| 1 | #if GLM_ARCH & GLM_ARCH_SSE2_BIT |
| 2 | |
| 3 | #include "type_mat4x4.hpp" |
| 4 | #include "../geometric.hpp" |
| 5 | #include "../simd/matrix.h" |
| 6 | #include <cstring> |
| 7 | |
| 8 | namespace glm{ |
| 9 | namespace detail |
| 10 | { |
| 11 | # if GLM_CONFIG_ALIGNED_GENTYPES == GLM_ENABLE |
| 12 | template<qualifier Q> |
| 13 | struct compute_matrixCompMult<4, 4, float, Q, true> |
| 14 | { |
| 15 | static_assert(detail::is_aligned<Q>::value, "Specialization requires aligned"); |
| 16 | |
| 17 | GLM_FUNC_QUALIFIER static mat<4, 4, float, Q> call(mat<4, 4, float, Q> const& x, mat<4, 4, float, Q> const& y) |
| 18 | { |
| 19 | mat<4, 4, float, Q> Result; |
| 20 | glm_mat4_matrixCompMult( |
| 21 | &x[0].data, |
| 22 | &y[0].data, |
| 23 | &Result[0].data); |
| 24 | return Result; |
| 25 | } |
| 26 | }; |
| 27 | # endif |
| 28 | |
| 29 | template<qualifier Q> |
| 30 | struct compute_transpose<4, 4, float, Q, true> |
| 31 | { |
| 32 | GLM_FUNC_QUALIFIER static mat<4, 4, float, Q> call(mat<4, 4, float, Q> const& m) |
| 33 | { |
| 34 | mat<4, 4, float, Q> Result; |
| 35 | glm_mat4_transpose(&m[0].data, &Result[0].data); |
| 36 | return Result; |
| 37 | } |
| 38 | }; |
| 39 | |
| 40 | template<qualifier Q> |
| 41 | struct compute_transpose<3, 3, float, Q, true> |
| 42 | { |
| 43 | GLM_FUNC_QUALIFIER static mat<3, 3, float, Q> call(mat<3, 3, float, Q> const& m) |
| 44 | { |
| 45 | mat<3, 3, float, Q> Result; |
| 46 | glm_mat3_transpose(&m[0].data, &Result[0].data); |
| 47 | return Result; |
| 48 | } |
| 49 | }; |
| 50 | |
| 51 | template<qualifier Q> |
| 52 | struct compute_determinant<4, 4, float, Q, true> |
| 53 | { |
| 54 | GLM_FUNC_QUALIFIER static float call(mat<4, 4, float, Q> const& m) |
| 55 | { |
| 56 | return _mm_cvtss_f32(glm_mat4_determinant(&m[0].data)); |
| 57 | } |
| 58 | }; |
| 59 | |
| 60 | template<qualifier Q> |
| 61 | struct compute_inverse<4, 4, float, Q, true> |
| 62 | { |
| 63 | GLM_FUNC_QUALIFIER static mat<4, 4, float, Q> call(mat<4, 4, float, Q> const& m) |
| 64 | { |
| 65 | mat<4, 4, float, Q> Result; |
| 66 | glm_mat4_inverse(&m[0].data, &Result[0].data); |
| 67 | return Result; |
| 68 | } |
| 69 | }; |
| 70 | }//namespace detail |
| 71 | |
| 72 | # if GLM_CONFIG_ALIGNED_GENTYPES == GLM_ENABLE |
| 73 | template<> |
| 74 | GLM_FUNC_QUALIFIER mat<4, 4, float, aligned_lowp> outerProduct<4, 4, float, aligned_lowp>(vec<4, float, aligned_lowp> const& c, vec<4, float, aligned_lowp> const& r) |
| 75 | { |
| 76 | __m128 NativeResult[4]; |
| 77 | glm_mat4_outerProduct(c.data, r.data, NativeResult); |
| 78 | mat<4, 4, float, aligned_lowp> Result; |
| 79 | std::memcpy(&Result[0], &NativeResult[0], sizeof(Result)); |
