* 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.
*/
#ifndef K_MAX_SHAPE_DIM
#define K_MAX_SHAPE_DIM 0
#endif
#include "catlass/arch/arch.hpp"
#include "catlass/catlass.hpp"
#include "catlass/conv/block/block_conv.hpp"
#include "catlass/conv/block/block_swizzle.hpp"
#include "catlass/conv/device/device_conv.hpp"
#include "catlass/conv/dispatch_policy.hpp"
#include "catlass/conv/kernel/conv3d_bias.hpp"
#include "catlass/conv_coord.hpp"
#include "catlass/gemm/gemm_type.hpp"
#include "catlass/layout/layout.hpp"
#include "catlass/status.hpp"
#include "golden.hpp"
#include "helper.hpp"
using namespace Catlass;
struct Options {
const std::string HELPER =
"24_conv_bias batch di cin1 hi wi cin0 cout kd kh kw sD sH sW dD dH dW pD pH pW [device_id]";
std::vector<uint32_t> fmapRelated = {1, 1, 1, 2, 9, 16};
std::vector<uint32_t> filterRelated = {1, 1, 1, 1};
std::vector<uint32_t> strides = {1, 1, 1};
std::vector<uint32_t> pads = {0, 0, 0};
std::vector<uint32_t> dilations = {1, 1, 1};
int32_t deviceId{0};
Options() = default;
int Parse(int argc, const char **argv) {
enum class ArgsIndex {
BATCH_INDEX = 1,
DI_INDEX,
CIN1_INDEX,
HI_INDEX,
WI_INDEX,
CIN0_INDEX,
COUT_INDEX,
KD_INDEX,
KH_INDEX,
KW_INDEX,
SD_INDEX,
SH_INDEX,
SW_INDEX,
DD_INDEX,
DH_INDEX,
DW_INDEX,
PD_INDEX,
PH_INDEX,
PW_INDEX,
DEVICE_ID_INDEX,
ARGS_MAX
};
if (argc > static_cast<uint32_t>(ArgsIndex::ARGS_MAX)
|| argc < static_cast<uint32_t>(ArgsIndex::DEVICE_ID_INDEX)) {
std::cerr << HELPER << std::endl;
return 0;
}
fmapRelated[0] = std::atoi(argv[static_cast<uint32_t>(ArgsIndex::BATCH_INDEX)]);
fmapRelated[1] = std::atoi(argv[static_cast<uint32_t>(ArgsIndex::DI_INDEX)]);
fmapRelated[2] = std::atoi(argv[static_cast<uint32_t>(ArgsIndex::CIN1_INDEX)]);
fmapRelated[3] = std::atoi(argv[static_cast<uint32_t>(ArgsIndex::HI_INDEX)]);
fmapRelated[4] = std::atoi(argv[static_cast<uint32_t>(ArgsIndex::WI_INDEX)]);
fmapRelated[5] = std::atoi(argv[static_cast<uint32_t>(ArgsIndex::CIN0_INDEX)]);
filterRelated[0] = std::atoi(argv[static_cast<uint32_t>(ArgsIndex::KD_INDEX)]);
filterRelated[1] = std::atoi(argv[static_cast<uint32_t>(ArgsIndex::KH_INDEX)]);
filterRelated[2] = std::atoi(argv[static_cast<uint32_t>(ArgsIndex::KW_INDEX)]);
filterRelated[3] = std::atoi(argv[static_cast<uint32_t>(ArgsIndex::COUT_INDEX)]);
strides[0] = std::atoi(argv[static_cast<uint32_t>(ArgsIndex::SD_INDEX)]);
strides[1] = std::atoi(argv[static_cast<uint32_t>(ArgsIndex::SH_INDEX)]);
strides[2] = std::atoi(argv[static_cast<uint32_t>(ArgsIndex::SW_INDEX)]);
dilations[0] = std::atoi(argv[static_cast<uint32_t>(ArgsIndex::DD_INDEX)]);
dilations[1] = std::atoi(argv[static_cast<uint32_t>(ArgsIndex::DH_INDEX)]);
dilations[2] = std::atoi(argv[static_cast<uint32_t>(ArgsIndex::DW_INDEX)]);
pads[0] = std::atoi(argv[static_cast<uint32_t>(ArgsIndex::PD_INDEX)]);
pads[1] = std::atoi(argv[static_cast<uint32_t>(ArgsIndex::PH_INDEX)]);
