aclnnTransQuantParamV3

📄 View Source Code

Applicable Products

Function

• Description: Converts the data type of the quantization parameter scale from FLOAT32 to UINT64 or INT64 as required by the hardware. Compared with aclnnTransQuantParamV2, this version adds the roundMode input to select the conversion mode for data values during the data type conversion process.

• Formula:

  1. out is in 64-bit format and is initialized to 0.

  2. If round_mode is 1, scale is rounded to the upper 19 bits. If round_mode is 0, no processing is performed.

     $$scale = Round(scale)$$

  3. scale is truncated to the upper 19 bits, stored at bit position 32 of out, and bit 46 is set to 1.

     $$out = out\ |\ (scale\ \&\ 0xFFFFE000)\ |\ (1\ll46)$$

  4. Perform subsequent calculations based on the offset value:

     • If offset does not exist, no further computation is performed.

     • If offset exists:

       1. Process the offset value as int, with a range of [-256, 255].

          $$offset = Max(Min(INT(Round(offset)),255),-256)$$

       2. Retain the lower 9 bits of offset and store them in bits 37 to 45 of out.

          $$out = (out\ \&\ 0x4000FFFFFFFF)\ |\ ((offset\ \&\ 0x1FF)\ll37)$$

Function Prototype

Each operator is divided into a two-phase API. The "aclnnTransQuantParamV3GetWorkspaceSize" API must be called first to obtain the workspace size required for computation and the executor that encapsulates the operator computation flow. Then, the "aclnnTransQuantParamV3" API is called to perform the computation.

aclnnStatus aclnnTransQuantParamV3GetWorkspaceSize(
  const aclTensor* scale,
  const aclTensor* offset,
  int64_t          roundMode,
  const aclTensor* out,
  uint64_t*        workspaceSize,
  aclOpExecutor**  executor)

aclnnStatus aclnnTransQuantParamV3(
  void                *workspace,
  uint64_t             workspaceSize,
  aclOpExecutor       *executor,
  const aclrtStream    stream)

aclnnTransQuantParamV3GetWorkspaceSize

• Parameters:

  Parameter	Input/Output	Description	Instruction	Data Type	Data Format	Dimension (Shape)	Non-contiguous Tensor
  scale (aclTensor*)	Input	The scale value in quantization. Corresponds to `scale` in the formula.	Empty tensors are not supported. The shape is 1D (t,), where t = 1 or n, or 2D (1, n), where n is consistent with the shape n of the right matrix in the matmul computation. The user must ensure that the scale data contains no NaN or Inf.	FLOAT32	ND	1-2	×
  offset (aclTensor*)	Input	Optional parameter, the offset value in quantization. Corresponds to `offset` in the formula.	Empty tensors are not supported. The shape is 1D (t,) or 2D (1, n), where t = 1 or n, and n matches the shape n of the right matrix in the matmul computation. The user must ensure that the offset data contains no NaN or Inf values.	FLOAT32	ND	1-2	×
  roundMode (int64_t)	Input	Rounding mode for filling FP32 to FP19 in quantization computation. Corresponds to `roundMode` in the formula description.	Only the following values are supported: 0 (compatible with V2), 1 (improved computation precision).	-	-	-	-
  out (aclTensor*)	Output	Quantization computation output. Corresponds to `out` in the corresponding formula.	Empty tensors are not supported. When the shape of the input scale is 1D, the shape of out is also 1D, and the size of that dimension is the maximum of the single-dimension shape sizes of scale and offset (if not nullptr). When the shape of the input scale is 2D, the shape of out is exactly the same as the shape of the input scale in both dimensions and size.	UINT64, INT64	ND	1-2	√
  workspaceSize (uint64_t*)	Output	Returns the workspace size to be allocated 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 input scale or out is a null pointer.
  ACLNN_ERR_PARAM_INVALID	161002	The data type or data format of scale, offset, or out is not within the supported range.
  		The shape of scale or offset is not within the supported range.
  		The value of roundMode is not within the supported range.

