Flutter Engine
 
Loading...
Searching...
No Matches
compute_unittests.cc
Go to the documentation of this file.
1// Copyright 2013 The Flutter Authors. All rights reserved.
2// Use of this source code is governed by a BSD-style license that can be
3// found in the LICENSE file.
4
5#include "flutter/fml/synchronization/waitable_event.h"
6#include "flutter/testing/testing.h"
7#include "gmock/gmock.h"
8#include "impeller/core/host_buffer.h"
9#include "impeller/fixtures/sample.comp.h"
10#include "impeller/fixtures/stage1.comp.h"
11#include "impeller/fixtures/stage2.comp.h"
12#include "impeller/playground/compute_playground_test.h"
13#include "impeller/renderer/command_buffer.h"
14#include "impeller/renderer/compute_pipeline_builder.h"
15#include "impeller/renderer/pipeline_library.h"
16#include "impeller/renderer/prefix_sum_test.comp.h"
17#include "impeller/renderer/threadgroup_sizing_test.comp.h"
18
19namespace impeller {
20namespace testing {
21using ComputeTest = ComputePlaygroundTest;
22INSTANTIATE_COMPUTE_SUITE(ComputeTest);
23
24TEST_P(ComputeTest, CapabilitiesReportSupport) {
25 auto context = GetContext();
26 ASSERT_TRUE(context);
27 ASSERT_TRUE(context->GetCapabilities()->SupportsCompute());
28}
29
30TEST_P(ComputeTest, CanCreateComputePass) {
31 using CS = SampleComputeShader;
32 auto context = GetContext();
33 auto host_buffer = HostBuffer::Create(context->GetResourceAllocator(),
34 context->GetIdleWaiter());
35 ASSERT_TRUE(context);
36 ASSERT_TRUE(context->GetCapabilities()->SupportsCompute());
37
38 using SamplePipelineBuilder = ComputePipelineBuilder<CS>;
39 auto pipeline_desc =
40 SamplePipelineBuilder::MakeDefaultPipelineDescriptor(*context);
41 ASSERT_TRUE(pipeline_desc.has_value());
42 auto compute_pipeline =
43 context->GetPipelineLibrary()->GetPipeline(pipeline_desc).Get();
44 ASSERT_TRUE(compute_pipeline);
45
46 auto cmd_buffer = context->CreateCommandBuffer();
47 auto pass = cmd_buffer->CreateComputePass();
48 ASSERT_TRUE(pass && pass->IsValid());
49
50 static constexpr size_t kCount = 5;
51
52 pass->SetPipeline(compute_pipeline);
53
54 CS::Info info{.count = kCount};
55 CS::Input0<kCount> input_0;
56 CS::Input1<kCount> input_1;
57 for (size_t i = 0; i < kCount; i++) {
58 input_0.elements[i] = Vector4(2.0 + i, 3.0 + i, 4.0 + i, 5.0 * i);
59 input_1.elements[i] = Vector4(6.0, 7.0, 8.0, 9.0);
60 }
61
62 input_0.fixed_array[1] = IPoint32(2, 2);
63 input_1.fixed_array[0] = UintPoint32(3, 3);
64 input_0.some_int = 5;
65 input_1.some_struct = CS::SomeStruct{.vf = Point(3, 4), .i = 42};
66
67 auto output_buffer = CreateHostVisibleDeviceBuffer<CS::Output<kCount>>(
68 context, "Output Buffer");
69
70 CS::BindInfo(*pass, host_buffer->EmplaceUniform(info));
71 CS::BindInput0(*pass, host_buffer->EmplaceStorageBuffer(input_0));
72 CS::BindInput1(*pass, host_buffer->EmplaceStorageBuffer(input_1));
73 CS::BindOutput(*pass, DeviceBuffer::AsBufferView(output_buffer));
74
75 ASSERT_TRUE(pass->Compute(ISize(kCount, 1)).ok());
76 ASSERT_TRUE(pass->EncodeCommands());
77
78 fml::AutoResetWaitableEvent latch;
79 ASSERT_TRUE(
80 context->GetCommandQueue()
81 ->Submit(
82 {cmd_buffer},
