aclnnGroupNormSwishGrad
Applicable Products
| Product | Supported |
|---|---|
| × | |
| √ | |
| √ | |
| × | |
| × | |
| × |
Function
Description: The backward operation of aclnnGroupNormSwish.
Function Prototype
Each operator is divided into a two-phase API. The "aclnnGroupNormSwishGradGetWorkspaceSize" API must be called first to obtain the workspace size required for computation and the executor that encapsulates the operator computation flow. Then, the "aclnnGroupNormSwishGrad" API is called to perform the computation.
aclnnStatus aclnnGroupNormSwishGradGetWorkspaceSize(
const aclTensor *dy,
const aclTensor *mean,
const aclTensor *rstd,
const aclTensor *x,
const aclTensor *gamma,
const aclTensor *beta,
int64_t numGroups,
char *dataFormatOptional,
double swishScale,
bool dgammaIsRequire,
bool dbetaIsRequire,
const aclTensor *dxOut,
const aclTensor *dgammaOut,
const aclTensor *dbetaOut,
uint64_t *workspaceSize,
aclOpExecutor **executor)aclnnStatus aclnnGroupNormSwishGrad(
void * workspace,
uint64_t workspaceSize,
aclOpExecutor *executor,
aclrtStream stream)aclnnGroupNormSwishGradGetWorkspaceSize
Parameters
Parameter Input/Output Description Instruction Data Type Data Format Dimension (Shape) Non-Contiguous Tensor dy (aclTensor*) Input Gradient for backward computation. - Empty tensor is not supported.
- Supports dimensions from 2D to 8D, where the first dimension is N and the second dimension is C.
FLOAT16, FLOAT, BFLOAT16 ND 2-8 √ mean (aclTensor*) Input. The second output of the forward computation, representing the mean of each group after the input is grouped. - Empty tensors are not supported.
- The data type is the same as gamma, and N is consistent with the 0th dimension of dy.
FLOAT16, FLOAT, BFLOAT16 ND 2 √ rstd (aclTensor*) Input. The third output of the forward computation, representing the reciprocal of the standard deviation for each group after grouping the input. - Not Supported Empty tensor.
- The data type is the same as gamma, where N is consistent with the 0th dimension of dy.
FLOAT16, FLOAT, BFLOAT16 ND 2 √ x (aclTensor*) Input. Forward input x. - Empty tensors are not supported.
- The data type and shape are the same as dy.
FLOAT16, FLOAT, BFLOAT16 ND 2-8 √ gamma (aclTensor*) Input Scale factor for each channel. - Null tensor is not supported.
- The data type and dimensions are the same as dy, and the number of elements must equal C.
FP16, FP32, BF16 ND 1 √ beta (aclTensor*) Input Bias coefficient for each channel. - Empty tensors are not supported.
- The data type and dimension are the same as dy, and the number of elements must equal C.
FLOAT16, FLOAT, BFLOAT16 ND 1 √ numGroups (int64_t) Input Divides the C dimension of the input gradOut into groups. The group must be greater than 0, and C must be divisible by group with a ratio not exceeding 4000. INT64 - - - dataFormatOptional (char*) Input Data format. Recommended value: NCHW. CHAR - - - swishScale (double) Input Computation coefficient. Recommended value: 1.0. DOUBLE - - - dgammaIsRequire (bool) Input Specifies whether to output dgamma. Recommended value: TRUE. BOOL - - - dbetaIsRequire (bool) Input Whether to output dbeta. Recommended value: TRUE. BOOL - - - dxOut (aclTensor*) Output In Formula, out. Data type and shape are the same as x. FLOAT16, FLOAT, BFLOAT16 ND 2-8 x dgammaOut (aclTensor*) Output meanOut in the formula. The data type and shape are the same as gamma. FLOAT16, FLOAT, BFLOAT16 ND 2 x dbetaOut (aclTensor*) Output In Formula, rstdOut. The data type and shape are the same as gamma. FLOAT16, FLOAT, BFLOAT16 ND 2 x workspaceSize (uint64_t*) Output Returns the workspace size to be applied on the Device side. - - - - - executor (aclOpExecutor**) Output Returns the operator executor, which contains the operator computation flow. - - - - - Return Value
aclnnStatus: return code. For details, see aclnn Return Code.
