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

[Jan Vesely]

On Thu, 2014-09-04 at 12:35 -0500, Aaron Watry wrote:
> Uses the algorithm:
> tan(x) = sin(x) / sqrt(1-sin^2(x))
> 
> An alternative is:
> tan(x) = sin(x) / cos(x)
> 
> Which produces more verbose bitcode and longer assembly.

this is weird. both EG and SI have both sin and cos instructions. Is the
input normalization code so bad that we are better of doing MUL+SUB+SQRT
instead?

> 
> Either way, the generated bitcode seems pretty nasty and a more optimized
> but still precise-enough solution is welcome.
> 
> Signed-off-by: Aaron Watry <awatry at gmail.com>
> ---
>  generic/include/clc/clc.h        | 1 +
>  generic/include/clc/math/tan.h   | 2 ++
>  generic/include/clc/math/tan.inc | 1 +
>  generic/lib/SOURCES              | 1 +
>  generic/lib/math/tan.cl          | 8 ++++++++
>  generic/lib/math/tan.inc         | 8 ++++++++
>  6 files changed, 21 insertions(+)
>  create mode 100644 generic/include/clc/math/tan.h
>  create mode 100644 generic/include/clc/math/tan.inc
>  create mode 100644 generic/lib/math/tan.cl
>  create mode 100644 generic/lib/math/tan.inc
> 
> diff --git a/generic/include/clc/clc.h b/generic/include/clc/clc.h
> index 079c674..69e44b6 100644
> --- a/generic/include/clc/clc.h
> +++ b/generic/include/clc/clc.h
> @@ -60,6 +60,7 @@
>  #include <clc/math/sin.h>
>  #include <clc/math/sincos.h>
>  #include <clc/math/sqrt.h>
> +#include <clc/math/tan.h>
>  #include <clc/math/trunc.h>
>  #include <clc/math/native_cos.h>
>  #include <clc/math/native_divide.h>
> diff --git a/generic/include/clc/math/tan.h b/generic/include/clc/math/tan.h
> new file mode 100644
> index 0000000..d2d52a9
> --- /dev/null
> +++ b/generic/include/clc/math/tan.h
> @@ -0,0 +1,2 @@
> +#define __CLC_BODY <clc/math/tan.inc>
> +#include <clc/math/gentype.inc>
> diff --git a/generic/include/clc/math/tan.inc b/generic/include/clc/math/tan.inc
> new file mode 100644
> index 0000000..50c5b1d
> --- /dev/null
> +++ b/generic/include/clc/math/tan.inc
> @@ -0,0 +1 @@
> +_CLC_OVERLOAD _CLC_DECL __CLC_GENTYPE tan(__CLC_GENTYPE x);
> diff --git a/generic/lib/SOURCES b/generic/lib/SOURCES
> index 30e182f..0667d14 100644
> --- a/generic/lib/SOURCES
> +++ b/generic/lib/SOURCES
> @@ -48,6 +48,7 @@ math/pown.cl
>  math/sin.cl
>  math/sincos.cl
>  math/sincos_helpers.cl
> +math/tan.cl
>  relational/all.cl
>  relational/any.cl
>  relational/isequal.cl
> diff --git a/generic/lib/math/tan.cl b/generic/lib/math/tan.cl
> new file mode 100644
> index 0000000..a447999
> --- /dev/null
> +++ b/generic/lib/math/tan.cl
> @@ -0,0 +1,8 @@
> +#include <clc/clc.h>
> +
> +#ifdef cl_khr_fp64
> +#pragma OPENCL EXTENSION cl_khr_fp64 : enable
> +#endif
> +
> +#define __CLC_BODY <tan.inc>
> +#include <clc/math/gentype.inc>
> diff --git a/generic/lib/math/tan.inc b/generic/lib/math/tan.inc
> new file mode 100644
> index 0000000..8d9d9fe
> --- /dev/null
> +++ b/generic/lib/math/tan.inc
> @@ -0,0 +1,8 @@
> +/*
> + * Note: tan(x) = sin(x)/cos(x) also, but the final assembly ends up being
> + *       twice as long for R600 (maybe for others as well).
> + */
> +_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE tan(__CLC_GENTYPE x) {
> +  __CLC_GENTYPE sinx = sin(x);
> +  return sinx / sqrt( (__CLC_GENTYPE) 1.0 - (sinx*sinx) );
> +}

-- 
Jan Vesely <jan.vesely at rutgers.edu>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: This is a digitally signed message part
URL: <http://lists.llvm.org/pipermail/libclc-dev/attachments/20140905/e34e4e39/attachment.sig>