vulkan: add exp operation (llama/15456)

Co-authored-by: aeseulgi <redacted>

diff --git a/src/ggml-vulkan/ggml-vulkan.cpp b/src/ggml-vulkan/ggml-vulkan.cpp
index a5bb1820b3896912cf03b0e2825aed7d02ec29f2..2d5254dfd2a387cfa12a38076f791fe5fa8f425b 100644 (file)
--- a/src/ggml-vulkan/ggml-vulkan.cpp
+++ b/src/ggml-vulkan/ggml-vulkan.cpp
@@ -490,6 +490,7 @@ struct vk_device_struct {
     vk_pipeline pipeline_l2_norm_f32;
 
     // [src/dst 0=fp32,1=fp16]
+    vk_pipeline pipeline_exp[2];
     vk_pipeline pipeline_gelu[2];
     vk_pipeline pipeline_gelu_erf[2];
     vk_pipeline pipeline_gelu_quick[2];
@@ -3066,6 +3067,7 @@ 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(exp)
     CREATE_UNARY(gelu)
     CREATE_UNARY(gelu_erf)
     CREATE_UNARY(gelu_quick)
@@ -7133,6 +7135,8 @@ 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_SILU:
                 return ctx->device->pipeline_silu[dst->type == GGML_TYPE_F16];
             case GGML_UNARY_OP_GELU:
@@ -9738,6 +9742,7 @@ static bool ggml_vk_build_graph(ggml_backend_vk_context * ctx, ggml_cgraph * cgr
         return false;
     case GGML_OP_UNARY:
         switch (ggml_get_unary_op(node)) {
+        case GGML_UNARY_OP_EXP:
         case GGML_UNARY_OP_SILU:
         case GGML_UNARY_OP_GELU:
         case GGML_UNARY_OP_GELU_ERF:
@@ -10015,6 +10020,7 @@ static bool ggml_vk_build_graph(ggml_backend_vk_context * ctx, ggml_cgraph * cgr
         break;
     case GGML_OP_UNARY:
         switch (ggml_get_unary_op(node)) {
+        case GGML_UNARY_OP_EXP:
         case GGML_UNARY_OP_SILU:
         case GGML_UNARY_OP_GELU:
         case GGML_UNARY_OP_GELU_ERF:
@@ -10251,6 +10257,7 @@ static bool ggml_vk_compute_forward(ggml_backend_vk_context * ctx, ggml_cgraph *
         break;
     case GGML_OP_UNARY:
         switch (ggml_get_unary_op(tensor)) {
+        case GGML_UNARY_OP_EXP:
         case GGML_UNARY_OP_SILU:
         case GGML_UNARY_OP_GELU:
         case GGML_UNARY_OP_GELU_ERF:
@@ -11166,6 +11173,7 @@ static bool ggml_backend_vk_device_supports_op(ggml_backend_dev_t dev, const ggm
     switch (op->op) {
         case GGML_OP_UNARY:
             switch (ggml_get_unary_op(op)) {
+                case GGML_UNARY_OP_EXP:
                 case GGML_UNARY_OP_GELU:
                 case GGML_UNARY_OP_GELU_ERF:
                 case GGML_UNARY_OP_GELU_QUICK:
@@ -11965,6 +11973,9 @@ static void ggml_vk_check_results_0(ggml_backend_vk_context * ctx, ggml_cgraph *
         }
     } else if (tensor->op == GGML_OP_UNARY) {
         switch (ggml_get_unary_op(tensor)) {
+        case GGML_UNARY_OP_EXP:
+            tensor_clone = ggml_exp(ggml_ctx, src_clone[0]);
+            break;
         case GGML_UNARY_OP_SILU:
             tensor_clone = ggml_silu(ggml_ctx, src_clone[0]);
             break;

diff --git a/src/ggml-vulkan/vulkan-shaders/exp.comp b/src/ggml-vulkan/vulkan-shaders/exp.comp
new file mode 100644 (file)
index 0000000..abecd2d
--- /dev/null
+++ b/src/ggml-vulkan/vulkan-shaders/exp.comp
@@ -0,0 +1,20 @@
+#version 450
+
+#include "generic_head.comp"
+#include "types.comp"
+
+#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;
+    }
+    data_d[i] = D_TYPE(exp(float(data_a[i])));
+}

diff --git a/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp b/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp
index 75c572d6fc904e25a726f12de8998b14e3d65cda..4e0bffe6da698460067df6038bcd6e9cd3b9ad60 100644 (file)
--- a/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp
+++ b/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp
@@ -586,6 +586,8 @@ void process_shaders() {
 
     string_to_spv("upscale_f32", "upscale.comp", {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}});
 
+    string_to_spv("exp_f16",        "exp.comp",         {{"A_TYPE", "float16_t"},   {"D_TYPE", "float16_t"}});
+    string_to_spv("exp_f32",        "exp.comp",         {{"A_TYPE", "float"},       {"D_TYPE", "float"}});
     string_to_spv("gelu_f16",       "gelu.comp",        {{"A_TYPE", "float16_t"},   {"D_TYPE", "float16_t"}});
     string_to_spv("gelu_f32",       "gelu.comp",        {{"A_TYPE", "float"},       {"D_TYPE", "float"}});
     string_to_spv("gelu_erf_f16",   "gelu_erf.comp",    {{"A_TYPE", "float16_t"},   {"D_TYPE", "float16_t"}});