83 [&latch, output_buffer, &input_0,
84 &input_1](CommandBuffer::Status status) {
85 EXPECT_EQ(status, CommandBuffer::Status::kCompleted);
86
87 auto view = DeviceBuffer::AsBufferView(output_buffer);
88 EXPECT_EQ(view.GetRange().length, sizeof(CS::Output<kCount>));
89
90 CS::Output<kCount>* output =
91 reinterpret_cast<CS::Output<kCount>*>(
92 output_buffer->OnGetContents());
93 EXPECT_TRUE(output);
94 for (size_t i = 0; i < kCount; i++) {
95 Vector4 vector = output->elements[i];
96 Vector4 computed = input_0.elements[i] * input_1.elements[i];
97 EXPECT_EQ(vector,
98 Vector4(computed.x + 2 + input_1.some_struct.i,
99 computed.y + 3 + input_1.some_struct.vf.x,
100 computed.z + 5 + input_1.some_struct.vf.y,
101 computed.w));
102 }
103 latch.Signal();
104 })
105 .ok());
106
107 latch.Wait();
108}
109
110TEST_P(ComputeTest, CanComputePrefixSum) {
111 using CS = PrefixSumTestComputeShader;
112 auto context = GetContext();
113 auto host_buffer = HostBuffer::Create(context->GetResourceAllocator(),
114 context->GetIdleWaiter());
115 ASSERT_TRUE(context);
116 ASSERT_TRUE(context->GetCapabilities()->SupportsCompute());
117
118 using SamplePipelineBuilder = ComputePipelineBuilder<CS>;
119 auto pipeline_desc =
120 SamplePipelineBuilder::MakeDefaultPipelineDescriptor(*context);
121 ASSERT_TRUE(pipeline_desc.has_value());
122 auto compute_pipeline =
123 context->GetPipelineLibrary()->GetPipeline(pipeline_desc).Get();
124 ASSERT_TRUE(compute_pipeline);
125
126 auto cmd_buffer = context->CreateCommandBuffer();
127 auto pass = cmd_buffer->CreateComputePass();
128 ASSERT_TRUE(pass && pass->IsValid());
129
130 static constexpr size_t kCount = 5;
131
132 pass->SetPipeline(compute_pipeline);
133
134 CS::InputData<kCount> input_data;
135 input_data.count = kCount;
136 for (size_t i = 0; i < kCount; i++) {
137 input_data.data[i] = 1 + i;
138 }
139
140 auto output_buffer = CreateHostVisibleDeviceBuffer<CS::OutputData<kCount>>(
141 context, "Output Buffer");
142
143 CS::BindInputData(*pass, host_buffer->EmplaceStorageBuffer(input_data));
144 CS::BindOutputData(*pass, DeviceBuffer::AsBufferView(output_buffer));
145
146 ASSERT_TRUE(pass->Compute(ISize(kCount, 1)).ok());
147 ASSERT_TRUE(pass->EncodeCommands());
148
149 fml::AutoResetWaitableEvent latch;
150 ASSERT_TRUE(
151 context->GetCommandQueue()
152 ->Submit({cmd_buffer},
153 [&latch, output_buffer](CommandBuffer::Status status) {
154 EXPECT_EQ(status, CommandBuffer::Status::kCompleted);
155
156 auto view = DeviceBuffer::AsBufferView(output_buffer);
157 EXPECT_EQ(view.GetRange().length,
158 sizeof(CS::OutputData<kCount>));
159
160 CS::OutputData<kCount>* output =
161 reinterpret_cast<CS::OutputData<kCount>*>(
162 output_buffer->OnGetContents());
163 EXPECT_TRUE(output);
164
165 constexpr uint32_t expected[kCount] = {1, 3, 6, 10, 15};
166 for (size_t i = 0; i < kCount; i++) {
167 auto computed_sum = output->data[i];
168 EXPECT_EQ(computed_sum, expected[i]);
169 }
170 latch.Signal();
171 })
172 .ok());
173
174 latch.Wait();
175}
176
177TEST_P(ComputeTest, 1DThreadgroupSizingIsCorrect) {
178 using CS = ThreadgroupSizingTestComputeShader;
179 auto context = GetContext();
180 ASSERT_TRUE(context);
