CANN/sip二维FFT接口文档

FFT_2D

【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：
  asdFftMakePlan2D：初始化二维FFT配置。
  asdFftExecC2C：执行复数到复数的FFT变换。
  asdFftExecC2R：执行复数到实数的FFT变换。
  asdFftExecR2C：执行实数到复数的FFT变换。

• 计算公式：
  傅里叶变换（Fourier transform）是一种线性积分变换，用于信号在时域和频域之间的变换，在物理学和工程学中有许多应用。对应给定长度为N的信号，其离散形式DFT(Discrete Fourier Transform)表达式如下：

  

  将系数矩阵(NN)和时域信号(N1)看做两个Tensor，在NPU上直接使用矩阵乘，可完成DFT，但时间复杂度太高，因此需要快速傅里叶变换。其基本原理是利用三角函数在复数域的旋转对称性，将序列拆分成子序列，通过蝶形运算以降低计算的复杂度：
  

函数原型

AspbStatus asdFftMakePlan2D( asdFftHandle handle, int64_t fftSizeX, int64_t fftSizeY, asdFftType fftType, asdFftDirection direction, int32_t batchSize)

AspbStatus asdFftExecC2C( asdFftHandle handle, const aclTensor * input, const aclTensor * output)

AspbStatus asdFftExecC2R( asdFftHandle handle, const aclTensor * input, const aclTensor * output)

AspbStatus asdFftExecR2C( asdFftHandle handle, const aclTensor * input, const aclTensor * output)

asdFftMakePlan2D

• 参数说明：

  参数名	输入/输出	描述
  handle（asdFftHandle）	输入	算子的句柄，需要手动申请创建asdFftHandle对象。
  fftSizeX（int64_t）	输入	对应公式中的'M'，FFT信号长度（第一维）。
  fftSizeY（int64_t）	输入	对应公式中的'N'，FFT信号长度（第二维）。
  fftType（asdFftType）	输入	FFT变换类型 ASCEND_FFT_C2C：复数到复数的快速傅里叶变换。 ASCEND_FFT_C2R：复数到实数的快速傅里叶变换。 ASCEND_FFT_R2C：实数到复数的快速傅里叶变换。
  direction（asdFftDirection）	输入	选择FFT执行正向变换或反向变换 ASCEND_FFT_FORWARD：正向快速傅里叶变换。 ASCEND_FFT_INVERSE：逆向快速傅里叶变换。
  batchSize（int32_t）	输入	FFT变换批处理操作中的数据批次数量。

• 返回值：

  返回状态码，具体参见SiP返回码。

asdFftExecC2C

• 参数说明：

  参数名	输入/输出	描述
  handle（asdFftHandle）	输入	算子的句柄，需要手动申请创建asdFftHandle对象。
  input（ const aclTensor *）	输入	对应公式中的'x'。 数据类型支持COMPLEX64。 数据格式支持ND。 输入的shape为（batchSize，fftSizeX，fftSizeY）。
  output（const aclTensor *）	输出	对应公式中的'y'。 数据类型支持COMPLEX64。 数据格式支持ND。 输出的shape为（batchSize，fftSizeX，fftSizeY）。

• 返回值：

  返回状态码，具体参见SiP返回码。

asdFftExecC2R

• 参数说明：

  参数名	输入/输出	描述
  handle（asdFftHandle）	输入	算子的句柄，需要手动申请创建asdFftHandle对象。
  input（ const aclTensor *）	输入	对应公式中的'x'。 数据类型支持COMPLEX64。 数据格式支持ND。 输入的shape为（batchSize，fftSizeX，fftSizeY/2+1）。
  output（const aclTensor *）	输出	对应公式中的'y'。 数据类型支持FLOAT32。 数据格式支持ND。 输入的shape为（batchSize，fftSizeX，fftSizeY）。

• 返回值：

  返回状态码，具体参见SiP返回码。

asdFftExecR2C

• 参数说明：

  参数名	输入/输出	描述
  handle（asdFftHandle）	输入	算子的句柄，需要手动申请创建asdFftHandle对象。
  input（ const aclTensor *）	输入	对应公式中的'x'。 数据类型支持FLOAT32。 数据格式支持ND。 输入的shape为（batchSize，fftSizeX，fftSizeY）。
  output（const aclTensor *）	输出	对应公式中的'y'。 数据类型支持COMPLEX64。 数据格式支持ND。 输入的shape为（batchSize，fftSizeX，fftSizeY/2+1）。

