Skip to content

[ET-VK] Move Tensor.* to namespace api #4124

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
wants to merge 2 commits into from
Closed

[ET-VK] Move Tensor.* to namespace api #4124

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
d2e9c7d
[ET-VK] Move `Tensor.*` to namespace `api`
jorgep31415 Jul 2, 2024
e1bc5da
Update on "[ET-VK] Move `Tensor.*` to namespace `api`"
jorgep31415 Jul 2, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
187 changes: 93 additions & 94 deletions backends/vulkan/runtime/api/Tensor.cpp
View file
Open in desktop

Large diffs are not rendered by default.

113 changes: 52 additions & 61 deletions backends/vulkan/runtime/api/Tensor.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,6 +15,7 @@
#include <executorch/backends/vulkan/runtime/api/Types.h>

namespace vkcompute {
namespace api {

/*
* Given the sizes of a tensor and the GPU memory layout, calculate the strides
Expand All @@ -29,7 +30,7 @@ namespace vkcompute {
*/
std::vector<int64_t> calculate_strides(
const std::vector<int64_t>& sizes,
const api::GPUMemoryLayout memory_layout,
const GPUMemoryLayout memory_layout,
const bool texel_strides = true);

/*
Expand All @@ -48,27 +49,24 @@ std::vector<int64_t> calculate_strides(
*/
std::vector<int64_t> calculate_padded_sizes(
const std::vector<int64_t>& sizes,
const api::GPUMemoryLayout memory_layout);
const GPUMemoryLayout memory_layout);

/*
* Given the padded sizes of a tensor and the GPU memory layout, calculate the
* 3D image extents required to store the tensor data as an image texture.
*/
utils::uvec3 calculate_image_extents(
const std::vector<int64_t>& padded_sizes,
const api::GPUMemoryLayout memory_layout);
const GPUMemoryLayout memory_layout);

struct LastAccess {
api::PipelineStageFlags stage;
api::MemoryAccessFlags access;
PipelineStageFlags stage;
MemoryAccessFlags access;

LastAccess()
: stage{api::PipelineStage::NO_STAGE},
access{api::MemoryAccessType::NONE} {}
: stage{PipelineStage::NO_STAGE}, access{MemoryAccessType::NONE} {}

LastAccess(
api::PipelineStageFlags stage_flags,
api::MemoryAccessFlags access_flags)
LastAccess(PipelineStageFlags stage_flags, MemoryAccessFlags access_flags)
: stage{stage_flags}, access{access_flags} {}
};

Expand All @@ -78,11 +76,11 @@ class vTensorStorage final {
vTensorStorage() = default;

vTensorStorage(
api::Context* context,
const api::StorageType storage_type,
const api::GPUMemoryLayout gpu_memory_layout,
Context* context,
const StorageType storage_type,
const GPUMemoryLayout gpu_memory_layout,
const std::vector<int64_t>& sizes,
const api::ScalarType dtype,
const ScalarType dtype,
const bool allocate_memory = true);

vTensorStorage(const vTensorStorage& other) = delete;
Expand All @@ -97,17 +95,17 @@ class vTensorStorage final {

private:
// Context
api::Context* context_{};
Context* context_{};

api::StorageType storage_type_;
StorageType storage_type_;

// Resource sizings
utils::uvec3 image_extents_{};
int64_t buffer_length_{};

// GPU Storage
mutable api::VulkanImage image_;
mutable api::VulkanBuffer buffer_;
mutable VulkanImage image_;
mutable VulkanBuffer buffer_;

// Last Access - used to insert memory barriers
LastAccess last_access_;
Expand All @@ -118,9 +116,9 @@ class vTensorStorage final {

// Memory barrier insertion
void transition(
api::PipelineBarrier&,
const api::PipelineStageFlags,
const api::MemoryAccessFlags);
PipelineBarrier&,
const PipelineStageFlags,
const MemoryAccessFlags);

// Validation
void verify() const;
Expand All @@ -132,8 +130,8 @@ class vTensorStorage final {

void discard_and_reallocate(
const std::vector<int64_t>& padded_sizes,
const api::GPUMemoryLayout gpu_memory_layout,
const api::ScalarType dtype);
const GPUMemoryLayout gpu_memory_layout,
const ScalarType dtype);
};

class vTensor final {
Expand All @@ -146,11 +144,11 @@ class vTensor final {

public:
explicit vTensor(
api::Context* context,
Context* context,
const std::vector<int64_t>& sizes,
const api::ScalarType dtype,
const api::StorageType storage_type = api::kTexture3D,
const api::GPUMemoryLayout memory_layout = api::kChannelsPacked,
const ScalarType dtype,
const StorageType storage_type = kTexture3D,
const GPUMemoryLayout memory_layout = kChannelsPacked,
const bool allocate_memory = true);

vTensor(const vTensor& other) = delete;
Expand All @@ -160,8 +158,8 @@ class vTensor final {
vTensor& operator=(vTensor&& other) = default;

private:
api::ScalarType dtype_;
api::GPUMemoryLayout memory_layout_;
ScalarType dtype_;
GPUMemoryLayout memory_layout_;

