About preset events(PAPI_HP_OPS) for half-precision floating-point operations in the papi library

[masahiko yamada]

Dear papi library development team,

When and at what version level is the half-precision floating-point preset event PAPI_HP_OPS supported in the papi library?

The ARM processor supports events for the following half-precision floating point operations:.

FP_HP_SCALE_OPS_SPEC, FP_HP_FIXED_OPS_SPEC

The FUJITSU A64FX processor implements FP_HP_SCALE_OPS_SPEC, FP_HP_FIXED_OPS_SPEC.

If possible, PAPI_HP_OPS should be used.

The following settings will be made.

PAPI_HP_OPS = FP_HP_SCALE_OPS_SPEC x VL / 128 + FP_HP_FIXED_OPS_SPEC

The VL is 512 for the FUJITSU A64FX processor.

Best regards.

Masahiko Yamada