fabi/Xenia-Canary
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge branch 'master' of https://github.com/xenia-project/xenia into canary_experimental

Browse Source
This commit is contained in:
Gliniak
2022-07-30 12:42:51 +02:00
parent 0c3019981c 7595cdb52b
commit 0e3403d6da
15 changed files with 1307 additions and 359 deletions
Download Patch File Download Diff File Expand all files Collapse all files

133
src/xenia/gpu/vulkan/vulkan_command_processor.cc
View File

@@ -2171,7 +2171,9 @@ bool VulkanCommandProcessor::IssueDraw(xenos::PrimitiveType prim_type,
// TODO(Triang3l): Tessellation, geometry-type-specific vertex shader,
// vertex shader as compute.
if (primitive_processing_result.host_vertex_shader_type !=
Shader::HostVertexShaderType::kVertex) {
Shader::HostVertexShaderType::kVertex &&
primitive_processing_result.host_vertex_shader_type !=
Shader::HostVertexShaderType::kPointListAsTriangleStrip) {
return false;
}
@@ -2179,7 +2181,7 @@ bool VulkanCommandProcessor::IssueDraw(xenos::PrimitiveType prim_type,
vertex_shader_modification =
pipeline_cache_->GetCurrentVertexShaderModification(
*vertex_shader, primitive_processing_result.host_vertex_shader_type,
interpolator_mask);
interpolator_mask, ps_param_gen_pos != UINT32_MAX);
pixel_shader_modification =
pixel_shader ? pipeline_cache_->GetCurrentPixelShaderModification(
*pixel_shader, interpolator_mask, ps_param_gen_pos)
@@ -2348,6 +2350,7 @@ bool VulkanCommandProcessor::IssueDraw(xenos::PrimitiveType prim_type,
}
const ui::vulkan::VulkanProvider& provider = GetVulkanProvider();
const VkPhysicalDeviceFeatures& device_features = provider.device_features();
const VkPhysicalDeviceLimits& device_limits =
provider.device_properties().limits;
@@ -2382,11 +2385,23 @@ bool VulkanCommandProcessor::IssueDraw(xenos::PrimitiveType prim_type,
UpdateDynamicState(viewport_info, primitive_polygonal,
normalized_depth_control);
auto vgt_draw_initiator = regs.Get<reg::VGT_DRAW_INITIATOR>();
// Whether to load the guest 32-bit (usually big-endian) vertex index
// indirectly in the vertex shader if full 32-bit indices are not supported by
// the host.
bool shader_32bit_index_dma =
!device_features.fullDrawIndexUint32 &&
primitive_processing_result.index_buffer_type ==
PrimitiveProcessor::ProcessedIndexBufferType::kGuestDMA &&
vgt_draw_initiator.index_size == xenos::IndexFormat::kInt32 &&
primitive_processing_result.host_vertex_shader_type ==
Shader::HostVertexShaderType::kVertex;
// Update system constants before uploading them.
bool vertex_shader_index_load;
UpdateSystemConstantValues(primitive_polygonal, primitive_processing_result,
viewport_info, used_texture_mask,
vertex_shader_index_load);
shader_32bit_index_dma, viewport_info,
used_texture_mask);
// Update uniform buffers and descriptor sets after binding the pipeline with
// the new layout.
@@ -2453,13 +2468,13 @@ bool VulkanCommandProcessor::IssueDraw(xenos::PrimitiveType prim_type,
// Draw.
if (primitive_processing_result.index_buffer_type ==
PrimitiveProcessor::ProcessedIndexBufferType::kNone ||
vertex_shader_index_load) {
shader_32bit_index_dma) {
deferred_command_buffer_.CmdVkDraw(
primitive_processing_result.host_draw_vertex_count, 1, 0, 0);
} else {
std::pair<VkBuffer, VkDeviceSize> index_buffer;
switch (primitive_processing_result.index_buffer_type) {
case PrimitiveProcessor::ProcessedIndexBufferType::kGuest:
case PrimitiveProcessor::ProcessedIndexBufferType::kGuestDMA:
index_buffer.first = shared_memory_->buffer();
index_buffer.second = primitive_processing_result.guest_index_base;
break;
@@ -2467,7 +2482,8 @@ bool VulkanCommandProcessor::IssueDraw(xenos::PrimitiveType prim_type,
index_buffer = primitive_processor_->GetConvertedIndexBuffer(
primitive_processing_result.host_index_buffer_handle);
break;
case PrimitiveProcessor::ProcessedIndexBufferType::kHostBuiltin:
case PrimitiveProcessor::ProcessedIndexBufferType::kHostBuiltinForAuto:
case PrimitiveProcessor::ProcessedIndexBufferType::kHostBuiltinForDMA:
index_buffer = primitive_processor_->GetBuiltinIndexBuffer(
primitive_processing_result.host_index_buffer_handle);
break;
@@ -3342,8 +3358,8 @@ void VulkanCommandProcessor::UpdateDynamicState(
void VulkanCommandProcessor::UpdateSystemConstantValues(
bool primitive_polygonal,
const PrimitiveProcessor::ProcessingResult& primitive_processing_result,
const draw_util::ViewportInfo& viewport_info, uint32_t used_texture_mask,
bool& vertex_shader_index_load_out) {
bool shader_32bit_index_dma, const draw_util::ViewportInfo& viewport_info,
uint32_t used_texture_mask) {
#if XE_UI_VULKAN_FINE_GRAINED_DRAW_SCOPES
SCOPE_profile_cpu_f("gpu");
#endif // XE_UI_VULKAN_FINE_GRAINED_DRAW_SCOPES
@@ -3367,51 +3383,17 @@ void VulkanCommandProcessor::UpdateSystemConstantValues(
// Flags.
