From 06a233c46482089db2c24494a881c2421a1d74da Mon Sep 17 00:00:00 2001 From: Artem Zyktin Date: Thu, 28 May 2026 13:24:38 +0800 Subject: [PATCH] Update operations --- src/tenvex/internal/add_impl.hpp | 1 - src/tenvex/internal/cross_impl.hpp | 2 +- src/tenvex/internal/dot.h | 18 +++-- src/tenvex/internal/dot_impl.hpp | 25 ++++++- src/tenvex/internal/expression.h | 2 +- src/tenvex/internal/magnitude.h | 5 +- src/tenvex/internal/magnitude_impl.hpp | 11 ++- src/tenvex/internal/mul.h | 4 ++ src/tenvex/internal/mul_impl.hpp | 7 ++ src/tenvex/internal/norm.h | 2 +- src/tenvex/internal/norm_impl.hpp | 19 ++--- src/tenvex/tenvex.h | 5 ++ src/tenvex/vec4.h | 37 ++++++++-- src/tenvex/vec4_impl.hpp | 96 ++++++++++++++++++++++++++ 14 files changed, 207 insertions(+), 27 deletions(-) create mode 100644 src/tenvex/vec4_impl.hpp diff --git a/src/tenvex/internal/add_impl.hpp b/src/tenvex/internal/add_impl.hpp index 0267546..c124370 100644 --- a/src/tenvex/internal/add_impl.hpp +++ b/src/tenvex/internal/add_impl.hpp @@ -18,7 +18,6 @@ TNVX_INLINE __m128 Add::eval() const noexcept { return _mm_add_ps(_l.eval(), _r.eval()); - } template diff --git a/src/tenvex/internal/cross_impl.hpp b/src/tenvex/internal/cross_impl.hpp index f0a45c6..4f723ca 100644 --- a/src/tenvex/internal/cross_impl.hpp +++ b/src/tenvex/internal/cross_impl.hpp @@ -38,7 +38,7 @@ __m128 Cross::eval() const noexcept } template -Cross cross(const Expr& l, const Expr& r) noexcept +Cross cross(const Expr& l, const Expr& r) noexcept { return { l.self(), r.self() }; } diff --git a/src/tenvex/internal/dot.h b/src/tenvex/internal/dot.h index 974067b..1727731 100644 --- a/src/tenvex/internal/dot.h +++ b/src/tenvex/internal/dot.h @@ -1,15 +1,12 @@ #pragma once #include "expression.h" -#include "scalar.h" -#include "../vec4.h" namespace tnvx { template -requires (!std::same_as && !std::same_as) -struct Dot : Expr +struct Dot : Expr> { inline Dot(const L& TNVX_RESTRICT l, const R& TNVX_RESTRICT r) noexcept; @@ -17,6 +14,9 @@ struct Dot : Expr [[nodiscard]] inline __m128 eval() const noexcept; + [[nodiscard]] inline + operator float() const noexcept; + private: tnvx_type_storage _l; tnvx_type_storage _r; @@ -24,7 +24,15 @@ private: template [[nodiscard]] inline -Dot dot(const Expr& l, const Expr& r) noexcept; +Dot dot(const Expr& l, const Expr& r) noexcept; + +template +[[nodiscard]] inline +float operator*(const Dot& l, float r) noexcept; + +template +[[nodiscard]] inline +float operator*(float l, const Dot& r) noexcept; } diff --git a/src/tenvex/internal/dot_impl.hpp b/src/tenvex/internal/dot_impl.hpp index ebbd27f..bc3bce5 100644 --- a/src/tenvex/internal/dot_impl.hpp +++ b/src/tenvex/internal/dot_impl.hpp @@ -29,9 +29,32 @@ __m128 Dot::eval() const noexcept } template -Dot dot(const Expr& l, const Expr& r) noexcept +TNVX_INLINE +Dot::operator float() const noexcept +{ + return _mm_cvtss_f32(eval()); +} + +template +TNVX_INLINE +Dot dot(const Expr& l, const Expr& r) noexcept { return { l.self(), r.self() }; } +template +TNVX_INLINE +float operator*(const Dot& l, float r) noexcept +{ + return static_cast(l) * r; +} + +template +float operator*(float l, const Dot& r) noexcept +{ + return l * static_cast(r); +} + + + } diff --git a/src/tenvex/internal/expression.h b/src/tenvex/internal/expression.h index 14cccd5..4c3a61e 100644 --- a/src/tenvex/internal/expression.h +++ b/src/tenvex/internal/expression.h @@ -29,7 +29,7 @@ concept expression = requires(T a) } && std::derived_from>; -class