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
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
|
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <atomic>
#include <functional>
#include <mutex>
#include <optional>
#include <string>
#include <thread>
#include <unordered_set>
#include <boost/functional/hash.hpp>
#include "common/alignment.h"
#include "common/assert.h"
#include "common/logging/log.h"
#include "common/scope_exit.h"
#include "core/core.h"
#include "core/frontend/emu_window.h"
#include "video_core/engines/kepler_compute.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/engines/shader_type.h"
#include "video_core/memory_manager.h"
#include "video_core/renderer_opengl/gl_rasterizer.h"
#include "video_core/renderer_opengl/gl_shader_cache.h"
#include "video_core/renderer_opengl/gl_shader_decompiler.h"
#include "video_core/renderer_opengl/gl_shader_disk_cache.h"
#include "video_core/renderer_opengl/gl_state_tracker.h"
#include "video_core/renderer_opengl/utils.h"
#include "video_core/shader/shader_ir.h"
namespace OpenGL {
using Tegra::Engines::ShaderType;
using VideoCommon::Shader::ConstBufferLocker;
using VideoCommon::Shader::ProgramCode;
using VideoCommon::Shader::ShaderIR;
namespace {
constexpr u32 STAGE_MAIN_OFFSET = 10;
constexpr u32 KERNEL_MAIN_OFFSET = 0;
constexpr VideoCommon::Shader::CompilerSettings COMPILER_SETTINGS{};
/// Gets the address for the specified shader stage program
GPUVAddr GetShaderAddress(Core::System& system, Maxwell::ShaderProgram program) {
const auto& gpu{system.GPU().Maxwell3D()};
const auto& shader_config{gpu.regs.shader_config[static_cast<std::size_t>(program)]};
return gpu.regs.code_address.CodeAddress() + shader_config.offset;
}
/// Gets if the current instruction offset is a scheduler instruction
constexpr bool IsSchedInstruction(std::size_t offset, std::size_t main_offset) {
// Sched instructions appear once every 4 instructions.
constexpr std::size_t SchedPeriod = 4;
const std::size_t absolute_offset = offset - main_offset;
return (absolute_offset % SchedPeriod) == 0;
}
/// Calculates the size of a program stream
std::size_t CalculateProgramSize(const ProgramCode& program) {
constexpr std::size_t start_offset = 10;
// This is the encoded version of BRA that jumps to itself. All Nvidia
// shaders end with one.
constexpr u64 self_jumping_branch = 0xE2400FFFFF07000FULL;
constexpr u64 mask = 0xFFFFFFFFFF7FFFFFULL;
std::size_t offset = start_offset;
while (offset < program.size()) {
const u64 instruction = program[offset];
if (!IsSchedInstruction(offset, start_offset)) {
if ((instruction & mask) == self_jumping_branch) {
// End on Maxwell's "nop" instruction
break;
}
if (instruction == 0) {
break;
}
}
offset++;
}
// The last instruction is included in the program size
return std::min(offset + 1, program.size());
}
/// Gets the shader program code from memory for the specified address
ProgramCode GetShaderCode(Tegra::MemoryManager& memory_manager, const GPUVAddr gpu_addr,
const u8* host_ptr) {
ProgramCode code(VideoCommon::Shader::MAX_PROGRAM_LENGTH);
ASSERT_OR_EXECUTE(host_ptr != nullptr, {
std::fill(code.begin(), code.end(), 0);
return code;
});
memory_manager.ReadBlockUnsafe(gpu_addr, code.data(), code.size() * sizeof(u64));
code.resize(CalculateProgramSize(code));
return code;
}
/// Gets the shader type from a Maxwell program type
constexpr GLenum GetGLShaderType(ShaderType shader_type) {
