1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
|
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <algorithm>
#include <cstddef>
#include <memory>
#include <vector>
#include "common/microprofile.h"
#include "core/core.h"
#include "core/memory.h"
#include "video_core/engines/kepler_compute.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/memory_manager.h"
#include "video_core/renderer_vulkan/fixed_pipeline_state.h"
#include "video_core/renderer_vulkan/maxwell_to_vk.h"
#include "video_core/renderer_vulkan/vk_compute_pipeline.h"
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/renderer_vulkan/vk_device.h"
#include "video_core/renderer_vulkan/vk_graphics_pipeline.h"
#include "video_core/renderer_vulkan/vk_pipeline_cache.h"
#include "video_core/renderer_vulkan/vk_rasterizer.h"
#include "video_core/renderer_vulkan/vk_renderpass_cache.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_update_descriptor.h"
#include "video_core/renderer_vulkan/wrapper.h"
#include "video_core/shader/compiler_settings.h"
#include "video_core/shader/memory_util.h"
namespace Vulkan {
MICROPROFILE_DECLARE(Vulkan_PipelineCache);
using Tegra::Engines::ShaderType;
using VideoCommon::Shader::GetShaderAddress;
using VideoCommon::Shader::GetShaderCode;
using VideoCommon::Shader::KERNEL_MAIN_OFFSET;
using VideoCommon::Shader::ProgramCode;
using VideoCommon::Shader::STAGE_MAIN_OFFSET;
namespace {
constexpr VkDescriptorType UNIFORM_BUFFER = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
constexpr VkDescriptorType STORAGE_BUFFER = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER;
constexpr VkDescriptorType UNIFORM_TEXEL_BUFFER = VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER;
constexpr VkDescriptorType COMBINED_IMAGE_SAMPLER = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
constexpr VkDescriptorType STORAGE_IMAGE = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE;
constexpr VideoCommon::Shader::CompilerSettings compiler_settings{
VideoCommon::Shader::CompileDepth::FullDecompile};
constexpr std::size_t GetStageFromProgram(std::size_t program) {
return program == 0 ? 0 : program - 1;
}
constexpr ShaderType GetStageFromProgram(Maxwell::ShaderProgram program) {
return static_cast<ShaderType>(GetStageFromProgram(static_cast<std::size_t>(program)));
}
ShaderType GetShaderType(Maxwell::ShaderProgram program) {
switch (program) {
case Maxwell::ShaderProgram::VertexB:
return ShaderType::Vertex;
case Maxwell::ShaderProgram::TesselationControl:
return ShaderType::TesselationControl;
case Maxwell::ShaderProgram::TesselationEval:
return ShaderType::TesselationEval;
case Maxwell::ShaderProgram::Geometry:
return ShaderType::Geometry;
case Maxwell::ShaderProgram::Fragment:
return ShaderType::Fragment;
default:
UNIMPLEMENTED_MSG("program={}", static_cast<u32>(program));
return ShaderType::Vertex;
}
}
template <VkDescriptorType descriptor_type, class Container>
void AddBindings(std::vector<VkDescriptorSetLayoutBinding>& bindings, u32& binding,
VkShaderStageFlags stage_flags, const Container& container) {
const u32 num_entries = static_cast<u32>(std::size(container));
for (std::size_t i = 0; i < num_entries; ++i) {
u32 count = 1;
if constexpr (descriptor_type == VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER) {
// Combined image samplers can be arrayed.
