Hi.
Ping :-)
(I hope I am not bothering you with it)
Giuliano.
On Thu, Jan 10, 2019 at 6:51 PM Giuliano Belinassi
wrote:
>
> Previously, the tests 'sinhatanh-2.c' and 'sinhatanh-3.c' did not count
> the number of functions found in the tree-dump. This patch address this
> issue.
>
> 2019-01-10 G
Hi,
Please check line 4102 in match.pd. Notice that
(if (flag_unsafe_math_optimizations)
And my code is inside that if :-)
Giuliano.
On Thu, Nov 29, 2018 at 2:37 AM Jeff Law wrote:
>
> On 11/27/18 12:38 PM, Giuliano Augusto Faulin Belinassi wrote:
> > Only do this optimizatio
Only do this optimization if funsafe-math and -fno-math-errno are
enabled, as pointed in the previous iteration.
Also added one more test case to ensure that fno-math-errno is
required for the optimization.
Special thanks for Wilco Dijsktra for all his help :-)
gcc/ChangeLog
2018-11-27 Giuliano
Hi,
Sorry, but I am a little confused.
On Wed, Nov 14, 2018 at 11:28 AM Wilco Dijkstra wrote:
>
> Hi,
>
>
> > Indeed. After plotting the graph of both functions, it is very clear
> > that this check isn't required. Sorry about that.
>
> It wouldn't be clear from the graph, you need to check that
On Wed, Nov 14, 2018 at 7:41 AM Richard Biener
wrote:
>
> On Tue, Nov 13, 2018 at 10:25 PM Giuliano Augusto Faulin Belinassi
> wrote:
> >
> > Only do the optimization if flag_signed_zeros &&
> > !flag_finite_math_only is set, as suggested in the previous itera
Only do the optimization if flag_signed_zeros &&
!flag_finite_math_only is set, as suggested in the previous iteration.
Before, the patch did the optimization even when -fno-signed-zeros and
-ffinite-math-only was set. This could generate badly incorrect
results for targets that do not support inf
Hi. Sorry for the late reply :P
> But the max. error in sinh/cosh/atanh is less than 2 ULP, with some math
> ibraries. It could be < 1 ULP, in theory, so sinh(atanh(x)) less than
>2 ULP even.
Sorry, but doesn't the user agree to sacrifice precision for
performance when -ffast-math is enabled?
>
Improves the numerical precision, as suggested on the previous iteration.
Before, the code used 1 - x*x, which can be problematic when x is near
1. Wilco Dijkstra suggested to replace it with 2*(1-fabs(x)) -
(1-fabs(x))*(1-fabs(x)) when fabs(x) > 0.5 to greatly improve
precision, however after som
So I did some further investigation comparing the ULP error.
With the formula that Wilco Dijkstra provided, there are cases where
the substitution is super precise.
With floats:
with input : = 9.9940395355224609375000e-01
sinh: before: = 2.896310058593750
;
> On Fri, Oct 19, 2018 at 09:21:07AM -0300, Giuliano Augusto Faulin Belinassi
> wrote:
> > > Did you enable FMA? I'd expect 1 - x*x to be accurate with FMA, so the
> > > relative error
> > > should be much better. If there is no FMA, 2*(1-fabs(x)) - (1-fabs(x)
Hello,
> Did you enable FMA? I'd expect 1 - x*x to be accurate with FMA, so the
> relative error
> should be much better. If there is no FMA, 2*(1-fabs(x)) - (1-fabs(x))^2
> should be
> more accurate when abs(x)>0.5 and still much faster.
No, but I will check how to enable it if FMA is availabl
ecise approximation). Well, I will try
check that.
But yes, with regard to compatibility this may be a problem.
On Wed, Oct 17, 2018 at 6:42 PM Jeff Law wrote:
>
> On 10/17/18 3:25 PM, Giuliano Augusto Faulin Belinassi wrote:
> >> Hmm, do we have problems as we get close to -1 or 1 where the
> Hmm, do we have problems as we get close to -1 or 1 where the outputs of
> the two forms might diverge?
Well, I did some minor testing with that with input x around nextafter(1, -1);
There are a minor imprecision when comparing directly with
sinh(atanh(x)) and cosh(atanh(x)).
* On 32-bits floa
re about that?
Since the only point of this patch is to simplify these kind of
expressions, and it is not being simplified at all in your arch as the
asm dump suggests, then it seems safe to skip all sinatan-*.c tests.
On Tue, Oct 16, 2018 at 1:25 PM Christophe Lyon
wrote:
>
> On
Hello, Christophe
Could you please dump the assembly of cosatanf here?
On Tue, Oct 16, 2018 at 12:23 PM Christophe Lyon
wrote:
>
> On Tue, 16 Oct 2018 at 17:15, Giuliano Augusto Faulin Belinassi
> wrote:
> >
> > Hello. Sorry for the late reply.
> >
> > &
ne 58 to:
int __attribute__ ((optimize("O0")))
in sinatan-1.c?
On Fri, Oct 12, 2018 at 3:24 PM Christophe Lyon
wrote:
>
> On Fri, 12 Oct 2018 at 20:01, Giuliano Augusto Faulin Belinassi
> wrote:
> >
> > > fc is built with:
> > >mov r0, #0
> &g
ter option?
