From e3ae7e3db56dcbe784dc481959e57f1adc0c47e5 Mon Sep 17 00:00:00 2001 From: Christophe Riccio Date: Wed, 5 Jan 2011 16:56:07 +0000 Subject: [PATCH] Added determinant SSE draft --- glm/core/intrinsic_matrix.inl | 313 ++++++++++++++++++++++++++++++++++ 1 file changed, 313 insertions(+) diff --git a/glm/core/intrinsic_matrix.inl b/glm/core/intrinsic_matrix.inl index 367b7929..548772e4 100644 --- a/glm/core/intrinsic_matrix.inl +++ b/glm/core/intrinsic_matrix.inl @@ -12,6 +12,20 @@ static const __m128 pi = _mm_set_ps1(3.141592653589793238462643383279f); static const __m128 _m128_rad_ps = _mm_set_ps1(3.141592653589793238462643383279f / 180.f); static const __m128 _m128_deg_ps = _mm_set_ps1(180.f / 3.141592653589793238462643383279f); +template +inline matType _mm_comp_mul_ps +( + __m128 const in1[4], + __m128 const in2[4], + __m128 out[4] +) +{ + out[0] = _mm_mul_ps(in1[0], in2[0]); + out[1] = _mm_mul_ps(in1[1], in2[1]); + out[2] = _mm_mul_ps(in1[2], in2[2]); + out[3] = _mm_mul_ps(in1[3], in2[3]); +} + inline void _mm_add_ps(__m128 in1[4], __m128 in2[4], __m128 out[4]) { { @@ -167,6 +181,297 @@ inline void _mm_transpose_ps(__m128 const in[4], __m128 out[4]) out[3] = _mm_shuffle_ps(tmp2, tmp3, 0xDD); } +inline __m128 _mm_det_ps(__m128 const in[4]) +{ + __m128 Fac0; + { + // valType SubFactor00 = m[2][2] * m[3][3] - m[3][2] * m[2][3]; + // valType SubFactor00 = m[2][2] * m[3][3] - m[3][2] * m[2][3]; + // valType SubFactor06 = m[1][2] * m[3][3] - m[3][2] * m[1][3]; + // valType SubFactor13 = m[1][2] * m[2][3] - m[2][2] * m[1][3]; + + __m128 Swp0a = _mm_shuffle_ps(in[3], in[2], _MM_SHUFFLE(3, 3, 3, 3)); + __m128 Swp0b = _mm_shuffle_ps(in[3], in[2], _MM_SHUFFLE(2, 2, 2, 2)); + + __m128 Swp00 = _mm_shuffle_ps(in[2], in[1], _MM_SHUFFLE(2, 2, 2, 2)); + __m128 Swp01 = _mm_shuffle_ps(Swp0a, Swp0a, _MM_SHUFFLE(2, 0, 0, 0)); + __m128 Swp02 = _mm_shuffle_ps(Swp0b, Swp0b, _MM_SHUFFLE(2, 0, 0, 0)); + __m128 Swp03 = _mm_shuffle_ps(in[2], in[1], _MM_SHUFFLE(3, 3, 3, 3)); + + __m128 Mul00 = _mm_mul_ps(Swp00, Swp01); + __m128 Mul01 = _mm_mul_ps(Swp02, Swp03); + Fac0 = _mm_sub_ps(Mul00, Mul01); + + bool stop = true; + } + + __m128 Fac1; + { + // valType SubFactor01 = m[2][1] * m[3][3] - m[3][1] * m[2][3]; + // valType SubFactor01 = m[2][1] * m[3][3] - m[3][1] * m[2][3]; + // valType SubFactor07 = m[1][1] * m[3][3] - m[3][1] * m[1][3]; + // valType SubFactor14 = m[1][1] * m[2][3] - m[2][1] * m[1][3]; + + __m128 Swp0a = _mm_shuffle_ps(in[3], in[2], _MM_SHUFFLE(3, 3, 3, 3)); + __m128 Swp0b = _mm_shuffle_ps(in[3], in[2], _MM_SHUFFLE(1, 1, 1, 1)); + + __m128 Swp00 = _mm_shuffle_ps(in[2], in[1], _MM_SHUFFLE(1, 1, 1, 1)); + __m128 Swp01 = _mm_shuffle_ps(Swp0a, Swp0a, _MM_SHUFFLE(2, 0, 0, 0)); + __m128 Swp02 = _mm_shuffle_ps(Swp0b, Swp0b, _MM_SHUFFLE(2, 0, 0, 0)); + __m128 Swp03 = _mm_shuffle_ps(in[2], in[1], _MM_SHUFFLE(3, 3, 3, 3)); + + __m128 Mul00 = _mm_mul_ps(Swp00, Swp01); + __m128 Mul01 = _mm_mul_ps(Swp02, Swp03); + Fac1 = _mm_sub_ps(Mul00, Mul01); + + bool stop = true; + } + + + __m128 Fac2; + { + // valType SubFactor02 = m[2][1] * m[3][2] - m[3][1] * m[2][2]; + // valType SubFactor02 = m[2][1] * m[3][2] - m[3][1] * m[2][2]; + // valType SubFactor08 = m[1][1] * m[3][2] - m[3][1] * m[1][2]; + // valType SubFactor15 = m[1][1] * m[2][2] - m[2][1] * m[1][2]; + + __m128 Swp0a = _mm_shuffle_ps(in[3], in[2], _MM_SHUFFLE(2, 2, 2, 2)); + __m128 Swp0b = _mm_shuffle_ps(in[3], in[2], _MM_SHUFFLE(1, 1, 1, 1)); + + __m128 Swp00 = _mm_shuffle_ps(in[2], in[1], _MM_SHUFFLE(1, 1, 1, 1)); + __m128 Swp01 = _mm_shuffle_ps(Swp0a, Swp0a, _MM_SHUFFLE(2, 0, 0, 0)); + __m128 Swp02 = _mm_shuffle_ps(Swp0b, Swp0b, _MM_SHUFFLE(2, 0, 0, 0)); + __m128 Swp03 = _mm_shuffle_ps(in[2], in[1], _MM_SHUFFLE(2, 2, 2, 2)); + + __m128 Mul00 = _mm_mul_ps(Swp00, Swp01); + __m128 Mul01 = _mm_mul_ps(Swp02, Swp03); + Fac2 = _mm_sub_ps(Mul00, Mul01); + + bool stop = true; + } + + __m128 Fac3; + { + // valType SubFactor03 = m[2][0] * m[3][3] - m[3][0] * m[2][3]; + // valType SubFactor03 = m[2][0] * m[3][3] - m[3][0] * m[2][3]; + // valType SubFactor09 = m[1][0] * m[3][3] - m[3][0] * m[1][3]; + // valType SubFactor16 = m[1][0] * m[2][3] - m[2][0] * m[1][3]; + + __m128 Swp0a = _mm_shuffle_ps(in[3], in[2], _MM_SHUFFLE(3, 3, 3, 3)); + __m128 Swp0b = _mm_shuffle_ps(in[3], in[2], _MM_SHUFFLE(0, 0, 0, 0)); + + __m128 Swp00 = _mm_shuffle_ps(in[2], in[1], _MM_SHUFFLE(0, 0, 0, 0)); + __m128 Swp01 = _mm_shuffle_ps(Swp0a, Swp0a, _MM_SHUFFLE(2, 0, 0, 0)); + __m128 Swp02 = _mm_shuffle_ps(Swp0b, Swp0b, _MM_SHUFFLE(2, 0, 0, 0)); + __m128 Swp03 = _mm_shuffle_ps(in[2], in[1], _MM_SHUFFLE(3, 3, 3, 3)); + + __m128 Mul00 = _mm_mul_ps(Swp00, Swp01); + __m128 Mul01 = _mm_mul_ps(Swp02, Swp03); + Fac3 = _mm_sub_ps(Mul00, Mul01); + + bool stop = true; + } + + __m128 Fac4; + { + // valType SubFactor04 = m[2][0] * m[3][2] - m[3][0] * m[2][2]; + // valType SubFactor04 = m[2][0] * m[3][2] - m[3][0] * m[2][2]; + // valType SubFactor10 = m[1][0] * m[3][2] - m[3][0] * m[1][2]; + // valType SubFactor17 = m[1][0] * m[2][2] - m[2][0] * m[1][2]; + + __m128 Swp0a = _mm_shuffle_ps(in[3], in[2], _MM_SHUFFLE(2, 2, 2, 2)); + __m128 Swp0b = _mm_shuffle_ps(in[3], in[2], _MM_SHUFFLE(0, 0, 0, 0)); + + __m128 