The first-phase API performs input parameter validation and returns an error in the following scenarios:
Return Code Error Code Description ACLNN_ERR_PARAM_NULLPTR 161001 The passed dy, mean, rstd, x, gamma, beta, dxOut, dgammaOut, or dbetaOut is a null pointer. ACLNN_ERR_PARAM_INVALID 161002 The data type of dy is not within the supported range. The data types of mean, rstd, x, gamma, and beta differ from that of dy. The data type of dxOut differs from that of dy.
aclnnGroupNormSwishGrad
Parameters
Parameter Input/Output Description workspace Input The workspace memory address applied on the Device side. workspaceSize Input The workspace size applied on the Device side, obtained from the first-phase API aclnnGroupNormSwishGradGetWorkspaceSize. executor Input Operator executor that contains the operator computation flow. stream Input Specifies the stream for task execution. Return Value
aclnnStatus: return code. For details, see aclnn Return Code.
Constraints
Deterministic computation
- aclnnGroupNormSwishGrad defaults to a non-deterministic implementation. Deterministic computation can be enabled through aclrtCtxSetSysParamOpt.
Input shape constraints:
numGroups is greater than 0.
C is divisible by group.
The number of elements of dy equals .
The number of elements of mean equals .
The number of elements of rstd equals .
The number of elements of x equals .
The number of elements in gamma equals C.
The number of elements in beta equals C.
The ratio of C to group does not exceed 4000.
Example
The following provides example code for reference only. For details about compilation and running, see Compile and Run Samples.
#include <iostream>
#include <vector>
#include "acl/acl.h"
#include "aclnnop/aclnn_group_norm_swish_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::vector<int64_t>& shape) {
int64_t shape_size = 1;
for (auto i : shape) {
shape_size *= i;
}
return shape_size;
}
int Init(int32_t deviceId, aclrtStream* stream) {
// Boilerplate for resource initialization.
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);
// Call aclrtMalloc to apply for device-side memory.
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);
// Copy host-side data to device-side memory using aclrtMemcpy.
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);
// Compute the strides of a contiguous tensor.
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];
}
// Call the aclCreateTensor API to create an aclTensor.
*tensor = aclCreateTensor(shape.data(), shape.size(), dataType, strides.data(), 0, aclFormat::ACL_FORMAT_ND,
shape.data(), shape.size(), *deviceAddr);
return 0;
}
int main() {
// 1. (Boilerplate) Initialize the device/stream. See the ACL API manual.
// Fill in the deviceId based on your actual device.
int32_t deviceId = 0;
aclrtStream stream;
auto ret = Init(deviceId, &stream);
// Handle the check based on your own needs.
CHECK_RET(ret == 0, LOG_PRINT("Init acl failed. ERROR: %d\n", ret); return ret);
// 2. Construct the inputs and outputs. Customize the construction based on the API interface.
