CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-30 08:42:00 +00:00
Code Issues Packages Projects Releases Wiki Activity

sync : ggml

Browse Source
This commit is contained in:
Georgi Gerganov
2024-08-27 22:01:45 +03:00
parent 3246fe84d7
commit 231cff5f6f
21 changed files with 1422 additions and 178 deletions
Download Patch File Download Diff File Expand all files Collapse all files

62
ggml/src/ggml-vulkan.cpp
View File

@@ -188,6 +188,8 @@ struct vk_device_struct {
vk_pipeline pipeline_upscale_f32;
vk_pipeline pipeline_scale_f32;
vk_pipeline pipeline_sqr_f32;
vk_pipeline pipeline_sin_f32;
vk_pipeline pipeline_cos_f32;
vk_pipeline pipeline_clamp_f32;
vk_pipeline pipeline_pad_f32;
vk_pipeline pipeline_repeat_f32;
@@ -1702,6 +1704,8 @@ static void ggml_vk_load_shaders(vk_device& device) {
ggml_vk_create_pipeline(device, device->pipeline_scale_f32, "scale_f32", scale_f32_len, scale_f32_data, "main", 2, sizeof(vk_op_unary_push_constants), {512, 1, 1}, {}, 1);
ggml_vk_create_pipeline(device, device->pipeline_sqr_f32, "sqr_f32", sqr_f32_len, sqr_f32_data, "main", 2, sizeof(vk_op_unary_push_constants), {512, 1, 1}, {}, 1);
ggml_vk_create_pipeline(device, device->pipeline_sin_f32, "sin_f32", sin_f32_len, sin_f32_data, "main", 2, sizeof(vk_op_unary_push_constants), {512, 1, 1}, {}, 1);
ggml_vk_create_pipeline(device, device->pipeline_cos_f32, "cos_f32", cos_f32_len, cos_f32_data, "main", 2, sizeof(vk_op_unary_push_constants), {512, 1, 1}, {}, 1);
ggml_vk_create_pipeline(device, device->pipeline_clamp_f32, "clamp_f32", clamp_f32_len, clamp_f32_data, "main", 2, sizeof(vk_op_unary_push_constants), {512, 1, 1}, {}, 1);
@@ -4023,6 +4027,16 @@ static vk_pipeline ggml_vk_op_get_pipeline(ggml_backend_vk_context * ctx, const
return ctx->device->pipeline_sqr_f32;
}
return nullptr;
case GGML_OP_SIN:
if (src0->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_F32) {
return ctx->device->pipeline_sin_f32;
}
return nullptr;
case GGML_OP_COS:
if (src0->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_F32) {
return ctx->device->pipeline_cos_f32;
}
return nullptr;
case GGML_OP_CLAMP:
if (src0->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_F32) {
return ctx->device->pipeline_clamp_f32;
@@ -4171,6 +4185,8 @@ static bool ggml_vk_op_supports_incontiguous(ggml_op op) {
case GGML_OP_UPSCALE:
case GGML_OP_SCALE:
case GGML_OP_SQR:
case GGML_OP_SIN:
case GGML_OP_COS:
case GGML_OP_CLAMP:
case GGML_OP_PAD:
case GGML_OP_REPEAT:
@@ -4381,6 +4397,8 @@ static void ggml_vk_op_f32(ggml_backend_vk_context * ctx, vk_context& subctx, co
case GGML_OP_MUL:
case GGML_OP_SCALE:
case GGML_OP_SQR:
case GGML_OP_SIN:
case GGML_OP_COS:
case GGML_OP_CLAMP:
case GGML_OP_PAD:
case GGML_OP_REPEAT:
@@ -4598,6 +4616,32 @@ static void ggml_vk_sqr(ggml_backend_vk_context * ctx, vk_context& subctx, const
}, dryrun);
}
static void ggml_vk_sin(ggml_backend_vk_context * ctx, vk_context& subctx, const ggml_tensor * src0, ggml_tensor * dst) {
const uint32_t src0_type_size = ggml_type_size(src0->type);
const uint32_t dst_type_size = ggml_type_size(dst->type);
ggml_vk_op_f32<vk_op_unary_push_constants>(ctx, subctx, src0, nullptr, nullptr, dst, GGML_OP_SIN, {
(uint32_t)ggml_nelements(src0),