On Fri, Oct 12, 2018 at 12:57 PM Jeff Law wrote:
>
> On 10/12/18 9:51 AM, Giuliano Augusto Faulin Belinassi wrote:
> > Hello
> > What is the output of these functions on such arch? Since the
> > test didn't fail for the sinatan counterpart, an possible
Hello
What is the output of these functions on such arch? Since the
test didn't fail for the sinatan counterpart, an possible explanation
would be that the calculation of the sqrf, sqrt and sqrtl (lines
62-64) yielded a number that is far behind of what it should be.
However, I am still not su
orted to try to keep
compatibility, but I think this is safe :-).
On Fri, Oct 12, 2018 at 1:09 AM Jeff Law wrote:
>
> On 8/7/18 2:00 PM, Giuliano Augusto Faulin Belinassi wrote:
> > Related with bug 86829, but for hyperbolic trigonometric functions.
> > This patch adds substitution rul
Fixed all issues pointed in the previous iteration.
There is now a significant change regarding how the sin(atan(x))
constant is calculated, as now it checks for which values such that
computing 1 + x*x won't overflow. There are two reasons for this
change: (1) Avoid an intermediate infinity value
Thank you for the review. I will address all these issues :-).
> Imagine a pause here while I lookup isolation of radicals It's been
> a long time... OK. Sure. I see what you're doing here...
Sorry, but I really did not understand your comment. Should I write a
shorter comment for that f
Pinging match.pd and real.c maintainers, as suggested in IRC. Sorry if
it is inappropriate
On Mon, Sep 17, 2018 at 9:46 AM Giuliano Augusto Faulin Belinassi
wrote:
>
> Ping.
>
> On Mon, Sep 3, 2018 at 4:11 PM, Giuliano Augusto Faulin Belinassi
> wrote:
> > Fixed the
Ping.
On Mon, Sep 3, 2018 at 4:11 PM, Giuliano Augusto Faulin Belinassi
wrote:
> Fixed the issues pointed by the previous discussions. Closes PR86829.
>
> Adds substitution rules for sin(atan(x)) and cos(atan(x)), being
> careful with overflow issues by constructing a assumed
Fixed the issues pointed by the previous discussions. Closes PR86829.
Adds substitution rules for sin(atan(x)) and cos(atan(x)), being
careful with overflow issues by constructing a assumed convergence
constant (see comment in real.c).
2018-09-03 Giuliano Belinassi
* match.pd: add simplifi
There was a small typo in the current version of trunk, more precisely
in genmatch.c and exp_unst.ads. This patch fixes it.
Changelog:
2018-08-23 Giuliano Belinassi
* genmatch.c: Fix typo 'exapnded' to 'expanded'.
* exp_unst.ads: Likewise.
I don't think this requires a test case :P
In
2018 06:02 AM, Richard Biener wrote:
>> On Tue, Aug 21, 2018 at 11:27 PM Jeff Law wrote:
>>>
>>> On 08/21/2018 02:08 PM, Giuliano Augusto Faulin Belinassi wrote:
>>>>> Just as an example, compare the results for
>>>>> x = 0x1.fp1023
>
>> On Mon, Aug 20, 2018 at 9:40 PM Jeff Law wrote:
>>>
>>> On 08/04/2018 07:22 AM, Giuliano Augusto Faulin Belinassi wrote:
>>>> Closes bug #86829
>>>>
>>>> Description: Adds substitution rules for both sin(atan(x)) and
>>>> c
ping
On Sat, Aug 4, 2018 at 10:22 AM, Giuliano Augusto Faulin Belinassi
wrote:
> Closes bug #86829
>
> Description: Adds substitution rules for both sin(atan(x)) and
> cos(atan(x)). These formulas are replaced by x / sqrt(x*x + 1) and 1 /
> sqrt(x*x + 1) respectively, providing up
t;
> paul
>
>
>> On Aug 7, 2018, at 4:42 PM, Giuliano Augusto Faulin Belinassi
>> wrote:
>>
>> That is a good question because I didn't know that such targets
>> exists. Any suggestion?
>>
>>
>> On Tue, Aug 7, 2018 at 5:29 PM,
That is a good question because I didn't know that such targets
exists. Any suggestion?
On Tue, Aug 7, 2018 at 5:29 PM, Paul Koning wrote:
>
>
>> On Aug 7, 2018, at 4:00 PM, Giuliano Augusto Faulin Belinassi
>> wrote:
>>
>> Related with bug 86829, but for
Related with bug 86829, but for hyperbolic trigonometric functions.
This patch adds substitution rules to both sinh(tanh(x)) -> x / sqrt(1
- x*x) and cosh(tanh(x)) -> 1 / sqrt(1 - x*x). Notice that the both
formulas has division by 0, but it causes no harm because 1/(+0) ->
+infinity, thus the math
Closes bug #86829
Description: Adds substitution rules for both sin(atan(x)) and
cos(atan(x)). These formulas are replaced by x / sqrt(x*x + 1) and 1 /
sqrt(x*x + 1) respectively, providing up to 10x speedup. This identity
can be proved mathematically.
Changelog:
2018-08-03 Giuliano Belinassi
32 matches
Mail list logo
