mirror of
https://codeberg.org/ziglang/zig
synced 2026-11-18 07:05:05 +00:00
A tag with "f32" assumes 32-bit wide floating point registers, while "f64" assumes 64-bit wide. Absence of either "f32" or "f64" means that the target is FPR width agnostic (O32) or the width is 64-bit by definition (R6, N64/N32). Tags with "n8" use IEEE 754-2008 NaN encoding. Similar to FPR width, it is omitted on tags that use 2008 encoding by definition (R6). Tags with "r6" need to exist because R6 is effectively an incompatible ISA; it re-encodes some of the instruction encoding space from earlier revisions. To be clear, as ridiculous as this is, the list of new tags here *is* already the curated list; the full list of tags that are architecturally valid is even longer, but I deliberately left out those that would have no realistic use. Even with that, we'll have to add more tags later for microMIPS and nanoMIPS... I expect https://github.com/ziglang/zig/issues/21507 to make all of this less painful once implemented. On the plus side, we can now fully determine the float ABI and standard dynamic linker path from the target triple for all supported targets. Besides just being nicer, this also means we no longer have a chicken-and-egg problem in target resolution when we want the ABI to affect default CPU features. closes https://codeberg.org/ziglang/zig/issues/31325
3067 lines
105 KiB
3067 lines
105 KiB