|
native 0.0.1
Vectors, masks and wide register packs for C++26
|
Raw OS query results retained for diagnostics and enhanced BF16 behavior. More...
Public Attributes | |
| bool | baseline_observed = false |
| True only after the OS baseline FP/Advanced SIMD query succeeds. | |
| bool | fp = false |
| OS permits baseline floating-point instructions and their register state. | |
| bool | asimd = false |
| OS permits baseline Advanced SIMD instructions and their register state. | |
| bool | fp16_observed = false |
| True only after an OS half-arithmetic feature query succeeds. | |
| bool | scalar_fp16 = false |
| OS advertises scalar half-precision arithmetic (FPHP/FEAT_FP16). | |
| bool | vector_fp16 = false |
| OS advertises Advanced SIMD half-precision arithmetic (ASIMDHP/FEAT_FP16). | |
| bool | bf16_observed = false |
| True only after an OS BF16 feature query succeeds. | |
| bool | bf16 = false |
| OS advertises non-streaming Advanced SIMD BF16 instructions (FEAT_BF16). | |
| bool | ebf16_observed = false |
| True only after an OS enhanced BF16 feature query succeeds. | |
| bool | ebf16 = false |
| isa< arm > | extra_observed {} |
Raw OS query results retained for diagnostics and enhanced BF16 behavior.
Definition at line 46 of file native.arm.features.ccm.
| bool native::arm_capabilities::raw_observations::ebf16 = false |
OS advertises enhanced BF16 behavior, usable when the caller sets FPCR.EBF. Also exposed as arm_feature::ebf16; NEON_BF16 does not require it.
Definition at line 67 of file native.arm.features.ccm.
Additional independent instruction-feature queries. A feature must occur in both sets; failed queries never authorize it.
Definition at line 70 of file native.arm.features.ccm.