doubled 0.3.0

burrbull · Aug 8, 2022 · 62c1610 · 62c1610
1 parent f81b595
commit 62c1610
Show file tree

Hide file tree

Showing 12 changed files with 44 additions and 37 deletions.
diff --git a/Cargo.toml b/Cargo.toml
@@ -11,8 +11,7 @@ license = "MIT OR Apache-2.0"
 readme = "README.md"
 
 [dependencies.doubled]
-git = "https://github.com/burrbull/doubled-rs"
-branch = "portable-simd"
+version = "0.3.0"
 features = ["simd"]
 
 [features]

diff --git a/src/common.rs b/src/common.rs
@@ -18,15 +18,15 @@ impl SqrtAsDoubled for f32 {
     #[inline]
     fn sqrt_as_doubled(self) -> Doubled<Self> {
         let t = self.sqrt();
-        ((self + t.mul_as_doubled(t)) * t.recpre()).scale(0.5)
+        ((self + t.mul_as_doubled(t)) * t.recip_as_doubled()).scale(0.5)
     }
 }
 
 impl SqrtAsDoubled for f64 {
     #[inline]
     fn sqrt_as_doubled(self) -> Doubled<Self> {
         let t = self.sqrt();
-        ((self + t.mul_as_doubled(t)) * t.recpre()).scale(0.5)
+        ((self + t.mul_as_doubled(t)) * t.recip_as_doubled()).scale(0.5)
     }
 }
 
@@ -52,15 +52,15 @@ impl Sqrt for Doubled<f32> {
     #[inline]
     fn sqrt(self) -> Self {
         let t = f32::from(self).sqrt();
-        ((self + t.mul_as_doubled(t)) * t.recpre()).scale(0.5)
+        ((self + t.mul_as_doubled(t)) * t.recip_as_doubled()).scale(0.5)
     }
 }
 
 impl Sqrt for Doubled<f64> {
     #[inline]
     fn sqrt(self) -> Self {
         let t = f64::from(self).sqrt();
-        ((self + t.mul_as_doubled(t)) * t.recpre()).scale(0.5)
+        ((self + t.mul_as_doubled(t)) * t.recip_as_doubled()).scale(0.5)
     }
 }
 

diff --git a/src/f32/u05.rs b/src/f32/u05.rs
@@ -185,7 +185,7 @@ pub fn sqrtf(mut d: f32) -> f32 {
     x *= 1.5 - 0.5 * d * x * x;
     x *= (1.5 - 0.5 * d * x * x) * d;
 
-    let d2 = (d + x.mul_as_doubled(x)) * x.recpre();
+    let d2 = (d + x.mul_as_doubled(x)) * x.recip_as_doubled();
 
     if (d == 0.) || (d == f32::INFINITY) {
         d

diff --git a/src/f32/u10.rs b/src/f32/u10.rs
@@ -225,7 +225,7 @@ pub fn tanf(d: f32) -> f32 {
     x = t * x;
 
     if (q & 1) != 0 {
-        x = x.recpre();
+        x = x.recip();
     }
 
     if d.is_neg_zero() {
@@ -439,7 +439,7 @@ fn test_atanf() {
 pub fn sinhf(x: f32) -> f32 {
     let mut y = fabsfk(x);
     let mut d = expk2f(Doubled::from(y));
-    d = d.sub_checked(d.recpre());
+    d = d.sub_checked(d.recip());
     y = f32::from(d) * 0.5;
 
     y = if fabsfk(x) > 89. { f32::INFINITY } else { y };
@@ -466,7 +466,7 @@ fn test_sinhf() {
 pub fn coshf(x: f32) -> f32 {
     let mut y = fabsfk(x);
     let mut d = expk2f(Doubled::from(y));
-    d = d.add_checked(d.recpre());
+    d = d.add_checked(d.recip());
     y = f32::from(d) * 0.5;
 
     y = if fabsfk(x) > 89. { f32::INFINITY } else { y };
@@ -490,7 +490,7 @@ fn test_coshf() {
 pub fn tanhf(x: f32) -> f32 {
     let mut y = fabsfk(x);
     let mut d = expk2f(Doubled::from(y));
-    let e = d.recpre();
+    let e = d.recip();
     d = d.sub_checked(e) / d.add_checked(e);
     y = f32::from(d);
 
