x64: matmul: blocking: update for 4k LDA

src/cpu/x64/matmul/brgemm_matmul_utils.cpp

@@ -1,5 +1,5 @@
 /*******************************************************************************
-* Copyright 2021-2023 Intel Corporation
+* Copyright 2021-2024 Intel Corporation
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
@@ -1477,7 +1477,9 @@ float matmul_amx_blocking_params_t::get_thread_balance_scores() {
 // returns score for current blocking parameters' values in range [0, 1]
 // for copied data reusage
 float matmul_amx_blocking_params_t::get_copied_data_reusage_scores() {
-    const dim_t effective_m_chunk_sz = 64 * 4;
+    // Values based on measured performance
+    const bool is_lda_4k = (current_lda_ * a_dt_sz) % 4096 == 0;
+    const dim_t effective_m_chunk_sz = 64 * (is_lda_4k ? 1 : 4);
     const dim_t desired_M_chunk_size = is_runtime_M
             ? effective_m_chunk_sz
             : nstl::min(M, effective_m_chunk_sz);