aclnnTransQuantParamV3

• Parameters:

  Parameter	Input/Output	Description
  workspace	Input	Workspace memory address allocated on the Device side.
  workspaceSize	Input	Workspace size allocated on the Device side, obtained from the first-phase API aclnnTransQuantParamV3GetWorkspaceSize.
  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

• Atlas inference products, Atlas A2 training products/Atlas A2 inference products, Atlas A3 training products/Atlas A3 inference products, Ascend 950PR/Ascend 950DT: This API supports use with matmul class operators (such as aclnnQuantMatmulV4).

• Atlas inference products, Atlas A2 training products/Atlas A2 inference products, Atlas A3 training products/Atlas A3 inference products: This API does not support use with grouped matmul class operators (such as aclnnGroupedMatmulV4).

• The shapes of scale, offset, and out are described as follows:

  • When there is no offset, the out shape is the same as the scale shape.

    • If out is used as a matmul class operator input (such as aclnnQuantMatmulV4), its shape supports 1D (1,), (n,) or 2D (1, n), where n is consistent with the shape n of the right matrix (corresponding parameter x2) in the matmul computation.

    • If out is used as a grouped matmul class operator input (such as aclnnGroupedMatmulV4), it is only used when the grouping mode is m-axis grouping (corresponding parameter groupType is 0). Its shape supports 1D (g,) or 2D (g, 1), (g, n), where n is consistent with the shape n of the right matrix (corresponding parameter x2) in the grouped matmul computation, and g is consistent with the number of groups (corresponding parameter groupListOptional's shape size) in the grouped matmul computation.

  • When offset is present, it is only used as a matmul class operator input (such as aclnnQuantMatmulV4):

    • The shapes of offset, scale, and out support 1D (1,), (n,) or 2D (1, n), where n is consistent with the shape n of the right matrix (corresponding parameter x2) in the matmul computation.

    • When the input scale's shape is 1D, the out shape is also 1D, and its shape size is the maximum of the single-dimension shape sizes of scale and offset.

    • When the shape of the input scale is 2D, the shape of out has exactly the same dimensions and size as the shape of the input scale.

• Deterministic computation:

  • aclnnTransQuantParamV3 defaults to a deterministic implementation.

Example

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

#include <memory>
#include <iostream>
#include <vector>
#include "acl/acl.h"
#include "aclnnop/aclnn_trans_quant_param_v3.h"

#define CHECK_RET(cond, return_expr) \
    do {                             \
        if (!(cond)) {               \
            return_expr;             \
        }                            \
    } while (0)

#define CHECK_FREE_RET(cond, return_expr) \
    do {                                  \
        if (!(cond)) {                    \
            Finalize(deviceId, stream);   \
            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;
}

int Init(int32_t deviceId, aclrtStream* stream)
{
    // Boilerplate, 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 the host-side data to the 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);

    // Calculate 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 the aclCreateTensor API.
    *tensor = aclCreateTensor(
        shape.data(), shape.size(), dataType, strides.data(), 0, aclFormat::ACL_FORMAT_ND, shape.data(), shape.size(),
        *deviceAddr);
    return 0;
}

void Finalize(int32_t deviceId, aclrtStream stream)
{
    aclrtDestroyStream(stream);
    aclrtResetDevice(deviceId);
    aclFinalize();
}

