* Copyright (c) 2026 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 sspmv_kernel.cpp
* \brief SSPMV Kernel for ascend950 (DAV_3510)
*/
#include "kernel_operator.h"
#include "simt_api/asc_simt.h"
#include "common/helper/kernel_constant.h"
#include "cann_ops_blas_common.h"
#include "sspmv_tiling_data.h"
template <bool UPLO_IS_UPPER>
__simt_vf__ __aicore__ LAUNCH_BOUND(SIMT_MAX_THREAD_NUM) inline void SspmvSimtCompute(
uint32_t n, float alpha, float beta, int64_t incx, int64_t incy, __gm__ const float* aGm,
__gm__ const float* xGm, __gm__ float* yGm)
{
for (uint32_t row = blockIdx.x * blockDim.x + threadIdx.x; row < n; row += gridDim.x * blockDim.x) {
int64_t symStart;
if constexpr (UPLO_IS_UPPER) {
symStart = static_cast<int64_t>(row) * (row + 1) / 2;
} else {
symStart = static_cast<int64_t>(row) * n - static_cast<int64_t>(row) * (row - 1) / 2;
}
float acc = 0.0f;
int64_t colStart = 0;
for (uint32_t jCol = 0; jCol < n; ++jCol) {
int64_t gmIdx;
if constexpr (UPLO_IS_UPPER) {
gmIdx = (row <= jCol) ? colStart + row : symStart + jCol;
colStart += (jCol + 1);
} else {
gmIdx = (row >= jCol) ? colStart + (row - jCol) : symStart + (jCol - row);
colStart += (n - jCol);
}
float xVal = (incx >= 0) ? xGm[jCol * incx] : xGm[(n - 1 - jCol) * (-incx)];
acc += aGm[gmIdx] * xVal;
}
int64_t yIdx = (incy >= 0) ? (row * incy) : ((n - 1 - row) * (-incy));
float yVal = yGm[yIdx];
yGm[yIdx] = alpha * acc + beta * yVal;
}
}
__global__ __aicore__ void sspmv_kernel(GM_ADDR a, GM_ADDR x, GM_ADDR y, GM_ADDR workSpace,
const SspmvTilingData tiling)
{
KERNEL_TASK_TYPE_DEFAULT(KERNEL_TYPE_AIV_ONLY);
if (tiling.uplo == ACLBLAS_UPPER) {
asc_vf_call<SspmvSimtCompute<true>>(
dim3{tiling.nthreads, 1, 1}, tiling.n, tiling.alpha, tiling.beta, tiling.incx, tiling.incy,
reinterpret_cast<__gm__ const float*>(a), reinterpret_cast<__gm__ const float*>(x),
reinterpret_cast<__gm__ float*>(y));
} else {
asc_vf_call<SspmvSimtCompute<false>>(
dim3{tiling.nthreads, 1, 1}, tiling.n, tiling.alpha, tiling.beta, tiling.incx, tiling.incy,
reinterpret_cast<__gm__ const float*>(a), reinterpret_cast<__gm__ const float*>(x),
reinterpret_cast<__gm__ float*>(y));
}
}
void sspmv_kernel_do(
GM_ADDR a, GM_ADDR x, GM_ADDR y, GM_ADDR workSpace,
uint32_t numBlocks, const SspmvTilingData& tiling, void* stream)
{
sspmv_kernel<<<numBlocks, nullptr, stream>>>(a, x, y, workSpace, tiling);
}