aclnnGroupNormalizationGrad【免费下载链接】ops-nn本项目是CANN提供的神经网络类计算算子库实现网络在NPU上加速计算。项目地址: https://gitcode.com/cann/ops-nn产品支持情况产品是否支持Atlas A2 训练系列产品/Atlas 800I A2 推理产品√功能说明算子功能完成 Group Normalization 的反向。计算公式$$ \hat{x} (x - mean) \cdot rstd $$$$ s_1 \sum(dy \cdot gamma), \quad s_2 \sum(dy \cdot gamma \cdot \hat{x}) $$$$ dx \frac{rstd}{M} \cdot gamma \cdot (M \cdot dy - s_1 - \hat{x} \cdot s_2) $$函数原型每个算子分为两段式接口必须先调用aclnnGroupNormalizationGradGetWorkspaceSize接口获取入参并根据计算流程计算所需workspace大小再调用aclnnGroupNormalizationGrad接口执行计算。aclnnStatus aclnnGroupNormalizationGradGetWorkspaceSize( const aclTensor *x, const aclTensor *dy, const aclTensor *gamma, const aclTensor *mean, const aclTensor *rstd, aclTensor *dx, uint64_t *workspaceSize, aclOpExecutor **executor)aclnnStatus aclnnGroupNormalizationGrad( void *workspace, uint64_t workspaceSize, aclOpExecutor *executor, const aclrtStream stream)aclnnGroupNormalizationGradGetWorkspaceSize参数说明参数名输入/输出描述使用说明数据类型数据格式维度(shape)非连续Tensorx输入公式中的 x前向输入。dtype 需与 dy 保持一致。shape 需与 dy 相同。FLOAT、BFLOAT16、FLOAT16ND3-8×dy输入公式中的 dy上游梯度。数据类型与 x 的数据类型满足互推导关系。FLOAT、BFLOAT16、FLOAT16ND3-8×gamma输入已广播到 [N, G, M] 的缩放系数。dtype 需与 x 保持一致。shape 需与 x 相同。FLOAT、BFLOAT16、FLOAT16ND3-8×mean输入每个 group 的均值形状为 [N, G]。dtype 需与 x 保持一致。FLOAT、BFLOAT16、FLOAT16ND2×rstd输入每个 group 的标准差倒数形状为 [N, G]。dtype 需与 x 保持一致。FLOAT、BFLOAT16、FLOAT16ND2×dx输出公式中的 dx输入梯度。dtype 需与 x 相同。shape 需与 x 相等。FLOAT、BFLOAT16、FLOAT16ND3-8×workspaceSize输出返回需要在 Device 侧申请的 workspace 大小。-----executor输出返回 op 执行器包含了算子计算流程。-----返回值aclnnStatus返回状态码具体参见aclnn返回码。 第一段接口会完成入参校验出现以下场景时报错返回码错误码描述ACLNN_ERR_PARAM_NULLPTR161001传入的 x、dy、gamma、mean 或 rstd 是空指针。ACLNN_ERR_PARAM_INVALID161002x、dy、gamma、mean 或 rstd 的数据类型不在支持的范围之内。x、dy、gamma、mean 或 rstd 的 shape 超过 8 维或 x、dy、gamma 的 shape 低于 3 维。x、dy、gamma 与 dx 数据类型不一致。x、dy、gamma 的 shape 不一致。aclnnGroupNormalizationGrad参数说明参数名输入/输出描述workspace输入在 Device 侧申请的 workspace 内存地址。workspaceSize输入在 Device 侧申请的 workspace 大小由第一段接口 aclnnGroupNormalizationGradGetWorkspaceSize 获取。executor输入op 执行器包含了算子计算流程。stream输入指定执行任务的 Stream。返回值aclnnStatus返回状态码具体参见aclnn返回码。约束说明无。调用示例示例代码如下仅供参考具体编译和执行过程请参考编译与运行样例。#include iostream #include vector #include acl/acl.h #include aclnn_group_normalization_grad.h #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) { 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); 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); 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); 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]; } *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); std::vectorint64_t xShape {2, 4, 128}; std::vectorint64_t meanShape {2, 4}; void* xDeviceAddr nullptr; void* dyDeviceAddr nullptr; void* gammaDeviceAddr nullptr; void* meanDeviceAddr nullptr; void* rstdDeviceAddr nullptr; void* dxDeviceAddr nullptr; aclTensor* x nullptr; aclTensor* dy nullptr; aclTensor* gamma nullptr; aclTensor* mean nullptr; aclTensor* rstd nullptr; aclTensor* dx nullptr; auto size GetShapeSize(xShape); std::vectorfloat xHostData(size); std::vectorfloat