vulkan: Increase workgroup size for GLU, for performance

Author: jeffbolznv

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

@@ -2761,8 +2761,8 @@ static void ggml_vk_load_shaders(vk_device& device) {
 #undef CREATE_UNARY
 
 #define CREATE_GLU(name)  \
-    ggml_vk_create_pipeline(device, device->pipeline_ ## name [0], #name "_f32", name ## _f32_len, name ## _f32_data, "main", 3, sizeof(vk_op_glu_push_constants), {1, 1, 1}, { device->subgroup_size }, 1);  \
-    ggml_vk_create_pipeline(device, device->pipeline_ ## name [1], #name "_f16", name ## _f16_len, name ## _f16_data, "main", 3, sizeof(vk_op_glu_push_constants), {1, 1, 1}, { device->subgroup_size }, 1);
+    ggml_vk_create_pipeline(device, device->pipeline_ ## name [0], #name "_f32", name ## _f32_len, name ## _f32_data, "main", 3, sizeof(vk_op_glu_push_constants), {1, 1, 1}, {}, 1);  \
+    ggml_vk_create_pipeline(device, device->pipeline_ ## name [1], #name "_f16", name ## _f16_len, name ## _f16_data, "main", 3, sizeof(vk_op_glu_push_constants), {1, 1, 1}, {}, 1);
 
     CREATE_GLU(geglu)
     CREATE_GLU(reglu)

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

@@ -1,12 +1,12 @@
 #extension GL_EXT_shader_16bit_storage : require
 
-layout(local_size_x_id = 0, local_size_y = 1, local_size_z = 1) in;
+layout(local_size_x = 512, local_size_y = 1, local_size_z = 1) in;
 
 layout (binding = 0) readonly buffer A {A_TYPE data_a[];};
 layout (binding = 1) readonly buffer B {A_TYPE data_b[];};
 layout (binding = 2) writeonly buffer D {D_TYPE data_d[];};
 
-layout (constant_id = 0) const uint BLOCK_SIZE = 32;
+const uint BLOCK_SIZE = 512;
 
 layout (push_constant) uniform parameter
 {