On 07/14/2011 11:40 AM, Richard Guenther wrote:
> (look how the vectorizer
> for example uses new target hooks instead of generating vectorized RTL
> and then using rtx_cost).
But wouldn't we then end up with just a bunch of special purpose tree_cost
functions
again?! Then we would again be doome
On Thu, Jul 14, 2011 at 11:40 AM, Richard Guenther
wrote:
> Generating RTL from GIMPLE passes just to be able to use rtx_cost is,
> well ... gross.
Indeed. And it is one of the major blockers for fully separating the
RTL, GIMPLE and target code off into separate modules. It would be
great to get
On Wed, Jul 13, 2011 at 09:58:08AM -0700, Richard Henderson wrote:
> Why the force_operand? You've got register inputs. Either the target
> is going to support the operation or it isn't.
I agree that it doesn't seem to be necessary. I've used force_operand
since ivopts (add_cost) is doing it wit
On Wed, Jul 13, 2011 at 11:49 PM, Steven Bosscher wrote:
> On Wed, Jul 13, 2011 at 4:34 PM, Richard Guenther
> wrote:
>> On Wed, Jul 13, 2011 at 3:13 PM, Andreas Krebbel
>> wrote:
>>> Hi,
>>>
>>> the widening_mul pass might increase the number of multiplications in
>>> the code by transforming
>
On 07/13/2011 06:58 PM, Richard Henderson wrote:
> Why the force_operand? You've got register inputs. Either the target
> is going to support the operation or it isn't.
> Seems to me you can check the availability of the operation in the
> optab and pass that gen_rtx_fmt_ee result to rtx_cost di
On 07/13/2011 04:34 PM, Richard Guenther wrote:
> On Wed, Jul 13, 2011 at 3:13 PM, Andreas Krebbel
> wrote:
>> Hi,
>>
>> the widening_mul pass might increase the number of multiplications in
>> the code by transforming
>>
>> a = b * c
>> d = a + 2
>> e = a + 3
>>
>> into:
>>
>> d = b * c + 2
>> e
On Wed, Jul 13, 2011 at 4:34 PM, Richard Guenther
wrote:
> On Wed, Jul 13, 2011 at 3:13 PM, Andreas Krebbel
> wrote:
>> Hi,
>>
>> the widening_mul pass might increase the number of multiplications in
>> the code by transforming
>>
>> a = b * c
>> d = a + 2
>> e = a + 3
>>
>> into:
>>
>> d = b * c
On 07/13/2011 06:13 AM, Andreas Krebbel wrote:
> + force_operand (gen_rtx_fmt_ee (code, mode,
> +gen_raw_REG (mode, LAST_VIRTUAL_REGISTER + 1),
> +gen_raw_REG (mode, LAST_VIRTUAL_REGISTER + 2)),
> + NULL_RTX);
> + break;
> + c
Richard Guenther wrote:
> On Wed, Jul 13, 2011 at 3:13 PM, Andreas Krebbel
> wrote:
>> Hi,
>>
>> the widening_mul pass might increase the number of multiplications in
>> the code by transforming
>>
>> a = b * c
>> d = a + 2
>> e = a + 3
>>
>> into:
>>
>> d = b * c + 2
>> e = b * c + 3
>>
>> under
On Wed, Jul 13, 2011 at 3:13 PM, Andreas Krebbel
wrote:
> Hi,
>
> the widening_mul pass might increase the number of multiplications in
> the code by transforming
>
> a = b * c
> d = a + 2
> e = a + 3
>
> into:
>
> d = b * c + 2
> e = b * c + 3
>
> under the assumption that an FMA instruction is n
Hi,
the widening_mul pass might increase the number of multiplications in
the code by transforming
a = b * c
d = a + 2
e = a + 3
into:
d = b * c + 2
e = b * c + 3
under the assumption that an FMA instruction is not more expensive
than a simple add. This certainly isn't always true. While e.g
11 matches
Mail list logo
