cassdlcm/mediapipe
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
2044e45b1a
mediapipe/mediapipe2/framework/formats/tensor.cc
AK391 2044e45b1a examples
2021-06-10 23:01:19 +00:00

486 lines
19 KiB
C++
Raw Blame History

// Copyright 2020 The MediaPipe Authors.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#include "mediapipe/framework/formats/tensor.h"
#include <cstdint>
#include <utility>
#include "absl/synchronization/mutex.h"
#include "mediapipe/framework/port.h"
#include "mediapipe/framework/port/logging.h"
#if MEDIAPIPE_METAL_ENABLED
#include <mach/mach_init.h>
#include <mach/vm_map.h>
#else
#include <cstdlib>
#endif // MEDIAPIPE_METAL_ENABLED
namespace mediapipe {
// Zero and negative values are not checked here.
bool IsPowerOfTwo(int v) { return (v & (v - 1)) == 0; }
int BhwcBatchFromShape(const Tensor::Shape& shape) {
LOG_IF(FATAL, shape.dims.empty())
<< "Tensor::Shape must be non-empty to retrieve a named dimension";
return shape.dims[0];
}
int BhwcHeightFromShape(const Tensor::Shape& shape) {
LOG_IF(FATAL, shape.dims.empty())
<< "Tensor::Shape must be non-empty to retrieve a named dimension";
return shape.dims.size() < 4 ? 1 : shape.dims[shape.dims.size() - 3];
}
int BhwcWidthFromShape(const Tensor::Shape& shape) {
LOG_IF(FATAL, shape.dims.empty())
<< "Tensor::Shape must be non-empty to retrieve a named dimension";
return shape.dims.size() < 3 ? 1 : shape.dims[shape.dims.size() - 2];
}
int BhwcDepthFromShape(const Tensor::Shape& shape) {
LOG_IF(FATAL, shape.dims.empty())
<< "Tensor::Shape must be non-empty to retrieve a named dimension";
return shape.dims.size() < 2 ? 1 : shape.dims[shape.dims.size() - 1];
}
// TODO: Match channels count and padding for Texture2D:
// 1) support 1/2/4 channesl texture for 1/2/3-4 depth.
// 2) Allocate cpu_buffer_ with padded amount of memory
// 3) pad/"unpad" the bitmap after transfer CPU <-> GPU
#if MEDIAPIPE_METAL_ENABLED
namespace {
// MTLBuffer can use existing properly aligned and allocated CPU memory.
size_t AlignToPageSize(size_t size) {
auto page_size = getpagesize();
return (size + page_size - 1) / page_size * page_size;
}
void* AllocateVirtualMemory(size_t size) {
vm_address_t data;
auto error = vm_allocate(mach_task_self(), &data, AlignToPageSize(size),
VM_FLAGS_ANYWHERE);
LOG_IF(FATAL, error != KERN_SUCCESS)
<< "Can't allocate virtual memory for Tensor.";
return reinterpret_cast<void*>(data);
}
void DeallocateVirtualMemory(void* pointer, size_t size) {
vm_deallocate(mach_task_self(), reinterpret_cast<vm_address_t>(pointer),
size);
}
} // namespace
Tensor::MtlBufferView Tensor::GetMtlBufferReadView(
id<MTLCommandBuffer> command_buffer) const {
LOG_IF(FATAL, valid_ == kValidNone)
<< "Tensor must be written prior to read from.";
LOG_IF(FATAL, !(valid_ & (kValidCpu | kValidMetalBuffer)))
<< "Tensor conversion between different GPU resources is not supported "
"yet.";
auto lock(absl::make_unique<absl::MutexLock>(&view_mutex_));
valid_ |= kValidMetalBuffer;
AllocateMtlBuffer([command_buffer device]);
return {metal_buffer_, std::move(lock)};
}
Tensor::MtlBufferView Tensor::GetMtlBufferWriteView(
id<MTLCommandBuffer> command_buffer) const {
// Don't overwrite command buffer at which the metal buffer has been written
// so we can wait until completed.
