On 12/04/13 11:01, Richard Sandiford wrote:
Ping for this patch, which is the only one of the series that hasn't
been approved.
Thanks,
Richard
Richard Sandiford <rdsandif...@googlemail.com> writes:
I have a patch to upgrade most genrecog warnings into errors. This patch
fixes those for tilegx. There seemed to be two sources of warnings:
- the intrinsics often used matched pointer_operands in an addition,
so that the destination accepted constant pointers. I think the
direct translation would be pmode_register_operand, but since these
additions have a specific mode, I think a modeful register_operand
is more natural.
- some instructions used reg_or_0_operand as a destination.
Tested by building tilegx-elf with the warnings turned to errors, and by
comparing the before and after assembly output at -O2 for gcc.c-torture,
gcc.dg and g++.dg. OK to install?
Thanks,
Richard
gcc/
* config/tilegx/tilegx.md (insn_ld_add<bitsuffix>): Use
register_operand rather than pointer_operand. Add modes to the
operands.
(insn_ldna_add<bitsuffix>): Likewise.
(insn_ld<I124MODE:n><s>_add<I48MODE:bitsuffix>): Likewise.
(insn_ldnt_add<bitsuffix>): Likewise.
(insn_ldnt<I124MODE:n><s>_add<I48MODE:bitsuffix>): Likewise.
(insn_ld_add_L2<bitsuffix>): Likewise.
(insn_ldna_add_L2<bitsuffix>): Likewise.
(insn_ld<I124MODE:n><s>_add_L2<I48MODE:bitsuffix>): Likewise.
(insn_ldnt_add_L2<bitsuffix>): Likewise.
(insn_ldnt<I124MODE:n><s>_add_L2<I48MODE:bitsuffix>): Likewise.
(insn_ld_add_miss<bitsuffix>): Likewise.
(insn_ldna_add_miss<bitsuffix>): Likewise.
(insn_ld<I124MODE:n><s>_add_miss<I48MODE:bitsuffix>): Likewise.
(insn_ldnt_add_miss<bitsuffix>): Likewise.
(insn_ldnt<I124MODE:n><s>_add_miss<I48MODE:bitsuffix>): Likewise.
(insn_st_add<bitsuffix>): Likewise.
(insn_st<I124MODE:n>_add<I48MODE:bitsuffix>): Likewise.
(*insn_st<I124MODE:n>_add<I48MODE:bitsuffix>): Likewise.
(insn_stnt_add<bitsuffix>): Likewise.
(insn_stnt<I124MODE:n>_add<I48MODE:bitsuffix>): Likewise.
(*insn_stnt<I124MODE:n>_add<I48MODE:bitsuffix>): Likewise.
(vec_pack_<pack_optab>_v4hi): Use register_operand rather than
reg_or_0_operand for operand 0.
(insn_v2<pack_insn>): Likewise.
(vec_pack_hipart_v4hi): Likewise.
(insn_v2packh): Likewise.
(vec_pack_ssat_v2si): Likewise.
(insn_v4packsc): Likewise.
This looks pretty mechanical. Hopefully there wasn't a compelling
reason to use pointer_operand instead of either register_operand or
other alternatives.
Let's give Walt a bit more time to chime in just in case there was a
particular reason for the prior choice of pointer_operand. Perhaps
Monday morning. If you haven't heard from Walt by then, consider the
patch approved by me.
jeff
