CANN/sip BLAS点积算子文档

Dot【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产品支持情况产品是否支持Atlas 200I/500 A2 推理产品×Atlas 推理系列产品×Atlas 训练系列产品×Atlas A3 训练系列产品/Atlas A3 推理系列产品√Atlas A2 训练系列产品/Atlas A2 推理系列产品√Ascend 950PR/Ascend 950DT×功能说明接口功能asdBlasMakeDotPlan初始化该句柄对应的Dot算子配置。asdBlasSdot计算两个实数向量的点积。asdBlasCdotu计算两个复数向量的点积。asdBlasCdotc计算一个复数向量取共轭后和另一个复数向量的点积。计算公式asdBlasSdot的公式$$ result\sum _{i1}^n(x[i] * y[i]) $$示例 输入“x”为 [1.0, 2.0] 输入“y”为 [1.0, 2.0] 调用asdBlasSdot算子后输出“result”为 5.0asdBlasCdotu的公式$$ result\sum _{i1}^n(conj(x[i]) * y[i]) $$ 其中x[i]和y[i]是复数。 示例 输入“x”为 [ 0.15540.8840j, -0.3564-0.2552j] 输入“y”为 [-0.14041.3380j, -0.48760.1842j] 调用asdBlasCdotu算子后输出“result”为 1.2877-0.1420jasdBlasCdotc的公式$$ result\sum _{i1}^n(conj(x[i]) * y[i]) $$ 其中x[i]和y[i]是复数,conj共轭操作。 示例 输入“x”为 [ 0.15540.8840j, -0.3564-0.2552j] 输入“y”为 [-0.14041.3380j, -0.48760.1842j] 调用asdBlasCdotc算子后输出“result”为 1.2877-0.1420j函数原型AspbStatus asdBlasMakeDotPlan( asdBlasHandle handle)AspbStatus asdBlasSdot( asdBlasHandle handle, const int64_t n, aclTensor * x, const int64_t incx, aclTensor * y, const int64_t incy, aclTensor * result)AspbStatus asdBlasCdotu( asdBlasHandle handle, const int64_t n, aclTensor * x, const int64_t incx, aclTensor * y, const int64_t incy, aclTensor * result)AspbStatus asdBlasCdotc( asdBlasHandle handle, const int64_t n, aclTensor * x, const int64_t incx, aclTensor * y, const int64_t incy, aclTensor * result)asdBlasMakeDotPlan参数说明参数名输入/输出描述handleasdBlasHandle输入算子的句柄返回值返回状态码具体参见SiP返回码。asdBlasSdot asdBlasCdotu asdBlasCdotc参数说明参数名输入/输出描述handleasdBlasHandle输入算子的句柄nint64_t输入向量x或向量y中的元素个数。xaclTensor *输入对应公式中的x。asdBlasSdot支持的数据类型支持FLOAT32。asdBlasCdotu asdBlasCdotc支持的数据类型支持COMPLEX64。数据格式支持ND。shape为[n]。incxint64_t输入向量x相邻元素间的内存地址偏移量当前约束为1。yaclTensor *输入对应公式中的y。asdBlasSdot支持的数据类型支持FLOAT32。asdBlasCdotu asdBlasCdotc支持的数据类型支持COMPLEX64。数据格式支持ND。shape为[n]。incyint64_t输入向量y相邻元素间的内存地址偏移量当前约束为1。resultaclTensor *输出表示输出的结果对应公式中的result。数据类型支持FLOAT32只包含一个元素。数据格式支持ND。shape为[1]。返回值返回状态码具体参见SiP返回码。约束说明输入的元素个数n当前覆盖支持[16.71e06]。算子输入shape为[n]输出shape为[1]。算子实际计算时不支持ND高维度运算不支持维度≥3的运算。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。asdBlasSdot#include iostream #include vector #include asdsip.h #include acl/acl.h #include acl_meta.h using namespace AsdSip; #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) #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::vectorint64_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::vectorT hostData, const std::vectorint64_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::vectorint64_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(int argc, char **argv) { // 设置算子使用的device id int deviceId 0; //固定写法创造执行流 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 int64_t n 5; int64_t incx 1; int64_t incy 1; int64_t xSize 5; std::vectorfloat tensorInXData; tensorInXData.reserve(xSize); for (int64_t i 0; i xSize; i) { tensorInXData[i] 1.0 i; } int64_t ySize 5; std::vectorfloat tensorInYData; tensorInYData.reserve(xSize); for (int64_t i 0; i