convert vello_cpu to use #[simd]

Cargo.toml

@@ -119,7 +119,8 @@ rayon = { version = "1.12.0" }
 thread_local = "1.1.9"
 crossbeam-channel = "0.5.15"
 ordered-channel = { version = "1.2.0", features = ["crossbeam-channel"] }
-fearless_simd = { version = "0.7.0", default-features = false }
+fearless_simd = { version = "1.0.0-rc.1", default-features = false }
+fearless_simd_macros = "0.1.0"
 
 # Unlike the other crates, please do not update these before a release
 # unless absolutely necessary.
@@ -177,3 +178,7 @@ codegen-units = 1
 inherits = "release"
 debug = true
 strip = "none"
+
+[patch.crates-io]
+fearless_simd = { git = "https://github.com/linebender/fearless_simd.git", branch = "main" }
+fearless_simd_macros = { git = "https://github.com/linebender/fearless_simd.git", branch = "main" }

vello_common/src/flatten_simd.rs

@@ -369,12 +369,12 @@ fn is_finite_simd<S: Simd>(x: f32x4<S>) -> mask32x4<S> {
 /// SIMD version of [`approx_parabola_integral`].
 #[inline(always)]
 fn approx_parabola_integral_simd<S: Simd, F: SimdFloat<S, Element = f32>>(x: F) -> F {
-    let simd = x.witness();
+    let simd = x.token();
 
     const D: f32 = 0.67;
     const D_POWI_4: f32 = 0.201_511_2;
 
-    let temp = F::splat(x.witness(), 0.25)
+    let temp = F::splat(simd, 0.25)
         .mul_add(x * x, F::splat(simd, D_POWI_4))
         .sqrt()
         .sqrt();

vello_common/src/pixmap.rs

@@ -415,6 +415,7 @@ impl Pixmap {
     /// If both, premultiplied and unpremultiplied RGBA are acceptable formats, it is
     /// recommended to choose [`ImageAlphaType::AlphaPremultiplied`], as the data can
     /// be returned as is without any additional post-processing.
+    #[inline]
     pub fn take_rgba8(mut self, alpha_type: ImageAlphaType) -> Vec<u8> {
         if self.may_have_transparency && alpha_type == ImageAlphaType::Alpha {
             unpremultiply_rgba8(bytemuck::cast_slice_mut(&mut self.buf));