct DrawParameters;
+struct FramebufferCacheKey;
class RasterizerOpenGL : public VideoCore::RasterizerInterface {
public:
@@ -195,11 +196,12 @@ private:
OGLVertexArray>
vertex_array_cache;
+ std::map<FramebufferCacheKey, OGLFramebuffer> framebuffer_cache;
+
std::array<SamplerInfo, Tegra::Engines::Maxwell3D::Regs::NumTextureSamplers> texture_samplers;
static constexpr std::size_t STREAM_BUFFER_SIZE = 128 * 1024 * 1024;
OGLBufferCache buffer_cache;
- OGLFramebuffer framebuffer;
PrimitiveAssembler primitive_assembler{buffer_cache};
GLint uniform_buffer_alignment;
@@ -214,6 +216,8 @@ private:
void SetupShaders(GLenum primitive_mode);
+ void SetupCachedFramebuffer(const FramebufferCacheKey& fbkey, OpenGLState& current_state);
+
enum class AccelDraw { Disabled, Arrays, Indexed };
AccelDraw accelerate_draw = AccelDraw::Disabled;
diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp
index 038b25c75..aea6bf1af 100644
--- a/src/video_core/renderer_opengl/gl_shader_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp
@@ -2,7 +2,9 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
+#include <boost/functional/hash.hpp>
#include "common/assert.h"
+#include "common/hash.h"
#include "core/core.h"
#include "core/memory.h"
#include "video_core/engines/maxwell_3d.h"
@@ -66,14 +68,17 @@ CachedShader::CachedShader(VAddr addr, Maxwell::ShaderProgram program_type)
// stage here.
setup.SetProgramB(GetShaderCode(GetShaderAddress(Maxwell::ShaderProgram::VertexB)));
case Maxwell::ShaderProgram::VertexB:
+ CalculateProperties();
program_result = GLShader::GenerateVertexShader(setup);
gl_type = GL_VERTEX_SHADER;
break;
case Maxwell::ShaderProgram::Geometry:
+ CalculateProperties();
program_result = GLShader::GenerateGeometryShader(setup);
gl_type = GL_GEOMETRY_SHADER;
break;
case Maxwell::ShaderProgram::Fragment:
+ CalculateProperties();
program_result = GLShader::GenerateFragmentShader(setup);
gl_type = GL_FRAGMENT_SHADER;
break;
@@ -140,6 +145,46 @@ GLuint CachedShader::LazyGeometryProgram(OGLProgram& target_program,
return target_program.handle;
};
+static bool IsSchedInstruction(std::size_t offset, std::size_t main_offset) {
+ // sched instructions appear once every 4 instructions.
+ static constexpr std::size_t SchedPeriod = 4;
+ const std::size_t absolute_offset = offset - main_offset;
+ return (absolute_offset % SchedPeriod) == 0;
+}
+
+static std::size_t CalculateProgramSize(const GLShader::ProgramCode& program) {
+ constexpr std::size_t start_offset = 10;
+ std::size_t offset = start_offset;
+ std::size_t size = start_offset * sizeof(u64);
+ while (offset < program.size()) {
+ const u64 inst = program[offset];
+ if (!IsSchedInstruction(offset, start_offset)) {
+ if (inst == 0 || (inst >> 52) == 0x50b) {
+ break;
+ }
+ }
+ size += sizeof(inst);
+ offset++;
+ }
+ return size;
+}
+
+void CachedShader::CalculateProperties() {
+ setup.program.real_size = CalculateProgramSize(setup.program.code);
+ setup.program.real_size_b = 0;
+ setup.program.unique_identifier = Common::CityHash64(
+ reinterpret_cast<const char*>(setup.program.code.data()), setup.program.real_size);
+ if (program_type == Maxwell::ShaderProgram::VertexA) {
+ std::size_t seed = 0;
+ boost::hash_combine(seed, setup.program.unique_identifier);
+ setup.program.real_size_b = CalculateProgramSize(setup.program.code_b);
+ const u64 identifier_b = Common::CityHash64(
+ reinterpret_cast<const char*>(setup.program.code_b.data()), setup.program.real_size_b);
+ boost::hash_combine(seed, identifier_b);
+ setup.program.unique_identifier = static_cast<u64>(seed);
+ }
+}
+
ShaderCacheOpenGL::ShaderCacheOpenGL(RasterizerOpenGL& rasterizer) : RasterizerCache{rasterizer} {}
Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) {
diff --git a/src/video_core/renderer_opengl/gl_shader_cache.h b/src/video_core/renderer_opengl/gl_shader_cache.h
index 08f470de3..b4ef6030d 100644
--- a/src/video_core/renderer_opengl/gl_shader_cache.h
+++ b/src/video_core/renderer_opengl/gl_shader_cache.h
@@ -81,6 +81,8 @@ private:
GLuint LazyGeometryProgram(OGLProgram& target_program, const std::string& glsl_topology,
u32 max_vertices, const std::string& debug_name);
+ void CalculateProperties();
+
VAddr addr;
std::size_t shader_length;
Maxwell::ShaderProgram program_type;
diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
index 8d68156bf..a5cfa0070 100644
--- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
@@ -50,6 +50,14 @@ public:
using std::runtime_error::runtime_error;
};
+/// Generates code to use for a swizzle operation.
