cann/sip asdConvolve卷积滤波算子

asdConvolve

【免费下载链接】sip本项目是CANN提供的一款高效、可靠的高性能信号处理算子加速库，基于华为Ascend AI处理器，专门为信号处理领域而设计。项目地址: https://gitcode.com/cann/sip

产品支持情况

功能说明

• 接口功能：对给定的信号进行一维滤波操作。

• 计算公式： $$ w(k)=\sum _{j}u(j)v(k-j+1) $$ 其中，w(k)为输出的k位置的元素，u(j)为输入位置为j的一维信号，v(k-j+1)为位置为k-j+1的滤波卷积核。一维信号为复数向量，滤波卷积核为实数向量。

  示例： 输入“u”为： [[1.+1.j 2.+2.j] [1.+1.j 2.+2.j]] 输入“v”为： [1. 2. 3. 4.] 调用“asdConvolve”算子后，输出“result”为： [[4.+4.j, 7.+7.j], [4.+4.j, 7.+7.j]]

函数原型

AspbStatus asdConvolve( const aclTensor * signal, const aclTensor * kernel, aclTensor * output, asdConvolveMode_t mode, void *stream, void * workspace)

asdConvolve

• 参数说明：

  参数名	输入/输出	描述
  signal（aclTensor *）	输入	输入的一维信号。 支持的数据类型为COMPLEX32、COMPLEX64。 输入信号shape为[BatchCount, n]。
  kernel（aclTensor *）	输入	输入的滤波卷积核。 支持的数据类型为FLOAT16、FLOAT32。 输入滤波卷积核shape为[k]。
  output（aclTensor *）	输入/输出	输入/输出信号。 支持的数据类型为COMPLEX32、COMPLEX64。 输出shape与输入shape保持一致。
  mode（asdConvolveMode_t）	输入	滤波卷积模式，当前仅支持ASD_CONVOLVE_SAME，即输入和输出的向量维度保持一致。
  stream（void*）	输入	算子执行时的stream。
  workspace（void*）	输入	算子所需的Workspace指针。

• 返回值：

  返回状态码，具体参见SiP返回码。

约束说明

无

调用示例

示例代码如下，该样例旨在提供快速上手、开发和调试算子的最小化实现，其核心目标是使用最精简的代码展示算子的核心功能，而非提供生产级的安全保障。不推荐用户直接将示例代码作为业务代码，若用户将示例代码应用在自身的真实业务场景中且发生了安全问题，则需用户自行承担。

