ShuriZma
/
xenia
mirror of https://github.com/xenia-project/xenia.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

[Vulkan] Remove UB-based independent blend logic 

On Vulkan, unlike Direct3D, not writing to a color target in the fragment shader produces an undefined result.
This commit is contained in:
Triang3l 2022-06-25 20:57:44 +03:00
parent d8b2944caa
commit a5c8df7a37
3 changed files with 9 additions and 44 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
src/xenia/gpu/spirv_shader_translator.cc
View File

@ -1270,19 +1270,16 @@ void SpirvShaderTranslator::StartFragmentShaderBeforeMain() {
"xe_out_fragment_data_2",
"xe_out_fragment_data_3",
};
uint32_t fragment_data_outputs_written =
current_shader().writes_color_targets() &
~GetSpirvShaderModification().pixel.color_outputs_disabled;
for (uint32_t i = 0; i < xenos::kMaxColorRenderTargets; ++i) {
if (!(fragment_data_outputs_written & (uint32_t(1) << i))) {
continue;
}
spv::Id output_fragment_data_rt =
builder_->createVariable(spv::NoPrecision, spv::StorageClassOutput,
type_float4_, kFragmentDataNames[i]);
output_fragment_data_[i] = output_fragment_data_rt;
uint32_t color_targets_remaining = current_shader().writes_color_targets();
uint32_t color_target_index;
while (xe::bit_scan_forward(color_targets_remaining, &color_target_index)) {
color_targets_remaining &= ~(UINT32_C(1) << color_target_index);
spv::Id output_fragment_data_rt = builder_->createVariable(
spv::NoPrecision, spv::StorageClassOutput, type_float4_,
kFragmentDataNames[color_target_index]);
output_fragment_data_[color_target_index] = output_fragment_data_rt;
builder_->addDecoration(output_fragment_data_rt, spv::DecorationLocation,
int(i));
int(color_target_index));
// Make invariant as pixel shaders may be used for various precise
// computations.
builder_->addDecoration(output_fragment_data_rt, spv::DecorationInvariant);

5
src/xenia/gpu/spirv_shader_translator.h
View File

@ -46,11 +46,6 @@ class SpirvShaderTranslator : public ShaderTranslator {
struct PixelShaderModification {
// Dynamically indexable register count from SQ_PROGRAM_CNTL.
uint32_t dynamic_addressable_register_count : 8;
// Color outputs removed from the shader to implement a zero color write
// mask when independent blending (and thus independent write masks) is
// not supported without switching to a render pass with some attachments
// actually excluded.
uint32_t color_outputs_disabled : 4;
} pixel;
uint64_t value = 0;

27
src/xenia/gpu/vulkan/vulkan_pipeline_cache.cc
View File

@ -141,33 +141,6 @@ VulkanPipelineCache::GetCurrentPixelShaderModification(
shader.GetDynamicAddressableRegisterCount(
sq_program_cntl.ps_num_reg)));
const ui::vulkan::VulkanProvider& provider =
command_processor_.GetVulkanProvider();
const VkPhysicalDeviceFeatures& device_features = provider.device_features();
if (!device_features.independentBlend) {
// Since without independent blending, the write mask is common for all
// attachments, but the render pass may still include the attachments from
// previous draws (to prevent excessive render pass changes potentially
// doing stores and loads), disable writing to render targets with a
// completely empty write mask by removing the output from the shader.
// Only explicitly excluding render targets that the shader actually writes
// to, for better pipeline storage compatibility between devices with and
// without independent blending (so in the usual situation - the shader
// doesn't write to any render targets disabled via the color mask - no
// explicit disabling of shader outputs will be needed, and the disabled
// output mask will be 0).
uint32_t color_targets_remaining = shader.writes_color_targets();
uint32_t color_target_index;
while (xe::bit_scan_forward(color_targets_remaining, &color_target_index)) {
color_targets_remaining &= ~(uint32_t(1) << color_target_index);
if (!(normalized_color_mask &
(uint32_t(0b1111) << (4 * color_target_index)))) {
modification.pixel.color_outputs_disabled |= uint32_t(1)
<< color_target_index;
}
}
}
return modification;
}