* 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 <cstring>
#include <iostream>
#include <vector>
#include <cstdlib>
#include <cmath>
#include <string>
#include <tiling/platform/platform_ascendc.h>
#include <acl/acl.h>
#include "catlass/catlass.hpp"
#include "catlass/arch/arch.hpp"
#include "catlass/layout/layout.hpp"
#include "catlass/gemm/block/block_mmad.hpp"
#include "catlass/gemm/block/block_swizzle.hpp"
#include "catlass/gemm/dispatch_policy.hpp"
#include "catlass/gemm/kernel/w4a8_matmul.hpp"
#include "catlass/gemm/gemm_type.hpp"
#include "catlass/gemm/device/device_gemm.hpp"
#include "golden.hpp"
#include "helper.hpp"
using namespace Catlass;
using Options = GemmOptions;
void Run(Options const &options)
{
aclrtStream stream{nullptr};
ACL_CHECK(aclInit(nullptr));
ACL_CHECK(aclrtSetDevice(options.deviceId));
ACL_CHECK(aclrtCreateStream(&stream));
auto aicoreNum = platform_ascendc::PlatformAscendCManager::GetInstance()->GetCoreNumAic();
uint32_t m = options.problemShape.m();
uint32_t n = options.problemShape.n();
uint32_t k = options.problemShape.k();
float scalar = 1.5;
using LayoutA = layout::RowMajor;
using LayoutPrologueB = layout::RowMajor;
using LayoutB = layout::RowMajor;
using LayoutC = layout::RowMajor;
LayoutA layoutA{m, k};
LayoutPrologueB layoutPrologueB{k, n, (n + 1) / 2 * 2};
LayoutC layoutC{m, n};
size_t lenA = static_cast<size_t>(m) * k;
size_t lenB = static_cast<size_t>(k) * n;
size_t lenC = static_cast<size_t>(m) * n;
uint64_t sizeA = layoutA.Capacity() * sizeof(int8_t);
uint64_t sizeB = layoutPrologueB.Capacity() / 2 * sizeof(int8_t);
uint64_t sizeC = layoutC.Capacity() * sizeof(fp16_t);
uint64_t goldenSize = layoutC.Capacity() * sizeof(float);
void *hostA = nullptr;
ACL_CHECK(aclrtMallocHost(&hostA, sizeA));
std::string inputA_path = "../../examples/32_w4a8_matmul/data/inputA.dat";
ReadFile(inputA_path, hostA, sizeA);
void *hostB = nullptr;
ACL_CHECK(aclrtMallocHost(&hostB, sizeB));
std::string inputB_path = "../../examples/32_w4a8_matmul/data/inputB.dat";
ReadFile(inputB_path, hostB, sizeB);
std::vector<float> hExpected(lenC);
std::string expected_path = "../../examples/32_w4a8_matmul/data/expected.dat";
ReadFile(expected_path, hExpected.data(), goldenSize);
using ElementA = int8_t;
using ElementPrologueB = AscendC::int4b_t;
using ElementB = int8_t;
using ElementC = half;
uint64_t fftsAddr{0};
uint32_t fftsLen{0};
rtGetC2cCtrlAddr(&fftsAddr, &fftsLen);
uint8_t *deviceA, *deviceB, *deviceC;
ACL_CHECK(aclrtMalloc((void **)&deviceA, sizeA, ACL_MEM_MALLOC_HUGE_FIRST));
ACL_CHECK(aclrtMalloc((void **)&deviceB, sizeB, ACL_MEM_MALLOC_HUGE_FIRST));
ACL_CHECK(aclrtMalloc((void **)&deviceC, sizeC, ACL_MEM_MALLOC_HUGE_FIRST));
ACL_CHECK(aclrtMemcpy(deviceA, sizeA, hostA, sizeA, ACL_MEMCPY_HOST_TO_DEVICE));
ACL_CHECK(aclrtMemcpy(deviceB, sizeB, hostB, sizeB, ACL_MEMCPY_HOST_TO_DEVICE));
using ArchTag = Arch::AtlasA2;
constexpr bool enableUnitFlag = false;
using DispatchPolicy = Gemm::MmadAtlasA2PingPongWithPrologue<enableUnitFlag>;
using L1TileShape = std::conditional_t<std::is_same_v<LayoutA, layout::ColumnMajor> &&
std::is_same_v<LayoutB, layout::ColumnMajor>, GemmShape<256, 128, 512>, GemmShape<128, 256, 512>>;
