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

[D3D12] Various refactoring mostly on the immediate drawer side, cache immdraw scissor, force stretch alpha to 1

Browse Source
This commit is contained in:
Triang3l
2020-09-20 15:15:08 +03:00
parent 6ba0620b10
commit 183c7eee2b
21 changed files with 145 additions and 117 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
src/xenia/gpu/d3d12/d3d12_command_processor.h
View File

@@ -376,7 +376,7 @@ class D3D12CommandProcessor : public CommandProcessor {
CommandAllocator* command_allocator_submitted_last_ = nullptr;
ID3D12GraphicsCommandList* command_list_ = nullptr;
ID3D12GraphicsCommandList1* command_list_1_ = nullptr;
std::unique_ptr<DeferredCommandList> deferred_command_list_ = nullptr;
std::unique_ptr<DeferredCommandList> deferred_command_list_;
// Should bindless textures and samplers be used - many times faster
// UpdateBindings than bindful (that becomes a significant bottleneck with
@@ -388,14 +388,12 @@ class D3D12CommandProcessor : public CommandProcessor {
// targets.
bool edram_rov_used_ = false;
std::unique_ptr<ui::d3d12::D3D12UploadBufferPool> constant_buffer_pool_ =
nullptr;
std::unique_ptr<ui::d3d12::D3D12UploadBufferPool> constant_buffer_pool_;
static constexpr uint32_t kViewBindfulHeapSize = 32768;
static_assert(kViewBindfulHeapSize <=
D3D12_MAX_SHADER_VISIBLE_DESCRIPTOR_HEAP_SIZE_TIER_1);
std::unique_ptr<ui::d3d12::D3D12DescriptorHeapPool> view_bindful_heap_pool_ =
nullptr;
std::unique_ptr<ui::d3d12::D3D12DescriptorHeapPool> view_bindful_heap_pool_;
// Currently bound descriptor heap - updated by RequestViewBindfulDescriptors.
ID3D12DescriptorHeap* view_bindful_heap_current_;
// Rationale: textures have 4 KB alignment in guest memory, and there can be
@@ -426,7 +424,7 @@ class D3D12CommandProcessor : public CommandProcessor {
static constexpr uint32_t kSamplerHeapSize = 2000;
static_assert(kSamplerHeapSize <= D3D12_MAX_SHADER_VISIBLE_SAMPLER_HEAP_SIZE);
std::unique_ptr<ui::d3d12::D3D12DescriptorHeapPool>
sampler_bindful_heap_pool_ = nullptr;
sampler_bindful_heap_pool_;
ID3D12DescriptorHeap* sampler_bindful_heap_current_;
ID3D12DescriptorHeap* sampler_bindless_heap_current_ = nullptr;
D3D12_CPU_DESCRIPTOR_HANDLE sampler_bindless_heap_cpu_start_;
@@ -452,15 +450,15 @@ class D3D12CommandProcessor : public CommandProcessor {
ID3D12RootSignature* root_signature_bindless_vs_ = nullptr;
ID3D12RootSignature* root_signature_bindless_ds_ = nullptr;
std::unique_ptr<SharedMemory> shared_memory_ = nullptr;
std::unique_ptr<SharedMemory> shared_memory_;
std::unique_ptr<PipelineCache> pipeline_cache_ = nullptr;
std::unique_ptr<PipelineCache> pipeline_cache_;
std::unique_ptr<TextureCache> texture_cache_ = nullptr;
std::unique_ptr<TextureCache> texture_cache_;
std::unique_ptr<RenderTargetCache> render_target_cache_ = nullptr;
std::unique_ptr<RenderTargetCache> render_target_cache_;
std::unique_ptr<PrimitiveConverter> primitive_converter_ = nullptr;
std::unique_ptr<PrimitiveConverter> primitive_converter_;
// Mip 0 contains the normal gamma ramp (256 entries), mip 1 contains the PWL
// ramp (128 entries). DXGI_FORMAT_R10G10B10A2_UNORM 1D.

