* 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.
*/
#include "kernel_operator.h"
#include "common/helper/kernel_utils.h"
#include "common/iterator/iterator.h"
#include "common/compute/simd.h"
#ifdef __CCE_KT_TEST__
#undef __aicore__
#define __aicore__
#else
#ifndef __aicore__
#define __aicore__ [aicore]
#endif
#endif
__aicore__ __inline__ __attribute__((always_inline)) void copy_vec_gm2ub(
AscendC::LocalTensor<float> dst,
AscendC::GlobalTensor<float> src,
uint32_t len)
{
uint16_t nBurst = 1;
uint32_t lenBurst = len * sizeof(float);
uint8_t leftPaddingNum = 0;
uint8_t rightPaddingNum = 8 - len % 8;
uint32_t srcGap = 0;
uint32_t dstGap = 0;
gm_to_ub_align<ArchType::ASCEND_V220, float>(
dst, src,
0,
nBurst, lenBurst, leftPaddingNum, rightPaddingNum, srcGap, dstGap);
}
__aicore__ __inline__ __attribute__((always_inline)) void copy_vec_ub2gm(
AscendC::GlobalTensor<float> dst,
AscendC::LocalTensor<float> src,
uint32_t len)
{
uint16_t nBurst = 1;
uint32_t lenBurst = len * sizeof(float);
uint8_t leftPaddingNum = 0;
uint8_t rightPaddingNum = 0;
uint32_t srcGap = 0;
uint32_t dstGap = 0;
ub_to_gm_align<ArchType::ASCEND_V220, float>(
dst, src,
0,
nBurst, lenBurst, leftPaddingNum, rightPaddingNum, srcGap, dstGap);
}
__aicore__ __inline__ __attribute__((always_inline)) void cdot_single_iteration_aiv(
AscendC::GlobalTensor<float> gm_x, AscendC::GlobalTensor<float> gm_y, AscendC::GlobalTensor<float> gm_result,
AscendC::LocalTensor<float> ub_x, AscendC::LocalTensor<float> ub_y, AscendC::LocalTensor<float> ub_x_real,
AscendC::LocalTensor<float> ub_x_imag, AscendC::LocalTensor<float> ub_y_real, AscendC::LocalTensor<float> ub_y_imag,
AscendC::LocalTensor<float> ub_acc_real, AscendC::LocalTensor<float> ub_acc_imag, uint32_t len, uint32_t event_id,
uint32_t conj)
{
uint32_t repeatTime = (len + 63) / 64;
uint32_t computeRepeat = (len / 2 + 63) / 64;
uint32_t realOffset = 0;
uint32_t imagOffset = computeRepeat * 64;
AscendC::Duplicate<float, false>(ub_x, (float)0.0, 64, computeRepeat * 2, 1, 8);
AscendC::Duplicate<float, false>(ub_y, (float)0.0, 64, computeRepeat * 2, 1, 8);
SET_FLAG(V, MTE2, event_id);
WAIT_FLAG(V, MTE2, event_id);
copy_vec_gm2ub(ub_x, gm_x, len);
copy_vec_gm2ub(ub_y, gm_y, len);
SET_FLAG(MTE2, V, event_id);
WAIT_FLAG(MTE2, V, event_id);
AscendC::Duplicate<float, false>(ub_x_real, (float)0.0, 64, computeRepeat, 1, 8);
AscendC::Duplicate<float, false>(ub_x_imag, (float)0.0, 64, computeRepeat, 1, 8);
AscendC::Duplicate<float, false>(ub_y_real, (float)0.0, 64, computeRepeat, 1, 8);
AscendC::Duplicate<float, false>(ub_y_imag, (float)0.0, 64, computeRepeat, 1, 8);