pads[2] = std::atoi(argv[static_cast<uint32_t>(ArgsIndex::PW_INDEX)]);
if (argc == static_cast<uint32_t>(ArgsIndex::ARGS_MAX)) {
deviceId = std::atoi(argv[static_cast<uint32_t>(ArgsIndex::DEVICE_ID_INDEX)]);
}
return 0;
}
};
static void Run(const Options &options) {
aclrtStream stream{nullptr};
ACL_CHECK(aclInit(nullptr));
ACL_CHECK(aclrtSetDevice(options.deviceId));
ACL_CHECK(aclrtCreateStream(&stream));
Conv3dParams problemShape = Conv3dParams::MakeConvCoord(
options.fmapRelated.data(), options.filterRelated.data(), options.pads.data(), options.strides.data(),
options.dilations.data()
);
uint32_t n = problemShape.batch();
uint32_t di = problemShape.di();
uint32_t cin1 = problemShape.cin1();
uint32_t hi = problemShape.hi();
uint32_t wi = problemShape.wi();
uint32_t cin0 = problemShape.cin0();
uint32_t kdc1khkw = problemShape.kdc1khkw();
uint32_t n1 = problemShape.n1();
uint32_t n0 = problemShape.n0();
uint32_t dout = problemShape.dout();
uint32_t ho = problemShape.ho();
uint32_t wo = problemShape.wo();
uint32_t cout1 = problemShape.cout1();
uint32_t cout0 = problemShape.cout0();
uint32_t cout = problemShape.cout();
size_t lenFmap = static_cast<size_t>(n) * di * cin1 * hi * wi * cin0;
size_t lenFilter = static_cast<size_t>(kdc1khkw) * n1 * n0 * cin0;
size_t lenBias = static_cast<size_t>(cout);
size_t lenOut = static_cast<size_t>(n) * dout * cout1 * ho * wo * cout0;
size_t sizeFmap = lenFmap * sizeof(fp16_t);
size_t sizeFilter = lenFilter * sizeof(fp16_t);
size_t sizeOut = lenOut * sizeof(fp16_t);
size_t sizeBias = lenBias * sizeof(fp16_t);
using LayoutFmap = layout::NDC1HWC0;
using LayoutFilter = layout::KDC1KHKWN1N0C0;
using LayoutOut = layout::NDC1HWC0;
using LayoutBias = layout::VectorLayout;
LayoutFmap layoutFmap{n, di, cin1, hi * wi, cin0};
LayoutFilter layoutFilter{kdc1khkw, n1, n0, cin0};
LayoutOut layoutOut{n, dout, cout1, ho * wo, cout0};
std::vector<fp16_t> hostFmap(lenFmap);
std::vector<fp16_t> hostFilter(lenFilter);
std::vector<fp16_t> hostBias(lenBias);
ReadFile("./data/fmap.bin", hostFmap.data(), sizeFmap);
ReadFile("./data/weight.bin", hostFilter.data(), sizeFilter);
ReadFile("./data/bias.bin", hostBias.data(), sizeBias);
uint8_t *deviceFmap{nullptr};
ACL_CHECK(aclrtMalloc(reinterpret_cast<void **>(&deviceFmap), sizeFmap, ACL_MEM_MALLOC_HUGE_FIRST));
ACL_CHECK(aclrtMemcpy(deviceFmap, sizeFmap, hostFmap.data(), sizeFmap, ACL_MEMCPY_HOST_TO_DEVICE));
uint8_t *deviceFilter{nullptr};
ACL_CHECK(aclrtMalloc(reinterpret_cast<void **>(&deviceFilter), sizeFilter, ACL_MEM_MALLOC_HUGE_FIRST));
ACL_CHECK(aclrtMemcpy(deviceFilter, sizeFilter, hostFilter.data(), sizeFilter, ACL_MEMCPY_HOST_TO_DEVICE));
uint8_t *deviceBias{nullptr};
ACL_CHECK(aclrtMalloc(reinterpret_cast<void **>(&deviceBias), sizeBias, ACL_MEM_MALLOC_HUGE_FIRST));
ACL_CHECK(aclrtMemcpy(deviceBias, sizeBias, hostBias.data(), sizeBias, ACL_MEMCPY_HOST_TO_DEVICE));
