aclnnDynamicQuant

📄 View Source Code

Applicable Products

Function

• Description: Performs per-token symmetric dynamic quantization on the input tensor.

• Formula:

  • If smoothScalesOptional is not input, then
  $$scaleOut=row\_max(abs(x))/dtypeMax$$ $$yOut=round(x/scaleOut)$$
  • If smoothScalesOptional is input, then
  $$input = x\cdot smoothScalesOptional$$ $$scaleOut=row\_max(abs(input))/dtypeMax$$ $$yOut=round(input/scaleOut)$$

  Where row_max represents the maximum value of each row, and dtypeMax is the maximum value of the output data type.

Function Prototype

Each operator is divided into a two-phase API. You must first call the "aclnnDynamicQuantGetWorkspaceSize" API to obtain the workspace size required for computation and the executor that encapsulates the operator computation flow, and then call the "aclnnDynamicQuant" API to perform the computation.

aclnnStatus aclnnDynamicQuantGetWorkspaceSize(
  const aclTensor* x,
  const aclTensor* smoothScalesOptional,
  const aclTensor* yOut,
  const aclTensor* scaleOut,
  uint64_t*        workspaceSize,
  aclOpExecutor**  executor)

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

aclnnDynamicQuantGetWorkspaceSize

• Parameters:

  Parameter	Input/Output	Description	Instruction	Data Type	Data Format	Dimension (Shape)	Non-contiguous Tensor
  x (aclTensor*)	Input	The input tensor of the operator. Corresponds to `x` in the formula.	Supports empty tensors.	FLOAT16, BFLOAT16	ND	2-8	√
  smoothScalesOptional (aclTensor*)	Input	SmoothScales input to the operator. Corresponds to `smoothScalesOptional` in the corresponding formula.	Supports empty tensor. The data type must be the same as that of x. The shape dimension is the same as the last dimension of x.	FLOAT16, BFLOAT16	ND	1	√
  yOut (aclTensor*)	Output	Quantized output tensor, corresponding to `yOut` in the formula.	Supports empty tensors. When the type is INT32, the last dimension of the shape is 1/8 of the last dimension of x, and the other dimensions are consistent with x. For other types, the shape is consistent with x.	INT4, INT8, FLOAT8_E4M3FN, FLOAT8_E5M2, HIFLOAT8, INT32	ND	2-8	√
  scaleOut (aclTensor*)	Output	Scale used for quantization. Corresponds to `scaleOut` in the formula.	Supports empty tensor. The shape has the same number of dimensions as x, excluding the last dimension.	FLOAT32	ND	1-7	√
  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.	-	-	-	-	-

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

    The data type of the output yOut supports only INT4, INT8, and INT32.

  • Atlas inference products and Atlas training products:

    • The data type of the input x and smoothScalesOptional supports only FLOAT16.

    • The data type of the output parameter yOut only supports INT8.

    • The input parameter smoothScalesOptional is a reserved parameter and does not participate in computation in the current version.

• 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 x or out parameter passed in is a null pointer.
  ACLNN_ERR_PARAM_INVALID	161002	The data type, data format, or dimension of a parameter is not within the supported range.
  		When the data type of yOut is INT4, the last dimension of x or yOut is not an even number.
  		When the data type of yOut is INT32, the last dimension of x is not divisible by 8.
  ACLNN_ERR_INNER_CREATE_EXECUTOR	561001	Internal creation of aclOpExecutor fails.

aclnnDynamicQuant

• Parameters:

  Parameter	Input/Output	Description
  workspace	Input	Address of the workspace memory applied for on the Device side.
  workspaceSize	Input	Size of the workspace applied for on the Device side, obtained through the first-phase API aclnnDynamicQuantGetWorkspaceSize.
  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 A2 training products/Atlas A2 inference products, Atlas A3 training products/Atlas A3 inference products, Ascend 950PR/Ascend 950DT:

  • When the data type of yOut is INT4, the last dimension of x and yOut must be divisible by 2.

  • When the data type of yOut is INT32, the last dimension of x must be divisible by 8.

• Atlas inference products and Atlas training products: The last dimension of input x must be divisible by 32. Currently, only symmetric quantization is supported, and the BFLOAT16 data type is not supported.

