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

ggml: Add op l2_norm

Browse source 
Signed-off-by: Molly Sophia <mollysophia379@gmail.com>
This commit is contained in:
Molly Sophia 2025-01-15 20:42:40 +08:00
parent 19d3c8293b
commit 5445300758
16 changed files with 489 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

23
ggml/src/ggml-vulkan/ggml-vulkan.cpp
View file

@ -239,6 +239,7 @@ struct vk_device_struct {
vk_pipeline pipeline_norm_f32;
vk_pipeline pipeline_group_norm_f32;
vk_pipeline pipeline_rms_norm_f32;
vk_pipeline pipeline_l2_norm_f32;
vk_pipeline pipeline_gelu_f32;
vk_pipeline pipeline_gelu_quick_f32;
vk_pipeline pipeline_silu_f32;
@ -2069,6 +2070,7 @@ static void ggml_vk_load_shaders(vk_device& device) {
ggml_vk_create_pipeline(device, device->pipeline_norm_f32, "norm_f32", norm_f32_len, norm_f32_data, "main", 2, sizeof(vk_op_push_constants), {1, 1, 1}, {}, 1);
ggml_vk_create_pipeline(device, device->pipeline_group_norm_f32, "group_norm_f32", group_norm_f32_len, group_norm_f32_data, "main", 2, sizeof(vk_op_push_constants), {1, 1, 1}, {}, 1);
ggml_vk_create_pipeline(device, device->pipeline_rms_norm_f32, "rms_norm_f32", rms_norm_f32_len, rms_norm_f32_data, "main", 2, sizeof(vk_op_push_constants), {1, 1, 1}, {}, 1);
ggml_vk_create_pipeline(device, device->pipeline_l2_norm_f32, "l2_norm_f32", l2_norm_f32_len, l2_norm_f32_data, "main", 2, sizeof(vk_op_push_constants), {1, 1, 1}, {}, 1);
ggml_vk_create_pipeline(device, device->pipeline_cpy_f32_f32, "cpy_f32_f32", cpy_f32_f32_len, cpy_f32_f32_data, "main", 2, sizeof(vk_op_unary_push_constants), {512, 1, 1}, {}, 1);
ggml_vk_create_pipeline(device, device->pipeline_cpy_f32_f16, "cpy_f32_f16", cpy_f32_f16_len, cpy_f32_f16_data, "main", 2, sizeof(vk_op_unary_push_constants), {512, 1, 1}, {}, 1);
@ -5203,6 +5205,11 @@ static vk_pipeline ggml_vk_op_get_pipeline(ggml_backend_vk_context * ctx, const
return ctx->device->pipeline_rms_norm_f32;
}
return nullptr;
case GGML_OP_L2_NORM:
if (src0->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_F32) {
return ctx->device->pipeline_l2_norm_f32;
}
return nullptr;
case GGML_OP_UNARY:
switch (ggml_get_unary_op(dst)) {
case GGML_UNARY_OP_SILU:
@ -5542,6 +5549,7 @@ static void ggml_vk_op_f32(ggml_backend_vk_context * ctx, vk_context& subctx, co
switch (op) {
case GGML_OP_NORM:
case GGML_OP_RMS_NORM:
case GGML_OP_L2_NORM:
case GGML_OP_SOFT_MAX:
case GGML_OP_SUM_ROWS:
{
@ -6058,6 +6066,11 @@ static void ggml_vk_rms_norm(ggml_backend_vk_context * ctx, vk_context& subctx,
ggml_vk_op_f32<vk_op_push_constants>(ctx, subctx, src0, nullptr, nullptr, dst, GGML_OP_RMS_NORM, { (uint32_t)src0->ne[0], (uint32_t)src0->ne[1], op_params[0], 0.0f }, dryrun);
}
static void ggml_vk_l2_norm(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;
ggml_vk_op_f32<vk_op_push_constants>(ctx, subctx, src0, nullptr, nullptr, dst, GGML_OP_L2_NORM, { (uint32_t)src0->ne[0], (uint32_t)src0->ne[1], op_params[0], 0.0f }, dryrun);
}
static void ggml_vk_unary(ggml_backend_vk_context * ctx, vk_context& subctx, const ggml_tensor * src0, ggml_tensor * dst, bool dryrun = false) {
ggml_vk_op_f32<vk_op_push_constants>(ctx, subctx, src0, nullptr, nullptr, dst, GGML_OP_UNARY, { (uint32_t)ggml_nelements(src0), 0, 0.0f, 0.0f }, dryrun);
}
@ -7023,6 +7036,7 @@ static bool ggml_vk_build_graph(ggml_backend_vk_context * ctx, ggml_tensor * nod
case GGML_OP_NORM:
case GGML_OP_GROUP_NORM:
case GGML_OP_RMS_NORM:
case GGML_OP_L2_NORM:
case GGML_OP_DIAG_MASK_INF:
case GGML_OP_SOFT_MAX:
case GGML_OP_ROPE:
@ -7075,6 +7089,7 @@ static bool ggml_vk_build_graph(ggml_backend_vk_context * ctx, ggml_tensor * nod
case GGML_OP_NORM:
case GGML_OP_GROUP_NORM:
case GGML_OP_RMS_NORM:
case GGML_OP_L2_NORM:
case GGML_OP_UNARY:
case GGML_OP_DIAG_MASK_INF:
case GGML_OP_SOFT_MAX:
@ -7171,6 +7186,10 @@ static bool ggml_vk_build_graph(ggml_backend_vk_context * ctx, ggml_tensor * nod
case GGML_OP_RMS_NORM:
ggml_vk_rms_norm(ctx, compute_ctx, src0, node, dryrun);
break;
case GGML_OP_L2_NORM:
ggml_vk_l2_norm(ctx, compute_ctx, src0, node, dryrun);
break;
case GGML_OP_UNARY:
switch (ggml_get_unary_op(node)) {
@ -7305,6 +7324,7 @@ static bool ggml_vk_compute_forward(ggml_backend_vk_context * ctx, ggml_tensor *
case GGML_OP_NORM:
case GGML_OP_GROUP_NORM:
case GGML_OP_RMS_NORM:
case GGML_OP_L2_NORM:
case GGML_OP_DIAG_MASK_INF:
case GGML_OP_SOFT_MAX:
case GGML_OP_ROPE:
@ -8223,6 +8243,7 @@ static bool ggml_backend_vk_device_supports_op(ggml_backend_dev_t dev, const ggm
case GGML_OP_NORM:
case GGML_OP_GROUP_NORM:
case GGML_OP_RMS_NORM:
case GGML_OP_L2_NORM:
return ggml_is_contiguous(op->src[0]);
case GGML_OP_ADD:
case GGML_OP_ACC:
@ -8747,6 +8768,8 @@ static void ggml_vk_check_results_0(ggml_tensor * tensor) {
tensor_clone = ggml_group_norm(ggml_ctx, src0_clone, *(int *)tensor->op_params, ((float *)tensor->op_params)[1]);
} else if (tensor->op == GGML_OP_RMS_NORM) {
tensor_clone = ggml_rms_norm(ggml_ctx, src0_clone, *(float *)tensor->op_params);
} else if (tensor->op == GGML_OP_L2_NORM) {
tensor_clone = ggml_l2_norm(ggml_ctx, src0_clone, *(float *)tensor->op_params);
} else if (tensor->op == GGML_OP_SOFT_MAX) {
if (src1 != nullptr) {
tensor_clone = ggml_soft_max_ext(ggml_ctx, src0_clone, src1_clone, ((float *)tensor->op_params)[0], ((float *)tensor->op_params)[1]);