1
src/xenia/gpu/d3d12/d3d12_graphics_system.cc
View File

@@ -142,6 +142,7 @@ X_STATUS D3D12GraphicsSystem::Setup(cpu::Processor* processor,
stretch_pipeline_desc.VS.BytecodeLength = sizeof(fullscreen_vs);
stretch_pipeline_desc.PS.pShaderBytecode = stretch_ps;
stretch_pipeline_desc.PS.BytecodeLength = sizeof(stretch_ps);
// The shader will set alpha to 1, don't use output-merger to preserve it.
stretch_pipeline_desc.BlendState.RenderTarget[0].RenderTargetWriteMask =
D3D12_COLOR_WRITE_ENABLE_ALL;
stretch_pipeline_desc.SampleMask = UINT_MAX;

4
src/xenia/gpu/d3d12/pipeline_cache.h
View File

@@ -247,7 +247,7 @@ class PipelineCache {
uint32_t resolution_scale_;
// Reusable shader translator.
std::unique_ptr<DxbcShaderTranslator> shader_translator_ = nullptr;
std::unique_ptr<DxbcShaderTranslator> shader_translator_;
// Command processor thread DXIL conversion/disassembly interfaces, if DXIL
// disassembly is enabled.
@@ -344,7 +344,7 @@ class PipelineCache {
// Manual-reset event set when the last queued pipeline state object is
// created and there are no more pipeline state objects to create. This is
// triggered by the thread creating the last pipeline state object.
std::unique_ptr<xe::threading::Event> creation_completion_event_ = nullptr;
std::unique_ptr<xe::threading::Event> creation_completion_event_;
// Whether setting the event on completion is queued. Protected with
// creation_request_lock_, notify_one creation_request_cond_ when set.
bool creation_completion_set_event_ = false;

2
src/xenia/gpu/d3d12/primitive_converter.h
View File

@@ -107,7 +107,7 @@ class PrimitiveConverter {
Memory& memory_;
TraceWriter& trace_writer_;
std::unique_ptr<ui::d3d12::D3D12UploadBufferPool> buffer_pool_ = nullptr;
std::unique_ptr<ui::d3d12::D3D12UploadBufferPool> buffer_pool_;
// Static index buffers for emulating unsupported primitive types when drawing
// without an index buffer.

2
src/xenia/gpu/d3d12/render_target_cache.h
View File

@@ -603,7 +603,7 @@ class RenderTargetCache {
// For traces only.
ID3D12Resource* edram_snapshot_download_buffer_ = nullptr;
std::unique_ptr<ui::d3d12::D3D12UploadBufferPool>
edram_snapshot_restore_pool_ = nullptr;
edram_snapshot_restore_pool_;
};
} // namespace d3d12

BIN
src/xenia/gpu/d3d12/shaders/dxbc/stretch_gamma_ps.cso
View File

Binary file not shown.