uint32_t flags = 0;
// Vertex index shader loading.
bool vertex_shader_index_load = false;
// Only for ProcessedIndexBufferType kGuest since kHostConverted indices may
// be not loaded into the GPU memory (only read on the CPU), though
// kHostConverted must never be used for point lists and rectangle lists
// without geometry shaders anyway. For regular 32-bit index fetching without
// fullDrawIndexUint32, kHostConverted indices are already byte-swapped and
// truncated to 24 bits, so indirect fetch is not needed.
if (shader_32bit_index_dma) {
flags |= SpirvShaderTranslator::kSysFlag_VertexIndexLoad;
}
if (primitive_processing_result.index_buffer_type ==
PrimitiveProcessor::ProcessedIndexBufferType::kGuest) {
switch (primitive_processing_result.host_vertex_shader_type) {
case Shader::HostVertexShaderType::kVertex: {
// For guest (usually big-endian) 32-bit indices when they're not
// supported by the device.
if (vgt_draw_initiator.index_size == xenos::IndexFormat::kInt32) {
const ui::vulkan::VulkanProvider& provider = GetVulkanProvider();
const VkPhysicalDeviceFeatures& device_features =
provider.device_features();
if (!device_features.fullDrawIndexUint32) {
vertex_shader_index_load = true;
flags |= SpirvShaderTranslator::kSysFlag_VertexIndexLoad;
}
}
} break;
// kMemexportCompute never comes out of the PrimitiveProcessor, as
// memexport compute shaders are executed alongside their vertex
// counterparts, since they may still result in drawing.
case Shader::HostVertexShaderType::kPointListAsTriangleStrip:
case Shader::HostVertexShaderType::kRectangleListAsTriangleStrip: {
// Always loading the guest index buffer indirectly if it's used, as
// host indexing contains a part needed specifically for the host for
// the construction of the primitive - host vertices don't map 1:1 to
// guest ones.
vertex_shader_index_load = true;
flags |=
SpirvShaderTranslator::kSysFlag_ComputeOrPrimitiveVertexIndexLoad;
if (vgt_draw_initiator.index_size == xenos::IndexFormat::kInt32) {
flags |= SpirvShaderTranslator ::
kSysFlag_ComputeOrPrimitiveVertexIndexLoad32Bit;
}
} break;
default:
break;
PrimitiveProcessor::ProcessedIndexBufferType::kHostBuiltinForDMA) {
flags |= SpirvShaderTranslator::kSysFlag_ComputeOrPrimitiveVertexIndexLoad;
if (vgt_draw_initiator.index_size == xenos::IndexFormat::kInt32) {
flags |= SpirvShaderTranslator ::
kSysFlag_ComputeOrPrimitiveVertexIndexLoad32Bit;
}
}
vertex_shader_index_load_out = vertex_shader_index_load;
// W0 division control.
// http://www.x.org/docs/AMD/old/evergreen_3D_registers_v2.pdf
// 8: VTX_XY_FMT = true: the incoming XY have already been multiplied by 1/W0.
@@ -3466,9 +3448,9 @@ void VulkanCommandProcessor::UpdateSystemConstantValues(
// Index or tessellation edge factor buffer endianness.
dirty |= system_constants_.vertex_index_endian !=
primitive_processing_result.host_index_endian;
primitive_processing_result.host_shader_index_endian;
system_constants_.vertex_index_endian =
primitive_processing_result.host_index_endian;
primitive_processing_result.host_shader_index_endian;
// Vertex index offset.
dirty |= system_constants_.vertex_base_index != vgt_indx_offset;
@@ -3482,6 +3464,49 @@ void VulkanCommandProcessor::UpdateSystemConstantValues(
system_constants_.ndc_offset[i] = viewport_info.ndc_offset[i];
}
// Point size.
if (vgt_draw_initiator.prim_type == xenos::PrimitiveType::kPointList) {
auto pa_su_point_minmax = regs.Get<reg::PA_SU_POINT_MINMAX>();
auto pa_su_point_size = regs.Get<reg::PA_SU_POINT_SIZE>();
float point_vertex_diameter_min =
float(pa_su_point_minmax.min_size) * (2.0f / 16.0f);
float point_vertex_diameter_max =
float(pa_su_point_minmax.max_size) * (2.0f / 16.0f);
float point_constant_diameter_x =
float(pa_su_point_size.width) * (2.0f / 16.0f);
float point_constant_diameter_y =
float(pa_su_point_size.height) * (2.0f / 16.0f);
dirty |= system_constants_.point_vertex_diameter_min !=
point_vertex_diameter_min;
dirty |= system_constants_.point_vertex_diameter_max !=
point_vertex_diameter_max;
dirty |= system_constants_.point_constant_diameter[0] !=
point_constant_diameter_x;
dirty |= system_constants_.point_constant_diameter[1] !=
point_constant_diameter_y;
system_constants_.point_vertex_diameter_min = point_vertex_diameter_min;
system_constants_.point_vertex_diameter_max = point_vertex_diameter_max;
system_constants_.point_constant_diameter[0] = point_constant_diameter_x;
system_constants_.point_constant_diameter[1] = point_constant_diameter_y;
// 2 because 1 in the NDC is half of the viewport's axis, 0.5 for diameter
// to radius conversion to avoid multiplying the per-vertex diameter by an
// additional constant in the shader.
