Hi, This is my proposal for ulimately getting rid of the nasty store_fixed_bit_field recursion.
IMHO, the root of the recursion trouble is here:
@@ -1007,12 +1013,8 @@ store_fixed_bit_field (rtx op0, unsigned
if (MEM_P (op0))
{
mode = GET_MODE (op0);
if (GET_MODE_BITSIZE (mode) == 0
|| GET_MODE_BITSIZE (mode)> GET_MODE_BITSIZE (word_mode))
mode = word_mode;
mode = get_best_mode (bitsize, bitnum, bitregion_start, bitregion_end,
MEM_ALIGN (op0), mode, MEM_VOLATILE_P (op0));
But, because now we always have bitregion_start and bitregion_end to limit
the access size, it is no longer necessary to restrict the largest mode, that
get_best_mode may return.
This patch is very similar to the previous patch, which split up the
extract_fixed_bit_field,
This time, I just split up store_fixed_bit_field and use
store_fixed_bit_field_1 to force the
strict-volatile-bitfield mode it necessary, and let get_best_mode find a mode,
that is
can be used to access the field, which is no longer impacted by the memory
context's selected
mode in this case.
I tried this patch with an ARM-Cross compiler and a large eCos application, to
see if anything
changes in the generated code with this patch, but 2 MB of code stays binary
the same,
that's a good sign.
I added the new Ada test case, and the test case from PR59134, which does no
longer re-produce
after my previous C++ memory model patch, but this "fix" was more or less by
chance.
Boot-Strap on X86_64-pc-linux-gnu (languages=all,ada,go) and regression-tests
still running.
Ok for trunk (when the tests succeed)?
Thanks
Bernd. 2013-12-03 Bernd Edlinger <bernd.edlin...@hotmail.de>
PR middle-end/59134
* expmed.c (store_fixed_bit_field_1): New function.
(store_bit_field): Use narrow_bit_field_mem and
store_fixed_bit_field_1 for -fstrict-volatile-bitfields.
(extract_bit_field): Likewise. Use narrow_bit_field_mem and
extract_fixed_bit_field_1 to do the extraction.
(store_fixed_bit_field): Simplify mode selection algorithm.
Call store_fixed_bit_field_1 to do the real work.
(extract_fixed_bit_field_1): New function.
testsuite:
2013-12-03 Bernd Edlinger <bernd.edlin...@hotmail.de>
PR middle-end/59134
* gcc.c-torture/compile/pr59134.c: New test.
* gnat.dg/misaligned_volatile.adb: New test.
patch-bitfields-update-2.diff
Description: Binary data