181 ASSERT_TRUE(context->GetCapabilities()->SupportsCompute());
182
183 using SamplePipelineBuilder = ComputePipelineBuilder<CS>;
184 auto pipeline_desc =
185 SamplePipelineBuilder::MakeDefaultPipelineDescriptor(*context);
186 ASSERT_TRUE(pipeline_desc.has_value());
187 auto compute_pipeline =
188 context->GetPipelineLibrary()->GetPipeline(pipeline_desc).Get();
189 ASSERT_TRUE(compute_pipeline);
190
191 auto cmd_buffer = context->CreateCommandBuffer();
192 auto pass = cmd_buffer->CreateComputePass();
193 ASSERT_TRUE(pass && pass->IsValid());
194
195 static constexpr size_t kCount = 2048;
196
197 pass->SetPipeline(compute_pipeline);
198
199 auto output_buffer = CreateHostVisibleDeviceBuffer<CS::OutputData<kCount>>(
200 context, "Output Buffer");
201
202 CS::BindOutputData(*pass, DeviceBuffer::AsBufferView(output_buffer));
203
204 ASSERT_TRUE(pass->Compute(ISize(kCount, 1)).ok());
205 ASSERT_TRUE(pass->EncodeCommands());
206
207 fml::AutoResetWaitableEvent latch;
208 ASSERT_TRUE(
209 context->GetCommandQueue()
210 ->Submit({cmd_buffer},
211 [&latch, output_buffer](CommandBuffer::Status status) {
212 EXPECT_EQ(status, CommandBuffer::Status::kCompleted);
213
214 auto view = DeviceBuffer::AsBufferView(output_buffer);
215 EXPECT_EQ(view.GetRange().length,
216 sizeof(CS::OutputData<kCount>));
217
218 CS::OutputData<kCount>* output =
219 reinterpret_cast<CS::OutputData<kCount>*>(
220 output_buffer->OnGetContents());
221 EXPECT_TRUE(output);
222 EXPECT_EQ(output->data[kCount - 1], kCount - 1);
223 latch.Signal();
224 })
225 .ok());
226
227 latch.Wait();
228}
229
230TEST_P(ComputeTest, CanComputePrefixSumLargeInteractive) {
231 using CS = PrefixSumTestComputeShader;
232
233 auto context = GetContext();
234 auto host_buffer = HostBuffer::Create(context->GetResourceAllocator(),
235 context->GetIdleWaiter());
236
237 ASSERT_TRUE(context);
238 ASSERT_TRUE(context->GetCapabilities()->SupportsCompute());
239
240 auto callback = [&](RenderPass& render_pass) -> bool {
241 using SamplePipelineBuilder = ComputePipelineBuilder<CS>;
242 auto pipeline_desc =
243 SamplePipelineBuilder::MakeDefaultPipelineDescriptor(*context);
244 auto compute_pipeline =
245 context->GetPipelineLibrary()->GetPipeline(pipeline_desc).Get();
246
247 auto cmd_buffer = context->CreateCommandBuffer();
248 auto pass = cmd_buffer->CreateComputePass();
249
250 static constexpr size_t kCount = 1023;
251
252 pass->SetPipeline(compute_pipeline);
253
254 CS::InputData<kCount> input_data;
255 input_data.count = kCount;
256 for (size_t i = 0; i < kCount; i++) {
257 input_data.data[i] = 1 + i;
258 }
259
260 auto output_buffer = CreateHostVisibleDeviceBuffer<CS::OutputData<kCount>>(
261 context, "Output Buffer");
262
263 CS::BindInputData(*pass, host_buffer->EmplaceStorageBuffer(input_data));
264 CS::BindOutputData(*pass, DeviceBuffer::AsBufferView(output_buffer));
265
266 pass->Compute(ISize(kCount, 1));
267 pass->EncodeCommands();
268 host_buffer->Reset();
269 return context->GetCommandQueue()->Submit({cmd_buffer}).ok();
270 };
271 ASSERT_TRUE(OpenPlaygroundHere(callback));
272}
273
274TEST_P(ComputeTest, MultiStageInputAndOutput) {
