yuzu-emu
/
yuzu
Archived
1
0
Fork 0
Code

shader: Split profile and runtime information in separate structs

This commit is contained in:
ReinUsesLisp 2021-05-21 02:12:32 -03:00 committed by ameerj
parent eb15667905
commit 9e7b6622c2
14 changed files with 300 additions and 308 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

40
src/shader_recompiler/backend/glasm/emit_context.cpp
View File

@ -23,23 +23,25 @@ std::string_view InterpDecorator(Interpolation interp) {
}
} // Anonymous namespace
EmitContext::EmitContext(IR::Program& program, Bindings& bindings, const Profile& profile_)
: info{program.info}, profile{profile_} {
EmitContext::EmitContext(IR::Program& program, Bindings& bindings, const Profile& profile_,
const RuntimeInfo& runtime_info_)
: profile{profile_}, runtime_info{runtime_info_} {
// FIXME: Temporary partial implementation
const auto& info{program.info};
u32 cbuf_index{};
for (const auto& desc : program.info.constant_buffer_descriptors) {
for (const auto& desc : info.constant_buffer_descriptors) {
if (desc.count != 1) {
throw NotImplementedException("Constant buffer descriptor array");
}
Add("CBUFFER c{}[]={{program.buffer[{}]}};", desc.index, cbuf_index);
++cbuf_index;
}
for (const auto& desc : program.info.storage_buffers_descriptors) {
for (const auto& desc : info.storage_buffers_descriptors) {
if (desc.count != 1) {
throw NotImplementedException("Storage buffer descriptor array");
}
}
if (const size_t num = program.info.storage_buffers_descriptors.size(); num > 0) {
if (const size_t num = info.storage_buffers_descriptors.size(); num > 0) {
Add("PARAM c[{}]={{program.local[0..{}]}};", num, num - 1);
}
stage = program.stage;
@ -67,8 +69,8 @@ EmitContext::EmitContext(IR::Program& program, Bindings& bindings, const Profile
break;
}
const std::string_view attr_stage{stage == Stage::Fragment ? "fragment" : "vertex"};
for (size_t index = 0; index < program.info.input_generics.size(); ++index) {
const auto& generic{program.info.input_generics[index]};
for (size_t index = 0; index < info.input_generics.size(); ++index) {
const auto& generic{info.input_generics[index]};
if (generic.used) {
Add("{}ATTRIB in_attr{}[]={{{}.attrib[{}..{}]}};",
InterpDecorator(generic.interpolation), index, attr_stage, index, index);
@ -101,8 +103,8 @@ EmitContext::EmitContext(IR::Program& program, Bindings& bindings, const Profile
index, index);
}
}
for (size_t index = 0; index < program.info.stores_frag_color.size(); ++index) {
if (!program.info.stores_frag_color[index]) {
for (size_t index = 0; index < info.stores_frag_color.size(); ++index) {
if (!info.stores_frag_color[index]) {
continue;
}
if (index == 0) {
@ -111,28 +113,28 @@ EmitContext::EmitContext(IR::Program& program, Bindings& bindings, const Profile
Add("OUTPUT frag_color{}=result.color[{}];", index, index);
}
}
for (size_t index = 0; index < program.info.stores_generics.size(); ++index) {
if (program.info.stores_generics[index]) {
for (size_t index = 0; index < info.stores_generics.size(); ++index) {
if (info.stores_generics[index]) {
Add("OUTPUT out_attr{}[]={{result.attrib[{}..{}]}};", index, index, index);
}
}
image_buffer_bindings.reserve(program.info.image_buffer_descriptors.size());
for (const auto& desc : program.info.image_buffer_descriptors) {
image_buffer_bindings.reserve(info.image_buffer_descriptors.size());
for (const auto& desc : info.image_buffer_descriptors) {
image_buffer_bindings.push_back(bindings.image);
bindings.image += desc.count;
}
image_bindings.reserve(program.info.image_descriptors.size());
for (const auto& desc : program.info.image_descriptors) {
image_bindings.reserve(info.image_descriptors.size());
for (const auto& desc : info.image_descriptors) {
image_bindings.push_back(bindings.image);
bindings.image += desc.count;
}
texture_buffer_bindings.reserve(program.info.texture_buffer_descriptors.size());
for (const auto& desc : program.info.texture_buffer_descriptors) {
texture_buffer_bindings.reserve(info.texture_buffer_descriptors.size());
for (const auto& desc : info.texture_buffer_descriptors) {
texture_buffer_bindings.push_back(bindings.texture);
bindings.texture += desc.count;
}
texture_bindings.reserve(program.info.texture_descriptors.size());
for (const auto& desc : program.info.texture_descriptors) {
texture_bindings.reserve(info.texture_descriptors.size());
for (const auto& desc : info.texture_descriptors) {
texture_bindings.push_back(bindings.texture);
bindings.texture += desc.count;
}