count = container[i].size;
}
VkDescriptorSetLayoutBinding& entry = bindings.emplace_back();
entry.binding = binding++;
entry.descriptorType = descriptor_type;
entry.descriptorCount = count;
entry.stageFlags = stage_flags;
entry.pImmutableSamplers = nullptr;
}
}
u32 FillDescriptorLayout(const ShaderEntries& entries,
std::vector<VkDescriptorSetLayoutBinding>& bindings,
Maxwell::ShaderProgram program_type, u32 base_binding) {
const ShaderType stage = GetStageFromProgram(program_type);
const VkShaderStageFlags flags = MaxwellToVK::ShaderStage(stage);
u32 binding = base_binding;
AddBindings<UNIFORM_BUFFER>(bindings, binding, flags, entries.const_buffers);
AddBindings<STORAGE_BUFFER>(bindings, binding, flags, entries.global_buffers);
AddBindings<UNIFORM_TEXEL_BUFFER>(bindings, binding, flags, entries.texel_buffers);
AddBindings<COMBINED_IMAGE_SAMPLER>(bindings, binding, flags, entries.samplers);
AddBindings<STORAGE_IMAGE>(bindings, binding, flags, entries.images);
return binding;
}
} // Anonymous namespace
std::size_t GraphicsPipelineCacheKey::Hash() const noexcept {
const u64 hash = Common::CityHash64(reinterpret_cast<const char*>(this), sizeof *this);
return static_cast<std::size_t>(hash);
}
bool GraphicsPipelineCacheKey::operator==(const GraphicsPipelineCacheKey& rhs) const noexcept {
return std::memcmp(&rhs, this, sizeof *this) == 0;
}
std::size_t ComputePipelineCacheKey::Hash() const noexcept {
const u64 hash = Common::CityHash64(reinterpret_cast<const char*>(this), sizeof *this);
return static_cast<std::size_t>(hash);
}
bool ComputePipelineCacheKey::operator==(const ComputePipelineCacheKey& rhs) const noexcept {
return std::memcmp(&rhs, this, sizeof *this) == 0;
}
CachedShader::CachedShader(Core::System& system, Tegra::Engines::ShaderType stage,
GPUVAddr gpu_addr, VAddr cpu_addr, ProgramCode program_code,
u32 main_offset)
: RasterizerCacheObject{cpu_addr}, gpu_addr{gpu_addr}, program_code{std::move(program_code)},
registry{stage, GetEngine(system, stage)}, shader_ir{this->program_code, main_offset,
compiler_settings, registry},
entries{GenerateShaderEntries(shader_ir)} {}
CachedShader::~CachedShader() = default;
Tegra::Engines::ConstBufferEngineInterface& CachedShader::GetEngine(
Core::System& system, Tegra::Engines::ShaderType stage) {
if (stage == Tegra::Engines::ShaderType::Compute) {
return system.GPU().KeplerCompute();
} else {
return system.GPU().Maxwell3D();
}
}
VKPipelineCache::VKPipelineCache(Core::System& system, RasterizerVulkan& rasterizer,
const VKDevice& device, VKScheduler& scheduler,
VKDescriptorPool& descriptor_pool,
VKUpdateDescriptorQueue& update_descriptor_queue,
VKRenderPassCache& renderpass_cache)
: RasterizerCache{rasterizer}, system{system}, device{device}, scheduler{scheduler},
descriptor_pool{descriptor_pool}, update_descriptor_queue{update_descriptor_queue},
renderpass_cache{renderpass_cache} {}
VKPipelineCache::~VKPipelineCache() = default;
std::array<Shader, Maxwell::MaxShaderProgram> VKPipelineCache::GetShaders() {
const auto& gpu = system.GPU().Maxwell3D();
std::array<Shader, Maxwell::MaxShaderProgram> shaders;
for (std::size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) {
const auto program{static_cast<Maxwell::ShaderProgram>(index)};
// Skip stages that are not enabled
if (!gpu.regs.IsShaderConfigEnabled(index)) {
continue;
}
auto& memory_manager{system.GPU().MemoryManager()};
const GPUVAddr program_addr{GetShaderAddress(system, program)};
const std::optional cpu_addr = memory_manager.GpuToCpuAddress(program_addr);
ASSERT(cpu_addr);
auto shader = cpu_addr ? TryGet(*cpu_addr) : null_shader;
if (!shader) {
const auto host_ptr{memory_manager.GetPointer(program_addr)};
// No shader found - create a new one
constexpr u32 stage_offset = STAGE_MAIN_OFFSET;
const auto stage = static_cast<Tegra::Engines::ShaderType>(index == 0 ? 0 : index - 1);
