thek0tyara/llama-cpp-turboquant
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 4

Introduced simplified interface for creation of algobase

Browse source
This commit is contained in:
Alejandro Saucedo 2020-08-31 09:35:40 +01:00
parent 13206a9d9b
commit f29e56fcb8
4 changed files with 173 additions and 23 deletions
Download patch file Download diff file Expand all files Collapse all files

94
single_include/kompute/Kompute.hpp
View file

@ -685,6 +685,23 @@ class Manager
SPDLOG_DEBUG("Kompute Manager evalOp running sequence SUCCESS");
}
/**
* Operation that adds extra operations to existing or new created
* sequences.
*
* @param tensors The tensors to be used in the operation recorded
* @param sequenceName The name of the sequence to be retrieved or created
* @param TArgs Template parameters that will be used to initialise
* Operation to allow for extensible configurations on initialisation
*/
template<typename T, typename... TArgs>
void evalOpDefault(std::vector<std::shared_ptr<Tensor>> tensors,
TArgs&&... params)
{
SPDLOG_DEBUG("Kompute Manager evalOp Default triggered");
this->evalOp<T>(tensors, KP_DEFAULT_SESSION, std::forward<TArgs>(params)...);
}
private:
// -------------- OPTIONALLY OWNED RESOURCES
std::shared_ptr<vk::Instance> mInstance = nullptr;
@ -845,9 +862,44 @@ class OpAlgoBase : public OpBase
std::shared_ptr<vk::Device> device,
std::shared_ptr<vk::CommandBuffer> commandBuffer,
std::vector<std::shared_ptr<Tensor>>& tensors,
bool copyOutputData = false,
std::string shaderFilePath = "",
const std::vector<char>& shaderDataRaw = {});
bool copyOutputData);
/**
* Constructor that enables a file to be passed to the operation with
* the contents of the shader. This can be either in raw format or in
* compiled SPIR-V binary format.
*
* @param physicalDevice Vulkan physical device used to find device queues
* @param device Vulkan logical device for passing to Algorithm
* @param commandBuffer Vulkan Command Buffer to record commands into
* @param tensors Tensors that are to be used in this operation
* @param copyOutputData Whether to map device data for all output tensors back to their host data vectors
* @param shaderFilePath Optional parameter to specify the shader to load (either in spirv or raw format)
*/
OpAlgoBase(std::shared_ptr<vk::PhysicalDevice> physicalDevice,
std::shared_ptr<vk::Device> device,
std::shared_ptr<vk::CommandBuffer> commandBuffer,
std::vector<std::shared_ptr<Tensor>>& tensors,
bool copyOutputData,
std::string shaderFilePath);
/**
* Constructor that enables raw shader data to be passed to the main operation
* which can be either in raw shader glsl code or in compiled SPIR-V binary.
*
* @param physicalDevice Vulkan physical device used to find device queues
* @param device Vulkan logical device for passing to Algorithm
* @param commandBuffer Vulkan Command Buffer to record commands into
* @param tensors Tensors that are to be used in this operation
* @param copyOutputData Whether to map device data for all output tensors back to their host data vectors
* @param shaderDataRaw Optional parameter to specify the shader data either in binary or raw form
*/
OpAlgoBase(std::shared_ptr<vk::PhysicalDevice> physicalDevice,
std::shared_ptr<vk::Device> device,
std::shared_ptr<vk::CommandBuffer> commandBuffer,
std::vector<std::shared_ptr<Tensor>>& tensors,
bool copyOutputData,
const std::vector<char>& shaderDataRaw);
/**
* Default destructor, which is in charge of destroying the algorithm
@ -920,12 +972,10 @@ OpAlgoBase<tX, tY, tZ>::OpAlgoBase(std::shared_ptr<vk::PhysicalDevice> physicalD
std::shared_ptr<vk::Device> device,
std::shared_ptr<vk::CommandBuffer> commandBuffer,
std::vector<std::shared_ptr<Tensor>>& tensors,
bool copyOutputData,
std::string shaderFilePath,
const std::vector<char>& shaderDataRaw)
bool copyOutputData)
: OpBase(physicalDevice, device, commandBuffer, tensors, false)
{
SPDLOG_DEBUG("Kompute OpAlgoBase constructor with params numTensors: {} copyOutputData: {}, shaderFilePath: {}", tensors.size(), copyOutputData, shaderFilePath);
SPDLOG_DEBUG("Kompute OpAlgoBase constructor with params numTensors: {} copyOutputData: {}, shaderFilePath: {}", tensors.size(), copyOutputData);
// The dispatch size is set up based on either explicitly provided template
// parameters or by default it would take the shape and size of the tensors
@ -947,13 +997,39 @@ OpAlgoBase<tX, tY, tZ>::OpAlgoBase(std::shared_ptr<vk::PhysicalDevice> physicalD
this->mY,
this->mZ);
this->mShaderFilePath = shaderFilePath;
this->mCopyOutputData = copyOutputData;
this->mShaderDataRaw = shaderDataRaw;
this->mAlgorithm = std::make_shared<Algorithm>(device, commandBuffer);
}
template<uint32_t tX, uint32_t tY, uint32_t tZ>
OpAlgoBase<tX, tY, tZ>::OpAlgoBase(std::shared_ptr<vk::PhysicalDevice> physicalDevice,
std::shared_ptr<vk::Device> device,
std::shared_ptr<vk::CommandBuffer> commandBuffer,
std::vector<std::shared_ptr<Tensor>>& tensors,
bool copyOutputData,
std::string shaderFilePath)
: OpAlgoBase(physicalDevice, device, commandBuffer, tensors, copyOutputData)
{
SPDLOG_DEBUG("Kompute OpAlgoBase shaderFilePath constructo with shaderfile path: {}", shaderFilePath);
this->mShaderFilePath = shaderFilePath;
}
template<uint32_t tX, uint32_t tY, uint32_t tZ>
OpAlgoBase<tX, tY, tZ>::OpAlgoBase(std::shared_ptr<vk::PhysicalDevice> physicalDevice,
std::shared_ptr<vk::Device> device,
std::shared_ptr<vk::CommandBuffer> commandBuffer,
std::vector<std::shared_ptr<Tensor>>& tensors,
bool copyOutputData,
const std::vector<char>& shaderDataRaw)
: OpAlgoBase(physicalDevice, device, commandBuffer, tensors, copyOutputData)
{
SPDLOG_DEBUG("Kompute OpAlgoBase shaderFilePath constructo with shader raw data length: {}", shaderDataRaw.size());
this->mShaderDataRaw = shaderDataRaw;
}
template<uint32_t tX, uint32_t tY, uint32_t tZ>
OpAlgoBase<tX, tY, tZ>::~OpAlgoBase()
{

