| Message ID | alpine.LNX.2.00.1011242015550.6883@monoid.intra.ispras.ru |
|---|---|
| State | New |
| Headers | show |
On 11/24/2010 12:28 PM, Alexander Monakov wrote: > Hi, > > Selective scheduler uses NOPs as temporary placeholders for recently removed > insns. It does not expect them to be present as "normal" insns, but that > happens in the testcase with -O0 -fno-dce. > > Fortunately, we can just pick an unrecognizable pattern for a local NOP. The > patch implements that. Bootstrapped and regtested on x86_64-linux, > bootstrapped on ia64 with sel-sched at -O2 and regtest there is in progress. > > 2010-11-24 Alexander Monakov<amonakov@ispras.ru> > > PR rtl-optimization/46585 > * sel-sched-ir.c (return_regset_to_pool): Verify that RS is not NULL. > (vinsn_init): Skip DF initialization for local NOPs. > (vinsn_delete): Don't try to free regsets for local NOPs. > (setup_nop_and_exit_insns): Change definition of nop_pattern. > > testsuite: > * gcc.dg/pr46585.c: New. > Although the usage of const as a pattern is not documented in GCC manual, it is OK. I definitely remember that const as a pattern is already used in the compiler (only don't remember where :) Thanks for the patch.
On Nov 24, 2010, at 2:15 PM, Vladimir Makarov wrote: >> > Although the usage of const as a pattern is not documented in GCC manual, it is OK. I definitely remember that const as a pattern is already used in the compiler (only don't remember where :) The "nop" pattern description, section 16.9 ("Standard pattern names for generation") mentions it. paul
diff --git a/gcc/sel-sched-ir.c b/gcc/sel-sched-ir.c index 71c02c4..e6e8d44 100644 --- a/gcc/sel-sched-ir.c +++ b/gcc/sel-sched-ir.c @@ -942,6 +942,7 @@ get_clear_regset_from_pool (void) void return_regset_to_pool (regset rs) { + gcc_assert (rs); regset_pool.diff--; if (regset_pool.n == regset_pool.s) @@ -1175,6 +1176,9 @@ vinsn_init (vinsn_t vi, insn_t insn, bool force_unique_p) VINSN_COUNT (vi) = 0; vi->cost = -1; + if (INSN_NOP_P (insn)) + return; + if (DF_INSN_UID_SAFE_GET (INSN_UID (insn)) != NULL) init_id_from_df (VINSN_ID (vi), insn, force_unique_p); else @@ -1256,9 +1260,12 @@ vinsn_delete (vinsn_t vi) { gcc_assert (VINSN_COUNT (vi) == 0); - return_regset_to_pool (VINSN_REG_SETS (vi)); - return_regset_to_pool (VINSN_REG_USES (vi)); - return_regset_to_pool (VINSN_REG_CLOBBERS (vi)); + if (!INSN_NOP_P (VINSN_INSN_RTX (vi))) + { + return_regset_to_pool (VINSN_REG_SETS (vi)); + return_regset_to_pool (VINSN_REG_USES (vi)); + return_regset_to_pool (VINSN_REG_CLOBBERS (vi)); + } free (vi); } @@ -5606,7 +5614,7 @@ setup_nop_and_exit_insns (void) gcc_assert (nop_pattern == NULL_RTX && exit_insn == NULL_RTX); - nop_pattern = gen_nop (); + nop_pattern = constm1_rtx; start_sequence (); emit_insn (nop_pattern); diff --git a/gcc/testsuite/gcc.dg/pr46585.c b/gcc/testsuite/gcc.dg/pr46585.c index e69de29..32befdf 100644 --- a/gcc/testsuite/gcc.dg/pr46585.c +++ b/gcc/testsuite/gcc.dg/pr46585.c @@ -0,0 +1,11 @@ +/* { dg-do compile { target powerpc*-*-* ia64-*-* i?86-*-* x86_64-*-* } } */ +/* { dg-options "-fno-dce -fschedule-insns -fselective-scheduling" } */ +void +foo (void) +{ + switch (0) + { + default: + break; + } +}
Hi, Selective scheduler uses NOPs as temporary placeholders for recently removed insns. It does not expect them to be present as "normal" insns, but that happens in the testcase with -O0 -fno-dce. Fortunately, we can just pick an unrecognizable pattern for a local NOP. The patch implements that. Bootstrapped and regtested on x86_64-linux, bootstrapped on ia64 with sel-sched at -O2 and regtest there is in progress. 2010-11-24 Alexander Monakov <amonakov@ispras.ru> PR rtl-optimization/46585 * sel-sched-ir.c (return_regset_to_pool): Verify that RS is not NULL. (vinsn_init): Skip DF initialization for local NOPs. (vinsn_delete): Don't try to free regsets for local NOPs. (setup_nop_and_exit_insns): Change definition of nop_pattern. testsuite: * gcc.dg/pr46585.c: New.