Jeff Law <l...@redhat.com> writes:
> On 09/18/14 04:25, Richard Sandiford wrote:
>> This is the main patch for the bug. We should treat a register as invalid
>> for a mode change if simplify_subreg_regno cannot provide a new register
>> number for the result. We should treat a class as invalid for a mode change
>> if all registers in the class are invalid. This is an extension of the old
>> CANNOT_CHANGE_MODE_CLASS-based check (simplify_subreg_regno checks C_C_C_M).
>>
>> I forgot to say that the patch is a prerequisite to removing aarch64's
>> C_C_C_M. There are other prerequisites too, but removing C_C_C_M without
>> this patch caused regressions in the existing testsuite, which is why no
>> new tests are needed.
>>
>>
>> gcc/
>> * hard-reg-set.h: Include hash-table.h.
>> (target_hard_regs): Add a finalize method and a x_simplifiable_subregs
>> field.
>> * target-globals.c (target_globals::~target_globals): Handle
>> hard_regs->finalize.
>> * rtl.h (subreg_shape): New structure.
>> (shape_of_subreg): New function.
>> (simplifiable_subregs): Declare.
>> * reginfo.c (simplifiable_subreg): New structure.
>> (simplifiable_subregs_hasher): Likewise.
>> (simplifiable_subregs): New function.
>> (invalid_mode_changes): Delete.
>> (alid_mode_changes, valid_mode_changes_obstack): New variables.
>> (record_subregs_of_mode): Remove subregs_of_mode parameter.
>> Record valid mode changes in valid_mode_changes.
>> (find_subregs_of_mode): Remove subregs_of_mode parameter.
>> Update calls to record_subregs_of_mode.
>> (init_subregs_of_mode): Remove invalid_mode_changes and bitmap
>> handling. Initialize new variables. Update call to
>> find_subregs_of_mode.
>> (invalid_mode_change_p): Check new variables instead of
>> invalid_mode_changes.
>> (finish_subregs_of_mode): Finalize new variables instead of
>> invalid_mode_changes.
>> (target_hard_regs::finalize): New function.
>> * ira-costs.c (print_allocno_costs): Call invalid_mode_change_p
>> even when CLASS_CANNOT_CHANGE_MODE is undefined.
>>
>> Index: gcc/rtl.h
>> ===================================================================
>> --- gcc/rtl.h 2014-09-15 11:55:40.459855161 +0100
>> +++ gcc/rtl.h 2014-09-15 12:26:21.249077760 +0100
>> +/* Return the shape of a SUBREG rtx. */
>> +
>> +static inline subreg_shape
>> +shape_of_subreg (const_rtx x)
>> +{
>> + return subreg_shape (GET_MODE (SUBREG_REG (x)),
>> + SUBREG_BYTE (x), GET_MODE (x));
>> +}
>> +
> Is there some reason you don't have a constructor that accepts a
> const_rtx?
I was worried that by allowing implicit const_rtx->subreg_shape
conversions, it would be less obvious that the rtx has to have
code SUBREG. I.e. a checked conversion would be hidden in the
constructor rather than being explicit.
If with David's new rtx hierarchy we end up with an rtx_subreg
subclass then I agree we should have a constructor that takes
one of those.
Thanks,
Richard
