1 /*

     2  * simd.h: abstraction of Intel SSE instruction set

     3  *

     4  * This file is part of Pyrit Ray Tracer.

     5  *

     6  * Copyright 2008  Radek Brich

     7  *

     8  * Permission is hereby granted, free of charge, to any person obtaining a copy

     9  * of this software and associated documentation files (the "Software"), to deal

    10  * in the Software without restriction, including without limitation the rights

    11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell

    12  * copies of the Software, and to permit persons to whom the Software is

    13  * furnished to do so, subject to the following conditions:

    14  *

    15  * The above copyright notice and this permission notice shall be included in

    16  * all copies or substantial portions of the Software.

    17  *

    18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR

    19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,

    20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE

    21  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER

    22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,

    23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN

    24  * THE SOFTWARE.

    25  */

    26 

    27 #ifndef SIMD_H

    28 #define SIMD_H

    29 

    30 #include "common.h"

    31 

    32 #ifndef NO_SIMD

    33 

    34 #include <xmmintrin.h>

    35 

    36 typedef __m128 mfloat4;

    37 

    38 #define mZero  _mm_set_ps1(0.0f)

    39 #define mOne   _mm_set_ps1(1.0f)

    40 #define mTwo   _mm_set_ps1(2.0f)

    41 #define mEps   _mm_set_ps1(Eps)

    42 #define mMEps  _mm_set_ps1(-Eps)

    43 #define mInf   _mm_set_ps1(Inf)

    44 #define mMInf  _mm_set_ps1(-Inf)

    45 #define mAllSet  _mm_cmplt_ps(mZero, mOne)

    46 

    47 #define mset1 _mm_set_ps1

    48 #define mset _mm_set_ps

    49 

    50 #define madd _mm_add_ps

    51 #define msub _mm_sub_ps

    52 #define mmul _mm_mul_ps

    53 #define mdiv _mm_div_ps

    54 #define msqrt _mm_sqrt_ps

    55 

    56 #define mand _mm_and_ps

    57 #define mor  _mm_or_ps

    58 #define mcmpgt _mm_cmpgt_ps

    59 #define mcmplt _mm_cmplt_ps

    60 #define mcmpge _mm_cmpge_ps

    61 #define mcmple _mm_cmple_ps

    62 #define mcmpeq _mm_cmpeq_ps

    63 #define mcmpneq _mm_cmpneq_ps

    64 #define mmin _mm_min_ps

    65 #define mmax _mm_max_ps

    66 #define mmovemask _mm_movemask_ps

    67 

    68 inline const mfloat4 mselect(const mfloat4& mask, const mfloat4& a, const mfloat4& b)

    69 {

    70 	return _mm_or_ps(_mm_and_ps(mask, a), _mm_andnot_ps(mask, b));

    71 }

    72 

    73 inline const mfloat4 mfastpow(const mfloat4& base, const mfloat4& exponent)

    74 {

    75     __m128 denom = _mm_mul_ps(exponent, base);

    76     denom = _mm_sub_ps(exponent, denom);

    77     denom = _mm_add_ps(base, denom);

    78     return _mm_mul_ps(base, _mm_rcp_ps(denom));

    79 }

    80 #endif

    81 

    82 #endif