ProgramCode code = GetShaderCode(memory_manager, program_addr, host_ptr, false);
shader = std::make_shared<CachedShader>(system, stage, program_addr, *cpu_addr,
std::move(code), stage_offset);
if (cpu_addr) {
Register(shader);
} else {
null_shader = shader;
}
}
shaders[index] = std::move(shader);
}
return last_shaders = shaders;
}
VKGraphicsPipeline& VKPipelineCache::GetGraphicsPipeline(const GraphicsPipelineCacheKey& key) {
MICROPROFILE_SCOPE(Vulkan_PipelineCache);
if (last_graphics_pipeline && last_graphics_key == key) {
return *last_graphics_pipeline;
}
last_graphics_key = key;
const auto [pair, is_cache_miss] = graphics_cache.try_emplace(key);
auto& entry = pair->second;
if (is_cache_miss) {
LOG_INFO(Render_Vulkan, "Compile 0x{:016X}", key.Hash());
const auto [program, bindings] = DecompileShaders(key);
entry = std::make_unique<VKGraphicsPipeline>(device, scheduler, descriptor_pool,
update_descriptor_queue, renderpass_cache, key,
bindings, program);
}
return *(last_graphics_pipeline = entry.get());
}
VKComputePipeline& VKPipelineCache::GetComputePipeline(const ComputePipelineCacheKey& key) {
MICROPROFILE_SCOPE(Vulkan_PipelineCache);
const auto [pair, is_cache_miss] = compute_cache.try_emplace(key);
auto& entry = pair->second;
if (!is_cache_miss) {
return *entry;
}
LOG_INFO(Render_Vulkan, "Compile 0x{:016X}", key.Hash());
auto& memory_manager = system.GPU().MemoryManager();
const auto program_addr = key.shader;
const auto cpu_addr = memory_manager.GpuToCpuAddress(program_addr);
ASSERT(cpu_addr);
auto shader = cpu_addr ? TryGet(*cpu_addr) : null_kernel;
if (!shader) {
// No shader found - create a new one
const auto host_ptr = memory_manager.GetPointer(program_addr);
ProgramCode code = GetShaderCode(memory_manager, program_addr, host_ptr, true);
shader = std::make_shared<CachedShader>(system, Tegra::Engines::ShaderType::Compute,
program_addr, *cpu_addr, std::move(code),
KERNEL_MAIN_OFFSET);
if (cpu_addr) {
Register(shader);
} else {
null_kernel = shader;
}
}
Specialization specialization;
specialization.workgroup_size = key.workgroup_size;
specialization.shared_memory_size = key.shared_memory_size;
const SPIRVShader spirv_shader{Decompile(device, shader->GetIR(), ShaderType::Compute,
shader->GetRegistry(), specialization),
shader->GetEntries()};
entry = std::make_unique<VKComputePipeline>(device, scheduler, descriptor_pool,
update_descriptor_queue, spirv_shader);
return *entry;
}
void VKPipelineCache::Unregister(const Shader& shader) {
bool finished = false;
const auto Finish = [&] {
// TODO(Rodrigo): Instead of finishing here, wait for the fences that use this pipeline and
// flush.
if (finished) {
return;
}
finished = true;
scheduler.Finish();
};
const GPUVAddr invalidated_addr = shader->GetGpuAddr();
for (auto it = graphics_cache.begin(); it != graphics_cache.end();) {
auto& entry = it->first;
if (std::find(entry.shaders.begin(), entry.shaders.end(), invalidated_addr) ==
entry.shaders.end()) {
++it;
continue;
}
Finish();
it = graphics_cache.erase(it);
}
for (auto it = compute_cache.begin(); it != compute_cache.end();) {
auto& entry = it->first;
if (entry.shader != invalidated_addr) {
++it;
continue;
}
Finish();
it = compute_cache.erase(it);
}
RasterizerCache::Unregister(shader);
}
std::pair<SPIRVProgram, std::vector<VkDescriptorSetLayoutBinding>>
VKPipelineCache::DecompileShaders(const GraphicsPipelineCacheKey& key) {
const auto& fixed_state = key.fixed_state;
auto& memory_manager = system.GPU().MemoryManager();
const auto& gpu = system.GPU().Maxwell3D();
Specialization specialization;
if (fixed_state.rasterizer.Topology() == Maxwell::PrimitiveTopology::Points) {
float point_size;
std::memcpy(&point_size, &fixed_state.rasterizer.point_size, sizeof(float));
specialization.point_size = point_size;
