This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC 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]

[PATCH] Speedup DOM and tree-cfg a lot (partial of PR 15524)


I noticed this when I was looking into why GCC was failing to bootstrap
on OpenBSD 3.1 because of memory hog, and I noticed that something was
changed in the middle-end to speed up cases with large amount of switch
statements.  I looked into that PR and decided to test the testcase and
I noticed it was slow so I decided to file a bug and see if there was
any easy thing to fix.  I noticed that find_case_label_for_value was
calling simple_cst_equal on CASE_LOW which is not needed, only a
call to tree_int_cst_compare is needed as CASE_LOW is only ever
an INTEGER_CST as defined by c-tree.texi.

This patch changes the call to simple_cst_equal to be tree_int_cst_compare
to get a speedup (I did not test how much though), it should also speed
up bootstrap of gcc as there are some large switch statements in
insn-attrtab.c.


OK? Bootstrapped on powerpc-apple-darwin with no regressions.


Thanks, Andrew Pinski


ChangeLog:


	* tree-cfg.c (find_case_label_for_value): Replace call to
	simple_cst_equal with tree_int_cst_compare.


Index: tree-cfg.c =================================================================== RCS file: /cvs/gcc/gcc/gcc/tree-cfg.c,v retrieving revision 2.2 diff -u -p -r2.2 tree-cfg.c --- tree-cfg.c 15 May 2004 23:07:51 -0000 2.2 +++ tree-cfg.c 18 May 2004 22:25:34 -0000 @@ -1958,7 +1958,7 @@ find_case_label_for_value (tree switch_e else if (CASE_HIGH (t) == NULL) { /* A `normal' case label. */ - if (simple_cst_equal (CASE_LOW (t), val) == 1) + if (tree_int_cst_compare (CASE_LOW (t), val) == 0) return t; } else




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