FusedMulAdd

Applicability

Product	Supported
Atlas 350 Accelerator Card	√
Atlas A3 training product / Atlas A3 inference product	√
Atlas A2 training product / Atlas A2 inference product	√
Atlas 200I/500 A2 inference product	√
Atlas inference product AI Core	√
Atlas inference product Vector Core	x
Atlas training product	x

Function Usage

Multiplies src0 and dst element-wise, adds the result to src1, and stores the final result back to dst. The formula is as follows:

$\text{[math]}$

Prototype

Computation of the first n pieces of data of a tensor

        
             template <typename T>
__aicore__ inline void FusedMulAdd(const LocalTensor<T>& dst, const LocalTensor<T>& src0, const LocalTensor<T>& src1, const int32_t& count)

High-dimensional tensor sharding computation

Bitwise mask mode

          
               template <typename T, bool isSetMask = true>
__aicore__ inline void FusedMulAdd(const LocalTensor<T>& dst, const LocalTensor<T>& src0, const LocalTensor<T>& src1, uint64_t mask[], const uint8_t repeatTime, const BinaryRepeatParams& repeatParams)

Contiguous mask mode

          
               template <typename T, bool isSetMask = true>
__aicore__ inline void FusedMulAdd(const LocalTensor<T>& dst, const LocalTensor<T>& src0, const LocalTensor<T>& src1, uint64_t mask, const uint8_t repeatTime, const BinaryRepeatParams& repeatParams)

Parameters

**Table 1** Template parameters
Parameter	Description
T	Operand data type. For the Atlas 350 Accelerator Card, the supported data types are half, bfloat16_t, float, int64_t, and uint64_t. For the Atlas A3 training product / Atlas A3 inference product , the supported data types are half and float. For the Atlas A2 training product / Atlas A2 inference product , the supported data types are half and float. For the Atlas 200I/500 A2 inference product , the supported data types are half and float. For the Atlas inference product AI Core, the supported data types are half and float.
isSetMask	Indicates whether to set mask inside the API. true: sets mask inside the API. false: sets mask outside the API. Developers need to use the SetVectorMask API to set the mask value. In this mode, the mask value in the input parameter of this API must be set to the placeholder MASK_PLACEHOLDER.

**Table 2** Parameters
Parameter	Input/Output	Description
dst	Input/Output	Destination operand. The type is LocalTensor, and TPosition can be VECIN, VECCALC, or VECOUT. The start address of LocalTensor must be 32-byte aligned.
src0/src1	Input	Source operands. The type is LocalTensor, and TPosition can be VECIN, VECCALC, or VECOUT. The start address of LocalTensor must be 32-byte aligned. The two source operands must have the same data type as the destination operand.
count	Input	Number of elements involved in the computation.
mask[]/mask	Input	mask controls the elements that participate in computation in each iteration. Bitwise mode: controls the elements that participate in computation by bit. If a bit is set to 1, the corresponding element participates in the computation. If a bit is set to 0, the corresponding element is masked from the computation. The mask value is an array. The array length and the value range of the array elements are related to the operand data type. When the operand is 16-bit, the array length is 2, with mask[0] and mask[1] each in the range [0, 2⁶⁴ – 1], and they cannot both be 0 at the same time. When the operand is 32-bit, the array length is 1, with mask[0] in the range (0, 2⁶⁴ – 1]. When the operand is 64-bit, the array length is 1, with mask[0] in the range (0, 2³² – 1]. For example, if mask = [0, 8] and 8 = 0b1000, only the fourth element participates in computation. Contiguous mode: indicates the number of contiguous elements that participate in computation. The value range is related to the operand data type. The maximum number of elements that can be processed in each iteration varies according to the data type. When the operand is 16-bit, mask ∈ [1, 128]. When the operand is 32-bit, mask ∈ [1, 64]. When the operand is 64-bit, mask ∈ [1, 32].
repeatTime	Input	Number of iteration repeats. The Vector Unit reads 256 bytes of contiguous data for computation each time. To read the complete data for processing, the unit needs to read the input data in multiple repeats. repeatTime indicates the number of iterations. For details about this parameter, see High-dimensional Sharding APIs.
repeatParams	Input	Parameters that control the operand address strides. They are of the BinaryRepeatParams type, and contain such parameters as those that specify the address stride of the operand for the same data block between adjacent iterations and address stride of the operand between different data blocks in a single iteration. For details about the address stride of the operand between adjacent iterations, see repeatStride. For details about the address stride of the operand between different data blocks in a single iteration, see dataBlockStride.