(uint32_t)src0->ne[0], (uint32_t)src0->ne[1], (uint32_t)src0->ne[2], (uint32_t)src0->ne[3], (uint32_t)src0->nb[0] / src0_type_size, (uint32_t)src0->nb[1] / src0_type_size, (uint32_t)src0->nb[2] / src0_type_size, (uint32_t)src0->nb[3] / src0_type_size,
(uint32_t) dst->ne[0], (uint32_t) dst->ne[1], (uint32_t) dst->ne[2], (uint32_t) dst->ne[3], (uint32_t) dst->nb[0] / dst_type_size, (uint32_t) dst->nb[1] / dst_type_size, (uint32_t) dst->nb[2] / dst_type_size, (uint32_t) dst->nb[3] / dst_type_size,
0,
0.0f, 0.0f,
});
}
static void ggml_vk_cos(ggml_backend_vk_context * ctx, vk_context& subctx, const ggml_tensor * src0, ggml_tensor * dst) {
const uint32_t src0_type_size = ggml_type_size(src0->type);
const uint32_t dst_type_size = ggml_type_size(dst->type);
ggml_vk_op_f32<vk_op_unary_push_constants>(ctx, subctx, src0, nullptr, nullptr, dst, GGML_OP_COS, {
(uint32_t)ggml_nelements(src0),
(uint32_t)src0->ne[0], (uint32_t)src0->ne[1], (uint32_t)src0->ne[2], (uint32_t)src0->ne[3], (uint32_t)src0->nb[0] / src0_type_size, (uint32_t)src0->nb[1] / src0_type_size, (uint32_t)src0->nb[2] / src0_type_size, (uint32_t)src0->nb[3] / src0_type_size,
(uint32_t) dst->ne[0], (uint32_t) dst->ne[1], (uint32_t) dst->ne[2], (uint32_t) dst->ne[3], (uint32_t) dst->nb[0] / dst_type_size, (uint32_t) dst->nb[1] / dst_type_size, (uint32_t) dst->nb[2] / dst_type_size, (uint32_t) dst->nb[3] / dst_type_size,
0,
0.0f, 0.0f,
});
}
static void ggml_vk_clamp(ggml_backend_vk_context * ctx, vk_context& subctx, const ggml_tensor * src0, ggml_tensor * dst, bool dryrun = false) {
float * op_params = (float *)dst->op_params;
const uint32_t src0_type_size = ggml_type_size(src0->type);
@@ -5658,6 +5702,8 @@ static void ggml_vk_build_graph(ggml_backend_vk_context * ctx, ggml_tensor * nod
case GGML_OP_UPSCALE:
case GGML_OP_SCALE:
case GGML_OP_SQR:
case GGML_OP_SIN:
case GGML_OP_COS:
case GGML_OP_CLAMP:
case GGML_OP_PAD:
case GGML_OP_CPY:
@@ -5735,6 +5781,14 @@ static void ggml_vk_build_graph(ggml_backend_vk_context * ctx, ggml_tensor * nod
case GGML_OP_SQR:
ggml_vk_sqr(ctx, compute_ctx, src0, node, dryrun);
break;
case GGML_OP_SIN:
ggml_vk_sin(ctx, compute_ctx, src0, node);
break;
case GGML_OP_COS:
ggml_vk_cos(ctx, compute_ctx, src0, node);
break;
case GGML_OP_CLAMP:
ggml_vk_clamp(ctx, compute_ctx, src0, node, dryrun);
@@ -5851,6 +5905,8 @@ static bool ggml_vk_compute_forward(ggml_backend_vk_context * ctx, ggml_tensor *
case GGML_OP_UPSCALE:
case GGML_OP_SCALE:
case GGML_OP_SQR:
case GGML_OP_SIN:
case GGML_OP_COS:
case GGML_OP_CLAMP:
case GGML_OP_PAD:
case GGML_OP_CPY:
@@ -6582,6 +6638,8 @@ GGML_CALL static bool ggml_backend_vk_supports_op(ggml_backend_t backend, const
case GGML_OP_UPSCALE:
case GGML_OP_SCALE:
case GGML_OP_SQR:
case GGML_OP_SIN:
case GGML_OP_COS:
case GGML_OP_CLAMP:
case GGML_OP_PAD:
case GGML_OP_CONT:
@@ -7024,6 +7082,10 @@ static void ggml_vk_check_results_0(ggml_tensor * tensor) {
tensor_clone = ggml_scale(ggml_ctx, src0_clone, ((float *)tensor->op_params)[0]);
} else if (tensor->op == GGML_OP_SQR) {
tensor_clone = ggml_sqr(ggml_ctx, src0_clone);
} else if (tensor->op == GGML_OP_SIN) {
tensor_clone = ggml_sin(ggml_ctx, src0_clone);
} else if (tensor->op == GGML_OP_COS) {
tensor_clone = ggml_cos(ggml_ctx, src0_clone);
} else if (tensor->op == GGML_OP_CLAMP) {
tensor_clone = ggml_clamp(ggml_ctx, src0_clone, ((float *)tensor->op_params)[0], ((float *)tensor->op_params)[1]);
} else if (tensor->op == GGML_OP_PAD) {