std::vector<int64_t> dyShape = {2, 3, 4};
std::vector<int64_t> meanShape = {2, 1};
std::vector<int64_t> rstdShape = {2, 1};
std::vector<int64_t> xShape = {2, 3, 4};
std::vector<int64_t> gammaShape = {3};
std::vector<int64_t> betaShape = {3};
std::vector<int64_t> dxOutShape = {2, 3, 4};
std::vector<int64_t> dgammaOutShape = {3};
std::vector<int64_t> dbetaOutShape = {3};
void* dyDeviceAddr = nullptr;
void* meanDeviceAddr = nullptr;
void* rstdDeviceAddr = nullptr;
void* xDeviceAddr = nullptr;
void* gammaDeviceAddr = nullptr;
void* betaDeviceAddr = nullptr;
void* dxOutDeviceAddr = nullptr;
void* dgammaOutDeviceAddr = nullptr;
void* dbetaOutDeviceAddr = nullptr;
aclTensor* dy = nullptr;
aclTensor* mean = nullptr;
aclTensor* rstd = nullptr;
aclTensor* x = nullptr;
aclTensor* gamma = nullptr;
aclTensor* beta = nullptr;
aclTensor* dxOut = nullptr;
aclTensor* dgammaOut = nullptr;
aclTensor* dbetaOut = nullptr;
std::vector<float> dyHostData = {1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0, 12.0,
13.0, 14.0, 15.0, 16.0, 17.0, 18.0, 19.0, 20.0, 21.0, 22.0, 23.0, 24.0};
std::vector<float> meanHostData = {2.0, 2};
std::vector<float> rstdHostData = {2.0, 2};
std::vector<float> xHostData = {1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0, 12.0,
13.0, 14.0, 15.0, 16.0, 17.0, 18.0, 19.0, 20.0, 21.0, 22.0, 23.0, 24.0};
std::vector<float> gammaHostData = {2.0, 2, 2};
std::vector<float> betaHostData = {2.0, 2, 2};
std::vector<float> dxOutHostData = {1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0, 12.0,
13.0, 14.0, 15.0, 16.0, 17.0, 18.0, 19.0, 20.0, 21.0, 22.0, 23.0, 24.0};
std::vector<float> dgammaOutHostData = {2.0, 2, 2};
std::vector<float> dbetaOutHostData = {2.0, 2, 2};
int64_t numGroups = 1;
char* dataFormatOptional = nullptr;
float swishScale = 1.0f;
bool dgammaIsRequire = true;
bool dbetaIsRequire = true;
// Create the dy aclTensor.
ret = CreateAclTensor(dyHostData, dyShape, &dyDeviceAddr, aclDataType::ACL_FLOAT, &dy);
CHECK_RET(ret == ACL_SUCCESS, return ret);
// Create the mean aclTensor.
ret = CreateAclTensor(meanHostData, meanShape, &meanDeviceAddr, aclDataType::ACL_FLOAT, &mean);
CHECK_RET(ret == ACL_SUCCESS, return ret);
// Create the rstd aclTensor.
ret = CreateAclTensor(rstdHostData, rstdShape, &rstdDeviceAddr, aclDataType::ACL_FLOAT, &rstd);
CHECK_RET(ret == ACL_SUCCESS, return ret);
// Create x aclTensor.
ret = CreateAclTensor(xHostData, xShape, &xDeviceAddr, aclDataType::ACL_FLOAT, &x);
CHECK_RET(ret == ACL_SUCCESS, return ret);
// Create gamma aclTensor.
ret = CreateAclTensor(gammaHostData, gammaShape, &gammaDeviceAddr, aclDataType::ACL_FLOAT, &gamma);
CHECK_RET(ret == ACL_SUCCESS, return ret);
// Create beta aclTensor.
ret = CreateAclTensor(betaHostData, betaShape, &betaDeviceAddr, aclDataType::ACL_FLOAT, &beta);
CHECK_RET(ret == ACL_SUCCESS, return ret);
// Create dxOut aclTensor.
ret = CreateAclTensor(dxOutHostData, dxOutShape, &dxOutDeviceAddr, aclDataType::ACL_FLOAT, &dxOut);
CHECK_RET(ret == ACL_SUCCESS, return ret);
// Create dgammaOut aclTensor.
ret = CreateAclTensor(dgammaOutHostData, dgammaOutShape, &dgammaOutDeviceAddr, aclDataType::ACL_FLOAT, &dgammaOut);
CHECK_RET(ret == ACL_SUCCESS, return ret);
// Create dbetaOut aclTensor.