switch (shader_type) {
case ShaderType::Vertex:
return GL_VERTEX_SHADER;
case ShaderType::Geometry:
return GL_GEOMETRY_SHADER;
case ShaderType::Fragment:
return GL_FRAGMENT_SHADER;
case ShaderType::Compute:
return GL_COMPUTE_SHADER;
default:
return GL_NONE;
}
}
/// Hashes one (or two) program streams
u64 GetUniqueIdentifier(ShaderType shader_type, bool is_a, const ProgramCode& code,
const ProgramCode& code_b = {}) {
u64 unique_identifier = boost::hash_value(code);
if (is_a) {
// VertexA programs include two programs
boost::hash_combine(unique_identifier, boost::hash_value(code_b));
}
return unique_identifier;
}
constexpr const char* GetShaderTypeName(ShaderType shader_type) {
switch (shader_type) {
case ShaderType::Vertex:
return "VS";
case ShaderType::TesselationControl:
return "HS";
case ShaderType::TesselationEval:
return "DS";
case ShaderType::Geometry:
return "GS";
case ShaderType::Fragment:
return "FS";
case ShaderType::Compute:
return "CS";
}
return "UNK";
}
constexpr ShaderType GetShaderType(Maxwell::ShaderProgram program_type) {
switch (program_type) {
case Maxwell::ShaderProgram::VertexA:
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;
}
return {};
}
std::string MakeShaderID(u64 unique_identifier, ShaderType shader_type) {
return fmt::format("{}{:016X}", GetShaderTypeName(shader_type), unique_identifier);
}
std::shared_ptr<ConstBufferLocker> MakeLocker(const ShaderDiskCacheEntry& entry) {
const VideoCore::GuestDriverProfile guest_profile{entry.texture_handler_size};
auto locker = std::make_shared<ConstBufferLocker>(entry.type, guest_profile);
locker->SetBoundBuffer(entry.bound_buffer);
for (const auto& [address, value] : entry.keys) {
const auto [buffer, offset] = address;
locker->InsertKey(buffer, offset, value);
}
for (const auto& [offset, sampler] : entry.bound_samplers) {
locker->InsertBoundSampler(offset, sampler);
}
for (const auto& [key, sampler] : entry.bindless_samplers) {
const auto [buffer, offset] = key;
locker->InsertBindlessSampler(buffer, offset, sampler);
}
return locker;
}
std::shared_ptr<OGLProgram> BuildShader(const Device& device, ShaderType shader_type,
u64 unique_identifier, const ShaderIR& ir,
bool hint_retrievable = false) {
LOG_INFO(Render_OpenGL, "{}", MakeShaderID(unique_identifier, shader_type));
const std::string glsl = DecompileShader(device, ir, shader_type);
OGLShader shader;
shader.Create(glsl.c_str(), GetGLShaderType(shader_type));
auto program = std::make_shared<OGLProgram>();
program->Create(true, hint_retrievable, shader.handle);
return program;
}
std::unordered_set<GLenum> GetSupportedFormats() {
GLint num_formats;
glGetIntegerv(GL_NUM_PROGRAM_BINARY_FORMATS, &num_formats);
std::vector<GLint> formats(num_formats);
glGetIntegerv(GL_PROGRAM_BINARY_FORMATS, formats.data());
std::unordered_set<GLenum> supported_formats;
for (const GLint format : formats) {
supported_formats.insert(static_cast<GLenum>(format));
}
return supported_formats;
}
} // Anonymous namespace
CachedShader::CachedShader(const u8* host_ptr, VAddr cpu_addr, std::size_t size_in_bytes,
std::shared_ptr<VideoCommon::Shader::ConstBufferLocker> locker,
ShaderEntries entries, std::shared_ptr<OGLProgram> program)
: RasterizerCacheObject{host_ptr}, locker{std::move(locker)}, entries{std::move(entries)},
cpu_addr{cpu_addr}, size_in_bytes{size_in_bytes}, program{std::move(program)} {}
CachedShader::~CachedShader() = default;
GLuint CachedShader::GetHandle() const {
