aclnnTransQuantParam

📄 View Source Code

Applicable Products

Function

• Description: Converts the input scale data from FLOAT32 type to UINT64 type required by the hardware, and stores it in quantParam.

• Formula:

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

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

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

  3. Perform subsequent calculations based on the offset value:

     • If offset does not exist, no further calculation 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 9 bits of offset and store them in bits 37 to 45 of out.

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

Function Prototype

aclnnStatus aclnnTransQuantParam(
  const float  *scaleArray,
  uint64_t      scaleSize,
  const float  *offsetArray,
  uint64_t      offsetSize,
  uint64_t    **quantParam,
  uint64_t     *quantParamSize)

aclnnTransQuantParam

• Parameters:

  Parameter	Input/Output	Description	Instruction	Data Type	Data Format	Dimension (Shape)	Non-contiguous Tensor
  scaleArray (float*)	Input	Represents the memory pointer to the scale data, corresponding to `scale` in the formula.	Ensure that the scale data contains no NaN or inf values.	-	-	-	-
  scaleSize (uint64_t)	Input	Indicates the number of scale data.	Ensure that scaleSize and scaleArray contain the same number of elements.	-	-	-	-
  offsetArray (float*)	Input	Indicates the memory that stores the offset data, corresponding to `offset` in the formula.	Ensure that the offset data contains no NaN or inf values.	-	-	-	-
  offsetSize (uint64_t)	Input	Indicates the number of offset data.	You need to ensure that `offsetSize` and `offsetArray` contain the same number of elements.	-	-	-	-
  quantParam (uint64_t**)	Output	Indicates the address pointing to the memory that stores the converted quantParam data, corresponding to `out` in the formula.	-	-	-	-	-
  quantParamSize (uint64_t*)	Output	Indicates the number of elements storing quantParam data.	You need to ensure that `quantParamSize` matches the number of elements contained in `quantParam`.	-	-	-	-

• Return value

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

The API performs input parameter validation and returns an error in the following scenarios:

Constraints

• Deterministic computation:

  • aclnnTransQuantParam 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 "acl/acl.h"
#include "aclnnop/aclnn_trans_quant_param.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)

int main()
{
    float scaleArray[3] = {1.0, 1.0, 1.0};
    uint64_t scaleSize = 3;
    float offsetArray[3] = {1.0, 1.0, 1.0};
    uint64_t offsetSize = 3;
    uint64_t* result = nullptr;
    uint64_t resultSize = 0;
    auto ret = aclnnTransQuantParam(scaleArray, scaleSize, offsetArray, offsetSize, &result, &resultSize);
    CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnTransQuantParam failed. ERROR: %d\n", ret); return ret);
    for (auto i = 0; i < resultSize; i++) {
        LOG_PRINT("result[%d] is: %ld\n", i, result[i]);
    }
    free(result);
    return 0;
}