Documentation
Download
文档中心
aclnnCtcLoss

aclnnCtcLoss

📄 View Source Code

Applicable Products

ProductSupported
Ascend 950PR/Ascend 950DT√
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: Computes the connectionist temporal classification loss value.

  • Formula:

Define ykty_{k}^{t} as the probability that the true character is kk at time tt. (Generally, ykty_{k}^{t} is an element in the output matrix after softmax.) The set of all sequences that can be formed from the character set L′L^{'} is called L′TL^{'T}, and any sequence in L′TL^{'T} is called a path, denoted as ππ. The distribution of ππ is given by formula (1):

p(π∣x)=∏t=1Tyπtt,∀π∈L′T.(1)p(π|x)=\prod_{t=1}^{T}y^{t}_{π_{t}} , \forall π \in L'^{T}. \tag{1}

Define a many-to-one mapping B: L′T→L≤TL^{'T} \to L^{\leq T}. The conditional probability of l∈L≤Tl \in L^{\leq T} obtained through mapping B is equal to the sum of probabilities of all possible paths corresponding to ll, as shown in formula (2):

p(l∣x)=∑π∈B−1(l)p(π∣x).(2)p(l|x)=\sum_{π \in B^{-1}(l)}p(π|x).\tag{2}

The task of finding the path of ll that maximizes p(l∣x)p(l|x) is called decoding, as shown in formula (3):

h(x)=l∈L≤Targ max p(l∣x).(3)h(x)=^{arg \ max}_{l \in L^{ \leq T}} \ p(l|x).\tag{3}

When zeroInfinity is True

h(x)={0,h(x)==Inf or h(x)==−Infh(x), else h(x)=\begin{cases}0,&h(x) == Inf \text{ or } h(x) == -Inf \\h(x),&\text { else }\end{cases}

Function Prototype

Each operator is divided into a two-phase API. The "aclnnCtcLossGetWorkspaceSize" API must be called first to obtain the workspace size required for computation and the executor that encapsulates the operator computation flow. Then, the "aclnnCtcLoss" API is called to perform the computation.

aclnnStatus aclnnCtcLossGetWorkspaceSize(
    const aclTensor*     logProbs,
    const aclTensor*     targets,
    const aclIntArray*   inputLengths,
    const aclIntArray*   targetLengths,
    int64_t              blank,
    bool                 zeroInfinity,
    aclTensor*           negLogLikelihoodOut,
    aclTensor*           logAlphaOut,
    uint64_t*            workspaceSize,
    aclOpExecutor**      executor)
aclnnStatus aclnnCtcLoss(
    void*          workspace,
    uint64_t       workspaceSize,
    aclOpExecutor* executor,
    aclrtStream    stream)

aclnnCtcLossGetWorkspaceSize

  • Parameters:

