Cleaning up expand optabs code

[Richard Henderson]

On 03/22/2011 08:08 AM, Richard Sandiford wrote:
>>> +  for (i = 0; i + 1 < nops; i++)
>>> +    if (ops[i].commutative < MAX_EXPAND_OPERANDS
>>> +       && swap_commutative_operands_with_target (ops[ops[i].commutative].value,
>>> +                                                 ops[i].value,
>>> +                                                 ops[i + 1].value))
>>
>> with the assumption of i & i+1 being related, to be a pretty strong
>> assumption.
> 
> OK.  I saw this as the expand equivalent of the "+" constraint, which
> makes the same assumption.  I don't think we really support gaps between
> commutative operands in any meaningful way.

You mean "%" I assume.  Yes, it does concern operands N and N+1.  But it's
documented that way, whereas you have .commutative and documented that as
the the operand number that gets swapped.  Which begs the question of what
is that third operand above, and how it would get specified.

Not that it matters now...

> 	* optabs.h (emit_unop_insn, maybe_emit_unop_insn): Change insn code
> 	parameter from "int" to "enum insn_code".
> 	(expand_operand_type): New enum.
> 	(expand_operand): New structure.
> 	(create_expand_operand): New function.
> 	(create_fixed_operand, create_output_operand): Likewise
> 	(create_input_operand, create_convert_operand_to): Likewise.
> 	(create_convert_operand_from, create_address_operand): Likewise.
> 	(create_integer_operand): Likewise.
> 	(create_convert_operand_from_type, maybe_legitimize_operands): Declare.
> 	(maybe_gen_insn, maybe_expand_insn, maybe_expand_jump_insn): Likewise.
> 	(expand_insn, expand_jump_insn): Likewise.
> 	* builtins.c (expand_builtin_prefetch): Use the new interfaces.
> 	(expand_builtin_interclass_mathfn, expand_builtin_strlen): Likewise.
> 	(expand_movstr, expand_builtin___clear_cache): Likewise.
> 	(expand_builtin_lock_release): Likewise.
> 	* explow.c (allocate_dynamic_stack_space): Likewise.
> 	(probe_stack_range): Likewise.  Allow check_stack to FAIL,
> 	and use the default handling in that case.
> 	* expmed.c (check_predicate_volatile_ok): Delete.
> 	(store_bit_field_1, extract_bit_field_1): Use the new interfaces.
> 	(emit_cstore): Likewise.
> 	* expr.c (emit_block_move_via_movmem): Likewise.
> 	(set_storage_via_setmem, expand_assignment): Likewise.
> 	(emit_storent_insn, try_casesi): Likewise.
> 	(emit_single_push_insn): Likewise.  Allow the expansion to fail.
> 	* optabs.c (expand_widen_pattern_expr, expand_ternary_op): Likewise.
> 	(expand_vec_shift_expr, expand_binop_directly): Likewise.
> 	(expand_twoval_unop, expand_twoval_binop): Likewise.
> 	(expand_unop_direct, emit_indirect_jump): Likewise.
> 	(emit_conditional_move, vector_compare_rtx): Likewise.
> 	(expand_vec_cond_expr, expand_val_compare_and_swap_1): Likewise.
> 	(expand_sync_operation, expand_sync_fetch_operation): Likewise.
> 	(expand_sync_lock_test_and_set): Likewise.
> 	(maybe_emit_unop_insn): Likewise.  Change icode to an insn_code.
> 	(emit_unop_insn): Likewise.
> 	(expand_copysign_absneg): Change icode to an insn_code.
> 	(create_convert_operand_from_type): New function.
> 	(maybe_legitimize_operand, maybe_legitimize_operands): Likewise.
> 	(maybe_gen_insn, maybe_expand_insn, maybe_expand_jump_insn): Likewise.
> 	(expand_insn, expand_jump_insn): Likewise.
> 	* config/i386/i386.md (setmem<mode>): Use nonmemory_operand rather
> 	than const_int_operand for operand 2.

Ok.  Watch out for other target problems this week.


r~