[Libclc-dev] [PATCH 2/2] AMDGPU: Implement accurate double precision sqrt

[Jan Vesely]

On Wed, 2015-05-13 at 03:34 +0000, Tom Stellard wrote:
> ---
>  amdgcn/lib/SOURCES      |  1 +
>  amdgcn/lib/math/sqrt.cl | 59 +++++++++++++++++++++++++++++++++++++++++++++++++

I see you created new folder for amdgcn, is the implementation expected
to be different for fp64 enabled EG and NI asics?

>  2 files changed, 60 insertions(+)
>  create mode 100644 amdgcn/lib/SOURCES
>  create mode 100644 amdgcn/lib/math/sqrt.cl
> 
> diff --git a/amdgcn/lib/SOURCES b/amdgcn/lib/SOURCES
> new file mode 100644
> index 0000000..64f874c
> --- /dev/null
> +++ b/amdgcn/lib/SOURCES
> @@ -0,0 +1 @@
> +math/sqrt.cl
> diff --git a/amdgcn/lib/math/sqrt.cl b/amdgcn/lib/math/sqrt.cl
> new file mode 100644
> index 0000000..3e5b17c
> --- /dev/null
> +++ b/amdgcn/lib/math/sqrt.cl
> @@ -0,0 +1,59 @@
> +/*
> + * Copyright (c) 2015 Advanced Micro Devices, Inc.
> + *
> + * Permission is hereby granted, free of charge, to any person obtaining a copy
> + * of this software and associated documentation files (the "Software"), to deal
> + * in the Software without restriction, including without limitation the rights
> + * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
> + * copies of the Software, and to permit persons to whom the Software is
> + * furnished to do so, subject to the following conditions:
> + *
> + * The above copyright notice and this permission notice shall be included in
> + * all copies or substantial portions of the Software.
> + *
> + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
> + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
> + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
> + * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
> + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
> + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
> + * THE SOFTWARE.
> + */
> +
> +#include <clc/clc.h>
> +#include "../../../generic/lib/clcmacro.h"
> +#include "math/clc_sqrt.h"
> +
> +_CLC_DEFINE_UNARY_BUILTIN(float, sqrt, __clc_sqrt, float)
> +
> +#ifdef cl_khr_fp64
> +
> +#pragma OPENCL EXTENSION cl_khr_fp64 : enable
> +
> +
> +_CLC_OVERLOAD _CLC_DEF double sqrt(double x) {
> +
> +  uint vcc = x < 0x1p-767;
> +  uint exp0 = vcc ? 0x100 : 0;
> +  unsigned exp1 = vcc ? 0xffffff80 : 0;
> +
> +  double v01 = ldexp(x, exp0);
> +  double v23 = __builtin_amdgpu_rsq(v01);
> +  double v45 = v01 * v23;
> +  v23 = v23 * 0.5;
> +
> +  double v67 = fma(-v23, v45, 0.5);
> +  v45 = fma(v45, v67, v45);
> +  double v89 = fma(-v45, v45, v01);
> +  v23 = fma(v23, v67, v23);
> +  v45 = fma(v89, v23, v45);
> +  v67 = fma(-v45, v45, v01);
> +  v23 = fma(v67, v23, v45);
> +
> +  v23 = ldexp(v23, exp1);
> +  return ((x == __builtin_inf()) || (x == 0.0)) ? v01 : v23;
> +}
> +
> +_CLC_UNARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, double, sqrt, double);
> +
> +#endif

-- 
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/20150521/94f8e53b/attachment.sig>