float point_screen_diameter_to_ndc_radius_x =
(/* 0.5f * 2.0f * */ float(texture_cache_->draw_resolution_scale_x())) /
std::max(viewport_info.xy_extent[0], uint32_t(1));
float point_screen_diameter_to_ndc_radius_y =
(/* 0.5f * 2.0f * */ float(texture_cache_->draw_resolution_scale_y())) /
std::max(viewport_info.xy_extent[1], uint32_t(1));
dirty |= system_constants_.point_screen_diameter_to_ndc_radius[0] !=
point_screen_diameter_to_ndc_radius_x;
dirty |= system_constants_.point_screen_diameter_to_ndc_radius[1] !=
point_screen_diameter_to_ndc_radius_y;
system_constants_.point_screen_diameter_to_ndc_radius[0] =
point_screen_diameter_to_ndc_radius_x;
system_constants_.point_screen_diameter_to_ndc_radius[1] =
point_screen_diameter_to_ndc_radius_y;
}
// Texture signedness / gamma.
{
uint32_t textures_remaining = used_texture_mask;

4
src/xenia/gpu/vulkan/vulkan_command_processor.h
View File

@@ -436,8 +436,8 @@ class VulkanCommandProcessor : public CommandProcessor {
void UpdateSystemConstantValues(
bool primitive_polygonal,
const PrimitiveProcessor::ProcessingResult& primitive_processing_result,
const draw_util::ViewportInfo& viewport_info, uint32_t used_texture_mask,
bool& vertex_shader_index_load_out);
bool shader_32bit_index_dma, const draw_util::ViewportInfo& viewport_info,
uint32_t used_texture_mask);
bool UpdateBindings(const VulkanShader* vertex_shader,
const VulkanShader* pixel_shader);
// Allocates a descriptor set and fills one or two VkWriteDescriptorSet

348
src/xenia/gpu/vulkan/vulkan_pipeline_cache.cc
View File

@@ -118,7 +118,7 @@ VulkanShader* VulkanPipelineCache::LoadShader(xenos::ShaderType shader_type,
SpirvShaderTranslator::Modification
VulkanPipelineCache::GetCurrentVertexShaderModification(
const Shader& shader, Shader::HostVertexShaderType host_vertex_shader_type,
uint32_t interpolator_mask) const {
uint32_t interpolator_mask, bool ps_param_gen_used) const {
assert_true(shader.type() == xenos::ShaderType::kVertex);
assert_true(shader.is_ucode_analyzed());
const auto& regs = register_file_;
@@ -133,6 +133,16 @@ VulkanPipelineCache::GetCurrentVertexShaderModification(
modification.vertex.interpolator_mask = interpolator_mask;
if (host_vertex_shader_type ==
Shader::HostVertexShaderType::kPointListAsTriangleStrip) {
modification.vertex.output_point_parameters = uint32_t(ps_param_gen_used);
} else {
modification.vertex.output_point_parameters =
uint32_t((shader.writes_point_size_edge_flag_kill_vertex() & 0b001) &&
regs.Get<reg::VGT_DRAW_INITIATOR>().prim_type ==
xenos::PrimitiveType::kPointList);
}
return modification;
}
@@ -284,6 +294,8 @@ bool VulkanPipelineCache::ConfigurePipeline(
if (GetGeometryShaderKey(
description.geometry_shader,
SpirvShaderTranslator::Modification(vertex_shader->modification()),
SpirvShaderTranslator::Modification(
pixel_shader ? pixel_shader->modification() : 0),
geometry_shader_key)) {
geometry_shader = GetGeometryShader(geometry_shader_key);
if (geometry_shader == VK_NULL_HANDLE) {
@@ -496,6 +508,7 @@ bool VulkanPipelineCache::GetCurrentStateDescription(
PipelinePrimitiveTopology primitive_topology;
switch (primitive_processing_result.host_primitive_type) {
case xenos::PrimitiveType::kPointList:
geometry_shader = PipelineGeometryShader::kPointList;
primitive_topology = PipelinePrimitiveTopology::kPointList;
break;
case xenos::PrimitiveType::kLineList:
@@ -815,10 +828,22 @@ bool VulkanPipelineCache::ArePipelineRequirementsMet(
bool VulkanPipelineCache::GetGeometryShaderKey(
PipelineGeometryShader geometry_shader_type,
SpirvShaderTranslator::Modification vertex_shader_modification,
SpirvShaderTranslator::Modification pixel_shader_modification,
GeometryShaderKey& key_out) {
if (geometry_shader_type == PipelineGeometryShader::kNone) {
return false;
}
// For kPointListAsTriangleStrip, output_point_parameters has a different
// meaning (the coordinates, not the size). However, the AsTriangleStrip host
// vertex shader types are needed specifically when geometry shaders are not
// supported as fallbacks.