17
src/include/kompute/Manager.hpp
View file

@ -95,6 +95,23 @@ class Manager
SPDLOG_DEBUG("Kompute Manager evalOp running sequence SUCCESS");
}
/**
* Operation that adds extra operations to existing or new created
* sequences.
*
* @param tensors The tensors to be used in the operation recorded
* @param sequenceName The name of the sequence to be retrieved or created
* @param TArgs Template parameters that will be used to initialise
* Operation to allow for extensible configurations on initialisation
*/
template<typename T, typename... TArgs>
void evalOpDefault(std::vector<std::shared_ptr<Tensor>> tensors,
TArgs&&... params)
{
SPDLOG_DEBUG("Kompute Manager evalOp Default triggered");
this->evalOp<T>(tensors, KP_DEFAULT_SESSION, std::forward<TArgs>(params)...);
}
private:
// -------------- OPTIONALLY OWNED RESOURCES
std::shared_ptr<vk::Instance> mInstance = nullptr;

77
src/include/kompute/operations/OpAlgoBase.hpp
View file

@ -57,9 +57,44 @@ class OpAlgoBase : public OpBase
std::shared_ptr<vk::Device> device,
std::shared_ptr<vk::CommandBuffer> commandBuffer,
std::vector<std::shared_ptr<Tensor>>& tensors,
bool copyOutputData = false,
std::string shaderFilePath = "",
const std::vector<char>& shaderDataRaw = {});
bool copyOutputData);
/**
* Constructor that enables a file to be passed to the operation with
* the contents of the shader. This can be either in raw format or in
* compiled SPIR-V binary format.
*
* @param physicalDevice Vulkan physical device used to find device queues
* @param device Vulkan logical device for passing to Algorithm
* @param commandBuffer Vulkan Command Buffer to record commands into
* @param tensors Tensors that are to be used in this operation
* @param copyOutputData Whether to map device data for all output tensors back to their host data vectors
* @param shaderFilePath Optional parameter to specify the shader to load (either in spirv or raw format)
*/
OpAlgoBase(std::shared_ptr<vk::PhysicalDevice> physicalDevice,
std::shared_ptr<vk::Device> device,
std::shared_ptr<vk::CommandBuffer> commandBuffer,
std::vector<std::shared_ptr<Tensor>>& tensors,
bool copyOutputData,
std::string shaderFilePath);
/**
* Constructor that enables raw shader data to be passed to the main operation
* which can be either in raw shader glsl code or in compiled SPIR-V binary.
*
* @param physicalDevice Vulkan physical device used to find device queues
* @param device Vulkan logical device for passing to Algorithm
* @param commandBuffer Vulkan Command Buffer to record commands into
* @param tensors Tensors that are to be used in this operation
* @param copyOutputData Whether to map device data for all output tensors back to their host data vectors
* @param shaderDataRaw Optional parameter to specify the shader data either in binary or raw form
*/
OpAlgoBase(std::shared_ptr<vk::PhysicalDevice> physicalDevice,
std::shared_ptr<vk::Device> device,
std::shared_ptr<vk::CommandBuffer> commandBuffer,
std::vector<std::shared_ptr<Tensor>>& tensors,
bool copyOutputData,
const std::vector<char>& shaderDataRaw);
/**
* Default destructor, which is in charge of destroying the algorithm
