From 55357750066b44b196ebc4b826bb6ff1cd5e6939 Mon Sep 17 00:00:00 2001
From: Nathan Vegdahl <cessen@cessen.com>
Date: Fri, 15 Jul 2022 15:20:44 -0700
Subject: [PATCH] RMath: implement Bool4 type.

---
 sub_crates/rmath/src/wide4.rs | 162 +++++++++++++++++++++++++++++++++-
 1 file changed, 158 insertions(+), 4 deletions(-)

diff --git a/sub_crates/rmath/src/wide4.rs b/sub_crates/rmath/src/wide4.rs
index 9cb9e8e..e846753 100644
--- a/sub_crates/rmath/src/wide4.rs
+++ b/sub_crates/rmath/src/wide4.rs
@@ -1,12 +1,14 @@
-use std::ops::{AddAssign, DivAssign, MulAssign, SubAssign};
+use std::ops::{
+    AddAssign, BitAndAssign, BitOrAssign, BitXorAssign, DivAssign, MulAssign, SubAssign,
+};
 
 use approx::ulps_eq;
 
 use crate::{difference_of_products, two_prod, two_sum};
 
-pub use fallback::Float4;
+pub use fallback::{Bool4, Float4};
 mod fallback {
-    use std::ops::{Add, Div, Mul, Neg, Sub};
+    use std::ops::{Add, BitAnd, BitOr, BitXor, Div, Mul, Neg, Not, Sub};
 
     use crate::FMulAdd;
 
@@ -78,6 +80,64 @@ mod fallback {
         //     a.max(b)
         // }
 
+        //-----------------------------------------------------
+        // Comparisons.
+
+        /// Less than.
+        #[inline(always)]
+        pub fn lt(self, rhs: Self) -> Bool4 {
+            Bool4([
+                self.0[0] < rhs.0[0],
+                self.0[1] < rhs.0[1],
+                self.0[2] < rhs.0[2],
+                self.0[3] < rhs.0[3],
+            ])
+        }
+
+        /// Less than or equal.
+        #[inline(always)]
+        pub fn lte(self, rhs: Self) -> Bool4 {
+            Bool4([
+                self.0[0] <= rhs.0[0],
+                self.0[1] <= rhs.0[1],
+                self.0[2] <= rhs.0[2],
+                self.0[3] <= rhs.0[3],
+            ])
+        }
+
+        /// Greater than.
+        #[inline(always)]
+        pub fn gt(self, rhs: Self) -> Bool4 {
+            Bool4([
+                self.0[0] > rhs.0[0],
+                self.0[1] > rhs.0[1],
+                self.0[2] > rhs.0[2],
+                self.0[3] > rhs.0[3],
+            ])
+        }
+
+        /// Greater than or equal.
+        #[inline(always)]
+        pub fn gte(self, rhs: Self) -> Bool4 {
+            Bool4([
+                self.0[0] >= rhs.0[0],
+                self.0[1] >= rhs.0[1],
+                self.0[2] >= rhs.0[2],
+                self.0[3] >= rhs.0[3],
+            ])
+        }
+
+        /// Equal.
+        #[inline(always)]
+        pub fn eq(self, rhs: Self) -> Bool4 {
+            Bool4([
+                self.0[0] == rhs.0[0],
+                self.0[1] == rhs.0[1],
+                self.0[2] == rhs.0[2],
+                self.0[3] == rhs.0[3],
+            ])
+        }
+
         //-----------------------------------------------------
         // Individual components.
 
@@ -265,14 +325,87 @@ mod fallback {
     }
 
     impl FMulAdd for Float4 {
+        #[inline(always)]
         fn fma(self, b: Self, c: Self) -> Self {
             self.mul_add(b, c)
         }
     }
+
+    //---------------------------------------------------------
+
+    #[derive(Debug, Copy, Clone)]
+    #[repr(transparent)]
+    pub struct Bool4([bool; 4]);
+
+    impl Bool4 {
+        #[inline(always)]
+        pub fn to_bools(self) -> [bool; 4] {
+            self.0
+        }
+
+        #[inline(always)]
+        pub fn to_bitmask(self) -> u8 {
+            self.0[0] as u8
+                | ((self.0[1] as u8) << 1)
+                | ((self.0[2] as u8) << 2)
+                | ((self.0[3] as u8) << 3)
+        }
+    }
+
+    impl BitAnd for Bool4 {
+        type Output = Self;
+
+        #[inline(always)]
+        fn bitand(self, rhs: Self) -> Self {
+            Self([
+                self.0[0] & rhs.0[0],
+                self.0[1] & rhs.0[1],
+                self.0[2] & rhs.0[2],
+                self.0[3] & rhs.0[3],
+            ])
+        }
+    }
+
+    impl BitOr for Bool4 {
+        type Output = Self;
+
+        #[inline(always)]
+        fn bitor(self, rhs: Self) -> Self {
+            Self([
+                self.0[0] | rhs.0[0],
+                self.0[1] | rhs.0[1],
+                self.0[2] | rhs.0[2],
+                self.0[3] | rhs.0[3],
+            ])
+        }
+    }
+
+    impl BitXor for Bool4 {
+        type Output = Self;
+
+        #[inline(always)]
+        fn bitxor(self, rhs: Self) -> Self {
+            Self([
+                self.0[0] ^ rhs.0[0],
+                self.0[1] ^ rhs.0[1],
+                self.0[2] ^ rhs.0[2],
+                self.0[3] ^ rhs.0[3],
+            ])
+        }
+    }
+
+    impl Not for Bool4 {
+        type Output = Self;
+
+        #[inline(always)]
+        fn not(self) -> Self {
+            Self([!self.0[0], !self.0[1], !self.0[2], !self.0[3]])
+        }
+    }
 }
 
 //-------------------------------------------------------------
-// Float4 impls that don't depend on its inner representation.
+// Impls that don't depend on inner representation.
 
 impl Float4 {
     /// 3D dot product (only uses the first 3 components).
@@ -562,3 +695,24 @@ impl DivAssign<f32> for Float4 {
         *self = *self / rhs;
     }
 }
+
+impl BitAndAssign for Bool4 {
+    #[inline(always)]
+    fn bitand_assign(&mut self, rhs: Self) {
+        *self = *self & rhs;
+    }
+}
+
+impl BitOrAssign for Bool4 {
+    #[inline(always)]
+    fn bitor_assign(&mut self, rhs: Self) {
+        *self = *self | rhs;
+    }
+}
+
+impl BitXorAssign for Bool4 {
+    #[inline(always)]
+    fn bitxor_assign(&mut self, rhs: Self) {
+        *self = *self ^ rhs;
+    }
+}