• 返回值：

  返回状态码，具体参见SiP返回码。

约束说明

asdFftMakePlan2D

• fftSizeX、fftSizeY需保证不超过$2^{27}$且分解质因数后不包含超过199的质因子。
• batchSize在存储允许范围内应无额外约束。
• 输入的元素个数理论支持[1，$2^{30}$]。
• 输入的元素不支持inf、-inf和nan，如果输入中包含这些值， 那么结果为未定义。

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。

• C2C_2D

#include <iostream> #include <vector> #include "asdsip.h" #include "acl/acl.h" #include "aclnn/acl_meta.h" using namespace AsdSip; #define CHECK_RET(cond, return_expr) \ do { \ if (!(cond)) { \ return_expr; \ } \ } while (0) #define LOG_PRINT(message, ...) \ do { \ printf(message, ##__VA_ARGS__); \ } while (0) #define ASD_STATUS_CHECK(err) \ do { \ AsdSip::AspbStatus err_ = (err); \ if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \ std::cout << "Execute failed." << std::endl; \ exit(-1); \ } \ } 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，AscendCL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; } int main() { int32_t 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); // 创造tensor的Host侧数据 // int batch = 2, Nfft1 = 199 * 199, Nfft2 = 4096; int batch = 2, Nfft1 = 64, Nfft2 = 64; // core dd const int64_t inSignal = Nfft2; const int64_t outSignal = Nfft2; const int64_t tensorInSize = batch * Nfft1 * inSignal; const int64_t tensorOutSize = batch * Nfft1 * outSignal; std::vector<int64_t> selfShape = {batch, Nfft1, inSignal}; std::vector<int64_t> outShape = {batch, Nfft1, outSignal}; std::vector<std::complex<float>> inputHostData(tensorInSize, std::complex<float>(0, 0)); for (int i = 0; i < tensorInSize; i++) { inputHostData[i] = std::complex<float>(i, i + 1); } std::vector<std::complex<float>> outHostData(tensorOutSize, std::complex<float>(0, 0)); void *inputDeviceAddr = nullptr; void *outDeviceAddr = nullptr; aclTensor *input = nullptr; aclTensor *out = nullptr; ret = CreateAclTensor(inputHostData, selfShape, &inputDeviceAddr, aclDataType::ACL_COMPLEX64, &input); CHECK_RET(ret == ::ACL_SUCCESS, return ret); ret = CreateAclTensor(outHostData, outShape, &outDeviceAddr, aclDataType::ACL_COMPLEX64, &out); CHECK_RET(ret == ::ACL_SUCCESS, return ret); asdFftHandle handle; asdFftCreate(handle); asdFftMakePlan2D(handle, Nfft1, Nfft2, asdFftType::ASCEND_FFT_C2C, asdFftDirection::ASCEND_FFT_FORWARD, batch); size_t work_size; asdFftGetWorkspaceSize(handle, work_size); void *workspaceAddr = nullptr; if (work_size > 0) { ret = aclrtMalloc(&workspaceAddr, static_cast<int64_t>(work_size), ACL_MEM_MALLOC_HUGE_FIRST); CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("allocate workspace failed. ERROR: %d\n", ret); return ret); } asdFftSetWorkspace(handle, (uint8_t *)workspaceAddr); asdFftSetStream(handle, stream); ASD_STATUS_CHECK(asdFftExecC2C(handle, input, out)); ret = aclrtSynchronizeStream(stream); CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("aclrtSynchronizeStream failed. ERROR: %d\n", ret); return ret); asdFftDestroy(handle); auto size = GetShapeSize(outShape); std::vector<std::complex<float>> outData(size, 0); ret = aclrtMemcpy(outData.data(), outData.size() * sizeof(outData[0]), outDeviceAddr, size * sizeof(outData[0]), ACL_MEMCPY_DEVICE_TO_HOST); // 打印输出tensor值中前16个 for (int64_t i = 0; i < 16; i++) { std::cout << static_cast<std::complex<float>>(outData[i]) << "\t"; } std::cout << "\nend result" << std::endl; std::cout << "Execute successfully." << std::endl; aclDestroyTensor(input); aclDestroyTensor(out); aclrtFree(inputDeviceAddr); aclrtFree(outDeviceAddr); if (work_size > 0) { aclrtFree(workspaceAddr); } aclrtDestroyStream(stream); aclrtResetDevice(deviceId); aclFinalize(); return 0; }