6
src/shader_recompiler/backend/glasm/emit_context.h
View File

@ -16,6 +16,7 @@
namespace Shader {
struct Info;
struct Profile;
struct RuntimeInfo;
} // namespace Shader
namespace Shader::Backend {
@ -31,7 +32,8 @@ namespace Shader::Backend::GLASM {
class EmitContext {
public:
explicit EmitContext(IR::Program& program, Bindings& bindings, const Profile& profile_);
explicit EmitContext(IR::Program& program, Bindings& bindings, const Profile& profile_,
const RuntimeInfo& runtime_info_);
template <typename... Args>
void Add(const char* format_str, IR::Inst& inst, Args&&... args) {
@ -56,8 +58,8 @@ public:
std::string code;
RegAlloc reg_alloc{*this};
const Info& info;
const Profile& profile;
const RuntimeInfo& runtime_info;
std::vector<u32> texture_buffer_bindings;
std::vector<u32> image_buffer_bindings;

19
src/shader_recompiler/backend/glasm/emit_glasm.cpp
View File

@ -374,8 +374,9 @@ std::string_view GetTessSpacing(TessSpacing spacing) {
}
} // Anonymous namespace
std::string EmitGLASM(const Profile& profile, IR::Program& program, Bindings& bindings) {
EmitContext ctx{program, bindings, profile};
std::string EmitGLASM(const Profile& profile, const RuntimeInfo& runtime_info, IR::Program& program,
Bindings& bindings) {
EmitContext ctx{program, bindings, profile, runtime_info};
Precolor(ctx, program);
EmitCode(ctx, program);
std::string header{StageHeader(program.stage)};
@ -385,18 +386,18 @@ std::string EmitGLASM(const Profile& profile, IR::Program& program, Bindings& bi
header += fmt::format("VERTICES_OUT {};", program.invocations);
break;
case Stage::TessellationEval:
header +=
fmt::format("TESS_MODE {};"
"TESS_SPACING {};"
"TESS_VERTEX_ORDER {};",
GetTessMode(profile.tess_primitive), GetTessSpacing(profile.tess_spacing),
profile.tess_clockwise ? "CW" : "CCW");
header += fmt::format("TESS_MODE {};"
"TESS_SPACING {};"
"TESS_VERTEX_ORDER {};",
GetTessMode(runtime_info.tess_primitive),
GetTessSpacing(runtime_info.tess_spacing),
runtime_info.tess_clockwise ? "CW" : "CCW");
break;
case Stage::Geometry:
header += fmt::format("PRIMITIVE_IN {};"
"PRIMITIVE_OUT {};"
"VERTICES_OUT {};",
InputPrimitive(profile.input_topology),
InputPrimitive(runtime_info.input_topology),
OutputPrimitive(program.output_topology), program.output_vertices);
break;
case Stage::Compute:

6
src/shader_recompiler/backend/glasm/emit_glasm.h
View File

@ -12,12 +12,12 @@
namespace Shader::Backend::GLASM {
[[nodiscard]] std::string EmitGLASM(const Profile& profile, IR::Program& program,
Bindings& binding);
[[nodiscard]] std::string EmitGLASM(const Profile& profile, const RuntimeInfo& runtime_info,
IR::Program& program, Bindings& bindings);
[[nodiscard]] inline std::string EmitGLASM(const Profile& profile, IR::Program& program) {
Bindings binding;
return EmitGLASM(profile, program, binding);
return EmitGLASM(profile, {}, program, binding);
}
} // namespace Shader::Backend::GLASM

26
src/shader_recompiler/backend/spirv/emit_context.cpp
View File

@ -136,7 +136,7 @@ Id DefineInput(EmitContext& ctx, Id type, bool per_invocation,
break;
case Stage::Geometry:
if (per_invocation) {
const u32 num_vertices{NumVertices(ctx.profile.input_topology)};
const u32 num_vertices{NumVertices(ctx.runtime_info.input_topology)};
type = ctx.TypeArray(type, ctx.Const(num_vertices));
}
break;
@ -161,8 +161,8 @@ void DefineGenericOutput(EmitContext& ctx, size_t index, std::optional<u32> invo
while (element < 4) {
const u32 remainder{4 - element};
const TransformFeedbackVarying* xfb_varying{};
if (!ctx.profile.xfb_varyings.empty()) {
xfb_varying = &ctx.profile.xfb_varyings[base_attr_index + element];
if (!ctx.runtime_info.xfb_varyings.empty()) {
xfb_varying = &ctx.runtime_info.xfb_varyings[base_attr_index + element];
xfb_varying = xfb_varying && xfb_varying->components > 0 ? xfb_varying : nullptr;
}
const u32 num_components{xfb_varying ? xfb_varying->components : remainder};
@ -208,7 +208,7 @@ Id GetAttributeType(EmitContext& ctx, AttributeType type) {
}
std::optional<AttrInfo> AttrTypes(EmitContext& ctx, u32 index) {
const AttributeType type{ctx.profile.generic_input_types.at(index)};
const AttributeType type{ctx.runtime_info.generic_input_types.at(index)};
switch (type) {
case AttributeType::Float:
return AttrInfo{ctx.input_f32, ctx.F32[1], false};
@ -441,13 +441,15 @@ void VectorTypes::Define(Sirit::Module& sirit_ctx, Id base_type, std::string_vie
}
}
EmitContext::EmitContext(const Profile& profile_, IR::Program& program, Bindings& binding)
: Sirit::Module(profile_.supported_spirv), profile{profile_}, stage{program.stage} {
EmitContext::EmitContext(const Profile& profile_, const RuntimeInfo& runtime_info_,
IR::Program& program, Bindings& bindings)
: Sirit::Module(profile_.supported_spirv), profile{profile_},
runtime_info{runtime_info_}, stage{program.stage} {
const bool is_unified{profile.unified_descriptor_binding};
u32& uniform_binding{is_unified ? binding.unified : binding.uniform_buffer};
u32& storage_binding{is_unified ? binding.unified : binding.storage_buffer};
u32& texture_binding{is_unified ? binding.unified : binding.texture};
u32& image_binding{is_unified ? binding.unified : binding.image};
u32& uniform_binding{is_unified ? bindings.unified : bindings.uniform_buffer};
u32& storage_binding{is_unified ? bindings.unified : bindings.storage_buffer};
u32& texture_binding{is_unified ? bindings.unified : bindings.texture};
u32& image_binding{is_unified ? bindings.unified : bindings.image};
AddCapability(spv::Capability::Shader);
DefineCommonTypes(program.info);
DefineCommonConstants();
@ -1211,7 +1213,7 @@ void EmitContext::DefineInputs(const Info& info) {
if (!generic.used) {
continue;
}
const AttributeType input_type{profile.generic_input_types[index]};
const AttributeType input_type{runtime_info.generic_input_types[index]};
if (input_type == AttributeType::Disabled) {
continue;
}
@ -1256,7 +1258,7 @@ void EmitContext::DefineOutputs(const IR::Program& program) {
if (info.stores_position || stage == Stage::VertexB) {
output_position = DefineOutput(*this, F32[4], invocations, spv::BuiltIn::Position);
}
if (info.stores_point_size || profile.fixed_state_point_size) {
if (info.stores_point_size || runtime_info.fixed_state_point_size) {
if (stage == Stage::Fragment) {
throw NotImplementedException("Storing PointSize in fragment stage");
}

