This is the mail archive of the mailing list for the glibc project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Use .S sources for x86/x86_64 expl

On 05/05/2012 06:22 PM, Joseph S. Myers wrote:
The x86/x86_64 implementation of expl uses a long inline asm inside a
C source file.

I don't think this is generally a good practice, it may offer some
marginal convenience for accessing operands where the precise syntax
varies between PIC and non-PIC, or x86 and x86_64, but it makes the
code less flexible and harder to modify than either plain C code or a
pure .S file, and we have macros to simplify writing such things as
PIC accesses in .S files.  When you want to add additional constant
operands, or checks directly on the binary representation of the
floating-point value, as part of fixing bugs, requiring everything to
be specified as asm operands rather than just writing the desired
instruction sequences directly rather gets in the way.

Thus, I propose this patch which moves expl to using separate .S files
for x86 and x86_64, the usual approach for the bulk of the other
x86/x86_64 long double functions with assembly implementations, which
works fine for them.  Tested x86 and x86_64.

The idea was to be able to share the file between i386 and x86-64. When I ported glibc to x86-64, I added those files where it was easy to do such a mix. Duplication in general is bad.

If this mix now makes things too complicated, then I'm fine with reversing it, and thus the patch is fine.

 Andreas Jaeger aj@{,} Twitter/Identica: jaegerandi
  SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany
   GF: Jeff Hawn,Jennifer Guild,Felix Imendörffer,HRB16746 (AG Nürnberg)
    GPG fingerprint = 93A3 365E CE47 B889 DF7F  FED1 389A 563C C272 A126

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]