| 80 | return Result; |
| 81 | } |
| 82 | |
| 83 | template<> |
| 84 | GLM_FUNC_QUALIFIER mat<4, 4, float, aligned_mediump> outerProduct<4, 4, float, aligned_mediump>(vec<4, float, aligned_mediump> const& c, vec<4, float, aligned_mediump> const& r) |
| 85 | { |
| 86 | __m128 NativeResult[4]; |
| 87 | glm_mat4_outerProduct(c.data, r.data, NativeResult); |
| 88 | mat<4, 4, float, aligned_mediump> Result; |
| 89 | std::memcpy(&Result[0], &NativeResult[0], sizeof(Result)); |
| 90 | return Result; |
| 91 | } |
| 92 | |
| 93 | template<> |
| 94 | GLM_FUNC_QUALIFIER mat<4, 4, float, aligned_highp> outerProduct<4, 4, float, aligned_highp>(vec<4, float, aligned_highp> const& c, vec<4, float, aligned_highp> const& r) |
| 95 | { |
| 96 | __m128 NativeResult[4]; |
| 97 | glm_mat4_outerProduct(c.data, r.data, NativeResult); |
| 98 | mat<4, 4, float, aligned_highp> Result; |
| 99 | std::memcpy(&Result[0], &NativeResult[0], sizeof(Result)); |
| 100 | return Result; |
| 101 | } |
| 102 | # endif |
| 103 | }//namespace glm |
| 104 | |
| 105 | #elif GLM_ARCH & GLM_ARCH_NEON_BIT |
| 106 | |
| 107 | namespace glm { |
| 108 | #if GLM_LANG & GLM_LANG_CXX11_FLAG |
| 109 | template <qualifier Q> |
| 110 | GLM_FUNC_QUALIFIER |
| 111 | typename std::enable_if<detail::is_aligned<Q>::value, mat<4, 4, float, Q>>::type |
| 112 | operator*(mat<4, 4, float, Q> const & m1, mat<4, 4, float, Q> const & m2) |
| 113 | { |
| 114 | auto MulRow = [&](int l) { |
| 115 | float32x4_t const SrcA = m2[l].data; |
| 116 | |
| 117 | float32x4_t r = neon::mul_lane(m1[0].data, SrcA, 0); |
| 118 | r = neon::madd_lane(r, m1[1].data, SrcA, 1); |
| 119 | r = neon::madd_lane(r, m1[2].data, SrcA, 2); |
| 120 | r = neon::madd_lane(r, m1[3].data, SrcA, 3); |
| 121 | |
| 122 | return r; |
| 123 | }; |
| 124 | |
| 125 | mat<4, 4, float, aligned_highp> Result; |
| 126 | Result[0].data = MulRow(0); |
| 127 | Result[1].data = MulRow(1); |
| 128 | Result[2].data = MulRow(2); |
| 129 | Result[3].data = MulRow(3); |
| 130 | |
| 131 | return Result; |
| 132 | } |
| 133 | #endif // CXX11 |
| 134 | |
| 135 | namespace detail |
| 136 | { |
| 137 | template<qualifier Q> |
| 138 | struct compute_inverse<4, 4, float, Q, true> |
| 139 | { |
| 140 | GLM_FUNC_QUALIFIER static mat<4, 4, float, Q> call(mat<4, 4, float, Q> const& m) |
| 141 | { |
| 142 | float32x4_t const& m0 = m[0].data; |
| 143 | float32x4_t const& m1 = m[1].data; |
| 144 | float32x4_t const& m2 = m[2].data; |
| 145 | float32x4_t const& m3 = m[3].data; |
| 146 | |
| 147 | // m[2][2] * m[3][3] - m[3][2] * m[2][3]; |
| 148 | // m[2][2] * m[3][3] - m[3][2] * m[2][3]; |
| 149 | // m[1][2] * m[3][3] - m[3][2] * m[1][3]; |