    Parameter Input/Output Description Instruction Data Type Data Format Dimension (Shape) Non-contiguous Tensor
    logProbs (aclTensor*) Input Represents the logarithmic probability of the output, which is y in the formula. - FLOAT16, FLOAT, BFLOAT16 ND (T,N,C) or (T,C)
    T is the input length, N is the batch size, and C is the number of classes, which must be greater than 0, including the blank label.    		 √
    targets (aclTensor*) Input Represents the labels containing the target sequence, i.e., π in the formula. When the shape is (N, S), S is a value not less than the maximum value in targetLengths; or when the shape is (SUM(targetLengths)), assuming targets are unpadded and concatenated in dimension 1. When logProbs is 2-dimensional, N = 1. INT64, INT32 ND - √
    inputLengths (aclIntArray*) Input Indicates the actual lengths of the input sequences. T in the formula is an element in inputLengths. The array length is N, and each value in the array must be less than or equal to T. When logProbs is 2D, N = 1. INT64, INT32 - - -
    targetLengths (aclIntArray*) Input Indicates the actual length of the target sequence. The length of l in the formula is the element in targetLengths. The array length is N. When the shape of targets is (N, S), each value in the array must be less than or equal to S. When logProbs is 2-dimensional, N = 1. INT64, INT32 - - -
    blank (int64_t) Input Represents the blank label. The value must be greater than or equal to 0 and less than C. INT64 - - -
    zeroInfinity (bool) Input Indicates whether to zero infinite losses and associated gradients, i.e., zeroInfinity in the formula. - BOOL - - -
    negLogLikelihoodOut (aclTensor*) Output Represents the output loss value, i.e., h in the formula. The data type must be consistent with logProbs. When logProbs is 3D, negLogLikelihoodOut is a tensor with shape (N); otherwise, negLogLikelihoodOut is a 0D tensor. Consistent with logProbs. ND - √
    logAlphaOut (aclTensor*) Output Represents the probability of a possible trace from input to target, i.e., p(l|x) in the formula. The data type must be consistent with logProbs. When logProbs is 2-dimensional, N = 1. Consistent with logProbs. ND - √
    workspaceSize (uint64_t*) Output Returns the workspace size to be applied on the Device side. - - - - -
    executor (aclOpExecutor**) Output Returns the op executor, which contains the operator computation flow. - - - - -

  • logAlphaOut:

    • Ascend 950PR/Ascend 950DT, Atlas A2 training products/Atlas A2 inference products, Atlas A3 training products/Atlas A3 inference products : The shape is (N,T,(2∗max(targetLengths)+8)/8∗8N, T, (2*max(targetLengths)+8)/8*8).

  • 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 Value Error Code Description
    ACLNN_ERR_PARAM_NULLPTR 161001 The passed logProbs, targets, inputLengths, targetLengths, negLogLikelihoodOut, and logAlphaOut are null pointers.
    ACLNN_ERR_PARAM_INVALID 161002 The data type of logProbs, targets, inputLengths, or targetLengths is not within the supported range.
    The tensors logProbs, targets, inputLengths, targetLengths, negLogLikelihoodOut, and logAlphaOut do not meet the corresponding shape requirements, or the lengths of the ArrayLists inputLengths and targetLengths do not meet the requirements.
    blank does not meet the value range requirement.

aclnnCtcLoss

  • Parameters:

    Parameter Input/Output Description
    workspace Input Address of the workspace memory applied on the Device side.
    workspaceSize Input Size of the workspace applied on the Device side, obtained through the first-phase API aclnnCtcLossGetWorkspaceSize.
    executor Input Operator executor, which contains the operator computation process.
    stream Input Specifies the stream for task execution.

  • Return value

    aclnnStatus: return code. For details, see aclnn Return Code.

Constraints

  • Value Range Constraints:

    • The value range of targets must be [0,C−1][0, C - 1] and exclude the value corresponding to blank, where CC represents the last dimension of logProbs, i.e., the number of classes.

    • The value range of inputLengths must be [1,T][1, T], where TT represents the 0th dimension of logProbs, i.e., the input length.

    • The value range of targetLengths must be greater than or equal to 1.

    • The elements in targetLengths must be less than or equal to the corresponding elements in inputLengths.

    If the first three value range constraints are not met, out-of-bounds behavior may occur on CPU/GPU, causing the computation results of negLogLikelihoodOut and logAlphaOut to potentially differ from those on CPU/GPU. If the fourth value range constraint is not met, the computation result of logAlphaOut on the corresponding batch will differ from that on CPU/GPU.

  • Deterministic computation:

    • aclnnCtcLoss defaults to a non-deterministic implementation. You can enable deterministic computation through aclrtCtxSetSysParamOpt.

Example

The following provides example code for reference only. For details about compilation and running, see Compile and Run Samples.