if (vertex_shader_modification.vertex.host_vertex_shader_type ==
Shader::HostVertexShaderType::kPointListAsTriangleStrip ||
vertex_shader_modification.vertex.host_vertex_shader_type ==
Shader::HostVertexShaderType::kRectangleListAsTriangleStrip) {
assert_always();
return false;
}
GeometryShaderKey key;
key.type = geometry_shader_type;
// TODO(Triang3l): Once all needed inputs and outputs are added, uncomment the
@@ -832,9 +857,8 @@ bool VulkanPipelineCache::GetGeometryShaderKey(
key.has_vertex_kill_and =
/* vertex_shader_modification.vertex.vertex_kill_and */ 0;
key.has_point_size =
/* vertex_shader_modification.vertex.output_point_size */ 0;
key.has_point_coordinates =
/* pixel_shader_modification.pixel.param_gen_point */ 0;
vertex_shader_modification.vertex.output_point_parameters;
key.has_point_coordinates = pixel_shader_modification.pixel.param_gen_point;
key_out = key;
return true;
}
@@ -853,6 +877,13 @@ VkShaderModule VulkanPipelineCache::GetGeometryShader(GeometryShaderKey key) {
spv::ExecutionMode output_primitive_execution_mode = spv::ExecutionMode(0);
uint32_t output_max_vertices = 0;
switch (key.type) {
case PipelineGeometryShader::kPointList:
// Point to a strip of 2 triangles.
input_primitive_execution_mode = spv::ExecutionModeInputPoints;
input_primitive_vertex_count = 1;
output_primitive_execution_mode = spv::ExecutionModeOutputTriangleStrip;
output_max_vertices = 4;
break;
case PipelineGeometryShader::kRectangleList:
// Triangle to a strip of 2 triangles.
input_primitive_execution_mode = spv::ExecutionModeTriangles;
@@ -901,6 +932,7 @@ VkShaderModule VulkanPipelineCache::GetGeometryShader(GeometryShaderKey key) {
spv::Id type_bool4 = builder.makeVectorType(type_bool, 4);
spv::Id type_int = builder.makeIntType(32);
spv::Id type_float = builder.makeFloatType(32);
spv::Id type_float2 = builder.makeVectorType(type_float, 2);
spv::Id type_float4 = builder.makeVectorType(type_float, 4);
spv::Id type_clip_distances =
clip_distance_count
@@ -912,9 +944,54 @@ VkShaderModule VulkanPipelineCache::GetGeometryShader(GeometryShaderKey key) {
? builder.makeArrayType(
type_float, builder.makeUintConstant(cull_distance_count), 0)
: spv::NoType;
spv::Id type_point_coordinates = key.has_point_coordinates
? builder.makeVectorType(type_float, 2)
: spv::NoType;
// System constants.
// For points:
// - float2 point_constant_diameter
// - float2 point_screen_diameter_to_ndc_radius
enum PointConstant : uint32_t {
kPointConstantConstantDiameter,
kPointConstantScreenDiameterToNdcRadius,
kPointConstantCount,
};
spv::Id type_system_constants = spv::NoType;
if (key.type == PipelineGeometryShader::kPointList) {
id_vector_temp.clear();
id_vector_temp.resize(kPointConstantCount);
id_vector_temp[kPointConstantConstantDiameter] = type_float2;
id_vector_temp[kPointConstantScreenDiameterToNdcRadius] = type_float2;
type_system_constants =
builder.makeStructType(id_vector_temp, "XeSystemConstants");
builder.addMemberName(type_system_constants, kPointConstantConstantDiameter,
"point_constant_diameter");
builder.addMemberDecoration(
type_system_constants, kPointConstantConstantDiameter,
spv::DecorationOffset,
int(offsetof(SpirvShaderTranslator::SystemConstants,
point_constant_diameter)));
builder.addMemberName(type_system_constants,
kPointConstantScreenDiameterToNdcRadius,
"point_screen_diameter_to_ndc_radius");
builder.addMemberDecoration(
type_system_constants, kPointConstantScreenDiameterToNdcRadius,
spv::DecorationOffset,
int(offsetof(SpirvShaderTranslator::SystemConstants,
point_screen_diameter_to_ndc_radius)));
}
spv::Id uniform_system_constants = spv::NoResult;
if (type_system_constants != spv::NoType) {
builder.addDecoration(type_system_constants, spv::DecorationBlock);
uniform_system_constants = builder.createVariable(
spv::NoPrecision, spv::StorageClassUniform, type_system_constants,
"xe_uniform_system_constants");
builder.addDecoration(uniform_system_constants,
spv::DecorationDescriptorSet,
int(SpirvShaderTranslator::kDescriptorSetConstants));
builder.addDecoration(uniform_system_constants, spv::DecorationBinding,
int(SpirvShaderTranslator::kConstantBufferSystem));
// Generating SPIR-V 1.0, no need to add bindings to the entry point's
// interface until SPIR-V 1.4.
}
// Inputs and outputs - matching glslang order, in gl_PerVertex gl_in[],
// user-defined outputs, user-defined inputs, out gl_PerVertex.
