vulkan: Workaround compiler bug in conv2d coopmat2 path (llama/24924)

* vulkan: Workaround compiler bug in conv2d coopmat2 path

* apply same workaround to CONV_3D

* Apply suggestion from @jeffbolznv

diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/conv2d_mm.comp b/ggml/src/ggml-vulkan/vulkan-shaders/conv2d_mm.comp
index 1428ef68d81ca787f29c4ceae606b8a6118a715a..99400098bf2b708d59b414b1ec1749990d278911 100644 (file)
--- a/ggml/src/ggml-vulkan/vulkan-shaders/conv2d_mm.comp
+++ b/ggml/src/ggml-vulkan/vulkan-shaders/conv2d_mm.comp
@@ -158,7 +158,7 @@ const uint32_t Csh_stride = BS_NPQ;
 #ifdef COOPMAT
 const uint32_t Csh_len    = BS_K * Csh_stride;
 #else
-const uint32_t Csh_len    = csh_store != 0 ? BS_K * Csh_stride : 1;
+const uint32_t Csh_len    = csh_store != 0 ? BS_K * Csh_stride : 8; // 8 to workaround compiler bug
 #endif
 shared SHMEM_TYPE Csh[Csh_len];  // K x NPQ
 #endif

diff --git a/ggml/src/ggml-vulkan/vulkan-shaders/conv3d_mm.comp b/ggml/src/ggml-vulkan/vulkan-shaders/conv3d_mm.comp
index a9712eb3acfce3d95dc7af43ca9424393c6f0d52..f66f299f6dae4d1c7ae1139e8fb580292dae77d4 100644 (file)
--- a/ggml/src/ggml-vulkan/vulkan-shaders/conv3d_mm.comp
+++ b/ggml/src/ggml-vulkan/vulkan-shaders/conv3d_mm.comp
@@ -144,7 +144,7 @@ const uint32_t Csh_stride = BS_NPQ;
 #ifdef COOPMAT
 const uint32_t Csh_len    = BS_K * Csh_stride;
 #else
-const uint32_t Csh_len    = csh_store != 0 ? BS_K * Csh_stride : 1;
+const uint32_t Csh_len    = csh_store != 0 ? BS_K * Csh_stride : 8; // 8 to workaround compiler bug
 #endif
 shared SHMEM_TYPE Csh[Csh_len];  // K x NPQ
 #endif