  • Atlas A2 training products/Atlas A2 inference products, Atlas A3 training products/Atlas A3 inference products:

    #include <iostream>
#include <vector>
#include "acl/acl.h"
#include "aclnnop/aclnn_ctc_loss.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);
  // Apply for device-side memory by calling aclrtMalloc.
  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 by calling 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];
  }

  // Create an aclTensor by calling aclCreateTensor.
  *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 device ID based on your actual device.
  // 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 needs.
  CHECK_RET(ret == 0, LOG_PRINT("Init acl failed. ERROR: %d\n", ret); return ret);
  // 2. Construct the input and output. Customize the construction based on the API interface.
  std::vector<int64_t> logProbsShape = {12, 4, 5};
  std::vector<int64_t> targetsShape = {4, 7};
  std::vector<int64_t> negLoglikelihoodOutShape = {4};
  std::vector<int64_t> logAlphaOutShape = {4, 12, 16};
  void*logProbsDeviceAddr = nullptr;
  void* targetsDeviceAddr = nullptr;
  void*negLoglikelihoodOutDeviceAddr = nullptr;
  void* logAlphaOutDeviceAddr = nullptr;
  aclTensor*logProbs = nullptr;
  aclTensor* targets = nullptr;
  aclIntArray*inputLengths = nullptr;
  aclIntArray* targetLengths = nullptr;
  aclTensor*negLoglikelihoodOut = nullptr;
  aclTensor* logAlphaOut = nullptr;
  std::vector<float> logProbsHostData = {
    -1.0894, -2.7162, -0.9764, -1.9126, -2.6162,
    -2.0684, -2.4871, -2.0866, -1.7205, -0.7187,
    -2.4423, -1.2017, -1.4653, -1.1821, -2.5942,
    -2.4670, -2.7257, -1.4135, -2.1042, -0.7248,

    -3.7759, -1.3742, -1.2549, -1.5807, -1.4562,
    -1.3826, -1.8995, -1.8527, -0.9493, -2.8895,
    -1.6316, -2.6603, -2.5014, -0.6992, -1.8609,
    -1.9269, -2.2350, -0.8073, -1.8906, -1.8947,

    -0.3468, -2.5855, -2.0723, -2.7147, -3.6668,
    -0.9541, -1.7258, -2.0693, -1.6378, -2.1531,
    -3.5386, -3.4830, -0.2532, -2.0557, -3.3261,
    -1.1480, -1.8080, -0.8244, -3.2414, -3.1909,

    -0.8866, -0.7540, -4.4312, -3.4634, -2.6000,
    -1.2785, -1.8347, -3.3122, -0.7620, -2.8349,
    -1.4975, -1.3865, -0.9645, -3.8171, -2.0939,
    -2.3536, -2.0773, -1.4981, -0.8372, -2.0938,

    -1.2186, -0.8285, -2.9399, -2.1159, -2.3620,
    -2.3139, -0.6503, -2.7249, -1.2340, -3.7927,
    -0.7143, -2.5084, -3.2826, -2.6651, -1.1334,
    -1.6965, -1.9728, -2.3849, -1.6052, -0.9554,

    -1.6384, -1.2596, -2.1680, -1.8476, -1.3866,
    -3.0455, -0.5737, -2.5339, -2.1118, -1.6681,
    -2.4675, -2.8842, -0.4329, -3.6266, -1.6925,
    -3.1023, -2.7696, -1.2755, -0.6470, -2.4143,

    -2.0107, -2.0912, -1.3053, -0.8557, -3.0683,
    -1.2872, -3.6523, -1.6703, -2.7596, -0.8063,
    -2.4633, -1.2959, -1.6153, -2.3072, -1.0705,
    -3.0543, -0.6473, -1.1650, -2.9025, -2.7710,

    -3.5519, -2.0400, -1.8667, -1.4289, -0.8050,
    -1.4602, -0.7452, -1.5754, -3.1624, -3.1247,
    -1.4677, -1.2725, -2.9575, -1.8883, -1.2513,
    -1.2164, -1.5894, -2.2217, -2.3714, -1.2110,