@@ -977,6 +1054,8 @@ VkShaderModule VulkanPipelineCache::GetGeometryShader(GeometryShaderKey key) {
type_array_in_gl_per_vertex, "gl_in");
main_interface.push_back(in_gl_per_vertex);
uint32_t output_location = 0;
// Interpolators outputs.
std::array<spv::Id, xenos::kMaxInterpolators> out_interpolators;
for (uint32_t i = 0; i < key.interpolator_count; ++i) {
@@ -984,23 +1063,28 @@ VkShaderModule VulkanPipelineCache::GetGeometryShader(GeometryShaderKey key) {
spv::NoPrecision, spv::StorageClassOutput, type_float4,
fmt::format("xe_out_interpolator_{}", i).c_str());
out_interpolators[i] = out_interpolator;
builder.addDecoration(out_interpolator, spv::DecorationLocation, i);
builder.addDecoration(out_interpolator, spv::DecorationLocation,
int(output_location));
builder.addDecoration(out_interpolator, spv::DecorationInvariant);
main_interface.push_back(out_interpolator);
++output_location;
}
// Point coordinate output.
spv::Id out_point_coordinates = spv::NoResult;
if (key.has_point_coordinates) {
out_point_coordinates = builder.createVariable(
spv::NoPrecision, spv::StorageClassOutput, type_point_coordinates,
"xe_out_point_coordinates");
out_point_coordinates =
builder.createVariable(spv::NoPrecision, spv::StorageClassOutput,
type_float2, "xe_out_point_coordinates");
builder.addDecoration(out_point_coordinates, spv::DecorationLocation,
key.interpolator_count);
int(output_location));
builder.addDecoration(out_point_coordinates, spv::DecorationInvariant);
main_interface.push_back(out_point_coordinates);
++output_location;
}
uint32_t input_location = 0;
// Interpolator inputs.
std::array<spv::Id, xenos::kMaxInterpolators> in_interpolators;
for (uint32_t i = 0; i < key.interpolator_count; ++i) {
@@ -1010,8 +1094,10 @@ VkShaderModule VulkanPipelineCache::GetGeometryShader(GeometryShaderKey key) {
0),
fmt::format("xe_in_interpolator_{}", i).c_str());
in_interpolators[i] = in_interpolator;
builder.addDecoration(in_interpolator, spv::DecorationLocation, i);
builder.addDecoration(in_interpolator, spv::DecorationLocation,
int(input_location));
main_interface.push_back(in_interpolator);
++input_location;
}
// Point size input.
@@ -1023,8 +1109,9 @@ VkShaderModule VulkanPipelineCache::GetGeometryShader(GeometryShaderKey key) {
0),
"xe_in_point_size");
builder.addDecoration(in_point_size, spv::DecorationLocation,
key.interpolator_count);
int(input_location));
main_interface.push_back(in_point_size);
++input_location;
}
// out gl_PerVertex.
@@ -1198,6 +1285,231 @@ VkShaderModule VulkanPipelineCache::GetGeometryShader(GeometryShaderKey key) {
}
switch (key.type) {
case PipelineGeometryShader::kPointList: {
// Expand the point sprite, with left-to-right, top-to-bottom UVs.
spv::Id const_int_0 = builder.makeIntConstant(0);
spv::Id const_int_1 = builder.makeIntConstant(1);
spv::Id const_float_0 = builder.makeFloatConstant(0.0f);
// Load the point diameter in guest pixels.
id_vector_temp.clear();
id_vector_temp.reserve(2);
id_vector_temp.push_back(
builder.makeIntConstant(int32_t(kPointConstantConstantDiameter)));
id_vector_temp.push_back(const_int_0);
spv::Id point_guest_diameter_x = builder.createLoad(
builder.createAccessChain(spv::StorageClassUniform,
uniform_system_constants, id_vector_temp),
spv::NoPrecision);
id_vector_temp.back() = const_int_1;
spv::Id point_guest_diameter_y = builder.createLoad(
builder.createAccessChain(spv::StorageClassUniform,
uniform_system_constants, id_vector_temp),
spv::NoPrecision);
if (key.has_point_size) {
// The vertex shader's header writes -1.0 to point_size by default, so
// any non-negative value means that it was overwritten by the
// translated vertex shader, and needs to be used instead of the
// constant size. The per-vertex diameter is already clamped in the
// vertex shader (combined with making it non-negative).
id_vector_temp.clear();
// 0 is the input primitive vertex index.
id_vector_temp.push_back(const_int_0);
spv::Id point_vertex_diameter = builder.createLoad(
builder.createAccessChain(spv::StorageClassInput, in_point_size,
id_vector_temp),
spv::NoPrecision);
spv::Id point_vertex_diameter_written =
builder.createBinOp(spv::OpFOrdGreaterThanEqual, type_bool,
point_vertex_diameter, const_float_0);
point_guest_diameter_x = builder.createTriOp(
spv::OpSelect, type_float, point_vertex_diameter_written,
point_vertex_diameter, point_guest_diameter_x);
point_guest_diameter_y = builder.createTriOp(
spv::OpSelect, type_float, point_vertex_diameter_written,
point_vertex_diameter, point_guest_diameter_y);
}
// 4D5307F1 has zero-size snowflakes, drop them quicker, and also drop
// points with a constant size of zero since point lists may also be used
// as just "compute" with memexport.