command_buffer_ = command_buffer;
return GetMtlBufferWriteView([command_buffer device]);
}
Tensor::MtlBufferView Tensor::GetMtlBufferWriteView(
id<MTLDevice> device) const {
auto lock(absl::make_unique<absl::MutexLock>(&view_mutex_));
valid_ = kValidMetalBuffer;
AllocateMtlBuffer(device);
return {metal_buffer_, std::move(lock)};
}
void Tensor::AllocateMtlBuffer(id<MTLDevice> device) const {
device_ = device;
if (!cpu_buffer_) {
// It also means that the metal buffer is not allocated yet.
cpu_buffer_ = AllocateVirtualMemory(bytes());
}
if (!metal_buffer_) {
metal_buffer_ =
[device_ newBufferWithBytesNoCopy:cpu_buffer_
length:AlignToPageSize(bytes())
options:MTLResourceStorageModeShared |
MTLResourceCPUCacheModeDefaultCache
deallocator:^(void* pointer, NSUInteger length) {
DeallocateVirtualMemory(pointer, length);
}];
}
}
#endif // MEDIAPIPE_METAL_ENABLED
#if MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_30
Tensor::OpenGlTexture2dView Tensor::GetOpenGlTexture2dReadView() const {
LOG_IF(FATAL, valid_ == kValidNone)
<< "Tensor must be written prior to read from.";
LOG_IF(FATAL, !(valid_ & (kValidCpu | kValidOpenGlTexture2d)))
<< "Tensor conversion between different GPU resources is not supported "
"yet.";
auto lock = absl::make_unique<absl::MutexLock>(&view_mutex_);
AllocateOpenGlTexture2d();
if (!(valid_ & kValidOpenGlTexture2d)) {
const int padded_size =
texture_height_ * texture_width_ * 4 * element_size();
auto temp_buffer = absl::make_unique<uint8_t[]>(padded_size);
uint8_t* dest_buffer = temp_buffer.get();
uint8_t* src_buffer = reinterpret_cast<uint8_t*>(cpu_buffer_);
const int num_elements = BhwcWidthFromShape(shape_) *
BhwcHeightFromShape(shape_) *
BhwcBatchFromShape(shape_);
const int actual_depth_size = BhwcDepthFromShape(shape_) * element_size();
const int padded_depth_size =
(BhwcDepthFromShape(shape_) + 3) / 4 * 4 * element_size();
for (int e = 0; e < num_elements; e++) {
std::memcpy(dest_buffer, src_buffer, actual_depth_size);
src_buffer += actual_depth_size;
dest_buffer += padded_depth_size;
}
// Transfer from CPU memory into GPU memory.
glBindTexture(GL_TEXTURE_2D, opengl_texture2d_);
// Set alignment for the proper value (default) to avoid address sanitizer
// error "out of boundary reading".
glPixelStorei(GL_UNPACK_ALIGNMENT, 4);
glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, texture_width_, texture_height_,
GL_RGBA, GL_FLOAT, temp_buffer.get());
glBindTexture(GL_TEXTURE_2D, 0);
valid_ |= kValidOpenGlTexture2d;
}
return {opengl_texture2d_, std::move(lock)};
}
Tensor::OpenGlTexture2dView Tensor::GetOpenGlTexture2dWriteView() const {
auto lock = absl::make_unique<absl::MutexLock>(&view_mutex_);
AllocateOpenGlTexture2d();
valid_ = kValidOpenGlTexture2d;
return {opengl_texture2d_, std::move(lock)};
}
Tensor::OpenGlTexture2dView::Layout
Tensor::OpenGlTexture2dView::GetLayoutDimensions(const Tensor::Shape& shape,
int* width, int* height) {
static int max_size = 0;
if (max_size == 0) {
int max_texture_size;
glGetIntegerv(GL_MAX_TEXTURE_SIZE, &max_texture_size);
int max_renderbuffer_size;
glGetIntegerv(GL_MAX_RENDERBUFFER_SIZE, &max_renderbuffer_size);
int max_viewport_dims[2];
glGetIntegerv(GL_MAX_VIEWPORT_DIMS, max_viewport_dims);
max_size = std::min(std::min(max_texture_size, max_renderbuffer_size),
std::min(max_viewport_dims[0], max_viewport_dims[1]));
}
const int num_slices = (BhwcDepthFromShape(shape) + 3) / 4;
const int num_elements = BhwcBatchFromShape(shape) *
BhwcHeightFromShape(shape) *
BhwcWidthFromShape(shape);
const int num_pixels = num_slices * num_elements;
int w = BhwcWidthFromShape(shape) * num_slices;
if (w <= max_size) {
int h = (num_pixels + w - 1) / w;
if (h <= max_size) {
*width = w;
*height = h;
return Tensor::OpenGlTexture2dView::Layout::kAligned;
}
}
// The best performance of a compute shader can be achived with textures'
// width multiple of 256. Making minimum fixed width of 256 waste memory for
// small tensors. The optimal balance memory-vs-performance is power of 2.