• Deterministic computation:

  • aclnnDynamicQuant 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 <iostream>
#include <vector>
#include "acl/acl.h"
#include "aclnnop/aclnn_dynamic_quant.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<int8_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("mean result[%ld] is: %d\n", i, resultData[i]);
    }
}

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

    // 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];
    }

    // 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.
    // Set deviceId based on the actual device.
    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);

    // 2. Construct the input and output. Custom construction is required based on the API interface.
    int rowNum = 4;
    int rowLen = 32;
    std::vector<int64_t> xShape = {rowNum, rowLen};
    std::vector<int64_t> smoothShape = {rowLen};
    std::vector<int64_t> yShape = {rowNum, rowLen};
    std::vector<int64_t> scaleShape = {rowNum};

    void* xDeviceAddr = nullptr;
    void* smoothDeviceAddr = nullptr;
    void* yDeviceAddr = nullptr;
    void* scaleDeviceAddr = nullptr;

    aclTensor* x = nullptr;
    aclTensor* smooth = nullptr;
    aclTensor* y = nullptr;
    aclTensor* scale = nullptr;

    std::vector<aclFloat16> xHostData;
    std::vector<aclFloat16> smoothHostData;
    std::vector<int8_t> yHostData;
    std::vector<float> scaleHostData;
    for (int i = 0; i < rowNum; ++i) {
        for (int j = 0; j < rowLen; ++j) {
            float value1 = i * rowLen + j;
            xHostData.push_back(aclFloatToFloat16(value1));
            yHostData.push_back(0);
        }
        scaleHostData.push_back(0);
    }

    for (int k = 0; k < rowLen; ++k) {
        float value2 = k * rowLen + 1;
        smoothHostData.push_back(aclFloatToFloat16(value2));
    }

    // Create the x aclTensor.
    ret = CreateAclTensor(xHostData, xShape, &xDeviceAddr, aclDataType::ACL_FLOAT16, &x);
    CHECK_RET(ret == ACL_SUCCESS, return ret);
    // Create the smooth aclTensor.
    ret = CreateAclTensor(smoothHostData, smoothShape, &smoothDeviceAddr, aclDataType::ACL_FLOAT16, &smooth);
    CHECK_RET(ret == ACL_SUCCESS, return ret);
    // Create the y aclTensor.
    ret = CreateAclTensor(yHostData, yShape, &yDeviceAddr, aclDataType::ACL_INT8, &y);
    CHECK_RET(ret == ACL_SUCCESS, return ret);
    // Create the scale aclTensor.
    ret = CreateAclTensor(scaleHostData, scaleShape, &scaleDeviceAddr, aclDataType::ACL_FLOAT, &scale);
    CHECK_RET(ret == ACL_SUCCESS, return ret);

    // 3. Call the CANN operator library API. Replace with the actual API name.
    uint64_t workspaceSize = 0;
    aclOpExecutor* executor;

    // Call the first-phase API of aclnnDynamicQuant.
    ret = aclnnDynamicQuantGetWorkspaceSize(x, smooth, y, scale, &workspaceSize, &executor);
    CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnDynamicQuantGetWorkspaceSize 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 aclnnDynamicQuant.
    ret = aclnnDynamicQuant(workspaceAddr, workspaceSize, executor, stream);
    CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnDynamicQuant 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 values and copy the result from the device memory to the host side. Modify based on the specific API definition.
    PrintOutResult(yShape, &yDeviceAddr);

    // 6. Release aclTensor. Modify based on the specific API definition.
    aclDestroyTensor(x);
    aclDestroyTensor(smooth);
    aclDestroyTensor(y);
    aclDestroyTensor(scale);

    // 7. Release device resources.
    aclrtFree(xDeviceAddr);
    aclrtFree(smoothDeviceAddr);
    aclrtFree(yDeviceAddr);
    aclrtFree(scaleDeviceAddr);
    if (workspaceSize > 0) {
        aclrtFree(workspaceAddr);
    }
    aclrtDestroyStream(stream);
    aclrtResetDevice(deviceId);
    aclFinalize();

    return 0;
}