Strmm【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asdBlasMakeStrmmPlan初始化该句柄对应的Strmm算子配置。asdBlasStrmm单精度其功能是将一个三角矩阵A乘一个矩阵B得到一个新的矩阵C。计算公式 $$ c \begin{cases} alphaop(A)B if side ASDBLAS_SIDE_LEFT \ alphaBop(A) if side ASDBLAS_SIDE_RIGHT \ \end{cases} $$示例输入“A”为[ [ 1, 0 ], [ 3, 4 ] ]输入“B”为[ [ 1, 2 ], [ 3, 4 ] ]输入“side”为L“uplo”为L输入“trans”为N输入“diag”为N。输入“n”为2输入“lda”为2输入“ldb”为2输入“ldc”为2。输入“alpha”为2.345。调用“asdBlasStrmm”算子后输出“C”为[ [ 2.3450, 4.6900],[35.1750, 51.5900] ]函数原型AspbStatus asdBlasMakeStrmmPlan( asdBlasHandle handle)AspbStatus asdBlasStrmm( asdBlasHandle handle, asdBlasSideMode_t side, asdBlasFillMode_t uplo, asdBlasOperation_t trans, asdBlasDiagType_t diag, const int64_t m, const int64_t n, const float * alpha, aclTensor * A, const int64_t lda, aclTensor * B, const int64_t ldb, aclTensor * C, const int64_t ldc)asdBlasMakeStrmmPlan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。asdBlasStrmm参数说明参数名输入/输出描述handleasdBlasHandle输入算子的句柄sideasdBlasSideMode_t输入指定矩阵A是乘法左侧还是右侧。。ASDBLAS_SIDE_LEFT:左侧ASDBLAS_SIDE_RIGHT:右侧uploasdBlasFillMode_t输入指定矩阵A的存储格式。ASDBLAS_FILL_MODE_LOWER:下三角ASDBLAS_FILL_MODE_UPPER:上三角diagasdBlasDiagType_t输入指定是否假定矩阵A的对角线元素为1。ASDBLAS_DIAG_NON_UNIT:不假定为1ASDBLAS_DIAG_UNIT:假定为1transasdBlasOperation_t输入指定是否对矩阵A进行转置。ASDBLAS_OP_N:不转置ASDBLAS_OP_T:转置mint64_t输入矩阵B和C的行数。nint64_t输入矩阵B和C的列数。alphafloat *输入公式中的alpha用于计算矩阵乘法的系数。AaclTensor *输入对应公式中的A。数据类型支持FLOAT32。数据格式支持ND。当A为乘法左矩阵时shape为[mm]当A为乘法右矩阵时shape为[nn]ldaint64_t输入表示张量A中元素的间隔当前约束为m/n当sideASDBLAS_SIDE_LEFT时为msideASDBLAS_SIDE_RIGHT时为n。BaclTensor *输入对应公式中的B。数据类型支持FLOAT32。数据格式支持ND。shape为[mn]ldbint64_t输入表示张量B中元素的间隔当前约束为m。CaclTensor *输出对应公式中的C。数据类型支持FLOAT32。数据格式支持ND。shape为[mn]ldcint64_t输入表示张量C中元素的间隔当前约束为m。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mn当前覆盖支持[18193]。当side ASDBLAS_SIDE_LEFT时算子输入shape为[mm]、[mn]输出shape为[mn]。当side ASDBLAS_SIDE_RIGHT时算子输入shape为[nn]、[mn]输出shape为[mn]。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。#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) { 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); asdBlasSideMode_t side asdBlasSideMode_t::ASDBLAS_SIDE_LEFT; asdBlasFillMode_t uplo asdBlasFillMode_t::ASDBLAS_FILL_MODE_LOWER; asdBlasOperation_t trans asdBlasOperation_t::ASDBLAS_OP_N; asdBlasDiagType_t diag asdBlasDiagType_t::ASDBLAS_DIAG_NON_UNIT; const int64_t m 5; const int64_t n 5; float alpha 1.0; int64_t lda m; int64_t ldb m; int64_t ldc m; const int64_t tensorASize m * m; std::vectorfloat tensorInAData(tensorASize, 0.0); for (int64_t i 0; i m; i) { for (int64_t j 0; j m; j) { tensorInAData[m * i j] i; } } const int64_t tensorBSize m * n; std::vectorfloat tensorInBData(tensorBSize, 0.0); for (int64_t i 0; i m; i) { for (int64_t j 0; j n; j) { tensorInBData[n * i j] i; } } const int64_t tensorCSize m * n; std::vectorfloat tensorCData(tensorCSize, 0.0); std::cout side static_castint32_t(side) std::endl; std::cout uplo static_castint32_t(uplo) std::endl; std::cout trans static_castint32_t(trans) std::endl; std::cout diag static_castint32_t(diag) std::endl; std::cout ------- input A ------- std::endl; for (int64_t i 0; i m; i) { for (int64_t j 0; j m; j) std::cout tensorInAData[i * m j] ; std::cout std::endl; } std::cout ------- input B ------- std::endl; for (int64_t i 0; i m; i) { for (int64_t j 0; j n; j) std::cout tensorInBData[i * n j] ; std::cout std::endl; } std::vectorint64_t aShape {tensorASize}; std::vectorint64_t bShape {tensorBSize}; std::vectorint64_t cShape {tensorCSize}; aclTensor *inputA nullptr; aclTensor *inputB nullptr; aclTensor *outputC nullptr; void *inputADeviceAddr nullptr; void *inputBDeviceAddr nullptr; void *outputCDeviceAddr nullptr; ret CreateAclTensor(tensorInAData, aShape, inputADeviceAddr, aclDataType::ACL_FLOAT, inputA); CHECK_RET(ret ::ACL_SUCCESS, return ret); ret CreateAclTensor(tensorInBData, bShape, inputBDeviceAddr, aclDataType::ACL_FLOAT, inputB); CHECK_RET(ret ::ACL_SUCCESS, return ret); ret CreateAclTensor(tensorCData, cShape, outputCDeviceAddr, aclDataType::ACL_FLOAT, outputC); CHECK_RET(ret ::ACL_SUCCESS, return ret); asdBlasHandle handle; asdBlasCreate(handle); size_t lwork 0; void *buffer nullptr; asdBlasMakeStrmmPlan(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( asdBlasStrmm(handle, side, uplo, trans, diag, m, n, alpha, inputA, lda, inputB, ldb, outputC, ldc)); asdBlasSynchronize(handle); asdBlasDestroy(handle); ret aclrtMemcpy(tensorCData.data(), tensorCSize * sizeof(float), outputCDeviceAddr, tensorCSize * 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 ------- output C ------- std::endl; for (int64_t i 0; i m; i) { for (int64_t j 0; j n; j) { std::cout tensorCData[i * n j] ; } std::cout std::endl; } std::cout Execute successfully. std::endl; aclDestroyTensor(inputA); aclDestroyTensor(inputB); aclDestroyTensor(outputC); aclrtFree(inputADeviceAddr); aclrtFree(inputBDeviceAddr); aclrtFree(outputCDeviceAddr); 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