diff options
author | Tom Stellard <thomas.stellard@amd.com> | 2015-07-10 13:37:08 +0000 |
---|---|---|
committer | Tom Stellard <thomas.stellard@amd.com> | 2015-07-10 13:37:08 +0000 |
commit | 4f8d26230c83d630a7dbc0a7411beb9f25c4c6e0 (patch) | |
tree | de7058137b81c6fda68c4e147657def43bc52afe /libclc/r600 | |
parent | 2ef5ec6b2bc0c0bcc5b7b09004de4038fe1f901e (diff) | |
download | llvm-4f8d26230c83d630a7dbc0a7411beb9f25c4c6e0.zip llvm-4f8d26230c83d630a7dbc0a7411beb9f25c4c6e0.tar.gz llvm-4f8d26230c83d630a7dbc0a7411beb9f25c4c6e0.tar.bz2 |
R600: Implement accurate double precision sqrt v2
v2:
- Use same implementation for R600 and gcn.
llvm-svn: 241907
Diffstat (limited to 'libclc/r600')
-rw-r--r-- | libclc/r600/lib/SOURCES | 1 | ||||
-rw-r--r-- | libclc/r600/lib/math/sqrt.cl | 59 |
2 files changed, 60 insertions, 0 deletions
diff --git a/libclc/r600/lib/SOURCES b/libclc/r600/lib/SOURCES index 0f1227e..5cdb14a 100644 --- a/libclc/r600/lib/SOURCES +++ b/libclc/r600/lib/SOURCES @@ -1,6 +1,7 @@ atomic/atomic.cl math/ldexp.cl math/nextafter.cl +math/sqrt.cl workitem/get_num_groups.ll workitem/get_group_id.ll workitem/get_local_size.ll diff --git a/libclc/r600/lib/math/sqrt.cl b/libclc/r600/lib/math/sqrt.cl new file mode 100644 index 0000000..3e5b17c --- /dev/null +++ b/libclc/r600/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 |