    -2.0843, -0.6515, -1.4252, -2.9402, -2.7964,
    -1.5261, -2.5471, -1.7167, -1.9846, -0.9488,
    -1.4847, -1.7093, -1.4095, -1.7293, -1.7675,
    -0.9203, -4.2299, -1.8740, -1.4076, -1.6671,

    -1.9052, -0.8330, -2.1839, -2.2459, -1.6193,
    -2.9108, -1.2114, -1.4616, -1.7297, -1.4330,
    -2.2656, -0.7878, -1.8533, -1.8711, -2.0349,
    -2.2457, -2.1395, -1.4509, -0.7538, -2.6381,

    -0.8078, -2.1054, -2.6703, -1.1108, -3.3867,
    -1.7774, -1.8426, -1.9473, -1.3293, -1.3273,
    -1.3490, -1.9842, -2.5357, -2.2161, -0.8800,
    -1.5412, -1.8003, -2.7603, -0.8606, -2.0066,

    -1.8342, -2.2741, -1.8348, -1.5833, -0.9877,
    -3.5196, -2.3361, -0.9124, -0.9307, -2.5531,
    -1.4862, -1.2153, -1.4453, -3.4462, -1.5625,
    -2.6455, -1.4153, -1.3079, -1.1568, -2.2897};
  std::vector<int64_t> targetsHostData = {
    1, 2, 1, 1, 2, 4, 1,
    2, 2, 2, 2, 2, 2, 3,
    4, 2, 1, 4, 3, 1, 4,
    4, 1, 4, 2, 2, 2, 3};

  std::vector<float> negLoglikelihoodOutHostData = {0, 0, 0, 0};
  std::vector<float> logAlphaOutHostData = {
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,

    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,

    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,

    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0};
  // Create the logProbs aclTensor.
  ret = CreateAclTensor(logProbsHostData, logProbsShape, &logProbsDeviceAddr, aclDataType::ACL_FLOAT, &logProbs);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create the targets aclTensor.
  ret = CreateAclTensor(targetsHostData, targetsShape, &targetsDeviceAddr, aclDataType::ACL_INT64, &targets);
  CHECK_RET(ret == ACL_SUCCESS, return ret);

  std::vector<int64_t> inputLengthsSizeData = {10,10,10,10};
  inputLengths = aclCreateIntArray(inputLengthsSizeData.data(), 4);
  CHECK_RET(inputLengths != nullptr, return ACL_ERROR_BAD_ALLOC);
  std::vector<int64_t> targetLengthsSizeData = {2, 3, 1, 5};
  targetLengths = aclCreateIntArray(targetLengthsSizeData.data(), 4);
  CHECK_RET(targetLengths != nullptr, return ACL_ERROR_BAD_ALLOC);

  // Create the negLoglikelihoodOut aclTensor.
  ret = CreateAclTensor(negLoglikelihoodOutHostData, negLoglikelihoodOutShape, &negLoglikelihoodOutDeviceAddr, aclDataType::ACL_FLOAT, &negLoglikelihoodOut);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create the logAlphaOut aclTensor.
  ret = CreateAclTensor(logAlphaOutHostData, logAlphaOutShape, &logAlphaOutDeviceAddr, aclDataType::ACL_FLOAT, &logAlphaOut);
  CHECK_RET(ret == ACL_SUCCESS, return ret);

  // 3. Call the CANN operator library API. Replace it with the specific API.
  uint64_t workspaceSize = 0;
  aclOpExecutor* executor;
  // Call the first-phase API of aclnnCtcLoss.
  ret = aclnnCtcLossGetWorkspaceSize(logProbs, targets, inputLengths, targetLengths, 0, false, negLoglikelihoodOut, logAlphaOut, &workspaceSize, &executor);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnCtcLossGetWorkspaceSize failed. ERROR: %d\n", ret); return ret);
  // Apply for device memory based on the workspaceSize calculated 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 aclnnCtcLoss.
  ret = aclnnCtcLoss(workspaceAddr, workspaceSize, executor, stream);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnCtcLoss 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 negLoglikelihoodOut value, copy the result from the device memory to the host memory, and modify based on the specific API definition.
  auto size = GetShapeSize(negLoglikelihoodOutShape);
  std::vector<float> resultData(size, 0);
  ret = aclrtMemcpy(resultData.data(), resultData.size() * sizeof(resultData[0]), negLoglikelihoodOutDeviceAddr, 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("negLoglikelihoodOut result[%ld] is: %f\n", i, resultData[i]);
  }