4
src/shader_recompiler/backend/spirv/emit_context.h
View File

@ -103,7 +103,8 @@ struct GenericElementInfo {
class EmitContext final : public Sirit::Module {
public:
explicit EmitContext(const Profile& profile, IR::Program& program, Bindings& binding);
explicit EmitContext(const Profile& profile, const RuntimeInfo& runtime_info,
IR::Program& program, Bindings& binding);
~EmitContext();
[[nodiscard]] Id Def(const IR::Value& value);
@ -150,6 +151,7 @@ public:
}
const Profile& profile;
const RuntimeInfo& runtime_info;
Stage stage{};
Id void_id{};

20
src/shader_recompiler/backend/spirv/emit_spirv.cpp
View File

@ -226,16 +226,17 @@ void DefineEntryPoint(const IR::Program& program, EmitContext& ctx, Id main) {
case Stage::TessellationEval:
execution_model = spv::ExecutionModel::TessellationEvaluation;
ctx.AddCapability(spv::Capability::Tessellation);
ctx.AddExecutionMode(main, ExecutionMode(ctx.profile.tess_primitive));
ctx.AddExecutionMode(main, ExecutionMode(ctx.profile.tess_spacing));
ctx.AddExecutionMode(main, ctx.profile.tess_clockwise ? spv::ExecutionMode::VertexOrderCw
: spv::ExecutionMode::VertexOrderCcw);
ctx.AddExecutionMode(main, ExecutionMode(ctx.runtime_info.tess_primitive));
ctx.AddExecutionMode(main, ExecutionMode(ctx.runtime_info.tess_spacing));
ctx.AddExecutionMode(main, ctx.runtime_info.tess_clockwise
? spv::ExecutionMode::VertexOrderCw
: spv::ExecutionMode::VertexOrderCcw);
break;
case Stage::Geometry:
execution_model = spv::ExecutionModel::Geometry;
ctx.AddCapability(spv::Capability::Geometry);
ctx.AddCapability(spv::Capability::GeometryStreams);
switch (ctx.profile.input_topology) {
switch (ctx.runtime_info.input_topology) {
case InputTopology::Points:
ctx.AddExecutionMode(main, spv::ExecutionMode::InputPoints);
break;
@ -279,7 +280,7 @@ void DefineEntryPoint(const IR::Program& program, EmitContext& ctx, Id main) {
if (program.info.stores_frag_depth) {
ctx.AddExecutionMode(main, spv::ExecutionMode::DepthReplacing);
}
if (ctx.profile.force_early_z) {
if (ctx.runtime_info.force_early_z) {
ctx.AddExecutionMode(main, spv::ExecutionMode::EarlyFragmentTests);
}
break;
@ -402,7 +403,7 @@ void SetupCapabilities(const Profile& profile, const Info& info, EmitContext& ct
if (info.uses_sample_id) {
ctx.AddCapability(spv::Capability::SampleRateShading);
}
if (!ctx.profile.xfb_varyings.empty()) {
if (!ctx.runtime_info.xfb_varyings.empty()) {
ctx.AddCapability(spv::Capability::TransformFeedback);
}
if (info.uses_derivatives) {
@ -433,8 +434,9 @@ void PatchPhiNodes(IR::Program& program, EmitContext& ctx) {
}
} // Anonymous namespace
std::vector<u32> EmitSPIRV(const Profile& profile, IR::Program& program, Bindings& binding) {
EmitContext ctx{profile, program, binding};
std::vector<u32> EmitSPIRV(const Profile& profile, const RuntimeInfo& runtime_info,
IR::Program& program, Bindings& bindings) {
EmitContext ctx{profile, runtime_info, program, bindings};
const Id main{DefineMain(ctx, program)};
DefineEntryPoint(program, ctx, main);
if (profile.support_float_controls) {

