search

loongson / la-abi-specs

25 stars 6 forks source link

How does loongarch deal with herb sutter's P0709? It needs an extra register for passing the error. #11

Open trcrsired opened 3 months ago

trcrsired commented 3 months ago

https://www.open-std.org/jtc1/sc22/wg14/www/docs/n2289.pdf

For fails(E) returns, it is proposed for at least AArch64, ARM, x86 and x64, that the discriminant be returned via the CPU's carry flag. This is because compilers can often fold the setting or clearing of the CPU's carry flag into the ordering of other operations, thus making this a zero runtime overhead choice of discriminant1 . On other architectures such as RISC-V (which has no status register), an extra register would make more sense. It doesn't matter what an architecture chooses, so long as it is consistent across all compilers.

how do you define this?

xen0n commented 3 months ago

I would say that an extra GPR would be appropriate for LoongArch. Because while there is a "flags" register on current Loongson-3-series models, it's part of LBT context, so the base ABI cannot rely on that.