| Message ID | alpine.DEB.2.02.1310260837400.32346@stedding.saclay.inria.fr |
|---|---|
| State | New |
| Headers | show |
On 10/26/13 01:15, Marc Glisse wrote: > Hello, > > this patch teaches gcc that free kills the memory its argument points > to. The equality test is probably too strict, I guess we can loosen it > later (unless you have suggestions?). > > Note that the corresponding code for BUILT_IN_MEMCPY and others seems > suspicious to me, it looks like it is testing for equality between a > pointer and a mem_ref, which is unlikely to happen. > > Bootstrap+testsuite on x86_64-unknown-linux-gnu. > > 2013-10-27 Marc Glisse <marc.glisse@inria.fr> > > PR tree-optimization/19831 > gcc/ > * tree-ssa-alias.c (stmt_kills_ref_p_1): Handle BUILT_IN_FREE. > > gcc/testsuite/ > * gcc.dg/tree-ssa/alias-25.c: New file. OK for the trunk. I agree the MEM_REF* and VA_END cases look strange and I think they're wrong as well. Did you happen to try fixing them and running the testsuite to see if anything fails? ISTM your testcase could be tweaked to test conclusively if they're doing the right thing -- and regardless of what's found that test can/should make its way into the testsuite. Jeff
Index: gcc/testsuite/gcc.dg/tree-ssa/alias-25.c =================================================================== --- gcc/testsuite/gcc.dg/tree-ssa/alias-25.c (revision 0) +++ gcc/testsuite/gcc.dg/tree-ssa/alias-25.c (working copy) @@ -0,0 +1,12 @@ +/* { dg-do compile } */ +/* { dg-options "-O1 -fdump-tree-optimized" } */ + +void f (long *p) { + *p = 42; + p[4] = 42; + __builtin_free (p); +} + +/* { dg-final { scan-tree-dump-not "= 42" "optimized" } } */ +/* { dg-final { cleanup-tree-dump "optimized" } } */ + Property changes on: gcc/testsuite/gcc.dg/tree-ssa/alias-25.c ___________________________________________________________________ Added: svn:keywords ## -0,0 +1 ## +Author Date Id Revision URL \ No newline at end of property Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Index: gcc/tree-ssa-alias.c =================================================================== --- gcc/tree-ssa-alias.c (revision 204088) +++ gcc/tree-ssa-alias.c (working copy) @@ -2053,20 +2053,30 @@ stmt_kills_ref_p_1 (gimple stmt, ao_ref } } if (is_gimple_call (stmt)) { tree callee = gimple_call_fndecl (stmt); if (callee != NULL_TREE && DECL_BUILT_IN_CLASS (callee) == BUILT_IN_NORMAL) switch (DECL_FUNCTION_CODE (callee)) { + case BUILT_IN_FREE: + { + tree ptr = gimple_call_arg (stmt, 0); + tree base = ao_ref_base (ref); + if (base && TREE_CODE (base) == MEM_REF + && TREE_OPERAND (base, 0) == ptr) + return true; + break; + } + case BUILT_IN_MEMCPY: case BUILT_IN_MEMPCPY: case BUILT_IN_MEMMOVE: case BUILT_IN_MEMSET: case BUILT_IN_MEMCPY_CHK: case BUILT_IN_MEMPCPY_CHK: case BUILT_IN_MEMMOVE_CHK: case BUILT_IN_MEMSET_CHK: { tree dest = gimple_call_arg (stmt, 0);