42
src/xenia/gpu/d3d12/shaders/dxbc/stretch_gamma_ps.h
View File

@@ -1,11 +1,11 @@
// generated from `xb buildhlsl`
// source: stretch_gamma.ps.hlsl
const uint8_t stretch_gamma_ps[] = {
0x44, 0x58, 0x42, 0x43, 0x8D, 0x1B, 0xAE, 0x52, 0x96, 0x87, 0x7B, 0x8F,
0xCB, 0xD3, 0xBC, 0x4B, 0xCF, 0xAF, 0x66, 0xA9, 0x01, 0x00, 0x00, 0x00,
0x10, 0x05, 0x00, 0x00, 0x05, 0x00, 0x00, 0x00, 0x34, 0x00, 0x00, 0x00,
0x44, 0x58, 0x42, 0x43, 0x46, 0xCD, 0x15, 0xA8, 0x4D, 0x5B, 0xFD, 0x07,
0x4C, 0x62, 0xA7, 0xD0, 0x61, 0x89, 0x85, 0xA6, 0x01, 0x00, 0x00, 0x00,
0x24, 0x05, 0x00, 0x00, 0x05, 0x00, 0x00, 0x00, 0x34, 0x00, 0x00, 0x00,
0x14, 0x02, 0x00, 0x00, 0x48, 0x02, 0x00, 0x00, 0x7C, 0x02, 0x00, 0x00,
0x74, 0x04, 0x00, 0x00, 0x52, 0x44, 0x45, 0x46, 0xD8, 0x01, 0x00, 0x00,
0x88, 0x04, 0x00, 0x00, 0x52, 0x44, 0x45, 0x46, 0xD8, 0x01, 0x00, 0x00,
0x01, 0x00, 0x00, 0x00, 0x2C, 0x01, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00,
0x3C, 0x00, 0x00, 0x00, 0x01, 0x05, 0xFF, 0xFF, 0x00, 0x05, 0x00, 0x00,
0xB0, 0x01, 0x00, 0x00, 0x13, 0x13, 0x44, 0x25, 0x3C, 0x00, 0x00, 0x00,
@@ -16,7 +16,7 @@ const uint8_t stretch_gamma_ps[] = {
0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0xF4, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
0x05, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF,
0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x0C, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0x00, 0x00, 0x00,
0x02, 0x00, 0x00, 0x00, 0x05, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
0xFF, 0xFF, 0xFF, 0xFF, 0x01, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
@@ -54,8 +54,8 @@ const uint8_t stretch_gamma_ps[] = {
0x20, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x03, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0F, 0x00, 0x00, 0x00,
0x53, 0x56, 0x5F, 0x54, 0x61, 0x72, 0x67, 0x65, 0x74, 0x00, 0xAB, 0xAB,
0x53, 0x48, 0x45, 0x58, 0xF0, 0x01, 0x00, 0x00, 0x51, 0x00, 0x00, 0x00,
0x7C, 0x00, 0x00, 0x00, 0x6A, 0x08, 0x00, 0x01, 0x59, 0x00, 0x00, 0x07,
0x53, 0x48, 0x45, 0x58, 0x04, 0x02, 0x00, 0x00, 0x51, 0x00, 0x00, 0x00,
0x81, 0x00, 0x00, 0x00, 0x6A, 0x08, 0x00, 0x01, 0x59, 0x00, 0x00, 0x07,
0x46, 0x8E, 0x30, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x5A, 0x00, 0x00, 0x06, 0x46, 0x6E, 0x30, 0x00, 0x00, 0x00, 0x00, 0x00,
@@ -67,20 +67,20 @@ const uint8_t stretch_gamma_ps[] = {
0x55, 0x55, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x62, 0x10, 0x00, 0x03,
0x32, 0x10, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x65, 0x00, 0x00, 0x03,
0xF2, 0x20, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x68, 0x00, 0x00, 0x02,
0x02, 0x00, 0x00, 0x00, 0x48, 0x00, 0x00, 0x0D, 0xF2, 0x00, 0x10, 0x00,
0x02, 0x00, 0x00, 0x00, 0x48, 0x00, 0x00, 0x0D, 0x72, 0x00, 0x10, 0x00,
0x00, 0x00, 0x00, 0x00, 0x46, 0x10, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00,
0x46, 0x7E, 0x20, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x60, 0x20, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x01, 0x40, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0A,
0x12, 0x00, 0x10, 0x00, 0x01, 0x00, 0x00, 0x00, 0x0A, 0x80, 0x30, 0x80,
0x82, 0x00, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0A, 0x80, 0x30, 0x80,
0x41, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x01, 0x40, 0x00, 0x00, 0x00, 0x00, 0x80, 0x3F,
0x38, 0x00, 0x00, 0x09, 0x22, 0x00, 0x10, 0x00, 0x01, 0x00, 0x00, 0x00,
0x38, 0x00, 0x00, 0x09, 0x12, 0x00, 0x10, 0x00, 0x01, 0x00, 0x00, 0x00,
0x0A, 0x80, 0x30, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x01, 0x40, 0x00, 0x00, 0x00, 0x00, 0x00, 0x3F,
0x32, 0x00, 0x00, 0x09, 0x72, 0x00, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00,
0x46, 0x02, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x06, 0x00, 0x10, 0x00,
0x01, 0x00, 0x00, 0x00, 0x56, 0x05, 0x10, 0x00, 0x01, 0x00, 0x00, 0x00,
0x46, 0x02, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0xF6, 0x0F, 0x10, 0x00,
0x00, 0x00, 0x00, 0x00, 0x06, 0x00, 0x10, 0x00, 0x01, 0x00, 0x00, 0x00,
0x48, 0x00, 0x00, 0x0D, 0x12, 0x00, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00,
0x0A, 0x00, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x46, 0x7E, 0x20, 0x00,
0x01, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x00, 0x60, 0x20, 0x00,
@@ -94,19 +94,21 @@ const uint8_t stretch_gamma_ps[] = {
0x00, 0x00, 0x00, 0x00, 0x46, 0x7E, 0x20, 0x00, 0x01, 0x00, 0x00, 0x00,
0x01, 0x00, 0x00, 0x00, 0x00, 0x60, 0x20, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x01, 0x40, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x36, 0x00, 0x00, 0x05, 0xF2, 0x20, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00,
0x46, 0x0E, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x3E, 0x00, 0x00, 0x01,
0x53, 0x54, 0x41, 0x54, 0x94, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00,
0x02, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
0x03, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x36, 0x00, 0x00, 0x05, 0x72, 0x20, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00,
0x46, 0x02, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x36, 0x00, 0x00, 0x05,
0x82, 0x20, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x40, 0x00, 0x00,
0x00, 0x00, 0x80, 0x3F, 0x3E, 0x00, 0x00, 0x01, 0x53, 0x54, 0x41, 0x54,
0x94, 0x00, 0x00, 0x00, 0x0A, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
};