ASSERT(point_size != 0.0f);
}
for (std::size_t i = 0; i < Maxwell::NumVertexAttributes; ++i) {
const auto& attribute = fixed_state.vertex_input.attributes[i];
specialization.enabled_attributes[i] = attribute.enabled.Value() != 0;
specialization.attribute_types[i] = attribute.Type();
}
specialization.ndc_minus_one_to_one = fixed_state.rasterizer.ndc_minus_one_to_one;
SPIRVProgram program;
std::vector<VkDescriptorSetLayoutBinding> bindings;
for (std::size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) {
const auto program_enum = static_cast<Maxwell::ShaderProgram>(index);
// Skip stages that are not enabled
if (!gpu.regs.IsShaderConfigEnabled(index)) {
continue;
}
const GPUVAddr gpu_addr = GetShaderAddress(system, program_enum);
const auto cpu_addr = memory_manager.GpuToCpuAddress(gpu_addr);
ASSERT(cpu_addr);
const auto shader = TryGet(*cpu_addr);
ASSERT(shader);
const std::size_t stage = index == 0 ? 0 : index - 1; // Stage indices are 0 - 5
const auto program_type = GetShaderType(program_enum);
const auto& entries = shader->GetEntries();
program[stage] = {
Decompile(device, shader->GetIR(), program_type, shader->GetRegistry(), specialization),
entries};
if (program_enum == Maxwell::ShaderProgram::VertexA) {
// VertexB was combined with VertexA, so we skip the VertexB iteration
++index;
}
const u32 old_binding = specialization.base_binding;
specialization.base_binding =
FillDescriptorLayout(entries, bindings, program_enum, specialization.base_binding);
ASSERT(old_binding + entries.NumBindings() == specialization.base_binding);
}
return {std::move(program), std::move(bindings)};
}
template <VkDescriptorType descriptor_type, class Container>
void AddEntry(std::vector<VkDescriptorUpdateTemplateEntry>& template_entries, u32& binding,
u32& offset, const Container& container) {
static constexpr u32 entry_size = static_cast<u32>(sizeof(DescriptorUpdateEntry));
const u32 count = static_cast<u32>(std::size(container));
if constexpr (descriptor_type == COMBINED_IMAGE_SAMPLER) {
for (u32 i = 0; i < count; ++i) {
const u32 num_samplers = container[i].size;
VkDescriptorUpdateTemplateEntry& entry = template_entries.emplace_back();
entry.dstBinding = binding;
entry.dstArrayElement = 0;
entry.descriptorCount = num_samplers;
entry.descriptorType = descriptor_type;
entry.offset = offset;
entry.stride = entry_size;
++binding;
offset += num_samplers * entry_size;
}
return;
}
if constexpr (descriptor_type == UNIFORM_TEXEL_BUFFER) {
// Nvidia has a bug where updating multiple uniform texels at once causes the driver to
// crash.
for (u32 i = 0; i < count; ++i) {
VkDescriptorUpdateTemplateEntry& entry = template_entries.emplace_back();
entry.dstBinding = binding + i;
entry.dstArrayElement = 0;
entry.descriptorCount = 1;
entry.descriptorType = descriptor_type;
entry.offset = offset + i * entry_size;
entry.stride = entry_size;
}
} else if (count > 0) {
VkDescriptorUpdateTemplateEntry& entry = template_entries.emplace_back();
entry.dstBinding = binding;
entry.dstArrayElement = 0;
entry.descriptorCount = count;
entry.descriptorType = descriptor_type;
entry.offset = offset;
entry.stride = entry_size;
}
offset += count * entry_size;
binding += count;
}
void FillDescriptorUpdateTemplateEntries(
const ShaderEntries& entries, u32& binding, u32& offset,
std::vector<VkDescriptorUpdateTemplateEntryKHR>& template_entries) {
AddEntry<UNIFORM_BUFFER>(template_entries, offset, binding, entries.const_buffers);
AddEntry<STORAGE_BUFFER>(template_entries, offset, binding, entries.global_buffers);
AddEntry<UNIFORM_TEXEL_BUFFER>(template_entries, offset, binding, entries.texel_buffers);
AddEntry<COMBINED_IMAGE_SAMPLER>(template_entries, offset, binding, entries.samplers);
AddEntry<STORAGE_IMAGE>(template_entries, offset, binding, entries.images);
}
} // namespace Vulkan
|