+static std::string GetSwizzle(u64 elem) {
+ ASSERT(elem <= 3);
+ std::string swizzle = ".";
+ swizzle += "xyzw"[elem];
+ return swizzle;
+}
+
/// Translate topology
static std::string GetTopologyName(Tegra::Shader::OutputTopology topology) {
switch (topology) {
@@ -920,7 +928,7 @@ private:
case Attribute::Index::FrontFacing:
// TODO(Subv): Find out what the values are for the other elements.
ASSERT(stage == Maxwell3D::Regs::ShaderStage::Fragment);
- return "vec4(0, 0, 0, uintBitsToFloat(gl_FrontFacing ? 1 : 0))";
+ return "vec4(0, 0, 0, intBitsToFloat(gl_FrontFacing ? -1 : 0))";
default:
const u32 index{static_cast<u32>(attribute) -
static_cast<u32>(Attribute::Index::Attribute_0)};
@@ -1004,14 +1012,6 @@ private:
}
}
- /// Generates code to use for a swizzle operation.
- static std::string GetSwizzle(u64 elem) {
- ASSERT(elem <= 3);
- std::string swizzle = ".";
- swizzle += "xyzw"[elem];
- return swizzle;
- }
-
ShaderWriter& shader;
ShaderWriter& declarations;
std::vector<GLSLRegister> regs;
@@ -1343,7 +1343,7 @@ private:
regs.SetRegisterToInteger(dest, true, 0, result, 1, 1);
}
- void WriteTexsInstruction(const Instruction& instr, const std::string& texture) {
+ void WriteTexsInstructionFloat(const Instruction& instr, const std::string& texture) {
// TEXS has two destination registers and a swizzle. The first two elements in the swizzle
// go into gpr0+0 and gpr0+1, and the rest goes into gpr28+0 and gpr28+1
@@ -1368,6 +1368,38 @@ private:
}
}
+ void WriteTexsInstructionHalfFloat(const Instruction& instr, const std::string& texture) {
+ // TEXS.F16 destionation registers are packed in two registers in pairs (just like any half
+ // float instruction).