• C2R_2D

#include <iostream> #include <vector> #include "asdsip.h" #include "acl/acl.h" #include "aclnn/acl_meta.h" using namespace AsdSip; #define CHECK_RET(cond, return_expr) \ do { \ if (!(cond)) { \ return_expr; \ } \ } while (0) #define LOG_PRINT(message, ...) \ do { \ printf(message, ##__VA_ARGS__); \ } while (0) #define ASD_STATUS_CHECK(err) \ do { \ AsdSip::AspbStatus err_ = (err); \ if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \ std::cout << "Execute failed." << std::endl; \ exit(-1); \ } \ } 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，AscendCL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; } int main() { int32_t 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); // 创造tensor的Host侧数据 // int batch = 32, Nfft = 128; // int batch = 32, Nfft = 8192; // int batch = 32, Nfft = 15000; // 创造tensor的Host侧数据 int batch = 2, Nfft1 = 128, Nfft2 = 128; const int64_t inSignal = Nfft2 / 2 + 1; const int64_t outSignal = Nfft2; const int64_t tensorInSize = batch * Nfft1 * inSignal; const int64_t tensorOutSize = batch * Nfft1 * outSignal; std::vector<int64_t> selfShape = {batch, Nfft1, inSignal}; std::vector<int64_t> outShape = {batch, Nfft1, outSignal}; std::vector<std::complex<float>> inputHostData(tensorInSize, std::complex<float>(0, 0)); for (int i = 0; i < tensorInSize; i++) { inputHostData[i] = std::complex<float>(i, i + 1); } std::vector<float> outHostData(tensorOutSize, 0); void *inputDeviceAddr = nullptr; void *outDeviceAddr = nullptr; aclTensor *input = nullptr; aclTensor *out = nullptr; ret = CreateAclTensor(inputHostData, selfShape, &inputDeviceAddr, aclDataType::ACL_COMPLEX64, &input); CHECK_RET(ret == ::ACL_SUCCESS, return ret); ret = CreateAclTensor(outHostData, outShape, &outDeviceAddr, aclDataType::ACL_FLOAT, &out); CHECK_RET(ret == ::ACL_SUCCESS, return ret); asdFftHandle handle; asdFftCreate(handle); asdFftMakePlan2D(handle, Nfft1, Nfft2, asdFftType::ASCEND_FFT_C2R, asdFftDirection::ASCEND_FFT_FORWARD, batch); size_t work_size; asdFftGetWorkspaceSize(handle, work_size); void *workspaceAddr = nullptr; if (work_size > 0) { ret = aclrtMalloc(&workspaceAddr, static_cast<int64_t>(work_size), ACL_MEM_MALLOC_HUGE_FIRST); CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("allocate workspace failed. ERROR: %d\n", ret); return ret); } asdFftSetWorkspace(handle, (uint8_t *)workspaceAddr); asdFftSetStream(handle, stream); ASD_STATUS_CHECK(asdFftExecC2R(handle, input, out)); ret = aclrtSynchronizeStream(stream); CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("aclrtSynchronizeStream failed. ERROR: %d\n", ret); return ret); asdFftDestroy(handle); auto size = GetShapeSize(outShape); std::vector<float> outData(size, 0); ret = aclrtMemcpy(outData.data(), outData.size() * sizeof(outData[0]), outDeviceAddr, size * sizeof(outData[0]), ACL_MEMCPY_DEVICE_TO_HOST); // 打印输出tensor值中前16个 for (int64_t i = 0; i < 16; i++) { std::cout << static_cast<float>(outData[i]) << "\t"; } std::cout << "\nend result" << std::endl; std::cout << "Execute successfully." << std::endl; aclDestroyTensor(input); aclDestroyTensor(out); aclrtFree(inputDeviceAddr); aclrtFree(outDeviceAddr); if (work_size > 0) { aclrtFree(workspaceAddr); } aclrtDestroyStream(stream); aclrtResetDevice(deviceId); aclFinalize(); return 0; }

