* Copyright (c) 2025 Huawei Technologies Co., Ltd.
* This program is free software, you can redistribute it and/or modify it under the terms and conditions of
* CANN Open Software License Agreement Version 2.0 (the "License").
* Please refer to the License for details. You may not use this file except in compliance with the License.
* THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
* INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
* See LICENSE in the root of the software repository for the full text of the License.
*/
* \file bmm_tiling.cpp
* \brief
*/
#include <iostream>
#include <algorithm>
#include "math_util.h"
#include "../../detail/host_log.h"
#include "matmul_tiling_algorithm.h"
#include "include/adv_api/matmul/bmm_tiling.h"
using namespace std;
namespace matmul_tiling {
int32_t MultiCoreMatmulTiling::SetDim(int32_t dim)
{
this->blockDim = dim;
return 0;
}
int32_t MultiCoreMatmulTiling::SetSingleShape(int32_t singleMIn, int32_t singleNIn, int32_t singleKIn)
{
TILING_LOG_DEBUG(" Set SingleCore M: %d", singleMIn);
TILING_LOG_DEBUG(" Set SingleCore N: %d", singleNIn);
TILING_LOG_DEBUG(" Set SingleCore K: %d", singleKIn);
if (singleMIn > 0) {
this->singleCoreM = singleMIn;
}
if (singleNIn > 0) {
this->singleCoreN = singleNIn;
}
if (singleKIn > 0) {
this->singleCoreK = singleKIn;
}
return 0;
}
int32_t MultiCoreMatmulTiling::SetAlignSplit(int32_t alignM, int32_t alignN, int32_t alignK)
{
if (alignM > 0) {
this->alignSingleM = alignM;
}
if (alignN > 0) {
this->alignSingleN = alignN;
}
if (alignK > 0) {
this->alignSingleK = alignK;
}
return 0;
}
int32_t MultiCoreMatmulTiling::SetShape(int32_t m, int32_t n, int32_t k)
{
TILING_LOG_DEBUG(" Set Single M: %d", m);
TILING_LOG_DEBUG(" Set Single N: %d", n);
TILING_LOG_DEBUG(" Set Single K: %d", k);
this->singleM = m;
this->singleN = n;
this->singleK = k;
if (this->orgM == -1 && this->orgN == -1 && this->orgKa == -1 && this->orgKb == -1) {
this->orgM = m != -1 ? m : this->orgM;
this->orgN = n != -1 ? n : this->orgN;
this->orgKa = k != -1 ? k : this->orgKa;
this->orgKb = k != -1 ? k : this->orgKb;
}
return 0;
}
int64_t MultiCoreMatmulTiling::GetTiling(optiling::TCubeTiling& tiling)
{
int32_t ret = Compute();
if (ret == -1) {
TILING_LOG_INFO("Cannot deduce tiling params from given info.");
return ret;
}
SetFinalTiling(tiling);
PrintTilingDataInfo(tiling);
return ret;
}
int64_t MultiCoreMatmulTiling::GetTiling(AscendC::tiling::TCubeTiling& tiling)
{
optiling::TCubeTiling mmTiling;
int64_t ret = GetTiling(mmTiling);
mmTiling.SaveToBuffer(&tiling, sizeof(TCubeTiling));
return ret;
}
int64_t MultiCoreMatmulTiling::Compute()
{
if (!CheckSetParam()) {
TILING_LOG_INFO("Tiling compute params check don't pass.");
return -1;
}
MatmulTilingAlgorithm algoIns(this);
int ret = algoIns.Process();
return ret;
}
int32_t MultiCoreMatmulTiling::GetSingleShape(int32_t& shapeM, int32_t& shapeN, int32_t& shapeK)
{
shapeM = this->tiling_.get_singleCoreM();
shapeN = this->tiling_.get_singleCoreN();
shapeK = this->tiling_.get_singleCoreK();
return 0;
}
int32_t MultiCoreMatmulTiling::GetCoreNum(int32_t& dim, int32_t& mDim, int32_t& nDim)
{
dim = this->tiling_.get_usedCoreNum();
mDim = MathUtil::CeilDivision(this->singleM, this->tiling_.get_singleCoreM());
