1 /* ----------------------------------------------------------------------------- 2 3 Copyright (c) 2006 Simon Brown si@sjbrown.co.uk 4 5 Permission is hereby granted, free of charge, to any person obtaining 6 a copy of this software and associated documentation files (the 7 "Software"), to deal in the Software without restriction, including 8 without limitation the rights to use, copy, modify, merge, publish, 9 distribute, sublicense, and/or sell copies of the Software, and to 10 permit persons to whom the Software is furnished to do so, subject to 11 the following conditions: 12 13 The above copyright notice and this permission notice shall be included 14 in all copies or substantial portions of the Software. 15 16 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS 17 OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 18 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. 19 IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY 20 CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, 21 TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE 22 SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. 23 24 -------------------------------------------------------------------------- */ 25 26 #ifndef SQUISH_SIMD_SSE_H 27 #define SQUISH_SIMD_SSE_H 28 29 #include <xmmintrin.h> 30 #if ( SQUISH_USE_SSE > 1 ) 31 #include <emmintrin.h> 32 #endif 33 #include <cassert> 34 35 #define SQUISH_SSE_SPLAT( a ) \ 36 ( ( a ) | ( ( a ) << 2 ) | ( ( a ) << 4 ) | ( ( a ) << 6 ) ) 37 38 namespace squish { 39 40 #define VEC4_CONST( X ) Vec4( _mm_set1_ps( X ) ) 41 42 class Vec4 43 { 44 public: 45 typedef Vec4 const& Arg; 46 Vec4()47 Vec4() {} 48 Vec4(__m128 v)49 explicit Vec4( __m128 v ) : m_v( v ) {} 50 Vec4(Vec4 const & arg)51 Vec4( Vec4 const& arg ) : m_v( arg.m_v ) {} 52 53 Vec4& operator=( Vec4 const& arg ) 54 { 55 m_v = arg.m_v; 56 return *this; 57 } 58 Vec4(const float * v)59 Vec4( const float * v ) 60 { 61 m_v = _mm_load_ps( v ); 62 } 63 Vec4(float x,float y,float z,float w)64 Vec4( float x, float y, float z, float w ) 65 { 66 m_v = _mm_setr_ps( x, y, z, w ); 67 } 68 GetVec3()69 Vec3 GetVec3() const 70 { 71 SQUISH_ALIGN_16 float c[4]; 72 _mm_store_ps( c, m_v ); 73 return Vec3( c[0], c[1], c[2] ); 74 } 75 SplatX()76 Vec4 SplatX() const { return Vec4( _mm_shuffle_ps( m_v, m_v, SQUISH_SSE_SPLAT( 0 ) ) ); } SplatY()77 Vec4 SplatY() const { return Vec4( _mm_shuffle_ps( m_v, m_v, SQUISH_SSE_SPLAT( 1 ) ) ); } SplatZ()78 Vec4 SplatZ() const { return Vec4( _mm_shuffle_ps( m_v, m_v, SQUISH_SSE_SPLAT( 2 ) ) ); } SplatW()79 Vec4 SplatW() const { return Vec4( _mm_shuffle_ps( m_v, m_v, SQUISH_SSE_SPLAT( 3 ) ) ); } 80 81 Vec4& operator+=( Arg v ) 82 { 83 m_v = _mm_add_ps( m_v, v.m_v ); 84 return *this; 85 } 86 87 Vec4& operator-=( Arg v ) 88 { 89 m_v = _mm_sub_ps( m_v, v.m_v ); 90 return *this; 91 } 92 93 Vec4& operator*=( Arg v ) 94 { 95 m_v = _mm_mul_ps( m_v, v.m_v ); 96 return *this; 97 } 98 99 friend Vec4 operator+( Vec4::Arg left, Vec4::Arg