aclnnCumprod&aclnnInplaceCumprod

Supported Products

Product	Supported
Atlas A3 training products/Atlas A3 inference products	√
Atlas A2 training products/Atlas A2 inference products	√
Atlas 200I/500 A2 inference products	×
Atlas inference products	×
Atlas training products	×

Function

Description: The aclnnCumprod API is added. The cumprod function calculates the cumulative product of the input tensor along a specified dimension. For example, if a tensor represents a series of values, cumprod can calculate the product sequence of these values from the start position to the current position.
Formulas:
- One-dimensional tensor (vector)
  
  For a one-dimensional tensor, the cumulative product $y=[y_1,y_2,y_3...,y_n]$ is calculated as follows:
  
  $y_1=x_1$ $y_2=x_1 \times x_2$ $y_3=x_1 \times x_2\times x_3$ ... $y_n=x_1\times x_2\times x_3\times x_n$
  
  $y_i=\prod_{j=1}^ix_j$ , where $i=1,2...,n$ .
- High-dimensional tensor (using a two-dimensional tensor as an example, dim=0 along the row direction)
  
  For a two-dimensional tensor:
  $X=\begin{bmatrix}x_{11}&x_{12}&...&x_{1m}\\x_{21}&x_{22}&...&x_{2m}\\...&...&...&...&\\x_{n1}&x_{n2}&...&x_{nm}&\end{bmatrix}$
  The result tensor after calculation is as follows:
  $Y=\begin{bmatrix}y_{11}&y_{12}&...&y_{1m}\\y_{21}&y_{22}&...&y_{2m}\\...&...&...&...&\\y_{n1}&y_{n2}&...&y_{nm}&\end{bmatrix}$
  For the first column (j = 1):
  $y_{i1}=x_{11}\times x_{21}\times ...\times x_{i1} (for i = 1, 2,..., n)$
  Therefore, for any column j, the following rule also applies:
  $y_{ij}=\prod_{k=1}^{i} x_{kj}$
- High-dimensional tensor (using a two-dimensional tensor as an example, dim=1 along the column direction)
  
  Therefore, for any column j, the following rule also applies:
  $y_{ij}=\prod_{k=1}^{j} x_{ik}$
- Other parameters can be deduced based on the preceding rules.

Prototype

Each operator has two-phase API calls. First, aclnnCumprodGetWorkspaceSize or aclnnInplaceCumprodGetWorkspaceSize is called to obtain the workspace size required for computation and the executor that contains the operator computation process. Then, aclnnCumprod or aclnnInplaceCumprod is called to perform computation.

aclnnStatus aclnnCumprodGetWorkspaceSize(const aclTensor* input, const aclScalar* dim, const aclDataType dtype, aclTensor* out, uint64_t* workspaceSize, aclOpExecutor** executor)
aclnnStatus aclnnCumprod(void* workspace, uint64_t workspaceSize, aclOpExecutor* executor, aclrtStream stream)
aclnnStatus aclnnInplaceCumprodGetWorkspaceSize(aclTensor* input, const aclScalar* dim, uint64_t* workspaceSize, aclOpExecutor** executor)
aclnnStatus aclnnInplaceCumprod(void* workspace, uint64_t workspaceSize, aclOpExecutor* executor, aclrtStream stream)

aclnnCumprodGetWorkspaceSize

Parameters
- input (aclTensor*, computation input): current input value (indicating the data for which the cumulative product needs to be calculated), aclTensor on the device. Non-contiguous tensors are supported. Empty tensors are supported. The data type can be FLOAT, FLOAT16, BFLOAT16, DOUBLE, INT8, INT16, INT32, INT64, UINT8, UINT16, UINT32, or UINT64. The data format can be ND.
- dim (aclScalar*, computation input): current input value, specifying the dimension for calculating the cumulative product. For a two-dimensional tensor, dim=0 indicates that the calculation is performed along the row direction, and dim=1 indicates that the calculation is performed along the column direction. It is an aclScalar on the device. The value range is [-rank(input), rank(input)). The data type can be INT32.
- dtype (aclDataType, computation input): data type of the input during computation. If the value is ACL_DT_UNDEFINED, the original type of the input is used for computation. If a specific type is specified (within the range of data types supported by the input), the input is converted to this type before computation.
- out (aclTensor*, computation output): cumulative product result, aclTensor on the device. The data format can be ND. When dtype is ACL_DT_UNDEFINED, the data type must be the same as that of the input. When dtype is specified, the data type must be the same as that of dtype. The shape of out must be identical to that of input.
- workspaceSize (uint64_t*, output): size of the workspace to be allocated on the device.
- executor (aclOpExecutor**, output): operator executor, containing the operator computation process.
Returns

aclnnStatus: status code. For details, see aclnn Return Codes.

