/dports/devel/libflatarray/libflatarray-0.3.0/src/detail/ |
H A D | short_vec_neon_float_4.hpp | 65 val1 = vsubq_f32(val1, other.val1); in operator -=() 71 return short_vec<float, 4>(vsubq_f32(val1, other.val1)); in operator -()
|
/dports/graphics/waifu2x-ncnn-vulkan/waifu2x-ncnn-vulkan-20210521/src/ncnn/src/layer/arm/ |
H A D | convolution_3x3.h | 1985 float32x4_t _r_0_m_6 = vsubq_f32(_r_00, _r_66); in conv3x3s1_winograd64_neon4() 1986 float32x4_t _r_7_m_1 = vsubq_f32(_r_77, _r_11); in conv3x3s1_winograd64_neon4() 1988 float32x4_t _r_4_m_2 = vsubq_f32(_r_44, _r_22); in conv3x3s1_winograd64_neon4() 1989 float32x4_t _r_3_m_5 = vsubq_f32(_r_33, _r_55); in conv3x3s1_winograd64_neon4() 2004 float32x4_t _tmp2 = vsubq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_neon4() 2019 float32x4_t _tmp4 = vsubq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_neon4() 2032 float32x4_t _tmp6 = vsubq_f32(_tmp56a, _tmp56b); in conv3x3s1_winograd64_neon4() 2083 float32x4_t _t_0_m_6 = vsubq_f32(_t_00, _t_66); in conv3x3s1_winograd64_neon4() 2084 float32x4_t _t_7_m_1 = vsubq_f32(_t_77, _t_11); in conv3x3s1_winograd64_neon4() 2086 float32x4_t _t_4_m_2 = vsubq_f32(_t_44, _t_22); in conv3x3s1_winograd64_neon4() [all …]
|
H A D | selu_arm.cpp | 55 _nps = vsubq_f32(_nps, _one); in forward_inplace()
|
/dports/benchmarks/vkpeak/vkpeak-20210430/ncnn/src/layer/arm/ |
H A D | convolution_3x3.h | 1985 float32x4_t _r_0_m_6 = vsubq_f32(_r_00, _r_66); in conv3x3s1_winograd64_neon4() 1986 float32x4_t _r_7_m_1 = vsubq_f32(_r_77, _r_11); in conv3x3s1_winograd64_neon4() 1988 float32x4_t _r_4_m_2 = vsubq_f32(_r_44, _r_22); in conv3x3s1_winograd64_neon4() 1989 float32x4_t _r_3_m_5 = vsubq_f32(_r_33, _r_55); in conv3x3s1_winograd64_neon4() 2004 float32x4_t _tmp2 = vsubq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_neon4() 2019 float32x4_t _tmp4 = vsubq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_neon4() 2032 float32x4_t _tmp6 = vsubq_f32(_tmp56a, _tmp56b); in conv3x3s1_winograd64_neon4() 2083 float32x4_t _t_0_m_6 = vsubq_f32(_t_00, _t_66); in conv3x3s1_winograd64_neon4() 2084 float32x4_t _t_7_m_1 = vsubq_f32(_t_77, _t_11); in conv3x3s1_winograd64_neon4() 2086 float32x4_t _t_4_m_2 = vsubq_f32(_t_44, _t_22); in conv3x3s1_winograd64_neon4() [all …]
|
H A D | selu_arm.cpp | 55 _nps = vsubq_f32(_nps, _one); in forward_inplace()
|
/dports/misc/ncnn/ncnn-20211208/src/layer/arm/ |
H A D | convolution_3x3.h | 1985 float32x4_t _r_0_m_6 = vsubq_f32(_r_00, _r_66); in conv3x3s1_winograd64_neon4() 1986 float32x4_t _r_7_m_1 = vsubq_f32(_r_77, _r_11); in conv3x3s1_winograd64_neon4() 1988 float32x4_t _r_4_m_2 = vsubq_f32(_r_44, _r_22); in conv3x3s1_winograd64_neon4() 1989 float32x4_t _r_3_m_5 = vsubq_f32(_r_33, _r_55); in conv3x3s1_winograd64_neon4() 2004 float32x4_t _tmp2 = vsubq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_neon4() 2019 float32x4_t _tmp4 = vsubq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_neon4() 2032 float32x4_t _tmp6 = vsubq_f32(_tmp56a, _tmp56b); in conv3x3s1_winograd64_neon4() 2083 float32x4_t _t_0_m_6 = vsubq_f32(_t_00, _t_66); in conv3x3s1_winograd64_neon4() 2084 float32x4_t _t_7_m_1 = vsubq_f32(_t_77, _t_11); in conv3x3s1_winograd64_neon4() 2086 float32x4_t _t_4_m_2 = vsubq_f32(_t_44, _t_22); in conv3x3s1_winograd64_neon4() [all …]