PIPE_BARRIER(V);
uint32_t mask = 0;
uint64_t rsvdCnt = 0;
AscendC::GatherMask<float>(ub_x_real, ub_x, 1, false, mask,
{1, static_cast<uint16_t>(repeatTime), 8, 8}, rsvdCnt);
AscendC::GatherMask<float>(ub_x_imag, ub_x, 2, false, mask,
{1, static_cast<uint16_t>(repeatTime), 8, 8}, rsvdCnt);
AscendC::GatherMask<float>(ub_y_real, ub_y, 1, false, mask,
{1, static_cast<uint16_t>(repeatTime), 8, 8}, rsvdCnt);
AscendC::GatherMask<float>(ub_y_imag, ub_y, 2, false, mask,
{1, static_cast<uint16_t>(repeatTime), 8, 8}, rsvdCnt);
PIPE_BARRIER(V);
if (conj) {
muls_v<ArchType::ASCEND_V220, float>(ub_x_imag, ub_x_imag, -1.0f, computeRepeat, 1, 1, 8, 8);
PIPE_BARRIER(V);
}
uint64_t mul_mask = 64;
AscendC::Mul(ub_x, ub_x_real, ub_y_real, mul_mask, computeRepeat, {1, 1, 1, 8, 8, 8});
AscendC::Mul(ub_y, ub_x_real, ub_y_imag, mul_mask, computeRepeat, {1, 1, 1, 8, 8, 8});
PIPE_BARRIER(V);
mul_v<ArchType::ASCEND_V220, float>(ub_x_real, ub_x_imag, ub_y_imag, computeRepeat, 1, 1, 1, 8, 8, 8);
mul_v<ArchType::ASCEND_V220, float>(ub_y_real, ub_x_imag, ub_y_real, computeRepeat, 1, 1, 1, 8, 8, 8);
PIPE_BARRIER(V);
sub_v<ArchType::ASCEND_V220, float>(ub_x_real, ub_x, ub_x_real, computeRepeat, 1, 1, 1, 8, 8, 8);
add_v<ArchType::ASCEND_V220, float>(ub_y_real, ub_y_real, ub_y, computeRepeat, 1, 1, 1, 8, 8, 8);
AscendC::Duplicate<float, false>(ub_x_imag, (float)0.0, 64, 1, 1, 8);
AscendC::Duplicate<float, false>(ub_y_imag, (float)0.0, 64, 1, 1, 8);
PIPE_BARRIER(V);
cadd_v<ArchType::ASCEND_V220, float>(ub_x_imag, ub_x_real, computeRepeat, 1, 1, 8);
cadd_v<ArchType::ASCEND_V220, float>(ub_y_imag, ub_y_real, computeRepeat, 1, 1, 8);
if (computeRepeat > 0) {
PIPE_BARRIER(V);
cadd_v<ArchType::ASCEND_V220, float>(ub_x_imag, ub_x_imag, 1, 1, 1, 8);
cadd_v<ArchType::ASCEND_V220, float>(ub_y_imag, ub_y_imag, 1, 1, 1, 8);
}
PIPE_BARRIER(V);
add_v<ArchType::ASCEND_V220, float>(ub_acc_real, ub_acc_real, ub_x_imag, 1, 1, 1, 1, 8, 8, 8);
add_v<ArchType::ASCEND_V220, float>(ub_acc_imag, ub_acc_imag, ub_y_imag, 1, 1, 1, 1, 8, 8, 8);
}
__aicore__ __inline__ __attribute__((always_inline)) void cdot_process_aiv(
AscendC::GlobalTensor<float> gm_x, AscendC::GlobalTensor<float> gm_y, AscendC::GlobalTensor<float> gm_result,
AscendC::GlobalTensor<float> gm_workpsace, uint32_t offset, uint32_t compute_num, uint32_t conj,
uint32_t vec_core_num)
{
auto core_idx = AscendC::GetBlockIdx();
uint32_t maxDataCount = 23 * 1024 / 4;
AsdopsBuffer<ArchType::ASCEND_V220> buf;
AscendC::LocalTensor<float> ub_x_ping = buf.GetBuffer<BufferType::ASCEND_UB, float>(0);