// sizes of the tensor in NCHW dimension order
std::vector<int64_t> sizes_;
Expand All @@ -181,10 +179,10 @@ class vTensor final {
* Refer to the comments for the corresponding *_ubo() functions for more
* context about the data contained in each buffer.
*/
api::ParamsBuffer sizes_uniform_;
api::ParamsBuffer texture_limits_uniform_;
api::ParamsBuffer texel_strides_uniform_;
api::ParamsBuffer ntexels_uniform_;
ParamsBuffer sizes_uniform_;
ParamsBuffer texture_limits_uniform_;
ParamsBuffer texel_strides_uniform_;
ParamsBuffer ntexels_uniform_;

vTensorStorage storage_;

Expand All @@ -193,56 +191,48 @@ class vTensor final {
Texture Access
*/

inline api::VulkanImage& image() const& {
inline VulkanImage& image() const& {
return storage_.image_;
}

api::VulkanImage& image(
api::PipelineBarrier&,
const api::PipelineStageFlags) &;
VulkanImage& image(PipelineBarrier&, const PipelineStageFlags) &;

api::VulkanImage& image(
api::PipelineBarrier&,
const api::PipelineStageFlags,
const api::MemoryAccessFlags) &;
VulkanImage&
image(PipelineBarrier&, const PipelineStageFlags, const MemoryAccessFlags) &;

inline api::VulkanBuffer& buffer() const& {
inline VulkanBuffer& buffer() const& {
return storage_.buffer_;
}

api::VulkanBuffer& buffer(
api::PipelineBarrier&,
const api::PipelineStageFlags) &;
VulkanBuffer& buffer(PipelineBarrier&, const PipelineStageFlags) &;

api::VulkanBuffer& buffer(
api::PipelineBarrier&,
const api::PipelineStageFlags,
const api::MemoryAccessFlags) &;
VulkanBuffer&
buffer(PipelineBarrier&, const PipelineStageFlags, const MemoryAccessFlags) &;

/*
Metadata
*/

inline api::StorageType storage_type() const {
inline StorageType storage_type() const {
return storage_.storage_type_;
}

inline bool has_buffer_storage() const {
return storage_.storage_type_ == api::kBuffer;
return storage_.storage_type_ == kBuffer;
}

inline const utils::uvec3& image_extents() const {
return storage_.image_extents_;
}

/*
* Extract an `api::ScalarType` from the TensorOptions member
* Extract an `ScalarType` from the TensorOptions member
*/
inline api::ScalarType dtype() const {
inline ScalarType dtype() const {
return dtype_;
}

inline api::GPUMemoryLayout gpu_memory_layout() const {
inline GPUMemoryLayout gpu_memory_layout() const {
return memory_layout_;
}

Expand All @@ -267,7 +257,7 @@ class vTensor final {
* Note that dimensions that are not present in the tensor's sizes are set to
* a size of 1.
*/
const api::BufferBindInfo sizes_ubo();
const BufferBindInfo sizes_ubo();

/*
* Returns a GPU buffer containing the virtual image extents of the tensor.
Expand All @@ -278,18 +268,18 @@ class vTensor final {
*
* This buffer should only be used to
*/
const api::BufferBindInfo texture_limits_ubo();
const BufferBindInfo texture_limits_ubo();

/*
* Returns the strides of the texel buffer used to store the tensor, as
* calculated by calculate_strides().
*/
const api::BufferBindInfo texel_strides_ubo();
const BufferBindInfo texel_strides_ubo();

/*
* Returns the number of texels in the texel buffer used to store the tensor.
*/
const api::BufferBindInfo ntexels_ubo();
const BufferBindInfo ntexels_ubo();

inline const utils::ivec3 texture_limits() const {
return texture_limits_.limits;
Expand All @@ -300,7 +290,7 @@ class vTensor final {
}

inline size_t nbytes() const {
return api::element_size(dtype()) * numel();
return element_size(dtype()) * numel();
}

/*
Expand All @@ -322,7 +312,7 @@ class vTensor final {
* Return nbytes but based on padded_sizes_ instead of sizes_
*/
inline VkDeviceSize gpu_nbytes() const {
return api::element_size(dtype()) * gpu_numel();
return element_size(dtype()) * gpu_numel();
}

/*
Expand All @@ -338,7 +328,7 @@ class vTensor final {
/*
* Binds the underlying resource to the given memory allocation
*/
void bind_allocation(const api::Allocation& allocation);
void bind_allocation(const Allocation& allocation);

private:
/*
Expand All @@ -362,4 +352,5 @@ class vTensor final {
void virtual_resize(const std::vector<int64_t>& new_sizes);
};

} // namespace api
} // namespace vkcompute
4 changes: 2 additions & 2 deletions backends/vulkan/runtime/graph/ComputeGraph.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -36,7 +36,7 @@ namespace vkcompute {
graph_->values_in_use_--; \
}

VALUE_PTR_CLASS_IMPL(vTensorPtr, vTensor, Tensor)
VALUE_PTR_CLASS_IMPL(vTensorPtr, api::vTensor, Tensor)
VALUE_PTR_CLASS_IMPL(TensorRefPtr, TensorRef, TensorRef)
VALUE_PTR_CLASS_IMPL(StagingPtr, api::StorageBuffer, Staging)
VALUE_PTR_CLASS_IMPL(IntListPtr, std::vector<int64_t>, IntList)
Expand Down Expand Up @@ -151,7 +151,7 @@ ValueRef ComputeGraph::add_tensor(

ValueRef idx(static_cast<int>(values_.size()));
check_no_active_value_ptrs();
values_.emplace_back(vTensor(
values_.emplace_back(api::vTensor(
context(), sizes, dtype, storage_type, memory_layout, allocate_memory));

if (!allocate_memory) {
Expand Down
Loading
Loading