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

Added baseline functionality including new memory models

Browse source
This commit is contained in:
Alejandro Saucedo 2021-02-24 22:26:02 +00:00
parent 9aae5d69db
commit 635fdb02be
22 changed files with 283 additions and 1919 deletions
Download patch file Download diff file Expand all files Collapse all files

44
src/include/kompute/operations/OpMult.hpp
View file

@ -4,14 +4,12 @@
#include "kompute/Core.hpp"
#if RELEASE
#include "kompute/shaders/shaderopmult.hpp"
#endif
#include "kompute/Algorithm.hpp"
#include "kompute/Tensor.hpp"
#include "kompute/operations/OpAlgoCreate.hpp"
#include "kompute/operations/OpAlgoDispatch.hpp"
namespace kp {
@ -19,15 +17,9 @@ namespace kp {
* Operation that performs multiplication on two tensors and outpus on third
* tensor.
*/
class OpMult : public OpAlgoCreate
class OpMult : public OpAlgoDispatch
{
public:
/**
* Base constructor, should not be used unless explicitly intended.
*/
OpMult() {
}
/**
* Default constructor with parameters that provides the bare minimum
@ -40,46 +32,30 @@ class OpMult : public OpAlgoCreate
* @param tensors Tensors that are to be used in this operation
* @param komputeWorkgroup Optional parameter to specify the layout for processing
*/
OpMult(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,
const Workgroup& komputeWorkgroup = {})
: OpAlgoCreate(physicalDevice, device, commandBuffer, tensors, "", komputeWorkgroup)
OpMult(std::vector<std::shared_ptr<Tensor>> tensors, std::shared_ptr<Algorithm> algorithm)
: OpAlgoDispatch(tensors, algorithm)
{
KP_LOG_DEBUG("Kompute OpMult constructor with params");
#ifndef RELEASE
this->mShaderFilePath = "shaders/glsl/opmult.comp.spv";
#endif
}
if (tensors.size() != 3) {
throw std::runtime_error("Kompute OpMult expected 3 tensors but got " + tensors.size());
}
#if RELEASE
/**
* If RELEASE=1 it will be using the static version of the shader which is
* loaded using this file directly. Otherwise it should not override the function.
*/
std::vector<uint32_t> fetchSpirvBinaryData() override
{
KP_LOG_WARN(
"Kompute OpMult Running shaders directly from header");
return std::vector<uint32_t>(
std::vector<uint32_t> spirv(
(uint32_t*)shader_data::shaders_glsl_opmult_comp_spv,
(uint32_t*)(shader_data::shaders_glsl_opmult_comp_spv +
kp::shader_data::shaders_glsl_opmult_comp_spv_len));
algorithm->rebuild(tensors, spirv, Workgroup({tensors[0]->size()}));
}
#endif
/**
* Default destructor, which is in charge of destroying the algorithm
* components but does not destroy the underlying tensors
*/
~OpMult() override {
virtual ~OpMult() override {
KP_LOG_DEBUG("Kompute OpMult destructor started");
}
};
} // End namespace kp