// The texture width and height are choosen to be closer to square.
float power = std::log2(std::sqrt(static_cast<float>(num_pixels)));
w = 1 << static_cast<int>(power);
int h = (num_pixels + w - 1) / w;
LOG_IF(FATAL, w > max_size || h > max_size)
<< "The tensor can't fit into OpenGL Texture2D View.";
*width = w;
*height = h;
return Tensor::OpenGlTexture2dView::Layout::kLinearized;
}
void Tensor::AllocateOpenGlTexture2d() const {
if (opengl_texture2d_ == GL_INVALID_INDEX) {
gl_context_ = mediapipe::GlContext::GetCurrent();
LOG_IF(FATAL, !gl_context_) << "GlContext is not bound to the thread.";
glGenTextures(1, &opengl_texture2d_);
glBindTexture(GL_TEXTURE_2D, opengl_texture2d_);
// Texture2D represents a buffer with computable data so should be fetched
// but not sampled - can affect performance. Also on GLES2.0 sampling is not
// supported from floating point textures.
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
OpenGlTexture2dView::GetLayoutDimensions(shape_, &texture_width_,
&texture_height_);
if (gl_context_->GetGlVersion() != mediapipe::GlVersion::kGLES2) {
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_BASE_LEVEL, 0);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAX_LEVEL, 0);
glTexStorage2D(GL_TEXTURE_2D, 1, GL_RGBA32F, texture_width_,
texture_height_);
} else {
// GLES2.0 supports only clamp addressing mode for NPOT textures.
// If any of dimensions is NPOT then both addressing modes are clamp.
if (!IsPowerOfTwo(texture_width_) || !IsPowerOfTwo(texture_height_)) {
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
}
// We assume all contexts will have the same extensions, so we only check
// once for OES_texture_float extension, to save time.
static bool has_oes_extension =
gl_context_->HasGlExtension("OES_texture_float");
LOG_IF(FATAL, !has_oes_extension)
<< "OES_texture_float extension required in order to use MP tensor "
<< "with GLES 2.0";
// Allocate the image data; note that it's no longer RGBA32F, so will be
// lower precision.
glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA, texture_width_, texture_height_,
0, GL_RGBA, GL_FLOAT, 0 /* data */);
}
glBindTexture(GL_TEXTURE_2D, 0);
glGenFramebuffers(1, &frame_buffer_);
}
}
#endif // MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_30
#if MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_31
Tensor::OpenGlBufferView Tensor::GetOpenGlBufferReadView() const {
LOG_IF(FATAL, valid_ == kValidNone)
<< "Tensor must be written prior to read from.";
LOG_IF(FATAL, !(valid_ & (kValidCpu | kValidOpenGlBuffer)))
<< "Tensor conversion between different GPU resources is not supported "
"yet.";
auto lock(absl::make_unique<absl::MutexLock>(&view_mutex_));
AllocateOpenGlBuffer();
if (!(valid_ & kValidOpenGlBuffer)) {
glBindBuffer(GL_SHADER_STORAGE_BUFFER, opengl_buffer_);
void* ptr =
glMapBufferRange(GL_SHADER_STORAGE_BUFFER, 0, bytes(),
GL_MAP_INVALIDATE_BUFFER_BIT | GL_MAP_WRITE_BIT);
std::memcpy(ptr, cpu_buffer_, bytes());
glUnmapBuffer(GL_SHADER_STORAGE_BUFFER);
valid_ |= kValidOpenGlBuffer;
}
return {opengl_buffer_, std::move(lock)};
}
Tensor::OpenGlBufferView Tensor::GetOpenGlBufferWriteView() const {
auto lock(absl::make_unique<absl::MutexLock>(&view_mutex_));
AllocateOpenGlBuffer();
valid_ = kValidOpenGlBuffer;
return {opengl_buffer_, std::move(lock)};
}
void Tensor::AllocateOpenGlBuffer() const {
if (opengl_buffer_ == GL_INVALID_INDEX) {
gl_context_ = mediapipe::GlContext::GetCurrent();
LOG_IF(FATAL, !gl_context_) << "GlContext is not bound to the thread.";
glGenBuffers(1, &opengl_buffer_);
glBindBuffer(GL_SHADER_STORAGE_BUFFER, opengl_buffer_);
glBufferData(GL_SHADER_STORAGE_BUFFER, bytes(), NULL, GL_STREAM_COPY);
}
}
#endif // MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_31
Tensor& Tensor::operator=(Tensor&& src) {
if (this != &src) {
Invalidate();
Move(&src);
}
return *this;
}
void Tensor::Move(Tensor* src) {
valid_ = src->valid_;
src->valid_ = kValidNone;
shape_ = src->shape();
element_type_ = src->element_type();
src->element_type_ = ElementType::kNone; // Mark as invalidated.
cpu_buffer_ = src->cpu_buffer_;
src->cpu_buffer_ = nullptr;
#if MEDIAPIPE_METAL_ENABLED
device_ = src->device_;
command_buffer_ = src->command_buffer_;
metal_buffer_ = src->metal_buffer_;
src->metal_buffer_ = nil;
#endif // MEDIAPIPE_METAL_ENABLED
#if MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_30
gl_context_ = std::move(src->gl_context_);
frame_buffer_ = src->frame_buffer_;
src->frame_buffer_ = GL_INVALID_INDEX;
opengl_texture2d_ = src->opengl_texture2d_;
src->opengl_texture2d_ = GL_INVALID_INDEX;
texture_width_ = src->texture_width_;
texture_height_ = src->texture_height_;
#if MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_31
opengl_buffer_ = src->opengl_buffer_;
src->opengl_buffer_ = GL_INVALID_INDEX;
#endif // MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_31
#endif // MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_30
}
Tensor::Tensor(ElementType element_type, const Shape& shape)
: element_type_(element_type), shape_(shape) {}
void Tensor::Invalidate() {
#if MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_30
GLuint cleanup_gl_tex = GL_INVALID_INDEX;
GLuint cleanup_gl_fb = GL_INVALID_INDEX;
GLuint cleanup_gl_buf = GL_INVALID_INDEX;
#endif // MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_30
{
absl::MutexLock lock(&view_mutex_);
#if MEDIAPIPE_METAL_ENABLED
// If memory is allocated and not owned by the metal buffer.
// TODO: Re-design cpu buffer memory management.
if (cpu_buffer_ && !metal_buffer_) {
DeallocateVirtualMemory(cpu_buffer_, AlignToPageSize(bytes()));
}
metal_buffer_ = nil;
#else
if (cpu_buffer_) {
free(cpu_buffer_);
}
#endif // MEDIAPIPE_METAL_ENABLED
cpu_buffer_ = nullptr;
// Don't need to wait for the resource to be deleted bacause if will be
// released on last reference deletion inside the OpenGL driver.
#if MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_30
std::swap(cleanup_gl_tex, opengl_texture2d_);
std::swap(cleanup_gl_fb, frame_buffer_);
#if MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_31
std::swap(cleanup_gl_buf, opengl_buffer_);
#endif // MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_31
#endif // MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_30
}
// Do not hold the view mutex while invoking GlContext::RunWithoutWaiting,
// since that method may acquire the context's own lock.