|
H A D | selu_arm.cpp | 55 _nps = vsubq_f32(_nps, _one); in forward_inplace()
|
/dports/graphics/vapoursynth-waifu2x-ncnn-vulkan/vapoursynth-waifu2x-ncnn-vulkan-r4/deps/ncnn/src/layer/arm/ |
H A D | convolution_3x3.h | 1985 float32x4_t _r_0_m_6 = vsubq_f32(_r_00, _r_66); in conv3x3s1_winograd64_neon4() 1986 float32x4_t _r_7_m_1 = vsubq_f32(_r_77, _r_11); in conv3x3s1_winograd64_neon4() 1988 float32x4_t _r_4_m_2 = vsubq_f32(_r_44, _r_22); in conv3x3s1_winograd64_neon4() 1989 float32x4_t _r_3_m_5 = vsubq_f32(_r_33, _r_55); in conv3x3s1_winograd64_neon4() 2004 float32x4_t _tmp2 = vsubq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_neon4() 2019 float32x4_t _tmp4 = vsubq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_neon4() 2032 float32x4_t _tmp6 = vsubq_f32(_tmp56a, _tmp56b); in conv3x3s1_winograd64_neon4() 2083 float32x4_t _t_0_m_6 = vsubq_f32(_t_00, _t_66); in conv3x3s1_winograd64_neon4() 2084 float32x4_t _t_7_m_1 = vsubq_f32(_t_77, _t_11); in conv3x3s1_winograd64_neon4() 2086 float32x4_t _t_4_m_2 = vsubq_f32(_t_44, _t_22); in conv3x3s1_winograd64_neon4() [all …]
|
H A D | selu_arm.cpp | 55 _nps = vsubq_f32(_nps, _one); in forward_inplace()
|
/dports/graphics/realsr-ncnn-vulkan/realsr-ncnn-vulkan-20210210/src/ncnn/src/layer/arm/ |
H A D | convolution_3x3.h | 1985 float32x4_t _r_0_m_6 = vsubq_f32(_r_00, _r_66); in conv3x3s1_winograd64_neon4() 1986 float32x4_t _r_7_m_1 = vsubq_f32(_r_77, _r_11); in conv3x3s1_winograd64_neon4() 1988 float32x4_t _r_4_m_2 = vsubq_f32(_r_44, _r_22); in conv3x3s1_winograd64_neon4() 1989 float32x4_t _r_3_m_5 = vsubq_f32(_r_33, _r_55); in conv3x3s1_winograd64_neon4() 2004 float32x4_t _tmp2 = vsubq_f32(_tmp12a, _tmp12b); in conv3x3s1_winograd64_neon4() 2019 float32x4_t _tmp4 = vsubq_f32(_tmp34a, _tmp34b); in conv3x3s1_winograd64_neon4() 2032 float32x4_t _tmp6 = vsubq_f32(_tmp56a, _tmp56b); in conv3x3s1_winograd64_neon4() 2083 float32x4_t _t_0_m_6 = vsubq_f32(_t_00, _t_66); in conv3x3s1_winograd64_neon4() 2084 float32x4_t _t_7_m_1 = vsubq_f32(_t_77, _t_11); in conv3x3s1_winograd64_neon4() 2086 float32x4_t _t_4_m_2 = vsubq_f32(_t_44, _t_22); in conv3x3s1_winograd64_neon4() [all …]
|
H A D | selu_arm.cpp | 55 _nps = vsubq_f32(_nps, _one); in forward_inplace()
|
/dports/graphics/opencv/opencv-4.5.3/3rdparty/carotene/src/ |
H A D | phase.cpp | 103 a = vbslq_f32(vcgeq_f32(ax, ay), a, vsubq_f32(_90, a)); \ 104 a = vbslq_f32(vcltq_f32(v_x, z), vsubq_f32(_180, a), a); \ 105 a = vbslq_f32(vcltq_f32(v_y, z), vsubq_f32(_360, a), a); \