275 using CS1 = Stage1ComputeShader;
276 using Stage1PipelineBuilder = ComputePipelineBuilder<CS1>;
277 using CS2 = Stage2ComputeShader;
278 using Stage2PipelineBuilder = ComputePipelineBuilder<CS2>;
279
280 auto context = GetContext();
281 auto host_buffer = HostBuffer::Create(context->GetResourceAllocator(),
282 context->GetIdleWaiter());
283 ASSERT_TRUE(context);
284 ASSERT_TRUE(context->GetCapabilities()->SupportsCompute());
285
286 auto pipeline_desc_1 =
287 Stage1PipelineBuilder::MakeDefaultPipelineDescriptor(*context);
288 ASSERT_TRUE(pipeline_desc_1.has_value());
289 auto compute_pipeline_1 =
290 context->GetPipelineLibrary()->GetPipeline(pipeline_desc_1).Get();
291 ASSERT_TRUE(compute_pipeline_1);
292
293 auto pipeline_desc_2 =
294 Stage2PipelineBuilder::MakeDefaultPipelineDescriptor(*context);
295 ASSERT_TRUE(pipeline_desc_2.has_value());
296 auto compute_pipeline_2 =
297 context->GetPipelineLibrary()->GetPipeline(pipeline_desc_2).Get();
298 ASSERT_TRUE(compute_pipeline_2);
299
300 auto cmd_buffer = context->CreateCommandBuffer();
301 auto pass = cmd_buffer->CreateComputePass();
302 ASSERT_TRUE(pass && pass->IsValid());
303
304 static constexpr size_t kCount1 = 5;
305 static constexpr size_t kCount2 = kCount1 * 2;
306
307 CS1::Input<kCount1> input_1;
308 input_1.count = kCount1;
309 for (size_t i = 0; i < kCount1; i++) {
310 input_1.elements[i] = i;
311 }
312
313 CS2::Input<kCount2> input_2;
314 input_2.count = kCount2;
315 for (size_t i = 0; i < kCount2; i++) {
316 input_2.elements[i] = i;
317 }
318
319 auto output_buffer_1 = CreateHostVisibleDeviceBuffer<CS1::Output<kCount2>>(
320 context, "Output Buffer Stage 1");
321 auto output_buffer_2 = CreateHostVisibleDeviceBuffer<CS2::Output<kCount2>>(
322 context, "Output Buffer Stage 2");
323
324 {
325 pass->SetPipeline(compute_pipeline_1);
326
327 CS1::BindInput(*pass, host_buffer->EmplaceStorageBuffer(input_1));
328 CS1::BindOutput(*pass, DeviceBuffer::AsBufferView(output_buffer_1));
329
330 ASSERT_TRUE(pass->Compute(ISize(512, 1)).ok());
331 pass->AddBufferMemoryBarrier();
332 }
333
334 {
335 pass->SetPipeline(compute_pipeline_2);
336
337 CS1::BindInput(*pass, DeviceBuffer::AsBufferView(output_buffer_1));
338 CS2::BindOutput(*pass, DeviceBuffer::AsBufferView(output_buffer_2));
339 ASSERT_TRUE(pass->Compute(ISize(512, 1)).ok());
340 }
341
342 ASSERT_TRUE(pass->EncodeCommands());
343
344 fml::AutoResetWaitableEvent latch;
345 ASSERT_TRUE(
346 context->GetCommandQueue()
347 ->Submit({cmd_buffer},
348 [&latch, &output_buffer_1,
349 &output_buffer_2](CommandBuffer::Status status) {
350 EXPECT_EQ(status, CommandBuffer::Status::kCompleted);
351
352 CS1::Output<kCount2>* output_1 =
353 reinterpret_cast<CS1::Output<kCount2>*>(
354 output_buffer_1->OnGetContents());
355 EXPECT_TRUE(output_1);
356 EXPECT_EQ(output_1->count, 10u);
357 EXPECT_THAT(
358 output_1->elements,
359 ::testing::ElementsAre(0, 0, 2, 3, 4, 6, 6, 9, 8, 12));
360
361 CS2::Output<kCount2>* output_2 =
362 reinterpret_cast<CS2::Output<kCount2>*>(
363 output_buffer_2->OnGetContents());
364 EXPECT_TRUE(output_2);
365 EXPECT_EQ(output_2->count, 10u);
366 EXPECT_THAT(output_2->elements,