  // 6. Obtain the output logAlphaOut value, copy the result from the device memory to the host memory, and modify based on the specific API definition.
  auto size1 = GetShapeSize(logAlphaOutShape);
  std::vector<float> resultData1(size1, 0);
  ret = aclrtMemcpy(resultData1.data(), resultData1.size() * sizeof(resultData1[0]), logAlphaOutDeviceAddr, size1 * 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 < size1; i++) {
    LOG_PRINT("logAlphaOut result[%ld] is: %f\n", i, resultData1[i]);
  }

  // 7. Release aclTensor and IntArray, and modify based on the specific API definition.
  aclDestroyTensor(logProbs);
  aclDestroyTensor(targets);
  aclDestroyIntArray(inputLengths);
  aclDestroyIntArray(targetLengths);
  aclDestroyTensor(negLoglikelihoodOut);
  aclDestroyTensor(logAlphaOut);

  // 8. Release device resources. Modify based on the specific API definition.
  aclrtFree(logProbsDeviceAddr);
  aclrtFree(targetsDeviceAddr);
  aclrtFree(negLoglikelihoodOutDeviceAddr);
  aclrtFree(logAlphaOutDeviceAddr);
  if (workspaceSize > 0) {
    aclrtFree(workspaceAddr);
  }
  aclrtDestroyStream(stream);
  aclrtResetDevice(deviceId);
  aclFinalize();
  return 0;
}

  • Ascend 950PR/Ascend 950DT:

    #include <iostream>
#include <vector>
#include "acl/acl.h"
#include "aclnnop/aclnn_ctc_loss.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);

  // Call aclrtMemcpy to copy host-side data to the device-side 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 the 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 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 needs.
  CHECK_RET(ret == 0, LOG_PRINT("Init acl failed. ERROR: %d\n", ret); return ret);
  // 2. Construct the input and output. Custom construction is required based on the API.
  std::vector<int64_t> logProbsShape = {12, 4, 5};
  std::vector<int64_t> targetsShape = {4, 7};
  std::vector<int64_t> negLoglikelihoodOutShape = {4};
  std::vector<int64_t> logAlphaOutShape = {4, 12, 11};
  void* logProbsDeviceAddr = nullptr;
  void* targetsDeviceAddr = nullptr;
  void* negLoglikelihoodOutDeviceAddr = nullptr;
  void* logAlphaOutDeviceAddr = nullptr;
  aclTensor* logProbs = nullptr;
  aclTensor* targets = nullptr;
  aclIntArray* inputLengths = nullptr;
  aclIntArray* targetLengths = nullptr;
  aclTensor* negLoglikelihoodOut = nullptr;
  aclTensor* logAlphaOut = nullptr;
  std::vector<float> logProbsHostData = {
    -1.0894, -2.7162, -0.9764, -1.9126, -2.6162,
    -2.0684, -2.4871, -2.0866, -1.7205, -0.7187,
    -2.4423, -1.2017, -1.4653, -1.1821, -2.5942,
    -2.4670, -2.7257, -1.4135, -2.1042, -0.7248,

    -3.7759, -1.3742, -1.2549, -1.5807, -1.4562,
    -1.3826, -1.8995, -1.8527, -0.9493, -2.8895,
    -1.6316, -2.6603, -2.5014, -0.6992, -1.8609,
    -1.9269, -2.2350, -0.8073, -1.8906, -1.8947,