+
+ std::array<std::string, 4> components;
+ u32 written_components = 0;
+
+ for (u32 component = 0; component < 4; ++component) {
+ if (!instr.texs.IsComponentEnabled(component))
+ continue;
+ components[written_components++] = texture + GetSwizzle(component);
+ }
+ if (written_components == 0)
+ return;
+
+ const auto BuildComponent = [&](std::string low, std::string high, bool high_enabled) {
+ return "vec2(" + low + ", " + (high_enabled ? high : "0") + ')';
+ };
+
+ regs.SetRegisterToHalfFloat(
+ instr.gpr0, 0, BuildComponent(components[0], components[1], written_components > 1),
+ Tegra::Shader::HalfMerge::H0_H1, 1, 1);
+
+ if (written_components > 2) {
+ ASSERT(instr.texs.HasTwoDestinations());
+ regs.SetRegisterToHalfFloat(
+ instr.gpr28, 0,
+ BuildComponent(components[2], components[3], written_components > 3),
+ Tegra::Shader::HalfMerge::H0_H1, 1, 1);
+ }
+ }
+
static u32 TextureCoordinates(Tegra::Shader::TextureType texture_type) {
switch (texture_type) {
case Tegra::Shader::TextureType::Texture1D:
@@ -1649,7 +1681,7 @@ private:
for (size_t i = 0; i < coord_count; ++i) {
const bool last = (i == (coord_count - 1)) && (coord_count > 1);
coord += regs.GetRegisterAsFloat(last ? last_coord_register : coord_register + i);
- if (!last) {
+ if (i < coord_count - 1) {
coord += ',';
}
}
@@ -1670,6 +1702,99 @@ private:
is_array, (coord_count > 2 ? 1 : 0)));
}
+ std::pair<std::string, std::string> GetTLD4Code(const Instruction& instr,
+ const Tegra::Shader::TextureType texture_type,
+ const bool depth_compare, const bool is_array) {
+
+ const size_t coord_count = TextureCoordinates(texture_type);
+ const size_t total_coord_count = coord_count + (is_array ? 1 : 0);
+ const size_t total_reg_count = total_coord_count + (depth_compare ? 1 : 0);
+
+ constexpr std::array<const char*, 5> coord_container{
+ {"", "", "vec2 coord = vec2(", "vec3 coord = vec3(", "vec4 coord = vec4("}};
+
+ // If enabled arrays index is always stored in the gpr8 field
+ const u64 array_register = instr.gpr8.Value();
+ // First coordinate index is the gpr8 or gpr8 + 1 when arrays are used
+ const u64 coord_register = array_register + (is_array ? 1 : 0);
+
+ std::string coord = coord_container[total_coord_count];
+ for (size_t i = 0; i < coord_count;) {
+ coord += regs.GetRegisterAsFloat(coord_register + i);
+ ++i;
+ if (i != coord_count) {
+ coord += ',';
+ }
+ }
+
+ if (is_array) {
+ coord += ',' + regs.GetRegisterAsInteger(array_register);
+ }
+ coord += ");";
+
+ const std::string sampler =
+ GetSampler(instr.sampler, texture_type, is_array, depth_compare);
+
+ std::string texture = "textureGather(" + sampler + ", coord, ";
+ if (depth_compare) {
+ // Depth is always stored in the register signaled by gpr20
+ texture += regs.GetRegisterAsFloat(instr.gpr20.Value()) + ')';
+ } else {
+ texture += std::to_string(instr.tld4.component) + ')';
+ }
+ return std::make_pair(coord, texture);
+ }
+
+ std::pair<std::string, std::string> GetTLDSCode(const Instruction& instr,
+ const Tegra::Shader::TextureType texture_type,
+ const bool is_array) {
+
+ const size_t coord_count = TextureCoordinates(texture_type);
+ const size_t total_coord_count = coord_count + (is_array ? 1 : 0);
+ const bool lod_enabled =
+ instr.tlds.GetTextureProcessMode() == Tegra::Shader::TextureProcessMode::LL;
+
+ constexpr std::array<const char*, 4> coord_container{
+ {"", "int coord = (", "ivec2 coord = ivec2(", "ivec3 coord = ivec3("}};
+
+ std::string coord = coord_container[total_coord_count];
+
+ // If enabled arrays index is always stored in the gpr8 field
+ const u64 array_register = instr.gpr8.Value();
+
+ // if is array gpr20 is used
+ const u64 coord_register = is_array ? instr.gpr20.Value() : instr.gpr8.Value();
+
+ const u64 last_coord_register =
+ ((coord_count > 2) || (coord_count == 2 && !lod_enabled)) && !is_array
+ ? static_cast<u64>(instr.gpr20.Value())
+ : coord_register + 1;
+
+ for (size_t i = 0; i < coord_count; ++i) {
+ const bool last = (i == (coord_count - 1)) && (coord_count > 1);
+ coord += regs.GetRegisterAsInteger(last ? last_coord_register : coord_register + i);
+ if (i < coord_count - 1) {
+ coord += ',';
+ }
+ }
+ if (is_array) {
+ coord += ',' + regs.GetRegisterAsInteger(array_register);
+ }
+ coord += ");";
+
+ const std::string sampler = GetSampler(instr.sampler, texture_type, is_array, false);
+
+ std::string texture = "texelFetch(" + sampler + ", coords";
+
+ if (lod_enabled) {
+ // When lod is used always is in grp20
+ texture += ", " + regs.GetRegisterAsInteger(instr.gpr20) + ')';
+ } else {
+ texture += ", 0)";
+ }
+ return std::make_pair(coord, texture);
+ }
+
/**
* Compiles a single instruction from Tegra to GLSL.