AscendC::LocalTensor<float> ub_x_pong = buf.GetBuffer<BufferType::ASCEND_UB, float>(23 * 1024);
AscendC::LocalTensor<float> ub_y_ping = buf.GetBuffer<BufferType::ASCEND_UB, float>(23 * 2 * 1024);
AscendC::LocalTensor<float> ub_y_pong = buf.GetBuffer<BufferType::ASCEND_UB, float>(23 * 3 * 1024);
AscendC::LocalTensor<float> ub_x_real_ping = buf.GetBuffer<BufferType::ASCEND_UB, float>(23 * 4 * 1024);
AscendC::LocalTensor<float> ub_x_real_pong = buf.GetBuffer<BufferType::ASCEND_UB, float>((92 + 11.5 * 1) * 1024);
AscendC::LocalTensor<float> ub_x_imag_ping = buf.GetBuffer<BufferType::ASCEND_UB, float>((92 + 11.5 * 2) * 1024);
AscendC::LocalTensor<float> ub_x_imag_pong = buf.GetBuffer<BufferType::ASCEND_UB, float>((92 + 11.5 * 3) * 1024);
AscendC::LocalTensor<float> ub_y_real_ping = buf.GetBuffer<BufferType::ASCEND_UB, float>((92 + 11.5 * 4) * 1024);
AscendC::LocalTensor<float> ub_y_real_pong = buf.GetBuffer<BufferType::ASCEND_UB, float>((92 + 11.5 * 5) * 1024);
AscendC::LocalTensor<float> ub_y_imag_ping = buf.GetBuffer<BufferType::ASCEND_UB, float>((92 + 11.5 * 6) * 1024);
AscendC::LocalTensor<float> ub_y_imag_pong = buf.GetBuffer<BufferType::ASCEND_UB, float>((92 + 11.5 * 7) * 1024);
AscendC::LocalTensor<float> ub_acc_real_ping = buf.GetBuffer<BufferType::ASCEND_UB, float>((92 + 11.5 * 8) * 1024);
AscendC::LocalTensor<float> ub_acc_imag_ping = buf.GetBuffer<BufferType::ASCEND_UB, float>((92 + 11.5 * 8 + 1) * 1024);
AscendC::LocalTensor<float> ub_acc_real_pong = buf.GetBuffer<BufferType::ASCEND_UB, float>((92 + 11.5 * 8 + 2) * 1024);
AscendC::LocalTensor<float> ub_acc_imag_pong = buf.GetBuffer<BufferType::ASCEND_UB, float>((92 + 11.5 * 8 + 3) * 1024);
AscendC::Duplicate<float, false>(ub_acc_real_ping, (float)0.0, 64, 1, 1, 8);
AscendC::Duplicate<float, false>(ub_acc_imag_ping, (float)0.0, 64, 1, 1, 8);
AscendC::Duplicate<float, false>(ub_acc_real_pong, (float)0.0, 64, 1, 1, 8);
AscendC::Duplicate<float, false>(ub_acc_imag_pong, (float)0.0, 64, 1, 1, 8);
PIPE_BARRIER(V);
uint32_t repeatTimes = compute_num / maxDataCount;
uint32_t remainNum = compute_num % maxDataCount;
uint32_t ping_flag = 1;
uint32_t curr_offset = offset;
if (repeatTimes > 0) {
SET_FLAG(V, MTE2, EVENT_ID0);
SET_FLAG(V, MTE2, EVENT_ID1);
for (uint32_t i = 0; i < repeatTimes; i++) {
AscendC::LocalTensor<float> ub_x = ping_flag ? ub_x_ping : ub_x_pong;
AscendC::LocalTensor<float> ub_y = ping_flag ? ub_y_ping : ub_y_pong;
AscendC::LocalTensor<float> ub_x_real = ping_flag ? ub_x_real_ping : ub_x_real_pong;
AscendC::LocalTensor<float> ub_x_imag = ping_flag ? ub_x_imag_ping : ub_x_imag_pong;
