ggml-vulkan: Add ELU op support (#20183)
* ggml-Vulkan: add ELU support * ggml-Vulkan: remove extra spaces and variables * ggml-Vulkan: fix format issue * ggml-Vulkan: fix format issue * fix whitespace issue * Update Vulkan.csv and ops.md
This commit is contained in:
GiantPrince
GitHub
parent
cd18a50ea5
commit
d088d5b74f
14
docs/ops.md
14
docs/ops.md
|
|
@ -37,11 +37,11 @@ Legend:
|
|||
| CROSS_ENTROPY_LOSS | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
|
||||
| CROSS_ENTROPY_LOSS_BACK | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
|
||||
| CUMSUM | ❌ | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ |
|
||||
| DIAG | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
|
||||
| DIAG | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ |
|
||||
| DIAG_MASK_INF | ❌ | ✅ | ✅ | ✅ | ❌ | 🟡 | ✅ | ✅ | ❌ | ❌ | ❌ |
|
||||
| DIV | ❌ | ✅ | ✅ | ✅ | 🟡 | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ |
|
||||
| DUP | ❌ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | ✅ | ✅ | ❌ | ❌ | ❌ |
|
||||
| ELU | ❌ | ✅ | ✅ | 🟡 | 🟡 | ❌ | ✅ | ❌ | ✅ | ❌ | ❌ |
|
||||
| ELU | ❌ | ✅ | ✅ | 🟡 | 🟡 | ❌ | ✅ | 🟡 | ✅ | ❌ | ❌ |
|
||||
| EXP | ❌ | ✅ | ✅ | 🟡 | 🟡 | ❌ | ✅ | 🟡 | ✅ | ❌ | ❌ |
|
||||
| EXPM1 | ❌ | ❌ | ✅ | 🟡 | 🟡 | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ |
|
||||
| FILL | ❌ | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ |
|
||||
|
|
@ -54,7 +54,7 @@ Legend:
|
|||
| GELU | ❌ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | ✅ | 🟡 | ✅ | ❌ | ❌ |
|
||||
| GELU_ERF | ❌ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | ✅ | 🟡 | ✅ | ❌ | ❌ |
|
||||
| GELU_QUICK | ❌ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | ✅ | 🟡 | ✅ | ❌ | ❌ |
|
||||
| GET_ROWS | ❌ | 🟡 | ✅ | 🟡 | ✅ | 🟡 | 🟡 | 🟡 | 🟡 | ❌ | ❌ |
|
||||
| GET_ROWS | ❌ | 🟡 | ✅ | 🟡 | ✅ | 🟡 | 🟡 | ✅ | 🟡 | ❌ | ❌ |
|
||||
| GET_ROWS_BACK | ❌ | ❌ | 🟡 | 🟡 | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
|
||||
| GROUP_NORM | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ | ❌ |
|
||||
| HARDSIGMOID | ❌ | ✅ | ✅ | 🟡 | 🟡 | ❌ | ✅ | 🟡 | ✅ | ❌ | ❌ |
|
||||
|
|
@ -90,7 +90,7 @@ Legend:
|
|||
| RWKV_WKV6 | ❌ | ❌ | ✅ | ✅ | ✅ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
|
||||
| RWKV_WKV7 | ❌ | ❌ | ✅ | ✅ | ✅ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
|
||||
| SCALE | ❌ | 🟡 | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ |
|
||||
| SET | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | 🟡 | ❌ | ❌ | ❌ | ❌ |
|
||||
| SET | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | 🟡 | ✅ | ❌ | ❌ | ❌ |
|
||||
| SET_ROWS | ❌ | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | 🟡 | ❌ | ❌ |
