H3cJP/yuzu
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

gl_shader_decompiler: Implement gl_ViewportIndex and gl_Layer in vertex shaders

Browse Source 
This commit implements gl_ViewportIndex and gl_Layer in vertex and
geometry shaders. In the case it's used in a vertex shader, it requires
ARB_shader_viewport_layer_array. This extension is available on AMD and
Nvidia devices (mesa and proprietary drivers), but not available on
Intel on any platform. At the moment of writing this description I don't
know if this is a hardware limitation or a driver limitation.

In the case that ARB_shader_viewport_layer_array is not available,
writes to these registers on a vertex shader are ignored, with the
appropriate logging.
This commit is contained in:
ReinUsesLisp 2019-07-07 20:36:42 -03:00
parent 8070cb3f6b
commit c9d886c84e
10 changed files with 136 additions and 40 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
src/video_core/engines/shader_bytecode.h
View File

@ -78,7 +78,7 @@ union Attribute {
constexpr explicit Attribute(u64 value) : value(value) {}
enum class Index : u64 {
PointSize = 6,
LayerViewportPointSize = 6,
Position = 7,
Attribute_0 = 8,
Attribute_31 = 39,

2
src/video_core/renderer_opengl/gl_device.cpp
View File

@ -26,6 +26,7 @@ Device::Device() {
uniform_buffer_alignment = GetInteger<std::size_t>(GL_UNIFORM_BUFFER_OFFSET_ALIGNMENT);
max_vertex_attributes = GetInteger<u32>(GL_MAX_VERTEX_ATTRIBS);
max_varyings = GetInteger<u32>(GL_MAX_VARYING_VECTORS);
has_vertex_viewport_layer = GLAD_GL_ARB_shader_viewport_layer_array;
has_variable_aoffi = TestVariableAoffi();
has_component_indexing_bug = TestComponentIndexingBug();
}
@ -34,6 +35,7 @@ Device::Device(std::nullptr_t) {
uniform_buffer_alignment = 0;
max_vertex_attributes = 16;
max_varyings = 15;
has_vertex_viewport_layer = true;
has_variable_aoffi = true;
has_component_indexing_bug = false;
}

5
src/video_core/renderer_opengl/gl_device.h
View File

@ -26,6 +26,10 @@ public:
return max_varyings;
}
bool HasVertexViewportLayer() const {
return has_vertex_viewport_layer;
}
bool HasVariableAoffi() const {
return has_variable_aoffi;
}
@ -41,6 +45,7 @@ private:
std::size_t uniform_buffer_alignment{};
u32 max_vertex_attributes{};
u32 max_varyings{};
bool has_vertex_viewport_layer{};
bool has_variable_aoffi{};
bool has_component_indexing_bug{};
};

7
src/video_core/renderer_opengl/gl_shader_cache.cpp
View File

@ -182,8 +182,11 @@ CachedProgram SpecializeShader(const std::string& code, const GLShader::ShaderEn
const auto texture_buffer_usage{variant.texture_buffer_usage};
std::string source = "#version 430 core\n"
"#extension GL_ARB_separate_shader_objects : enable\n\n";
source += fmt::format("#define EMULATION_UBO_BINDING {}\n", base_bindings.cbuf++);
"#extension GL_ARB_separate_shader_objects : enable\n";
if (entries.shader_viewport_layer_array) {
source += "#extension GL_ARB_shader_viewport_layer_array : enable\n";
}
source += fmt::format("\n#define EMULATION_UBO_BINDING {}\n", base_bindings.cbuf++);
for (const auto& cbuf : entries.const_buffers) {
source +=

