* 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.
*/
#include "impl/kernel/matmul_allreduce.h"
#include "impl/kernel/allgather_matmul.h"
#include "impl/kernel/matmul_reduce_scatter.h"
#include "impl/kernel/allgather_matmul_with_gather_result.h"
#include "impl/kernel/matmul_reduce_scatter_padding_ab.h"
#include "impl/kernel/matmul_reduce_scatter_padding_a.h"
#include "impl/kernel/matmul_reduce_scatter_padding_b.h"
#include "impl/kernel/allgather_matmul_padding.h"
using namespace AscendC;
using ElementA = bfloat16_t;
using ElementB = bfloat16_t;
using ElementC = bfloat16_t;
using LayoutA0 = Catlass::layout::RowMajor;
using LayoutB0 = Catlass::layout::RowMajor;
using LayoutA1 = Catlass::layout::ColumnMajor;
using LayoutB1 = Catlass::layout::ColumnMajor;
using LayoutC = Catlass::layout::RowMajor;
void LaunchMatmulAllReduceBF16(
void *stream, uint64_t fftsAddr,
uint8_t *a, uint8_t *b, uint8_t *c,
uint8_t *gatherA, uint8_t *workspace,
uint8_t *symmetricPtr, CocTilingParams& cocTiling,
uint32_t transA, uint32_t transB)
{
(void)gatherA;
(void)workspace;
if (!transA && !transB) {
MatmulAllReduce<ElementA, LayoutA0, ElementB, LayoutB0, ElementC, LayoutC>
<<<BLOCK_NUM, nullptr, stream>>>(fftsAddr, a, b, c, symmetricPtr, cocTiling);
} else if (!transA && transB) {
MatmulAllReduce<ElementA, LayoutA0, ElementB, LayoutB1, ElementC, LayoutC>
<<<BLOCK_NUM, nullptr, stream>>>(fftsAddr, a, b, c, symmetricPtr, cocTiling);
} else if (transA && !transB) {
MatmulAllReduce<ElementA, LayoutA1, ElementB, LayoutB0, ElementC, LayoutC>
<<<BLOCK_NUM, nullptr, stream>>>(fftsAddr, a, b, c, symmetricPtr, cocTiling);
} else {
MatmulAllReduce<ElementA, LayoutA1, ElementB, LayoutB1, ElementC, LayoutC>
<<<BLOCK_NUM, nullptr, stream>>>(fftsAddr, a, b, c, symmetricPtr, cocTiling);
}
}
void LaunchAllGatherMatmulBF16(
void *stream, uint64_t fftsAddr,
uint8_t *a, uint8_t *b, uint8_t *c,
uint8_t *gatherA, uint8_t *workspace,
uint8_t *symmetricPtr, CocTilingParams& cocTiling,
uint32_t transA, uint32_t transB)
{
(void)gatherA;
(void)workspace;
if (!transA && !transB) {
AllGatherMatmul<ElementA, LayoutA0, ElementB, LayoutB0, ElementC, LayoutC>
<<<BLOCK_NUM, nullptr, stream>>>(fftsAddr, a, b, c, symmetricPtr, cocTiling);
} else if (!transA && transB) {
AllGatherMatmul<ElementA, LayoutA0, ElementB, LayoutB1, ElementC, LayoutC>
<<<BLOCK_NUM, nullptr, stream>>>(fftsAddr, a, b, c, symmetricPtr, cocTiling);
}
}
void LaunchMatmulReduceScatterBF16(
void *stream, uint64_t fftsAddr,
uint8_t *a, uint8_t *b, uint8_t *c,
uint8_t *gatherA, uint8_t *workspace,
uint8_t *symmetricPtr, CocTilingParams& cocTiling,
uint32_t transA, uint32_t transB)
{
(void)gatherA;
(void)workspace;