right ) 100 { 101 return Vec4( _mm_add_ps( left.m_v, right.m_v ) ); 102 } 103 104 friend Vec4 operator-( Vec4::Arg left, Vec4::Arg right ) 105 { 106 return Vec4( _mm_sub_ps( left.m_v, right.m_v ) ); 107 } 108 109 friend Vec4 operator*( Vec4::Arg left, Vec4::Arg right ) 110 { 111 return Vec4( _mm_mul_ps( left.m_v, right.m_v ) ); 112 } 113 114 //! Returns a*b + c MultiplyAdd(Vec4::Arg a,Vec4::Arg b,Vec4::Arg c)115 friend Vec4 MultiplyAdd( Vec4::Arg a, Vec4::Arg b, Vec4::Arg c ) 116 { 117 return Vec4( _mm_add_ps( _mm_mul_ps( a.m_v, b.m_v ), c.m_v ) ); 118 } 119 120 //! Returns -( a*b - c ) NegativeMultiplySubtract(Vec4::Arg a,Vec4::Arg b,Vec4::Arg c)121 friend Vec4 NegativeMultiplySubtract( Vec4::Arg a, Vec4::Arg b, Vec4::Arg c ) 122 { 123 return Vec4( _mm_sub_ps( c.m_v, _mm_mul_ps( a.m_v, b.m_v ) ) ); 124 } 125 Reciprocal(Vec4::Arg v)126 friend Vec4 Reciprocal( Vec4::Arg v ) 127 { 128 // get the reciprocal estimate 129 __m128 estimate = _mm_rcp_ps( v.m_v ); 130 131 // one round of Newton-Rhaphson refinement 132 __m128 diff = _mm_sub_ps( _mm_set1_ps( 1.0f ), _mm_mul_ps( estimate, v.m_v ) ); 133 return Vec4( _mm_add_ps( _mm_mul_ps( diff, estimate ), estimate ) ); 134 } 135 Min(Vec4::Arg left,Vec4::Arg right)136 friend Vec4 Min( Vec4::Arg left, Vec4::Arg right ) 137 { 138 return Vec4( _mm_min_ps( left.m_v, right.m_v ) ); 139 } 140 Max(Vec4::Arg left,Vec4::Arg right)141 friend Vec4 Max( Vec4::Arg left, Vec4::Arg right ) 142 { 143 return Vec4( _mm_max_ps( left.m_v, right.m_v ) ); 144 } 145 Truncate(Vec4::Arg v)146 friend Vec4 Truncate( Vec4::Arg v ) 147 { 148 #if ( SQUISH_USE_SSE == 1 ) 149 // convert to ints 150 __m128 input = v.m_v; 151 __m64 lo = _mm_cvttps_pi32( input ); 152 __m64 hi = _mm_cvttps_pi32( _mm_movehl_ps( input, input ) ); 153 154 // convert to floats 155 __m128 part = _mm_movelh_ps( input, _mm_cvtpi32_ps( input, hi ) ); 156 __m128 truncated = _mm_cvtpi32_ps( part, lo ); 157 158 // clear out the MMX multimedia state to allow FP calls later 159 _mm_empty(); 160 return Vec4( truncated ); 161 #else 162 // use SSE2 instructions 163 return Vec4( _mm_cvtepi32_ps( _mm_cvttps_epi32( v.m_v ) ) ); 164 #endif 165 } 166 CompareEqual(Vec4::Arg left,Vec4::Arg right)167 friend Vec4 CompareEqual( Vec4::Arg left, Vec4::Arg right ) 168 { 169 return Vec4( _mm_cmpeq_ps( left.m_v, right.m_v ) ); 170 } 171 Select(Vec4::Arg off,Vec4::Arg on,Vec4::Arg bits)172 friend Vec4 Select( Vec4::Arg off, Vec4::Arg on, Vec4::Arg bits ) 173 { 174 __m128 a = _mm_andnot_ps( bits.m_v, off.m_v ); 175 __m128 b = _mm_and_ps( bits.m_v, on.m_v ); 176 177 return Vec4( _mm_or_ps( a, b ) ); 178 } 179 CompareAnyLessThan(Vec4::Arg left,Vec4::Arg right)180 friend bool CompareAnyLessThan( Vec4::Arg left, Vec4::Arg right ) 181 { 182 __m128 bits = _mm_cmplt_ps( left.m_v, right.m_v ); 183 int value = _mm_movemask_ps( bits ); 184 return value != 0; 185 } 186 187 private: 188 __m128 m_v; 189 }; 190 191 } // namespace squish 192 193 #endif // ndef SQUISH_SIMD_SSE_H 194