This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Fully Pruned vs. Semi Pruned, some actual numbers
- From: Daniel Berlin <dberlin at dberlin dot org>
- To: GCC Patches <gcc-patches at gcc dot gnu dot org>
- Cc: Chris Lattner <sabre at nondot dot org>,Jeff Law <law at redhat dot com>
- Date: Tue, 20 Jan 2004 16:58:49 -0500
- Subject: Fully Pruned vs. Semi Pruned, some actual numbers
Contrary to what the papers for SSAPRE also state, semi-pruned doesn't
buy us anything.
Before
[dberlin@dberlin gcc]$ pcregrep -h "Reloads:[^0]" *.pre|sort|uniq -c
1 fprintf (tree_dump_file.3317_1595, "Reloads:%d\n",
T.3345_1594);
543 Reloads:1
10 Reloads:10
1 Reloads:11
2 Reloads:16
104 Reloads:2
17 Reloads:3
19 Reloads:4
20 Reloads:5
5 Reloads:6
3 Reloads:7
6 Reloads:8
1 Reloads:9
After
[dberlin@dberlin gcc]$ pcregrep -h "Reloads:[^0]" *.pre|sort|uniq -c
1 fprintf (tree_dump_file.3317_1595, "Reloads:%d\n",
T.3345_1594);
543 Reloads:1
10 Reloads:10
1 Reloads:11
2 Reloads:16
104 Reloads:2
16 Reloads:3
20 Reloads:4
20 Reloads:5
5 Reloads:6
3 Reloads:7
6 Reloads:8
So we make one more reload total with full-pruned.
I'll note that cutely, because we have files with the same name in
different front ends, causing the stats files to be overwritten
depending on what frontend is being compiled, these end up being the
stats from a mix of the backend + different pieces of different
frontends.
But it's the same mix in both cases.
--Dan