shader: Rework varyings and implement passthrough geometry shaders

Put all varyings into a single std::bitset with helpers to access it.

Implement passthrough geometry shaders using host's.
This commit is contained in:
ReinUsesLisp 2021-06-24 02:41:09 -03:00 committed by ameerj
parent 4f052a1f39
commit 7dafa96ab5
29 changed files with 351 additions and 337 deletions

View file

@ -229,6 +229,7 @@ add_library(shader_recompiler STATIC
program_header.h
runtime_info.h
shader_info.h
varying_state.h
)
target_link_libraries(shader_recompiler PUBLIC common fmt::fmt sirit)

View file

@ -83,14 +83,13 @@ 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 < info.input_generics.size(); ++index) {
const auto& generic{info.input_generics[index]};
if (generic.used) {
for (size_t index = 0; index < IR::NUM_GENERICS; ++index) {
if (info.loads.Generic(index)) {
Add("{}ATTRIB in_attr{}[]={{{}.attrib[{}..{}]}};",
InterpDecorator(generic.interpolation), index, attr_stage, index, index);
InterpDecorator(info.interpolation[index]), index, attr_stage, index, index);
}
}
if (IsInputArray(stage) && info.loads_position) {
if (IsInputArray(stage) && info.loads.AnyComponent(IR::Attribute::PositionX)) {
Add("ATTRIB vertex_position=vertex.position;");
}
if (info.uses_invocation_id) {
@ -102,7 +101,7 @@ EmitContext::EmitContext(IR::Program& program, Bindings& bindings, const Profile
if (info.stores_tess_level_inner) {
Add("OUTPUT result_patch_tessinner[]={{result.patch.tessinner[0..1]}};");
}
if (info.stores_clip_distance) {
if (info.stores.ClipDistances()) {
Add("OUTPUT result_clip[]={{result.clip[0..7]}};");
}
for (size_t index = 0; index < info.uses_patches.size(); ++index) {
@ -124,8 +123,8 @@ EmitContext::EmitContext(IR::Program& program, Bindings& bindings, const Profile
Add("OUTPUT frag_color{}=result.color[{}];", index, index);
}
}
for (size_t index = 0; index < info.stores_generics.size(); ++index) {
if (info.stores_generics[index]) {
for (size_t index = 0; index < IR::NUM_GENERICS; ++index) {
if (info.stores.Generic(index)) {
Add("OUTPUT out_attr{}[]={{result.attrib[{}..{}]}};", index, index, index);
}
}

View file

@ -296,8 +296,10 @@ void SetupOptions(const IR::Program& program, const Profile& profile,
if (info.uses_sparse_residency) {
header += "OPTION EXT_sparse_texture2;";
}
if (((info.stores_viewport_index || info.stores_layer) && stage != Stage::Geometry) ||
info.stores_viewport_mask) {
const bool stores_viewport_layer{info.stores[IR::Attribute::ViewportIndex] ||
info.stores[IR::Attribute::Layer]};
if ((stage != Stage::Geometry && stores_viewport_layer) ||
info.stores[IR::Attribute::ViewportMask]) {
if (profile.support_viewport_index_layer_non_geometry) {
header += "OPTION NV_viewport_array2;";
}

View file

@ -261,7 +261,7 @@ void EmitGetAttributeIndexed(EmitContext& ctx, IR::Inst& inst, ScalarS32 offset,
fmt::format("{}.z", value), fmt::format("{}.w", value)};
read(compare_index, values);
}};
if (ctx.info.loads_position) {
if (ctx.info.loads.AnyComponent(IR::Attribute::PositionX)) {
const u32 index{static_cast<u32>(IR::Attribute::PositionX)};
if (IsInputArray(ctx.stage)) {
read_swizzled(index, fmt::format("vertex_position{}", VertexIndex(ctx, vertex)));
@ -269,8 +269,8 @@ void EmitGetAttributeIndexed(EmitContext& ctx, IR::Inst& inst, ScalarS32 offset,
read_swizzled(index, fmt::format("{}.position", ctx.attrib_name));
}
}
for (u32 index = 0; index < ctx.info.input_generics.size(); ++index) {
if (!ctx.info.input_generics[index].used) {
for (u32 index = 0; index < static_cast<u32>(IR::NUM_GENERICS); ++index) {
if (!ctx.info.loads.Generic(index)) {
continue;
}
read_swizzled(index, fmt::format("in_attr{}{}[0]", index, VertexIndex(ctx, vertex)));