if (!transA && !transB) {
MatmulReduceScatter<ElementA, LayoutA0, ElementB, LayoutB0, ElementC, LayoutC>
<<<BLOCK_NUM, nullptr, stream>>>(fftsAddr, a, b, c, symmetricPtr, cocTiling);
} else if (!transA && transB) {
MatmulReduceScatter<ElementA, LayoutA0, ElementB, LayoutB1, ElementC, LayoutC>
<<<BLOCK_NUM, nullptr, stream>>>(fftsAddr, a, b, c, symmetricPtr, cocTiling);
} else if (transA && !transB) {
MatmulReduceScatter<ElementA, LayoutA1, ElementB, LayoutB0, ElementC, LayoutC>
<<<BLOCK_NUM, nullptr, stream>>>(fftsAddr, a, b, c, symmetricPtr, cocTiling);
} else {
MatmulReduceScatter<ElementA, LayoutA1, ElementB, LayoutB1, ElementC, LayoutC>
<<<BLOCK_NUM, nullptr, stream>>>(fftsAddr, a, b, c, symmetricPtr, cocTiling);
}
}
void LaunchAllGatherMatmulWithGatherResultBF16(
void *stream, uint64_t fftsAddr,
uint8_t *a, uint8_t *b, uint8_t *c,
uint8_t *gatherA, uint8_t *workspace,
uint8_t *symmetricPtr, CocTilingParams& cocTiling,
uint32_t transA, uint32_t transB)
{
(void)workspace;
if (!transA && !transB) {
AllGatherMatmulWithGatherResult<ElementA, LayoutA0, ElementB, LayoutB0, ElementC, LayoutC>
<<<BLOCK_NUM, nullptr, stream>>>(fftsAddr, a, b, c, gatherA, symmetricPtr, cocTiling);
} else if (!transA && transB) {
AllGatherMatmulWithGatherResult<ElementA, LayoutA0, ElementB, LayoutB1, ElementC, LayoutC>
<<<BLOCK_NUM, nullptr, stream>>>(fftsAddr, a, b, c, gatherA, symmetricPtr, cocTiling);
}
}
void LaunchMatmulReduceScatterPaddingABBF16(
void *stream, uint64_t fftsAddr,
uint8_t *a, uint8_t *b, uint8_t *c,
uint8_t *aWorkSpace, uint8_t *bWorkSpace,
uint8_t *symmetricPtr, CocTilingParams& cocTiling,
uint32_t transA, uint32_t transB)
{
if (!transA && !transB) {
MatmulReduceScatterPaddingAB<ElementA, LayoutA0, ElementB, LayoutB0, ElementC, LayoutC>
<<<BLOCK_NUM, nullptr, stream>>>(fftsAddr, a, b, c, aWorkSpace, bWorkSpace, symmetricPtr, cocTiling);
} else if (!transA && transB) {
MatmulReduceScatterPaddingAB<ElementA, LayoutA0, ElementB, LayoutB1, ElementC, LayoutC>
<<<BLOCK_NUM, nullptr, stream>>>(fftsAddr, a, b, c, aWorkSpace, bWorkSpace, symmetricPtr, cocTiling);
} else if (transA && !transB) {
MatmulReduceScatterPaddingAB<ElementA, LayoutA1, ElementB, LayoutB0, ElementC, LayoutC>
<<<BLOCK_NUM, nullptr, stream>>>(fftsAddr, a, b, c, aWorkSpace, bWorkSpace, symmetricPtr, cocTiling);
} else {
MatmulReduceScatterPaddingAB<ElementA, LayoutA1, ElementB, LayoutB1, ElementC, LayoutC>
<<<BLOCK_NUM, nullptr, stream>>>(fftsAddr, a, b, c, aWorkSpace, bWorkSpace, symmetricPtr, cocTiling);
}
}
void LaunchMatmulReduceScatterPaddingABF16(
void *stream, uint64_t fftsAddr,
uint8_t *a, uint8_t *b, uint8_t *c,
uint8_t *aWorkSpace, uint8_t *bWorkSpace,