104
src/video_core/renderer_opengl/gl_shader_decompiler.cpp
View File

@ -14,6 +14,7 @@
#include "common/alignment.h"
#include "common/assert.h"
#include "common/common_types.h"
#include "common/logging/log.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/renderer_opengl/gl_device.h"
#include "video_core/renderer_opengl/gl_rasterizer.h"
@ -244,6 +245,8 @@ public:
usage.is_read, usage.is_written);
}
entries.clip_distances = ir.GetClipDistances();
entries.shader_viewport_layer_array =
stage == ShaderStage::Vertex && (ir.UsesLayer() || ir.UsesPointSize());
entries.shader_length = ir.GetLength();
return entries;
}
@ -280,22 +283,34 @@ private:
}
void DeclareVertexRedeclarations() {
bool clip_distances_declared = false;
code.AddLine("out gl_PerVertex {{");
++code.scope;
code.AddLine("vec4 gl_Position;");
for (const auto o : ir.GetOutputAttributes()) {
if (o == Attribute::Index::PointSize)
code.AddLine("float gl_PointSize;");
if (!clip_distances_declared && (o == Attribute::Index::ClipDistances0123 ||
o == Attribute::Index::ClipDistances4567)) {
for (const auto attribute : ir.GetOutputAttributes()) {
if (attribute == Attribute::Index::ClipDistances0123 ||
attribute == Attribute::Index::ClipDistances4567) {
code.AddLine("float gl_ClipDistance[];");
clip_distances_declared = true;
break;
}
}
if (stage != ShaderStage::Vertex || device.HasVertexViewportLayer()) {
if (ir.UsesLayer()) {
code.AddLine("int gl_Layer;");
}
if (ir.UsesViewportIndex()) {
code.AddLine("int gl_ViewportIndex;");
}
} else if (stage == ShaderStage::Vertex && !device.HasVertexViewportLayer()) {
LOG_ERROR(
Render_OpenGL,
"GL_ARB_shader_viewport_layer_array is not available and its required by a shader");
}
if (ir.UsesPointSize()) {
code.AddLine("int gl_PointSize;");
}
--code.scope;
code.AddLine("}};");
@ -803,6 +818,45 @@ private:
return CastOperand(VisitOperand(operation, operand_index), type);
}
std::optional<std::pair<std::string, bool>> GetOutputAttribute(const AbufNode* abuf) {
switch (const auto attribute = abuf->GetIndex()) {
case Attribute::Index::Position:
return std::make_pair("gl_Position"s + GetSwizzle(abuf->GetElement()), false);
case Attribute::Index::LayerViewportPointSize:
switch (abuf->GetElement()) {
case 0:
UNIMPLEMENTED();
return {};
case 1:
if (stage == ShaderStage::Vertex && !device.HasVertexViewportLayer()) {
return {};
}
return std::make_pair("gl_Layer", true);
case 2:
if (stage == ShaderStage::Vertex && !device.HasVertexViewportLayer()) {
return {};
}
return std::make_pair("gl_ViewportIndex", true);
case 3:
UNIMPLEMENTED_MSG("Requires some state changes for gl_PointSize to work in shader");
return std::make_pair("gl_PointSize", false);
}
return {};
case Attribute::Index::ClipDistances0123:
return std::make_pair(fmt::format("gl_ClipDistance[{}]", abuf->GetElement()), false);
case Attribute::Index::ClipDistances4567:
return std::make_pair(fmt::format("gl_ClipDistance[{}]", abuf->GetElement() + 4),
false);
default:
if (IsGenericAttribute(attribute)) {
return std::make_pair(
GetOutputAttribute(attribute) + GetSwizzle(abuf->GetElement()), false);
}
UNIMPLEMENTED_MSG("Unhandled output attribute: {}", static_cast<u32>(attribute));
return {};
}
}
std::string CastOperand(const std::string& value, Type type) const {
switch (type) {
case Type::Bool:
@ -999,6 +1053,8 @@ private:
const Node& src = operation[1];
std::string target;
bool is_integer = false;
if (const auto gpr = std::get_if<GprNode>(&*dest)) {
if (gpr->GetIndex() == Register::ZeroIndex) {
// Writing to Register::ZeroIndex is a no op
@ -1007,26 +1063,12 @@ private:
target = GetRegister(gpr->GetIndex());
} else if (const auto abuf = std::get_if<AbufNode>(&*dest)) {
UNIMPLEMENTED_IF(abuf->IsPhysicalBuffer());
target = [&]() -> std::string {
switch (const auto attribute = abuf->GetIndex(); abuf->GetIndex()) {
case Attribute::Index::Position:
return "gl_Position"s + GetSwizzle(abuf->GetElement());
case Attribute::Index::PointSize:
return "gl_PointSize";
case Attribute::Index::ClipDistances0123:
return fmt::format("gl_ClipDistance[{}]", abuf->GetElement());
case Attribute::Index::ClipDistances4567:
return fmt::format("gl_ClipDistance[{}]", abuf->GetElement() + 4);
default:
if (IsGenericAttribute(attribute)) {
return GetOutputAttribute(attribute) + GetSwizzle(abuf->GetElement());
}
UNIMPLEMENTED_MSG("Unhandled output attribute: {}",
static_cast<u32>(attribute));
return "0";
}
}();
const auto result = GetOutputAttribute(abuf);
if (!result) {
return {};
}
target = result->first;
is_integer = result->second;
} else if (const auto lmem = std::get_if<LmemNode>(&*dest)) {
target = fmt::format("{}[ftou({}) / 4]", GetLocalMemory(), Visit(lmem->GetAddress()));
} else if (const auto gmem = std::get_if<GmemNode>(&*dest)) {
@ -1038,7 +1080,11 @@ private:
UNREACHABLE_MSG("Assign called without a proper target");
}
code.AddLine("{} = {};", target, Visit(src));
if (is_integer) {
code.AddLine("{} = ftoi({});", target, Visit(src));
} else {
code.AddLine("{} = {};", target, Visit(src));
}
return {};
}