View file

@ -212,22 +212,22 @@ std::string_view OutputPrimitive(OutputTopology topology) {
}
void SetupLegacyOutPerVertex(EmitContext& ctx, std::string& header) {
if (!ctx.info.stores_legacy_varyings) {
if (!ctx.info.stores.Legacy()) {
return;
}
if (ctx.info.stores_fixed_fnc_textures) {
if (ctx.info.stores.FixedFunctionTexture()) {
header += "vec4 gl_TexCoord[8];";
}
if (ctx.info.stores_color_front_diffuse) {
if (ctx.info.stores.AnyComponent(IR::Attribute::ColorFrontDiffuseR)) {
header += "vec4 gl_FrontColor;";
}
if (ctx.info.stores_color_front_specular) {
if (ctx.info.stores.AnyComponent(IR::Attribute::ColorFrontSpecularR)) {
header += "vec4 gl_FrontSecondaryColor;";
}
if (ctx.info.stores_color_back_diffuse) {
if (ctx.info.stores.AnyComponent(IR::Attribute::ColorBackDiffuseR)) {
header += "vec4 gl_BackColor;";
}
if (ctx.info.stores_color_back_specular) {
if (ctx.info.stores.AnyComponent(IR::Attribute::ColorBackSpecularR)) {
header += "vec4 gl_BackSecondaryColor;";
}
}
@ -237,32 +237,32 @@ void SetupOutPerVertex(EmitContext& ctx, std::string& header) {
return;
}
header += "out gl_PerVertex{vec4 gl_Position;";
if (ctx.info.stores_point_size) {
if (ctx.info.stores[IR::Attribute::PointSize]) {
header += "float gl_PointSize;";
}
if (ctx.info.stores_clip_distance) {
if (ctx.info.stores.ClipDistances()) {
header += "float gl_ClipDistance[];";
}
if (ctx.info.stores_viewport_index && ctx.profile.support_viewport_index_layer_non_geometry &&
ctx.stage != Stage::Geometry) {
if (ctx.info.stores[IR::Attribute::ViewportIndex] &&
ctx.profile.support_viewport_index_layer_non_geometry && ctx.stage != Stage::Geometry) {
header += "int gl_ViewportIndex;";
}
SetupLegacyOutPerVertex(ctx, header);
header += "};";
if (ctx.info.stores_viewport_index && ctx.stage == Stage::Geometry) {
if (ctx.info.stores[IR::Attribute::ViewportIndex] && ctx.stage == Stage::Geometry) {
header += "out int gl_ViewportIndex;";
}
}
void SetupLegacyInPerFragment(EmitContext& ctx, std::string& header) {
if (!ctx.info.loads_legacy_varyings) {
if (!ctx.info.loads.Legacy()) {
return;
}
header += "in gl_PerFragment{";
if (ctx.info.loads_fixed_fnc_textures) {
if (ctx.info.loads.FixedFunctionTexture()) {
header += "vec4 gl_TexCoord[8];";
}
if (ctx.info.loads_color_front_diffuse) {
if (ctx.info.loads.AnyComponent(IR::Attribute::ColorFrontDiffuseR)) {
header += "vec4 gl_Color;";
}
header += "};";
@ -325,14 +325,13 @@ EmitContext::EmitContext(IR::Program& program, Bindings& bindings, const Profile
SetupOutPerVertex(*this, header);
SetupLegacyInPerFragment(*this, header);
for (size_t index = 0; index < info.input_generics.size(); ++index) {
const auto& generic{info.input_generics[index]};
if (!generic.used || !runtime_info.previous_stage_stores_generic[index]) {
for (size_t index = 0; index < IR::NUM_GENERICS; ++index) {
if (!info.loads.Generic(index) || !runtime_info.previous_stage_stores.Generic(index)) {
continue;
}
header +=
fmt::format("layout(location={}){}in vec4 in_attr{}{};", index,
InterpDecorator(generic.interpolation), index, InputArrayDecorator(stage));
header += fmt::format("layout(location={}){}in vec4 in_attr{}{};", index,
InterpDecorator(info.interpolation[index]), index,
InputArrayDecorator(stage));
}
for (size_t index = 0; index < info.uses_patches.size(); ++index) {
if (!info.uses_patches[index]) {
@ -349,11 +348,10 @@ EmitContext::EmitContext(IR::Program& program, Bindings& bindings, const Profile
header += fmt::format("layout(location={})out vec4 frag_color{};", index, index);
}
}
for (size_t index = 0; index < info.stores_generics.size(); ++index) {
if (!info.stores_generics[index]) {
continue;
for (size_t index = 0; index < IR::NUM_GENERICS; ++index) {
if (info.stores.Generic(index)) {
DefineGenericOutput(index, program.invocations);
}
DefineGenericOutput(index, program.invocations);
}
DefineConstantBuffers(bindings);
DefineStorageBuffers(bindings);
@ -398,14 +396,14 @@ void EmitContext::SetupExtensions() {
header += "#extension GL_NV_shader_thread_shuffle : enable\n";
}
}
if ((info.stores_viewport_index || info.stores_layer) &&
if ((info.stores[IR::Attribute::ViewportIndex] || info.stores[IR::Attribute::Layer]) &&
profile.support_viewport_index_layer_non_geometry && stage != Stage::Geometry) {
header += "#extension GL_ARB_shader_viewport_layer_array : enable\n";
}
if (info.uses_sparse_residency && profile.support_gl_sparse_textures) {
header += "#extension GL_ARB_sparse_texture2 : enable\n";
}
if (info.stores_viewport_mask && profile.support_viewport_mask) {
if (info.stores[IR::Attribute::ViewportMask] && profile.support_viewport_mask) {
header += "#extension GL_NV_viewport_array2 : enable\n";
}
if (info.uses_typeless_image_reads) {
@ -535,20 +533,20 @@ void EmitContext::DefineHelperFunctions() {
fmt::format("float IndexedAttrLoad(int offset{}){{int base_index=offset>>2;uint "
"masked_index=uint(base_index)&3u;switch(base_index>>2){{",
vertex_arg)};
if (info.loads_position) {
if (info.loads.AnyComponent(IR::Attribute::PositionX)) {
const auto position_idx{is_array ? "gl_in[vertex]." : ""};
func += fmt::format("case {}:return {}{}[masked_index];",
static_cast<u32>(IR::Attribute::PositionX) >> 2, position_idx,
position_name);
}
const u32 base_attribute_value = static_cast<u32>(IR::Attribute::Generic0X) >> 2;
for (u32 i = 0; i < info.input_generics.size(); ++i) {
if (!info.input_generics[i].used) {
for (u32 index = 0; index < IR::NUM_GENERICS; ++index) {
if (!info.loads.Generic(index)) {
continue;
}
const auto vertex_idx{is_array ? "[vertex]" : ""};
func += fmt::format("case {}:return in_attr{}{}[masked_index];",
base_attribute_value + i, i, vertex_idx);
base_attribute_value + index, index, vertex_idx);
}
func += "default: return 0.0;}}";
header += func;

View file

@ -171,7 +171,7 @@ void EmitCode(EmitContext& ctx, const IR::Program& program) {
}
std::string GlslVersionSpecifier(const EmitContext& ctx) {
if (ctx.uses_y_direction || ctx.info.stores_legacy_varyings || ctx.info.loads_legacy_varyings) {
if (ctx.uses_y_direction || ctx.info.stores.Legacy() || ctx.info.loads.Legacy()) {
return " compatibility";
}
return "";

View file

@ -179,7 +179,7 @@ void EmitGetAttribute(EmitContext& ctx, IR::Inst& inst, IR::Attribute attr,
const char swizzle{"xyzw"[element]};
if (IR::IsGeneric(attr)) {
const u32 index{IR::GenericAttributeIndex(attr)};
if (!ctx.runtime_info.previous_stage_stores_generic[index]) {
if (!ctx.runtime_info.previous_stage_stores.Generic(index)) {
ctx.AddF32("{}=0.f;", inst, attr);
return;
}

View file

@ -20,8 +20,8 @@ void InitializeOutputVaryings(EmitContext& ctx) {
if (ctx.stage == Stage::VertexB || ctx.stage == Stage::Geometry) {
ctx.Add("gl_Position=vec4(0,0,0,1);");
}
for (size_t index = 0; index < ctx.info.stores_generics.size(); ++index) {
if (!ctx.info.stores_generics[index]) {
for (size_t index = 0; index < IR::NUM_GENERICS; ++index) {
if (!ctx.info.stores.Generic(index)) {
continue;
}
const auto& info_array{ctx.output_generics.at(index)};

