[dataflow]: PATCH COMMITTED to fix dumps and a register allocation problem.

Kenneth Zadeck zadeck@naturalbridge.com
Sat Nov 4 19:12:00 GMT 2006

This patch fixes two problems:

First, the register usage information is modified by local_alloc before
global alloc sees it.  This patch keeps global from recomputing that
information from scratch when it redoes the data flow analysis since the
changes that are made by local are somewhat non standard.

The other issue fixed by this patch is to print the hard register names
in the dump file for all of the register sets. 

2006-11-04  Kenneth Zadeck <zadeck@naturalbridge.com>
    * df-scan.c (df_scan_start_dump): Changed print routine to print
    hard register names.
    * df-core.c (df_print_regset): New function.
    * global.c (global_alloc): Do not recompute register infomation
    when recomputing dataflow.
    * df.h (DF_RI_NO_UPDATE): New flag. 
    * df-problems.c (df_lr_top_dump, df_lr_bottom_dump,
    df_ur_top_dump, df_ur_bottom_dump, df_live_top_dump,
    df_live_bottom_dump, df_urec_top_dump, df_urec_bottom_dump):
    Changed to use df_print_regset.
    (df_ri_alloc, df_ri_compute): Changed to not update if
    DF_RI_NO_UPDATE flag.
    * cfgrtl.c (print_rtl_with_bb): Changed to print preds and
    successor blocks in dump file.

Bootstrapped and tested on x86-64-linux.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: debug1.diff
Type: text/x-patch
Size: 10021 bytes
Desc: not available
URL: <http://gcc.gnu.org/pipermail/gcc-patches/attachments/20061104/061e542b/attachment.bin>

More information about the Gcc-patches mailing list