ySize; i) { tensorInYData[i] 10.0 i; } int64_t resultSize 1; std::vectorfloat resultData; resultData.reserve(resultSize); std::cout ------- input x ------- std::endl; for (int64_t i 0; i xSize; i) { std::cout tensorInXData[i] ; } std::cout std::endl; std::cout ------- input y ------- std::endl; for (int64_t i 0; i ySize; i) { std::cout tensorInYData[i] ; } std::cout std::endl; // 创造输入/输出tensor std::vectorint64_t xShape {xSize}; std::vectorint64_t yShape {ySize}; std::vectorint64_t resultShape {resultSize}; aclTensor *inputX nullptr; aclTensor *inputY nullptr; aclTensor *result nullptr; void *inputXDeviceAddr nullptr; void *inputYDeviceAddr nullptr; void *resultDeviceAddr nullptr; ret CreateAclTensor(tensorInXData, xShape, inputXDeviceAddr, aclDataType::ACL_FLOAT, inputX); CHECK_RET(ret ::ACL_SUCCESS, return ret); ret CreateAclTensor(tensorInYData, yShape, inputYDeviceAddr, aclDataType::ACL_FLOAT, inputY); CHECK_RET(ret ::ACL_SUCCESS, return ret); ret CreateAclTensor(resultData, resultShape, resultDeviceAddr, aclDataType::ACL_FLOAT, result); CHECK_RET(ret ::ACL_SUCCESS, return ret); // 创建算子执行句柄 asdBlasHandle handle; asdBlasCreate(handle); // 创造算子执行所需workspace size_t lwork 0; void *buffer nullptr; asdBlasMakeDotPlan(handle); asdBlasGetWorkspaceSize(handle, lwork); std::cout lwork lwork std::endl; if (lwork 0) { ret aclrtMalloc(buffer, static_castint64_t(lwork), ACL_MEM_MALLOC_HUGE_FIRST); CHECK_RET(ret ::ACL_SUCCESS, LOG_PRINT(allocate workspace failed. ERROR: %d\n, ret); return ret); } asdBlasSetWorkspace(handle, buffer); // 配置算子执行信息 asdBlasSetStream(handle, stream); // 调用接口执行算子固定调用逻辑 ASD_STATUS_CHECK(asdBlasSdot(handle, n, inputX, incx, inputY, incy, result)); asdBlasSynchronize(handle); // 调度算子后销毁算子句柄 asdBlasDestroy(handle); // 将输出tensor的Device侧数据复制到Host侧内存上 ret aclrtMemcpy(resultData.data(), resultSize * sizeof(float), resultDeviceAddr, resultSize * sizeof(float), 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 (int64_t i 0; i 1; i) { std::cout resultData[i] ; } std::cout std::endl; std::cout Execute successfully. std::endl; // 资源释放 aclDestroyTensor(inputX); aclDestroyTensor(inputY); aclDestroyTensor(result); aclrtFree(inputXDeviceAddr); aclrtFree(inputYDeviceAddr); aclrtFree(resultDeviceAddr); aclrtDestroyStream(stream); aclrtResetDevice(deviceId); aclFinalize(); return 0; }asdBlasCdotu#include iostream #include vector #include cmath #include random #include complex #include asdsip.h #include acl/acl.h #include acl_meta.h using namespace AsdSip; #define ASD_STATUS_CHECK(err) \ do { \ AsdSip::AspbStatus err_ (err); \ if (err_ ! AsdSip::ErrorType::ACL_SUCCESS) { \ std::cout Execute failed. std::endl; \ exit(-1); \ } else { \ std::cout Execute successfully. std::endl; \ } \ } while (0) void printTensor(const std::complexfloat *tensorData, int64_t tensorSize) { for (int64_t i 0; i tensorSize; i) { std::cout tensorData[i] ; } std::cout std::endl; } #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::vectorint64_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::vectorT