6
src/shader_recompiler/backend/spirv/emit_spirv.h
View File

@ -16,12 +16,12 @@
namespace Shader::Backend::SPIRV {
[[nodiscard]] std::vector<u32> EmitSPIRV(const Profile& profile, IR::Program& program,
Bindings& binding);
[[nodiscard]] std::vector<u32> EmitSPIRV(const Profile& profile, const RuntimeInfo& runtime_info,
IR::Program& program, Bindings& bindings);
[[nodiscard]] inline std::vector<u32> EmitSPIRV(const Profile& profile, IR::Program& program) {
Bindings binding;
return EmitSPIRV(profile, program, binding);
return EmitSPIRV(profile, {}, program, binding);
}
} // namespace Shader::Backend::SPIRV

4
src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp
View File

@ -17,7 +17,7 @@ struct AttrInfo {
};
std::optional<AttrInfo> AttrTypes(EmitContext& ctx, u32 index) {
const AttributeType type{ctx.profile.generic_input_types.at(index)};
const AttributeType type{ctx.runtime_info.generic_input_types.at(index)};
switch (type) {
case AttributeType::Float:
return AttrInfo{ctx.input_f32, ctx.F32[1], false};
@ -468,7 +468,7 @@ Id EmitIsHelperInvocation(EmitContext& ctx) {
}
Id EmitYDirection(EmitContext& ctx) {
return ctx.Const(ctx.profile.y_negate ? -1.0f : 1.0f);
return ctx.Const(ctx.runtime_info.y_negate ? -1.0f : 1.0f);
}
Id EmitLoadLocal(EmitContext& ctx, Id word_offset) {

15
src/shader_recompiler/backend/spirv/emit_spirv_special.cpp
View File

@ -18,8 +18,8 @@ void ConvertDepthMode(EmitContext& ctx) {
}
void SetFixedPipelinePointSize(EmitContext& ctx) {
if (ctx.profile.fixed_state_point_size) {
const float point_size{*ctx.profile.fixed_state_point_size};
if (ctx.runtime_info.fixed_state_point_size) {
const float point_size{*ctx.runtime_info.fixed_state_point_size};
ctx.OpStore(ctx.output_point_size, ctx.Const(point_size));
}
}
@ -62,7 +62,10 @@ Id ComparisonFunction(EmitContext& ctx, CompareFunction comparison, Id operand_1
}
void AlphaTest(EmitContext& ctx) {
const auto comparison{*ctx.profile.alpha_test_func};
if (!ctx.runtime_info.alpha_test_func) {
return;
}
const auto comparison{*ctx.runtime_info.alpha_test_func};
if (comparison == CompareFunction::Always) {
return;
}
@ -76,7 +79,7 @@ void AlphaTest(EmitContext& ctx) {
const Id true_label{ctx.OpLabel()};
const Id discard_label{ctx.OpLabel()};
const Id alpha_reference{ctx.Const(ctx.profile.alpha_test_reference)};
const Id alpha_reference{ctx.Const(ctx.runtime_info.alpha_test_reference)};
const Id condition{ComparisonFunction(ctx, comparison, alpha, alpha_reference)};
ctx.OpSelectionMerge(true_label, spv::SelectionControlMask::MaskNone);
@ -113,7 +116,7 @@ void EmitPrologue(EmitContext& ctx) {
}
void EmitEpilogue(EmitContext& ctx) {
if (ctx.stage == Stage::VertexB && ctx.profile.convert_depth_mode) {
if (ctx.stage == Stage::VertexB && ctx.runtime_info.convert_depth_mode) {
ConvertDepthMode(ctx);
}
if (ctx.stage == Stage::Fragment) {
@ -122,7 +125,7 @@ void EmitEpilogue(EmitContext& ctx) {
}
void EmitEmitVertex(EmitContext& ctx, const IR::Value& stream) {
if (ctx.profile.convert_depth_mode) {
if (ctx.runtime_info.convert_depth_mode) {
ConvertDepthMode(ctx);
}
if (stream.IsImmediate()) {

13
src/shader_recompiler/profile.h
View File

@ -81,19 +81,22 @@ struct Profile {
bool support_viewport_mask{};
bool support_typeless_image_loads{};
bool support_demote_to_helper_invocation{};
bool warp_size_potentially_larger_than_guest{};
bool support_int64_atomics{};
bool warp_size_potentially_larger_than_guest{};
bool lower_left_origin_mode{};
// FClamp is broken and OpFMax + OpFMin should be used instead
/// OpFClamp is broken and OpFMax + OpFMin should be used instead
bool has_broken_spirv_clamp{};
// Offset image operands with an unsigned type do not work
/// Offset image operands with an unsigned type do not work
bool has_broken_unsigned_image_offsets{};
// Signed instructions with unsigned data types are misinterpreted
/// Signed instructions with unsigned data types are misinterpreted
bool has_broken_signed_operations{};
// Ignores SPIR-V ordered vs unordered using GLSL semantics
/// Ignores SPIR-V ordered vs unordered using GLSL semantics
bool ignore_nan_fp_comparisons{};
};
struct RuntimeInfo {
std::array<AttributeType, 32> generic_input_types{};
bool convert_depth_mode{};
bool force_early_z{};