#if MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_30
if (cleanup_gl_tex != GL_INVALID_INDEX || cleanup_gl_fb != GL_INVALID_INDEX ||
cleanup_gl_buf != GL_INVALID_INDEX)
gl_context_->RunWithoutWaiting([cleanup_gl_tex, cleanup_gl_fb
#if MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_31
,
cleanup_gl_buf
#endif // MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_31
]() {
glDeleteTextures(1, &cleanup_gl_tex);
glDeleteFramebuffers(1, &cleanup_gl_fb);
#if MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_31
glDeleteBuffers(1, &cleanup_gl_buf);
#endif // MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_31
});
#endif // MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_30
}
Tensor::CpuReadView Tensor::GetCpuReadView() const {
auto lock = absl::make_unique<absl::MutexLock>(&view_mutex_);
LOG_IF(FATAL, valid_ == kValidNone)
<< "Tensor must be written prior to read from.";
AllocateCpuBuffer();
if (!(valid_ & kValidCpu)) {
// GPU-to-CPU synchronization and read-back.
#if MEDIAPIPE_METAL_ENABLED
if (valid_ & kValidMetalBuffer) {
LOG_IF(FATAL, !command_buffer_) << "Metal -> CPU synchronization "
"requires MTLCommandBuffer to be set.";
if (command_buffer_) {
[command_buffer_ waitUntilCompleted];
}
}
#endif // MEDIAPIPE_METAL_ENABLED
#if MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_30
#if MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_31
// TODO: we cannot just grab the GL context's lock while holding
// the view mutex here.
if (valid_ & kValidOpenGlBuffer) {
gl_context_->Run([this]() {
glBindBuffer(GL_SHADER_STORAGE_BUFFER, opengl_buffer_);
const void* ptr = glMapBufferRange(GL_SHADER_STORAGE_BUFFER, 0, bytes(),
GL_MAP_READ_BIT);
std::memcpy(cpu_buffer_, ptr, bytes());
glUnmapBuffer(GL_SHADER_STORAGE_BUFFER);
});
} else
#endif // MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_31
// Transfer data from texture if not transferred from SSBO/MTLBuffer
// yet.
if (valid_ & kValidOpenGlTexture2d) {
gl_context_->Run([this]() {
const int padded_size =
texture_height_ * texture_width_ * 4 * element_size();
auto temp_buffer = absl::make_unique<uint8_t[]>(padded_size);
uint8_t* buffer = temp_buffer.get();
glBindFramebuffer(GL_FRAMEBUFFER, frame_buffer_);
glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0,
GL_TEXTURE_2D, opengl_texture2d_, 0);
glPixelStorei(GL_PACK_ALIGNMENT, 4);
glReadPixels(0, 0, texture_width_, texture_height_, GL_RGBA, GL_FLOAT,
buffer);
uint8_t* dest_buffer = reinterpret_cast<uint8_t*>(cpu_buffer_);
const int actual_depth_size =
BhwcDepthFromShape(shape_) * element_size();
const int num_slices = (BhwcDepthFromShape(shape_) + 3) / 4;
const int padded_depth_size = num_slices * 4 * element_size();
const int num_elements = BhwcWidthFromShape(shape_) *
BhwcHeightFromShape(shape_) *
BhwcBatchFromShape(shape_);
for (int e = 0; e < num_elements; e++) {
std::memcpy(dest_buffer, buffer, actual_depth_size);
dest_buffer += actual_depth_size;
buffer += padded_depth_size;
}
});
}
#endif // MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_30
valid_ |= kValidCpu;
}
return {cpu_buffer_, std::move(lock)};
}
Tensor::CpuWriteView Tensor::GetCpuWriteView() const {
auto lock = absl::make_unique<absl::MutexLock>(&view_mutex_);
AllocateCpuBuffer();
valid_ = kValidCpu;
return {cpu_buffer_, std::move(lock)};
}
void Tensor::AllocateCpuBuffer() const {
if (!cpu_buffer_) {
#if MEDIAPIPE_METAL_ENABLED
cpu_buffer_ = AllocateVirtualMemory(bytes());
#else
cpu_buffer_ = malloc(bytes());
#endif // MEDIAPIPE_METAL_ENABLED
}
}
} // namespace mediapipe
Reference in New Issue View Git Blame Copy Permalink