On Tue, 1 Apr 2025, Jakub Jelinek wrote:
> On Tue, Apr 01, 2025 at 01:36:23PM +0800, Hongtao Liu wrote:
> > >Changing ix86_valid_target_attribute_inner_p might be even better because
> > >OPT_msse4 is RejectNegative option, so !value for it looks weird.
> > msse4 is defined as ix86_opt_isa in ix86_valid_target_attribute_inner_p
> >
> > 1055 IX86_ATTR_ISA ("sse4", OPT_msse4),
> >
> > and would be handled in ix86_handle_option
> >
> > 1282 else if (type == ix86_opt_isa)
> > 1283 {
> > 1284 struct cl_decoded_option decoded;
> > 1285
> > 1286 generate_option (opt, NULL, opt_set_p, CL_TARGET,
> > &decoded);
> > 1287 ix86_handle_option (opts, opts_set,
> > 1288 &decoded, input_location);
> > 1289 }
> >
> > So I think it's already correct here.
>
> The entries are
> msse4
> Target RejectNegative Mask(ISA_SSE4_2) Var(ix86_isa_flags) Save
> Support MMX, SSE, SSE2, SSE3, SSSE3, SSE4.1 and SSE4.2 built-in functions and
> code generation.
>
> mno-sse4
> Target RejectNegative InverseMask(ISA_SSE4_1) Var(ix86_isa_flags) Save
> Do not support SSE4.1 and SSE4.2 built-in functions and code generation.
>
> so if it is turned into something without RejectNegative,
> it will be wrong in the Mask argument, either for the positive or for the
> negative case.
>
> I wonder though if msse4 should not be a Target RejectNegative Alias to
> msse4.2
> and mno-sse4 RejectNegative Alias to mno-sse4.1.
> Though, unsure if one will still be able to specify it in #pragma GCC target
> or target attribute...
Yeah, I think the lack of Alias(-msse4.1,-msse4.2) aka alias to
enable/disable two options is what makes it difficult. I'll post
a v2 momentarily fixing up ix86_valid_target_attribute_inner_p
instead. That at least seems quite a safe change (also for branches).
Richard.