|
H A D | remap.cpp | 148 float32x4_t v_dst = vmlaq_f32(v_dst_0, vsubq_f32(v_dst_1, v_dst_0), v_coeff.val[1]); in remapLinearReplicate() 161 v_dst = vmlaq_f32(v_dst_0, vsubq_f32(v_dst_1, v_dst_0), v_coeff.val[1]); in remapLinearReplicate() 246 float32x4_t v_dst = vmlaq_f32(v_dst_0, vsubq_f32(v_dst_1, v_dst_0), v_coeff.val[1]); in remapLinearConst() 259 v_dst = vmlaq_f32(v_dst_0, vsubq_f32(v_dst_1, v_dst_0), v_coeff.val[1]); in remapLinearConst() 538 v_coeff.val[0] = vsubq_f32(v_table.val[0], vcvtq_f32_s32(v_src_x)); in remapLinear() 539 v_coeff.val[1] = vsubq_f32(v_table.val[1], vcvtq_f32_s32(v_src_y)); in remapLinear() 620 v_coeff.val[0] = vsubq_f32(v_table.val[0], vcvtq_f32_s32(v_src_x0)); in remapLinear() 621 v_coeff.val[1] = vsubq_f32(v_table.val[1], vcvtq_f32_s32(v_src_y0)); in remapLinear()
|
/dports/science/gromacs/gromacs-2021.4/src/gromacs/simd/impl_arm_neon/ |
H A D | impl_arm_neon_util_float.h | 238 vst1q_f32(base + align * offset[0], vsubq_f32(vld1q_f32(base + align * offset[0]), t4)); in transposeScatterDecrU() 239 vst1q_f32(base + align * offset[1], vsubq_f32(vld1q_f32(base + align * offset[1]), t5)); in transposeScatterDecrU() 240 vst1q_f32(base + align * offset[2], vsubq_f32(vld1q_f32(base + align * offset[2]), t6)); in transposeScatterDecrU() 241 vst1q_f32(base + align * offset[3], vsubq_f32(vld1q_f32(base + align * offset[3]), t7)); in transposeScatterDecrU()
|
/dports/math/cglm/cglm-0.8.4/include/cglm/ |
H A D | vec4.h | 370 vst1q_f32(dest, vsubq_f32(vld1q_f32(a), vld1q_f32(b))); in glm_vec4_sub() 392 vst1q_f32(dest, vsubq_f32(vld1q_f32(v), vdupq_n_f32(s))); in glm_vec4_subs() 549 vsubq_f32(vld1q_f32(a), in glm_vec4_subadd() 752 return glmm_norm(vsubq_f32(glmm_load(a), glmm_load(b))); in glm_vec4_distance() 774 return glmm_norm2(vsubq_f32(glmm_load(a), glmm_load(b))); in glm_vec4_distance2()
|
/dports/games/OpenTomb/OpenTomb-win32-2018-02-03_alpha/extern/bullet/LinearMath/ |
H A D | btQuaternion.h | 177 mVec128 = vsubq_f32(mVec128, q.mVec128); 274 A0 = vsubq_f32(A0, A3); // AB03 = AB0 - AB3 480 return btQuaternion(vsubq_f32(mVec128, q2.mVec128)); 642 A0 = vsubq_f32(A0, A3); // AB03 = AB0 - AB3 729 A1 = vsubq_f32(A1, A3); // AB123 = AB12 - AB3 812 A1 = vsubq_f32(A1, A3); // AB123 = AB12 - AB3
|
/dports/devel/bullet/bullet3-3.21/src/Bullet3Common/ |
H A D | b3Quaternion.h | 201 mVec128 = vsubq_f32(mVec128, q.mVec128); 298 A0 = vsubq_f32(A0, A3); // AB03 = AB0 - AB3 478 return b3Quaternion(vsubq_f32(mVec128, q2.mVec128)); 633 A0 = vsubq_f32(A0, A3); // AB03 = AB0 - AB3 720 A1 = vsubq_f32(A1, A3); // AB123 = AB12 - AB3 803 A1 = vsubq_f32(A1, A3); // AB123 = AB12 - AB3
|
/dports/devel/bullet/bullet3-3.21/src/LinearMath/ |
H A D | btQuaternion.h | 222 mVec128 = vsubq_f32(mVec128, q.mVec128); 319 A0 = vsubq_f32(A0, A3); // AB03 = AB0 - AB3 531 return btQuaternion(vsubq_f32(mVec128, q2.mVec128)); 703 A0 = vsubq_f32(A0, A3); // AB03 = AB0 - AB3 790 A1 = vsubq_f32(A1, A3); // AB123 = AB12 - AB3 873 A1 = vsubq_f32(A1, A3); // AB123 = AB12 - AB3