@@ -534,7 +534,11 @@ fn logk2f(d: Doubled<f32>) -> Doubled<f32> {
 pub fn asinhf(x: f32) -> f32 {
     let mut y = fabsfk(x);
 
-    let mut d = if y > 1. { x.recpre() } else { Doubled::from(y) };
+    let mut d = if y > 1. {
+        x.recip_as_doubled()
+    } else {
+        Doubled::from(y)
+    };
     d = (d.square() + 1.).sqrt();
     d = if y > 1. { d * y } else { d };
 
@@ -1400,7 +1404,7 @@ pub fn erff(a: f32) -> f32 {
         t2 = t2.square();
         t2 = t2.square();
         t2 = t2.square();
-        t2 = t2.recpre();
+        t2 = t2.recip();
     } else if x > 4. {
         t2 = Doubled::from(0.);
     } else {

diff --git a/src/f32x.rs b/src/f32x.rs
@@ -526,7 +526,7 @@ macro_rules! impl_math_f32 {
             #[inline]
             fn sqrt_as_doubled(self) -> Doubled<Self> {
                 let t = self.sqrt();
-                ((self + t.mul_as_doubled(t)) * t.recpre_as_doubled()).scale(Self::splat(0.5))
+                ((self + t.mul_as_doubled(t)) * t.recip_as_doubled()).scale(Self::splat(0.5))
             }
         }
 

diff --git a/src/f32x/u05_impl.rs b/src/f32x/u05_impl.rs
@@ -154,7 +154,7 @@ macro_rules! impl_math_f32_u05 {
             x *= F32x::splat(1.5) - HALF * d * x * x;
             x *= d;
 
-            let d2 = (d + x.mul_as_doubled(x)) * x.recpre_as_doubled();
+            let d2 = (d + x.mul_as_doubled(x)) * x.recip_as_doubled();
 
             x = F32x::from(d2) * q;
 

diff --git a/src/f32x/u10_impl.rs b/src/f32x/u10_impl.rs
@@ -461,7 +461,7 @@ macro_rules! impl_math_f32_u10 {
                 .add_checked(F32x::splat(0.333_333_611_488_342_285_156_25).add_checked(s * x) * s);
             x = t * x;
 
-            x = o.select_doubled(x.recpre(), x);
+            x = o.select_doubled(x.recip(), x);
 
             let u = F32x::from(x);
 
@@ -517,7 +517,7 @@ macro_rules! impl_math_f32_u10 {
                 .add_checked(F32x::splat(0.333_333_611_488_342_285_156_25).add_checked(s * x) * s);
             x = t * x;
 
-            x = o.select_doubled(x.recpre(), x);
+            x = o.select_doubled(x.recip(), x);
 
             let u = F32x::from(x);
 
@@ -737,7 +737,7 @@ macro_rules! impl_math_f32_u10 {
         pub fn sinhf(x: F32x) -> F32x {
             let mut y = x.abs();
             let d = expk2f(Doubled::from(y));
-            let d = d.sub_checked(d.recpre());
+            let d = d.sub_checked(d.recip());
             y = F32x::from(d) * HALF;
 
             y = (x.abs().simd_gt(F32x::splat(89.)) | y.is_nan()).select(INFINITY, y);
@@ -764,7 +764,7 @@ macro_rules! impl_math_f32_u10 {
         pub fn coshf(x: F32x) -> F32x {
             let mut y = x.abs();
             let d = expk2f(Doubled::from(y));
-            let d = d.add_checked(d.recpre());
+            let d = d.add_checked(d.recip());
             y = F32x::from(d) * HALF;
 
             y = (x.abs().simd_gt(F32x::splat(89.)) | y.is_nan()).select(INFINITY, y);
@@ -788,7 +788,7 @@ macro_rules! impl_math_f32_u10 {
         pub fn tanhf(x: F32x) -> F32x {
             let mut y = x.abs();
             let d = expk2f(Doubled::from(y));
-            let e = d.recpre();
+            let e = d.recip();
             let d = d.add_checked(-e) / d.add_checked(e);
             y = F32x::from(d);
 