Returns

None

Restrictions

For details about the operand address alignment requirements, see General Address Alignment Restrictions.
For details about the constraints on operand address overlapping, see General Address Overlapping Restrictions.

For the Atlas 350 Accelerator Card, uint64_t and int64_t support only the APIs that compute the first n pieces of data in a tensor.

Examples

Example of high-dimensional sharding computation API - contiguous mask mode (half precision)

        
             uint64_t mask = 128;
// repeatTime = 2. 128 elements are computed in one iteration, and 256 elements are computed in total.
// dstBlkStride, src0BlkStride, src1BlkStride = 1. Data is continuously read and written in a single iteration.
// dstRepStride, src0RepStride, src1RepStride = 8. Data is continuously read and written between adjacent iterations.
AscendC::FusedMulAdd(dstLocal, src0Local, src1Local, mask, 2, { 1, 1, 1, 8, 8, 8 });

Example of high-dimensional sharding computation API - bitwise mask mode (half-type input)

        
             uint64_t mask[2] = { UINT64_MAX, UINT64_MAX };
// repeatTime = 2. 128 elements are computed in one iteration, and 256 elements are computed in total.
// dstBlkStride, src0BlkStride, src1BlkStride = 1. Data is continuously read and written in a single iteration.
// dstRepStride, src0RepStride, src1RepStride = 8. Data is continuously read and written between adjacent iterations.
AscendC::FusedMulAdd(dstLocal, src0Local, src1Local, mask, 2, { 1, 1, 1, 8, 8, 8 });

Example of computing the first n pieces of data of a tensor (half-type input)

        
             AscendC::FusedMulAdd(dstLocal, src0Local, src1Local, 256);

Result example:

Input (src0Local):
[ 51.4     25.92    28.3     26.62    -4.523  -93.6      8.71    21.02
 -49.8    -80.4     95.94     4.74    70.25   -60.38    79.5    -18.19
  79.1    -66.7     62.5     83.6    -73.1     72.56   -63.03    81.3
 -62.28   -49.94   -78.7     56.2    -52.97    66.2     43.94    13.43
  43.06    51.16    34.16   -64.56   -73.1    -94.7    -68.94    73.06
 -93.4    -46.62   -83.75    15.46    32.88   -76.4     53.84    70.6
  -3.455  -88.4    -65.75   -16.1     88.9    -70.56   -69.44   -11.91
 -77.06    76.06    22.73    91.25   -96.06    64.4     13.1     30.56
 -99.2    -98.56    58.1     31.92   -56.47   -72.7    -42.94    49.1
  98.44    83.75    -4.336   30.03    33.2    -54.78   -21.19   -57.22
 -61.34   -39.8    -29.44   -16.12    38.4    -71.6    -28.52    63.62
  36.      61.38    26.69   -16.34    92.2    -67.7    -92.75   -41.16
 -85.44   -91.2    -22.31   -47.38   -27.28   -77.44    64.     -78.56
  52.22   -61.8    -84.94   -64.7     91.3     64.56    67.25    65.44
  58.7     64.7    -75.06   -44.7    -22.05    71.7     78.9     34.7
 -26.88    39.7    -83.3      0.6274 -34.56   -94.7     -7.027   86.1
 -15.18    63.47    39.     -53.1     54.53    75.44    32.53   -78.3
 -22.34    74.7     -4.312  -33.2      2.19    98.25   -14.66    34.88
   6.746   88.5    -55.03    88.25   -79.56   -61.62    82.3     47.47
 -17.19    45.72   -71.4    -93.5    -32.84   -40.      49.88    27.98
 -70.56   -47.6     76.25    54.62   -62.06   -13.484   86.94    21.81
 -54.53     2.236  -25.16    86.75   -45.97   -44.5    -54.     -53.6
 -23.33    58.62   -48.16   -17.52    87.75   -60.7    -80.6     74.5
  66.4     70.7    -34.28    -3.43   -88.5    -43.56   -22.9    -93.5
 -95.75   -90.6     57.97   -60.06    75.94   -92.3    -15.87    38.5
  37.34   -80.56    71.25    69.      36.25    55.53    74.75    31.1
  -8.445    2.152   95.75    -4.777    9.41    97.8    -64.75    90.94
  38.84    16.8    -17.44    87.     -11.336   98.1    -94.6    -76.2
 -20.14    18.1    -90.4     51.84   -22.88    20.33    45.38    96.06
 -68.56   -57.66    61.78   -78.3     76.     -26.23    27.36   -52.5
 -90.4     23.78   -47.7    -36.      68.4    -59.2    -59.28    32.12
 -44.84    -2.428   -9.266   57.44    66.25   -62.8     92.8     50.75  ]
Input (src1Local):
[-43.1     -90.       -0.7295   49.28    -52.12    -55.53     99.6
  94.4      62.56     20.67    -25.4     -70.6      43.44     57.97
   5.355    38.66    -67.3     -26.72     43.7     -81.06     54.47
 -71.3      84.8      92.9      49.88    -49.94     79.75    -71.8
   6.5      42.3      22.44      6.64    -83.6     -24.3     -97.06
  43.47    -31.06     -9.55     -7.734   -30.27     70.3      10.91
  55.72     60.03     85.8     -21.86    -34.28     -1.962   -11.18
 -20.4      53.34    -44.72      9.28    -40.44     19.42     62.66
 -84.75     39.1      45.9     -89.56     70.94     99.5      16.62
 -36.7     -26.2      87.06    -87.94     19.      -30.12     16.94
 -85.44    -17.06     33.28    -49.84    -24.78    -58.25     27.81
 -23.48     81.06     82.94    -35.88     -4.47    -74.7      85.
 -18.22    -67.5      76.5      96.1     -32.4     -45.56     21.53
 -28.5      88.       -0.1978  -20.34    -44.53    -13.27     -7.93
  33.3      89.8      49.12    -19.84     48.38     83.9      53.
 -65.6      62.97     76.75    -74.4     -23.19     73.1      -9.38
 -31.86     69.44    -52.47    -75.94     54.78     78.94    -74.9
  -0.01271  21.88    -82.9     -34.44     20.56     64.25      7.57
 -63.6     -78.44     19.06     67.25     34.62     82.4      15.6
  84.06     75.06    -97.94    -41.12    -77.75     35.3      88.
  -2.758    -4.36     34.8      73.94    -33.28    -24.5      53.38
 -54.3      19.14     57.1      43.4     -39.4      16.36     24.94
 -42.94    -26.25     27.92    -35.44     79.94     42.12    -62.72
  90.       98.75     -8.22     79.      -96.94    -91.1     -32.94
  64.4     -78.56    -49.56     10.23     82.9     -27.        7.023
 -42.7     -25.67    -42.34     22.72     98.56     69.2     -72.9
  60.28    -43.94     73.06    -28.31     37.5      84.8      -1.514
  81.7     -68.3      46.3      66.       86.44    -26.78    -52.72
  -3.766   -17.95     87.6     -93.5       9.13     58.25     44.62
   2.64     90.25    -42.16    -50.62     18.48     -3.156    36.16
  82.4      26.44     69.8     -47.56    -54.72     58.88    -16.77
  58.44     -5.62    -38.88     40.44    -87.94     33.25     68.94
 -73.2       3.64     45.3     -59.1     -69.9     -94.5      16.02
  12.11     91.56    -30.4     -47.56     56.84    -17.06     60.
 -31.33     50.      -40.44      9.17     31.9     -51.2     -34.72
 -11.43     19.58    -89.6     -61.53    -98.25     94.94     43.8
  -6.848   -99.       99.9     -28.66   ]
Input (dstLocal):
[-18.36     44.94     32.34    -50.      -99.5      89.4       1.568
  61.4     -36.      -46.28     54.88     14.92     61.2     -16.14
  42.72     87.25     -2.787   -66.8      58.3      23.94     89.