Swp00 = _mm_shuffle_ps(in[2], in[1], _MM_SHUFFLE(0, 0, 0, 0)); + __m128 Swp01 = _mm_shuffle_ps(Swp0a, Swp0a, _MM_SHUFFLE(2, 0, 0, 0)); + __m128 Swp02 = _mm_shuffle_ps(Swp0b, Swp0b, _MM_SHUFFLE(2, 0, 0, 0)); + __m128 Swp03 = _mm_shuffle_ps(in[2], in[1], _MM_SHUFFLE(2, 2, 2, 2)); + + __m128 Mul00 = _mm_mul_ps(Swp00, Swp01); + __m128 Mul01 = _mm_mul_ps(Swp02, Swp03); + Fac4 = _mm_sub_ps(Mul00, Mul01); + + bool stop = true; + } + + __m128 Fac5; + { + // valType SubFactor05 = m[2][0] * m[3][1] - m[3][0] * m[2][1]; + // valType SubFactor05 = m[2][0] * m[3][1] - m[3][0] * m[2][1]; + // valType SubFactor12 = m[1][0] * m[3][1] - m[3][0] * m[1][1]; + // valType SubFactor18 = m[1][0] * m[2][1] - m[2][0] * m[1][1]; + + __m128 Swp0a = _mm_shuffle_ps(in[3], in[2], _MM_SHUFFLE(1, 1, 1, 1)); + __m128 Swp0b = _mm_shuffle_ps(in[3], in[2], _MM_SHUFFLE(0, 0, 0, 0)); + + __m128 Swp00 = _mm_shuffle_ps(in[2], in[1], _MM_SHUFFLE(0, 0, 0, 0)); + __m128 Swp01 = _mm_shuffle_ps(Swp0a, Swp0a, _MM_SHUFFLE(2, 0, 0, 0)); + __m128 Swp02 = _mm_shuffle_ps(Swp0b, Swp0b, _MM_SHUFFLE(2, 0, 0, 0)); + __m128 Swp03 = _mm_shuffle_ps(in[2], in[1], _MM_SHUFFLE(1, 1, 1, 1)); + + __m128 Mul00 = _mm_mul_ps(Swp00, Swp01); + __m128 Mul01 = _mm_mul_ps(Swp02, Swp03); + Fac5 = _mm_sub_ps(Mul00, Mul01); + + bool stop = true; + } + + __m128 SignA = _mm_set_ps( 1.0f,-1.0f, 1.0f,-1.0f); + __m128 SignB = _mm_set_ps(-1.0f, 1.0f,-1.0f, 1.0f); + + // m[1][0] + // m[0][0] + // m[0][0] + // m[0][0] + __m128 Temp0 = _mm_shuffle_ps(in[1], in[0], _MM_SHUFFLE(0, 0, 0, 0)); + __m128 Vec0 = _mm_shuffle_ps(Temp0, Temp0, _MM_SHUFFLE(2, 2, 2, 0)); + + // m[1][1] + // m[0][1] + // m[0][1] + // m[0][1] + __m128 Temp1 = _mm_shuffle_ps(in[1], in[0], _MM_SHUFFLE(1, 1, 1, 1)); + __m128 Vec1 = _mm_shuffle_ps(Temp1, Temp1, _MM_SHUFFLE(2, 2, 2, 0)); + + // m[1][2] + // m[0][2] + // m[0][2] + // m[0][2] + __m128 Temp2 = _mm_shuffle_ps(in[1], in[0], _MM_SHUFFLE(2, 2, 2, 2)); + __m128 Vec2 = _mm_shuffle_ps(Temp2, Temp2, _MM_SHUFFLE(2, 2, 2, 0)); + + // m[1][3] + // m[0][3] + // m[0][3] + // m[0][3] + __m128 Temp3 = _mm_shuffle_ps(in[1], in[0], _MM_SHUFFLE(3, 3, 3, 3)); + __m128 Vec3 = _mm_shuffle_ps(Temp3, Temp3, _MM_SHUFFLE(2, 2, 2, 0)); + + // col0 + // + (Vec1[0] * Fac0[0] - Vec2[0] * Fac1[0] + Vec3[0] * Fac2[0]), + // - (Vec1[1] * Fac0[1] - Vec2[1] * Fac1[1] + Vec3[1] * Fac2[1]), + // + (Vec1[2] * Fac0[2] - Vec2[2] * Fac1[2] + Vec3[2] * Fac2[2]), + // - (Vec1[3] * Fac0[3] - Vec2[3] * Fac1[3] + Vec3[3] * Fac2[3]), + __m128 Mul00 = _mm_mul_ps(Vec1, Fac0); + __m128 Mul01 = _mm_mul_ps(Vec2, Fac1); + __m128 Mul02 = _mm_mul_ps(Vec3, Fac2); + __m128 Sub00 = _mm_sub_ps(Mul00, Mul01); + __m128 Add00 = _mm_add_ps(Sub00, Mul02); + __m128 Inv0 = _mm_mul_ps(SignB, Add00); + + // col1 + // - (Vec0[0] * Fac0[0] - Vec2[0] * Fac3[0] + Vec3[0] * Fac4[0]), + // + (Vec0[0] * Fac0[1] - Vec2[1] * Fac3[1] + Vec3[1] * Fac4[1]), + // - (Vec0[0] * Fac0[2] - Vec2[2] * Fac3[2] + Vec3[2] * Fac4[2]), + // + (Vec0[0] * Fac0[3] - Vec2[3] * Fac3[3] + Vec3[3] * Fac4[3]), + __m128 Mul03 = _mm_mul_ps(Vec0, Fac0); + __m128 Mul04 = _mm_mul_ps(Vec2, Fac3); + __m128 Mul05 = _mm_mul_ps(Vec3, Fac4); + __m128 Sub01 = _mm_sub_ps(Mul03, Mul04); + __m128 Add01 = _mm_add_ps(Sub01, Mul05); + __m128 Inv1 = _mm_mul_ps(SignA, Add01); + + // col2 + // + (Vec0[0] * Fac1[0] - Vec1[0] * Fac3[0] + Vec3[0] * Fac5[0]), + // - (Vec0[0] * Fac1[1] - Vec1[1] * Fac3[1] + Vec3[1] * Fac5[1]), + // + (Vec0[0] * Fac1[2] - Vec1[2] * Fac3[2] + Vec3[2] * Fac5[2]), + // - (Vec0[0] * Fac1[3] - Vec1[3] * Fac3[3] + Vec3[3] * Fac5[3]), + __m128 Mul06 = _mm_mul_ps(Vec0, Fac1); + __m128 Mul07 = _mm_mul_ps(Vec1, Fac3); + __m128 Mul08 = _mm_mul_ps(Vec3, Fac5); + __m128 Sub02 = _mm_sub_ps(Mul06, Mul07); + __m128 Add02 = _mm_add_ps(Sub02, Mul08); + __m128 Inv2 = _mm_mul_ps(SignB, Add02); + + // col3 + // - (Vec1[0] * Fac2[0] - Vec1[0] * Fac4[0] + Vec2[0] * Fac5[0]), + // + (Vec1[0] * Fac2[1] - Vec1[1] * Fac4[1] + Vec2[1] * Fac5[1]), + // - (Vec1[0] * Fac2[2] - Vec1[2] * Fac4[2] + Vec2[2] * Fac5[2]), + // + (Vec1[0] * Fac2[3] - Vec1[3] * Fac4[3] + Vec2[3] * Fac5[3])); + __m128 Mul09 = _mm_mul_ps(Vec0, Fac2); + __m128 Mul10 = _mm_mul_ps(Vec1, Fac4); + __m128 Mul11 = _mm_mul_ps(Vec2, Fac5); + __m128 Sub03 = _mm_sub_ps(Mul09, Mul10); + __m128 Add03 = _mm_add_ps(Sub03, Mul11); + __m128 Inv3 = _mm_mul_ps(SignA, Add03); + + __m128 Row0 = _mm_shuffle_ps(Inv0, Inv1, _MM_SHUFFLE(0, 0, 0, 0)); + __m128 Row1 = _mm_shuffle_ps(Inv2, Inv3, _MM_SHUFFLE(0, 0, 0, 0)); + __m128 Row2 = _mm_shuffle_ps(Row0, Row1, _MM_SHUFFLE(2, 0, 2, 0)); + + // valType Determinant = m[0][0] * Inverse[0][0] + // + m[0][1] * Inverse[1][0] + // + m[0][2] * Inverse[2][0] + // + m[0][3] * Inverse[3][0]; + __m128 Det0 = _mm_dot_ps(in[0], Row2); + return Det0; +} + +template +inline typename detail::tmat4x4::value_type _mm_det2_ps +( + __m128 const & m[4] +) +{ + GLM_STATIC_ASSERT(detail::type::is_float, "'determinant' only accept floating-point inputs"); + + //T SubFactor00 = m[2][2] * m[3][3] - m[3][2] * m[2][3]; + //T SubFactor01 = m[2][1] * m[3][3] - m[3][1] * m[2][3]; + //T SubFactor02 = m[2][1] * m[3][2] - m[3][1] * m[2][2]; + //T SubFactor03 = m[2][0] * m[3][3] - m[3][0] * m[2][3]; + //T