|
||||
| SGN | ❌ | ✅ | ✅ | 🟡 | 🟡 | ❌ | ✅ | ❌ | ✅ | ❌ | ❌ |
|
||||
| SIGMOID | ❌ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | ✅ | 🟡 | ✅ | ❌ | ❌ |
|
||||
|
|
@ -100,7 +100,7 @@ Legend:
|
|||
| SOFTPLUS | ❌ | ❌ | ✅ | 🟡 | 🟡 | ❌ | ✅ | 🟡 | ✅ | ❌ | ❌ |
|
||||
| SOFT_MAX | ❌ | 🟡 | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ |
|
||||
| SOFT_MAX_BACK | ❌ | ❌ | 🟡 | 🟡 | ❌ | ❌ | 🟡 | ✅ | ❌ | ❌ | ❌ |
|
||||
| SOLVE_TRI | ❌ | ❌ | ✅ | 🟡 | ❌ | ❌ | ❌ | 🟡 | ❌ | ❌ | ❌ |
|
||||
| SOLVE_TRI | ❌ | ❌ | ✅ | 🟡 | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ |
|
||||
| SQR | ❌ | ✅ | ✅ | ✅ | 🟡 | ✅ | ✅ | 🟡 | ✅ | ❌ | ❌ |
|
||||
| SQRT | ❌ | ✅ | ✅ | ✅ | 🟡 | ✅ | ✅ | 🟡 | ✅ | ❌ | ❌ |
|
||||
| SSM_CONV | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ | ❌ |
|
||||
|
|
@ -116,5 +116,5 @@ Legend:
|
|||
| TOP_K | ❌ | ❌ | ✅ | ❌ | ✅ | ❌ | 🟡 | 🟡 | ✅ | ❌ | ❌ |
|
||||
| TRI | ❌ | ❌ | ✅ | ✅ | ✅ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
|
||||
| TRUNC | ❌ | ❌ | ✅ | 🟡 | ❌ | ❌ | 🟡 | 🟡 | ✅ | ❌ | ❌ |
|
||||
| UPSCALE | ❌ | 🟡 | ✅ | ✅ | 🟡 | 🟡 | 🟡 | 🟡 | ❌ | ❌ | ❌ |
|
||||
| XIELU | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ |
|
||||
| UPSCALE | ❌ | 🟡 | ✅ | ✅ | 🟡 | 🟡 | 🟡 | ✅ | ❌ | ❌ | ❌ |
|
||||
| XIELU | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ |
|
||||
|
|
|
|||
9146
docs/ops/Vulkan.csv
9146
docs/ops/Vulkan.csv
File diff suppressed because it is too large
Load Diff
|
|
@ -744,6 +744,7 @@ struct vk_device_struct {
|
|||
|
||||
// [src/dst 0=fp32,1=fp16]
|
||||
vk_pipeline pipeline_exp[2];
|
||||
vk_pipeline pipeline_elu[2];
|
||||
vk_pipeline pipeline_gelu[2];
|
||||
vk_pipeline pipeline_gelu_erf[2];
|
||||
vk_pipeline pipeline_gelu_quick[2];
|
||||
|
|
@ -4373,6 +4374,7 @@ static void ggml_vk_load_shaders(vk_device& device) {
|
|||
ggml_vk_create_pipeline(device, device->pipeline_ ## name [0], #name "_f32", name ## _f32_len, name ## _f32_data, "main", 2, sizeof(vk_op_push_constants), {512, 1, 1}, {}, 1); \
|
||||
ggml_vk_create_pipeline(device, device->pipeline_ ## name [1], #name "_f16", name ## _f16_len, name ## _f16_data, "main", 2, sizeof(vk_op_push_constants), {512, 1, 1}, {}, 1);
|
||||
|
||||
CREATE_UNARY(elu)
|
||||
CREATE_UNARY(gelu)
|
||||
CREATE_UNARY(gelu_erf)
|
||||
CREATE_UNARY(gelu_quick)
|
||||
|
|
@ -9241,6 +9243,8 @@ static vk_pipeline ggml_vk_op_get_pipeline(ggml_backend_vk_context * ctx, const
|
|||
switch (ggml_get_unary_op(dst)) {
|
||||
case GGML_UNARY_OP_EXP:
|
||||
return ctx->device->pipeline_exp[dst->type == GGML_TYPE_F16];
|
||||
case GGML_UNARY_OP_ELU:
|
||||
return ctx->device->pipeline_elu[dst->type == GGML_TYPE_F16];
|
||||
