On Dec 15, 2006, at 4:19 AM, Simon Marlow wrote:
Peter Tanski wrote:
As part of testing Yasm, I looked at the next step for the
Windows native port: building and testing GHC using Yasm as the
assembler, which may benefit from changing the option flags in
compiler/main/ DriverPipeline.hs and compiler/main/DynFlags.hs .
I would like to change the flags to:
(1) move the hard-coded option parameters for 'runPhase As ...'
in DriverPipeline.hs into separate option sets in DynFlags.hs --
that way you only have to look in one place to change the hard-
coded options (or, in this case, to code them conditionally under
a preprocessor define USE_YASM);
(2) include parameters for Yasm ("-ax86", "-pgas", "-fwin32"); and,
(3) include Yasm as a configuration option (--enable-yasm).
An alternative implementation would be to add a special command
line option, --yasm, to enable use of yasm, although this would
require Makefile changes both for building and testing, in the
same way as using --pgma, --opta ... to use Yasm as is. What do
you think?
For the native Windows port, we should have a new target, probably
i386-unknown-win32 or i386-unknown-windows. Then we can have an
option in config.mk to control whether YASM is used, $(UseYASM),
which would be on by default for windows_TARGET_OS. Later we might
want to make it a configure option, but for now I'd leave it as a
build.mk option.
Right on. Would it be good to consolidate the hard-coded build flags
into one module, i.e., wired-in flags in DriverPipeline.hs should be
brought together in DynFlags.hs?
To sum up:
* new target: i386-unknown-windows (Yasm covers Win64, Intel and AMD)
* windows_TARGET_OS (new cpp define)
Other things may need changing:
* add code to DriverPhases.hs to allow ".asm" suffix? (Not necessary)
* SysTools.lhs:328-329 (as-prog,as-args) are NOT (gcc_prog,gcc_args)
* (potential bugger): remove default use of mangler on -O and -fvia-
C on -fglasgow-exts; Yasm chokes on /ARGS, /NO_ARGS gas directives,
may add to Yasm with special -D macro[=value] options
Cheers,
Pete
_______________________________________________
Cvs-ghc mailing list
[EMAIL PROTECTED]
http://www.haskell.org/mailman/listinfo/cvs-ghc