hostData, const std::vectorint64_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::vectorint64_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; } void printTensor(std::vectorstd::complexfloat 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 n 8; int64_t xSize 8; int64_t ySize 8; std::vectorstd::complexfloat tensorInXData; tensorInXData.reserve(xSize); for (int64_t i 0; i xSize; i) { tensorInXData[i] {2.0, (float)(1.0 i)}; } std::vectorstd::complexfloat tensorInYData; tensorInYData.reserve(ySize); for (int64_t i 0; i ySize; i) { tensorInYData[i] {3.0, 4.0}; } int64_t resultSize 1; std::vectorstd::complexfloat resultData; resultData.reserve(resultSize); std::cout ------- input TensorInX ------- std::endl; printTensor(tensorInXData.data(), xSize); std::cout ------- input TensorInY ------- std::endl; printTensor(tensorInYData.data(), ySize); std::vectorint64_t xShape {xSize}; std::vectorint64_t yShape {ySize}; std::vectorint64_t resultShape {resultSize}; aclTensor *inputX nullptr; aclTensor *inputY nullptr; aclTensor *result nullptr; void *inputXDeviceAddr nullptr; void *inputYDeviceAddr nullptr; void *resultDeviceAddr nullptr; ret CreateAclTensor(tensorInXData, xShape, inputXDeviceAddr, aclDataType::ACL_COMPLEX64, inputX); CHECK_RET(ret ::ACL_SUCCESS, return ret); ret CreateAclTensor(tensorInYData, yShape, inputYDeviceAddr, aclDataType::ACL_COMPLEX64, inputY); CHECK_RET(ret ::ACL_SUCCESS, return ret); ret CreateAclTensor(resultData, resultShape, resultDeviceAddr, aclDataType::ACL_COMPLEX64, result); CHECK_RET(ret ::ACL_SUCCESS, return ret); asdBlasHandle handle; asdBlasCreate(handle); size_t lwork 0; void *buffer nullptr; asdBlasMakeDotPlan(handle); asdBlasGetWorkspaceSize(handle, lwork); std::cout lwork lwork std::endl; if (lwork 0) { ret aclrtMalloc(buffer, static_castint64_t(lwork), ACL_MEM_MALLOC_HUGE_FIRST); CHECK_RET(ret ::ACL_SUCCESS, LOG_PRINT(allocate workspace failed. ERROR: %d\n, ret); return ret); } asdBlasSetWorkspace(handle, buffer); asdBlasSetStream(handle, stream); ASD_STATUS_CHECK(asdBlasCdotu(handle, n, inputX, 1, inputY, 1, result)); asdBlasSynchronize(handle); asdBlasDestroy(handle); ret aclrtMemcpy(resultData.data(), resultSize * sizeof(std::complexfloat), resultDeviceAddr, resultSize * sizeof(std::complexfloat), 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; printTensor(resultData.data(), resultSize); aclDestroyTensor(inputX); aclDestroyTensor(inputY); aclDestroyTensor(result); aclrtFree(inputXDeviceAddr); aclrtFree(inputYDeviceAddr); aclrtFree(resultDeviceAddr); aclrtDestroyStream(stream); aclrtResetDevice(deviceId); aclFinalize(); return 0; }asdBlasCdotc#include iostream #include vector #include complex #include cmath #include random #include asdsip.h #include acl/acl.h #include acl_meta.h using namespace AsdSip; #define ASD_STATUS_CHECK(err) \ do { \ AsdSip::AspbStatus err_ (err); \ if (err_ ! AsdSip::ErrorType::ACL_SUCCESS) { \ std::cout Execute failed. std::endl; \ exit(-1); \ } else { \ std::cout Execute successfully. std::endl; \ } \ } while (0) void printTensor(const std::complexfloat *tensorData, int64_t tensorSize) { for (int64_t i 0; i tensorSize; i) { std::cout tensorData[i] ; } std::cout std::endl; } #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::vectorint64_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::vectorT hostData, const std::vectorint64_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::vectorint64_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; } void printTensor(std::vectorstd::complexfloat 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 n 8; int64_t xSize 8; int64_t ySize 8; std::vectorstd::complexfloat tensorInXData; tensorInXData.reserve(xSize); for (int64_t i 0; i xSize; i) { tensorInXData[i] {2.0, (float)(1.0 i)}; } std::vectorstd::complexfloat tensorInYData; tensorInYData.reserve(ySize); for (int64_t i 0; i ySize; i) { tensorInYData[i] {3.0, 4.0}; } int64_t resultSize 1; std::vectorstd::complexfloat resultData; resultData.reserve(resultSize); std::cout ------- input TensorInX ------- std::endl; printTensor(tensorInXData.data(), xSize); std::cout ------- input TensorInY ------- std::endl; printTensor(tensorInYData.data(), ySize); std::vectorint64_t xShape {xSize}; std::vectorint64_t yShape {ySize}; std::vectorint64_t resultShape {resultSize}; aclTensor *inputX nullptr; aclTensor *inputY nullptr; aclTensor *result nullptr; void *inputXDeviceAddr nullptr; void *inputYDeviceAddr nullptr; void *resultDeviceAddr nullptr; ret CreateAclTensor(tensorInXData, xShape, inputXDeviceAddr, aclDataType::ACL_COMPLEX64, inputX); CHECK_RET(ret ::ACL_SUCCESS, return ret); ret CreateAclTensor(tensorInYData, yShape, inputYDeviceAddr, aclDataType::ACL_COMPLEX64, inputY); CHECK_RET(ret ::ACL_SUCCESS, return ret); ret CreateAclTensor(resultData, resultShape, resultDeviceAddr, aclDataType::ACL_COMPLEX64, result); CHECK_RET(ret ::ACL_SUCCESS, return ret); asdBlasHandle handle; asdBlasCreate(handle); size_t lwork 0; void *buffer nullptr; asdBlasMakeDotPlan(handle); asdBlasGetWorkspaceSize(handle, lwork); std::cout lwork lwork std::endl; if (lwork 0) { ret aclrtMalloc(buffer, static_castint64_t(lwork), ACL_MEM_MALLOC_HUGE_FIRST); CHECK_RET(ret ::ACL_SUCCESS, LOG_PRINT(allocate workspace failed. ERROR: %d\n, ret); return ret); } asdBlasSetWorkspace(handle, buffer); asdBlasSetStream(handle, stream); ASD_STATUS_CHECK(asdBlasCdotc(handle, n, inputX, 1, inputY, 1, result)); asdBlasSynchronize(handle); asdBlasDestroy(handle); ret aclrtMemcpy(resultData.data(), resultSize * sizeof(std::complexfloat), resultDeviceAddr, resultSize * sizeof(std::complexfloat), 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; printTensor(resultData.data(), resultSize); aclDestroyTensor(inputX); aclDestroyTensor(inputY); aclDestroyTensor(result); aclrtFree(inputXDeviceAddr); aclrtFree(inputYDeviceAddr); aclrtFree(resultDeviceAddr); 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创作声明：本文部分内容由AI辅助生成（AIGC），仅供参考