spv::Id point_size_not_zero = builder.createBinOp(
spv::OpLogicalAnd, type_bool,
builder.createBinOp(spv::OpFOrdGreaterThan, type_bool,
point_guest_diameter_x, const_float_0),
builder.createBinOp(spv::OpFOrdGreaterThan, type_bool,
point_guest_diameter_y, const_float_0));
spv::Block& point_size_zero_predecessor = *builder.getBuildPoint();
spv::Block& point_size_zero_then_block = builder.makeNewBlock();
spv::Block& point_size_zero_merge_block = builder.makeNewBlock();
{
std::unique_ptr<spv::Instruction> selection_merge_op(
std::make_unique<spv::Instruction>(spv::OpSelectionMerge));
selection_merge_op->addIdOperand(point_size_zero_merge_block.getId());
selection_merge_op->addImmediateOperand(
spv::SelectionControlDontFlattenMask);
point_size_zero_predecessor.addInstruction(
std::move(selection_merge_op));
}
{
std::unique_ptr<spv::Instruction> branch_conditional_op(
std::make_unique<spv::Instruction>(spv::OpBranchConditional));
branch_conditional_op->addIdOperand(point_size_not_zero);
branch_conditional_op->addIdOperand(
point_size_zero_merge_block.getId());
branch_conditional_op->addIdOperand(point_size_zero_then_block.getId());
branch_conditional_op->addImmediateOperand(2);
branch_conditional_op->addImmediateOperand(1);
point_size_zero_predecessor.addInstruction(
std::move(branch_conditional_op));
}
point_size_zero_then_block.addPredecessor(&point_size_zero_predecessor);
point_size_zero_merge_block.addPredecessor(&point_size_zero_predecessor);
builder.setBuildPoint(&point_size_zero_then_block);
builder.createNoResultOp(spv::OpReturn);
builder.setBuildPoint(&point_size_zero_merge_block);
// Transform the diameter in the guest screen coordinates to radius in the
// normalized device coordinates, and then to the clip space by
// multiplying by W.
id_vector_temp.clear();
id_vector_temp.reserve(2);
id_vector_temp.push_back(builder.makeIntConstant(
int32_t(kPointConstantScreenDiameterToNdcRadius)));
id_vector_temp.push_back(const_int_0);
spv::Id point_radius_x = builder.createBinOp(
spv::OpFMul, type_float, point_guest_diameter_x,
builder.createLoad(builder.createAccessChain(spv::StorageClassUniform,
uniform_system_constants,
id_vector_temp),
spv::NoPrecision));
builder.addDecoration(point_radius_x, spv::DecorationNoContraction);
id_vector_temp.back() = const_int_1;
spv::Id point_radius_y = builder.createBinOp(
spv::OpFMul, type_float, point_guest_diameter_y,
builder.createLoad(builder.createAccessChain(spv::StorageClassUniform,
uniform_system_constants,
id_vector_temp),
spv::NoPrecision));
builder.addDecoration(point_radius_y, spv::DecorationNoContraction);
id_vector_temp.clear();
id_vector_temp.reserve(2);
// 0 is the input primitive vertex index.
id_vector_temp.push_back(const_int_0);
id_vector_temp.push_back(const_member_in_gl_per_vertex_position);
spv::Id point_position = builder.createLoad(
builder.createAccessChain(spv::StorageClassInput, in_gl_per_vertex,
id_vector_temp),
spv::NoPrecision);
spv::Id point_w =
builder.createCompositeExtract(point_position, type_float, 3);
point_radius_x =
builder.createBinOp(spv::OpFMul, type_float, point_radius_x, point_w);
builder.addDecoration(point_radius_x, spv::DecorationNoContraction);
point_radius_y =
builder.createBinOp(spv::OpFMul, type_float, point_radius_y, point_w);
builder.addDecoration(point_radius_y, spv::DecorationNoContraction);
// Load the inputs for the guest point.
// Interpolators.
std::array<spv::Id, xenos::kMaxInterpolators> point_interpolators;
id_vector_temp.clear();
// 0 is the input primitive vertex index.
id_vector_temp.push_back(const_int_0);
for (uint32_t i = 0; i < key.interpolator_count; ++i) {
point_interpolators[i] = builder.createLoad(
builder.createAccessChain(spv::StorageClassInput,
in_interpolators[i], id_vector_temp),
spv::NoPrecision);
}
// Positions.
spv::Id point_x =
builder.createCompositeExtract(point_position, type_float, 0);
spv::Id point_y =
builder.createCompositeExtract(point_position, type_float, 1);
std::array<spv::Id, 2> point_edge_x, point_edge_y;
for (uint32_t i = 0; i < 2; ++i) {
spv::Op point_radius_add_op = i ? spv::OpFAdd : spv::OpFSub;
point_edge_x[i] = builder.createBinOp(point_radius_add_op, type_float,
point_x, point_radius_x);
builder.addDecoration(point_edge_x[i], spv::DecorationNoContraction);
point_edge_y[i] = builder.createBinOp(point_radius_add_op, type_float,
point_y, point_radius_y);
builder.addDecoration(point_edge_y[i], spv::DecorationNoContraction);
};
spv::Id point_z =
builder.createCompositeExtract(point_position, type_float, 2);
// Clip distances.