15
src/xenia/gpu/d3d12/shaders/dxbc/stretch_gamma_ps.txt
View File

@@ -17,7 +17,7 @@
// Name Type Format Dim ID HLSL Bind Count
// ------------------------------ ---------- ------- ----------- ------- -------------- ------
// xe_sampler_linear_clamp sampler NA NA S0 s0 1
// xe_texture texture float4 2d T0 t0 1
// xe_texture texture float3 2d T0 t0 1
// xe_gamma_ramp texture float3 1d T1 t1 1
// XeStretchGammaRootConstants cbuffer NA NA CB0 cb0 1
//
@@ -45,13 +45,14 @@ dcl_resource_texture1d (float,float,float,float) T1[1:1], space=0
dcl_input_ps linear v0.xy
dcl_output o0.xyzw
dcl_temps 2
sample_l r0.xyzw, v0.xyxx, T0[0].xyzw, S0[0], l(0.000000)
add r1.x, -CB0[0][0].x, l(1.000000)
mul r1.y, CB0[0][0].x, l(0.500000)
mad r0.xyz, r0.xyzx, r1.xxxx, r1.yyyy
sample_l r0.xyz, v0.xyxx, T0[0].xyzw, S0[0], l(0.000000)
add r0.w, -CB0[0][0].x, l(1.000000)
mul r1.x, CB0[0][0].x, l(0.500000)
mad r0.xyz, r0.xyzx, r0.wwww, r1.xxxx
sample_l r0.x, r0.x, T1[1].xyzw, S0[0], l(0.000000)
sample_l r0.y, r0.y, T1[1].xyzw, S0[0], l(0.000000)
sample_l r0.z, r0.z, T1[1].xyzw, S0[0], l(0.000000)
mov o0.xyzw, r0.xyzw
mov o0.xyz, r0.xyzx
mov o0.w, l(1.000000)
ret
// Approximately 9 instruction slots used
// Approximately 10 instruction slots used

BIN
src/xenia/gpu/d3d12/shaders/dxbc/stretch_ps.cso
View File

Binary file not shown.

