From b01b4fa7353d28dea553667e56619138f78addef Mon Sep 17 00:00:00 2001 From: Jeremy Gu <145739220+wgu9@users.noreply.github.com> Date: Wed, 24 Jun 2026 18:51:13 -0700 Subject: [PATCH] cuda : sanitize invalid Blackwell smpbo values --- ggml/src/ggml-cuda/ggml-cuda.cu | 5 +++++ 1 file changed, 5 insertions(+) diff --git a/ggml/src/ggml-cuda/ggml-cuda.cu b/ggml/src/ggml-cuda/ggml-cuda.cu index cca70592f807..48d5c0bc7768 100644 --- a/ggml/src/ggml-cuda/ggml-cuda.cu +++ b/ggml/src/ggml-cuda/ggml-cuda.cu @@ -289,6 +289,11 @@ static ggml_cuda_device_info ggml_cuda_init() { (size_t)(prop.totalGlobalMem / (1024 * 1024))); #else info.devices[id].smpbo = prop.sharedMemPerBlockOptin; + if (info.devices[id].smpbo == 0 || info.devices[id].smpbo > prop.sharedMemPerMultiprocessor) { + GGML_LOG_WARN("%s: device %d reported invalid sharedMemPerBlockOptin=%zu, falling back to sharedMemPerBlock=%zu\n", + __func__, id, info.devices[id].smpbo, (size_t) prop.sharedMemPerBlock); + info.devices[id].smpbo = prop.sharedMemPerBlock; + } info.devices[id].cc = 100*prop.major + 10*prop.minor; GGML_LOG_INFO(" Device %d: %s, compute capability %d.%d, VMM: %s, VRAM: %zu MiB\n", id, prop.name, prop.major, prop.minor, device_vmm ? "yes" : "no",