using L0TileShape = std::conditional_t<std::is_same_v<LayoutA, layout::ColumnMajor> &&
std::is_same_v<LayoutB, layout::ColumnMajor>, GemmShape<256, 128, 128>, GemmShape<128, 256, 128>>;
using PrologueSrcType = Gemm::GemmType<ElementPrologueB, LayoutPrologueB>;
using PrologueDstType = Gemm::GemmType<ElementB, LayoutB>;
using AType = Gemm::GemmType<ElementA, LayoutA>;
using BType = PrologueDstType;
using CType = Gemm::GemmType<ElementC, LayoutC>;
using PrologueA = void;
constexpr uint32_t computeLen = 24 * 1024;
using PrologueB = Gemm::Tile::TileCastInt4ToInt8<ArchTag, PrologueSrcType, PrologueDstType, computeLen>;
using TileCopy = Gemm::Tile::TileCopyWithPrologueDeqPerTensor<ArchTag, AType, BType, CType, PrologueA, PrologueB>;
using BlockMmad = Gemm::Block::BlockMmad<
DispatchPolicy,
L1TileShape, L0TileShape,
AType, BType, CType, void,
TileCopy
>;
if (options.problemShape.m() > options.problemShape.n()) {
using BlockScheduler = typename Gemm::Block::GemmIdentityBlockSwizzle<3, 0>;
using BlockEpilogue = void;
using MatmulKernel = Gemm::Kernel::W4A8Matmul<BlockMmad, BlockEpilogue, BlockScheduler>;
using MatmulAdapter = Gemm::Device::DeviceGemm<MatmulKernel>;
MatmulAdapter matmulOp;
typename MatmulKernel::Arguments arguments{
options.problemShape,
deviceA, layoutA,
deviceB, layoutPrologueB,
deviceC, layoutC,
scalar,
aicoreNum
};
matmulOp.CanImplement(arguments);
size_t sizeWorkspace = matmulOp.GetWorkspaceSize(arguments);
uint8_t *deviceWorkspace = nullptr;
if (sizeWorkspace > 0) {
ACL_CHECK(aclrtMalloc(reinterpret_cast<void **>(&deviceWorkspace), sizeWorkspace, ACL_MEM_MALLOC_HUGE_FIRST));
}
matmulOp.Initialize(arguments, deviceWorkspace);
matmulOp(stream, aicoreNum, fftsAddr);
if (sizeWorkspace > 0) {
ACL_CHECK(aclrtFree(deviceWorkspace));
}
} else {
using BlockScheduler = typename Gemm::Block::GemmIdentityBlockSwizzle<3, 1>;
using BlockEpilogue = void;
using MatmulKernel = Gemm::Kernel::W4A8Matmul<BlockMmad, BlockEpilogue, BlockScheduler>;
using MatmulAdapter = Gemm::Device::DeviceGemm<MatmulKernel>;
MatmulAdapter matmulOp;
typename MatmulKernel::Arguments arguments{
options.problemShape,
deviceA, layoutA,
deviceB, layoutPrologueB,
deviceC, layoutC,
scalar,
aicoreNum
};
matmulOp.CanImplement(arguments);
size_t sizeWorkspace = matmulOp.GetWorkspaceSize(arguments);
uint8_t *deviceWorkspace = nullptr;
if (sizeWorkspace > 0) {
ACL_CHECK(aclrtMalloc(reinterpret_cast<void **>(&deviceWorkspace), sizeWorkspace, ACL_MEM_MALLOC_HUGE_FIRST));
}
matmulOp.Initialize(arguments, deviceWorkspace);
matmulOp(stream, aicoreNum, fftsAddr);
if (sizeWorkspace > 0) {
ACL_CHECK(aclrtFree(deviceWorkspace));
}
}
ACL_CHECK(aclrtSynchronizeStream(stream));
ACL_CHECK(aclrtFree(deviceA));
ACL_CHECK(aclrtFree(deviceB));
std::vector<fp16_t> hostC(lenC);
ACL_CHECK(aclrtMemcpy(hostC.data(), sizeC, deviceC, sizeC, ACL_MEMCPY_DEVICE_TO_HOST));
ACL_CHECK(aclrtFree(deviceC));
std::vector<uint64_t> errorIndices = golden::CompareData(hostC, hExpected, k);
if (errorIndices.empty()) {
std::cout << "Compare success." << std::endl;
} else {
std::cerr << "Compare failed. Error count: " << errorIndices.size() << std::endl;
}
ACL_CHECK(aclrtFreeHost(hostA));
ACL_CHECK(aclrtFreeHost(hostB));
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;
}