int aclnnTransQuantParamV3Test(int32_t deviceId, aclrtStream& stream)
{
    auto ret = Init(deviceId, &stream);
    CHECK_RET(ret == ACL_SUCCESS, 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> offsetShape = {3};
    std::vector<int64_t> scaleShape = {3};
    std::vector<int64_t> outShape = {3};
    void* scaleDeviceAddr = nullptr;
    void* offsetDeviceAddr = nullptr;
    void* outDeviceAddr = nullptr;
    aclTensor* scale = nullptr;
    aclTensor* offset = nullptr;
    aclTensor* out = nullptr;
    std::vector<float> scaleHostData = {1, 1, 1};
    std::vector<float> offsetHostData = {1, 1, 1};
    std::vector<uint64_t> outHostData = {1, 1, 1};
    int64_t roundMode = 1;
    // Create the scale aclTensor.
    ret = CreateAclTensor(scaleHostData, scaleShape, &scaleDeviceAddr, aclDataType::ACL_FLOAT, &scale);
    std::unique_ptr<aclTensor, aclnnStatus (*)(const aclTensor*)> scaleTensorPtr(scale, aclDestroyTensor);
    std::unique_ptr<void, aclError (*)(void*)> scaleDeviceAddrPtr(scaleDeviceAddr, aclrtFree);
    CHECK_RET(ret == ACL_SUCCESS, return ret);
    // Create the offset aclTensor.
    ret = CreateAclTensor(offsetHostData, offsetShape, &offsetDeviceAddr, aclDataType::ACL_FLOAT, &offset);
    std::unique_ptr<aclTensor, aclnnStatus (*)(const aclTensor*)> offsetTensorPtr(offset, aclDestroyTensor);
    std::unique_ptr<void, aclError (*)(void*)> offsetDeviceAddrPtr(offsetDeviceAddr, aclrtFree);
    CHECK_RET(ret == ACL_SUCCESS, return ret);
    // Create the out aclTensor.
    ret = CreateAclTensor(outHostData, outShape, &outDeviceAddr, aclDataType::ACL_UINT64, &out);
    std::unique_ptr<aclTensor, aclnnStatus (*)(const aclTensor*)> outTensorPtr(out, aclDestroyTensor);
    std::unique_ptr<void, aclError (*)(void*)> outDeviceAddrPtr(outDeviceAddr, aclrtFree);
    CHECK_RET(ret == ACL_SUCCESS, return ret);

    // 3. Call the CANN operator library API. Replace it with the specific API name.
    uint64_t workspaceSize = 0;
    aclOpExecutor* executor;
    // Call the first-phase API of aclnnTransQuantParamV3.
    ret = aclnnTransQuantParamV3GetWorkspaceSize(scale, offset, roundMode, out, &workspaceSize, &executor);
    CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnTransQuantParamV3GetWorkspaceSize failed. ERROR: %d\n", ret);
              return ret);
    // Allocate device memory based on the workspaceSize calculated by the first-phase API.
    void* workspaceAddr = nullptr;
    std::unique_ptr<void, aclError (*)(void*)> workspaceAddrPtr(nullptr, aclrtFree);
    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);
        workspaceAddrPtr.reset(workspaceAddr);
    }
    // Call the second-phase API of aclnnTransQuantParamV3.
    ret = aclnnTransQuantParamV3(workspaceAddr, workspaceSize, executor, stream);
    CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnTransQuantParamV3 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-side memory to the host side. Modify this based on the specific API definition.
    auto size = GetShapeSize(outShape);
    std::vector<uint64_t> resultData(size, 0);
    ret = aclrtMemcpy(
        resultData.data(), resultData.size() * sizeof(resultData[0]), outDeviceAddr, 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 ret);
    for (int64_t i = 0; i < size; i++) {
        LOG_PRINT("result[%ld] is: %lu\n", i, resultData[i]);
    }

    return ACL_SUCCESS;
}

int main()
{
    // 1. (Boilerplate) Initialize the device/stream. See the ACL API manual.
    // Enter the actual device ID.
    int32_t deviceId = 0;
    aclrtStream stream;
    auto ret = aclnnTransQuantParamV3Test(deviceId, stream);
    CHECK_FREE_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnTransQuantParamV3Test failed. ERROR: %d\n", ret); return ret);

    Finalize(deviceId, stream);
    return 0;
}