30
src/xenia/gpu/d3d12/shaders/dxbc/stretch_ps.h
View File

@@ -1,11 +1,11 @@
// generated from `xb buildhlsl`
// source: stretch.ps.hlsl
const uint8_t stretch_ps[] = {
0x44, 0x58, 0x42, 0x43, 0x84, 0x8B, 0x62, 0x99, 0x4D, 0x9D, 0x1A, 0x13,
0x49, 0x13, 0xF7, 0x6B, 0x9F, 0x69, 0x25, 0x8A, 0x01, 0x00, 0x00, 0x00,
0xB0, 0x02, 0x00, 0x00, 0x05, 0x00, 0x00, 0x00, 0x34, 0x00, 0x00, 0x00,
0x44, 0x58, 0x42, 0x43, 0x0B, 0x39, 0x5E, 0x9C, 0x0F, 0x3E, 0xAC, 0xF6,
0x9E, 0xB3, 0x11, 0xC0, 0x9B, 0x25, 0x2E, 0x6E, 0x01, 0x00, 0x00, 0x00,
0xE0, 0x02, 0x00, 0x00, 0x05, 0x00, 0x00, 0x00, 0x34, 0x00, 0x00, 0x00,
0x14, 0x01, 0x00, 0x00, 0x48, 0x01, 0x00, 0x00, 0x7C, 0x01, 0x00, 0x00,
0x14, 0x02, 0x00, 0x00, 0x52, 0x44, 0x45, 0x46, 0xD8, 0x00, 0x00, 0x00,
0x44, 0x02, 0x00, 0x00, 0x52, 0x44, 0x45, 0x46, 0xD8, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
0x3C, 0x00, 0x00, 0x00, 0x01, 0x05, 0xFF, 0xFF, 0x00, 0x05, 0x00, 0x00,
0xAF, 0x00, 0x00, 0x00, 0x13, 0x13, 0x44, 0x25, 0x3C, 0x00, 0x00, 0x00,
@@ -16,7 +16,7 @@ const uint8_t stretch_ps[] = {
0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0xA4, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
0x05, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF,
0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x0C, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x78, 0x65, 0x5F, 0x73,
0x61, 0x6D, 0x70, 0x6C, 0x65, 0x72, 0x5F, 0x6C, 0x69, 0x6E, 0x65, 0x61,
0x72, 0x5F, 0x63, 0x6C, 0x61, 0x6D, 0x70, 0x00, 0x78, 0x65, 0x5F, 0x74,
@@ -33,26 +33,30 @@ const uint8_t stretch_ps[] = {
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x0F, 0x00, 0x00, 0x00, 0x53, 0x56, 0x5F, 0x54,
0x61, 0x72, 0x67, 0x65, 0x74, 0x00, 0xAB, 0xAB, 0x53, 0x48, 0x45, 0x58,
0x90, 0x00, 0x00, 0x00, 0x51, 0x00, 0x00, 0x00, 0x24, 0x00, 0x00, 0x00,
0xC0, 0x00, 0x00, 0x00, 0x51, 0x00, 0x00, 0x00, 0x30, 0x00, 0x00, 0x00,
0x6A, 0x08, 0x00, 0x01, 0x5A, 0x00, 0x00, 0x06, 0x46, 0x6E, 0x30, 0x00,
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x58, 0x18, 0x00, 0x07, 0x46, 0x7E, 0x30, 0x00,
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x55, 0x55, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x62, 0x10, 0x00, 0x03,
0x32, 0x10, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x65, 0x00, 0x00, 0x03,
0xF2, 0x20, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x48, 0x00, 0x00, 0x0D,
0xF2, 0x20, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x46, 0x10, 0x10, 0x00,
0x00, 0x00, 0x00, 0x00, 0x46, 0x7E, 0x20, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x00, 0x60, 0x20, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x01, 0x40, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0xF2, 0x20, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x68, 0x00, 0x00, 0x02,
0x01, 0x00, 0x00, 0x00, 0x48, 0x00, 0x00, 0x0D, 0x72, 0x00, 0x10, 0x00,
0x00, 0x00, 0x00, 0x00, 0x46, 0x10, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00,
0x46, 0x7E, 0x20, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x60, 0x20, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x01, 0x40, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x36, 0x00, 0x00, 0x05,
0x72, 0x20, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x46, 0x02, 0x10, 0x00,
0x00, 0x00, 0x00, 0x00, 0x36, 0x00, 0x00, 0x05, 0x82, 0x20, 0x10, 0x00,
0x00, 0x00, 0x00, 0x00, 0x01, 0x40, 0x00, 0x00, 0x00, 0x00, 0x80, 0x3F,
0x3E, 0x00, 0x00, 0x01, 0x53, 0x54, 0x41, 0x54, 0x94, 0x00, 0x00, 0x00,
0x02, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x04, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x02, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,