41
ggml/src/ggml-vulkan/vulkan-shaders/l2_norm.comp Normal file
View file

@ -0,0 +1,41 @@
#version 450
#include "generic_head.comp"
#include "types.comp"
#extension GL_EXT_control_flow_attributes : enable
#define BLOCK_SIZE 512
layout(local_size_x = BLOCK_SIZE, local_size_y = 1, local_size_z = 1) in;
layout (binding = 0) readonly buffer X {A_TYPE data_a[];};
layout (binding = 1) writeonly buffer D {D_TYPE data_d[];};
shared FLOAT_TYPE sum[BLOCK_SIZE];
void main() {
const uint row = gl_WorkGroupID.z * 262144 + gl_WorkGroupID.y * 512 + gl_WorkGroupID.x;
const uint tid = gl_LocalInvocationID.x;
sum[tid] = FLOAT_TYPE(0.0f); // partial sum for thread in warp
[[unroll]] for (uint col = tid; col < p.KX; col += BLOCK_SIZE) {
const FLOAT_TYPE xi = FLOAT_TYPE(data_a[row*p.KX + col]);
sum[tid] += xi * xi;
}
// sum up partial sums and write back result
barrier();
[[unroll]] for (int s = BLOCK_SIZE / 2; s > 0; s >>= 1) {
if (tid < s) {
sum[tid] += sum[tid + s];
}
barrier();
}
const FLOAT_TYPE scale = inversesqrt(max(sum[0], FLOAT_TYPE(p.param1)));
[[unroll]] for (uint col = tid; col < p.KX; col += BLOCK_SIZE) {
data_d[row*p.KX + col] = D_TYPE(scale * FLOAT_TYPE(data_a[row*p.KX + col]));
}
}

1
ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp
View file

@ -418,6 +418,7 @@ void process_shaders() {
string_to_spv("norm_f32", "norm.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float"}}));
string_to_spv("group_norm_f32", "group_norm.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float"}}));
string_to_spv("rms_norm_f32", "rms_norm.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float"}}));
string_to_spv("l2_norm_f32", "l2_norm.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float"}}));
string_to_spv("cpy_f32_f32", "copy.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}});
string_to_spv("cpy_f32_f16", "copy.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float16_t"}});