AscendC::LocalTensor<float> ub_y_real = ping_flag ? ub_y_real_ping : ub_y_real_pong;
AscendC::LocalTensor<float> ub_y_imag = ping_flag ? ub_y_imag_ping : ub_y_imag_pong;
AscendC::LocalTensor<float> ub_acc_real = ping_flag ? ub_acc_real_ping : ub_acc_real_pong;
AscendC::LocalTensor<float> ub_acc_imag = ping_flag ? ub_acc_imag_ping : ub_acc_imag_pong;
auto event_id = ping_flag ? EVENT_ID0 : EVENT_ID1;
WAIT_FLAG(V, MTE2, event_id);
cdot_single_iteration_aiv(gm_x[curr_offset], gm_y [curr_offset], gm_result, ub_x, ub_y, ub_x_real,
ub_x_imag, ub_y_real, ub_y_imag, ub_acc_real, ub_acc_imag, maxDataCount, event_id,
conj);
curr_offset += maxDataCount;
SET_FLAG(V, MTE2, event_id);
ping_flag = 1 - ping_flag;
}
WAIT_FLAG(V, MTE2, EVENT_ID0);
WAIT_FLAG(V, MTE2, EVENT_ID1);
}
if (remainNum > 0) {
AscendC::LocalTensor<float> ub_x = ping_flag ? ub_x_ping : ub_x_pong;
AscendC::LocalTensor<float> ub_y = ping_flag ? ub_y_ping : ub_y_pong;
AscendC::LocalTensor<float> ub_x_real = ping_flag ? ub_x_real_ping : ub_x_real_pong;
AscendC::LocalTensor<float> ub_x_imag = ping_flag ? ub_x_imag_ping : ub_x_imag_pong;
AscendC::LocalTensor<float> ub_y_real = ping_flag ? ub_y_real_ping : ub_y_real_pong;
AscendC::LocalTensor<float> ub_y_imag = ping_flag ? ub_y_imag_ping : ub_y_imag_pong;
AscendC::LocalTensor<float> ub_acc_real = ping_flag ? ub_acc_real_ping : ub_acc_real_pong;
AscendC::LocalTensor<float> ub_acc_imag = ping_flag ? ub_acc_imag_ping : ub_acc_imag_pong;
auto event_id = ping_flag ? EVENT_ID0 : EVENT_ID1;
cdot_single_iteration_aiv(gm_x[curr_offset], gm_y[curr_offset], gm_result, ub_x, ub_y, ub_x_real, ub_x_imag,
ub_y_real, ub_y_imag, ub_acc_real, ub_acc_imag, remainNum, event_id, conj);
}
PIPE_BARRIER(ALL);
if (repeatTimes > 0) {
add_v<ArchType::ASCEND_V220, float>(ub_acc_real_ping, ub_acc_real_ping, ub_acc_real_pong, 1, 1, 1, 1, 8, 8, 8);
add_v<ArchType::ASCEND_V220, float>(ub_acc_imag_ping, ub_acc_imag_ping, ub_acc_imag_pong, 1, 1, 1, 1, 8, 8, 8);
SET_FLAG(V, MTE3, EVENT_ID0);
WAIT_FLAG(V, MTE3, EVENT_ID0);
}
uint32_t imag_gm_offset = 40;
copy_vec_ub2gm(gm_workpsace[core_idx], ub_acc_real_ping, 1);
copy_vec_ub2gm(gm_workpsace[imag_gm_offset + core_idx], ub_acc_imag_ping, 1);
PIPE_BARRIER(ALL);
FftsCrossCoreSync<PIPE_MTE3, 0>(0);
WaitFlagDev(0);
if (core_idx == 0) {
AscendC::LocalTensor<float> ub_sum_real = buf.GetBuffer<BufferType::ASCEND_UB, float>(0);
AscendC::LocalTensor<float> ub_sum_imag = buf.GetBuffer<BufferType::ASCEND_UB, float>(64 * 4);
AscendC::LocalTensor<float> ub_result_real = buf.GetBuffer<BufferType::ASCEND_UB, float>(64 * 4 * 2);
