From 7c8cf299255ed744c84752072e9f13d6beba9a05 Mon Sep 17 00:00:00 2001 From: 0cc4m Date: Wed, 31 Jan 2024 06:26:32 +0100 Subject: [PATCH] Fix small matrix multiplication errors in AMD GPUs on Windows or with amdvlk Co-authored-by: Engininja2 <139037756+Engininja2@users.noreply.github.com> --- ggml-vulkan.cpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/ggml-vulkan.cpp b/ggml-vulkan.cpp index 9532e242b..bccc40bf5 100644 --- a/ggml-vulkan.cpp +++ b/ggml-vulkan.cpp @@ -817,7 +817,7 @@ static void ggml_vk_load_shaders() { // mulmat std::initializer_list warptile_l = { 128, 128, 128, 16, vk_device.subgroup_size * 2, 64, 2, 4, 4, vk_device.subgroup_size }; std::initializer_list warptile_m = { 128, 64, 64, 16, vk_device.subgroup_size, 32, 2, 4, 2, vk_device.subgroup_size }; - std::initializer_list warptile_s = { vk_device.subgroup_size, 32, 32, 8, 32, 32, 2, 2, 2, vk_device.subgroup_size }; + std::initializer_list warptile_s = { vk_device.subgroup_size, 32, 32, 16, 32, 32, 2, 2, 2, vk_device.subgroup_size }; std::array l_wg_denoms = {128, 128, 1 }; std::array m_wg_denoms = { 64, 64, 1 };