    -0.3468, -2.5855, -2.0723, -2.7147, -3.6668,
    -0.9541, -1.7258, -2.0693, -1.6378, -2.1531,
    -3.5386, -3.4830, -0.2532, -2.0557, -3.3261,
    -1.1480, -1.8080, -0.8244, -3.2414, -3.1909,

    -0.8866, -0.7540, -4.4312, -3.4634, -2.6000,
    -1.2785, -1.8347, -3.3122, -0.7620, -2.8349,
    -1.4975, -1.3865, -0.9645, -3.8171, -2.0939,
    -2.3536, -2.0773, -1.4981, -0.8372, -2.0938,

    -1.2186, -0.8285, -2.9399, -2.1159, -2.3620,
    -2.3139, -0.6503, -2.7249, -1.2340, -3.7927,
    -0.7143, -2.5084, -3.2826, -2.6651, -1.1334,
    -1.6965, -1.9728, -2.3849, -1.6052, -0.9554,

    -1.6384, -1.2596, -2.1680, -1.8476, -1.3866,
    -3.0455, -0.5737, -2.5339, -2.1118, -1.6681,
    -2.4675, -2.8842, -0.4329, -3.6266, -1.6925,
    -3.1023, -2.7696, -1.2755, -0.6470, -2.4143,

    -2.0107, -2.0912, -1.3053, -0.8557, -3.0683,
    -1.2872, -3.6523, -1.6703, -2.7596, -0.8063,
    -2.4633, -1.2959, -1.6153, -2.3072, -1.0705,
    -3.0543, -0.6473, -1.1650, -2.9025, -2.7710,

    -3.5519, -2.0400, -1.8667, -1.4289, -0.8050,
    -1.4602, -0.7452, -1.5754, -3.1624, -3.1247,
    -1.4677, -1.2725, -2.9575, -1.8883, -1.2513,
    -1.2164, -1.5894, -2.2217, -2.3714, -1.2110,

    -2.0843, -0.6515, -1.4252, -2.9402, -2.7964,
    -1.5261, -2.5471, -1.7167, -1.9846, -0.9488,
    -1.4847, -1.7093, -1.4095, -1.7293, -1.7675,
    -0.9203, -4.2299, -1.8740, -1.4076, -1.6671,

    -1.9052, -0.8330, -2.1839, -2.2459, -1.6193,
    -2.9108, -1.2114, -1.4616, -1.7297, -1.4330,
    -2.2656, -0.7878, -1.8533, -1.8711, -2.0349,
    -2.2457, -2.1395, -1.4509, -0.7538, -2.6381,

    -0.8078, -2.1054, -2.6703, -1.1108, -3.3867,
    -1.7774, -1.8426, -1.9473, -1.3293, -1.3273,
    -1.3490, -1.9842, -2.5357, -2.2161, -0.8800,
    -1.5412, -1.8003, -2.7603, -0.8606, -2.0066,

    -1.8342, -2.2741, -1.8348, -1.5833, -0.9877,
    -3.5196, -2.3361, -0.9124, -0.9307, -2.5531,
    -1.4862, -1.2153, -1.4453, -3.4462, -1.5625,
    -2.6455, -1.4153, -1.3079, -1.1568, -2.2897};
  std::vector<int64_t> targetsHostData = {
    1, 2, 1, 1, 2, 4, 1,
    2, 2, 2, 2, 2, 2, 3,
    4, 2, 1, 4, 3, 1, 4,
    4, 1, 4, 2, 2, 2, 3};

  std::vector<float> negLoglikelihoodOutHostData = {0, 0, 0, 0};
  std::vector<float> logAlphaOutHostData = {
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,