SubFactor04 = m[2][0] * m[3][2] - m[3][0] * m[2][2]; + //T SubFactor05 = m[2][0] * m[3][1] - m[3][0] * m[2][1]; + + // First 2 columns + __m128 Swp2A = _mm_shuffle_ps(m[2], m[2], _MM_SHUFFLE(0, 1, 1, 2)); + __m128 Swp3A = _mm_shuffle_ps(m[3], m[3], _MM_SHUFFLE(3, 2, 3, 3)); + __m128 MulA = __mm_mul_ps(Swp2A, Swp3A); + + // Second 2 columns + __m128 Swp2B = _mm_shuffle_ps(m[2], m[2], _MM_SHUFFLE(0, 1, 1, 2)); + __m128 Swp3B = _mm_shuffle_ps(m[3], m[3], _MM_SHUFFLE(3, 2, 3, 3)); + __m128 MulB = __mm_mul_ps(Swp2A, Swp3A); + + // Columns subtraction + __m128 SubAB = __mm_sub_ps(MulA, MulB); + + // Last 2 rows + __m128 Swp2C = _mm_shuffle_ps(m[2], m[2], _MM_SHUFFLE(1, 2, 0, 0)); + __m128 Swp3C = _mm_shuffle_ps(m[3], m[3], _MM_SHUFFLE(0, 0, 1, 2)); + __m128 MulC = __mm_mul_ps(Swp2C, Swp3C); + __m128 SwpD = __mm_hl_ps(MulC); + + //detail::tvec4 DetCof( + // + (m[1][1] * SubFactor00 - m[1][2] * SubFactor01 + m[1][3] * SubFactor02), + // - (m[1][0] * SubFactor00 - m[1][2] * SubFactor03 + m[1][3] * SubFactor04), + // + (m[1][0] * SubFactor01 - m[1][1] * SubFactor03 + m[1][3] * SubFactor05), + // - (m[1][0] * SubFactor02 - m[1][1] * SubFactor04 + m[1][2] * SubFactor05)); + + __128 SubFacA = _mm_shuffle_ps(SubAB, SubAB, _MM_SHUFFLE(2, 1, 0, 0)); + __128 SwpFacA = _mm_shuffle_ps(m[1], m[1], _MM_SHUFFLE(0, 0, 0, 1)); + __128 MulFacA = __mm_mul_ps(SwpFacA, SubFacA); + + __128 SubFacB = ; + __128 SwpFacB = _mm_shuffle_ps(m[1], m[1], _MM_SHUFFLE(1, 1, 2, 2)); + __128 MulFacB = __mm_mul_ps(SwpFacB, SubFacB); + + __128 SubRes = __mm_sub_ps(MulFacA, MulFacA); + + __128 SubFacC = ; + __128 SwpFacC = _mm_shuffle_ps(m[1], m[1], _MM_SHUFFLE(1, 1, 2, 2)); + __128 MulFacC = __mm_mul_ps(SwpFacC, SubFacC); + + __m128 AddRes = __mm_add_ps(SubRes, MulFacC); + __m128 DetCof = __mm_mul_ps(AddRes, _mm_set_ps( 1.0f,-1.0f, 1.0f,-1.0f)); + + //return m[0][0] * DetCof[0] + // + m[0][1] * DetCof[1] + // + m[0][2] * DetCof[2] + // + m[0][3] * DetCof[3]; + + return _mm_dot_ps(m[0], Signed); +} + inline void _mm_inverse_ps(__m128 const in[4], __m128 out[4]) { __m128 Fac0; @@ -702,3 +1007,11 @@ void _mm_rotate_ps(__m128 const in[4], float Angle, float const v[3], __m128 out //return Result; _mm_mul_ps(in, Result, out); } + +void _mm_outer_ps(__m128 const & c, __m128 const & r, __m128 out[4]) +{ + out[0] = _mm_mul_ps(c, _mm_shuffle_ps(r, r, _MM_SHUFFLE(0, 0, 0, 0)); + out[1] = _mm_mul_ps(c, _mm_shuffle_ps(r, r, _MM_SHUFFLE(1, 1, 1, 1)); + out[2] = _mm_mul_ps(c, _mm_shuffle_ps(r, r, _MM_SHUFFLE(2, 2, 2, 2)); + out[3] = _mm_mul_ps(c, _mm_shuffle_ps(r, r, _MM_SHUFFLE(3, 3, 3, 3)); +}