From: R0CKSTAR <redacted>
Date: Tue, 18 Mar 2025 18:28:26 +0000 (+0800)
Subject: musa: override warp_size of musa device to 32 (llama/12445)
X-Git-Tag: upstream/1.7.5~73
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=31b62276cfb47f005349f150e76375ff315e999f;p=pkg%2Fggml%2Fsources%2Fwhisper.cpp

musa: override warp_size of musa device to 32 (llama/12445)

Signed-off-by: Xiaodong Ye <redacted>
---

diff --git a/ggml/src/ggml-cuda/ggml-cuda.cu b/ggml/src/ggml-cuda/ggml-cuda.cu
index 8fb06382..5cb56df9 100644
--- a/ggml/src/ggml-cuda/ggml-cuda.cu
+++ b/ggml/src/ggml-cuda/ggml-cuda.cu
@@ -262,6 +262,8 @@ static ggml_cuda_device_info ggml_cuda_init() {
                       id, prop.name, prop.gcnArchName, info.devices[id].cc & 0xffff,
                       device_vmm ? "yes" : "no", prop.warpSize);
 #elif defined(GGML_USE_MUSA)
+        // FIXME: Ensure compatibility with varying warp sizes across different MUSA archs.
+        info.devices[id].warp_size = 32;
         // TODO: refine the .cc to reflect MUSA's actual CC capabilities
         info.devices[id].smpbo = prop.sharedMemPerBlockOptin;
         info.devices[id].cc = 100*prop.major + 10*prop.minor;