vulkan: implement SOFTPLUS

Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>

Author: giuseppe

ggml/src/ggml-vulkan/ggml-vulkan.cpp

@@ -664,6 +664,7 @@ struct vk_device_struct {
     vk_pipeline pipeline_hardsigmoid[2];
     vk_pipeline pipeline_hardswish[2];
     vk_pipeline pipeline_abs[2];
+    vk_pipeline pipeline_softplus[2];
 
     vk_pipeline pipeline_add1_f16_f16;
     vk_pipeline pipeline_add1_f16_f32;
@@ -3834,6 +3835,7 @@ static void ggml_vk_load_shaders(vk_device& device) {
     CREATE_UNARY(hardsigmoid)
     CREATE_UNARY(hardswish)
     CREATE_UNARY(abs)
+    CREATE_UNARY(softplus)
 #undef CREATE_UNARY
 
 #define CREATE_UNARY_RTE(name)  \
@@ -8258,6 +8260,8 @@ static vk_pipeline ggml_vk_op_get_pipeline(ggml_backend_vk_context * ctx, const
                 return ctx->device->pipeline_hardswish[dst->type == GGML_TYPE_F16];
             case GGML_UNARY_OP_ABS:
                 return ctx->device->pipeline_abs[dst->type == GGML_TYPE_F16];
+            case GGML_UNARY_OP_SOFTPLUS:
+                return ctx->device->pipeline_softplus[dst->type == GGML_TYPE_F16];
             default:
                 break;
         }
@@ -11279,6 +11283,7 @@ static bool ggml_vk_build_graph(ggml_backend_vk_context * ctx, ggml_cgraph * cgr
         case GGML_UNARY_OP_HARDSIGMOID:
         case GGML_UNARY_OP_HARDSWISH:
         case GGML_UNARY_OP_ABS:
+        case GGML_UNARY_OP_SOFTPLUS:
             break;
         default:
             return false;
@@ -11631,6 +11636,7 @@ static bool ggml_vk_build_graph(ggml_backend_vk_context * ctx, ggml_cgraph * cgr
         case GGML_UNARY_OP_HARDSIGMOID:
         case GGML_UNARY_OP_HARDSWISH:
         case GGML_UNARY_OP_ABS:
+        case GGML_UNARY_OP_SOFTPLUS:
             ggml_vk_unary(ctx, compute_ctx, src0, node);
             break;
         default:
@@ -11907,6 +11913,7 @@ static bool ggml_vk_compute_forward(ggml_backend_vk_context * ctx, ggml_cgraph *
         case GGML_UNARY_OP_HARDSIGMOID:
         case GGML_UNARY_OP_HARDSWISH:
         case GGML_UNARY_OP_ABS:
+        case GGML_UNARY_OP_SOFTPLUS:
             buf = tensor->buffer;
             break;
         default:
@@ -13509,6 +13516,7 @@ static bool ggml_backend_vk_device_supports_op(ggml_backend_dev_t dev, const ggm
                 case GGML_UNARY_OP_HARDSIGMOID:
                 case GGML_UNARY_OP_HARDSWISH:
                 case GGML_UNARY_OP_ABS:
+                case GGML_UNARY_OP_SOFTPLUS:
                     return ggml_is_contiguous(op->src[0]) &&
                            (op->src[0]->type == GGML_TYPE_F32 || op->src[0]->type == GGML_TYPE_F16) &&
                            (op->type == GGML_TYPE_F32 || op->type == GGML_TYPE_F16) &&
@@ -14422,6 +14430,9 @@ static void ggml_vk_check_results_0(ggml_backend_vk_context * ctx, ggml_cgraph *
             case GGML_UNARY_OP_ABS:
                 tensor_clone = ggml_abs(ggml_ctx, src_clone[0]);
                 break;
+            case GGML_UNARY_OP_SOFTPLUS:
+                tensor_clone = ggml_softplus(ggml_ctx, src_clone[0]);
+                break;
             default:
                 std::cerr << "Missing vk_check_results OP: " << ggml_op_name(tensor->op) << std::endl;
                 GGML_ABORT("fatal error");

ggml/src/ggml-vulkan/vulkan-shaders/softplus.comp

@@ -0,0 +1,23 @@
+#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;
+    }
+
+    const float x = float(data_a[i]);
+    const float result = (x > 20.0f) ? x : log(1.0f + exp(x));
+    data_d[i] = D_TYPE(result);
+}

ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp

@@ -842,6 +842,10 @@ 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("softplus_f16",   "softplus.comp",    {{"A_TYPE", "float16_t"},   {"D_TYPE", "float16_t"}});
+    string_to_spv("softplus_f32",   "softplus.comp",    {{"A_TYPE", "float"},       {"D_TYPE", "float"}});
+
     string_to_spv("add1_f16_f16",   "add1.comp",        {{"A_TYPE", "float16_t"},   {"B_TYPE", "float16_t"}, {"D_TYPE", "float16_t"}, {"FLOAT_TYPE", "float"}});
     string_to_spv("add1_f16_f32",   "add1.comp",        {{"A_TYPE", "float16_t"},   {"B_TYPE", "float"}, {"D_TYPE", "float16_t"}, {"FLOAT_TYPE", "float"}});
     string_to_spv("add1_f32_f32",   "add1.comp",        {{"A_TYPE", "float"},       {"B_TYPE", "float"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}});