On Tue, Feb 24, 2026 at 3:16 AM Samuel Thibault <[email protected]> wrote:
> Hello, > > [email protected], le mar. 24 févr. 2026 04:27:55 +0100, a ecrit: > > The commit message says "curr_spl is actually 4 bytes", but on x86_64, > > spl_t is typedef'd as unsigned long (i386/i386/spl.h line 34) > > I guess it's reading its own source code, where that was patched into > long. > That's what it thinks, too: Our CX8 review was wrong on two points. Damien replied saying the AI was > "hallucinating" — and he's right: > > 1. spl_t is int (4 bytes) on upstream. Our local branch had patched it > to unsigned long, and task 72 reviewed our local code, not upstream. The > spl.S > file doesn't need CX8 conversions. > 2. CX16 is impossible in one instruction. x86 addressing modes only > support scale factors of 1, 2, 4, 8. The CX(CX()) approach giving 4*4=16 is > the > correct solution. agape brent