if (!locker->IsConsistent()) {
std::abort();
}
return program->handle;
}
Shader CachedShader::CreateStageFromMemory(const ShaderParameters& params,
Maxwell::ShaderProgram program_type, ProgramCode code,
ProgramCode code_b) {
const auto shader_type = GetShaderType(program_type);
const std::size_t size_in_bytes = code.size() * sizeof(u64);
auto locker = std::make_shared<ConstBufferLocker>(shader_type, params.system.GPU().Maxwell3D());
const ShaderIR ir(code, STAGE_MAIN_OFFSET, COMPILER_SETTINGS, *locker);
// TODO(Rodrigo): Handle VertexA shaders
// std::optional<ShaderIR> ir_b;
// if (!code_b.empty()) {
// ir_b.emplace(code_b, STAGE_MAIN_OFFSET);
// }
auto program = BuildShader(params.device, shader_type, params.unique_identifier, ir);
ShaderDiskCacheEntry entry;
entry.type = shader_type;
entry.code = std::move(code);
entry.code_b = std::move(code_b);
entry.unique_identifier = params.unique_identifier;
entry.bound_buffer = locker->GetBoundBuffer();
entry.keys = locker->GetKeys();
entry.bound_samplers = locker->GetBoundSamplers();
entry.bindless_samplers = locker->GetBindlessSamplers();
params.disk_cache.SaveEntry(std::move(entry));
return std::shared_ptr<CachedShader>(new CachedShader(params.host_ptr, params.cpu_addr,
size_in_bytes, std::move(locker),
MakeEntries(ir), std::move(program)));
}
Shader CachedShader::CreateKernelFromMemory(const ShaderParameters& params, ProgramCode code) {
const std::size_t size_in_bytes = code.size() * sizeof(u64);
auto locker = std::make_shared<ConstBufferLocker>(Tegra::Engines::ShaderType::Compute,
params.system.GPU().KeplerCompute());
const ShaderIR ir(code, KERNEL_MAIN_OFFSET, COMPILER_SETTINGS, *locker);
auto program = BuildShader(params.device, ShaderType::Compute, params.unique_identifier, ir);
ShaderDiskCacheEntry entry;
entry.type = ShaderType::Compute;
entry.code = std::move(code);
entry.unique_identifier = params.unique_identifier;
entry.bound_buffer = locker->GetBoundBuffer();
entry.keys = locker->GetKeys();
entry.bound_samplers = locker->GetBoundSamplers();
entry.bindless_samplers = locker->GetBindlessSamplers();
params.disk_cache.SaveEntry(std::move(entry));
return std::shared_ptr<CachedShader>(new CachedShader(params.host_ptr, params.cpu_addr,
size_in_bytes, std::move(locker),
MakeEntries(ir), std::move(program)));
}
Shader CachedShader::CreateFromCache(const ShaderParameters& params,
const PrecompiledShader& precompiled_shader,
std::size_t size_in_bytes) {
return std::shared_ptr<CachedShader>(
new CachedShader(params.host_ptr, params.cpu_addr, size_in_bytes, precompiled_shader.locker,
precompiled_shader.entries, precompiled_shader.program));
}
ShaderCacheOpenGL::ShaderCacheOpenGL(RasterizerOpenGL& rasterizer, Core::System& system,
Core::Frontend::EmuWindow& emu_window, const Device& device)
: RasterizerCache{rasterizer}, system{system}, emu_window{emu_window}, device{device},
disk_cache{system} {}
void ShaderCacheOpenGL::LoadDiskCache(const std::atomic_bool& stop_loading,
const VideoCore::DiskResourceLoadCallback& callback) {
const std::optional transferable = disk_cache.LoadTransferable();
if (!transferable) {
return;
}
const std::vector gl_cache = disk_cache.LoadPrecompiled();
const auto supported_formats = GetSupportedFormats();
// Track if precompiled cache was altered during loading to know if we have to
// serialize the virtual precompiled cache file back to the hard drive
bool precompiled_cache_altered = false;