uint8_t *symmetricPtr, CocTilingParams& cocTiling,
uint32_t transA, uint32_t transB)
{
if (!transA && !transB) {
MatmulReduceScatterPaddingA<ElementA, LayoutA0, ElementB, LayoutB0, ElementC, LayoutC>
<<<BLOCK_NUM, nullptr, stream>>>(fftsAddr, a, b, c, aWorkSpace, bWorkSpace, symmetricPtr, cocTiling);
} else if (!transA && transB) {
MatmulReduceScatterPaddingA<ElementA, LayoutA0, ElementB, LayoutB1, ElementC, LayoutC>
<<<BLOCK_NUM, nullptr, stream>>>(fftsAddr, a, b, c, aWorkSpace, bWorkSpace, symmetricPtr, cocTiling);
} else if (transA && !transB) {
MatmulReduceScatterPaddingA<ElementA, LayoutA1, ElementB, LayoutB0, ElementC, LayoutC>
<<<BLOCK_NUM, nullptr, stream>>>(fftsAddr, a, b, c, aWorkSpace, bWorkSpace, symmetricPtr, cocTiling);
} else {
MatmulReduceScatterPaddingA<ElementA, LayoutA1, ElementB, LayoutB1, ElementC, LayoutC>
<<<BLOCK_NUM, nullptr, stream>>>(fftsAddr, a, b, c, aWorkSpace, bWorkSpace, symmetricPtr, cocTiling);
}
}
void LaunchMatmulReduceScatterPaddingBBF16(
void *stream, uint64_t fftsAddr,
uint8_t *a, uint8_t *b, uint8_t *c,
uint8_t *aWorkSpace, uint8_t *bWorkSpace,
uint8_t *symmetricPtr, CocTilingParams& cocTiling,
uint32_t transA, uint32_t transB)
{
if (!transA && !transB) {
MatmulReduceScatterPaddingB<ElementA, LayoutA0, ElementB, LayoutB0, ElementC, LayoutC>
<<<BLOCK_NUM, nullptr, stream>>>(fftsAddr, a, b, c, aWorkSpace, bWorkSpace, symmetricPtr, cocTiling);
} else if (!transA && transB) {
MatmulReduceScatterPaddingB<ElementA, LayoutA0, ElementB, LayoutB1, ElementC, LayoutC>
<<<BLOCK_NUM, nullptr, stream>>>(fftsAddr, a, b, c, aWorkSpace, bWorkSpace, symmetricPtr, cocTiling);
} else if (transA && !transB) {
MatmulReduceScatterPaddingB<ElementA, LayoutA1, ElementB, LayoutB0, ElementC, LayoutC>
<<<BLOCK_NUM, nullptr, stream>>>(fftsAddr, a, b, c, aWorkSpace, bWorkSpace, symmetricPtr, cocTiling);
} else {
MatmulReduceScatterPaddingB<ElementA, LayoutA1, ElementB, LayoutB1, ElementC, LayoutC>
<<<BLOCK_NUM, nullptr, stream>>>(fftsAddr, a, b, c, aWorkSpace, bWorkSpace, symmetricPtr, cocTiling);
}
}
void LaunchAllGatherMatmulPaddingBF16(
void *stream, uint64_t fftsAddr,
uint8_t *a, uint8_t *b, uint8_t *c,
uint8_t *gatherA, uint8_t *workspace,
uint8_t *symmetricPtr, CocTilingParams& cocTiling,
uint32_t transA, uint32_t transB)
{
(void)gatherA;
if (!transA && !transB) {
AllGatherMatmulPadding<ElementA, LayoutA0, ElementB, LayoutB0, ElementC, LayoutC>
<<<BLOCK_NUM, nullptr, stream>>>(fftsAddr, a, b, c, workspace, symmetricPtr, cocTiling);
} else if (!transA && transB) {
AllGatherMatmulPadding<ElementA, LayoutA0, ElementB, LayoutB1, ElementC, LayoutC>
<<<BLOCK_NUM, nullptr, stream>>>(fftsAddr, a, b, c, workspace, symmetricPtr, cocTiling);
}
}