26
src/video_core/renderer_opengl/gl_shader_cache.cpp
View File

@ -61,33 +61,15 @@ const Shader::Profile profile{
.support_viewport_mask = true,
.support_typeless_image_loads = true,
.support_demote_to_helper_invocation = false,
.warp_size_potentially_larger_than_guest = true,
.support_int64_atomics = false,
.warp_size_potentially_larger_than_guest = true,
.lower_left_origin_mode = true,
.has_broken_spirv_clamp = true,
.has_broken_unsigned_image_offsets = true,
.has_broken_signed_operations = true,
.ignore_nan_fp_comparisons = true,
.generic_input_types = {},
.convert_depth_mode = false,
.force_early_z = false,
.tess_primitive = {},
.tess_spacing = {},
.tess_clockwise = false,
.input_topology = Shader::InputTopology::Triangles,
.fixed_state_point_size = std::nullopt,
.alpha_test_func = Shader::CompareFunction::Always,
.alpha_test_reference = 0.0f,
.y_negate = false,
.xfb_varyings = {},
};
using Shader::Backend::GLASM::EmitGLASM;
@ -302,10 +284,10 @@ std::unique_ptr<GraphicsProgram> ShaderCache::CreateGraphicsProgram(
const size_t stage_index{index - 1};
infos[stage_index] = &program.info;
if (device.UseAssemblyShaders()) {
const std::string code{EmitGLASM(profile, program, binding)};
const std::string code{EmitGLASM(profile, {}, program, binding)};
assembly_programs[stage_index] = CompileProgram(code, AssemblyStage(stage_index));
} else {
const std::vector<u32> code{EmitSPIRV(profile, program, binding)};
const std::vector<u32> code{EmitSPIRV(profile, {}, program, binding)};
AddShader(Stage(stage_index), source_program.handle, code);
}
}

