Update fma target

Cargo.toml

@@ -2,7 +2,7 @@ workspace = { members = ["app"] }
 
 [package]
 name = "pic-scale"
-version = "0.2.6"
+version = "0.2.7"
 edition = "2021"
 description = "High performance image scaling"
 readme = "README.md"

src/math/bartlett.rs

@@ -41,7 +41,7 @@ where
     if x >= 0f32.as_() && x <= 1f32.as_() {
         return 2f32.as_() * x;
     }
-    return 2f32.as_() - 2f32.as_() * x;
+    2f32.as_() - 2f32.as_() * x
 }
 
 #[inline(always)]
@@ -61,5 +61,5 @@ where
     let fac = (x / (l - 1.0f32.as_()) - 0.5f32.as_()).abs();
     let w = 0.62f32.as_() - 0.4832.as_() * fac
         + 0.38f32.as_() * (2f32.as_() * V::const_pi() * fac).cos();
-    return w;
+    w
 }

src/neon/utils.rs

@@ -50,20 +50,17 @@ pub(crate) unsafe fn prefer_vfmaq_f32(
 pub(crate) unsafe fn load_3b_as_u16x4(src_ptr: *const u8) -> uint16x4_t {
     let out_reg: uint16x4_t;
     asm!("\
-         ldrb    {t1:w}, [{0}]
-         ldrb    {t2:w}, [{0}, #1]
-         ldrb    {t3:w}, [{0}, #2]
+         ldrh    {t1:w}, [{0}]
+         ldrb    {t2:w}, [{0}, #2]
 
-         ins {1:v}.h[0], {t1:w}
-         ins {1:v}.h[1], {t2:w}
-         ins {1:v}.h[2], {t3:w}
+         ins {1:v}.s[0], {t1:w}
+         ins {1:v}.h[2], {t2:w}
     \
     ",
     in(reg) src_ptr,
     out(vreg) out_reg,
     t1 = out(reg) _,
-    t2 = out(reg) _,
-    t3 = out(reg) _);
+    t2 = out(reg) _);
     out_reg
 }
 

src/sse/utils.rs

@@ -34,6 +34,7 @@ use std::arch::x86::*;
 use std::arch::x86_64::*;
 
 #[inline]
+#[target_feature(enable = "sse4.1")]
 pub unsafe fn _mm_prefer_fma_ps<const FMA: bool>(a: __m128, b: __m128, c: __m128) -> __m128 {
     if FMA {
         _mm_fma_psx(a, b, c)
@@ -43,7 +44,7 @@ pub unsafe fn _mm_prefer_fma_ps<const FMA: bool>(a: __m128, b: __m128, c: __m128
 }
 
 #[inline]
-#[target_feature(enable = "sse4.1")]
+#[target_feature(enable = "sse4.1,fma")]
 unsafe fn _mm_fma_psx(a: __m128, b: __m128, c: __m128) -> __m128 {
     _mm_fmadd_ps(b, c, a)
 }