#include <iostream> #include "asdsip.h" #include "filter_api.h" #include "acl/acl.h" #include "acl/acl_base.h" #include "acl_meta.h" #include <complex> #include <vector> using namespace AsdSip; using half = op::fp16_t; #define ASD_STATUS_CHECK(err) \ do { \ AsdSip::AspbStatus err_ = (err); \ if (err_ != AsdSip::ACL_SUCCESS) { \ std::cout << "Execute failed." << std::endl; \ exit(-1); \ } \ } while (0) #define CHECK_RET(cond, return_expr) \ do { \ if (!(cond)) { \ return_expr; \ } \ } while (0) #define LOG_PRINT(message, ...) \ do { \ printf(message, ##__VA_ARGS__); \ } while (0) int64_t GetShapeSize(const std::vector<int64_t> &shape) { int64_t shapeSize = 1; for (auto i : shape) { shapeSize *= i; } return shapeSize; } int Init(int32_t deviceId, aclrtStream *stream) { // 固定写法，acl初始化 auto ret = aclInit(nullptr); CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("aclInit failed. ERROR: %d\n", ret); return ret); ret = aclrtSetDevice(deviceId); CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("aclrtSetDevice failed. ERROR: %d\n", ret); return ret); ret = aclrtCreateStream(stream); CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("aclrtCreateStream failed. ERROR: %d\n", ret); return ret); return 0; } template <typename T> int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &shape, void **deviceAddr, aclDataType dataType, aclTensor **tensor) { auto size = GetShapeSize(shape) * sizeof(T); // 调用aclrtMalloc申请device侧内存 auto ret = aclrtMalloc(deviceAddr, size, ACL_MEM_MALLOC_HUGE_FIRST); CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("aclrtMalloc failed. ERROR: %d\n", ret); return ret); // 调用aclrtMemcpy将host侧数据复制到device侧内存上 ret = aclrtMemcpy(*deviceAddr, size, hostData.data(), size, ACL_MEMCPY_HOST_TO_DEVICE); CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("aclrtMemcpy failed. ERROR: %d\n", ret); return ret); // 计算连续tensor的strides std::vector<int64_t> strides(shape.size(), 1); for (int64_t i = shape.size() - 2; i >= 0; i--) { strides[i] = shape[i + 1] * strides[i + 1]; } // 调用aclCreateTensor接口创建aclTensor *tensor = aclCreateTensor(shape.data(), shape.size(), dataType, strides.data(), 0, aclFormat::ACL_FORMAT_ND, shape.data(), shape.size(), *deviceAddr); return 0; } template <typename T> void printTensor(std::vector<T> tensorData, int64_t tensorSize) { for (int64_t i = 0; i < tensorSize; i++) { std::cout << tensorData[i] << " "; } std::cout << std::endl; } int main(int argc, char **argv) { int deviceId = 0; aclrtStream stream; auto ret = Init(deviceId, &stream); CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("Init acl failed. ERROR: %d\n", ret); return ret); int64_t signalLen = 128; // 26208 int64_t kernelLen = 32; int64_t batchCount = 2; // 768 std::vector<std::complex<half>> tensorSignalData; tensorSignalData.reserve(signalLen * batchCount); std::vector<half> tensorKernelData; tensorKernelData.reserve(kernelLen); for (int64_t i = 0; i < signalLen * batchCount; i++) { tensorSignalData[i] = {(half)1.0, (half)1.0}; } for (int64_t i = 0; i < kernelLen; i++) { tensorKernelData[i] = (half)(1.0 + i); // tensorKernelData[i] = 1.0; } std::vector<std::complex<half>> tensorOutData; tensorOutData.reserve(signalLen * batchCount); for (int64_t i = 0; i < signalLen * batchCount; i++) { tensorOutData[i] = {(half)-1.0, (half)-1.0}; } std::vector<int64_t> signalShape = {batchCount, signalLen}; std::vector<int64_t> kernelShape = {kernelLen}; std::vector<int64_t> resultShape = {batchCount, signalLen}; aclTensor *signal = nullptr; aclTensor *kernel = nullptr; aclTensor *output = nullptr; void *signalDeviceAddr = nullptr; void *kernelDeviceAddr = nullptr; void *outputDeviceAddr = nullptr; ret = CreateAclTensor<std::complex<half>>( tensorSignalData, signalShape, &signalDeviceAddr, aclDataType::ACL_COMPLEX32, &signal); CHECK_RET(ret == ::ACL_SUCCESS, return ret); ret = CreateAclTensor<half>( tensorKernelData, kernelShape, &kernelDeviceAddr, aclDataType::ACL_FLOAT16, &kernel); CHECK_RET(ret == ::ACL_SUCCESS, return ret); ret = CreateAclTensor<std::complex<half>>( tensorOutData, resultShape, &outputDeviceAddr, aclDataType::ACL_COMPLEX32, &output); CHECK_RET(ret == ::ACL_SUCCESS, return ret); size_t lwork = 0; AsdSip::asdConvolveGetWorkspaceSize(signalLen, kernelLen, lwork); void *buffer = nullptr; std::cout << "lwork = " << lwork << std::endl; if (lwork > 0) { ret = aclrtMalloc(&buffer, static_cast<int64_t>(lwork), ACL_MEM_MALLOC_HUGE_FIRST); CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("allocate workspace failed. ERROR: %d\n", ret); return ret); } ASD_STATUS_CHECK(AsdSip::asdConvolve(signal, kernel, output, asdConvolveMode_t::ASD_CONVOLVE_SAME, stream, buffer)); ret = aclrtSynchronizeStream(stream); CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("aclrtSynchronizeStream failed. ERROR: %d\n", ret); return ret); ret = aclrtMemcpy(tensorOutData.data(), signalLen * batchCount * sizeof(std::complex<half>), outputDeviceAddr, signalLen * batchCount * sizeof(std::complex<half>), ACL_MEMCPY_DEVICE_TO_HOST); CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("copy result from device to host failed. ERROR: %d\n", ret); return ret); std::cout << "------- result -------" << std::endl; for (int batchIdx = 0; batchIdx < batchCount; batchIdx++) { for (int i = 0; i < signalLen; i++) { std::cout << "(" << (float)tensorOutData[batchIdx * signalLen + i].real() << "," << (float)tensorOutData[batchIdx * signalLen + i].imag() << ")" << " "; } std::cout << std::endl; } aclDestroyTensor(signal); aclDestroyTensor(kernel); aclDestroyTensor(output); aclrtFree(signalDeviceAddr); aclrtFree(kernelDeviceAddr); aclrtFree(outputDeviceAddr); aclrtDestroyStream(stream); aclrtResetDevice(deviceId); aclFinalize(); return 0; }

【免费下载链接】sip本项目是CANN提供的一款高效、可靠的高性能信号处理算子加速库，基于华为Ascend AI处理器，专门为信号处理领域而设计。项目地址: https://gitcode.com/cann/sip