* @param offset the offset of the Tegra shader instruction.
@@ -2766,33 +2891,33 @@ private:
const bool depth_compare =
instr.texs.UsesMiscMode(Tegra::Shader::TextureMiscMode::DC);
const auto process_mode = instr.texs.GetTextureProcessMode();
+
UNIMPLEMENTED_IF_MSG(instr.texs.UsesMiscMode(Tegra::Shader::TextureMiscMode::NODEP),
"NODEP is not implemented");
const auto scope = shader.Scope();
- const auto [coord, texture] =
+ auto [coord, texture] =
GetTEXSCode(instr, texture_type, process_mode, depth_compare, is_array);
shader.AddLine(coord);
- if (!depth_compare) {
- shader.AddLine("vec4 texture_tmp = " + texture + ';');
+ if (depth_compare) {
+ texture = "vec4(" + texture + ')';
+ }
+ shader.AddLine("vec4 texture_tmp = " + texture + ';');
+ if (instr.texs.fp32_flag) {
+ WriteTexsInstructionFloat(instr, "texture_tmp");
} else {
- shader.AddLine("vec4 texture_tmp = vec4(" + texture + ");");
+ WriteTexsInstructionHalfFloat(instr, "texture_tmp");
}
-
- WriteTexsInstruction(instr, "texture_tmp");
break;
}
case OpCode::Id::TLDS: {
const Tegra::Shader::TextureType texture_type{instr.tlds.GetTextureType()};
const bool is_array{instr.tlds.IsArrayTexture()};
- ASSERT(texture_type == Tegra::Shader::TextureType::Texture2D);
- ASSERT(is_array == false);
-
UNIMPLEMENTED_IF_MSG(instr.tlds.UsesMiscMode(Tegra::Shader::TextureMiscMode::NODEP),
"NODEP is not implemented");
UNIMPLEMENTED_IF_MSG(instr.tlds.UsesMiscMode(Tegra::Shader::TextureMiscMode::AOFFI),
@@ -2800,54 +2925,16 @@ private:
UNIMPLEMENTED_IF_MSG(instr.tlds.UsesMiscMode(Tegra::Shader::TextureMiscMode::MZ),
"MZ is not implemented");
- u32 extra_op_offset = 0;
-
- ShaderScopedScope scope = shader.Scope();
-
- switch (texture_type) {
- case Tegra::Shader::TextureType::Texture1D: {
- const std::string x = regs.GetRegisterAsInteger(instr.gpr8);
- shader.AddLine("float coords = " + x + ';');
- break;
- }
- case Tegra::Shader::TextureType::Texture2D: {
- UNIMPLEMENTED_IF_MSG(is_array, "Unhandled 2d array texture");
-
- const std::string x = regs.GetRegisterAsInteger(instr.gpr8);
- const std::string y = regs.GetRegisterAsInteger(instr.gpr20);
- // shader.AddLine("ivec2 coords = ivec2(" + x + ", " + y + ");");
- shader.AddLine("ivec2 coords = ivec2(" + x + ", " + y + ");");
- extra_op_offset = 1;
- break;
- }
- default:
- UNIMPLEMENTED_MSG("Unhandled texture type {}", static_cast<u32>(texture_type));
- }
- const std::string sampler =
- GetSampler(instr.sampler, texture_type, is_array, false);
+ const auto [coord, texture] = GetTLDSCode(instr, texture_type, is_array);
- const std::string texture = [&]() {
- switch (instr.tlds.GetTextureProcessMode()) {
- case Tegra::Shader::TextureProcessMode::LZ:
- return "texelFetch(" + sampler + ", coords, 0)";
- case Tegra::Shader::TextureProcessMode::LL:
- shader.AddLine(
- "float lod = " +
- regs.GetRegisterAsInteger(instr.gpr20.Value() + extra_op_offset) + ';');
- return "texelFetch(" + sampler + ", coords, lod)";
- default:
- UNIMPLEMENTED_MSG("Unhandled texture process mode {}",
- static_cast<u32>(instr.tlds.GetTextureProcessMode()));
- return "texelFetch(" + sampler + ", coords, 0)";
- }
- }();
+ const auto scope = shader.Scope();
- WriteTexsInstruction(instr, texture);
+ shader.AddLine(coord);
+ shader.AddLine("vec4 texture_tmp = " + texture + ';');