418
src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
View File

@ -89,6 +89,208 @@ Shader::CompareFunction MaxwellToCompareFunction(Maxwell::ComparisonOp compariso
UNIMPLEMENTED_MSG("Unimplemented comparison op={}", comparison);
return {};
}
static Shader::AttributeType CastAttributeType(const FixedPipelineState::VertexAttribute& attr) {
if (attr.enabled == 0) {
return Shader::AttributeType::Disabled;
}
switch (attr.Type()) {
case Maxwell::VertexAttribute::Type::SignedNorm:
case Maxwell::VertexAttribute::Type::UnsignedNorm:
case Maxwell::VertexAttribute::Type::UnsignedScaled:
case Maxwell::VertexAttribute::Type::SignedScaled:
case Maxwell::VertexAttribute::Type::Float:
return Shader::AttributeType::Float;
case Maxwell::VertexAttribute::Type::SignedInt:
return Shader::AttributeType::SignedInt;
case Maxwell::VertexAttribute::Type::UnsignedInt:
return Shader::AttributeType::UnsignedInt;
}
return Shader::AttributeType::Float;
}
std::vector<Shader::TransformFeedbackVarying> MakeTransformFeedbackVaryings(
const GraphicsPipelineCacheKey& key) {
static constexpr std::array VECTORS{
28, // gl_Position
32, // Generic 0
36, // Generic 1
40, // Generic 2
44, // Generic 3
48, // Generic 4
52, // Generic 5
56, // Generic 6
60, // Generic 7
64, // Generic 8
68, // Generic 9
72, // Generic 10
76, // Generic 11
80, // Generic 12
84, // Generic 13
88, // Generic 14
92, // Generic 15
96, // Generic 16
100, // Generic 17
104, // Generic 18
108, // Generic 19
112, // Generic 20
116, // Generic 21
120, // Generic 22
124, // Generic 23
128, // Generic 24
132, // Generic 25
136, // Generic 26
140, // Generic 27
144, // Generic 28
148, // Generic 29
152, // Generic 30
156, // Generic 31
160, // gl_FrontColor
164, // gl_FrontSecondaryColor
160, // gl_BackColor
164, // gl_BackSecondaryColor
192, // gl_TexCoord[0]
196, // gl_TexCoord[1]
200, // gl_TexCoord[2]
204, // gl_TexCoord[3]
208, // gl_TexCoord[4]
212, // gl_TexCoord[5]
216, // gl_TexCoord[6]
220, // gl_TexCoord[7]
};
std::vector<Shader::TransformFeedbackVarying> xfb(256);
for (size_t buffer = 0; buffer < Maxwell::NumTransformFeedbackBuffers; ++buffer) {
const auto& locations = key.state.xfb_state.varyings[buffer];
const auto& layout = key.state.xfb_state.layouts[buffer];
const u32 varying_count = layout.varying_count;
u32 highest = 0;
for (u32 offset = 0; offset < varying_count; ++offset) {
const u32 base_offset = offset;
const u8 location = locations[offset];
Shader::TransformFeedbackVarying varying;
varying.buffer = layout.stream;
varying.stride = layout.stride;
varying.offset = offset * 4;
varying.components = 1;
if (std::ranges::find(VECTORS, Common::AlignDown(location, 4)) != VECTORS.end()) {
UNIMPLEMENTED_IF_MSG(location % 4 != 0, "Unaligned TFB");
const u8 base_index = location / 4;
while (offset + 1 < varying_count && base_index == locations[offset + 1] / 4) {
++offset;
++varying.components;
}
}
xfb[location] = varying;
highest = std::max(highest, (base_offset + varying.components) * 4);
}
UNIMPLEMENTED_IF(highest != layout.stride);
}
return xfb;
}
Shader::RuntimeInfo MakeRuntimeInfo(const GraphicsPipelineCacheKey& key,
const Shader::IR::Program& program) {
Shader::RuntimeInfo info;
const Shader::Stage stage{program.stage};
const bool has_geometry{key.unique_hashes[4] != 0};
const bool gl_ndc{key.state.ndc_minus_one_to_one != 0};
const float point_size{Common::BitCast<float>(key.state.point_size)};
switch (stage) {
case Shader::Stage::VertexB:
if (!has_geometry) {
if (key.state.topology == Maxwell::PrimitiveTopology::Points) {
info.fixed_state_point_size = point_size;
}
if (key.state.xfb_enabled != 0) {
info.xfb_varyings = MakeTransformFeedbackVaryings(key);
}
info.convert_depth_mode = gl_ndc;
}
std::ranges::transform(key.state.attributes, info.generic_input_types.begin(),
&CastAttributeType);
break;
case Shader::Stage::TessellationEval:
// We have to flip tessellation clockwise for some reason...
info.tess_clockwise = key.state.tessellation_clockwise == 0;
info.tess_primitive = [&key] {
const u32 raw{key.state.tessellation_primitive.Value()};
switch (static_cast<Maxwell::TessellationPrimitive>(raw)) {
case Maxwell::TessellationPrimitive::Isolines:
return Shader::TessPrimitive::Isolines;
case Maxwell::TessellationPrimitive::Triangles:
return Shader::TessPrimitive::Triangles;
case Maxwell::TessellationPrimitive::Quads:
return Shader::TessPrimitive::Quads;
}
UNREACHABLE();
return Shader::TessPrimitive::Triangles;
}();
info.tess_spacing = [&] {
const u32 raw{key.state.tessellation_spacing};
switch (static_cast<Maxwell::TessellationSpacing>(raw)) {
case Maxwell::TessellationSpacing::Equal:
return Shader::TessSpacing::Equal;
case Maxwell::TessellationSpacing::FractionalOdd:
return Shader::TessSpacing::FractionalOdd;
case Maxwell::TessellationSpacing::FractionalEven:
return Shader::TessSpacing::FractionalEven;
}
UNREACHABLE();
return Shader::TessSpacing::Equal;
}();
break;
case Shader::Stage::Geometry:
if (program.output_topology == Shader::OutputTopology::PointList) {
info.fixed_state_point_size = point_size;
}
if (key.state.xfb_enabled != 0) {
info.xfb_varyings = MakeTransformFeedbackVaryings(key);
}
info.convert_depth_mode = gl_ndc;
break;
case Shader::Stage::Fragment:
info.alpha_test_func = MaxwellToCompareFunction(
key.state.UnpackComparisonOp(key.state.alpha_test_func.Value()));
