vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Fix matmul kernel, continue implementation

Browse source
This commit is contained in:
0cc4m 2023-06-10 16:24:37 +02:00
parent 061246fb07
commit 4a96d0eb7f
3 changed files with 49 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

2
Makefile
View file

@ -216,7 +216,7 @@ endif # LLAMA_METAL
ifdef LLAMA_VULKAN
CFLAGS += -DGGML_USE_VULKAN
LDFLAGS += -lvulkan
OBJS += ggml-vulkan.o ggml-vulkan-matmul-shader
OBJS += ggml-vulkan.o
ggml-vulkan.o: ggml-vulkan.cpp ggml-vulkan.h
$(CXX) $(CXXFLAGS) -c $< -o $@
ggml-vulkan-matmul-shader:

7
ggml-vulkan-matmul.glsl
View file

@ -12,6 +12,7 @@
// See the License for the specific language governing permissions and
// limitations under the License.
// Original at https://github.com/google/uVkCompute/blob/f3180c7e72ae639c0a7bc8cff7ed615b63ced27c/benchmarks/mmt/mmt_i8.glsl
// Modified by 0cc4m for FP32
#version 450 core
@ -22,6 +23,12 @@
#extension GL_KHR_shader_subgroup_basic : enable
#define WG_X 32
#define WG_Y 2
#define M0 32
#define N0 256
#define K0 16
layout(binding = 0) buffer InputA { vec4 x[]; } inputA;
layout(binding = 1) buffer InputB { vec4 x[]; } inputB;
layout(binding = 2) buffer Output { float x[]; } outputO;

43
ggml-vulkan.cpp
View file

@ -19,7 +19,8 @@
vk::Instance instance;
vk::PhysicalDevice physical_device;
vk::Device device;
vk::Device vk_device;
vk::Pipeline vk_pipeline_matmul;
VmaAllocation vk_buffer_qa_alloc, vk_buffer_a_alloc, vk_buffer_b_alloc, vk_buffer_c_alloc;
vk::Buffer vk_buffer_qa, vk_buffer_a, vk_buffer_b, vk_buffer_c;
@ -48,9 +49,47 @@ void ggml_vk_init(void) {
const float queue_priority = 1.0f;
vk::DeviceQueueCreateInfo device_queue_create_info(vk::DeviceQueueCreateFlags(), compute_queue_family_index, 1, &queue_priority);
vk::DeviceCreateInfo device_create_info(vk::DeviceCreateFlags(), device_queue_create_info);
device = physical_device.createDevice(device_create_info);
vk_device = physical_device.createDevice(device_create_info);
std::vector<char> matmul_shader_contents;
if (std::ifstream shader_file{ "ggml-vulkan-matmul.spv", std::ios::binary | std::ios::ate }) {
const size_t file_size = shader_file.tellg();
shader_file.seekg(0);
matmul_shader_contents.resize(file_size, '\0');
shader_file.read(matmul_shader_contents.data(), file_size);
}
vk::ShaderModuleCreateInfo shader_module_create_info(
vk::ShaderModuleCreateFlags(),
matmul_shader_contents.size(),
reinterpret_cast<const uint32_t*>(matmul_shader_contents.data())
);
vk::ShaderModule shader_module = vk_device.createShaderModule(shader_module_create_info);
const std::vector<vk::DescriptorSetLayoutBinding> descriptor_set_layout_binding = {
{0, vk::DescriptorType::eStorageBuffer, 1, vk::ShaderStageFlagBits::eCompute},
{1, vk::DescriptorType::eStorageBuffer, 1, vk::ShaderStageFlagBits::eCompute},
{2, vk::DescriptorType::eStorageBuffer, 1, vk::ShaderStageFlagBits::eCompute}
};
vk::DescriptorSetLayoutCreateInfo descriptor_set_layout_create_info(
vk::DescriptorSetLayoutCreateFlags(),
descriptor_set_layout_binding);
vk::DescriptorSetLayout descriptor_set_layout = vk_device.createDescriptorSetLayout(descriptor_set_layout_create_info);
vk::PipelineLayoutCreateInfo pipeline_layout_create_info(vk::PipelineLayoutCreateFlags(), descriptor_set_layout);
vk::PipelineLayout pipeline_layout = vk_device.createPipelineLayout(pipeline_layout_create_info);
vk::PipelineCache pipeline_cache = vk_device.createPipelineCache(vk::PipelineCacheCreateInfo());
vk::PipelineShaderStageCreateInfo pipeline_shader_create_info(
vk::PipelineShaderStageCreateFlags(),
vk::ShaderStageFlagBits::eCompute,
shader_module,
"main");
vk::ComputePipelineCreateInfo compute_pipeline_create_info(
vk::PipelineCreateFlags(), // Flags
pipeline_shader_create_info, // Shader Create Info struct
pipeline_layout); // Pipeline Layout
vk_pipeline_matmul = vk_device.createComputePipeline(pipeline_cache, compute_pipeline_create_info).value;
}
// static void ggml_cl_malloc(size_t req_size, size_t* cur_size, cl_mem_flags flags, cl_mem* buf) {