    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,

    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,

    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
    0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0};
  // Create logProbs aclTensor.
  ret = CreateAclTensor(logProbsHostData, logProbsShape, &logProbsDeviceAddr, aclDataType::ACL_FLOAT, &logProbs);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create targets aclTensor.
  ret = CreateAclTensor(targetsHostData, targetsShape, &targetsDeviceAddr, aclDataType::ACL_INT64, &targets);
  CHECK_RET(ret == ACL_SUCCESS, return ret);

  std::vector<int64_t> inputLengthsSizeData = {10,10,10,10};
  inputLengths = aclCreateIntArray(inputLengthsSizeData.data(), 4);
  CHECK_RET(inputLengths != nullptr, return ACL_ERROR_BAD_ALLOC);
  std::vector<int64_t> targetLengthsSizeData = {2, 3, 1, 5};
  targetLengths = aclCreateIntArray(targetLengthsSizeData.data(), 4);
  CHECK_RET(targetLengths != nullptr, return ACL_ERROR_BAD_ALLOC);

  // Create negLoglikelihoodOut aclTensor.
  ret = CreateAclTensor(negLoglikelihoodOutHostData, negLoglikelihoodOutShape, &negLoglikelihoodOutDeviceAddr, aclDataType::ACL_FLOAT, &negLoglikelihoodOut);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create logAlphaOut aclTensor.
  ret = CreateAclTensor(logAlphaOutHostData, logAlphaOutShape, &logAlphaOutDeviceAddr, aclDataType::ACL_FLOAT, &logAlphaOut);
  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 aclnnCtcLoss.
  ret = aclnnCtcLossGetWorkspaceSize(logProbs, targets, inputLengths, targetLengths, 0, false, negLoglikelihoodOut, logAlphaOut, &workspaceSize, &executor);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnCtcLossGetWorkspaceSize failed. ERROR: %d\n", ret); return ret);
  // Apply for device memory based on the workspaceSize calculated 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 aclnnCtcLoss.
  ret = aclnnCtcLoss(workspaceAddr, workspaceSize, executor, stream);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnCtcLoss failed. ERROR: %d\n", ret); return ret);
  // 4. (Boilerplate) Synchronize and 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 negLoglikelihoodOut value and copy the result from device memory to host. Modify based on the specific API definition.
  auto size = GetShapeSize(negLoglikelihoodOutShape);
  std::vector<float> resultData(size, 0);
  ret = aclrtMemcpy(resultData.data(), resultData.size() * sizeof(resultData[0]), negLoglikelihoodOutDeviceAddr, 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("negLoglikelihoodOut result[%ld] is: %f\n", i, resultData[i]);
  }

  // 6. Obtain the output logAlphaOut value and copy the result from device memory to host. Modify based on the specific API definition.
  auto size1 = GetShapeSize(logAlphaOutShape);
  std::vector<float> resultData1(size1, 0);
  ret = aclrtMemcpy(resultData1.data(), resultData1.size() * sizeof(resultData1[0]), logAlphaOutDeviceAddr, size1 * 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 < size1; i++) {
    LOG_PRINT("logAlphaOut result[%ld] is: %f\n", i, resultData1[i]);
  }

  // 7. Release aclTensor and IntArray. Modify based on the specific API definition.
  aclDestroyTensor(logProbs);
  aclDestroyTensor(targets);
  aclDestroyIntArray(inputLengths);
  aclDestroyIntArray(targetLengths);
  aclDestroyTensor(negLoglikelihoodOut);
  aclDestroyTensor(logAlphaOut);

  // 8. Release device resources. Modify based on the specific API definition.
  aclrtFree(logProbsDeviceAddr);
  aclrtFree(targetsDeviceAddr);
  aclrtFree(negLoglikelihoodOutDeviceAddr);
  aclrtFree(logAlphaOutDeviceAddr);
  if (workspaceSize > 0) {
    aclrtFree(workspaceAddr);
  }
  aclrtDestroyStream(stream);
  aclrtResetDevice(deviceId);
  aclFinalize();
  return 0;
}
This site uses cookies to improve your browsing experience. By continuing to browse this site, you agree to our use of cookies.View Details