ret = CreateAclTensor(dbetaOutHostData, dbetaOutShape, &dbetaOutDeviceAddr, aclDataType::ACL_FLOAT, &dbetaOut);
CHECK_RET(ret == ACL_SUCCESS, return ret);
// 3. Call the CANN operator library API. Replace with the specific API.
uint64_t workspaceSize = 0;
aclOpExecutor* executor;
// Call the first-phase API of aclnnGroupNormSwishGrad.
ret = aclnnGroupNormSwishGradGetWorkspaceSize(dy, mean, rstd, x, gamma, beta, numGroups, dataFormatOptional, swishScale, dgammaIsRequire, dbetaIsRequire, dxOut, dgammaOut, dbetaOut, &workspaceSize, &executor);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnGroupNormSwishGradGetWorkspaceSize failed. ERROR: %d\n", ret); return ret);
// Apply for device memory based on the workspaceSize computed by the first-phase API.
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;);
}
// Call the second-phase API of aclnnGroupNormSwishGrad.
ret = aclnnGroupNormSwishGrad(workspaceAddr, workspaceSize, executor, stream);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnGroupNormSwishGrad failed. ERROR: %d\n", ret); return ret);
// 4. (Boilerplate) Synchronously wait for task execution to complete.
ret = aclrtSynchronizeStream(stream);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtSynchronizeStream failed. ERROR: %d\n", ret); return ret);
// 5. Obtain the output value and copy the result from the device memory to the host memory. Modify based on the specific API definition.
auto size = GetShapeSize(dxOutShape);
ret = aclrtMemcpy(dxOutHostData.data(), dxOutHostData.size() * sizeof(dxOutHostData[0]), dxOutDeviceAddr, size * 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);
for (int64_t i = 0; i < size; i++) {
LOG_PRINT("dxOutHostData[%ld] is: %f\n", i, dxOutHostData[i]);
}
size = GetShapeSize(dgammaOutShape);
ret = aclrtMemcpy(dgammaOutHostData.data(), dgammaOutHostData.size() * sizeof(dgammaOutHostData[0]), dgammaOutDeviceAddr, size * 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);
for (int64_t i = 0; i < size; i++) {
LOG_PRINT("dgammaOutHostData[%ld] is: %f\n", i, dgammaOutHostData[i]);
}
size = GetShapeSize(dbetaOutShape);
ret = aclrtMemcpy(dbetaOutHostData.data(), dbetaOutHostData.size() * sizeof(dbetaOutHostData[0]), dbetaOutDeviceAddr, size * 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);
for (int64_t i = 0; i < size; i++) {
LOG_PRINT("dbetaOutHostData[%ld] is: %f\n", i, dbetaOutHostData[i]);
}
// 6. Release aclTensor and aclScalar. Modify based on the specific API definition.
aclDestroyTensor(dy);
aclDestroyTensor(mean);
aclDestroyTensor(rstd);
aclDestroyTensor(x);
aclDestroyTensor(gamma);
aclDestroyTensor(beta);
aclDestroyTensor(dxOut);
aclDestroyTensor(dgammaOut);
aclDestroyTensor(dbetaOut);
// 7. Release device resources. Modify based on the specific API definition.
aclrtFree(dyDeviceAddr);
aclrtFree(meanDeviceAddr);
aclrtFree(rstdDeviceAddr);
aclrtFree(xDeviceAddr);
aclrtFree(gammaDeviceAddr);
aclrtFree(betaDeviceAddr);
aclrtFree(dxOutDeviceAddr);
aclrtFree(dgammaOutDeviceAddr);
aclrtFree(dbetaOutDeviceAddr);
if (workspaceSize > 0) {
aclrtFree(workspaceAddr);
}
aclrtDestroyStream(stream);
aclrtResetDevice(deviceId);
aclFinalize();
return 0;
}