info.alpha_test_reference = Common::BitCast<float>(key.state.alpha_test_ref);
break;
default:
break;
}
switch (key.state.topology) {
case Maxwell::PrimitiveTopology::Points:
info.input_topology = Shader::InputTopology::Points;
break;
case Maxwell::PrimitiveTopology::Lines:
case Maxwell::PrimitiveTopology::LineLoop:
case Maxwell::PrimitiveTopology::LineStrip:
info.input_topology = Shader::InputTopology::Lines;
break;
case Maxwell::PrimitiveTopology::Triangles:
case Maxwell::PrimitiveTopology::TriangleStrip:
case Maxwell::PrimitiveTopology::TriangleFan:
case Maxwell::PrimitiveTopology::Quads:
case Maxwell::PrimitiveTopology::QuadStrip:
case Maxwell::PrimitiveTopology::Polygon:
case Maxwell::PrimitiveTopology::Patches:
info.input_topology = Shader::InputTopology::Triangles;
break;
case Maxwell::PrimitiveTopology::LinesAdjacency:
case Maxwell::PrimitiveTopology::LineStripAdjacency:
info.input_topology = Shader::InputTopology::LinesAdjacency;
break;
case Maxwell::PrimitiveTopology::TrianglesAdjacency:
case Maxwell::PrimitiveTopology::TriangleStripAdjacency:
info.input_topology = Shader::InputTopology::TrianglesAdjacency;
break;
}
info.force_early_z = key.state.early_z != 0;
info.y_negate = key.state.y_negate != 0;
return info;
}
} // Anonymous namespace
size_t ComputePipelineCacheKey::Hash() const noexcept {
@ -124,7 +326,7 @@ PipelineCache::PipelineCache(RasterizerVulkan& rasterizer_, Tegra::Engines::Maxw
serialization_thread(1, "yuzu:PipelineSerialization") {
const auto& float_control{device.FloatControlProperties()};
const VkDriverIdKHR driver_id{device.GetDriverID()};
base_profile = Shader::Profile{
profile = Shader::Profile{
.supported_spirv = device.IsKhrSpirv1_4Supported() ? 0x00010400U : 0x00010000U,
.unified_descriptor_binding = true,
.support_descriptor_aliasing = true,
@ -153,14 +355,10 @@ PipelineCache::PipelineCache(RasterizerVulkan& rasterizer_, Tegra::Engines::Maxw
.support_viewport_mask = device.IsNvViewportArray2Supported(),
.support_typeless_image_loads = device.IsFormatlessImageLoadSupported(),
.support_demote_to_helper_invocation = true,
.warp_size_potentially_larger_than_guest = device.IsWarpSizePotentiallyBiggerThanGuest(),
.support_int64_atomics = device.IsExtShaderAtomicInt64Supported(),
.warp_size_potentially_larger_than_guest = device.IsWarpSizePotentiallyBiggerThanGuest(),
.has_broken_spirv_clamp = driver_id == VK_DRIVER_ID_INTEL_PROPRIETARY_WINDOWS_KHR,
.has_broken_unsigned_image_offsets = false,
.generic_input_types{},
.fixed_state_point_size{},
.alpha_test_func{},
.xfb_varyings{},
};
}
@ -329,8 +527,8 @@ std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline(
const size_t stage_index{index - 1};
infos[stage_index] = &program.info;
const Shader::Profile profile{MakeProfile(key, program)};
const std::vector<u32> code{EmitSPIRV(profile, program, binding)};
const Shader::RuntimeInfo runtime_info{MakeRuntimeInfo(key, program)};
const std::vector<u32> code{EmitSPIRV(profile, runtime_info, program, binding)};
device.SaveShader(code);
modules[stage_index] = BuildShader(device, code);
if (device.HasDebuggingToolAttached()) {
@ -391,7 +589,7 @@ std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline(
Shader::Maxwell::Flow::CFG cfg{env, pools.flow_block, env.StartAddress()};
Shader::IR::Program program{TranslateProgram(pools.inst, pools.block, env, cfg)};
const std::vector<u32> code{EmitSPIRV(base_profile, program)};
const std::vector<u32> code{EmitSPIRV(profile, program)};
device.SaveShader(code);
vk::ShaderModule spv_module{BuildShader(device, code)};
if (device.HasDebuggingToolAttached()) {
@ -403,206 +601,4 @@ std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline(
thread_worker, program.info, std::move(spv_module));
}
static Shader::AttributeType CastAttributeType(const FixedPipelineState::VertexAttribute& attr) {
if (attr.enabled == 0) {
return Shader::AttributeType::Disabled;
}
switch (attr.Type()) {
case Maxwell::VertexAttribute::Type::SignedNorm:
case Maxwell::VertexAttribute::Type::UnsignedNorm:
case Maxwell::VertexAttribute::Type::UnsignedScaled:
case Maxwell::VertexAttribute::Type::SignedScaled:
case Maxwell::VertexAttribute::Type::Float:
return Shader::AttributeType::Float;
case Maxwell::VertexAttribute::Type::SignedInt:
return Shader::AttributeType::SignedInt;
case Maxwell::VertexAttribute::Type::UnsignedInt:
return Shader::AttributeType::UnsignedInt;
}
return Shader::AttributeType::Float;
}
static std::vector<Shader::TransformFeedbackVarying> MakeTransformFeedbackVaryings(
const GraphicsPipelineCacheKey& key) {
static constexpr std::array VECTORS{
28, // gl_Position
32, // Generic 0
36, // Generic 1
40, // Generic 2
44, // Generic 3
48, // Generic 4
52, // Generic 5
56, // Generic 6
60, // Generic 7
64, // Generic 8
68, // Generic 9
72, // Generic 10
76, // Generic 11
80, // Generic 12
84, // Generic 13
88, // Generic 14
92, // Generic 15
96, // Generic 16
100, // Generic 17
104, // Generic 18
108, // Generic 19
112, // Generic 20
116, // Generic 21
120, // Generic 22
124, // Generic 23
128, // Generic 24
132, // Generic 25
136, // Generic 26
140, // Generic 27
144, // Generic 28
148, // Generic 29
152, // Generic 30
156, // Generic 31
160, // gl_FrontColor
164, // gl_FrontSecondaryColor
160, // gl_BackColor
164, // gl_BackSecondaryColor
192, // gl_TexCoord[0]
196, // gl_TexCoord[1]
