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

Avoid requesting dedicated memory, VMA can decide that by itself

Browse source
This commit is contained in:
0cc4m 2023-06-30 21:20:19 +02:00
parent 4ea9b2fd4b
commit 36cd5d85e9
2 changed files with 19 additions and 18 deletions
Download patch file Download diff file Expand all files Collapse all files

1
Makefile
View file

@ -223,6 +223,7 @@ ggml-vulkan.o: ggml-vulkan.cpp ggml-vulkan.h
glslc -fshader-stage=compute --target-env=vulkan1.2 -O vk_shaders/matmul_f32.glsl -o vk_shaders/matmul_f32.spv glslc -fshader-stage=compute --target-env=vulkan1.2 -O vk_shaders/matmul_f32.glsl -o vk_shaders/matmul_f32.spv
glslc -fshader-stage=compute --target-env=vulkan1.2 -O vk_shaders/matmul_f16.glsl -o vk_shaders/matmul_f16.spv glslc -fshader-stage=compute --target-env=vulkan1.2 -O vk_shaders/matmul_f16.glsl -o vk_shaders/matmul_f16.spv
glslc -fshader-stage=compute --target-env=vulkan1.2 -O vk_shaders/f16_to_f32.glsl -o vk_shaders/f16_to_f32.spv glslc -fshader-stage=compute --target-env=vulkan1.2 -O vk_shaders/f16_to_f32.glsl -o vk_shaders/f16_to_f32.spv
glslc -fshader-stage=compute --target-env=vulkan1.2 -O vk_shaders/dequant_q4_0.glsl -o vk_shaders/dequant_q4_0.spv
endif endif
ifneq ($(filter aarch64%,$(UNAME_M)),) ifneq ($(filter aarch64%,$(UNAME_M)),)