Scalar; +struct Scalar; template concept vector_expression = expression && !std::same_as; diff --git a/src/tenvex/internal/magnitude.h b/src/tenvex/internal/magnitude.h index 8a30ffa..678c214 100644 --- a/src/tenvex/internal/magnitude.h +++ b/src/tenvex/internal/magnitude.h @@ -14,13 +14,16 @@ struct Magnitude : Expr> [[nodiscard]] inline __m128 eval() const noexcept; + [[nodiscard]] inline + operator float() const noexcept; + private: tnvx_type_storage _e; }; template [[nodiscard]] inline -Magnitude magnitude(const Expr e) noexcept; +Magnitude magnitude(const Expr& e) noexcept; } diff --git a/src/tenvex/internal/magnitude_impl.hpp b/src/tenvex/internal/magnitude_impl.hpp index 5d9db46..ce2723a 100644 --- a/src/tenvex/internal/magnitude_impl.hpp +++ b/src/tenvex/internal/magnitude_impl.hpp @@ -17,12 +17,19 @@ template TNVX_INLINE __m128 Magnitude::eval() const noexcept { - return _mm_sqrt_ps(Dot(_e, _e).eval()); + return _mm_sqrt_ss(Dot(_e, _e).eval()); } template TNVX_INLINE -Magnitude magnitude(const Expr e) noexcept +Magnitude::operator float() const noexcept +{ + return _mm_cvtss_f32(eval()); +} + +template +TNVX_INLINE +Magnitude magnitude(const Expr& e) noexcept { return { e.self() }; } diff --git a/src/tenvex/internal/mul.h b/src/tenvex/internal/mul.h index 50b35ff..3d08947 100644 --- a/src/tenvex/internal/mul.h +++ b/src/tenvex/internal/mul.h @@ -20,6 +20,10 @@ private: tnvx_type_storage _r; }; +//template +//[[nodiscard]] inline +//Mul operator*(const Expr& l, const Expr& r) noexcept; + template [[nodiscard]] inline Mul operator*(const Expr& l, float r) noexcept; diff --git a/src/tenvex/internal/mul_impl.hpp b/src/tenvex/internal/mul_impl.hpp index 60f31a9..8e6c50f 100644 --- a/src/tenvex/internal/mul_impl.hpp +++ b/src/tenvex/internal/mul_impl.hpp @@ -20,6 +20,13 @@ __m128 Mul::eval() const noexcept return _mm_mul_ps(_l.eval(), _r.eval()); } +//template +//TNVX_INLINE +//Mul operator*(const Expr& l, const Expr& r) noexcept +//{ +// return { l.self(), r.self() }; +//} + template TNVX_INLINE Mul operator*(const Expr& l, float r) noexcept diff --git a/src/tenvex/internal/norm.h b/src/tenvex/internal/norm.h index 8882029..dd0432d 100644 --- a/src/tenvex/internal/norm.h +++ b/src/tenvex/internal/norm.h @@ -20,7 +20,7 @@ private: template [[nodiscard]] inline -Norm norm(const Expr e) noexcept; +Norm norm(const Expr& e) noexcept; } diff --git a/src/tenvex/internal/norm_impl.hpp b/src/tenvex/internal/norm_impl.hpp index 215e9bf..d4af233 100644 --- a/src/tenvex/internal/norm_impl.hpp +++ b/src/tenvex/internal/norm_impl.hpp @@ -17,18 +17,19 @@ template TNVX_INLINE __m128 Norm::eval() const noexcept { - __m128 data = _e.eval(); - __m128 mul = _mm_mul_ps(data, data); - __m128 shuf = _mm_shuffle_ps(mul, mul, _MM_SHUFFLE(2, 3, 0, 1)); - __m128 sum = _mm_add_ps(mul, shuf); - shuf = _mm_movehl_ps(shuf, sum); - sum = _mm_add_ss(sum, shuf); - __m128 len = _mm_shuffle_ps(sum, sum, 0); - return _mm_div_ps(data, len); + __m128 data = _e.eval(); + __m128 mul = _mm_mul_ps(data, data); + __m128 shuf = _mm_shuffle_ps(mul, mul, _MM_SHUFFLE(2, 3, 0, 1)); + __m128 sum = _mm_add_ps(mul, shuf); + shuf = _mm_movehl_ps(shuf, sum); + sum = _mm_add_ss(sum, shuf); + __m128 len = _mm_shuffle_ps(sum, sum, 0); + __m128 sqrt_len = _mm_sqrt_ps(len); + return _mm_div_ps(data, sqrt_len); } template -Norm norm(const Expr e) noexcept +Norm norm(const Expr& e) noexcept { return { e.self() }; } diff --git a/src/tenvex/tenvex.h b/src/tenvex/tenvex.h