+ WriteTexsInstructionFloat(instr, "texture_tmp");
break;
}
case OpCode::Id::TLD4: {
- ASSERT(instr.tld4.texture_type == Tegra::Shader::TextureType::Texture2D);
- ASSERT(instr.tld4.array == 0);
UNIMPLEMENTED_IF_MSG(instr.tld4.UsesMiscMode(Tegra::Shader::TextureMiscMode::NODEP),
"NODEP is not implemented");
@@ -2857,56 +2944,29 @@ private:
"NDV is not implemented");
UNIMPLEMENTED_IF_MSG(instr.tld4.UsesMiscMode(Tegra::Shader::TextureMiscMode::PTP),
"PTP is not implemented");
+
+ auto texture_type = instr.tld4.texture_type.Value();
const bool depth_compare =
instr.tld4.UsesMiscMode(Tegra::Shader::TextureMiscMode::DC);
- auto texture_type = instr.tld4.texture_type.Value();
- u32 num_coordinates = TextureCoordinates(texture_type);
- if (depth_compare)
- num_coordinates += 1;
+ const bool is_array = instr.tld4.array != 0;
- const auto scope = shader.Scope();
-
- switch (num_coordinates) {
- case 2: {
- const std::string x = regs.GetRegisterAsFloat(instr.gpr8);
- const std::string y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1);
- shader.AddLine("vec2 coords = vec2(" + x + ", " + y + ");");
- break;
- }
- case 3: {
- const std::string x = regs.GetRegisterAsFloat(instr.gpr8);
- const std::string y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1);
- const std::string z = regs.GetRegisterAsFloat(instr.gpr8.Value() + 2);
- shader.AddLine("vec3 coords = vec3(" + x + ", " + y + ", " + z + ");");
- break;
- }
- default:
- UNIMPLEMENTED_MSG("Unhandled coordinates number {}",
- static_cast<u32>(num_coordinates));
- const std::string x = regs.GetRegisterAsFloat(instr.gpr8);
- const std::string y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1);
- shader.AddLine("vec2 coords = vec2(" + x + ", " + y + ");");
- texture_type = Tegra::Shader::TextureType::Texture2D;
- }
+ const auto [coord, texture] =
+ GetTLD4Code(instr, texture_type, depth_compare, is_array);
- const std::string sampler =
- GetSampler(instr.sampler, texture_type, false, depth_compare);
+ const auto scope = shader.Scope();
- const std::string texture = "textureGather(" + sampler + ", coords, " +
- std::to_string(instr.tld4.component) + ')';
+ shader.AddLine(coord);
+ std::size_t dest_elem{};
- if (depth_compare) {
- regs.SetRegisterToFloat(instr.gpr0, 0, texture, 1, 1, false);
- } else {
- std::size_t dest_elem{};
- for (std::size_t elem = 0; elem < 4; ++elem) {
- if (!instr.tex.IsComponentEnabled(elem)) {
- // Skip disabled components
- continue;
- }
- regs.SetRegisterToFloat(instr.gpr0, elem, texture, 1, 4, false, dest_elem);
- ++dest_elem;
+ shader.AddLine("vec4 texture_tmp = " + texture + ';');
+ for (std::size_t elem = 0; elem < 4; ++elem) {
+ if (!instr.tex.IsComponentEnabled(elem)) {
+ // Skip disabled components
+ continue;
}
+ regs.SetRegisterToFloat(instr.gpr0, elem, "texture_tmp", 1, 4, false,
+ dest_elem);
+ ++dest_elem;
}
break;
}
@@ -2920,27 +2980,31 @@ private:
const auto scope = shader.Scope();
+ std::string coords;
+
const bool depth_compare =
instr.tld4s.UsesMiscMode(Tegra::Shader::TextureMiscMode::DC);
- const std::string op_a = regs.GetRegisterAsFloat(instr.gpr8);
- const std::string op_b = regs.GetRegisterAsFloat(instr.gpr20);
- // TODO(Subv): Figure out how the sampler type is encoded in the TLD4S instruction.
