Hi Eric,
On 02/11/16 10:47, Eric Botcazou wrote:
This converts the preprocessor check for WORD_REGISTER_OPERATIONS into a
runtime one in rtlanal.c.
Since this one was in combination with an "#if defined" and used to guard an
if-statement I'd appreciate it if someone gave it a double-check that I
dind't screw up the intended behaviour.
Unfortunately I think you did, as the old version was:
#if WORD_REGISTER_OPERATIONS && defined (LOAD_EXTEND_OP)
/* If this is a typical RISC machine, we only have to worry
about the way loads are extended. */
if ((LOAD_EXTEND_OP (inner_mode) == SIGN_EXTEND
? val_signbit_known_set_p (inner_mode, nonzero)
: LOAD_EXTEND_OP (inner_mode) != ZERO_EXTEND)
|| !MEM_P (SUBREG_REG (x)))
#endif
and the new version is:
#ifdef LOAD_EXTEND_OP
/* If this is a typical RISC machine, we only have to worry
about the way loads are extended. */
if (WORD_REGISTER_OPERATIONS
&& ((LOAD_EXTEND_OP (inner_mode) == SIGN_EXTEND
? val_signbit_known_set_p (inner_mode, nonzero)
: LOAD_EXTEND_OP (inner_mode) != ZERO_EXTEND)
|| !MEM_P (SUBREG_REG (x))))
#endif
So if WORD_REGISTER_OPERATIONS is zero and LOAD_EXTEND_OP is defined, for
example on PowerPC, the block guarded by the condition is always executed in
the former case but never in the latter case.
I think you're right. I suppose the new condition should be:
#ifdef LOAD_EXTEND_OP
/* If this is a typical RISC machine, we only have to worry
about the way loads are extended. */
if (!WORD_REGISTER_OPERATIONS
|| ((LOAD_EXTEND_OP (inner_mode) == SIGN_EXTEND
? val_signbit_known_set_p (inner_mode, nonzero)
: LOAD_EXTEND_OP (inner_mode) != ZERO_EXTEND)
|| !MEM_P (SUBREG_REG (x))))
#endif
Would you prefer me to make this change or just revert the patch?
Thanks,
Kyrill
