> I suppose I was putting too much store by the expmed.c code. How does
> this version look? Tested on x86_64-linux-gnu, powerpc64-linux-gnu
> and cris-elf.
>
> Richard
>
>
> gcc/
> PR middle-end/55438
> * expmed.c (simple_mem_bitfield_p): New function, extracted from
> store_bit_field_1 and extract_bit_field_1. Use GET_MODE_ALIGNMENT
> rather than bitsize when checking the alignment.
I'd mention explicitly the function:
(store_bit_field_1): Call simple_mem_bitfield_p.
(extract_bit_field_1): Likewise.
> * stor-layout.c (bit_field_mode_iterator::bit_field_mode_iterator):
> Don't limit ALIGN_. Assume that memory is mapped in chunks of at
> least word size, regardless of BIGGEST_ALIGNMENT.
> (bit_field_mode_iterator::get_mode): Use GET_MODE_ALIGNMENT rather
> than unit when checking the alignment.
> (get_best_mode): Use GET_MODE_ALIGNMENT.
OK, modulo:
> Index: gcc/expmed.c
> ===================================================================
> --- gcc/expmed.c 2012-11-20 19:49:07.000000000 +0000
> +++ gcc/expmed.c 2012-11-27 17:27:38.381139339 +0000
> @@ -416,6 +416,21 @@ lowpart_bit_field_p (unsigned HOST_WIDE_
> else
> return bitnum % BITS_PER_WORD == 0;
> }
> +
> +/* Return true if OP is a memory and if a bitfield of size BITSIZE at
> + bit number BITNUM can be treated as a simple value of mode MODE. */
> +
> +static bool
> +simple_mem_bitfield_p (rtx op0, unsigned HOST_WIDE_INT bitsize,
> + unsigned HOST_WIDE_INT bitnum, enum machine_mode mode)
> +{
> + return (MEM_P (op0)
> + && bitnum % BITS_PER_UNIT == 0
> + && bitsize == GET_MODE_BITSIZE (mode)
> + && (!SLOW_UNALIGNED_ACCESS (mode, MEM_ALIGN (op0))
> + || (bitnum % GET_MODE_ALIGNMENT (mode) == 0
> + && MEM_ALIGN (op0) % GET_MODE_ALIGNMENT (mode) == 0)));
> +}
I think that the most idiomatic form is
MEM_ALIGN (op0) >= GET_MODE_ALIGNMENT (mode)
when it comes to alignments.
--
Eric Botcazou