case GGML_UNARY_OP_SILU:
|
||||
return ctx->device->pipeline_silu[dst->type == GGML_TYPE_F16];
|
||||
case GGML_UNARY_OP_GELU:
|
||||
|
|
@ -12852,6 +12856,7 @@ static bool ggml_vk_build_graph(ggml_backend_vk_context * ctx, ggml_cgraph * cgr
|
|||
}
|
||||
|
||||
switch (ggml_get_unary_op(node)) {
|
||||
case GGML_UNARY_OP_ELU:
|
||||
case GGML_UNARY_OP_EXP:
|
||||
case GGML_UNARY_OP_SILU:
|
||||
case GGML_UNARY_OP_GELU:
|
||||
|
|
@ -14951,6 +14956,7 @@ static bool ggml_backend_vk_device_supports_op(ggml_backend_dev_t dev, const ggm
|
|||
case GGML_OP_UNARY:
|
||||
switch (ggml_get_unary_op(op)) {
|
||||
case GGML_UNARY_OP_EXP:
|
||||
case GGML_UNARY_OP_ELU:
|
||||
case GGML_UNARY_OP_GELU:
|
||||
case GGML_UNARY_OP_GELU_ERF:
|
||||
case GGML_UNARY_OP_GELU_QUICK:
|
||||
|
|
@ -16074,6 +16080,9 @@ static void ggml_vk_check_results_0(ggml_backend_vk_context * ctx, ggml_cgraph *
|
|||
case GGML_UNARY_OP_EXP:
|
||||
tensor_clone = ggml_exp(ggml_ctx, src_clone[0]);
|
||||
break;
|
||||
case GGML_UNARY_OP_ELU:
|
||||
tensor_clone = ggml_elu(ggml_ctx, src_clone[0]);
|
||||
break;
|
||||
case GGML_UNARY_OP_SILU:
|
||||
tensor_clone = ggml_silu(ggml_ctx, src_clone[0]);
|
||||
break;
|
||||
|
|
|
|||
|
|
@ -0,0 +1,27 @@
|
|||
#version 450
|
||||
|
||||
#include "generic_head.glsl"
|
||||
#include "types.glsl"
|
||||
|
||||
#extension GL_EXT_control_flow_attributes : enable
|
||||
|
||||
layout(local_size_x = 512, 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[];};
|
||||
|
||||
void main() {
|
||||
const uint i = gl_GlobalInvocationID.z * 262144 + gl_GlobalInvocationID.y * 512 + gl_GlobalInvocationID.x;
|
||||
|
||||
if (i >= p.KX) {
|
||||
return;
|
||||
}
|
||||
|
||||
float x = float(data_a[i]);
|
||||
|
||||
if (x < 0.0f) {
|
||||
x = exp(x) - 1;
|
||||
}
|
||||
|
||||
data_d[i] = D_TYPE(x);
|
||||
}
|
||||
|
|
@ -867,6 +867,8 @@ void process_shaders() {
|
|||
string_to_spv("hardswish_f32", "hardswish.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}});
|
||||
string_to_spv("abs_f16", "abs.comp", {{"A_TYPE", "float16_t"}, {"D_TYPE", "float16_t"}});
|
||||
string_to_spv("abs_f32", "abs.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}});
|
||||
string_to_spv("elu_f16", "elu.comp", {{"A_TYPE", "float16_t"}, {"D_TYPE", "float16_t"}});
|
||||
string_to_spv("elu_f32", "elu.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}});
|
||||
string_to_spv("xielu_f16", "xielu.comp", {{"A_TYPE", "float16_t"}, {"D_TYPE", "float16_t"}});
|
||||
string_to_spv("xielu_f32", "xielu.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}});
|
||||
|
||||
|
|
|
|||
Loading…
Reference in New Issue