View file

@ -557,7 +557,7 @@ void EmitContext::DefineCommonConstants() {
}
void EmitContext::DefineInterfaces(const IR::Program& program) {
DefineInputs(program.info);
DefineInputs(program);
DefineOutputs(program);
}
@ -693,16 +693,16 @@ void EmitContext::DefineAttributeMemAccess(const Info& info) {
const Id compare_index{OpShiftRightArithmetic(U32[1], base_index, Const(2U))};
std::vector<Sirit::Literal> literals;
std::vector<Id> labels;
if (info.loads_position) {
if (info.loads.AnyComponent(IR::Attribute::PositionX)) {
literals.push_back(static_cast<u32>(IR::Attribute::PositionX) >> 2);
labels.push_back(OpLabel());
}
const u32 base_attribute_value = static_cast<u32>(IR::Attribute::Generic0X) >> 2;
for (u32 i = 0; i < info.input_generics.size(); ++i) {
if (!info.input_generics[i].used) {
for (u32 index = 0; index < static_cast<u32>(IR::NUM_GENERICS); ++index) {
if (!info.loads.Generic(index)) {
continue;
}
literals.push_back(base_attribute_value + i);
literals.push_back(base_attribute_value + index);
labels.push_back(OpLabel());
}
OpSelectionMerge(end_block, spv::SelectionControlMask::MaskNone);
@ -710,7 +710,7 @@ void EmitContext::DefineAttributeMemAccess(const Info& info) {
AddLabel(default_label);
OpReturnValue(Const(0.0f));
size_t label_index{0};
if (info.loads_position) {
if (info.loads.AnyComponent(IR::Attribute::PositionX)) {
AddLabel(labels[label_index]);
const Id pointer{is_array
? OpAccessChain(input_f32, input_position, vertex, masked_index)
@ -719,18 +719,18 @@ void EmitContext::DefineAttributeMemAccess(const Info& info) {
OpReturnValue(result);
++label_index;
}
for (size_t i = 0; i < info.input_generics.size(); i++) {
if (!info.input_generics[i].used) {
for (size_t index = 0; index < IR::NUM_GENERICS; ++index) {
if (!info.loads.Generic(index)) {
continue;
}
AddLabel(labels[label_index]);
const auto type{AttrTypes(*this, static_cast<u32>(i))};
const auto type{AttrTypes(*this, static_cast<u32>(index))};
if (!type) {
OpReturnValue(Const(0.0f));
++label_index;
continue;
}
const Id generic_id{input_generics.at(i)};
const Id generic_id{input_generics.at(index)};
const Id pointer{is_array
? OpAccessChain(type->pointer, generic_id, vertex, masked_index)
: OpAccessChain(type->pointer, generic_id, masked_index)};
@ -758,19 +758,19 @@ void EmitContext::DefineAttributeMemAccess(const Info& info) {
const Id compare_index{OpShiftRightArithmetic(U32[1], base_index, Const(2U))};
std::vector<Sirit::Literal> literals;
std::vector<Id> labels;
if (info.stores_position) {
if (info.stores.AnyComponent(IR::Attribute::PositionX)) {
literals.push_back(static_cast<u32>(IR::Attribute::PositionX) >> 2);
labels.push_back(OpLabel());
}
const u32 base_attribute_value = static_cast<u32>(IR::Attribute::Generic0X) >> 2;
for (size_t i = 0; i < info.stores_generics.size(); i++) {
if (!info.stores_generics[i]) {
for (size_t index = 0; index < IR::NUM_GENERICS; ++index) {
if (!info.stores.Generic(index)) {
continue;
}
literals.push_back(base_attribute_value + static_cast<u32>(i));
literals.push_back(base_attribute_value + static_cast<u32>(index));
labels.push_back(OpLabel());
}
if (info.stores_clip_distance) {
if (info.stores.ClipDistances()) {
literals.push_back(static_cast<u32>(IR::Attribute::ClipDistance0) >> 2);
labels.push_back(OpLabel());
literals.push_back(static_cast<u32>(IR::Attribute::ClipDistance4) >> 2);
@ -781,28 +781,28 @@ void EmitContext::DefineAttributeMemAccess(const Info& info) {
AddLabel(default_label);
OpReturn();
size_t label_index{0};
if (info.stores_position) {
if (info.stores.AnyComponent(IR::Attribute::PositionX)) {
AddLabel(labels[label_index]);
const Id pointer{OpAccessChain(output_f32, output_position, masked_index)};
OpStore(pointer, store_value);
OpReturn();
++label_index;
}
for (size_t i = 0; i < info.stores_generics.size(); ++i) {
if (!info.stores_generics[i]) {
for (size_t index = 0; index < IR::NUM_GENERICS; ++index) {
if (!info.stores.Generic(index)) {
continue;
}
if (output_generics[i][0].num_components != 4) {
if (output_generics[index][0].num_components != 4) {
throw NotImplementedException("Physical stores and transform feedbacks");
}
AddLabel(labels[label_index]);
const Id generic_id{output_generics[i][0].id};
const Id generic_id{output_generics[index][0].id};
const Id pointer{OpAccessChain(output_f32, generic_id, masked_index)};
OpStore(pointer, store_value);
OpReturn();
++label_index;
}
if (info.stores_clip_distance) {
if (info.stores.ClipDistances()) {
AddLabel(labels[label_index]);
const Id pointer{OpAccessChain(output_f32, clip_distances, masked_index)};
OpStore(pointer, store_value);
@ -1146,7 +1146,10 @@ void EmitContext::DefineImages(const Info& info, u32& binding) {
}
}
void EmitContext::DefineInputs(const Info& info) {
void EmitContext::DefineInputs(const IR::Program& program) {
const Info& info{program.info};
const VaryingState loads{info.loads.mask | info.passthrough.mask};
if (info.uses_workgroup_id) {
workgroup_id = DefineInput(*this, U32[3], false, spv::BuiltIn::WorkgroupId);
}
@ -1183,15 +1186,20 @@ void EmitContext::DefineInputs(const Info& info) {
fswzadd_lut_b =
ConstantComposite(F32[4], f32_minus_one, f32_minus_one, f32_one, f32_minus_one);
}
if (info.loads_primitive_id) {
if (loads[IR::Attribute::PrimitiveId]) {
primitive_id = DefineInput(*this, U32[1], false, spv::BuiltIn::PrimitiveId);
}
if (info.loads_position) {
if (loads.AnyComponent(IR::Attribute::PositionX)) {
const bool is_fragment{stage != Stage::Fragment};
const spv::BuiltIn built_in{is_fragment ? spv::BuiltIn::Position : spv::BuiltIn::FragCoord};
input_position = DefineInput(*this, F32[4], true, built_in);
if (profile.support_geometry_shader_passthrough) {
if (info.passthrough.AnyComponent(IR::Attribute::PositionX)) {
Decorate(input_position, spv::Decoration::PassthroughNV);
}
}
}
if (info.loads_instance_id) {
if (loads[IR::Attribute::InstanceId]) {
if (profile.support_vertex_instance_id) {
instance_id = DefineInput(*this, U32[1], true, spv::BuiltIn::InstanceId);
} else {
@ -1199,7 +1207,7 @@ void EmitContext::DefineInputs(const Info& info) {
base_instance = DefineInput(*this, U32[1], true, spv::BuiltIn::BaseInstance);
}
}
if (info.loads_vertex_id) {
if (loads[IR::Attribute::VertexId]) {
if (profile.support_vertex_instance_id) {
vertex_id = DefineInput(*this, U32[1], true, spv::BuiltIn::VertexId);
} else {
@ -1207,24 +1215,24 @@ void EmitContext::DefineInputs(const Info& info) {
base_vertex = DefineInput(*this, U32[1], true, spv::BuiltIn::BaseVertex);
}
}
if (info.loads_front_face) {
if (loads[IR::Attribute::FrontFace]) {
front_face = DefineInput(*this, U1, true, spv::BuiltIn::FrontFacing);
}
if (info.loads_point_coord) {
if (loads[IR::Attribute::PointSpriteS] || loads[IR::Attribute::PointSpriteT]) {
point_coord = DefineInput(*this, F32[2], true, spv::BuiltIn::PointCoord);
}
if (info.loads_tess_coord) {
if (loads[IR::Attribute::TessellationEvaluationPointU] ||
loads[IR::Attribute::TessellationEvaluationPointV]) {
tess_coord = DefineInput(*this, F32[3], false, spv::BuiltIn::TessCoord);
}
for (size_t index = 0; index < info.input_generics.size(); ++index) {
if (!runtime_info.previous_stage_stores_generic[index]) {
continue;
}
const InputVarying generic{info.input_generics[index]};
if (!generic.used) {
continue;
}
for (size_t index = 0; index < IR::NUM_GENERICS; ++index) {
const AttributeType input_type{runtime_info.generic_input_types[index]};
if (!runtime_info.previous_stage_stores.Generic(index)) {
continue;
}
if (!loads.Generic(index)) {
continue;
}
if (input_type == AttributeType::Disabled) {
continue;
}
@ -1234,10 +1242,13 @@ void EmitContext::DefineInputs(const Info& info) {
Name(id, fmt::format("in_attr{}", index));
input_generics[index] = id;
if (info.passthrough.Generic(index) && profile.support_geometry_shader_passthrough) {
Decorate(id, spv::Decoration::PassthroughNV);
}
if (stage != Stage::Fragment) {
continue;
}
switch (generic.interpolation) {
switch (info.interpolation[index]) {
case Interpolation::Smooth:
// Default
// Decorate(id, spv::Decoration::Smooth);
@ -1266,42 +1277,42 @@ void EmitContext::DefineInputs(const Info& info) {
void EmitContext::DefineOutputs(const IR::Program& program) {
const Info& info{program.info};
const std::optional<u32> invocations{program.invocations};
if (info.stores_position || stage == Stage::VertexB) {
if (info.stores.AnyComponent(IR::Attribute::PositionX) || stage == Stage::VertexB) {
output_position = DefineOutput(*this, F32[4], invocations, spv::BuiltIn::Position);
}
if (info.stores_point_size || runtime_info.fixed_state_point_size) {
if (info.stores[IR::Attribute::PointSize] || runtime_info.fixed_state_point_size) {
if (stage == Stage::Fragment) {
throw NotImplementedException("Storing PointSize in fragment stage");
}
output_point_size = DefineOutput(*this, F32[1], invocations, spv::BuiltIn::PointSize);
}
if (info.stores_clip_distance) {
if (info.stores.ClipDistances()) {
if (stage == Stage::Fragment) {
throw NotImplementedException("Storing ClipDistance in fragment stage");
}
const Id type{TypeArray(F32[1], Const(8U))};
clip_distances = DefineOutput(*this, type, invocations, spv::BuiltIn::ClipDistance);
}
if (info.stores_layer &&
if (info.stores[IR::Attribute::Layer] &&
(profile.support_viewport_index_layer_non_geometry || stage == Stage::Geometry)) {
if (stage == Stage::Fragment) {
throw NotImplementedException("Storing Layer in fragment stage");
}
layer = DefineOutput(*this, U32[1], invocations, spv::BuiltIn::Layer);
}
if (info.stores_viewport_index &&
if (info.stores[IR::Attribute::ViewportIndex] &&
(profile.support_viewport_index_layer_non_geometry || stage == Stage::Geometry)) {
if (stage == Stage::Fragment) {
throw NotImplementedException("Storing ViewportIndex in fragment stage");
}
viewport_index = DefineOutput(*this, U32[1], invocations, spv::BuiltIn::ViewportIndex);
}
if (info.stores_viewport_mask && profile.support_viewport_mask) {
if (info.stores[IR::Attribute::ViewportMask] && profile.support_viewport_mask) {
viewport_mask = DefineOutput(*this, TypeArray(U32[1], Const(1u)), std::nullopt,
spv::BuiltIn::ViewportMaskNV);
}
for (size_t index = 0; index < info.stores_generics.size(); ++index) {
if (info.stores_generics[index]) {
for (size_t index = 0; index < IR::NUM_GENERICS; ++index) {
if (info.stores.Generic(index)) {
DefineGenericOutput(*this, index, invocations);
}
}