• R2C_2D

#include <iostream> #include <vector> #include "asdsip.h" #include "acl/acl.h" #include "aclnn/acl_meta.h" using namespace AsdSip; #define CHECK_RET(cond, return_expr) \ do { \ if (!(cond)) { \ return_expr; \ } \ } while (0) #define LOG_PRINT(message, ...) \ do { \ printf(message, ##__VA_ARGS__); \ } while (0) #define ASD_STATUS_CHECK(err) \ do { \ AsdSip::AspbStatus err_ = (err); \ if (err_ != AsdSip::ErrorType::ACL_SUCCESS) { \ std::cout << "Execute failed." << std::endl; \ exit(-1); \ } \ } 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，AscendCL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; } int main() { int32_t 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); // 创造tensor的Host侧数据 int batch = 2, Nfft1 = 1024, Nfft2 = 4096; const int64_t inSignal = Nfft2; const int64_t outSignal = Nfft2 / 2 + 1; const int64_t tensorInSize = batch * Nfft1 * inSignal; const int64_t tensorOutSize = batch * Nfft1 * outSignal; std::vector<int64_t> selfShape = {batch, Nfft1, inSignal}; std::vector<int64_t> outShape = {batch, Nfft1, outSignal}; std::vector<float> inputHostData(tensorInSize, 0); for (int i = 0; i < tensorInSize; i++) { inputHostData[i] = i; } std::vector<std::complex<float>> outHostData(tensorOutSize, std::complex<float>(0, 0)); void *inputDeviceAddr = nullptr; void *outDeviceAddr = nullptr; aclTensor *input = nullptr; aclTensor *out = nullptr; ret = CreateAclTensor(inputHostData, selfShape, &inputDeviceAddr, aclDataType::ACL_FLOAT, &input); CHECK_RET(ret == ::ACL_SUCCESS, return ret); ret = CreateAclTensor(outHostData, outShape, &outDeviceAddr, aclDataType::ACL_COMPLEX64, &out); CHECK_RET(ret == ::ACL_SUCCESS, return ret); asdFftHandle handle; asdFftCreate(handle); asdFftMakePlan2D(handle, Nfft1, Nfft2, asdFftType::ASCEND_FFT_R2C, asdFftDirection::ASCEND_FFT_FORWARD, batch); size_t work_size; asdFftGetWorkspaceSize(handle, work_size); void *workspaceAddr = nullptr; if (work_size > 0) { ret = aclrtMalloc(&workspaceAddr, static_cast<int64_t>(work_size), ACL_MEM_MALLOC_HUGE_FIRST); CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("allocate workspace failed. ERROR: %d\n", ret); return ret); } asdFftSetWorkspace(handle, (uint8_t *)workspaceAddr); asdFftSetStream(handle, stream); ASD_STATUS_CHECK(asdFftExecR2C(handle, input, out)); ret = aclrtSynchronizeStream(stream); CHECK_RET(ret == ::ACL_SUCCESS, LOG_PRINT("aclrtSynchronizeStream failed. ERROR: %d\n", ret); return ret); asdFftDestroy(handle); auto size = GetShapeSize(outShape); std::vector<std::complex<float>> outData(size, 0); ret = aclrtMemcpy(outData.data(), outData.size() * sizeof(outData[0]), outDeviceAddr, size * sizeof(outData[0]), ACL_MEMCPY_DEVICE_TO_HOST); // 打印输出tensor值中前16个 for (int64_t i = 0; i < 16; i++) { std::cout << static_cast<std::complex<float>>(outData[i]) << "\t"; } std::cout << "\nend result" << std::endl; std::cout << "Execute successfully." << std::endl; aclDestroyTensor(input); aclDestroyTensor(out); aclrtFree(inputDeviceAddr); aclrtFree(outDeviceAddr); if (work_size > 0) { aclrtFree(workspaceAddr); } 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