| Summary: | Compiler does not take advantage of restrict | ||
|---|---|---|---|
| Product: | gcc | Reporter: | Mark Mitchell <mmitchel> |
| Component: | middle-end | Assignee: | Not yet assigned to anyone <unassigned> |
| Status: | RESOLVED DUPLICATE | ||
| Severity: | enhancement | CC: | acahalan, dnovillo, gcc-bugs, gcc, hoogerbrugge, ian, martin, mmitchel, pinskia, tim, tomash.brechko |
| Priority: | P3 | Keywords: | alias, missed-optimization |
| Version: | 4.3.0 | ||
| Target Milestone: | --- | ||
| Host: | i686-pc-linux-gnu | Target: | arm-none-linux-gnueabi |
| Build: | Known to work: | ||
| Known to fail: | Last reconfirmed: | 2007-09-01 11:15:21 | |
| Bug Depends on: | 14187, 16913, 29237 | ||
| Bug Blocks: | |||
Nearly a dup. Search for "restrict" and missed-optimization, alias for more of them. Well for the RTL level, with this code, one problem is that iv-opts messes up: ivtmp.34 = (long unsigned int) &(*(b + 64))[j]; There is a bug for the iv-opts issue too (though I have not had time to look into fixing it the correct way, iv-opts needs a real rewrite). And also PRE does not remap the pretmp for restrict (this already filed). Subject: Bug 33272 Author: jsm28 Date: Sun Sep 2 00:11:20 2007 New Revision: 128018 URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=128018 Log: PR middle-end/33272 * c-decl.c (grokdeclarator): Apply qualifiers to type of parameter decayed from array. testsuite: * gcc.dg/c99-arraydecl-3.c: New test. Added: trunk/gcc/testsuite/gcc.dg/c99-arraydecl-3.c Modified: trunk/gcc/ChangeLog trunk/gcc/c-decl.c trunk/gcc/testsuite/ChangeLog Subject: Bug 33272 Author: pinskia Date: Mon Nov 5 00:08:04 2007 New Revision: 129888 URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=129888 Log: Index: ChangeLog =================================================================== --- ChangeLog (revision 129887) +++ ChangeLog (working copy) @@ -6447,6 +6447,7 @@ 2007-09-02 Joseph Myers <joseph@codesourcery.com> + PR c/29237 PR middle-end/33272 * c-decl.c (grokdeclarator): Apply qualifiers to type of parameter decayed from array. Index: testsuite/ChangeLog =================================================================== --- testsuite/ChangeLog (revision 129887) +++ testsuite/ChangeLog (working copy) @@ -3041,6 +3041,7 @@ 2007-09-02 Joseph Myers <joseph@codesourcery.com> + PR C/29237 PR middle-end/33272 * gcc.dg/c99-arraydecl-3.c: New test. Modified: trunk/gcc/ChangeLog trunk/gcc/testsuite/ChangeLog |
Compiling this test case with -O2 -std=c99: void f(int *restrict r, int a[restrict 16][16], int b[restrict 16][16], int i, int j) { int x; *r = 0; for (x = 1; x < 16; ++x) *r = *r + a[i][x] * b[x][j]; } does not sink the store to "*r" outside of the loop; we end up writing to "*r" on every loop iteration. We do sink the store with "-fargument-noalias" so that is somehow stronger in this case than using "restrict". Despite other hard problems with "restrict", it seems like the special case of "all pointer arguments use restrict" should be equivalent to "-fargument-noalias is in effect".