On 27/04/12 16:17, Richard Earnshaw wrote:
This is OK.
Thanks, now committed.
It would be good to merge all the target32 movdi variants into one
pattern and then use alternative enabling to deal with the different
valid alternatives.
Yes, I'll take a look.
Andrew
On 16/04/12 13:41, Richard Earnshaw wrote:
P.S. This patch can't actually be committed until my NEON DImode
immediate constants patch is approved and committed. (Without that the
load #0 needs a constant pool, and loading constants this late has a bug
at -O0.)
neon-neg64.patch
2012-04-12
Ping.
On 10/04/12 14:00, Andrew Stubbs wrote:
Ping.
On 30/03/12 12:15, Andrew Stubbs wrote:
On 28/02/12 16:20, Andrew Stubbs wrote:
Hi all,
This patch implements 64-bit immediate constant loads in NEON.
The current state is that you can load const_vector, but not const_int.
This is clearly
On 30/03/12 12:15, Andrew Stubbs wrote:
On 28/02/12 16:20, Andrew Stubbs wrote:
Hi all,
This patch implements 64-bit immediate constant loads in NEON.
The current state is that you can load const_vector, but not const_int.
This is clearly not ideal. The result is a constant pool entry when
On 14/04/12 14:11, Andrew Stubbs wrote:
And now with the patch. :(
On 14/04/12 13:48, Andrew Stubbs wrote:
On 12/04/12 16:48, Richard Earnshaw wrote:
If negation in Neon needs a scratch register, it seems to me to be
somewhat odd that we're disparaging the ARM version.
Also, wouldn't it
On 12/04/12 16:48, Richard Earnshaw wrote:
If negation in Neon needs a scratch register, it seems to me to be
somewhat odd that we're disparaging the ARM version.
Also, wouldn't it be sensible to support a variant that was
early-clobber on operand 0, but loaded immediate zero into that value
And now with the patch. :(
On 14/04/12 13:48, Andrew Stubbs wrote:
On 12/04/12 16:48, Richard Earnshaw wrote:
If negation in Neon needs a scratch register, it seems to me to be
somewhat odd that we're disparaging the ARM version.
Also, wouldn't it be sensible to support a variant that was
Ping.
On 26/03/12 11:14, Andrew Stubbs wrote:
On 28/02/12 17:45, Andrew Stubbs wrote:
Hi all,
This patch adds a DImode negate pattern for NEON.
Unfortunately, the NEON vneg instruction only supports vectors, not
singletons, so there's no direct way to do it in DImode, and the
compiler ends
On 26/03/12 11:14, Andrew Stubbs wrote:
On 28/02/12 17:45, Andrew Stubbs wrote:
Hi all,
This patch adds a DImode negate pattern for NEON.
Unfortunately, the NEON vneg instruction only supports vectors, not
singletons, so there's no direct way to do it in DImode, and the
compiler ends up
On 12/04/12 16:48, Richard Earnshaw wrote:
If negation in Neon needs a scratch register, it seems to me to be
somewhat odd that we're disparaging the ARM version.
Good point, I'm really meaning to provide missing functionality in NEON
to prevent unnecessary moves, rather than cause them, so I
On 12/04/12 17:16, Andrew Stubbs wrote:
On 12/04/12 16:48, Richard Earnshaw wrote:
If negation in Neon needs a scratch register, it seems to me to be
somewhat odd that we're disparaging the ARM version.
Good point, I'm really meaning to provide missing functionality in NEON
to prevent
Ping.
On 30/03/12 12:15, Andrew Stubbs wrote:
On 28/02/12 16:20, Andrew Stubbs wrote:
Hi all,
This patch implements 64-bit immediate constant loads in NEON.
The current state is that you can load const_vector, but not const_int.
This is clearly not ideal. The result is a constant pool entry
On 28/02/12 16:20, Andrew Stubbs wrote:
Hi all,
This patch implements 64-bit immediate constant loads in NEON.
The current state is that you can load const_vector, but not const_int.
This is clearly not ideal. The result is a constant pool entry when it's
not necessary.
The patch disables the
On 08/03/12 18:03, Richard Henderson wrote:
On 03/08/12 08:19, Andrew Stubbs wrote:
+ (set_attr arch nota8,*,*,onlya8)
+ (set_attr_alternative insn_enabled
+ [(if_then_else (match_test TARGET_NEON)
+ (const_string yes) (const_string no))
+(const_string
On 03/27/12 13:23, Andrew Stubbs wrote:
gcc/
* config/arm/arm.md (arch): Add neon_onlya8 and neon_nota8.
(arch_enabled): Handle new arch types.
(one_cmpldi2): Add NEON support.
Looks good to me.
r~
On 28/02/12 17:45, Andrew Stubbs wrote:
Hi all,
This patch adds a DImode negate pattern for NEON.
Unfortunately, the NEON vneg instruction only supports vectors, not
singletons, so there's no direct way to do it in DImode, and the
compiler ends up moving the value back to core registers,
On 01/03/12 17:07, Richard Earnshaw wrote:
The RTL part of one_cmpldi2_internal and one_cmpldi2_neon are the same.
Given that we now have controls to determine when an alternative is
enabled it's generally better to have just one pattern here and turn on
the alternatives that are suitable
On 03/08/12 08:19, Andrew Stubbs wrote:
+ (set_attr arch nota8,*,*,onlya8)
+ (set_attr_alternative insn_enabled
+ [(if_then_else (match_test TARGET_NEON)
+(const_string yes) (const_string no))
+ (const_string yes)
+ (const_string yes)
+
On Wed 29 Feb 2012 18:00:19 GMT, Richard Earnshaw wrote:
Why can't we have a single insn that deals with the have-neon and
dont-have-neon cases?
Sorry, I'm not sure I follow?
There's one insn for the have-neon case, and one for the
don't-have-neon. The expander is necessary to prevent
On 01/03/12 12:57, Andrew Stubbs wrote:
On Wed 29 Feb 2012 18:00:19 GMT, Richard Earnshaw wrote:
Why can't we have a single insn that deals with the have-neon and
dont-have-neon cases?
Sorry, I'm not sure I follow?
There's one insn for the have-neon case, and one for the
On 28/02/12 17:45, Andrew Stubbs wrote:
This patch adds a DImode negate pattern for NEON.
Oops, that patch completely failed to allow for the fallback to
instructions that work in core registers. It also forgot to mention that
the CC register was clobbered.
This patch is the same except
On 29/02/12 14:48, Andrew Stubbs wrote:
Hi all,
This patch adds support for the DImode not operation in NEON.
Currently the compiler must move the value to core registers, invert it
there, and move it back again. This is bonkers because the VMVN
instruction will do the job perfectly.
22 matches
Mail list logo