View file

@ -300,7 +300,7 @@ private:
void DefineAttributeMemAccess(const Info& info);
void DefineGlobalMemoryFunctions(const Info& info);
void DefineInputs(const Info& info);
void DefineInputs(const IR::Program& program);
void DefineOutputs(const IR::Program& program);
};

View file

@ -281,11 +281,19 @@ void DefineEntryPoint(const IR::Program& program, EmitContext& ctx, Id main) {
ctx.AddExecutionMode(main, spv::ExecutionMode::OutputTriangleStrip);
break;
}
if (program.info.stores_point_size) {
if (program.info.stores[IR::Attribute::PointSize]) {
ctx.AddCapability(spv::Capability::GeometryPointSize);
}
ctx.AddExecutionMode(main, spv::ExecutionMode::OutputVertices, program.output_vertices);
ctx.AddExecutionMode(main, spv::ExecutionMode::Invocations, program.invocations);
if (program.is_geometry_passthrough) {
if (ctx.profile.support_geometry_shader_passthrough) {
ctx.AddExtension("SPV_NV_geometry_shader_passthrough");
ctx.AddCapability(spv::Capability::GeometryShaderPassthroughNV);
} else {
LOG_WARNING(Shader_SPIRV, "Geometry shader passthrough used with no support");
}
}
break;
case Stage::Fragment:
execution_model = spv::ExecutionModel::Fragment;
@ -377,20 +385,21 @@ void SetupCapabilities(const Profile& profile, const Info& info, EmitContext& ct
ctx.AddExtension("SPV_EXT_demote_to_helper_invocation");
ctx.AddCapability(spv::Capability::DemoteToHelperInvocationEXT);
}
if (info.stores_viewport_index) {
if (info.stores[IR::Attribute::ViewportIndex]) {
ctx.AddCapability(spv::Capability::MultiViewport);
}
if (info.stores_viewport_mask && profile.support_viewport_mask) {
if (info.stores[IR::Attribute::ViewportMask] && profile.support_viewport_mask) {
ctx.AddExtension("SPV_NV_viewport_array2");
ctx.AddCapability(spv::Capability::ShaderViewportMaskNV);
}
if (info.stores_layer || info.stores_viewport_index) {
if (info.stores[IR::Attribute::Layer] || info.stores[IR::Attribute::ViewportIndex]) {
if (profile.support_viewport_index_layer_non_geometry && ctx.stage != Stage::Geometry) {
ctx.AddExtension("SPV_EXT_shader_viewport_index_layer");
ctx.AddCapability(spv::Capability::ShaderViewportIndexLayerEXT);
}
}
if (!profile.support_vertex_instance_id && (info.loads_instance_id || info.loads_vertex_id)) {
if (!profile.support_vertex_instance_id &&
(info.loads[IR::Attribute::InstanceId] || info.loads[IR::Attribute::VertexId])) {
ctx.AddExtension("SPV_KHR_shader_draw_parameters");
ctx.AddCapability(spv::Capability::DrawParameters);
}

View file

@ -298,7 +298,7 @@ Id EmitGetAttribute(EmitContext& ctx, IR::Attribute attr, Id vertex) {
if (IR::IsGeneric(attr)) {
const u32 index{IR::GenericAttributeIndex(attr)};
const std::optional<AttrInfo> type{AttrTypes(ctx, index)};
if (!type || !ctx.runtime_info.previous_stage_stores_generic[index]) {
if (!type || !ctx.runtime_info.previous_stage_stores.Generic(index)) {
// Attribute is disabled
return ctx.Const(0.0f);
}

View file

@ -31,6 +31,10 @@ public:
return sph;
}
[[nodiscard]] const std::array<u32, 8>& GpPassthroughMask() const noexcept {
return gp_passthrough_mask;
}
[[nodiscard]] Stage ShaderStage() const noexcept {
return stage;
}
@ -41,6 +45,7 @@ public:
protected:
ProgramHeader sph{};
std::array<u32, 8> gp_passthrough_mask{};
Stage stage{};
u32 start_address{};
};

View file

@ -222,6 +222,8 @@ enum class Attribute : u64 {
FrontFace = 255,
};
constexpr size_t NUM_GENERICS = 32;
[[nodiscard]] bool IsGeneric(Attribute attribute) noexcept;
[[nodiscard]] u32 GenericAttributeIndex(Attribute attribute);
@ -230,6 +232,10 @@ enum class Attribute : u64 {
[[nodiscard]] std::string NameOf(Attribute attribute);
[[nodiscard]] constexpr IR::Attribute operator+(IR::Attribute attribute, size_t value) noexcept {
return static_cast<IR::Attribute>(static_cast<size_t>(attribute) + value);
}
} // namespace Shader::IR
template <>

View file

@ -27,6 +27,7 @@ struct Program {
u32 invocations{};
u32 local_memory_size{};
u32 shared_memory_size{};
bool is_geometry_passthrough{};
};
[[nodiscard]] std::string DumpProgram(const Program& program);

