Hello, Segher, On Jul 9, 2020, Segher Boessenkool <seg...@kernel.crashing.org> wrote:
>> The problem it addresses is that the current checking only tests for >> existence not for an incompatible/compatible setting. >> Currently both are reported as incompatible. https://gcc.gnu.org/pipermail/gcc-patches/2020-July/549760.html > Put that > ((rs6000_isa_flags & OPTION_MASK_MINIMAL_TOC) != 0) > check inside the block please, together with the CMODEL_SMALL check? We're trying to make sense of this suggestion of yours, and I confess I'm quite confused as to the intended effects of this piece of code, and similar ones in aix, freebsd64 and rtems rs6000 headers. Clearly, we wish to issue an error message if -mminimal-toc and -mcmodel other than small. But should we silently force the code model to small if -mno-minimal-toc is specified? (I believe that would be a consequence of the change you propose) Upon first reading your suggestion, I thought it was a think, and that you meant to move the rs6000_isa_flags_explicit test into the block, for that would: - set cmodel to small whenever -mminimal-toc is enabled, explicit or by default - error iff -mminimal-toc and non-small -mcmodel are explicitly given However, this could change behavior for the various headers that have an else block for the if that currently tests rs6000_isa_flags_explicit, which might be undesirable, and that's where things get confusing. It's not at all clear to me what's expected if e.g.: - -mminimal-toc is enabled by default (is it ever?) and a non-small code model is selected (explicitly or by default): error, silent change to small cmodel, silent disabling of minimal-toc, or something else? - -mno-minimal-toc is explicitly given along with some non-small code model: any reason to error out, force small cmode, force-enable minimal-toc, or something else? Could you please shed any light as to the intent, so that we can sort out the logic that will implement it? The logic in this code seems to be unchanged since it was first introduced by Alan Modra along with -mcmodel back in 2010, despite the many spelling changes as internal representation of flags and state and whatnot changed. I'm copying Alan as well, just in case he can recall or figure out what the intent was. AFAICT it first made linux64.h, and other headers got copies or variants thereof as the -mcmodel option got adopted by other systems. Thanks in advance, -- Alexandre Oliva, happy hacker https://FSFLA.org/blogs/lxo/ Free Software Activist GNU Toolchain Engineer
