Internal change.

PiperOrigin-RevId: 493973435
This commit is contained in:
Nikolay Chirkov 2022-12-08 12:59:46 -08:00 committed by Copybara-Service
parent a641ea12e1
commit 0fbaa8dc8a

View File

@ -0,0 +1,143 @@
#if !defined(MEDIAPIPE_NO_JNI) && \
(__ANDROID_API__ >= 26 || \
defined(__ANDROID_UNAVAILABLE_SYMBOLS_ARE_WEAK__))
#include <android/hardware_buffer.h>
#include <cstdint>
#include "mediapipe/framework/formats/tensor.h"
#include "mediapipe/framework/formats/tensor_data_types.h"
#include "mediapipe/gpu/gpu_test_base.h"
#include "mediapipe/gpu/shader_util.h"
#include "tensorflow/lite/delegates/gpu/gl/gl_call.h"
#include "testing/base/public/gunit.h"
// The test creates OpenGL ES buffer, fills the buffer with incrementing values
// 0.0, 0.1, 0.2 etc. with the compute shader on GPU.
// Then the test requests the CPU view and compares the values.
// Float32 and Float16 tests are there.
namespace {
using mediapipe::Float16;
using mediapipe::Tensor;
MATCHER_P(NearWithPrecision, precision, "") {
return std::abs(std::get<0>(arg) - std::get<1>(arg)) < precision;
}
#if MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_31
// Utility function to fill the GPU buffer.
void FillGpuBuffer(GLuint name, std::size_t size,
const Tensor::ElementType fmt) {
std::string shader_source;
if (fmt == Tensor::ElementType::kFloat32) {
shader_source = R"( #version 310 es
precision highp float;
layout(local_size_x = 1, local_size_y = 1) in;
layout(std430, binding = 0) buffer Output {float elements[];} output_data;
void main() {
uint v = gl_GlobalInvocationID.x * 2u;
output_data.elements[v] = float(v) / 10.0;
output_data.elements[v + 1u] = float(v + 1u) / 10.0;
})";
} else {
shader_source = R"( #version 310 es
precision highp float;
layout(local_size_x = 1, local_size_y = 1) in;
layout(std430, binding = 0) buffer Output {float elements[];} output_data;
void main() {
uint v = gl_GlobalInvocationID.x;
uint tmp = packHalf2x16(vec2((float(v)* 2.0 + 0.0) / 10.0,
(float(v) * 2.0 + 1.0) / 10.0));
output_data.elements[v] = uintBitsToFloat(tmp);
})";
}
GLuint shader;
MP_ASSERT_OK(TFLITE_GPU_CALL_GL(glCreateShader, &shader, GL_COMPUTE_SHADER));
const GLchar* sources[] = {shader_source.c_str()};
MP_ASSERT_OK(TFLITE_GPU_CALL_GL(glShaderSource, shader, 1, sources, nullptr));
MP_ASSERT_OK(TFLITE_GPU_CALL_GL(glCompileShader, shader));
GLint is_compiled = 0;
MP_ASSERT_OK(TFLITE_GPU_CALL_GL(glGetShaderiv, shader, GL_COMPILE_STATUS,
&is_compiled));
if (is_compiled == GL_FALSE) {
GLint max_length = 0;
MP_ASSERT_OK(TFLITE_GPU_CALL_GL(glGetShaderiv, shader, GL_INFO_LOG_LENGTH,
&max_length));
std::vector<GLchar> error_log(max_length);
glGetShaderInfoLog(shader, max_length, &max_length, error_log.data());
glDeleteShader(shader);
FAIL() << error_log.data();
return;
}
GLuint to_buffer_program;
MP_ASSERT_OK(TFLITE_GPU_CALL_GL(glCreateProgram, &to_buffer_program));
MP_ASSERT_OK(TFLITE_GPU_CALL_GL(glAttachShader, to_buffer_program, shader));
MP_ASSERT_OK(TFLITE_GPU_CALL_GL(glDeleteShader, shader));
MP_ASSERT_OK(TFLITE_GPU_CALL_GL(glLinkProgram, to_buffer_program));
MP_ASSERT_OK(
TFLITE_GPU_CALL_GL(glBindBufferBase, GL_SHADER_STORAGE_BUFFER, 0, name));
MP_ASSERT_OK(TFLITE_GPU_CALL_GL(glUseProgram, to_buffer_program));
MP_ASSERT_OK(TFLITE_GPU_CALL_GL(glDispatchCompute, size / 2, 1, 1));
MP_ASSERT_OK(TFLITE_GPU_CALL_GL(glBindBuffer, GL_SHADER_STORAGE_BUFFER, 0));
MP_ASSERT_OK(TFLITE_GPU_CALL_GL(glDeleteProgram, to_buffer_program));
}
class TensorAhwbGpuTest : public mediapipe::GpuTestBase {
public:
};
TEST_F(TensorAhwbGpuTest, TestGpuToCpuFloat32) {
Tensor::SetPreferredStorageType(Tensor::StorageType::kAhwb);
constexpr size_t num_elements = 20;
Tensor tensor{Tensor::ElementType::kFloat32, Tensor::Shape({num_elements})};
RunInGlContext([&tensor] {
auto ssbo_view = tensor.GetOpenGlBufferWriteView();
auto ssbo_name = ssbo_view.name();
EXPECT_GT(ssbo_name, 0);
FillGpuBuffer(ssbo_name, tensor.shape().num_elements(),
tensor.element_type());
});
auto ptr = tensor.GetCpuReadView().buffer<float>();
EXPECT_NE(ptr, nullptr);
std::vector<float> reference;
reference.resize(num_elements);
for (int i = 0; i < num_elements; i++) {
reference[i] = static_cast<float>(i) / 10.0f;
}
EXPECT_THAT(absl::Span<const float>(ptr, num_elements),
testing::Pointwise(testing::FloatEq(), reference));
}
TEST_F(TensorAhwbGpuTest, TestGpuToCpuFloat16) {
Tensor::SetPreferredStorageType(Tensor::StorageType::kAhwb);
constexpr size_t num_elements = 20;
Tensor tensor{Tensor::ElementType::kFloat16, Tensor::Shape({num_elements})};
RunInGlContext([&tensor] {
auto ssbo_view = tensor.GetOpenGlBufferWriteView();
auto ssbo_name = ssbo_view.name();
EXPECT_GT(ssbo_name, 0);
FillGpuBuffer(ssbo_name, tensor.shape().num_elements(),
tensor.element_type());
});
auto ptr = tensor.GetCpuReadView().buffer<Float16>();
EXPECT_NE(ptr, nullptr);
std::vector<Float16> reference;
reference.resize(num_elements);
for (int i = 0; i < num_elements; i++) {
reference[i] = static_cast<float>(i) / 10.0f;
}
// Precision is set to a reasonable value for Float16.
EXPECT_THAT(absl::Span<const Float16>(ptr, num_elements),
testing::Pointwise(NearWithPrecision(0.001), reference));
}
#endif // MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_31
} // namespace
#endif // !defined(MEDIAPIPE_NO_JNI) && (__ANDROID_API__ >= 26 ||
// defined(__ANDROID_UNAVAILABLE_SYMBOLS_ARE_WEAK__))