View file

@ -46,7 +46,7 @@ void CollectInterpolationInfo(Environment& env, IR::Program& program) {
return;
}
const ProgramHeader& sph{env.SPH()};
for (size_t index = 0; index < program.info.input_generics.size(); ++index) {
for (size_t index = 0; index < IR::NUM_GENERICS; ++index) {
std::optional<PixelImap> imap;
for (const PixelImap value : sph.ps.GenericInputMap(static_cast<u32>(index))) {
if (value == PixelImap::Unused) {
@ -60,7 +60,7 @@ void CollectInterpolationInfo(Environment& env, IR::Program& program) {
if (!imap) {
continue;
}
program.info.input_generics[index].interpolation = [&] {
program.info.interpolation[index] = [&] {
switch (*imap) {
case PixelImap::Unused:
case PixelImap::Perspective:
@ -140,6 +140,11 @@ IR::Program TranslateProgram(ObjectPool<IR::Inst>& inst_pool, ObjectPool<IR::Blo
program.output_topology = sph.common3.output_topology;
program.output_vertices = sph.common4.max_output_vertices;
program.invocations = sph.common2.threads_per_input_primitive;
program.is_geometry_passthrough = sph.common0.geometry_passthrough != 0;
if (program.is_geometry_passthrough) {
const auto mask{env.GpPassthroughMask()};
program.info.passthrough.mask |= ~Common::BitCast<std::bitset<256>>(mask);
}
break;
}
case Stage::Compute:
@ -194,12 +199,9 @@ IR::Program MergeDualVertexPrograms(IR::Program& vertex_a, IR::Program& vertex_b
result.stage = Stage::VertexB;
result.info = vertex_a.info;
result.local_memory_size = std::max(vertex_a.local_memory_size, vertex_b.local_memory_size);
for (size_t index = 0; index < 32; ++index) {
result.info.input_generics[index].used |= vertex_b.info.input_generics[index].used;
if (vertex_b.info.stores_generics[index]) {
result.info.stores_generics[index] = true;
}
}
result.info.loads.mask |= vertex_b.info.loads.mask;
result.info.stores.mask |= vertex_b.info.stores.mask;
Optimization::JoinTextureInfo(result.info, vertex_b.info);
Optimization::JoinStorageInfo(result.info, vertex_b.info);
Optimization::DeadCodeEliminationPass(result);

View file

@ -29,130 +29,6 @@ void AddConstantBufferDescriptor(Info& info, u32 index, u32 count) {
});
}
void GetAttribute(Info& info, IR::Attribute attr) {
if (IR::IsGeneric(attr)) {
info.input_generics.at(IR::GenericAttributeIndex(attr)).used = true;
return;
}
if (attr >= IR::Attribute::FixedFncTexture0S && attr <= IR::Attribute::FixedFncTexture9Q) {
info.loads_fixed_fnc_textures = true;
info.loads_legacy_varyings = true;
return;
}
switch (attr) {
case IR::Attribute::PrimitiveId:
info.loads_primitive_id = true;
break;
case IR::Attribute::PositionX:
case IR::Attribute::PositionY:
case IR::Attribute::PositionZ:
case IR::Attribute::PositionW:
info.loads_position = true;
break;
case IR::Attribute::ColorFrontDiffuseR:
case IR::Attribute::ColorFrontDiffuseG:
case IR::Attribute::ColorFrontDiffuseB:
case IR::Attribute::ColorFrontDiffuseA:
info.loads_color_front_diffuse = true;
info.loads_legacy_varyings = true;
break;
case IR::Attribute::PointSpriteS:
case IR::Attribute::PointSpriteT:
info.loads_point_coord = true;
break;
case IR::Attribute::TessellationEvaluationPointU:
case IR::Attribute::TessellationEvaluationPointV:
info.loads_tess_coord = true;
break;
case IR::Attribute::InstanceId:
info.loads_instance_id = true;
break;
case IR::Attribute::VertexId:
info.loads_vertex_id = true;
break;
case IR::Attribute::FrontFace:
info.loads_front_face = true;
break;
default:
throw NotImplementedException("Get attribute {}", attr);
}
}
void SetAttribute(Info& info, IR::Attribute attr) {
if (IR::IsGeneric(attr)) {
info.stores_generics[IR::GenericAttributeIndex(attr)] = true;
return;
}
if (attr >= IR::Attribute::FixedFncTexture0S && attr <= IR::Attribute::FixedFncTexture9Q) {
info.stores_fixed_fnc_textures = true;
info.stores_legacy_varyings = true;
return;
}
switch (attr) {
case IR::Attribute::Layer:
info.stores_layer = true;
break;
case IR::Attribute::ViewportIndex:
info.stores_viewport_index = true;
break;
case IR::Attribute::PointSize:
info.stores_point_size = true;
break;
case IR::Attribute::PositionX:
case IR::Attribute::PositionY:
case IR::Attribute::PositionZ:
case IR::Attribute::PositionW:
info.stores_position = true;
break;
case IR::Attribute::ColorFrontDiffuseR:
case IR::Attribute::ColorFrontDiffuseG:
case IR::Attribute::ColorFrontDiffuseB:
case IR::Attribute::ColorFrontDiffuseA:
info.stores_color_front_diffuse = true;
info.stores_legacy_varyings = true;
break;
case IR::Attribute::ColorFrontSpecularR:
case IR::Attribute::ColorFrontSpecularG:
case IR::Attribute::ColorFrontSpecularB:
case IR::Attribute::ColorFrontSpecularA:
info.stores_color_front_specular = true;
info.stores_legacy_varyings = true;
break;
case IR::Attribute::ColorBackDiffuseR:
case IR::Attribute::ColorBackDiffuseG:
case IR::Attribute::ColorBackDiffuseB:
case IR::Attribute::ColorBackDiffuseA:
info.stores_color_back_diffuse = true;
info.stores_legacy_varyings = true;
break;
case IR::Attribute::ColorBackSpecularR:
case IR::Attribute::ColorBackSpecularG:
case IR::Attribute::ColorBackSpecularB:
case IR::Attribute::ColorBackSpecularA:
info.stores_color_back_specular = true;
info.stores_legacy_varyings = true;
break;
case IR::Attribute::ClipDistance0:
case IR::Attribute::ClipDistance1:
case IR::Attribute::ClipDistance2:
case IR::Attribute::ClipDistance3:
case IR::Attribute::ClipDistance4:
case IR::Attribute::ClipDistance5:
case IR::Attribute::ClipDistance6:
case IR::Attribute::ClipDistance7:
info.stores_clip_distance = true;
break;
case IR::Attribute::FogCoordinate:
info.stores_fog_coordinate = true;
break;
case IR::Attribute::ViewportMask:
info.stores_viewport_mask = true;
break;
default:
throw NotImplementedException("Set attribute {}", attr);
}
}
void GetPatch(Info& info, IR::Patch patch) {
if (!IR::IsGeneric(patch)) {
throw NotImplementedException("Reading non-generic patch {}", patch);
@ -511,10 +387,10 @@ void VisitUsages(Info& info, IR::Inst& inst) {
info.uses_demote_to_helper_invocation = true;
break;
case IR::Opcode::GetAttribute:
GetAttribute(info, inst.Arg(0).Attribute());
info.loads.mask[static_cast<size_t>(inst.Arg(0).Attribute())] = true;
break;
case IR::Opcode::SetAttribute:
SetAttribute(info, inst.Arg(0).Attribute());
info.stores.mask[static_cast<size_t>(inst.Arg(0).Attribute())] = true;
break;
case IR::Opcode::GetPatch:
GetPatch(info, inst.Arg(0).Patch());
@ -943,26 +819,78 @@ void GatherInfoFromHeader(Environment& env, Info& info) {
if (!info.loads_indexed_attributes) {
return;
}
for (size_t i = 0; i < info.input_generics.size(); i++) {
info.input_generics[i].used |= header.ps.IsGenericVectorActive(i);
for (size_t index = 0; index < IR::NUM_GENERICS; ++index) {
const size_t offset{static_cast<size_t>(IR::Attribute::Generic0X) + index * 4};
const auto vector{header.ps.imap_generic_vector[index]};
info.loads.mask[offset + 0] = vector.x != PixelImap::Unused;
info.loads.mask[offset + 1] = vector.y != PixelImap::Unused;
info.loads.mask[offset + 2] = vector.z != PixelImap::Unused;
info.loads.mask[offset + 3] = vector.w != PixelImap::Unused;
}
info.loads_position |= header.ps.imap_systemb.position != 0;
return;
}
if (info.loads_indexed_attributes) {
for (size_t i = 0; i < info.input_generics.size(); i++) {
info.input_generics[i].used |= header.vtg.IsInputGenericVectorActive(i);
}
info.loads_position |= header.vtg.imap_systemb.position != 0;
}
if (info.stores_indexed_attributes) {
for (size_t i = 0; i < info.stores_generics.size(); i++) {
if (header.vtg.IsOutputGenericVectorActive(i)) {
info.stores_generics[i] = true;
for (size_t index = 0; index < IR::NUM_GENERICS; ++index) {
const IR::Attribute attribute{IR::Attribute::Generic0X + index * 4};
const auto mask = header.vtg.InputGeneric(index);
for (size_t i = 0; i < 4; ++i) {
info.loads.Set(attribute + i, mask[i]);
}
}
info.stores_clip_distance |= header.vtg.omap_systemc.clip_distances != 0;
info.stores_position |= header.vtg.omap_systemb.position != 0;
for (size_t index = 0; index < 8; ++index) {
const u16 mask{header.vtg.clip_distances};
info.loads.Set(IR::Attribute::ClipDistance0 + index, ((mask >> index) & 1) != 0);
}
info.loads.Set(IR::Attribute::PrimitiveId, header.vtg.imap_systemb.primitive_array_id != 0);
info.loads.Set(IR::Attribute::Layer, header.vtg.imap_systemb.rt_array_index != 0);
info.loads.Set(IR::Attribute::ViewportIndex, header.vtg.imap_systemb.viewport_index != 0);
info.loads.Set(IR::Attribute::PointSize, header.vtg.imap_systemb.point_size != 0);
info.loads.Set(IR::Attribute::PositionX, header.vtg.imap_systemb.position_x != 0);
info.loads.Set(IR::Attribute::PositionY, header.vtg.imap_systemb.position_y != 0);
info.loads.Set(IR::Attribute::PositionZ, header.vtg.imap_systemb.position_z != 0);
info.loads.Set(IR::Attribute::PositionW, header.vtg.imap_systemb.position_w != 0);
info.loads.Set(IR::Attribute::PointSpriteS, header.vtg.point_sprite_s != 0);
info.loads.Set(IR::Attribute::PointSpriteT, header.vtg.point_sprite_t != 0);
info.loads.Set(IR::Attribute::FogCoordinate, header.vtg.fog_coordinate != 0);
info.loads.Set(IR::Attribute::TessellationEvaluationPointU,
header.vtg.tessellation_eval_point_u != 0);
info.loads.Set(IR::Attribute::TessellationEvaluationPointV,
header.vtg.tessellation_eval_point_v != 0);
info.loads.Set(IR::Attribute::InstanceId, header.vtg.instance_id != 0);
info.loads.Set(IR::Attribute::VertexId, header.vtg.vertex_id != 0);
// TODO: Legacy varyings
}
if (info.stores_indexed_attributes) {
for (size_t index = 0; index < IR::NUM_GENERICS; ++index) {
const IR::Attribute attribute{IR::Attribute::Generic0X + index * 4};
const auto mask{header.vtg.OutputGeneric(index)};
for (size_t i = 0; i < 4; ++i) {
info.stores.Set(attribute + i, mask[i]);
}
}
for (size_t index = 0; index < 8; ++index) {
const u16 mask{header.vtg.omap_systemc.clip_distances};
info.stores.Set(IR::Attribute::ClipDistance0 + index, ((mask >> index) & 1) != 0);
}
info.stores.Set(IR::Attribute::PrimitiveId,
header.vtg.omap_systemb.primitive_array_id != 0);
info.stores.Set(IR::Attribute::Layer, header.vtg.omap_systemb.rt_array_index != 0);
info.stores.Set(IR::Attribute::ViewportIndex, header.vtg.omap_systemb.viewport_index != 0);
info.stores.Set(IR::Attribute::PointSize, header.vtg.omap_systemb.point_size != 0);
info.stores.Set(IR::Attribute::PositionX, header.vtg.omap_systemb.position_x != 0);
info.stores.Set(IR::Attribute::PositionY, header.vtg.omap_systemb.position_y != 0);
info.stores.Set(IR::Attribute::PositionZ, header.vtg.omap_systemb.position_z != 0);
info.stores.Set(IR::Attribute::PositionW, header.vtg.omap_systemb.position_w != 0);
info.stores.Set(IR::Attribute::PointSpriteS, header.vtg.omap_systemc.point_sprite_s != 0);
info.stores.Set(IR::Attribute::PointSpriteT, header.vtg.omap_systemc.point_sprite_t != 0);
info.stores.Set(IR::Attribute::FogCoordinate, header.vtg.omap_systemc.fog_coordinate != 0);
info.stores.Set(IR::Attribute::TessellationEvaluationPointU,
header.vtg.omap_systemc.tessellation_eval_point_u != 0);
info.stores.Set(IR::Attribute::TessellationEvaluationPointV,
header.vtg.omap_systemc.tessellation_eval_point_v != 0);
info.stores.Set(IR::Attribute::InstanceId, header.vtg.omap_systemc.instance_id != 0);
info.stores.Set(IR::Attribute::VertexId, header.vtg.omap_systemc.vertex_id != 0);
// TODO: Legacy varyings
}
}
} // Anonymous namespace