1
src/video_core/renderer_opengl/gl_shader_decompiler.h
View File

@ -78,6 +78,7 @@ struct ShaderEntries {
std::vector<ImageEntry> images;
std::vector<GlobalMemoryEntry> global_memory_entries;
std::array<bool, Maxwell::NumClipDistances> clip_distances{};
bool shader_viewport_layer_array{};
std::size_t shader_length{};
};

10
src/video_core/renderer_opengl/gl_shader_disk_cache.cpp
View File

@ -373,6 +373,12 @@ std::optional<ShaderDiskCacheDecompiled> ShaderDiskCacheOpenGL::LoadDecompiledEn
}
}
bool shader_viewport_layer_array{};
if (!LoadObjectFromPrecompiled(shader_viewport_layer_array)) {
return {};
}
entry.entries.shader_viewport_layer_array = shader_viewport_layer_array;
u64 shader_length{};
if (!LoadObjectFromPrecompiled(shader_length)) {
return {};
@ -445,6 +451,10 @@ bool ShaderDiskCacheOpenGL::SaveDecompiledFile(u64 unique_identifier, const std:
}
}
if (!SaveObjectToPrecompiled(entries.shader_viewport_layer_array)) {
return false;
}
if (!SaveObjectToPrecompiled(static_cast<u64>(entries.shader_length))) {
return false;
}

14
src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
View File

@ -430,20 +430,17 @@ private:
instance_index = DeclareBuiltIn(spv::BuiltIn::InstanceIndex, spv::StorageClass::Input,
t_in_uint, "instance_index");
bool is_point_size_declared = false;
bool is_clip_distances_declared = false;
for (const auto index : ir.GetOutputAttributes()) {
if (index == Attribute::Index::PointSize) {
is_point_size_declared = true;
} else if (index == Attribute::Index::ClipDistances0123 ||
index == Attribute::Index::ClipDistances4567) {
if (index == Attribute::Index::ClipDistances0123 ||
index == Attribute::Index::ClipDistances4567) {
is_clip_distances_declared = true;
}
}
std::vector<Id> members;
members.push_back(t_float4);
if (is_point_size_declared) {
if (ir.UsesPointSize()) {
members.push_back(t_float);
}
if (is_clip_distances_declared) {
@ -466,7 +463,7 @@ private:
position_index = MemberDecorateBuiltIn(spv::BuiltIn::Position, "position", true);
point_size_index =
MemberDecorateBuiltIn(spv::BuiltIn::PointSize, "point_size", is_point_size_declared);
MemberDecorateBuiltIn(spv::BuiltIn::PointSize, "point_size", ir.UsesPointSize());
clip_distances_index = MemberDecorateBuiltIn(spv::BuiltIn::ClipDistance, "clip_distances",
is_clip_distances_declared);
@ -712,7 +709,8 @@ private:
case Attribute::Index::Position:
return AccessElement(t_out_float, per_vertex, position_index,
abuf->GetElement());
case Attribute::Index::PointSize:
case Attribute::Index::LayerViewportPointSize:
UNIMPLEMENTED_IF(abuf->GetElement() != 3);
return AccessElement(t_out_float, per_vertex, point_size_index);
case Attribute::Index::ClipDistances0123:
return AccessElement(t_out_float, per_vertex, clip_distances_index,

16
src/video_core/shader/shader_ir.cpp
View File

@ -89,6 +89,22 @@ Node ShaderIR::GetPhysicalInputAttribute(Tegra::Shader::Register physical_addres
}
Node ShaderIR::GetOutputAttribute(Attribute::Index index, u64 element, Node buffer) {
if (index == Attribute::Index::LayerViewportPointSize) {
switch (element) {
case 0:
UNIMPLEMENTED();
break;
case 1:
uses_layer = true;
break;
case 2:
uses_viewport_index = true;
break;
case 3:
uses_point_size = true;
break;
}
}
if (index == Attribute::Index::ClipDistances0123 ||
index == Attribute::Index::ClipDistances4567) {
const auto clip_index =

15
src/video_core/shader/shader_ir.h
View File

@ -121,6 +121,18 @@ public:
return static_cast<std::size_t>(coverage_end * sizeof(u64));
}
bool UsesLayer() const {
return uses_layer;
}
bool UsesViewportIndex() const {
return uses_viewport_index;
}
bool UsesPointSize() const {
return uses_point_size;
}
bool HasPhysicalAttributes() const {
return uses_physical_attributes;
}
@ -343,6 +355,9 @@ private:
std::set<Image> used_images;
std::array<bool, Tegra::Engines::Maxwell3D::Regs::NumClipDistances> used_clip_distances{};
std::map<GlobalMemoryBase, GlobalMemoryUsage> used_global_memory;
bool uses_layer{};
bool uses_viewport_index{};
bool uses_point_size{};
bool uses_physical_attributes{}; // Shader uses AL2P or physical attribute read/writes
Tegra::Shader::Header header;