Load2DMX

Atlas 350 Accelerator Card

√

Atlas A3 training product/Atlas A3 inference product

x

Atlas A2 training product/Atlas A2 inference product

x

Atlas 200I/500 A2 inference product

x

Atlas inference product AI Core

x

Atlas inference product Vector Core

x

Atlas training product

x

T

Data types of the source and destination operands.

• Load2DMX API:

  For the Atlas 350 Accelerator Card, the supported data types are fp4x2_e2m1_t, fp4x2_e1m2_t, fp8_e4m3fn_t, and fp8_e5m2_t.

U

• For the Load2DMX API, U indicates the data type of dst. When src is of type fp8_e4m3fn_t or fp8_e5m2_t, U must be the MX data type corresponding to T, that is, AscendC::mx_fp8_e4m3_t or AscendC::mx_fp8_e5m2_t. Otherwise, the compilation fails. For other data types, T and U must be identical.

dst

Output

Destination operand, which is of the LocalTensor type.

The sequential arrangement of data is determined by TPosition of the destination operand. The constraints are as follows:

• When storing data in A2, use the ZZ or NZ format, with the fractal matrix size of 16 × (32 bytes/sizeof(T)).
• When storing data in B2, use the ZN format, with the fractal matrix size of (32 bytes/sizeof(T)) × 16.
• When storing data in A1 or B1, there is no format restriction. Generally, the data is in NZ format, with the fractal matrix size of 16 × (32 bytes/sizeof(T)).

src

Input

Source operand, which is of the LocalTensor or GlobalTensor type.

Its data type must be the same as that of dst.

srcMx

Input

Source operand, which is of the LocalTensor type. Only the fp8_e8m0_t type is supported.

loadDataParams

Input

LoadData parameter structure. Supported types are as follows:

• LoadData2DMxParams. For details, see Table 3.

For details, see ${INSTALL_DIR}/include/ascendc/basic_api/interface/kernel_struct_mm.h. Replace ${INSTALL_DIR} with the actual path for storing files after the CANN software is installed.