spv::Id point_clip_distances = spv::NoResult;
if (clip_distance_count) {
id_vector_temp.clear();
id_vector_temp.reserve(2);
// 0 is the input primitive vertex index.
id_vector_temp.push_back(const_int_0);
id_vector_temp.push_back(const_member_in_gl_per_vertex_clip_distance);
point_clip_distances = builder.createLoad(
builder.createAccessChain(spv::StorageClassInput, in_gl_per_vertex,
id_vector_temp),
spv::NoPrecision);
}
for (uint32_t i = 0; i < 4; ++i) {
// Same interpolators for the entire sprite.
for (uint32_t j = 0; j < key.interpolator_count; ++j) {
builder.createStore(point_interpolators[j], out_interpolators[j]);
}
// Top-left, bottom-left, top-right, bottom-right order (chosen
// arbitrarily, simply based on counterclockwise meaning front with
// frontFace = VkFrontFace(0), but faceness is ignored for non-polygon
// primitive types).
uint32_t point_vertex_x = i >> 1;
uint32_t point_vertex_y = i & 1;
// Point coordinates.
if (key.has_point_coordinates) {
id_vector_temp.clear();
id_vector_temp.reserve(2);
id_vector_temp.push_back(
builder.makeFloatConstant(float(point_vertex_x)));
id_vector_temp.push_back(
builder.makeFloatConstant(float(point_vertex_y)));
builder.createStore(
builder.makeCompositeConstant(type_float2, id_vector_temp),
out_point_coordinates);
}
// Position.
id_vector_temp.clear();
id_vector_temp.reserve(4);
id_vector_temp.push_back(point_edge_x[point_vertex_x]);
id_vector_temp.push_back(point_edge_y[point_vertex_y]);
id_vector_temp.push_back(point_z);
id_vector_temp.push_back(point_w);
spv::Id point_vertex_position =
builder.createCompositeConstruct(type_float4, id_vector_temp);
id_vector_temp.clear();
id_vector_temp.push_back(const_member_out_gl_per_vertex_position);
builder.createStore(
point_vertex_position,
builder.createAccessChain(spv::StorageClassOutput,
out_gl_per_vertex, id_vector_temp));
// Clip distances.
// TODO(Triang3l): Handle ps_ucp_mode properly, clip expanded points if
// needed.
if (clip_distance_count) {
id_vector_temp.clear();
id_vector_temp.push_back(
const_member_out_gl_per_vertex_clip_distance);
builder.createStore(
point_clip_distances,
builder.createAccessChain(spv::StorageClassOutput,
out_gl_per_vertex, id_vector_temp));
}
// Emit the vertex.
builder.createNoResultOp(spv::OpEmitVertex);
}
builder.createNoResultOp(spv::OpEndPrimitive);
} break;
case PipelineGeometryShader::kRectangleList: {
// Construct a strip with the fourth vertex generated by mirroring a
// vertex across the longest edge (the diagonal).
@@ -1308,8 +1620,8 @@ VkShaderModule VulkanPipelineCache::GetGeometryShader(GeometryShaderKey key) {
id_vector_temp.reserve(2);
id_vector_temp.push_back(const_float_0);
id_vector_temp.push_back(const_float_0);
const_point_coordinates_zero = builder.makeCompositeConstant(
type_point_coordinates, id_vector_temp);
const_point_coordinates_zero =
builder.makeCompositeConstant(type_float2, id_vector_temp);
}
// Emit the triangle in the strip that consists of the original vertices.
@@ -1491,8 +1803,8 @@ VkShaderModule VulkanPipelineCache::GetGeometryShader(GeometryShaderKey key) {
id_vector_temp.reserve(2);
id_vector_temp.push_back(const_float_0);
id_vector_temp.push_back(const_float_0);
const_point_coordinates_zero = builder.makeCompositeConstant(
type_point_coordinates, id_vector_temp);
const_point_coordinates_zero =
builder.makeCompositeConstant(type_float2, id_vector_temp);
}
// Build the triangle strip from the original quad vertices in the

4
src/xenia/gpu/vulkan/vulkan_pipeline_cache.h
View File

@@ -71,7 +71,7 @@ class VulkanPipelineCache {
SpirvShaderTranslator::Modification GetCurrentVertexShaderModification(
const Shader& shader,
Shader::HostVertexShaderType host_vertex_shader_type,
uint32_t interpolator_mask) const;
uint32_t interpolator_mask, bool ps_param_gen_used) const;
SpirvShaderTranslator::Modification GetCurrentPixelShaderModification(
const Shader& shader, uint32_t interpolator_mask,
uint32_t param_gen_pos) const;
@@ -92,6 +92,7 @@ class VulkanPipelineCache {
private:
enum class PipelineGeometryShader : uint32_t {
kNone,
kPointList,
kRectangleList,
kQuadList,
};
@@ -267,6 +268,7 @@ class VulkanPipelineCache {
static bool GetGeometryShaderKey(
PipelineGeometryShader geometry_shader_type,
SpirvShaderTranslator::Modification vertex_shader_modification,
SpirvShaderTranslator::Modification pixel_shader_modification,
GeometryShaderKey& key_out);
VkShaderModule GetGeometryShader(GeometryShaderKey key);

29
src/xenia/gpu/vulkan/vulkan_primitive_processor.cc
View File

@@ -27,17 +27,18 @@ namespace vulkan {
VulkanPrimitiveProcessor::~VulkanPrimitiveProcessor() { Shutdown(true); }
bool VulkanPrimitiveProcessor::Initialize() {
// TODO(Triang3l): fullDrawIndexUint32 feature check and indirect index fetch.