|
/dports/graphics/blender/blender-2.91.0/extern/bullet2/src/LinearMath/ |
H A D | btQuaternion.h | 222 mVec128 = vsubq_f32(mVec128, q.mVec128); 319 A0 = vsubq_f32(A0, A3); // AB03 = AB0 - AB3 531 return btQuaternion(vsubq_f32(mVec128, q2.mVec128)); 703 A0 = vsubq_f32(A0, A3); // AB03 = AB0 - AB3 790 A1 = vsubq_f32(A1, A3); // AB123 = AB12 - AB3 873 A1 = vsubq_f32(A1, A3); // AB123 = AB12 - AB3
|
/dports/devel/godot-tools/godot-3.2.3-stable/thirdparty/bullet/LinearMath/ |
H A D | btQuaternion.h | 222 mVec128 = vsubq_f32(mVec128, q.mVec128); 319 A0 = vsubq_f32(A0, A3); // AB03 = AB0 - AB3 531 return btQuaternion(vsubq_f32(mVec128, q2.mVec128)); 703 A0 = vsubq_f32(A0, A3); // AB03 = AB0 - AB3 790 A1 = vsubq_f32(A1, A3); // AB123 = AB12 - AB3 873 A1 = vsubq_f32(A1, A3); // AB123 = AB12 - AB3
|
/dports/devel/godot-tools/godot-3.2.3-stable/thirdparty/bullet/Bullet3Common/ |
H A D | b3Quaternion.h | 201 mVec128 = vsubq_f32(mVec128, q.mVec128); 298 A0 = vsubq_f32(A0, A3); // AB03 = AB0 - AB3 478 return b3Quaternion(vsubq_f32(mVec128, q2.mVec128)); 633 A0 = vsubq_f32(A0, A3); // AB03 = AB0 - AB3 720 A1 = vsubq_f32(A1, A3); // AB123 = AB12 - AB3 803 A1 = vsubq_f32(A1, A3); // AB123 = AB12 - AB3
|
/dports/graphics/urho3d/Urho3D-1.7.1/Source/ThirdParty/Bullet/src/LinearMath/ |
H A D | btQuaternion.h | 184 mVec128 = vsubq_f32(mVec128, q.mVec128); 281 A0 = vsubq_f32(A0, A3); // AB03 = AB0 - AB3 486 return btQuaternion(vsubq_f32(mVec128, q2.mVec128)); 663 A0 = vsubq_f32(A0, A3); // AB03 = AB0 - AB3 750 A1 = vsubq_f32(A1, A3); // AB123 = AB12 - AB3 833 A1 = vsubq_f32(A1, A3); // AB123 = AB12 - AB3
|
/dports/devel/py-bullet3/bullet3-3.21/src/Bullet3Common/ |
H A D | b3Quaternion.h | 201 mVec128 = vsubq_f32(mVec128, q.mVec128); 298 A0 = vsubq_f32(A0, A3); // AB03 = AB0 - AB3 478 return b3Quaternion(vsubq_f32(mVec128, q2.mVec128)); 633 A0 = vsubq_f32(A0, A3); // AB03 = AB0 - AB3 720 A1 = vsubq_f32(A1, A3); // AB123 = AB12 - AB3 803 A1 = vsubq_f32(A1, A3); // AB123 = AB12 - AB3
|
/dports/devel/py-bullet3/bullet3-3.21/src/LinearMath/ |
H A D | btQuaternion.h | 222 mVec128 = vsubq_f32(mVec128, q.mVec128); 319 A0 = vsubq_f32(A0, A3); // AB03 = AB0 - AB3 531 return btQuaternion(vsubq_f32(mVec128, q2.mVec128)); 703 A0 = vsubq_f32(A0, A3); // AB03 = AB0 - AB3 790 A1 = vsubq_f32(A1, A3); // AB123 = AB12 - AB3 873 A1 = vsubq_f32(A1, A3); // AB123 = AB12 - AB3
|
/dports/devel/godot/godot-3.2.3-stable/thirdparty/bullet/Bullet3Common/ |
H A D | b3Quaternion.h | 201 mVec128 = vsubq_f32(mVec128, q.mVec128); 298 A0 = vsubq_f32(A0, A3); // AB03 = AB0 - AB3 478 return b3Quaternion(vsubq_f32(mVec128, q2.mVec128)); 633 A0 = vsubq_f32(A0, A3); // AB03 = AB0 - AB3 720 A1 = vsubq_f32(A1, A3); // AB123 = AB12 - AB3 803 A1 = vsubq_f32(A1, A3); // AB123 = AB12 - AB3
|