View file

@ -34,6 +34,7 @@ struct Profile {
bool support_demote_to_helper_invocation{};
bool support_int64_atomics{};
bool support_derivative_control{};
bool support_geometry_shader_passthrough{};
bool support_gl_nv_gpu_shader_5{};
bool support_gl_amd_gpu_shader_half_float{};
bool support_gl_texture_shadow_lod{};

View file

@ -37,7 +37,9 @@ struct ProgramHeader {
BitField<15, 1, u32> kills_pixels;
BitField<16, 1, u32> does_global_store;
BitField<17, 4, u32> sass_version;
BitField<21, 5, u32> reserved;
BitField<21, 2, u32> reserved1;
BitField<24, 1, u32> geometry_passthrough;
BitField<25, 1, u32> reserved2;
BitField<26, 1, u32> does_load_or_store;
BitField<27, 1, u32> does_fp64;
BitField<28, 4, u32> stream_out_mask;
@ -79,24 +81,10 @@ struct ProgramHeader {
BitField<5, 1, u8> position_y;
BitField<6, 1, u8> position_z;
BitField<7, 1, u8> position_w;
BitField<0, 4, u8> first;
BitField<4, 4, u8> position;
u8 raw;
} imap_systemb;
union {
BitField<0, 1, u8> x;
BitField<1, 1, u8> y;
BitField<2, 1, u8> z;
BitField<3, 1, u8> w;
BitField<4, 1, u8> x2;
BitField<5, 1, u8> y2;
BitField<6, 1, u8> z2;
BitField<7, 1, u8> w2;
BitField<0, 4, u8> first;
BitField<4, 4, u8> second;
u8 raw;
} imap_generic_vector[16];
std::array<u8, 16> imap_generic_vector;
INSERT_PADDING_BYTES_NOINIT(2); // ImapColor
union {
@ -122,24 +110,10 @@ struct ProgramHeader {
BitField<5, 1, u8> position_y;
BitField<6, 1, u8> position_z;
BitField<7, 1, u8> position_w;
BitField<0, 4, u8> first;
BitField<4, 4, u8> position;
u8 raw;
} omap_systemb;
union {
BitField<0, 1, u8> x;
BitField<1, 1, u8> y;
BitField<2, 1, u8> z;
BitField<3, 1, u8> w;
BitField<4, 1, u8> x2;
BitField<5, 1, u8> y2;
BitField<6, 1, u8> z2;
BitField<7, 1, u8> w2;
BitField<0, 4, u8> first;
BitField<4, 4, u8> second;
u8 raw;
} omap_generic_vector[16];
std::array<u8, 16> omap_generic_vector;
INSERT_PADDING_BYTES_NOINIT(2); // OmapColor
@ -157,18 +131,24 @@ struct ProgramHeader {
INSERT_PADDING_BYTES_NOINIT(5); // OmapFixedFncTexture[10]
INSERT_PADDING_BYTES_NOINIT(1); // OmapReserved
[[nodiscard]] bool IsInputGenericVectorActive(size_t index) const {
if ((index & 1) == 0) {
return imap_generic_vector[index >> 1].first != 0;
}
return imap_generic_vector[index >> 1].second != 0;
[[nodiscard]] std::array<bool, 4> InputGeneric(size_t index) const noexcept {
const int data{imap_generic_vector[index >> 1] >> ((index % 2) * 4)};
return {
(data & 1) != 0,
(data & 2) != 0,
(data & 4) != 0,
(data & 8) != 0,
};
}
[[nodiscard]] bool IsOutputGenericVectorActive(size_t index) const {
if ((index & 1) == 0) {
return omap_generic_vector[index >> 1].first != 0;
}
return omap_generic_vector[index >> 1].second != 0;
[[nodiscard]] std::array<bool, 4> OutputGeneric(size_t index) const noexcept {
const int data{omap_generic_vector[index >> 1] >> ((index % 2) * 4)};
return {
(data & 1) != 0,
(data & 2) != 0,
(data & 4) != 0,
(data & 8) != 0,
};
}
} vtg;