@ -132,12 +167,10 @@ OpAlgoBase<tX, tY, tZ>::OpAlgoBase(std::shared_ptr<vk::PhysicalDevice> physicalD
std::shared_ptr<vk::Device> device,
std::shared_ptr<vk::CommandBuffer> commandBuffer,
std::vector<std::shared_ptr<Tensor>>& tensors,
bool copyOutputData,
std::string shaderFilePath,
const std::vector<char>& shaderDataRaw)
bool copyOutputData)
: OpBase(physicalDevice, device, commandBuffer, tensors, false)
{
SPDLOG_DEBUG("Kompute OpAlgoBase constructor with params numTensors: {} copyOutputData: {}, shaderFilePath: {}", tensors.size(), copyOutputData, shaderFilePath);
SPDLOG_DEBUG("Kompute OpAlgoBase constructor with params numTensors: {} copyOutputData: {}, shaderFilePath: {}", tensors.size(), copyOutputData);
// The dispatch size is set up based on either explicitly provided template
// parameters or by default it would take the shape and size of the tensors
@ -159,13 +192,39 @@ OpAlgoBase<tX, tY, tZ>::OpAlgoBase(std::shared_ptr<vk::PhysicalDevice> physicalD
this->mY,
this->mZ);
this->mShaderFilePath = shaderFilePath;
this->mCopyOutputData = copyOutputData;
this->mShaderDataRaw = shaderDataRaw;
this->mAlgorithm = std::make_shared<Algorithm>(device, commandBuffer);
}
template<uint32_t tX, uint32_t tY, uint32_t tZ>
OpAlgoBase<tX, tY, tZ>::OpAlgoBase(std::shared_ptr<vk::PhysicalDevice> physicalDevice,
std::shared_ptr<vk::Device> device,
std::shared_ptr<vk::CommandBuffer> commandBuffer,
std::vector<std::shared_ptr<Tensor>>& tensors,
bool copyOutputData,
std::string shaderFilePath)
: OpAlgoBase(physicalDevice, device, commandBuffer, tensors, copyOutputData)
{
SPDLOG_DEBUG("Kompute OpAlgoBase shaderFilePath constructo with shaderfile path: {}", shaderFilePath);
this->mShaderFilePath = shaderFilePath;
}
template<uint32_t tX, uint32_t tY, uint32_t tZ>
OpAlgoBase<tX, tY, tZ>::OpAlgoBase(std::shared_ptr<vk::PhysicalDevice> physicalDevice,
std::shared_ptr<vk::Device> device,
std::shared_ptr<vk::CommandBuffer> commandBuffer,
std::vector<std::shared_ptr<Tensor>>& tensors,
bool copyOutputData,
const std::vector<char>& shaderDataRaw)
: OpAlgoBase(physicalDevice, device, commandBuffer, tensors, copyOutputData)
{
SPDLOG_DEBUG("Kompute OpAlgoBase shaderFilePath constructo with shader raw data length: {}", shaderDataRaw.size());
this->mShaderDataRaw = shaderDataRaw;
}
template<uint32_t tX, uint32_t tY, uint32_t tZ>
OpAlgoBase<tX, tY, tZ>::~OpAlgoBase()
{

8
test/TestCustomOpRawShaders.cpp
View file

@ -10,7 +10,7 @@ TEST_CASE("op_custom_simple_raw_shader") {
std::shared_ptr<kp::Tensor> tensorA{ new kp::Tensor({ 3, 4, 5 })};
std::shared_ptr<kp::Tensor> tensorB{ new kp::Tensor({ 0, 0, 0 })};
mgr.evalOp<kp::OpCreateTensor>({ tensorA, tensorB });
mgr.evalOpDefault<kp::OpCreateTensor>({ tensorA, tensorB });
std::string shader(
"#version 450\n"
@ -24,11 +24,9 @@ TEST_CASE("op_custom_simple_raw_shader") {
"}\n"
);
mgr.evalOp<kp::OpAlgoBase<>>(
mgr.evalOpDefault<kp::OpAlgoBase<>>(
{ tensorA, tensorB },
"DEFAULT",
true,
"",
true, // Whether to copy output from device
std::vector<char>(shader.begin(), shader.end()));
REQUIRE(tensorA->data() == std::vector<uint32_t>{0, 1, 2});