| 150 | // m[1][2] * m[2][3] - m[2][2] * m[1][3]; |
| 151 | |
| 152 | float32x4_t Fac0; |
| 153 | { |
| 154 | float32x4_t w0 = vcombine_f32(neon::dup_lane(m2, 2), neon::dup_lane(m1, 2)); |
| 155 | float32x4_t w1 = neon::copy_lane(neon::dupq_lane(m3, 3), 3, m2, 3); |
| 156 | float32x4_t w2 = neon::copy_lane(neon::dupq_lane(m3, 2), 3, m2, 2); |
| 157 | float32x4_t w3 = vcombine_f32(neon::dup_lane(m2, 3), neon::dup_lane(m1, 3)); |
| 158 | Fac0 = w0 * w1 - w2 * w3; |
| 159 | } |
| 160 | |
| 161 | // m[2][1] * m[3][3] - m[3][1] * m[2][3]; |
| 162 | // m[2][1] * m[3][3] - m[3][1] * m[2][3]; |
| 163 | // m[1][1] * m[3][3] - m[3][1] * m[1][3]; |
| 164 | // m[1][1] * m[2][3] - m[2][1] * m[1][3]; |
| 165 | |
| 166 | float32x4_t Fac1; |
| 167 | { |
| 168 | float32x4_t w0 = vcombine_f32(neon::dup_lane(m2, 1), neon::dup_lane(m1, 1)); |
| 169 | float32x4_t w1 = neon::copy_lane(neon::dupq_lane(m3, 3), 3, m2, 3); |
| 170 | float32x4_t w2 = neon::copy_lane(neon::dupq_lane(m3, 1), 3, m2, 1); |
| 171 | float32x4_t w3 = vcombine_f32(neon::dup_lane(m2, 3), neon::dup_lane(m1, 3)); |
| 172 | Fac1 = w0 * w1 - w2 * w3; |
| 173 | } |
| 174 | |
| 175 | // m[2][1] * m[3][2] - m[3][1] * m[2][2]; |
| 176 | // m[2][1] * m[3][2] - m[3][1] * m[2][2]; |
| 177 | // m[1][1] * m[3][2] - m[3][1] * m[1][2]; |
| 178 | // m[1][1] * m[2][2] - m[2][1] * m[1][2]; |
| 179 | |
| 180 | float32x4_t Fac2; |
| 181 | { |
| 182 | float32x4_t w0 = vcombine_f32(neon::dup_lane(m2, 1), neon::dup_lane(m1, 1)); |
| 183 | float32x4_t w1 = neon::copy_lane(neon::dupq_lane(m3, 2), 3, m2, 2); |
| 184 | float32x4_t w2 = neon::copy_lane(neon::dupq_lane(m3, 1), 3, m2, 1); |
| 185 | float32x4_t w3 = vcombine_f32(neon::dup_lane(m2, 2), neon::dup_lane(m1, 2)); |
| 186 | Fac2 = w0 * w1 - w2 * w3; |
| 187 | } |
| 188 | |
| 189 | // m[2][0] * m[3][3] - m[3][0] * m[2][3]; |
| 190 | // m[2][0] * m[3][3] - m[3][0] * m[2][3]; |
| 191 | // m[1][0] * m[3][3] - m[3][0] * m[1][3]; |
| 192 | // m[1][0] * m[2][3] - m[2][0] * m[1][3]; |
| 193 | |
| 194 | float32x4_t Fac3; |
| 195 | { |
| 196 | float32x4_t w0 = vcombine_f32(neon::dup_lane(m2, 0), neon::dup_lane(m1, 0)); |
| 197 | float32x4_t w1 = neon::copy_lane(neon::dupq_lane(m3, 3), 3, m2, 3); |
| 198 | float32x4_t w2 = neon::copy_lane(neon::dupq_lane(m3, 0), 3, m2, 0); |
| 199 | float32x4_t w3 = vcombine_f32(neon::dup_lane(m2, 3), neon::dup_lane(m1, 3)); |
| 200 | Fac3 = w0 * w1 - w2 * w3; |
| 201 | } |
| 202 | |
| 203 | // m[2][0] * m[3][2] - m[3][0] * m[2][2]; |
| 204 | // m[2][0] * m[3][2] - m[3][0] * m[2][2]; |
| 205 | // m[1][0] * m[3][2] - m[3][0] * m[1][2]; |
| 206 | // m[1][0] * m[2][2] - m[2][0] * m[1][2]; |
| 207 | |
| 208 | float32x4_t Fac4; |
| 209 | { |