uint8_t *deviceOut{nullptr};
ACL_CHECK(aclrtMalloc(reinterpret_cast<void **>(&deviceOut), sizeOut, ACL_MEM_MALLOC_HUGE_FIRST));
auto aicCoreNum = platform_ascendc::PlatformAscendCManager::GetInstance()->GetCoreNumAic();
constexpr uint32_t l1AStages = 1;
constexpr uint32_t l1BStages = 1;
constexpr uint32_t l0AStages = 2;
constexpr uint32_t l0BStages = 2;
constexpr uint32_t l0CStages = 1;
constexpr bool enableUnitFlag = true;
using ArchTag = Arch::AtlasA2;
using DispatchPolicy =
Conv::ConvAtlasA2Pingpong<l1AStages, l1BStages, l0AStages, l0BStages, l0CStages, enableUnitFlag>;
using FmapType = Gemm::GemmType<half, LayoutFmap>;
using FilterType = Gemm::GemmType<half, LayoutFilter>;
using BiasType = Gemm::GemmType<half, LayoutBias>;
using OutType = Gemm::GemmType<half, LayoutOut>;
using CoreTileShape = ConvCoreShape<2, 2, 2, 2>;
using FmapL1TileShape = ConvFmapL1Shape<16, 1, 1>;
using FilterL1TileShape = ConvFilterL1Shape<1, 1, 16>;
using L0TileShape = ConvL0Shape<16, 16, 16>;
using BlockConv = Conv::Block::BlockConv<
DispatchPolicy, CoreTileShape, FmapL1TileShape, FilterL1TileShape, L0TileShape, FmapType, FilterType, OutType,
BiasType>;
using BlockEpilogue = void;
using BlockScheduler = typename Conv::Block::Conv3dIdentityBlockSwizzle<3, 0>;
using ConvKernel = Conv::Kernel::ConvBias<BlockConv, BlockEpilogue, BlockScheduler>;
using ConvAdapter = Conv::Device::DeviceConv<ConvKernel>;
ConvKernel::Arguments arguments{problemShape, deviceFmap, deviceFilter, deviceOut, deviceBias};
ConvAdapter conv_op;
conv_op.CanImplement(arguments);
size_t sizeWorkspace = conv_op.GetWorkspaceSize(arguments);
uint8_t *deviceWorkspace = nullptr;
if (sizeWorkspace > 0) {
ACL_CHECK(aclrtMalloc(reinterpret_cast<void **>(&deviceWorkspace), sizeWorkspace, ACL_MEM_MALLOC_HUGE_FIRST));
}
conv_op.Initialize(arguments, deviceWorkspace);
conv_op(stream, aicCoreNum);
ACL_CHECK(aclrtSynchronizeStream(stream));
if (sizeWorkspace > 0) {
ACL_CHECK(aclrtFree(deviceWorkspace));
}
std::vector<fp16_t> hostOut(lenOut);
ACL_CHECK(aclrtMemcpy(hostOut.data(), sizeOut, deviceOut, sizeOut, ACL_MEMCPY_DEVICE_TO_HOST));
std::vector<float> hostGolden(lenOut);
const size_t goldenSize = sizeOut * 2;
ReadFile("./data/golden.bin", hostGolden.data(), goldenSize);
std::vector<uint64_t> errorIndices = golden::CompareData(hostOut, hostGolden, kdc1khkw * cin0);
if (errorIndices.empty()) {
std::cout << "Compare success." << std::endl;
} else {
std::cerr << "Compare failed. Error count: " << errorIndices.size() << std::endl;
}
ACL_CHECK(aclrtFree(deviceFmap));
ACL_CHECK(aclrtFree(deviceFilter));
ACL_CHECK(aclrtFree(deviceBias));
ACL_CHECK(aclrtFree(deviceOut));
ACL_CHECK(aclrtDestroyStream(stream));
ACL_CHECK(aclrtResetDevice(options.deviceId));
ACL_CHECK(aclFinalize());
}
int main(int argc, const char **argv) {
Options options;
if (options.Parse(argc, argv) != 0) {
return -1;
}
Run(options);
return 0;
}