@@ -839,7 +839,7 @@ macro_rules! impl_math_f32_u10 {
             let mut y = x.abs();
             let o = y.simd_gt(ONE);
 
-            let mut d = o.select_doubled(x.recpre_as_doubled(), Doubled::from(y));
+            let mut d = o.select_doubled(x.recip_as_doubled(), Doubled::from(y));
             d = (d.square() + ONE).sqrt();
             d = o.select_doubled(d * y, d);
 
@@ -1778,7 +1778,7 @@ macro_rules! impl_math_f32_u10 {
                 t2 = t2.square();
                 t2 = t2.square();
                 t2 = t2.square();
-                t2 = t2.recpre();
+                t2 = t2.recip();
             } else {
                 let t = F32x::poly6(x, x2, x4,
                         o25.select_splat(-0.436_044_700_8_e-6, -0.113_001_284_8_e-6),
@@ -1825,7 +1825,7 @@ macro_rules! impl_math_f32_u10 {
                 s2 = s2.square();
                 s2 = s2.square();
                 s2 = s2.square();
-                s2 = s2.recpre();
+                s2 = s2.recip();
                 t2 = o25.select_doubled(s2, Doubled::from(expkf(t2)));
             }
 

diff --git a/src/f64/u05.rs b/src/f64/u05.rs
@@ -292,7 +292,7 @@ pub fn sqrt(mut d: f64) -> f64 {
     x = x * (1.5 - 0.5 * d * x * x);
     x = x * (1.5 - 0.5 * d * x * x) * d;
 
-    let d2 = (d + x.mul_as_doubled(x)) * x.recpre();
+    let d2 = (d + x.mul_as_doubled(x)) * x.recip_as_doubled();
 
     let ret = f64::from(d2) * q;
 

diff --git a/src/f64/u10.rs b/src/f64/u10.rs
@@ -583,7 +583,7 @@ fn test_atan() {
 pub fn sinh(x: f64) -> f64 {
     let mut y = fabsk(x);
     let mut d = expk2(Doubled::from(y));
-    d = d.sub_checked(d.recpre());
+    d = d.sub_checked(d.recip());
     y = f64::from(d) * 0.5;
 
     y = if fabsk(x) > 710. { f64::INFINITY } else { y };
@@ -610,7 +610,7 @@ fn test_sinh() {
 pub fn cosh(x: f64) -> f64 {
     let mut y = fabsk(x);
     let mut d = expk2(Doubled::from(y));
-    d = d.add_checked(d.recpre());
+    d = d.add_checked(d.recip());
     y = f64::from(d) * 0.5;
 
     y = if fabsk(x) > 710. { f64::INFINITY } else { y };
@@ -634,7 +634,7 @@ fn test_cosh() {
 pub fn tanh(x: f64) -> f64 {
     let mut y = fabsk(x);
     let mut d = expk2(Doubled::from(y));
-    let e = d.recpre();
+    let e = d.recip();
     d = d.sub_checked(e) / d.add_checked(e);
     y = f64::from(d);
 
@@ -691,7 +691,11 @@ fn logk2(d: Doubled<f64>) -> Doubled<f64> {
 pub fn asinh(x: f64) -> f64 {
     let mut y = fabsk(x);
 
-    let mut d = if y > 1. { x.recpre() } else { Doubled::from(y) };
+    let mut d = if y > 1. {
+        x.recip_as_doubled()
+    } else {
+        Doubled::from(y)
+    };
     d = (d.square() + 1.).sqrt();
     d = if y > 1. { d * y } else { d };
 
