ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-04-30 19:31:48 +00:00
Code Issues Packages Projects Releases Wiki Activity

Debug #733

Browse Source
This commit is contained in:
Iwan Kawrakow
2025-10-07 17:53:56 +03:00
parent f649e36a61
commit 51486dc3d4
1 changed files with 10 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
ggml/src/ggml-cuda/mmq.cuh
View File

@@ -4251,8 +4251,17 @@ void mul_mat_q_case(ggml_backend_cuda_context & ctx, const mmq_args & args, cuda
launch_mul_mat_q<type, 128>(ctx, args, stream); launch_mul_mat_q<type, 128>(ctx, args, stream);
break; break;
default: default:
fprintf(stderr, "mmq_x_best=%d\n", mmq_x_best); {
fprintf(stderr, "type=%s, mmq_x_best=%d\n", ggml_type_name(type), mmq_x_best);
fprintf(stderr, " id = %d, nsm = %d, cc = %d, smpbo = %d\n", id, nsm, cc, smpbo);
fprintf(stderr, " mmq_x_max = %d, mmq_y = %d, use_stream_k = %d\n", mmq_x_max, mmq_y, use_stream_k);
for (int mmq_x = 8; mmq_x <= mmq_x_max && nparts_best > 1; mmq_x += 8) {
const int granularity = mmq_get_granularity_host(mmq_x, cc);
const int shmem = mmq_get_shmem<type>(mmq_x, mmq_y, cc);
fprintf(stderr, " mmq_x = %d, granularity = %d, shmem = %d\n", mmq_x, granularity, shmem);
}
GGML_ABORT("fatal error"); GGML_ABORT("fatal error");
}
break; break;
} }
} }