fixed the perchannel block

luoyu-intel · Nov 15, 2023 · fea0280 · fea0280
1 parent 501153d
commit fea0280
Showing 1 changed file with 1 addition and 0 deletions.
diff --git a/onnxruntime/test/mlas/bench/bench_q4gemm.cpp b/onnxruntime/test/mlas/bench/bench_q4gemm.cpp
@@ -62,6 +62,7 @@ void Q4GEMM_Jblas(benchmark::State& state, int block_size, bool is_asym, MLAS_CO
   const size_t N = static_cast<size_t>(state.range(1));
   const size_t K = static_cast<size_t>(state.range(2));
   const size_t threads = static_cast<size_t>(state.range(3));
+  block_size = block_size == -1 ? K : block_size;
   const size_t pack_b_size = MlasNBitsGemmPackBSize(N, K, block_size, 4, is_asym, cmp_type);
 
   OrtThreadPoolParams tpo;