Mark generic functions #[inline]

Benchmarks for [1] seemed to indicate that repository organization for
some reason had an effect on performance, even though the exact same
rustc commands were running (though some with a different order). After
investigating more, it appears that dependencies may have an affect on
inlining thresholds for generic functions.

It is surprising that this happens, we more or less expect that public
functions will be standalone but everything they call will be inlined.
To help ensure this, mark all generic functions `#[inline]` if they
should be merged into the public function.

Zulip discussion at [2].

[1]: rust-lang#533
[2]: https://rust-lang.zulipchat.com/#narrow/channel/182449-t-compiler.2Fhelp/topic/Dependencies.20affecting.20codegen/with/513079387

Author: tgross35

src/math/fma.rs

@@ -29,6 +29,7 @@ pub fn fmaf128(x: f128, y: f128, z: f128) -> f128 {
 
 /// Fused multiply-add that works when there is not a larger float size available. Computes
 /// `(x * y) + z`.
+#[inline]
 pub fn fma_round<F>(x: F, y: F, z: F, _round: Round) -> FpResult<F>
 where
     F: Float,

src/math/fma_wide.rs

@@ -28,6 +28,7 @@ pub fn fmaf(x: f32, y: f32, z: f32) -> f32 {
 
 /// Fma implementation when a hardware-backed larger float type is available. For `f32` and `f64`,
 /// `f64` has enough precision to represent the `f32` in its entirety, except for double rounding.
+#[inline]
 pub fn fma_wide_round<F, B>(x: F, y: F, z: F, round: Round) -> FpResult<F>
 where
     F: Float + HFloat<D = B>,

src/math/generic/ceil.rs

@@ -10,10 +10,12 @@
 use super::super::support::{FpResult, Status};
 use super::super::{Float, Int, IntTy, MinInt};
 
+#[inline]
 pub fn ceil<F: Float>(x: F) -> F {
     ceil_status(x).val
 }
 
+#[inline]
 pub fn ceil_status<F: Float>(x: F) -> FpResult<F> {
     let zero = IntTy::<F>::ZERO;
 

src/math/generic/copysign.rs

@@ -1,6 +1,7 @@
 use super::super::Float;
 
 /// Copy the sign of `y` to `x`.
+#[inline]
 pub fn copysign<F: Float>(x: F, y: F) -> F {
     let mut ux = x.to_bits();
     let uy = y.to_bits();

src/math/generic/fabs.rs

@@ -1,6 +1,7 @@
 use super::super::Float;
 
 /// Absolute value.
+#[inline]
 pub fn fabs<F: Float>(x: F) -> F {
     let abs_mask = !F::SIGN_MASK;
     F::from_bits(x.to_bits() & abs_mask)

src/math/generic/fdim.rs

@@ -1,5 +1,6 @@
 use super::super::Float;
 
+#[inline]
 pub fn fdim<F: Float>(x: F, y: F) -> F {
     if x <= y { F::ZERO } else { x - y }
 }

src/math/generic/floor.rs

@@ -10,10 +10,12 @@
 use super::super::support::{FpResult, Status};
 use super::super::{Float, Int, IntTy, MinInt};
 
+#[inline]
 pub fn floor<F: Float>(x: F) -> F {
     floor_status(x).val
 }
 
+#[inline]
 pub fn floor_status<F: Float>(x: F) -> FpResult<F> {
     let zero = IntTy::<F>::ZERO;
 

src/math/generic/fmax.rs

@@ -16,7 +16,7 @@
 
 use super::super::Float;
 
-#[cfg_attr(all(test, assert_no_panic), no_panic::no_panic)]
+#[inline]
 pub fn fmax<F: Float>(x: F, y: F) -> F {
     let res = if x.is_nan() || x < y { y } else { x };
     // Canonicalize

src/math/generic/fmaximum.rs

@@ -11,6 +11,7 @@
 
 use super::super::Float;
 
+#[inline]
 pub fn fmaximum<F: Float>(x: F, y: F) -> F {
     let res = if x.is_nan() {
         x

src/math/generic/fmaximum_num.rs

@@ -13,6 +13,7 @@
 
 use super::super::Float;
 
+#[inline]
 pub fn fmaximum_num<F: Float>(x: F, y: F) -> F {
     let res =
         if x.is_nan() || x < y || (x.to_bits() == F::NEG_ZERO.to_bits() && y.is_sign_positive()) {