cpu: x64: matmul: disable parallel_k_reduction for int8

Author: xuxinzen

src/cpu/x64/matmul/brgemm_matmul_utils.cpp

@@ -892,6 +892,7 @@ float compute_blocking_heuristic_avx512(brgemm_matmul_conf_t &bgmmc,
         // Parallelize across K for shapes with big 'K' dimension
         bool bwd_w_par_k_blk = bgmmc.batch == 1
                 && bm_conf_utils.check_is_transposed(bgmmc.src_tag)
+                && !bm_conf_utils.is_int8()
                 && IMPLICATION(bm_conf_utils.is_bf16(), math::is_pow2(matmul.K))
                 && matmul.K >= 2048;
         if (bwd_w_par_k_blk) {