+
const std::string sampler = GetSampler(
instr.sampler, Tegra::Shader::TextureType::Texture2D, false, depth_compare);
- if (depth_compare) {
- // Note: TLD4S coordinate encoding works just like TEXS's
- const std::string op_y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1);
- shader.AddLine("vec3 coords = vec3(" + op_a + ", " + op_y + ", " + op_b + ");");
- } else {
- shader.AddLine("vec2 coords = vec2(" + op_a + ", " + op_b + ");");
- }
- std::string texture = "textureGather(" + sampler + ", coords, " +
- std::to_string(instr.tld4s.component) + ')';
- if (depth_compare) {
- texture = "vec4(" + texture + ')';
- }
- WriteTexsInstruction(instr, texture);
+ const std::string op_a = regs.GetRegisterAsFloat(instr.gpr8);
+ coords = "vec2 coords = vec2(" + op_a + ", ";
+ std::string texture = "textureGather(" + sampler + ", coords, ";
+
+ if (!depth_compare) {
+ const std::string op_b = regs.GetRegisterAsFloat(instr.gpr20);
+ coords += op_b + ");";
+ texture += std::to_string(instr.tld4s.component) + ')';
+ } else {
+ const std::string op_b = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1);
+ const std::string op_c = regs.GetRegisterAsFloat(instr.gpr20);
+ coords += op_b + ");";
+ texture += op_c + ')';
+ }
+ shader.AddLine(coords);
+ shader.AddLine("vec4 texture_tmp = " + texture + ';');
+ WriteTexsInstructionFloat(instr, "texture_tmp");
break;
}
case OpCode::Id::TXQ: {
diff --git a/src/video_core/renderer_opengl/gl_shader_gen.cpp b/src/video_core/renderer_opengl/gl_shader_gen.cpp
index 23ed91e27..5d0819dc5 100644
--- a/src/video_core/renderer_opengl/gl_shader_gen.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_gen.cpp
@@ -2,6 +2,7 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
+#include <fmt/format.h>
#include "common/assert.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/renderer_opengl/gl_shader_decompiler.h"
@@ -16,6 +17,8 @@ static constexpr u32 PROGRAM_OFFSET{10};
ProgramResult GenerateVertexShader(const ShaderSetup& setup) {
std::string out = "#version 430 core\n";
out += "#extension GL_ARB_separate_shader_objects : enable\n\n";
+ const std::string id = fmt::format("{:016x}", setup.program.unique_identifier);
+ out += "// Shader Unique Id: VS" + id + "\n\n";
out += Decompiler::GetCommonDeclarations();
out += R"(
@@ -84,6 +87,8 @@ void main() {
ProgramResult GenerateGeometryShader(const ShaderSetup& setup) {
// Version is intentionally skipped in shader generation, it's added by the lazy compilation.
std::string out = "#extension GL_ARB_separate_shader_objects : enable\n\n";
+ const std::string id = fmt::format("{:016x}", setup.program.unique_identifier);
+ out += "// Shader Unique Id: GS" + id + "\n\n";
out += Decompiler::GetCommonDeclarations();
out += "bool exec_geometry();\n";
@@ -117,6 +122,8 @@ void main() {
ProgramResult GenerateFragmentShader(const ShaderSetup& setup) {
std::string out = "#version 430 core\n";
out += "#extension GL_ARB_separate_shader_objects : enable\n\n";
+ const std::string id = fmt::format("{:016x}", setup.program.unique_identifier);
+ out += "// Shader Unique Id: FS" + id + "\n\n";
out += Decompiler::GetCommonDeclarations();
out += "bool exec_fragment();\n";
diff --git a/src/video_core/renderer_opengl/gl_shader_gen.h b/src/video_core/renderer_opengl/gl_shader_gen.h
index 4fa6d7612..fcc20d3b4 100644
--- a/src/video_core/renderer_opengl/gl_shader_gen.h
+++ b/src/video_core/renderer_opengl/gl_shader_gen.h
@@ -177,6 +177,9 @@ struct ShaderSetup {
struct {
ProgramCode code;
ProgramCode code_b; // Used for dual vertex shaders
+ u64 unique_identifier;
+ std::size_t real_size;
+ std::size_t real_size_b;
} program;
/// Used in scenarios where we have a dual vertex shaders
generated by cgit v1.2.3 (git 2.25.1) at 2024-11-02 11:43:25 +0100