   The first-phase API implements input parameter validation. The following error codes may be returned:
 161001 (ACLNN_ERR_PARAM_NULLPTR): 1. The passed input or dim is a null pointer.
 161002 (ACLNN_ERR_PARAM_INVALID): 1. The data type or format of input or dim is not supported.
                                   2. dim is incompatible with the shape of input.
                                   3. The shapes of out and input are inconsistent.

aclnnCumprod

Parameters
- workspace (void*, input): address of the workspace to be allocated on the device.
- workspaceSize (uint64_t, input): size of the workspace to be allocated on the device, which is obtained by calling the first-phase API aclnnCumprodGetWorkspaceSize.
- executor (aclOpExecutor *, input): operator executor, containing the operator computation process.
- stream (aclrtStream, input): stream for executing the task.
Returns

aclnnStatus: status code. For details, see aclnn Return Codes.

aclnnInplaceCumprodGetWorkspaceSize

Parameters
- input (aclTensor*, computation input | computation output): input and output tensor for the cumulative product, aclTensor on the device. Non-contiguous tensors are supported. Empty tensors are not supported. The data type can be FLOAT, FLOAT16, BFLOAT16, DOUBLE, INT8, INT16, INT32, INT64, UINT8, UINT16, UINT32, or UINT64. The data format can be ND.
- dim (aclScalar*, computation input): dimension for calculating the cumulative product. For a two-dimensional tensor, dim=0 indicates that the calculation is performed along the row direction, and dim=1 indicates that the calculation is performed along the column direction. It is an aclScalar on the device. The value range is [-rank(x), rank(x)]. The data type can be INT32.
- workspaceSize (uint64_t*, output): size of the workspace to be allocated on the device.
- executor (aclOpExecutor*, output): operator executor, containing the operator computation process.
Returns

aclnnStatus: status code. For details, see aclnn Return Codes.

   The first-phase API implements input parameter validation. The following error codes may be returned:
 161001 (ACLNN_ERR_PARAM_NULLPTR): 1. The passed input or dim is a null pointer.
 161002 (ACLNN_ERR_PARAM_INVALID): 1. The data type or format of input or dim is not supported.
                                   2. dim is incompatible with the shape of input.

aclnnInplaceCumprod

Parameters
- workspace (void*, input): address of the workspace to be allocated on the device.
- workspaceSize (uint64_t, input): size of the workspace to be allocated on the device, which is obtained by calling the first-phase API aclnnInplaceCumprodGetWorkspaceSize.
- executor (aclOpExecutor *, input): operator executor, containing the operator computation process.
- stream (aclrtStream, input): stream for executing the task.
Returns

aclnnStatus: status code. For details, see aclnn Return Codes.

Constraints

Deterministic computation:
- aclnnCumprod and aclnnInplaceCumprod default to a deterministic implementation.

Example

The following example is for reference only. For details, see Compilation and Running Sample.

#include <iostream>
#include <vector>
#include "acl/acl.h"
#include "aclnnop/aclnn_cumprod.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 shapeSize = 1;
    for (auto i : shape)
    {
        shapeSize *= i;
    }
    return shapeSize;
}

void PrintOutResult(std::vector<int64_t> &shape, void **deviceAddr)
{
    auto size = GetShapeSize(shape);
    std::vector<int64_t> resultData(size, 0);
    auto ret = aclrtMemcpy(resultData.data(), resultData.size() * sizeof(resultData[0]),
                           *deviceAddr, size * sizeof(resultData[0]), ACL_MEMCPY_DEVICE_TO_HOST);
    CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("copy result from device to host failed. ERROR: %d\n", ret); return);
    for (int64_t i = 0; i < size; i++)
    {
        LOG_PRINT("result[%ld] is: %d\n", i, resultData[i]);
    }
}

template<typename T>
void PrintOutFloatResult(std::vector<T> &shape, void **deviceAddr, const char *name)
{
    std::vector<float> resultData(shape.size(), 0);
    auto ret = aclrtMemcpy(resultData.data(), resultData.size() * sizeof(resultData[0]),
                           *deviceAddr, shape.size() * sizeof(resultData[0]), ACL_MEMCPY_DEVICE_TO_HOST);
    CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("copy result from device to host failed. ERROR: %d\n", ret); return);
    for (int64_t i = 0; i < shape.size(); i++)
    {
        LOG_PRINT("result var %s[%ld] is: %f\n", name, i, resultData[i]);
    }
}

int Init(int32_t deviceId, aclrtStream *stream)
{
    // (Boilerplate) Initialize AscendCL.
    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 allocate device 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);

