You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Optimize absl::Duration division and modulo: Allow the compiler to inline time_internal::IDivDuration, by splitting the slow path to a separate function.
With that change, the compiler can inline the fast path. This is specially important in the context of `Duration::operator%=`, because it allows proving that the return value is unused, therefore avoiding expensive multiplies and divides (e.g. `*q = num_hi / den_hi;`).
```
name old cpu/op new cpu/op delta
BM_Duration_Modulo 23.1ns ± 0% 22.5ns ± 0% -2.42% (p=0.000 n=20+16)
BM_Duration_Modulo_FastPath 7.05ns ± 0% 4.85ns ± 0% -31.17% (p=0.000 n=20+20)
name old time/op new time/op delta
BM_Duration_Modulo 23.1ns ± 0% 22.6ns ± 0% -2.43% (p=0.000 n=20+16)
BM_Duration_Modulo_FastPath 7.06ns ± 0% 4.86ns ± 0% -31.18% (p=0.000 n=20+20)
name old INSTRUCTIONS/op new INSTRUCTIONS/op delta
BM_Duration_Modulo 188 ± 0% 178 ± 0% -5.32% (p=0.000 n=20+20)
BM_Duration_Modulo_FastPath 84.0 ± 0% 62.0 ± 0% -26.19% (p=0.000 n=20+20)
name old CYCLES/op new CYCLES/op delta
BM_Duration_Modulo 73.8 ± 0% 72.1 ± 0% -2.27% (p=0.000 n=19+20)
BM_Duration_Modulo_FastPath 22.5 ± 0% 15.5 ± 0% -31.13% (p=0.000 n=19+20)
```
Note: We don't need to expose `absl::time_internal::IDivDuration` at all given that we have a public `absl::IDivDuration`.
PiperOrigin-RevId: 610710635
Change-Id: Ief7c3d5b1c000b397d931e9249edcaef96e7151e
0 commit comments