const ui::vulkan::VulkanProvider& provider =
command_processor_.GetVulkanProvider();
const VkPhysicalDeviceFeatures& device_features = provider.device_features();
const VkPhysicalDevicePortabilitySubsetFeaturesKHR*
device_portability_subset_features =
provider.device_portability_subset_features();
if (!InitializeCommon(true,
if (!InitializeCommon(device_features.fullDrawIndexUint32,
!device_portability_subset_features ||
device_portability_subset_features->triangleFans,
false, device_features.geometryShader)) {
false, device_features.geometryShader,
device_features.geometryShader,
device_features.geometryShader)) {
Shutdown();
return false;
}
@@ -128,9 +129,9 @@ void VulkanPrimitiveProcessor::EndFrame() {
frame_index_buffers_.clear();
}
bool VulkanPrimitiveProcessor::InitializeBuiltin16BitIndexBuffer(
uint32_t index_count, std::function<void(uint16_t*)> fill_callback) {
assert_not_zero(index_count);
bool VulkanPrimitiveProcessor::InitializeBuiltinIndexBuffer(
size_t size_bytes, std::function<void(void*)> fill_callback) {
assert_not_zero(size_bytes);
assert_true(builtin_index_buffer_ == VK_NULL_HANDLE);
assert_true(builtin_index_buffer_memory_ == VK_NULL_HANDLE);
assert_true(builtin_index_buffer_upload_ == VK_NULL_HANDLE);
@@ -141,7 +142,7 @@ bool VulkanPrimitiveProcessor::InitializeBuiltin16BitIndexBuffer(
const ui::vulkan::VulkanProvider::DeviceFunctions& dfn = provider.dfn();
VkDevice device = provider.device();
builtin_index_buffer_size_ = VkDeviceSize(sizeof(uint16_t) * index_count);
builtin_index_buffer_size_ = VkDeviceSize(size_bytes);
if (!ui::vulkan::util::CreateDedicatedAllocationBuffer(
provider, builtin_index_buffer_size_,
VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_BUFFER_USAGE_INDEX_BUFFER_BIT,
@@ -149,8 +150,8 @@ bool VulkanPrimitiveProcessor::InitializeBuiltin16BitIndexBuffer(
builtin_index_buffer_memory_)) {
XELOGE(
"Vulkan primitive processor: Failed to create the built-in index "
"buffer GPU resource with {} 16-bit indices",
index_count);
"buffer GPU resource with {} bytes",
size_bytes);
return false;
}
uint32_t upload_memory_type;
@@ -162,8 +163,8 @@ bool VulkanPrimitiveProcessor::InitializeBuiltin16BitIndexBuffer(
&upload_memory_type)) {
XELOGE(
"Vulkan primitive processor: Failed to create the built-in index "
"buffer upload resource with {} 16-bit indices",
index_count);
"buffer upload resource with {} bytes",
size_bytes);
ui::vulkan::util::DestroyAndNullHandle(dfn.vkDestroyBuffer, device,
builtin_index_buffer_);
ui::vulkan::util::DestroyAndNullHandle(dfn.vkFreeMemory, device,
@@ -176,8 +177,8 @@ bool VulkanPrimitiveProcessor::InitializeBuiltin16BitIndexBuffer(
VK_WHOLE_SIZE, 0, &mapping) != VK_SUCCESS) {
XELOGE(
"Vulkan primitive processor: Failed to map the built-in index buffer "
"upload resource with {} 16-bit indices",
index_count);
"upload resource with {} bytes",
size_bytes);
ui::vulkan::util::DestroyAndNullHandle(dfn.vkDestroyBuffer, device,
builtin_index_buffer_upload_);
ui::vulkan::util::DestroyAndNullHandle(dfn.vkFreeMemory, device,
@@ -188,7 +189,7 @@ bool VulkanPrimitiveProcessor::InitializeBuiltin16BitIndexBuffer(
builtin_index_buffer_memory_);
return false;
}
fill_callback(reinterpret_cast<uint16_t*>(mapping));
fill_callback(mapping);
ui::vulkan::util::FlushMappedMemoryRange(
provider, builtin_index_buffer_memory_, upload_memory_type);
dfn.vkUnmapMemory(device, builtin_index_buffer_upload_memory_);

5
src/xenia/gpu/vulkan/vulkan_primitive_processor.h
View File

@@ -56,9 +56,8 @@ class VulkanPrimitiveProcessor final : public PrimitiveProcessor {
}
protected:
bool InitializeBuiltin16BitIndexBuffer(
uint32_t index_count,
std::function<void(uint16_t*)> fill_callback) override;
bool InitializeBuiltinIndexBuffer(
size_t size_bytes, std::function<void(void*)> fill_callback) override;
void* RequestHostConvertedIndexBufferForCurrentFrame(
xenos::IndexFormat format, uint32_t index_count, bool coalign_for_simd,