367 ::testing::ElementsAre(0, 0, 4, 6, 8, 12, 12,
368 18, 16, 24));
369
370 latch.Signal();
371 })
372 .ok());
373
374 latch.Wait();
375}
376
377TEST_P(ComputeTest, CanCompute1DimensionalData) {
378 using CS = SampleComputeShader;
379 auto context = GetContext();
380 auto host_buffer = HostBuffer::Create(context->GetResourceAllocator(),
381 context->GetIdleWaiter());
382 ASSERT_TRUE(context);
383 ASSERT_TRUE(context->GetCapabilities()->SupportsCompute());
384
385 using SamplePipelineBuilder = ComputePipelineBuilder<CS>;
386 auto pipeline_desc =
387 SamplePipelineBuilder::MakeDefaultPipelineDescriptor(*context);
388 ASSERT_TRUE(pipeline_desc.has_value());
389 auto compute_pipeline =
390 context->GetPipelineLibrary()->GetPipeline(pipeline_desc).Get();
391 ASSERT_TRUE(compute_pipeline);
392
393 auto cmd_buffer = context->CreateCommandBuffer();
394 auto pass = cmd_buffer->CreateComputePass();
395 ASSERT_TRUE(pass && pass->IsValid());
396
397 static constexpr size_t kCount = 5;
398
399 pass->SetPipeline(compute_pipeline);
400
401 CS::Info info{.count = kCount};
402 CS::Input0<kCount> input_0;
403 CS::Input1<kCount> input_1;
404 for (size_t i = 0; i < kCount; i++) {
405 input_0.elements[i] = Vector4(2.0 + i, 3.0 + i, 4.0 + i, 5.0 * i);
406 input_1.elements[i] = Vector4(6.0, 7.0, 8.0, 9.0);
407 }
408
409 input_0.fixed_array[1] = IPoint32(2, 2);
410 input_1.fixed_array[0] = UintPoint32(3, 3);
411 input_0.some_int = 5;
412 input_1.some_struct = CS::SomeStruct{.vf = Point(3, 4), .i = 42};
413
414 auto output_buffer = CreateHostVisibleDeviceBuffer<CS::Output<kCount>>(
415 context, "Output Buffer");
416
417 CS::BindInfo(*pass, host_buffer->EmplaceUniform(info));
418 CS::BindInput0(*pass, host_buffer->EmplaceStorageBuffer(input_0));
419 CS::BindInput1(*pass, host_buffer->EmplaceStorageBuffer(input_1));
420 CS::BindOutput(*pass, DeviceBuffer::AsBufferView(output_buffer));
421
422 ASSERT_TRUE(pass->Compute(ISize(kCount, 1)).ok());
423 ASSERT_TRUE(pass->EncodeCommands());
424
425 fml::AutoResetWaitableEvent latch;
426 ASSERT_TRUE(
427 context->GetCommandQueue()
428 ->Submit(
429 {cmd_buffer},
430 [&latch, output_buffer, &input_0,
431 &input_1](CommandBuffer::Status status) {
432 EXPECT_EQ(status, CommandBuffer::Status::kCompleted);
433
434 auto view = DeviceBuffer::AsBufferView(output_buffer);
435 EXPECT_EQ(view.GetRange().length, sizeof(CS::Output<kCount>));
436
437 CS::Output<kCount>* output =
438 reinterpret_cast<CS::Output<kCount>*>(
439 output_buffer->OnGetContents());
440 EXPECT_TRUE(output);
441 for (size_t i = 0; i < kCount; i++) {
442 Vector4 vector = output->elements[i];
443 Vector4 computed = input_0.elements[i] * input_1.elements[i];
444 EXPECT_EQ(vector,
445 Vector4(computed.x + 2 + input_1.some_struct.i,
446 computed.y + 3 + input_1.some_struct.vf.x,
447 computed.z + 5 + input_1.some_struct.vf.y,
448 computed.w));
449 }
450 latch.Signal();
451 })
452 .ok());
453
454 latch.Wait();
455}
456
457TEST_P(ComputeTest, ReturnsEarlyWhenAnyGridDimensionIsZero) {
458 using CS = SampleComputeShader;
459 auto context = GetContext();
460 auto host_buffer = HostBuffer::Create(context->GetResourceAllocator(),
