[Libclc-dev] [PATCH 1/2] math: Add asin implementation

[Aaron Watry]

On Fri, Sep 5, 2014 at 12:54 PM, Jan Vesely <jan.vesely at rutgers.edu> wrote:
> On Thu, 2014-09-04 at 12:35 -0500, Aaron Watry wrote:
>> asin(x) = PI/2 - acos(x)
>
> LGTM.
>
> just out of curiosity.
> How does the precision compare to just using
> atan2(x, ( sqrt(1-x^2) ) )
> from 5) of your acos patch?
>
> I assume (PI/2 -) does not shift the balance.
>

The precision of both implementations looks ok.  The existing piglit
tests pass when tightened down to a tolerance of 1 ULP and fail at 0
ULP. Given that the spec gives us 4 ULP as allowed variance, it seems
like we're good.

I did the following, alternate implementations and did a quick check
on bitcode length and number of instructions on evergreen.  It seems
like the second variation gives us sufficient precision and fewer
hardware instructions for at least the tested architecture (CEDAR on
latest svn llvm/clang).

If you prefer, I can commit the second implementation instead.

--Aaron


diff --git a/generic/lib/math/asin.inc b/generic/lib/math/asin.inc
index f1a65b3..661663a 100644
--- a/generic/lib/math/asin.inc
+++ b/generic/lib/math/asin.inc
@@ -5,7 +5,15 @@
 #endif

 _CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE asin(__CLC_GENTYPE x) {
+#if 0
+  //Passes with 1ulp on evergreen, fails at 0
+  //(float16): 1786 DW on CEDAR, 22 GPRs, %694 is highest numbered
bitcode instr
   return ( (__CLC_GENTYPE)PI2 - acos(x));
+#else
+  //Passes with 1ulp on evergreen, fails at 0
+  //(float16): 1622 DW on CEDAR, 22 GPRs, %691 is highest numbered
bitcode instr
+  return atan2(x, sqrt((__CLC_GENTYPE)1.0 -(x*x) ) );
+#endif
 }

 #undef PI2




> jan
>
>
>>
>> We already have an implementation of acos(x), so just use that.
>>
>> Signed-off-by: Aaron Watry <awatry at gmail.com>
>> ---
>>  generic/include/clc/clc.h         |  1 +
>>  generic/include/clc/math/asin.h   |  2 ++
>>  generic/include/clc/math/asin.inc |  1 +
>>  generic/lib/SOURCES               |  1 +
>>  generic/lib/math/asin.cl          |  8 ++++++++
>>  generic/lib/math/asin.inc         | 11 +++++++++++
>>  6 files changed, 24 insertions(+)
>>  create mode 100644 generic/include/clc/math/asin.h
>>  create mode 100644 generic/include/clc/math/asin.inc
>>  create mode 100644 generic/lib/math/asin.cl
>>  create mode 100644 generic/lib/math/asin.inc
>>
>> diff --git a/generic/include/clc/clc.h b/generic/include/clc/clc.h
>> index 490893b..079c674 100644
>> --- a/generic/include/clc/clc.h
>> +++ b/generic/include/clc/clc.h
>> @@ -33,6 +33,7 @@
>>
>>  /* 6.11.2 Math Functions */
>>  #include <clc/math/acos.h>
>> +#include <clc/math/asin.h>
>>  #include <clc/math/atan.h>
>>  #include <clc/math/atan2.h>
>>  #include <clc/math/copysign.h>
>> diff --git a/generic/include/clc/math/asin.h b/generic/include/clc/math/asin.h
>> new file mode 100644
>> index 0000000..2a85872
>> --- /dev/null
>> +++ b/generic/include/clc/math/asin.h
>> @@ -0,0 +1,2 @@
>> +#define __CLC_BODY <clc/math/asin.inc>
>> +#include <clc/math/gentype.inc>
>> diff --git a/generic/include/clc/math/asin.inc b/generic/include/clc/math/asin.inc
>> new file mode 100644
>> index 0000000..b4ad8ff
>> --- /dev/null
>> +++ b/generic/include/clc/math/asin.inc
>> @@ -0,0 +1 @@
>> +_CLC_OVERLOAD _CLC_DECL __CLC_GENTYPE asin(__CLC_GENTYPE x);
>> diff --git a/generic/lib/SOURCES b/generic/lib/SOURCES
>> index 8eaaa61..30e182f 100644
>> --- a/generic/lib/SOURCES
>> +++ b/generic/lib/SOURCES
>> @@ -30,6 +30,7 @@ integer/sub_sat_if.ll
>>  integer/sub_sat_impl.ll
>>  integer/upsample.cl
>>  math/acos.cl
>> +math/asin.cl
>>  math/atan.cl
>>  math/atan2.cl
>>  math/copysign.cl
>> diff --git a/generic/lib/math/asin.cl b/generic/lib/math/asin.cl
>> new file mode 100644
>> index 0000000..d56dbd7
>> --- /dev/null
>> +++ b/generic/lib/math/asin.cl
>> @@ -0,0 +1,8 @@
>> +#include <clc/clc.h>
>> +
>> +#ifdef cl_khr_fp64
>> +#pragma OPENCL EXTENSION cl_khr_fp64 : enable
>> +#endif
>> +
>> +#define __CLC_BODY <asin.inc>
>> +#include <clc/math/gentype.inc>
>> diff --git a/generic/lib/math/asin.inc b/generic/lib/math/asin.inc
>> new file mode 100644
>> index 0000000..f1a65b3
>> --- /dev/null
>> +++ b/generic/lib/math/asin.inc
>> @@ -0,0 +1,11 @@
>> +#if __CLC_SCALAR_GENTYPE == double
>> +#define PI2 M_PI_2
>> +#else
>> +#define PI2 M_PI_2_F
>> +#endif
>> +
>> +_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE asin(__CLC_GENTYPE x) {
>> +  return ( (__CLC_GENTYPE)PI2 - acos(x));
>> +}
>> +
>> +#undef PI2
>
> --
> Jan Vesely <jan.vesely at rutgers.edu>