// Inform the frontend about shader build initialization
if (callback) {
callback(VideoCore::LoadCallbackStage::Build, 0, transferable->size());
}
std::mutex mutex;
std::size_t built_shaders = 0; // It doesn't have be atomic since it's used behind a mutex
std::atomic_bool gl_cache_failed = false;
const auto find_precompiled = [&gl_cache](u64 id) {
return std::find_if(gl_cache.begin(), gl_cache.end(),
[id](const auto& entry) { return entry.unique_identifier == id; });
};
const auto worker = [&](Core::Frontend::GraphicsContext* context, std::size_t begin,
std::size_t end) {
context->MakeCurrent();
SCOPE_EXIT({ return context->DoneCurrent(); });
for (std::size_t i = begin; i < end; ++i) {
if (stop_loading) {
return;
}
const auto& entry = (*transferable)[i];
const u64 unique_identifier = entry.unique_identifier;
const auto it = find_precompiled(unique_identifier);
const auto precompiled_entry = it != gl_cache.end() ? &*it : nullptr;
const bool is_compute = entry.type == ShaderType::Compute;
const u32 main_offset = is_compute ? KERNEL_MAIN_OFFSET : STAGE_MAIN_OFFSET;
auto locker = MakeLocker(entry);
const ShaderIR ir(entry.code, main_offset, COMPILER_SETTINGS, *locker);
std::shared_ptr<OGLProgram> program;
if (precompiled_entry) {
// If the shader is precompiled, attempt to load it with
program = GeneratePrecompiledProgram(entry, *precompiled_entry, supported_formats);
if (!program) {
gl_cache_failed = true;
}
}
if (!program) {
// Otherwise compile it from GLSL
program = BuildShader(device, entry.type, unique_identifier, ir, true);
}
PrecompiledShader shader;
shader.program = std::move(program);
shader.locker = std::move(locker);
shader.entries = MakeEntries(ir);
std::scoped_lock lock{mutex};
if (callback) {
callback(VideoCore::LoadCallbackStage::Build, ++built_shaders,
transferable->size());
}
runtime_cache.emplace(entry.unique_identifier, std::move(shader));
}
};
const auto num_workers{static_cast<std::size_t>(std::thread::hardware_concurrency() + 1ULL)};
const std::size_t bucket_size{transferable->size() / num_workers};
std::vector<std::unique_ptr<Core::Frontend::GraphicsContext>> contexts(num_workers);
std::vector<std::thread> threads(num_workers);
for (std::size_t i = 0; i < num_workers; ++i) {
const bool is_last_worker = i + 1 == num_workers;
const std::size_t start{bucket_size * i};
const std::size_t end{is_last_worker ? transferable->size() : start + bucket_size};
// On some platforms the shared context has to be created from the GUI thread
contexts[i] = emu_window.CreateSharedContext();
threads[i] = std::thread(worker, contexts[i].get(), start, end);
}
for (auto& thread : threads) {
thread.join();
}
if (gl_cache_failed) {
// Invalidate the precompiled cache if a shader dumped shader was rejected
disk_cache.InvalidatePrecompiled();
precompiled_cache_altered = true;
return;
}
if (stop_loading) {
return;
}
// TODO(Rodrigo): Do state tracking for transferable shaders and do a dummy draw
// before precompiling them
for (std::size_t i = 0; i < transferable->size(); ++i) {
const u64 id = (*transferable)[i].unique_identifier;
const auto it = find_precompiled(id);
if (it == gl_cache.end()) {
const GLuint program = runtime_cache.at(id).program->handle;
disk_cache.SavePrecompiled(id, program);
precompiled_cache_altered = true;
}
}
if (precompiled_cache_altered) {
disk_cache.SaveVirtualPrecompiledFile();
}
}
std::shared_ptr<OGLProgram> ShaderCacheOpenGL::GeneratePrecompiledProgram(