461 context->GetIdleWaiter());
462 ASSERT_TRUE(context);
463 ASSERT_TRUE(context->GetCapabilities()->SupportsCompute());
464
465 using SamplePipelineBuilder = ComputePipelineBuilder<CS>;
466 auto pipeline_desc =
467 SamplePipelineBuilder::MakeDefaultPipelineDescriptor(*context);
468 ASSERT_TRUE(pipeline_desc.has_value());
469 auto compute_pipeline =
470 context->GetPipelineLibrary()->GetPipeline(pipeline_desc).Get();
471 ASSERT_TRUE(compute_pipeline);
472
473 auto cmd_buffer = context->CreateCommandBuffer();
474 auto pass = cmd_buffer->CreateComputePass();
475 ASSERT_TRUE(pass && pass->IsValid());
476
477 static constexpr size_t kCount = 5;
478
479 pass->SetPipeline(compute_pipeline);
480
481 CS::Info info{.count = kCount};
482 CS::Input0<kCount> input_0;
483 CS::Input1<kCount> input_1;
484 for (size_t i = 0; i < kCount; i++) {
485 input_0.elements[i] = Vector4(2.0 + i, 3.0 + i, 4.0 + i, 5.0 * i);
486 input_1.elements[i] = Vector4(6.0, 7.0, 8.0, 9.0);
487 }
488
489 input_0.fixed_array[1] = IPoint32(2, 2);
490 input_1.fixed_array[0] = UintPoint32(3, 3);
491 input_0.some_int = 5;
492 input_1.some_struct = CS::SomeStruct{.vf = Point(3, 4), .i = 42};
493
494 auto output_buffer = CreateHostVisibleDeviceBuffer<CS::Output<kCount>>(
495 context, "Output Buffer");
496
497 CS::BindInfo(*pass, host_buffer->EmplaceUniform(info));
498 CS::BindInput0(*pass, host_buffer->EmplaceStorageBuffer(input_0));
499 CS::BindInput1(*pass, host_buffer->EmplaceStorageBuffer(input_1));
500 CS::BindOutput(*pass, DeviceBuffer::AsBufferView(output_buffer));
501
502 // Intentionally making the grid size zero in one dimension. No GPU will
503 // tolerate this.
504 EXPECT_FALSE(pass->Compute(ISize(0, 1)).ok());
505 pass->EncodeCommands();
506}
507
508} // namespace testing
509} // namespace impeller
impeller::DeviceBuffer::AsBufferView
static BufferView AsBufferView(std::shared_ptr< DeviceBuffer > buffer)
Create a buffer view of this entire buffer.
Definition device_buffer.cc:18
impeller::HostBuffer::Create
static std::shared_ptr< HostBuffer > Create(const std::shared_ptr< Allocator > &allocator, const std::shared_ptr< const IdleWaiter > &idle_waiter, size_t minimum_uniform_alignment)
Definition host_buffer.cc:21
impeller::RenderPass
Render passes encode render commands directed as one specific render target into an underlying comman...
Definition render_pass.h:30
compute_pipeline_builder.h
compute_playground_test.h
INSTANTIATE_COMPUTE_SUITE
#define INSTANTIATE_COMPUTE_SUITE(playground)
Definition compute_playground_test.h:56
i
int i
Definition fl_socket_accessible.cc:18
callback
FlutterDesktopBinaryReply callback
Definition flutter_windows_view_unittests.cc:52
host_buffer.h
command_buffer.h
impeller::testing::TEST_P
TEST_P(AiksTest, DrawAtlasNoColor)
Definition aiks_dl_atlas_unittests.cc:59
impeller::IPoint32
TPoint< int32_t > IPoint32
Definition point.h:329
impeller::UintPoint32
TPoint< uint32_t > UintPoint32
Definition point.h:330
pipeline_library.h
impeller::ComputePipelineBuilder
An optional (but highly recommended) utility for creating pipelines from reflected shader information...
Definition compute_pipeline_builder.h:26
waitable_event.h