View file

@ -10,6 +10,7 @@
#include <vector>
#include "common/common_types.h"
#include "shader_recompiler/varying_state.h"
namespace Shader {
@ -60,7 +61,7 @@ struct TransformFeedbackVarying {
struct RuntimeInfo {
std::array<AttributeType, 32> generic_input_types{};
std::bitset<32> previous_stage_stores_generic{};
VaryingState previous_stage_stores;
bool convert_depth_mode{};
bool force_early_z{};

View file

@ -9,6 +9,7 @@
#include "common/common_types.h"
#include "shader_recompiler/frontend/ir/type.h"
#include "shader_recompiler/varying_state.h"
#include <boost/container/small_vector.hpp>
#include <boost/container/static_vector.hpp>
@ -44,11 +45,6 @@ enum class Interpolation {
NoPerspective,
};
struct InputVarying {
Interpolation interpolation{Interpolation::Smooth};
bool used{false};
};
struct ConstantBufferDescriptor {
u32 index;
u32 count;
@ -121,18 +117,10 @@ struct Info {
bool uses_subgroup_shuffles{};
std::array<bool, 30> uses_patches{};
std::array<InputVarying, 32> input_generics{};
bool loads_primitive_id{};
bool loads_position{};
bool loads_color_front_diffuse{};
bool loads_fixed_fnc_textures{};
bool loads_point_coord{};
bool loads_instance_id{};
bool loads_vertex_id{};
bool loads_front_face{};
bool loads_legacy_varyings{};
bool loads_tess_coord{};
std::array<Interpolation, 32> interpolation{};
VaryingState loads;
VaryingState stores;
VaryingState passthrough;
bool loads_indexed_attributes{};
@ -140,21 +128,6 @@ struct Info {
bool stores_sample_mask{};
bool stores_frag_depth{};
std::bitset<32> stores_generics{};
bool stores_layer{};
bool stores_viewport_index{};
bool stores_point_size{};
bool stores_position{};
bool stores_color_front_diffuse{};
bool stores_color_front_specular{};
bool stores_color_back_diffuse{};
bool stores_color_back_specular{};
bool stores_fixed_fnc_textures{};
bool stores_clip_distance{};
bool stores_fog_coordinate{};
bool stores_viewport_mask{};
bool stores_legacy_varyings{};
bool stores_tess_level_outer{};
bool stores_tess_level_inner{};

View file

@ -0,0 +1,69 @@
// Copyright 2021 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <bitset>
#include <cstddef>
#include "shader_recompiler/frontend/ir/attribute.h"
namespace Shader {
struct VaryingState {
std::bitset<256> mask{};
void Set(IR::Attribute attribute, bool state = true) {
mask[static_cast<size_t>(attribute)] = state;
}
[[nodiscard]] bool operator[](IR::Attribute attribute) const noexcept {
return mask[static_cast<size_t>(attribute)];
}
[[nodiscard]] bool AnyComponent(IR::Attribute base) const noexcept {
return mask[static_cast<size_t>(base) + 0] || mask[static_cast<size_t>(base) + 1] ||
mask[static_cast<size_t>(base) + 2] || mask[static_cast<size_t>(base) + 3];
}
[[nodiscard]] bool AllComponents(IR::Attribute base) const noexcept {
return mask[static_cast<size_t>(base) + 0] && mask[static_cast<size_t>(base) + 1] &&
mask[static_cast<size_t>(base) + 2] && mask[static_cast<size_t>(base) + 3];
}
[[nodiscard]] bool IsUniform(IR::Attribute base) const noexcept {
return AnyComponent(base) == AllComponents(base);
}
[[nodiscard]] bool Generic(size_t index, size_t component) const noexcept {
return mask[static_cast<size_t>(IR::Attribute::Generic0X) + index * 4 + component];
}
[[nodiscard]] bool Generic(size_t index) const noexcept {
return Generic(index, 0) || Generic(index, 1) || Generic(index, 2) || Generic(index, 3);
}
[[nodiscard]] bool ClipDistances() const noexcept {
return AnyComponent(IR::Attribute::ClipDistance0) ||
AnyComponent(IR::Attribute::ClipDistance4);
}
[[nodiscard]] bool Legacy() const noexcept {
return AnyComponent(IR::Attribute::ColorFrontDiffuseR) ||
AnyComponent(IR::Attribute::ColorFrontSpecularR) ||
AnyComponent(IR::Attribute::ColorBackDiffuseR) ||
AnyComponent(IR::Attribute::ColorBackSpecularR) || FixedFunctionTexture();
}
[[nodiscard]] bool FixedFunctionTexture() const noexcept {
for (size_t index = 0; index < 10; ++index) {
if (AnyComponent(IR::Attribute::FixedFncTexture0S + index * 4)) {
return true;
}
}
return false;
}
};
} // namespace Shader

View file

@ -961,7 +961,11 @@ public:
SamplerIndex sampler_index;
INSERT_PADDING_WORDS_NOINIT(0x25);
INSERT_PADDING_WORDS_NOINIT(0x2);
std::array<u32, 8> gp_passthrough_mask;
INSERT_PADDING_WORDS_NOINIT(0x1B);
u32 depth_test_enable;
@ -1628,6 +1632,7 @@ ASSERT_REG_POSITION(zeta_width, 0x48a);
ASSERT_REG_POSITION(zeta_height, 0x48b);
ASSERT_REG_POSITION(zeta_depth, 0x48c);
ASSERT_REG_POSITION(sampler_index, 0x48D);
ASSERT_REG_POSITION(gp_passthrough_mask, 0x490);
ASSERT_REG_POSITION(depth_test_enable, 0x4B3);
ASSERT_REG_POSITION(independent_blend_enable, 0x4B9);
ASSERT_REG_POSITION(depth_write_enabled, 0x4BA);

View file

@ -61,10 +61,10 @@ Shader::RuntimeInfo MakeRuntimeInfo(const GraphicsPipelineKey& key,
bool glasm_use_storage_buffers, bool use_assembly_shaders) {
Shader::RuntimeInfo info;
if (previous_program) {
info.previous_stage_stores_generic = previous_program->info.stores_generics;
info.previous_stage_stores = previous_program->info.stores;
} else {
// Mark all stores as available
info.previous_stage_stores_generic.flip();
// Mark all stores as available for vertex shaders
info.previous_stage_stores.mask.set();
}
switch (program.stage) {
case Shader::Stage::VertexB:
@ -187,6 +187,7 @@ ShaderCache::ShaderCache(RasterizerOpenGL& rasterizer_, Core::Frontend::EmuWindo
.support_demote_to_helper_invocation = false,
.support_int64_atomics = false,
.support_derivative_control = device.HasDerivativeControl(),
.support_geometry_shader_passthrough = false, // TODO
.support_gl_nv_gpu_shader_5 = device.HasNvGpuShader5(),
.support_gl_amd_gpu_shader_half_float = device.HasAmdShaderHalfFloat(),
.support_gl_texture_shadow_lod = device.HasTextureShadowLod(),

