Force subgroup 32 on RX 5700 and subgroup 64 for im2col

Author: daniandtheweb

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

@@ -1543,11 +1543,18 @@ static void ggml_vk_load_shaders(vk_device& device) {
         device->pipeline_matmul_id_f32 = std::make_shared<vk_matmul_pipeline_struct>();
     }
 
+    vk::PhysicalDeviceProperties2 props2;
+    device->physical_device.getProperties2(&props2);
+    std::string device_name = props2.properties.deviceName.data();
     std::vector<std::future<void>> compiles;
     auto const &ggml_vk_create_pipeline = [&](vk_device& device, vk_pipeline& pipeline, const std::string &name, size_t spv_size, const void* spv_data, const std::string &entrypoint,
                                               uint32_t parameter_count, uint32_t push_constant_size, std::array<uint32_t, 3> wg_denoms, const std::vector<uint32_t>& specialization_constants,
                                               uint32_t align, bool disable_robustness = false, bool require_full_subgroups = false, uint32_t required_subgroup_size = 0) {
 
+        if (required_subgroup_size == 0) {
+            required_subgroup_size = (device_name.find("RX 5700") != std::string::npos) ? 32 : required_subgroup_size;
+        }
+
         if (!pipeline) {
             pipeline = std::make_shared<vk_pipeline_struct>();
             pipeline->name = name;
@@ -1573,6 +1580,20 @@ static void ggml_vk_load_shaders(vk_device& device) {
                                       parameter_count, wg_denoms, specialization_constants, disable_robustness, require_full_subgroups, required_subgroup_size));
     };
 
+    // New lambda for pipelines with subgroup size 64.
+    auto const &ggml_vk_create_pipeline_64 = [&](vk_device& device, vk_pipeline& pipeline,
+        const std::string &name, size_t spv_size, const void* spv_data,
+        const std::string &entrypoint, uint32_t parameter_count,
+        uint32_t push_constant_size, std::array<uint32_t, 3> wg_denoms,
+        const std::vector<uint32_t>& specialization_constants, uint32_t align,
+        bool disable_robustness = false, bool require_full_subgroups = false)
+    {
+        ggml_vk_create_pipeline(device, pipeline, name, spv_size, spv_data, entrypoint,
+                                  parameter_count, push_constant_size, wg_denoms,
+                                  specialization_constants, align, disable_robustness,
+                                  require_full_subgroups, 64);
+    };
+
 #if defined(VK_NV_cooperative_matrix2) && defined(GGML_VULKAN_COOPMAT2_GLSLC_SUPPORT)
     if (device->coopmat2) {
 
@@ -2151,11 +2172,21 @@ static void ggml_vk_load_shaders(vk_device& device) {
 
     ggml_vk_create_pipeline(device, device->pipeline_sum_rows_f32, "sum_rows_f32", sum_rows_f32_len, sum_rows_f32_data, "main", 2, sizeof(vk_op_push_constants), {1, 1, 1}, { device->subgroup_size }, 1);
 
-    ggml_vk_create_pipeline(device, device->pipeline_im2col_f32, "im2col_f32", im2col_f32_len, im2col_f32_data, "main", 2, sizeof(vk_op_im2col_push_constants), {512, 1, 1}, { device->subgroup_size }, 1, true);
-    if (device->float_controls_rte_fp16) {
-        ggml_vk_create_pipeline(device, device->pipeline_im2col_f32_f16, "im2col_f32_f16", im2col_f32_f16_rte_len, im2col_f32_f16_rte_data, "main", 2, sizeof(vk_op_im2col_push_constants), {512, 1, 1}, { device->subgroup_size }, 1, true);
+    // Workaround needed to speedup im2col on RX 5700
+    if (device_name.find("RX 5700")  != std::string::npos) {
+        ggml_vk_create_pipeline_64(device, device->pipeline_im2col_f32, "im2col_f32", im2col_f32_len, im2col_f32_data, "main", 2, sizeof(vk_op_im2col_push_constants), {512, 1, 1}, { device->subgroup_size }, 1, true);
+        if (device->float_controls_rte_fp16) {
+            ggml_vk_create_pipeline_64(device, device->pipeline_im2col_f32_f16, "im2col_f32_f16", im2col_f32_f16_rte_len, im2col_f32_f16_rte_data, "main", 2, sizeof(vk_op_im2col_push_constants), {512, 1, 1}, { device->subgroup_size }, 1, true);
+        } else {
+            ggml_vk_create_pipeline_64(device, device->pipeline_im2col_f32_f16, "im2col_f32_f16", im2col_f32_f16_len, im2col_f32_f16_data, "main", 2, sizeof(vk_op_im2col_push_constants), {512, 1, 1}, { device->subgroup_size }, 1, true);
+        }
     } else {
-        ggml_vk_create_pipeline(device, device->pipeline_im2col_f32_f16, "im2col_f32_f16", im2col_f32_f16_len, im2col_f32_f16_data, "main", 2, sizeof(vk_op_im2col_push_constants), {512, 1, 1}, { device->subgroup_size }, 1, true);
+        ggml_vk_create_pipeline(device, device->pipeline_im2col_f32, "im2col_f32", im2col_f32_len, im2col_f32_data, "main", 2, sizeof(vk_op_im2col_push_constants), {512, 1, 1}, { device->subgroup_size }, 1, true);
+        if (device->float_controls_rte_fp16) {
+            ggml_vk_create_pipeline(device, device->pipeline_im2col_f32_f16, "im2col_f32_f16", im2col_f32_f16_rte_len, im2col_f32_f16_rte_data, "main", 2, sizeof(vk_op_im2col_push_constants), {512, 1, 1}, { device->subgroup_size }, 1, true);
+        } else {
+            ggml_vk_create_pipeline(device, device->pipeline_im2col_f32_f16, "im2col_f32_f16", im2col_f32_f16_len, im2col_f32_f16_data, "main", 2, sizeof(vk_op_im2col_push_constants), {512, 1, 1}, { device->subgroup_size }, 1, true);
+        }
     }
 
     ggml_vk_create_pipeline(device, device->pipeline_timestep_embedding_f32, "timestep_embedding_f32", timestep_embedding_f32_len, timestep_embedding_f32_data, "main", 2, sizeof(vk_op_timestep_embedding_push_constants), {256, 1, 1}, {}, 1);