| 210 | float32x4_t w0 = vcombine_f32(neon::dup_lane(m2, 0), neon::dup_lane(m1, 0)); |
| 211 | float32x4_t w1 = neon::copy_lane(neon::dupq_lane(m3, 2), 3, m2, 2); |
| 212 | float32x4_t w2 = neon::copy_lane(neon::dupq_lane(m3, 0), 3, m2, 0); |
| 213 | float32x4_t w3 = vcombine_f32(neon::dup_lane(m2, 2), neon::dup_lane(m1, 2)); |
| 214 | Fac4 = w0 * w1 - w2 * w3; |
| 215 | } |
| 216 | |
| 217 | // m[2][0] * m[3][1] - m[3][0] * m[2][1]; |
| 218 | // m[2][0] * m[3][1] - m[3][0] * m[2][1]; |
| 219 | // m[1][0] * m[3][1] - m[3][0] * m[1][1]; |
| 220 | // m[1][0] * m[2][1] - m[2][0] * m[1][1]; |
| 221 | |
| 222 | float32x4_t Fac5; |
| 223 | { |
| 224 | float32x4_t w0 = vcombine_f32(neon::dup_lane(m2, 0), neon::dup_lane(m1, 0)); |
| 225 | float32x4_t w1 = neon::copy_lane(neon::dupq_lane(m3, 1), 3, m2, 1); |
| 226 | float32x4_t w2 = neon::copy_lane(neon::dupq_lane(m3, 0), 3, m2, 0); |
| 227 | float32x4_t w3 = vcombine_f32(neon::dup_lane(m2, 1), neon::dup_lane(m1, 1)); |
| 228 | Fac5 = w0 * w1 - w2 * w3; |
| 229 | } |
| 230 | |
| 231 | float32x4_t Vec0 = neon::copy_lane(neon::dupq_lane(m0, 0), 0, m1, 0); // (m[1][0], m[0][0], m[0][0], m[0][0]); |
| 232 | float32x4_t Vec1 = neon::copy_lane(neon::dupq_lane(m0, 1), 0, m1, 1); // (m[1][1], m[0][1], m[0][1], m[0][1]); |
| 233 | float32x4_t Vec2 = neon::copy_lane(neon::dupq_lane(m0, 2), 0, m1, 2); // (m[1][2], m[0][2], m[0][2], m[0][2]); |
| 234 | float32x4_t Vec3 = neon::copy_lane(neon::dupq_lane(m0, 3), 0, m1, 3); // (m[1][3], m[0][3], m[0][3], m[0][3]); |
| 235 | |
| 236 | float32x4_t Inv0 = Vec1 * Fac0 - Vec2 * Fac1 + Vec3 * Fac2; |
| 237 | float32x4_t Inv1 = Vec0 * Fac0 - Vec2 * Fac3 + Vec3 * Fac4; |
| 238 | float32x4_t Inv2 = Vec0 * Fac1 - Vec1 * Fac3 + Vec3 * Fac5; |
| 239 | float32x4_t Inv3 = Vec0 * Fac2 - Vec1 * Fac4 + Vec2 * Fac5; |
| 240 | |
| 241 | float32x4_t r0 = float32x4_t{-1, +1, -1, +1} * Inv0; |
| 242 | float32x4_t r1 = float32x4_t{+1, -1, +1, -1} * Inv1; |
| 243 | float32x4_t r2 = float32x4_t{-1, +1, -1, +1} * Inv2; |
| 244 | float32x4_t r3 = float32x4_t{+1, -1, +1, -1} * Inv3; |
| 245 | |
| 246 | float32x4_t det = neon::mul_lane(r0, m0, 0); |
| 247 | det = neon::madd_lane(det, r1, m0, 1); |
| 248 | det = neon::madd_lane(det, r2, m0, 2); |
| 249 | det = neon::madd_lane(det, r3, m0, 3); |
| 250 | |
| 251 | float32x4_t rdet = vdupq_n_f32(1 / vgetq_lane_f32(det, 0)); |
| 252 | |
| 253 | mat<4, 4, float, Q> r; |
| 254 | r[0].data = vmulq_f32(r0, rdet); |
| 255 | r[1].data = vmulq_f32(r1, rdet); |
| 256 | r[2].data = vmulq_f32(r2, rdet); |
| 257 | r[3].data = vmulq_f32(r3, rdet); |
| 258 | return r; |
| 259 | } |
| 260 | }; |
| 261 | }//namespace detail |
| 262 | }//namespace glm |
| 263 | #endif |