const ShaderDiskCacheEntry& entry, const ShaderDiskCachePrecompiled& precompiled_entry,
const std::unordered_set<GLenum>& supported_formats) {
if (supported_formats.find(precompiled_entry.binary_format) == supported_formats.end()) {
LOG_INFO(Render_OpenGL, "Precompiled cache entry with unsupported format, removing");
return {};
}
auto program = std::make_shared<OGLProgram>();
program->handle = glCreateProgram();
glProgramParameteri(program->handle, GL_PROGRAM_SEPARABLE, GL_TRUE);
glProgramBinary(program->handle, precompiled_entry.binary_format,
precompiled_entry.binary.data(),
static_cast<GLsizei>(precompiled_entry.binary.size()));
GLint link_status;
glGetProgramiv(program->handle, GL_LINK_STATUS, &link_status);
if (link_status == GL_FALSE) {
LOG_INFO(Render_OpenGL, "Precompiled cache rejected by the driver, removing");
return {};
}
return program;
}
Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) {
if (!system.GPU().Maxwell3D().dirty.flags[Dirty::Shaders]) {
return last_shaders[static_cast<std::size_t>(program)];
}
auto& memory_manager{system.GPU().MemoryManager()};
const GPUVAddr address{GetShaderAddress(system, program)};
// Look up shader in the cache based on address
const auto host_ptr{memory_manager.GetPointer(address)};
Shader shader{TryGet(host_ptr)};
if (shader) {
return last_shaders[static_cast<std::size_t>(program)] = shader;
}
// No shader found - create a new one
ProgramCode code{GetShaderCode(memory_manager, address, host_ptr)};
ProgramCode code_b;
if (program == Maxwell::ShaderProgram::VertexA) {
const GPUVAddr address_b{GetShaderAddress(system, Maxwell::ShaderProgram::VertexB)};
code_b = GetShaderCode(memory_manager, address_b, memory_manager.GetPointer(address_b));
}
const auto unique_identifier = GetUniqueIdentifier(
GetShaderType(program), program == Maxwell::ShaderProgram::VertexA, code, code_b);
const auto cpu_addr{*memory_manager.GpuToCpuAddress(address)};
const ShaderParameters params{system, disk_cache, device,
cpu_addr, host_ptr, unique_identifier};
const auto found = runtime_cache.find(unique_identifier);
if (found == runtime_cache.end()) {
shader = CachedShader::CreateStageFromMemory(params, program, std::move(code),
std::move(code_b));
} else {
const std::size_t size_in_bytes = code.size() * sizeof(u64);
shader = CachedShader::CreateFromCache(params, found->second, size_in_bytes);
}
Register(shader);
return last_shaders[static_cast<std::size_t>(program)] = shader;
}
Shader ShaderCacheOpenGL::GetComputeKernel(GPUVAddr code_addr) {
auto& memory_manager{system.GPU().MemoryManager()};
const auto host_ptr{memory_manager.GetPointer(code_addr)};
auto kernel = TryGet(host_ptr);
if (kernel) {
return kernel;
}
// No kernel found, create a new one
auto code{GetShaderCode(memory_manager, code_addr, host_ptr)};
const auto unique_identifier{GetUniqueIdentifier(ShaderType::Compute, false, code)};
const auto cpu_addr{*memory_manager.GpuToCpuAddress(code_addr)};
const ShaderParameters params{system, disk_cache, device,
cpu_addr, host_ptr, unique_identifier};
const auto found = runtime_cache.find(unique_identifier);
if (found == runtime_cache.end()) {
kernel = CachedShader::CreateKernelFromMemory(params, std::move(code));
} else {
const std::size_t size_in_bytes = code.size() * sizeof(u64);
kernel = CachedShader::CreateFromCache(params, found->second, size_in_bytes);
}
Register(kernel);
return kernel;
}
} // namespace OpenGL
|