From 9af9e80163ce549f0ac508fb379af1e751108f7a Mon Sep 17 00:00:00 2001 From: Eve <139727413+netrunnereve@users.noreply.github.com> Date: Sun, 8 Dec 2024 17:26:56 -0500 Subject: [PATCH] use subgroup min and max to check for gcn (requires https://github.com/ggerganov/llama.cpp/pull/10721) --- ggml/src/ggml-vulkan/ggml-vulkan.cpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/ggml/src/ggml-vulkan/ggml-vulkan.cpp b/ggml/src/ggml-vulkan/ggml-vulkan.cpp index 387dd8901..2202b39ef 100644 --- a/ggml/src/ggml-vulkan/ggml-vulkan.cpp +++ b/ggml/src/ggml-vulkan/ggml-vulkan.cpp @@ -1812,7 +1812,7 @@ static void ggml_vk_load_shaders(vk_device& device) { // AMD GCN graphics cards perform best when the number of rows per shader is doubled uint32_t rm = 1; - if ((device->vendor_id == VK_VENDOR_ID_AMD) && (device->subgroup_size == 64)) + if ((device->subgroup_size_control) && (device->vendor_id == VK_VENDOR_ID_AMD) && (device->subgroup_min_size == 64) && (device->subgroup_max_size == 64)) rm = 2; // computing additional rows per workgroup is a benefit for Q4_0 -> Q5_1, but not for Q8_0.