dyHostData(size); std::vectorfloat gammaHostData(size); std::vectorfloat dxHostData(size, 0.0f); auto meanSize GetShapeSize(meanShape); std::vectorfloat meanHostData(meanSize); std::vectorfloat rstdHostData(meanSize); for (int64_t i 0; i size; i) { xHostData[i] static_castfloat(i % 128) / 128.0f; dyHostData[i] 0.5f; gammaHostData[i] 1.0f; } for (int64_t i 0; i meanSize; i) { meanHostData[i] 0.0f; rstdHostData[i] 1.0f; } ret CreateAclTensor(xHostData, xShape, xDeviceAddr, aclDataType::ACL_FLOAT, x); CHECK_RET(ret ACL_SUCCESS, return ret); ret CreateAclTensor(dyHostData, xShape, dyDeviceAddr, aclDataType::ACL_FLOAT, dy); CHECK_RET(ret ACL_SUCCESS, return ret); ret CreateAclTensor(gammaHostData, xShape, gammaDeviceAddr, aclDataType::ACL_FLOAT, gamma); CHECK_RET(ret ACL_SUCCESS, return ret); ret CreateAclTensor(meanHostData, meanShape, meanDeviceAddr, aclDataType::ACL_FLOAT, mean); CHECK_RET(ret ACL_SUCCESS, return ret); ret CreateAclTensor(rstdHostData, meanShape, rstdDeviceAddr, aclDataType::ACL_FLOAT, rstd); CHECK_RET(ret ACL_SUCCESS, return ret); ret CreateAclTensor(dxHostData, xShape, dxDeviceAddr, aclDataType::ACL_FLOAT, dx); CHECK_RET(ret ACL_SUCCESS, return ret); uint64_t workspaceSize 0; aclOpExecutor* executor; ret aclnnGroupNormalizationGradGetWorkspaceSize(x, dy, gamma, mean, rstd, dx, workspaceSize, executor); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(aclnnGroupNormalizationGradGetWorkspaceSize failed. ERROR: %d\n, ret); return ret); void* workspaceAddr nullptr; if (workspaceSize 0) { ret aclrtMalloc(workspaceAddr, workspaceSize, ACL_MEM_MALLOC_HUGE_FIRST); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(allocate workspace failed. ERROR: %d\n, ret); return ret); } ret aclnnGroupNormalizationGrad(workspaceAddr, workspaceSize, executor, stream); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(aclnnGroupNormalizationGrad failed. ERROR: %d\n, ret); return ret); ret aclrtSynchronizeStream(stream); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(aclrtSynchronizeStream failed. ERROR: %d\n, ret); return ret); size GetShapeSize(xShape); std::vectorfloat resultData(size, 0); ret aclrtMemcpy(resultData.data(), resultData.size() * sizeof(resultData[0]), dxDeviceAddr, size * sizeof(resultData[0]), ACL_MEMCPY_DEVICE_TO_HOST); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(copy result failed. ERROR: %d\n, ret); return ret); for (int64_t i 0; i size; i) { LOG_PRINT(result[%ld] is: %f\n, i, resultData[i]); } aclDestroyTensor(x); aclDestroyTensor(dy); aclDestroyTensor(gamma); aclDestroyTensor(mean); aclDestroyTensor(rstd); aclDestroyTensor(dx); aclrtFree(xDeviceAddr); aclrtFree(dyDeviceAddr); aclrtFree(gammaDeviceAddr); aclrtFree(meanDeviceAddr); aclrtFree(rstdDeviceAddr); aclrtFree(dxDeviceAddr); if (workspaceSize 0) { aclrtFree(workspaceAddr); } aclrtDestroyStream(stream); aclrtResetDevice(deviceId); aclFinalize(); return 0; }【免费下载链接】ops-nn本项目是CANN提供的神经网络类计算算子库实现网络在NPU上加速计算。项目地址: https://gitcode.com/cann/ops-nn创作声明：本文部分内容由AI辅助生成（AIGC），仅供参考