nDim = MathUtil::CeilDivision(this->singleN, this->tiling_.get_singleCoreN());
return 0;
}
bool MultiCoreMatmulTiling::EnableL1BankConflictOptimise()
{
MatmulTilingAlgorithm algoIns(this);
bool ret = algoIns.EnableL1BankConflictOptimise();
return ret;
}
int32_t BatchMatmulTiling::SetBatch(int32_t batchMIn, int32_t batchNIn)
{
this->batchM = batchMIn;
this->batchN = batchNIn;
return 0;
}
int32_t BatchMatmulTiling::SetSingleBatch(int32_t singleMIn, int32_t singleNIn)
{
if (singleMIn > 0) {
this->singleBatchM = singleMIn;
}
if (singleNIn > 0) {
this->singleBatchN = singleNIn;
}
return 0;
}
int32_t BatchMatmulTiling::GetCoreNum(
int32_t& dim, int32_t& mDim, int32_t& nDim, int32_t& batchCoreM, int32_t& batchCoreN)
{
dim = this->tiling_.get_usedCoreNum();
mDim = MathUtil::CeilDivision(this->tiling_.get_M(), this->tiling_.get_singleCoreM());
nDim = MathUtil::CeilDivision(this->tiling_.get_N(), this->tiling_.get_singleCoreN());
batchCoreM = MathUtil::CeilDivision(this->batchM, this->singleBatchM);
batchCoreN = MathUtil::CeilDivision(this->batchN, this->singleBatchN);
return 0;
}
int64_t BatchMatmulTiling::GetTiling(optiling::TCubeTiling& tiling)
{
int32_t ret = Compute();
if (ret == -1) {
TILING_LOG_INFO("Cannot deduce tiling params from given info.");
return -1;
}
SetFinalTiling(tiling);
PrintTilingDataInfo(tiling);
return 0;
}
int64_t BatchMatmulTiling::GetTiling(AscendC::tiling::TCubeTiling& tiling)
{
optiling::TCubeTiling mmTiling;
int64_t ret = GetTiling(mmTiling);
mmTiling.SaveToBuffer(&tiling, sizeof(TCubeTiling));
return ret;
}
int64_t BatchMatmulTiling::Compute()
{
if (!CheckSetParam()) {
TILING_LOG_INFO("Tiling compute params check don't pass.");
return -1;
}
MatmulTilingAlgorithm algoIns(this);
int ret = algoIns.Process();
return ret;
}
}
extern "C" {
int32_t MultiCoreMatmulGetTmpBufSize(optiling::TCubeTiling& tiling, matmul_tiling::SysTilingTempBufSize& bufSize)
{
bufSize.l1Size = max(bufSize.l1Size, tiling.get_shareL1Size());
bufSize.l0cSize = max(bufSize.l0cSize, tiling.get_shareL0CSize());
bufSize.ubSize = max(bufSize.ubSize, max(tiling.get_shareUbSize(), 2 * tiling.get_transLength()));
return 0;
}
int32_t BatchMatmulGetTmpBufSize(optiling::TCubeTiling& tiling, matmul_tiling::SysTilingTempBufSize& bufSize)
{
bufSize.l1Size = max(bufSize.l1Size, tiling.get_shareL1Size());
bufSize.l0cSize = max(bufSize.l0cSize, tiling.get_shareL0CSize());
bufSize.ubSize = max(bufSize.ubSize, max(tiling.get_shareUbSize(), 2 * tiling.get_transLength()));
return 0;
}
int32_t MultiCoreMatmulGetTmpBufSizeV2(
AscendC::tiling::TCubeTiling& tiling, matmul_tiling::SysTilingTempBufSize& bufSize)
{
bufSize.l1Size = max(bufSize.l1Size, tiling.shareL1Size);
bufSize.l0cSize = max(bufSize.l0cSize, tiling.shareL0CSize);
bufSize.ubSize = max(bufSize.ubSize, max(tiling.shareUbSize, 2 * tiling.transLength));
return 0;
}
int32_t BatchMatmulGetTmpBufSizeV2(AscendC::tiling::TCubeTiling& tiling, matmul_tiling::SysTilingTempBufSize& bufSize)
{
bufSize.l1Size = max(bufSize.l1Size, tiling.shareL1Size);
bufSize.l0cSize = max(bufSize.l0cSize, tiling.shareL0CSize);
bufSize.ubSize = max(bufSize.ubSize, max(tiling.shareUbSize, 2 * tiling.transLength));
return 0;
}
};
