Ping. There was discussion of larger fixes for this including a
type promotion pass but this patch seems small, safe, and in line
with other platforms (like arm32).
Steve Ellcey
sell...@cavium.com
On Thu, 2017-09-14 at 11:43 -0700, Steve Ellcey wrote:
> On Thu, 2017-09-14 at 11:53 -0600, Jeff
On Thu, 2017-09-14 at 11:53 -0600, Jeff Law wrote:
>
>
> And I think that's starting to zero in on the problem --
> WORD_REGISTER_OPERATIONS is zero on aarch64 as you don't get extension
> to word_mode for W form registers.
>
> I wonder if what needs to happen is somehow look to extend that
On 09/14/2017 10:33 AM, Steve Ellcey wrote:
> On Thu, 2017-09-14 at 09:03 -0600, Jeff Law wrote:
>> On 09/13/2017 03:46 PM, Steve Ellcey wrote:
>>>
>>> In arm32 rtl expansion, when reading the QI memory location, I see
>>> these instructions get generated:
>>>
>>> (insn 10 3 11 2 (set (reg:SI
On 09/14/2017 10:33 AM, Steve Ellcey wrote:
> On Thu, 2017-09-14 at 09:03 -0600, Jeff Law wrote:
>> On 09/13/2017 03:46 PM, Steve Ellcey wrote:
>>>
>>> In arm32 rtl expansion, when reading the QI memory location, I see
>>> these instructions get generated:
>>>
>>> (insn 10 3 11 2 (set (reg:SI
On Thu, 2017-09-14 at 09:03 -0600, Jeff Law wrote:
> On 09/13/2017 03:46 PM, Steve Ellcey wrote:
> >
> > In arm32 rtl expansion, when reading the QI memory location, I see
> > these instructions get generated:
> >
> > (insn 10 3 11 2 (set (reg:SI 119)
> > (zero_extend:SI (mem:QI
On 09/13/2017 03:46 PM, Steve Ellcey wrote:
> On Wed, 2017-09-13 at 14:46 -0500, Segher Boessenkool wrote:
>> On Wed, Sep 13, 2017 at 06:13:50PM +0100, Kyrill Tkachov wrote:
>>>
>>> We are usually hesitant to add explicit subreg matching in the MD pattern
>>> (though I don't remember if there's
On Wed, Sep 13, 2017 at 10:41:33PM +, Wilco Dijkstra wrote:
> Steve Ellcey wrote:
>
> > And in aarch64 rtl expansion I see:
> >
> > (insn 10 9 11 (set (reg:QI 81)
> > (mem:QI (reg/v/f:DI 80 [ string ]) [0 *string_9(D)+0 S1 A8]))
> > "pr77729.c":3 -1
> > (nil))
>
> Yes using
On Wed, 2017-09-13 at 22:39 +, Wilco Dijkstra wrote:
> Steve Ellcey wrote:
>
> > And in aarch64 rtl expansion I see:
> >
> > (insn 10 9 11 (set (reg:QI 81)
> > (mem:QI (reg/v/f:DI 80 [ string ]) [0 *string_9(D)+0 S1
> A8])) "pr77729.c":3 -1
> > (nil))
>
> Yes using QI/HI mode
Steve Ellcey wrote:
> And in aarch64 rtl expansion I see:
>
> (insn 10 9 11 (set (reg:QI 81)
> (mem:QI (reg/v/f:DI 80 [ string ]) [0 *string_9(D)+0 S1 A8]))
> "pr77729.c":3 -1
> (nil))
Yes using QI/HI mode anywhere in the RTL seems perverse and incorrect given
AArch64
doesn't
On Wed, 2017-09-13 at 14:46 -0500, Segher Boessenkool wrote:
> On Wed, Sep 13, 2017 at 06:13:50PM +0100, Kyrill Tkachov wrote:
> >
> > We are usually hesitant to add explicit subreg matching in the MD pattern
> > (though I don't remember if there's a hard rule against it).
> > In this case this
Hi!
On Wed, Sep 13, 2017 at 06:13:50PM +0100, Kyrill Tkachov wrote:
> +;; Specialized OR instruction for combiner. The AND is masking out bits
> +;; not needed in the OR (doing a zero_extend). The zero_extend is not
> +;; needed because we know from the subreg that the upper part of the reg
>
Hi Steve,
On 13/09/17 17:50, Steve Ellcey wrote:
This is a patch for PR target/77729 on aarch64. The code is doing an
unneeded zero extend ('uxtb' in the original report, 'and' in the ToT
sources).
The patch looks a bit odd, it is a specialized define_insn for the combine
pass. At some
This is a patch for PR target/77729 on aarch64. The code is doing an
unneeded zero extend ('uxtb' in the original report, 'and' in the ToT sources).
The patch looks a bit odd, it is a specialized define_insn for the combine
pass. At some point in combine (I never did find out where), the
13 matches
Mail list logo