aclnnGroupNormalizationGrad【免费下载链接】ops-nn本项目是CANN提供的神经网络类计算算子库实现网络在NPU上加速计算。项目地址: https://gitcode.com/cann/ops-nn产品支持情况产品是否支持Atlas A2 训练系列产品/Atlas 800I A2 推理产品√功能说明算子功能完成 Group Normalization 的反向。计算公式$$ \hat{x} (x - mean) \cdot rstd $$$$ s_1 \sum(dy \cdot gamma), \quad s_2 \sum(dy \cdot gamma \cdot \hat{x}) $$$$ dx \frac{rstd}{M} \cdot gamma \cdot (M \cdot dy - s_1 - \hat{x} \cdot s_2) $$函数原型每个算子分为两段式接口必须先调用aclnnGroupNormalizationGradGetWorkspaceSize接口获取入参并根据计算流程计算所需workspace大小再调用aclnnGroupNormalizationGrad接口执行计算。aclnnStatus aclnnGroupNormalizationGradGetWorkspaceSize( const aclTensor *x, const aclTensor *dy, const aclTensor *gamma, const aclTensor *mean, const aclTensor *rstd, aclTensor *dx, uint64_t *workspaceSize, aclOpExecutor **executor)aclnnStatus aclnnGroupNormalizationGrad( void *workspace, uint64_t workspaceSize, aclOpExecutor *executor, const aclrtStream stream)aclnnGroupNormalizationGradGetWorkspaceSize参数说明参数名输入/输出描述使用说明数据类型数据格式维度(shape)非连续Tensorx输入公式中的 x前向输入。dtype 需与 dy 保持一致。shape 需与 dy 相同。FLOAT、BFLOAT16、FLOAT16ND3-8×dy输入公式中的 dy上游梯度。数据类型与 x 的数据类型满足互推导关系。FLOAT、BFLOAT16、FLOAT16ND3-8×gamma输入已广播到 [N, G, M] 的缩放系数。dtype 需与 x 保持一致。shape 需与 x 相同。FLOAT、BFLOAT16、FLOAT16ND3-8×mean输入每个 group 的均值形状为 [N, G]。dtype 需与 x 保持一致。FLOAT、BFLOAT16、FLOAT16ND2×rstd输入每个 group 的标准差倒数形状为 [N, G]。dtype 需与 x 保持一致。FLOAT、BFLOAT16、FLOAT16ND2×dx输出公式中的 dx输入梯度。dtype 需与 x 相同。shape 需与 x 相等。FLOAT、BFLOAT16、FLOAT16ND3-8×workspaceSize输出返回需要在 Device 侧申请的 workspace 大小。-----executor输出返回 op 执行器包含了算子计算流程。-----返回值aclnnStatus返回状态码具体参见aclnn返回码。 第一段接口会完成入参校验出现以下场景时报错返回码错误码描述ACLNN_ERR_PARAM_NULLPTR161001传入的 x、dy、gamma、mean 或 rstd 是空指针。ACLNN_ERR_PARAM_INVALID161002x、dy、gamma、mean 或 rstd 的数据类型不在支持的范围之内。x、dy、gamma、mean 或 rstd 的 shape 超过 8 维或 x、dy、gamma 的 shape 低于 3 维。x、dy、gamma 与 dx 数据类型不一致。x、dy、gamma 的 shape 不一致。aclnnGroupNormalizationGrad参数说明参数名输入/输出描述workspace输入在 Device 侧申请的 workspace 内存地址。workspaceSize输入在 Device 侧申请的 workspace 大小由第一段接口 aclnnGroupNormalizationGradGetWorkspaceSize 获取。executor输入op 执行器包含了算子计算流程。stream输入指定执行任务的 Stream。返回值aclnnStatus返回状态码具体参见aclnn返回码。约束说明无。调用示例示例代码如下仅供参考具体编译和执行过程请参考编译与运行样例。#include iostream #include vector #include acl/acl.h #include aclnn_group_normalization_grad.h #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) { 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); 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); 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); 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]; } *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); std::vectorint64_t xShape {2, 4, 128}; std::vectorint64_t meanShape {2, 4}; void* xDeviceAddr nullptr; void* dyDeviceAddr nullptr; void* gammaDeviceAddr nullptr; void* meanDeviceAddr nullptr; void* rstdDeviceAddr nullptr; void* dxDeviceAddr nullptr; aclTensor* x nullptr; aclTensor* dy nullptr; aclTensor* gamma nullptr; aclTensor* mean nullptr; aclTensor* rstd nullptr; aclTensor* dx nullptr; auto size GetShapeSize(xShape); std::vectorfloat xHostData(size); std::vectorfloat