36
ggml-vulkan.cpp
View file

@ -751,10 +751,10 @@ static void ggml_vk_mul_mat_f32(const ggml_tensor * src0, const ggml_tensor * sr
if (src0->backend == GGML_BACKEND_GPU) { if (src0->backend == GGML_BACKEND_GPU) {
d_X = *(vk_buffer*) src0->data; d_X = *(vk_buffer*) src0->data;
} else { } else {
ggml_vk_pool_malloc(ggml_type_size(src0->type) * x_ne, &d_X, VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT); ggml_vk_pool_malloc(ggml_type_size(src0->type) * x_ne, &d_X, 0);
} }
ggml_vk_pool_malloc(sizeof(float) * y_ne, &d_Y, VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT); ggml_vk_pool_malloc(sizeof(float) * y_ne, &d_Y, 0);
ggml_vk_pool_malloc(sizeof(float) * d_ne, &d_D, VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT); ggml_vk_pool_malloc(sizeof(float) * d_ne, &d_D, 0);
vk::Fence fence = vk_device.createFence(vk::FenceCreateInfo()); vk::Fence fence = vk_device.createFence(vk::FenceCreateInfo());
@ -833,10 +833,10 @@ static void ggml_vk_mul_mat_f16(const ggml_tensor * src0, const ggml_tensor * sr
if (src0->backend == GGML_BACKEND_GPU) { if (src0->backend == GGML_BACKEND_GPU) {
d_X = *(vk_buffer*) src0->data; d_X = *(vk_buffer*) src0->data;
} else { } else {
ggml_vk_pool_malloc(sizeof(ggml_fp16_t) * x_ne, &d_X, VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT); ggml_vk_pool_malloc(sizeof(ggml_fp16_t) * x_ne, &d_X, 0);
} }
ggml_vk_pool_malloc(sizeof(ggml_fp16_t) * y_ne, &d_Y, VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT); ggml_vk_pool_malloc(sizeof(ggml_fp16_t) * y_ne, &d_Y, 0);
ggml_vk_pool_malloc(sizeof(float) * d_ne, &d_D, VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT); ggml_vk_pool_malloc(sizeof(float) * d_ne, &d_D, 0);
bool src1_cont_rows = nb10 == sizeof(float); bool src1_cont_rows = nb10 == sizeof(float);
bool src1_cont_cols = (size_t)nb11 == ne11*sizeof(float); bool src1_cont_cols = (size_t)nb11 == ne11*sizeof(float);
@ -931,13 +931,13 @@ static void ggml_vk_mul_mat_q_f32(const ggml_tensor * src0, const ggml_tensor *
vk_buffer d_Y; vk_buffer d_Y;
vk_buffer d_D; vk_buffer d_D;
if (!mul_mat_vec) { if (!mul_mat_vec) {
ggml_vk_pool_malloc(sizeof(float) * x_ne, &d_X, VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT); ggml_vk_pool_malloc(sizeof(float) * x_ne, &d_X, 0);
} }
ggml_vk_pool_malloc(sizeof(float) * y_ne, &d_Y, VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT); ggml_vk_pool_malloc(sizeof(float) * y_ne, &d_Y, 0);
ggml_vk_pool_malloc(sizeof(float) * d_ne, &d_D, VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT); ggml_vk_pool_malloc(sizeof(float) * d_ne, &d_D, 0);
vk_buffer d_Q; vk_buffer d_Q;
if (src0->backend == GGML_BACKEND_CPU) { if (src0->backend == GGML_BACKEND_CPU) {
ggml_vk_pool_malloc(q_sz, &d_Q, VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT); ggml_vk_pool_malloc(q_sz, &d_Q, 0);
} }
vk_pipeline* to_fp32_vk = ggml_get_to_fp32_vk(type); vk_pipeline* to_fp32_vk = ggml_get_to_fp32_vk(type);
@ -1091,9 +1091,9 @@ void ggml_vk_test_matmul_f32(size_t m, size_t n, size_t k) {
vk_buffer d_X; vk_buffer d_X;
vk_buffer d_Y; vk_buffer d_Y;
vk_buffer d_D; vk_buffer d_D;
ggml_vk_pool_malloc(sizeof(float) * x_ne, &d_X, VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT); ggml_vk_pool_malloc(sizeof(float) * x_ne, &d_X, 0);
ggml_vk_pool_malloc(sizeof(float) * y_ne, &d_Y, VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT); ggml_vk_pool_malloc(sizeof(float) * y_ne, &d_Y, 0);
ggml_vk_pool_malloc(sizeof(float) * d_ne, &d_D, VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT); ggml_vk_pool_malloc(sizeof(float) * d_ne, &d_D, 0);
float* x = (float *) malloc(sizeof(float) * x_ne); float* x = (float *) malloc(sizeof(float) * x_ne);
float* y = (float *) malloc(sizeof(float) * y_ne); float* y = (float *) malloc(sizeof(float) * y_ne);
@ -1167,9 +1167,9 @@ void ggml_vk_test_matmul_f16(size_t m, size_t n, size_t k) {
vk_buffer d_X; vk_buffer d_X;
vk_buffer d_Y; vk_buffer d_Y;
vk_buffer d_D; vk_buffer d_D;
ggml_vk_pool_malloc(sizeof(ggml_fp16_t) * x_ne, &d_X, VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT); ggml_vk_pool_malloc(sizeof(ggml_fp16_t) * x_ne, &d_X, 0);
ggml_vk_pool_malloc(sizeof(ggml_fp16_t) * y_ne, &d_Y, VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT); ggml_vk_pool_malloc(sizeof(ggml_fp16_t) * y_ne, &d_Y, 0);
ggml_vk_pool_malloc(sizeof(float) * d_ne, &d_D, VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT); ggml_vk_pool_malloc(sizeof(float) * d_ne, &d_D, 0);
ggml_fp16_t* x = (ggml_fp16_t *) malloc(sizeof(ggml_fp16_t) * x_ne); ggml_fp16_t* x = (ggml_fp16_t *) malloc(sizeof(ggml_fp16_t) * x_ne);
ggml_fp16_t* y = (ggml_fp16_t *) malloc(sizeof(ggml_fp16_t) * y_ne); ggml_fp16_t* y = (ggml_fp16_t *) malloc(sizeof(ggml_fp16_t) * y_ne);
@ -1241,8 +1241,8 @@ void ggml_vk_test_matmul_f16(size_t m, size_t n, size_t k) {
void ggml_vk_test_f16_to_f32(size_t m) { void ggml_vk_test_f16_to_f32(size_t m) {
vk_buffer d_X; vk_buffer d_X;
vk_buffer d_D; vk_buffer d_D;
ggml_vk_pool_malloc(sizeof(ggml_fp16_t) * m, &d_X, VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT); ggml_vk_pool_malloc(sizeof(ggml_fp16_t) * m, &d_X, 0);
ggml_vk_pool_malloc(sizeof(float) * m, &d_D, VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT); ggml_vk_pool_malloc(sizeof(float) * m, &d_D, 0);
ggml_fp16_t* x = (ggml_fp16_t *) malloc(sizeof(ggml_fp16_t) * m); ggml_fp16_t* x = (ggml_fp16_t *) malloc(sizeof(ggml_fp16_t) * m);
float* d = (float *) malloc(sizeof(float) * m); float* d = (float *) malloc(sizeof(float) * m);