    // Call aclrtMemcpy to copy host data to the device memory.
    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 strides of the 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 aclCreateTensor 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;
}

template <typename T>
int CreateAclScalar(aclDataType dataType, T &hostData, aclScalar **scalar)
{
    *scalar = aclCreateScalar(&hostData, dataType);
    if (*scalar == nullptr)
    {
        return -1;
    }
    return 0;
}

int main()
{
    //1. (Boilerplate) Initialize the device and stream. For details, see the list of external AscendCL APIs. Set the device ID in use.
    int32_t deviceId = 0;
    aclrtStream stream;
    auto ret = Init(deviceId, &stream);
    CHECK_RET(ret == 0, LOG_PRINT("Init acl failed. ERROR: %d\n", ret); return ret);

    // 2. Construct the inputs and outputs based on the API definition.
    void *xDeviceAddr = nullptr;
    aclTensor *input = nullptr;
    std::vector<int64_t> xShape = {3};
    std::vector<int64_t> xHostData = {1,2,3};
    // Create the original input x.
    ret = CreateAclTensor(xHostData, xShape, &xDeviceAddr, aclDataType::ACL_INT64, &input);
    CHECK_RET(ret == ACL_SUCCESS, return ret);

    // Create an axis aclScalar.
    int32_t axis_value = 0;
    aclScalar *axis = nullptr;
    ret = CreateAclScalar(aclDataType::ACL_INT32, axis_value, &axis);
    CHECK_RET(ret == ACL_SUCCESS, return ret);

    // Create a result aclTensor.
    std::vector<int64_t> resultHostData(3, 0);
    std::vector<int64_t> resultShape = {3};
    void *resultDeviceAddr = nullptr;
    aclTensor *result = nullptr;
    ret = CreateAclTensor(resultHostData, resultShape, &resultDeviceAddr, aclDataType::ACL_INT64, &result);
    CHECK_RET(ret == ACL_SUCCESS, return ret);

    uint64_t workspaceSize = 0;
    aclOpExecutor *executor;
    aclDataType dtype = ACL_INT64;
    void *workspaceAddr = nullptr;
    // 3. Call the CANN operator library API.
    // Call the first-phase API of aclnnCumprod.
    ret = aclnnCumprodGetWorkspaceSize(input, axis, dtype, result, &workspaceSize, &executor);
    CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnCumprodGetWorkspaceSize failed. ERROR: %d\n", ret); return ret);

    // Allocate device memory based on workspaceSize computed by the first-phase API.
    if (workspaceSize > 0)
    {
        ret = aclrtMalloc(&workspaceAddr, workspaceSize, ACL_MEM_MALLOC_HUGE_FIRST);
        CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnCumprod allocate workspace failed. ERROR: %d\n", ret); return ret);
    }
    // Call the second-phase API of aclnnCumprod.
    ret = aclnnCumprod(workspaceAddr, workspaceSize, executor, stream);
    // 4. (Boilerplate) Synchronize the stream and wait for the task to complete.
    ret = aclrtSynchronizeStream(stream);
    // 5. Obtain the output value and copy the result from the device memory to the host. Modify the code based on the API definition.
    PrintOutResult(resultShape, &resultDeviceAddr);

    // 3. Call the CANN operator library API.
    // Call the first-phase API of aclnnInplaceCumprod.
    ret = aclnnInplaceCumprodGetWorkspaceSize(input, axis, &workspaceSize, &executor);
    CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnCumprodGetWorkspaceSize failed. ERROR: %d\n", ret); return ret);

    // Allocate device memory based on workspaceSize computed by the first-phase API.
    if (workspaceSize > 0)
    {
        ret = aclrtMalloc(&workspaceAddr, workspaceSize, ACL_MEM_MALLOC_HUGE_FIRST);
        CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnCumprod allocate workspace failed. ERROR: %d\n", ret); return ret);
    }
    // Call the second-phase API of aclnnInplaceCumprod.
    ret = aclnnInplaceCumprod(workspaceAddr, workspaceSize, executor, stream);
    // 4. (Boilerplate) Synchronize the stream and wait for the task to complete.
    ret = aclrtSynchronizeStream(stream);
    // 5. Obtain the output value and copy the result from the device memory to the host. Modify the code based on the API definition.
    PrintOutResult(resultShape, &xDeviceAddr);

    // 6. Release aclTensors and aclScalars. Modify the code based on the API definition.
    aclDestroyTensor(input);
    aclDestroyScalar(axis);

    // 7. Release device resources. Modify the code based on the API definition.
    aclrtFree(xDeviceAddr);
    aclrtFree(resultDeviceAddr);
    if (workspaceSize > 0)
    {
        aclrtFree(workspaceAddr);
    }
    aclrtDestroyStream(stream);
    aclrtResetDevice(deviceId);
    aclFinalize();
    return 0;
}