dyHostData(size); std::vectorfloat gammaHostData(size); std::vectorfloat dxHostData(size, 0.0f); auto meanSize GetShapeSize(meanShape); std::vectorfloat meanHostData(meanSize); std::vectorfloat rstdHostData(meanSize); for (int64_t i 0; i size; i) { xHostData[i] static_castfloat(i % 128) / 128.0f; dyHostData[i] 0.5f; gammaHostData[i] 1.0f; } for (int64_t i 0; i meanSize; i) { meanHostData[i] 0.0f; rstdHostData[i] 1.0f; } ret CreateAclTensor(xHostData, xShape, xDeviceAddr, aclDataType::ACL_FLOAT, x); CHECK_RET(ret ACL_SUCCESS, return ret); ret CreateAclTensor(dyHostData, xShape, dyDeviceAddr, aclDataType::ACL_FLOAT, dy); CHECK_RET(ret ACL_SUCCESS, return ret); ret CreateAclTensor(gammaHostData, xShape, gammaDeviceAddr, aclDataType::ACL_FLOAT, gamma); CHECK_RET(ret ACL_SUCCESS, return ret); ret CreateAclTensor(meanHostData, meanShape, meanDeviceAddr, aclDataType::ACL_FLOAT, mean); CHECK_RET(ret ACL_SUCCESS, return ret); ret CreateAclTensor(rstdHostData, meanShape, rstdDeviceAddr, aclDataType::ACL_FLOAT, rstd); CHECK_RET(ret ACL_SUCCESS, return ret); ret CreateAclTensor(dxHostData, xShape, dxDeviceAddr, aclDataType::ACL_FLOAT, dx); CHECK_RET(ret ACL_SUCCESS, return ret); uint64_t workspaceSize 0; aclOpExecutor* executor; ret aclnnGroupNormalizationGradGetWorkspaceSize(x, dy, gamma, mean, rstd, dx, workspaceSize, executor); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(aclnnGroupNormalizationGradGetWorkspaceSize failed. ERROR: %d\n, ret); return ret); void* workspaceAddr nullptr; if (workspaceSize 0) { ret aclrtMalloc(workspaceAddr, workspaceSize, ACL_MEM_MALLOC_HUGE_FIRST); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(allocate workspace failed. ERROR: %d\n, ret); return ret); } ret aclnnGroupNormalizationGrad(workspaceAddr, workspaceSize, executor, stream); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(aclnnGroupNormalizationGrad failed. ERROR: %d\n, ret); return ret); ret aclrtSynchronizeStream(stream); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(aclrtSynchronizeStream failed. ERROR: %d\n, ret); return ret); size GetShapeSize(xShape); std::vectorfloat resultData(size, 0); ret aclrtMemcpy(resultData.data(), resultData.size() * sizeof(resultData[0]), dxDeviceAddr, size * sizeof(resultData[0]), ACL_MEMCPY_DEVICE_TO_HOST); CHECK_RET(ret ACL_SUCCESS, LOG_PRINT(copy result failed. ERROR: %d\n, ret); return ret); for (int64_t i 0; i size; i) { LOG_PRINT(result[%ld] is: %f\n, i, resultData[i]); } aclDestroyTensor(x); aclDestroyTensor(dy); aclDestroyTensor(gamma); aclDestroyTensor(mean); aclDestroyTensor(rstd); aclDestroyTensor(dx); aclrtFree(xDeviceAddr); aclrtFree(dyDeviceAddr); aclrtFree(gammaDeviceAddr); aclrtFree(meanDeviceAddr); aclrtFree(rstdDeviceAddr); aclrtFree(dxDeviceAddr); if (workspaceSize 0) { aclrtFree(workspaceAddr); } aclrtDestroyStream(stream); aclrtResetDevice(deviceId); aclFinalize(); return 0; }【免费下载链接】ops-nn本项目是CANN提供的神经网络类计算算子库实现网络在NPU上加速计算。项目地址: https://gitcode.com/cann/ops-nn创作声明：本文部分内容由AI辅助生成（AIGC），仅供参考
)
)
