This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH] Add minimum insn match to crossjump
- From: Pat Haugen <pthaugen at us dot ibm dot com>
- To: gcc-patches at gcc dot gnu dot org
- Date: Tue, 6 Jul 2004 12:41:14 -0500
- Subject: Re: [PATCH] Add minimum insn match to crossjump
> The following patch imposes a minimum threshhold on the number of
> instructions to match before proceeding with the crossjump optimization
in
> cases where the block we're crossjumping from is not totally matched.
The
> thought here is to make sure there is some minimum number of instructions
> in the new block to make adding an extra branch along the one path worth
> while, it should also give the scheduler a little more to work with in
the
> new block. Runs of SPECInt and Skidmarks showed some minor improvement.
>
> I've bootstrapped/regtested on powerpc64-unknown-linux-gnu with no
> regressions.
>
> OK for mainline?
>
> -Pat
Here's the reworked patch after parameterizing the magic number for the
minimum instructions to match.
Bootstrapped/regtested on powerpc64-unknown-linux-gnu.
2004-07-06 Pat Haugen <pthaugen@us.ibm.com>
* params.def (PARAM_MIN_CROSSJUMP_INSNS): New.
* cfgcleanup.c (try_crossjump_to_edge): Add minimum insn match
threshhold.
* doc/invoke.texi (param): Document min-crossjump-insns.
Index: gcc/params.def
===================================================================
RCS file: /cvsroot/gcc/gcc/gcc/params.def,v
retrieving revision 1.41
diff -c -3 -p -r1.41 params.def
*** gcc/params.def 25 May 2004 12:54:54 -0000 1.41
--- gcc/params.def 6 Jul 2004 13:39:36 -0000
*************** DEFPARAM(PARAM_MAX_CROSSJUMP_EDGES,
*** 297,302 ****
--- 297,307 ----
"max-crossjump-edges",
"The maximum number of incoming edges to consider for crossjumping",
100)
+ /* The minimum number of matching instructions to consider for crossjumping. */
+ DEFPARAM(PARAM_MIN_CROSSJUMP_INSNS,
+ "min-crossjump-insns",
+ "The minimum number of matching instructions to consider for crossjumping",
+ 5)
/* The maximum length of path considered in cse. */
DEFPARAM(PARAM_MAX_CSE_PATH_LENGTH,
Index: gcc/cfgcleanup.c
===================================================================
RCS file: /cvsroot/gcc/gcc/gcc/cfgcleanup.c,v
retrieving revision 1.118
diff -c -3 -p -r1.118 cfgcleanup.c
*** gcc/cfgcleanup.c 16 Jun 2004 18:03:11 -0000 1.118
--- gcc/cfgcleanup.c 6 Jul 2004 13:39:36 -0000
*************** try_crossjump_to_edge (int mode, edge e1
*** 1528,1534 ****
/* ... and part the second. */
nmatch = flow_find_cross_jump (mode, src1, src2, &newpos1, &newpos2);
! if (!nmatch)
return false;
#ifndef CASE_DROPS_THROUGH
--- 1528,1540 ----
/* ... and part the second. */
nmatch = flow_find_cross_jump (mode, src1, src2, &newpos1, &newpos2);
!
! /* Don't proceed with the crossjump unless we found a sufficient number
! of matching instructions or the 'from' block was totally matched
! (such that its predecessors will hopefully be redirected and the
! block removed). */
! if ((nmatch < PARAM_VALUE (PARAM_MIN_CROSSJUMP_INSNS))
! && (newpos1 != BB_HEAD (src1)))
return false;
#ifndef CASE_DROPS_THROUGH
Index: gcc/doc/invoke.texi
===================================================================
RCS file: /cvsroot/gcc/gcc/gcc/doc/invoke.texi,v
retrieving revision 1.474
diff -c -3 -p -r1.474 invoke.texi
*** gcc/doc/invoke.texi 29 Jun 2004 01:53:04 -0000 1.474
--- gcc/doc/invoke.texi 6 Jul 2004 13:39:47 -0000
*************** the number of edges incoming to each blo
*** 5049,5054 ****
--- 5049,5060 ----
more aggressive optimization, making the compile time increase with
probably small improvement in executable size.
+ @item min-crossjump-insns
+ The minimum number of instructions which must be matched at the end
+ of two blocks before crossjumping will be performed on them. This
+ value is ignored in the case where all instructions in the block being
+ crossjumped from are matched. The default value is 5.
+
@item max-delay-slot-insn-search
The maximum number of instructions to consider when looking for an
instruction to fill a delay slot. If more than this arbitrary number of