 -78.56    -38.94    -78.25     52.6      77.94     84.4     -63.94
  90.7      51.97     84.1      99.3     -70.1      -3.691     4.16
  24.98    -91.94     91.2      65.3     -47.28     68.1     -60.2
 -90.4       0.1636    9.32     51.4      10.45     46.9      42.78
 -38.2     -39.25    -79.1      52.2      56.03    -20.72    -25.81
  27.5     -42.94    -71.6     -73.2     -16.45     99.      -16.
 -75.94    -18.44      6.92    -54.66     12.016    35.53     65.6
  84.4      55.28     34.16    -69.5      -0.4287  -83.1     -30.69
  86.06    -67.56    -97.56    -23.44     73.56    -29.84     -0.49
 -78.44    -17.45    -19.47     57.12     28.31    -86.8      95.44
  82.4      40.53     92.7      20.36    -77.75    -44.62    -21.
 -63.44     71.      -32.2      66.7      46.94    -95.3     -71.8
  -1.351    20.95    -84.44     33.28    -55.2      10.17     22.27
 -26.42    -76.06     90.       44.2     -80.4     -94.25     -9.055
  15.44     75.94    -40.47    -43.78     18.31     -5.586   -55.1
 -95.9      82.5      75.6     -76.25    -83.6     -70.      -54.16
  56.62     64.56    -97.06     93.5     -58.8       7.746   -12.164
  53.06    -56.72     97.7     -11.07     68.2     -77.3      77.8
 -41.44    -14.21    -48.56    -40.5      88.44     85.5     -92.25
  -8.39    -41.78     27.44     85.44     40.8     -80.2      38.94
  -7.598     3.83     74.      -97.06    -84.3      74.9     -88.6
 -92.3     -34.97     65.8     -60.8      39.06    -19.64     65.4
   5.336    -0.07324 -52.16     70.44     11.75      4.72     77.8
 -62.9      80.3     -24.66     62.84    -98.75    -58.88    -90.4
  31.53     70.44    -97.7      64.8      -7.203   -28.3      30.5
 -48.75    -89.06     25.94    -81.06    -62.72    -65.4      79.4
 -90.7      65.6      79.56    -21.22     -0.1489   83.44    -29.86
 -75.9      29.75     53.34     47.66     18.02     92.7      90.56
   8.2     -98.6     -40.53     15.484    28.47    -49.28     24.05
 -23.       -0.0836   50.5     -70.1     -96.9     -10.23    -19.34
 -86.6      63.38    -88.4      92.5     -16.52    -94.7     -92.1
  82.2     -35.94    -67.5     -46.03    -70.44    -54.44    -85.5
  48.72    -61.25     41.      -51.25    -50.4     -50.66     51.84
 -52.9      11.086   -74.1      50.66   ]
Output (dstLocal):
[ -987.      1075.       914.5    -1282.       398.     -8424.
   113.3     1384.      1856.      3740.      5240.         0.0625
  4344.      1032.      3402.     -1548.      -287.8     4428.
  3688.      1921.     -6452.     -5772.      2538.     -6272.
 -3226.     -3942.     -6560.     -3664.     -4796.      3482.
  3718.      1341.     -3104.      -213.1       45.06   -1570.
  6692.     -8640.     -4512.     -3484.     -6288.      2816.
  7624.        62.56     392.2    -3948.       528.5     3310.
  -159.      3354.      2634.      1228.      4648.     -3994.
  1458.       370.     -2204.     -3226.     -1582.     -6768.
  1652.      6472.      -193.     -2356.      1803.      -595.
 -3264.       402.5    -2036.     -4756.     -3708.      2696.
  3396.     -5868.       -22.92   -2554.      -990.5    -4740.
  1512.      5668.      1402.     -2932.       804.        92.9
 -3030.      1182.       631.5     3730.       986.5    -5372.
  2568.     -1374.      3824.     -6272.     -1908.      3156.
  3798.      1907.      1448.     -3274.       927.     -5184.
  3052.      7572.     -3696.        17.88   -1717.      5540.
  2964.     -3588.       757.      1448.     -1583.     -4852.
 -6808.     -2050.      1827.     -6676.      -789.       535.5
 -2019.     -1689.      3614.        32.06     257.5     5224.
   610.      7024.     -1129.     -4772.     -3228.      3798.
 -2938.      4356.      2176.      7504.     -2132.     -4468.
     1.906    491.8      113.44   -5576.     -1397.      -312.
   426.8    -6868.     -4232.     -3710.      1150.      3050.
 -3290.      4160.     -1454.     -4192.       556.      3880.
  -873.     -3454.      2116.     -2202.     -2810.       451.5
   390.8     4034.      6104.      1040.      6416.     -1966.
  5096.      -156.8    -1706.     -5264.     -1713.       847.
 -3522.      -328.8      -23.97   -3100.     -3370.      -107.2
   483.2    -4796.      5132.      5940.     -1564.      4416.
  3424.       286.8     8000.     -1292.     -1681.      9184.
 -6136.       739.5    -1667.     -1885.     -3706.      8208.
  -324.     -3214.     -2332.      5328.      5700.     -6252.
  2468.      4376.     -1637.        13.86    -707.5      -28.1
 -7180.      -115.7      571.5     4612.     -1222.      8488.
  3502.       196.2     1714.     -3564.      -135.      2706.
  4696.     -1763.       390.         2.129  -4520.     -3696.
  2146.      -302.5     -861.5    -8304.     -4252.      5064.
  5668.      1350.     -7212.      2476.      2216.      1937.
  6060.     -1086.      3390.      1909.     -5884.     -2896.
  3650.      1227.      2236.        24.12     564.      3022.
 -3512.      -795.5    -6780.      2542.    ]

Parent topic: Compound Computation