vulkan: fix coopmat1 without bf16 support (#19793)

diff --git a/ggml/src/ggml-vulkan/ggml-vulkan.cpp b/ggml/src/ggml-vulkan/ggml-vulkan.cpp
index a8840a0773bacf5da83edb0a1470bebd2663cf94..88b3e4e58eb16f10e1f78fa6ed67c1b88ad4b5d4 100644 (file)
--- a/ggml/src/ggml-vulkan/ggml-vulkan.cpp
+++ b/ggml/src/ggml-vulkan/ggml-vulkan.cpp
@@ -3780,10 +3780,12 @@ static void ggml_vk_load_shaders(vk_device& device) {
         && !device->coopmat_bf16_support
 #endif
         ) {
+        const uint32_t s_warptile_wm = device->subgroup_size == 8 ? 8 : 32;
+
         // use scalar tile sizes
         l_warptile = { 128, 128, 128, 16, subgroup_size_8 * 2, 64, 2, 4, 4, 1, subgroup_size_8 };
         m_warptile = { 128,  64,  64, 16, subgroup_size_8, 32, 2, 4, 2, 1, subgroup_size_8 };
-        s_warptile = { subgroup_size_16, 32, 32, 16, 32, 32, 2, 2, 2, 1, subgroup_size_8 };
+        s_warptile = { subgroup_size_32, 32, 32, 16, s_warptile_wm, 32, 2, 2, 2, 1, subgroup_size_8 };
 
         l_wg_denoms = {128, 128, 1 };
         m_wg_denoms = { 64,  64, 1 };