@@ -1839,7 +1843,7 @@ pub fn erf(a: f64) -> f64 {
         t2 = t2.square();
         t2 = t2.square();
         t2 = t2.square();
-        t2 = t2.recpre();
+        t2 = t2.recip();
     } else if x > 6. {
         t2 = Doubled::from(0.);
     } else {

diff --git a/src/f64x.rs b/src/f64x.rs
@@ -559,7 +559,7 @@ macro_rules! impl_math_f64 {
             #[inline]
             fn sqrt_as_doubled(self) -> Doubled<Self> {
                 let t = self.sqrt();
-                ((self + t.mul_as_doubled(t)) * t.recpre_as_doubled()).scale(Self::splat(0.5))
+                ((self + t.mul_as_doubled(t)) * t.recip_as_doubled()).scale(Self::splat(0.5))
             }
         }
 

diff --git a/src/f64x/u05_impl.rs b/src/f64x/u05_impl.rs
@@ -331,7 +331,7 @@ macro_rules! impl_math_f64_u05 {
             x *= F64x::splat(1.5) - HALF * d * x * x;
             x *= d;
 
-            let d2 = (d + x.mul_as_doubled(x)) * x.recpre_as_doubled();
+            let d2 = (d + x.mul_as_doubled(x)) * x.recip_as_doubled();
 
             x = F64x::from(d2) * q;
 

diff --git a/src/f64x/u10_impl.rs b/src/f64x/u10_impl.rs
@@ -944,7 +944,7 @@ macro_rules! impl_math_f64_u10 {
         pub fn sinh(x: F64x) -> F64x {
             let mut y = x.abs();
             let mut d = expk2(Doubled::from(y));
-            d = d.sub_checked(d.recpre());
+            d = d.sub_checked(d.recip());
             y = F64x::from(d) * HALF;
 
             y = (x.abs().simd_gt(F64x::splat(710.)) | y.is_nan()).select(INFINITY, y);
@@ -971,7 +971,7 @@ macro_rules! impl_math_f64_u10 {
         pub fn cosh(x: F64x) -> F64x {
             let mut y = x.abs();
             let mut d = expk2(Doubled::from(y));
-            d = d.add_checked(d.recpre());
+            d = d.add_checked(d.recip());
             y = F64x::from(d) * HALF;
 
             y = (x.abs().simd_gt(F64x::splat(710.)) | y.is_nan()).select(INFINITY, y);
@@ -995,7 +995,7 @@ macro_rules! impl_math_f64_u10 {
         pub fn tanh(x: F64x) -> F64x {
             let mut y = x.abs();
             let mut d = expk2(Doubled::from(y));
-            let e = d.recpre();
+            let e = d.recip();
             d = (d + (-e)) / (d + e);
             y = F64x::from(d);
 
@@ -1055,7 +1055,7 @@ macro_rules! impl_math_f64_u10 {
             let mut y = x.abs();
             let o = y.simd_gt(ONE);
 
-            let mut d = o.select_doubled(x.recpre_as_doubled(), Doubled::from(y));
+            let mut d = o.select_doubled(x.recip_as_doubled(), Doubled::from(y));
             d = (d.square() + ONE).sqrt();
             d = o.select_doubled(d * y, d);
 
@@ -2248,7 +2248,7 @@ macro_rules! impl_math_f64_u10 {
                 t2 = t2.square();
                 t2 = t2.square();
                 t2 = t2.square();
-                t2 = t2.recpre();
+                t2 = t2.recip();
             } else {
                 let t = F64x::poly21(x, x2, x4, x8, x16,
                     o25.select_splat(-0.208_327_100_252_522_209_7_e-14, -0.402_401_513_075_262_193_2_e-18),
@@ -2309,7 +2309,7 @@ macro_rules! impl_math_f64_u10 {
                 s2 = s2.square();
                 s2 = s2.square();
                 s2 = s2.square();
-                s2 = s2.recpre();
+                s2 = s2.recip();
                 t2 = o25.select_doubled(s2, Doubled::from(expk(t2)));
             }