200, // gl_TexCoord[2]
204, // gl_TexCoord[3]
208, // gl_TexCoord[4]
212, // gl_TexCoord[5]
216, // gl_TexCoord[6]
220, // gl_TexCoord[7]
};
std::vector<Shader::TransformFeedbackVarying> xfb(256);
for (size_t buffer = 0; buffer < Maxwell::NumTransformFeedbackBuffers; ++buffer) {
const auto& locations = key.state.xfb_state.varyings[buffer];
const auto& layout = key.state.xfb_state.layouts[buffer];
const u32 varying_count = layout.varying_count;
u32 highest = 0;
for (u32 offset = 0; offset < varying_count; ++offset) {
const u32 base_offset = offset;
const u8 location = locations[offset];
Shader::TransformFeedbackVarying varying;
varying.buffer = layout.stream;
varying.stride = layout.stride;
varying.offset = offset * 4;
varying.components = 1;
if (std::ranges::find(VECTORS, Common::AlignDown(location, 4)) != VECTORS.end()) {
UNIMPLEMENTED_IF_MSG(location % 4 != 0, "Unaligned TFB");
const u8 base_index = location / 4;
while (offset + 1 < varying_count && base_index == locations[offset + 1] / 4) {
++offset;
++varying.components;
}
}
xfb[location] = varying;
highest = std::max(highest, (base_offset + varying.components) * 4);
}
UNIMPLEMENTED_IF(highest != layout.stride);
}
return xfb;
}
Shader::Profile PipelineCache::MakeProfile(const GraphicsPipelineCacheKey& key,
const Shader::IR::Program& program) {
Shader::Profile profile{base_profile};
const Shader::Stage stage{program.stage};
const bool has_geometry{key.unique_hashes[4] != 0};
const bool gl_ndc{key.state.ndc_minus_one_to_one != 0};
const float point_size{Common::BitCast<float>(key.state.point_size)};
switch (stage) {
case Shader::Stage::VertexB:
if (!has_geometry) {
if (key.state.topology == Maxwell::PrimitiveTopology::Points) {
profile.fixed_state_point_size = point_size;
}
if (key.state.xfb_enabled != 0) {
profile.xfb_varyings = MakeTransformFeedbackVaryings(key);
}
profile.convert_depth_mode = gl_ndc;
}
std::ranges::transform(key.state.attributes, profile.generic_input_types.begin(),
&CastAttributeType);
break;
case Shader::Stage::TessellationEval:
// We have to flip tessellation clockwise for some reason...
profile.tess_clockwise = key.state.tessellation_clockwise == 0;
profile.tess_primitive = [&key] {
const u32 raw{key.state.tessellation_primitive.Value()};
switch (static_cast<Maxwell::TessellationPrimitive>(raw)) {
case Maxwell::TessellationPrimitive::Isolines:
return Shader::TessPrimitive::Isolines;
case Maxwell::TessellationPrimitive::Triangles:
return Shader::TessPrimitive::Triangles;
case Maxwell::TessellationPrimitive::Quads:
return Shader::TessPrimitive::Quads;
}
UNREACHABLE();
return Shader::TessPrimitive::Triangles;
}();
profile.tess_spacing = [&] {
const u32 raw{key.state.tessellation_spacing};
switch (static_cast<Maxwell::TessellationSpacing>(raw)) {
case Maxwell::TessellationSpacing::Equal:
return Shader::TessSpacing::Equal;
case Maxwell::TessellationSpacing::FractionalOdd:
return Shader::TessSpacing::FractionalOdd;
case Maxwell::TessellationSpacing::FractionalEven:
return Shader::TessSpacing::FractionalEven;
}
UNREACHABLE();
return Shader::TessSpacing::Equal;
}();
break;
case Shader::Stage::Geometry:
if (program.output_topology == Shader::OutputTopology::PointList) {
profile.fixed_state_point_size = point_size;
}
if (key.state.xfb_enabled != 0) {
profile.xfb_varyings = MakeTransformFeedbackVaryings(key);
}
profile.convert_depth_mode = gl_ndc;
break;
case Shader::Stage::Fragment:
profile.alpha_test_func = MaxwellToCompareFunction(
key.state.UnpackComparisonOp(key.state.alpha_test_func.Value()));
profile.alpha_test_reference = Common::BitCast<float>(key.state.alpha_test_ref);
break;
default:
break;
}
switch (key.state.topology) {
case Maxwell::PrimitiveTopology::Points:
profile.input_topology = Shader::InputTopology::Points;
break;
case Maxwell::PrimitiveTopology::Lines:
case Maxwell::PrimitiveTopology::LineLoop:
case Maxwell::PrimitiveTopology::LineStrip:
profile.input_topology = Shader::InputTopology::Lines;
break;
case Maxwell::PrimitiveTopology::Triangles:
case Maxwell::PrimitiveTopology::TriangleStrip:
case Maxwell::PrimitiveTopology::TriangleFan:
case Maxwell::PrimitiveTopology::Quads:
case Maxwell::PrimitiveTopology::QuadStrip:
case Maxwell::PrimitiveTopology::Polygon:
case Maxwell::PrimitiveTopology::Patches:
profile.input_topology = Shader::InputTopology::Triangles;
break;
case Maxwell::PrimitiveTopology::LinesAdjacency:
case Maxwell::PrimitiveTopology::LineStripAdjacency:
profile.input_topology = Shader::InputTopology::LinesAdjacency;
break;
case Maxwell::PrimitiveTopology::TrianglesAdjacency:
case Maxwell::PrimitiveTopology::TriangleStripAdjacency:
profile.input_topology = Shader::InputTopology::TrianglesAdjacency;
break;
}
profile.force_early_z = key.state.early_z != 0;
profile.y_negate = key.state.y_negate != 0;
return profile;
}
} // namespace Vulkan

5
src/video_core/renderer_vulkan/vk_pipeline_cache.h
View File

@ -129,9 +129,6 @@ private:
Shader::Environment& env,
bool build_in_parallel);
Shader::Profile MakeProfile(const GraphicsPipelineCacheKey& key,
const Shader::IR::Program& program);
const Device& device;
VKScheduler& scheduler;
DescriptorPool& descriptor_pool;
@ -148,7 +145,7 @@ private:
ShaderPools main_pools;
Shader::Profile base_profile;
Shader::Profile profile;
std::filesystem::path pipeline_cache_filename;
Common::ThreadWorker workers;