9
src/xenia/gpu/d3d12/shaders/dxbc/stretch_ps.txt
View File

@@ -7,7 +7,7 @@
// Name Type Format Dim ID HLSL Bind Count
// ------------------------------ ---------- ------- ----------- ------- -------------- ------
// xe_sampler_linear_clamp sampler NA NA S0 s0 1
// xe_texture texture float4 2d T0 t0 1
// xe_texture texture float3 2d T0 t0 1
//
//
//
@@ -30,6 +30,9 @@ dcl_sampler S0[0:0], mode_default, space=0
dcl_resource_texture2d (float,float,float,float) T0[0:0], space=0
dcl_input_ps linear v0.xy
dcl_output o0.xyzw
sample_l o0.xyzw, v0.xyxx, T0[0].xyzw, S0[0], l(0.000000)
dcl_temps 1
sample_l r0.xyz, v0.xyxx, T0[0].xyzw, S0[0], l(0.000000)
mov o0.xyz, r0.xyzx
mov o0.w, l(1.000000)
ret
// Approximately 2 instruction slots used
// Approximately 4 instruction slots used

6
src/xenia/gpu/d3d12/shaders/stretch.ps.hlsl
View File

@@ -1,6 +1,8 @@
Texture2D<float4> xe_texture : register(t0);
Texture2D<float3> xe_texture : register(t0);
SamplerState xe_sampler_linear_clamp : register(s0);
float4 main(float2 xe_texcoord : TEXCOORD) : SV_Target {
return xe_texture.SampleLevel(xe_sampler_linear_clamp, xe_texcoord, 0.0f);
// Force alpha to 1 to make sure the surface won't be translucent.
return float4(
xe_texture.SampleLevel(xe_sampler_linear_clamp, xe_texcoord, 0.0f), 1.0f);
}

11
src/xenia/gpu/d3d12/shaders/stretch_gamma.ps.hlsl
View File

@@ -1,4 +1,4 @@
Texture2D<float4> xe_texture : register(t0);
Texture2D<float3> xe_texture : register(t0);
Texture1D<float3> xe_gamma_ramp : register(t1);
SamplerState xe_sampler_linear_clamp : register(s0);
cbuffer XeStretchGammaRootConstants : register(b0) {
@@ -6,14 +6,15 @@ cbuffer XeStretchGammaRootConstants : register(b0) {
};
float4 main(float2 xe_texcoord : TEXCOORD) : SV_Target {
float4 color =
float3 color =
xe_texture.SampleLevel(xe_sampler_linear_clamp, xe_texcoord, 0.0f);
// The center of the first texel of the LUT contains the value for 0, and the
// center of the last texel contains the value for 1.
color.rgb = color.rgb * (1.0f - xe_gamma_ramp_inv_size) +
(0.5 * xe_gamma_ramp_inv_size);
color =
color * (1.0f - xe_gamma_ramp_inv_size) + (0.5 * xe_gamma_ramp_inv_size);
color.r = xe_gamma_ramp.SampleLevel(xe_sampler_linear_clamp, color.r, 0.0f).r;
color.g = xe_gamma_ramp.SampleLevel(xe_sampler_linear_clamp, color.g, 0.0f).g;
color.b = xe_gamma_ramp.SampleLevel(xe_sampler_linear_clamp, color.b, 0.0f).b;
return color;
// Force alpha to 1 to make sure the surface won't be translucent.
return float4(color, 1.0f);
}

3
src/xenia/gpu/d3d12/shared_memory.h
View File

@@ -212,8 +212,7 @@ class SharedMemory {
std::vector<UploadRange> upload_ranges_;
void GetRangesToUpload(uint32_t request_page_first,
uint32_t request_page_last);
std::unique_ptr<ui::d3d12::D3D12UploadBufferPool> upload_buffer_pool_ =
nullptr;
std::unique_ptr<ui::d3d12::D3D12UploadBufferPool> upload_buffer_pool_;
// GPU-written memory downloading for traces.
// Start page, length in pages.