AscendC::LocalTensor<float> ub_result_imag = buf.GetBuffer<BufferType::ASCEND_UB, float>(64 * 4 * 3);
AscendC::Duplicate<float, false>(ub_sum_real, (float)0.0, 64, 4, 1, 8);
SET_FLAG(V, MTE2, EVENT_ID0);
WAIT_FLAG(V, MTE2, EVENT_ID0);
copy_vec_gm2ub(ub_result_real, gm_workpsace, vec_core_num);
copy_vec_gm2ub(ub_result_imag, gm_workpsace[imag_gm_offset], vec_core_num);
SET_FLAG(MTE2, V, EVENT_ID0);
WAIT_FLAG(MTE2, V, EVENT_ID0);
cadd_v<ArchType::ASCEND_V220, float>(ub_sum_real, ub_result_real, 1, 1, 1, 8);
cadd_v<ArchType::ASCEND_V220, float>(ub_sum_imag, ub_result_imag, 1, 1, 1, 8);
SET_FLAG(V, MTE3, EVENT_ID0);
WAIT_FLAG(V, MTE3, EVENT_ID0);
copy_vec_ub2gm(gm_result, ub_sum_real, 1);
copy_vec_ub2gm(gm_result[1], ub_sum_imag, 1);
}
PIPE_BARRIER(ALL);
}
__global__ __aicore__ __vector__ void cdot(GM_ADDR x, GM_ADDR y, GM_ADDR result,
GM_ADDR workSpace, GM_ADDR tilingGm)
{
AscendC::SetMaskNorm();
SetVectorMask<float>((uint64_t)-1, (uint64_t)-1);
SetAtomicnone();
auto core_idx = AscendC::GetBlockIdx();
auto tiling_buf = reinterpret_cast<__gm__ uint8_t *>(tilingGm);
uint32_t n = (*(__gm__ uint32_t *)((__gm__ uint8_t *)tiling_buf));
uint32_t coreNum = (*(__gm__ uint32_t *)((__gm__ uint8_t *)tiling_buf + 4));
uint32_t isConj = (*(__gm__ uint32_t *)((__gm__ uint8_t *)tiling_buf + 8));
uint32_t offset = (*(__gm__ uint32_t *)((__gm__ uint8_t *)tiling_buf + 12 + 4 * core_idx));
uint32_t cal_num = (*(__gm__ uint32_t *)((__gm__ uint8_t *)tiling_buf + 12 + 40 * 4 + 4 * core_idx));
AscendC::GlobalTensor<float> x_tensor;
AscendC::GlobalTensor<float> y_tensor;
AscendC::GlobalTensor<float> result_tensor;
AscendC::GlobalTensor<float> workspace_tensor;
x_tensor.SetGlobalBuffer(reinterpret_cast<__gm__ float *>(x));
y_tensor.SetGlobalBuffer(reinterpret_cast<__gm__ float *>(y));
result_tensor.SetGlobalBuffer(reinterpret_cast<__gm__ float *>(result));
workspace_tensor.SetGlobalBuffer(reinterpret_cast<__gm__ float *>(workSpace));
AsdopsBuffer<ArchType::ASCEND_V220> buf;
AscendC::LocalTensor<float> ub_clear = buf.GetBuffer<BufferType::ASCEND_UB, float>(0);
AscendC::Duplicate<float, false>(ub_clear, (float)0.0, 64, 2, 1, 8);
SET_FLAG(V, MTE3, EVENT_ID0);
WAIT_FLAG(V, MTE3, EVENT_ID0);
copy_vec_ub2gm(workspace_tensor, ub_clear, coreNum * 2);
FftsCrossCoreSync<PIPE_MTE3, 0>(0);
WaitFlagDev(0);
cdot_process_aiv(x_tensor, y_tensor, result_tensor, workspace_tensor, offset, cal_num, isConj, coreNum);
}
void cdot_kernel_do(GM_ADDR x, GM_ADDR y, GM_ADDR result, GM_ADDR workSpace, GM_ADDR tilingGm,
uint32_t numBlocks, void *stream)
{
cdot<<<numBlocks, nullptr, stream>>>(x, y, result, workSpace, tilingGm);
}