index e891d73..c10d5f1 100644 --- a/src/tenvex/tenvex.h +++ b/src/tenvex/tenvex.h @@ -5,3 +5,8 @@ #include "internal/add.h" #include "internal/sub.h" #include "internal/mul.h" +#include "internal/dot.h" +#include "internal/cross.h" +#include "internal/norm.h" +#include "internal/magnitude.h" +#include "internal/scalar.h" diff --git a/src/tenvex/vec4.h b/src/tenvex/vec4.h index 98560d5..9554cd1 100644 --- a/src/tenvex/vec4.h +++ b/src/tenvex/vec4.h @@ -4,33 +4,60 @@ #include +#include + namespace tnvx { -struct alignas(16) vec4 : public Expr +struct alignas(16) vec4 : Expr { vec4() = default; - vec4(float x, float y, float z, float w); + inline + vec4(float x, float y, float z, float w) noexcept; template - vec4(const Expr& expr); + inline + vec4(const Expr& expr) noexcept; template - vec4& operator=(const Expr& expr); + inline + vec4& operator=(const Expr& expr) noexcept; + [[nodiscard]] inline __m128 eval() const noexcept; + [[nodiscard]] inline float magnitude() const noexcept; + + [[nodiscard]] inline vec4 norm() const noexcept; + [[nodiscard]] inline float x() const noexcept; + + [[nodiscard]] inline float y() const noexcept; + + [[nodiscard]] inline float z() const noexcept; + + [[nodiscard]] inline float w() const noexcept; + [[nodiscard]] inline + std::array data() const noexcept; + private: __m128 _data; }; -} // namespace tenvex +inline +bool operator==(const vec4& a, const vec4& b) noexcept; + +inline +bool approx_eq(const vec4& a, const vec4& b, float eps = 1e-3f) noexcept; + +} + +#include "vec4_impl.hpp" diff --git a/src/tenvex/vec4_impl.hpp b/src/tenvex/vec4_impl.hpp new file mode 100644 index 0000000..7078975 --- /dev/null +++ b/src/tenvex/vec4_impl.hpp @@ -0,0 +1,96 @@ +#pragma once + +#include "vec4.h" +#include "internal/magnitude.h" +#include "internal/norm.h" + +namespace tnvx +{ + +template +TNVX_INLINE +vec4::vec4(const Expr& expr) noexcept + : _data(expr.eval()) +{ +} + +template +TNVX_INLINE +vec4& vec4::operator=(const Expr& expr) noexcept +{ + _data = expr.eval(); + return *this; +} + +TNVX_INLINE +vec4::vec4(float x, float y, float z, float w) noexcept +{ + _data = _mm_set_ps(w, z, y, x); +} + +TNVX_INLINE +__m128 vec4::eval() const noexcept +{ + return _data; +} + +TNVX_INLINE +float vec4::magnitude() const noexcept +{ + return _mm_cvtss_f32(Magnitude(*this).eval()); +} + +TNVX_INLINE +vec4 vec4::norm() const noexcept +{ + return { Norm(*this) }; +} + +TNVX_INLINE +float vec4::x() const noexcept +{ + return _mm_cvtss_f32(_data); +} + +TNVX_INLINE +float vec4::y() const noexcept +{ + return _mm_cvtss_f32(_mm_shuffle_ps(_data, _data, _MM_SHUFFLE(1, 1, 1, 1))); +} + +TNVX_INLINE +float vec4::z() const noexcept +{ + return _mm_cvtss_f32(_mm_shuffle_ps(_data, _data, _MM_SHUFFLE(2, 2, 2, 2))); +} + +TNVX_INLINE +float vec4::w() const noexcept +{ + return _mm_cvtss_f32(_mm_shuffle_ps(_data, _data, _MM_SHUFFLE(3, 3, 3, 3))); +} + +TNVX_INLINE +std::array vec4::data() const noexcept +{ + std::array result; + _mm_storeu_ps(result.data(), _data); + return result; +} + +TNVX_INLINE +bool operator==(const vec4& a, const vec4& b) noexcept +{ + return _mm_movemask_ps(_mm_cmpeq_ps(a.eval(), b.eval())) == 0xF; +} + +TNVX_INLINE +bool approx_eq(const vec4& a, const vec4& b, float eps) noexcept +{ + __m128 epsilon = _mm_set1_ps(eps); + __m128 diff = _mm_sub_ps(a.eval(), b.eval()); + __m128 abs_diff = _mm_andnot_ps(_mm_set1_ps(-0.0f), diff); + return _mm_movemask_ps(_mm_cmplt_ps(abs_diff, epsilon)) == 0xF; +} + +}