View file

@ -487,10 +487,9 @@ void GraphicsPipeline::MakePipeline(VkRenderPass render_pass) {
static_vector<VkVertexInputBindingDivisorDescriptionEXT, 32> vertex_binding_divisors;
static_vector<VkVertexInputAttributeDescription, 32> vertex_attributes;
if (key.state.dynamic_vertex_input) {
const auto& input_attributes = stage_infos[0].input_generics;
for (size_t index = 0; index < key.state.attributes.size(); ++index) {
const u32 type = key.state.DynamicAttributeType(index);
if (!input_attributes[index].used || type == 0) {
if (!stage_infos[0].loads.Generic(index) || type == 0) {
continue;
}
vertex_attributes.push_back({
@ -526,10 +525,9 @@ void GraphicsPipeline::MakePipeline(VkRenderPass render_pass) {
});
}
}
const auto& input_attributes = stage_infos[0].input_generics;
for (size_t index = 0; index < key.state.attributes.size(); ++index) {
const auto& attribute = key.state.attributes[index];
if (!attribute.enabled || !input_attributes[index].used) {
if (!attribute.enabled || !stage_infos[0].loads.Generic(index)) {
continue;
}
vertex_attributes.push_back({

View file

@ -123,18 +123,21 @@ Shader::AttributeType AttributeType(const FixedPipelineState& state, size_t inde
return Shader::AttributeType::Disabled;
}
Shader::RuntimeInfo MakeRuntimeInfo(const GraphicsPipelineCacheKey& key,
Shader::RuntimeInfo MakeRuntimeInfo(std::span<const Shader::IR::Program> programs,
const GraphicsPipelineCacheKey& key,
const Shader::IR::Program& program,
const Shader::IR::Program* previous_program) {
Shader::RuntimeInfo info;
if (previous_program) {
info.previous_stage_stores_generic = previous_program->info.stores_generics;
info.previous_stage_stores = previous_program->info.stores;
if (previous_program->is_geometry_passthrough) {
info.previous_stage_stores.mask |= previous_program->info.passthrough.mask;
}
} else {
// Mark all stores as available
info.previous_stage_stores_generic.flip();
info.previous_stage_stores.mask.set();
}
const Shader::Stage stage{program.stage};
const bool has_geometry{key.unique_hashes[4] != 0};
const bool has_geometry{key.unique_hashes[4] != 0 && !programs[4].is_geometry_passthrough};
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) {
@ -302,6 +305,7 @@ PipelineCache::PipelineCache(RasterizerVulkan& rasterizer_, Tegra::Engines::Maxw
.support_demote_to_helper_invocation = true,
.support_int64_atomics = device.IsExtShaderAtomicInt64Supported(),
.support_derivative_control = true,
.support_geometry_shader_passthrough = device.IsNvGeometryShaderPassthroughSupported(),
.warp_size_potentially_larger_than_guest = device.IsWarpSizePotentiallyBiggerThanGuest(),
@ -518,7 +522,7 @@ std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline(
const size_t stage_index{index - 1};
infos[stage_index] = &program.info;
const Shader::RuntimeInfo runtime_info{MakeRuntimeInfo(key, program, previous_stage)};
const auto runtime_info{MakeRuntimeInfo(programs, key, program, previous_stage)};
const std::vector<u32> code{EmitSPIRV(profile, runtime_info, program, binding)};
device.SaveShader(code);
modules[stage_index] = BuildShader(device, code);

View file

@ -22,7 +22,7 @@
namespace VideoCommon {
constexpr std::array<char, 8> MAGIC_NUMBER{'y', 'u', 'z', 'u', 'c', 'a', 'c', 'h'};
constexpr u32 CACHE_VERSION = 4;
constexpr u32 CACHE_VERSION = 5;
constexpr size_t INST_SIZE = sizeof(u64);
@ -155,6 +155,10 @@ void GenericEnvironment::Serialize(std::ofstream& file) const {
.write(reinterpret_cast<const char*>(&shared_memory_size), sizeof(shared_memory_size));
} else {
file.write(reinterpret_cast<const char*>(&sph), sizeof(sph));
if (stage == Shader::Stage::Geometry) {
file.write(reinterpret_cast<const char*>(&gp_passthrough_mask),
sizeof(gp_passthrough_mask));
}
}
}
@ -202,6 +206,7 @@ GraphicsEnvironment::GraphicsEnvironment(Tegra::Engines::Maxwell3D& maxwell3d_,
u32 start_address_)
: GenericEnvironment{gpu_memory_, program_base_, start_address_}, maxwell3d{&maxwell3d_} {
gpu_memory->ReadBlock(program_base + start_address, &sph, sizeof(sph));
gp_passthrough_mask = maxwell3d->regs.gp_passthrough_mask;
switch (program) {
case Maxwell::ShaderProgram::VertexA:
stage = Shader::Stage::VertexA;
@ -319,6 +324,9 @@ void FileEnvironment::Deserialize(std::ifstream& file) {
.read(reinterpret_cast<char*>(&shared_memory_size), sizeof(shared_memory_size));
} else {
file.read(reinterpret_cast<char*>(&sph), sizeof(sph));
if (stage == Shader::Stage::Geometry) {
file.read(reinterpret_cast<char*>(&gp_passthrough_mask), sizeof(gp_passthrough_mask));
}
}
}

View file

@ -350,6 +350,10 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
LOG_INFO(Render_Vulkan, "Device doesn't support viewport masks");
}
if (!nv_geometry_shader_passthrough) {
LOG_INFO(Render_Vulkan, "Device doesn't support passthrough geometry shaders");
}
VkPhysicalDeviceUniformBufferStandardLayoutFeaturesKHR std430_layout;
if (khr_uniform_buffer_standard_layout) {
std430_layout = {
@ -768,6 +772,8 @@ std::vector<const char*> Device::LoadExtensions(bool requires_surface) {
};
test(nv_viewport_swizzle, VK_NV_VIEWPORT_SWIZZLE_EXTENSION_NAME, true);
test(nv_viewport_array2, VK_NV_VIEWPORT_ARRAY2_EXTENSION_NAME, true);
test(nv_geometry_shader_passthrough, VK_NV_GEOMETRY_SHADER_PASSTHROUGH_EXTENSION_NAME,
true);
test(khr_uniform_buffer_standard_layout,
VK_KHR_UNIFORM_BUFFER_STANDARD_LAYOUT_EXTENSION_NAME, true);
test(khr_spirv_1_4, VK_KHR_SPIRV_1_4_EXTENSION_NAME, true);

View file

@ -194,6 +194,11 @@ public:
return nv_viewport_array2;
}
/// Returns true if the device supports VK_NV_geometry_shader_passthrough.
bool IsNvGeometryShaderPassthroughSupported() const {
return nv_geometry_shader_passthrough;
}
/// Returns true if the device supports VK_KHR_uniform_buffer_standard_layout.
bool IsKhrUniformBufferStandardLayoutSupported() const {
return khr_uniform_buffer_standard_layout;
@ -363,6 +368,7 @@ private:
bool is_blit_depth_stencil_supported{}; ///< Support for blitting from and to depth stencil.
bool nv_viewport_swizzle{}; ///< Support for VK_NV_viewport_swizzle.
bool nv_viewport_array2{}; ///< Support for VK_NV_viewport_array2.
bool nv_geometry_shader_passthrough{}; ///< Support for VK_NV_geometry_shader_passthrough.
bool khr_uniform_buffer_standard_layout{}; ///< Support for scalar uniform buffer layouts.
bool khr_spirv_1_4{}; ///< Support for VK_KHR_spirv_1_4.
bool khr_workgroup_memory_explicit_layout{}; ///< Support for explicit workgroup layouts.