Re: [RFA] CFI improvements

[Andreas Jaeger <aj at suse dot de>]

Michal Ludvig <mludvig@suse.cz> writes:

> Hi,
> the attached patch adds some new features to the CFI engine:
> - register names can start with '%' as requested by GCC guys.
> - CIEs are not duplicated anymore but instead reused if possible.
> - tc_cfi_init() is moved from main() to dot_cfi_startproc() -
> otherwise it didn't set correct values on amd64 because the output
> format wasn't yet known in main().
> - .cfi_startproc takes optional parameter 'simple' that disables
> emitting of architecture-dependent initial instructions.
> - testsuite is modified to pass with CIE optimalization.
>
> I hope I haven't introduce any new C89 incompatibility (sorry for that one).
>
> Comments? Approvals?
>
> Michal Ludvig

> Index: dw2gencfi.c
> ===================================================================
> RCS file: /cvs/src/src/gas/dw2gencfi.c,v
> retrieving revision 1.2
> diff -u -p -r1.2 dw2gencfi.c
> --- dw2gencfi.c	20 May 2003 11:35:45 -0000	1.2
> +++ dw2gencfi.c	20 May 2003 12:04:30 -0000
> @@ -23,24 +23,55 @@
>  #include "as.h"
>  #include "dw2gencfi.h"
>  
> +struct cie_entry
> +{
> +  unsigned long offset;
> +  size_t size;
> +  void *data;
> +  struct cie_entry *next;
> +};
> +
> +struct cfi_data
> +{
> +  enum cfi_insn insn;
> +  long param[2];
> +  struct cfi_data *next;
> +};
> +
> +struct cfi_info
> +{
> +  addressT start_address;
> +  addressT end_address;
> +  addressT last_address;
> +  const char *labelname;
> +  struct cfi_data *data;
> +  struct cfi_info *next;
> +};
> +
> +/* Current open CFI entry.  */
> +static struct cfi_info *cfi_info;
> +
> +/* List of CIEs so that they could be reused.  */
> +static struct cie_entry *cie_root;
> +
>  /* Current target config.  */
>  static struct cfi_config current_config;
>  
>  /* This is the main entry point to the CFI machinery.  */
>  static void dot_cfi (int arg);
>  
> -const pseudo_typeS cfi_pseudo_table[] =
> -  {
> -    { "cfi_verbose", dot_cfi, CFI_verbose },
> -    { "cfi_startproc", dot_cfi, CFI_startproc },
> -    { "cfi_endproc", dot_cfi, CFI_endproc },
> -    { "cfi_def_cfa", dot_cfi, CFA_def_cfa },
> -    { "cfi_def_cfa_register", dot_cfi, CFA_def_cfa_register },
> -    { "cfi_def_cfa_offset", dot_cfi, CFA_def_cfa_offset },
> -    { "cfi_adjust_cfa_offset", dot_cfi, CFI_adjust_cfa_offset },
> -    { "cfi_offset", dot_cfi, CFA_offset },
> -    { NULL, NULL, 0 }
> -  };
> +const pseudo_typeS cfi_pseudo_table[] = {
> +  {"cfi_verbose", dot_cfi, CFI_verbose},
> +  {"cfi_startproc", dot_cfi, CFI_startproc},
> +  {"cfi_endproc", dot_cfi, CFI_endproc},
> +  {"cfi_def_cfa", dot_cfi, CFA_def_cfa},
> +  {"cfi_def_cfa_register", dot_cfi, CFA_def_cfa_register},
> +  {"cfi_def_cfa_offset", dot_cfi, CFA_def_cfa_offset},
> +  {"cfi_adjust_cfa_offset", dot_cfi, CFI_adjust_cfa_offset},
> +  {"cfi_offset", dot_cfi, CFA_offset},
> +  {"cfi_register", dot_cfi, CFA_register},
> +  {NULL, NULL, 0}
> +};

These look like whitespace changes.  Alan has done a number of
formatting changes, can you incorporate them cleanly into your patch
and then resend the patch?  Reverting Alan's formatting changes is not
the right way IMO.

> [...]
> +  char *simple = "simple";

const char?

Andreas
-- 
 Andreas Jaeger
  SuSE Labs aj@suse.de
   private aj@arthur.inka.de
    http://www.suse.de/~aj