AVX LCh

Cargo.toml

@@ -2,7 +2,7 @@ workspace = { members = ["src/app"] }
 
 [package]
 name = "colorutils-rs"
-version = "0.4.7"
+version = "0.4.8"
 edition = "2021"
 description = "High performance utilities for color format handling and conversion."
 readme = "README.md"

src/avx/cie.rs

@@ -0,0 +1,188 @@
+use crate::avx::math::*;
+use crate::avx::{_mm256_cube_ps, _mm256_prefer_fma_ps, _mm256_select_ps};
+use crate::luv::{
+    LUV_CUTOFF_FORWARD_Y, LUV_MULTIPLIER_FORWARD_Y, LUV_MULTIPLIER_INVERSE_Y, LUV_WHITE_U_PRIME,
+    LUV_WHITE_V_PRIME,
+};
+#[cfg(target_arch = "x86")]
+use std::arch::x86::*;
+#[cfg(target_arch = "x86_64")]
+use std::arch::x86_64::*;
+
+#[inline(always)]
+pub(crate) unsafe fn avx_lab_to_xyz(l: __m256, a: __m256, b: __m256) -> (__m256, __m256, __m256) {
+    let y = _mm256_mul_ps(
+        _mm256_add_ps(l, _mm256_set1_ps(16f32)),
+        _mm256_set1_ps(1f32 / 116f32),
+    );
+    let x = _mm256_add_ps(_mm256_mul_ps(a, _mm256_set1_ps(1f32 / 500f32)), y);
+    let z = _mm256_sub_ps(y, _mm256_mul_ps(b, _mm256_set1_ps(1f32 / 200f32)));
+    let x3 = _mm256_cube_ps(x);
+    let y3 = _mm256_cube_ps(y);
+    let z3 = _mm256_cube_ps(z);
+    let kappa = _mm256_set1_ps(0.008856f32);
+    let k_sub = _mm256_set1_ps(16f32 / 116f32);
+    let mult_1 = _mm256_set1_ps(1f32 / 7.787f32);
+    let low_x = _mm256_mul_ps(_mm256_sub_ps(x, k_sub), mult_1);
+    let low_y = _mm256_mul_ps(_mm256_sub_ps(y, k_sub), mult_1);
+    let low_z = _mm256_mul_ps(_mm256_sub_ps(z, k_sub), mult_1);
+
+    let x = _mm256_select_ps(_mm256_cmp_ps::<_CMP_GT_OS>(x3, kappa), x3, low_x);
+    let y = _mm256_select_ps(_mm256_cmp_ps::<_CMP_GT_OS>(y3, kappa), y3, low_y);
+    let z = _mm256_select_ps(_mm256_cmp_ps::<_CMP_GT_OS>(z3, kappa), z3, low_z);
+    let x = _mm256_mul_ps(x, _mm256_set1_ps(95.047f32 / 100f32));
+    let z = _mm256_mul_ps(z, _mm256_set1_ps(108.883f32 / 100f32));
+    (x, y, z)
+}
+
+#[inline(always)]
+pub(crate) unsafe fn avx_luv_to_xyz(l: __m256, u: __m256, v: __m256) -> (__m256, __m256, __m256) {
+    let zeros = _mm256_setzero_ps();
+    let zero_mask = _mm256_cmp_ps::<_CMP_EQ_OS>(l, zeros);
+    let l13 = _mm256_rcp_ps(_mm256_mul_ps(l, _mm256_set1_ps(13f32)));
+    let u = _mm256_prefer_fma_ps(_mm256_set1_ps(LUV_WHITE_U_PRIME), l13, u);
+    let v = _mm256_prefer_fma_ps(_mm256_set1_ps(LUV_WHITE_V_PRIME), l13, v);
+    let l_h = _mm256_mul_ps(
+        _mm256_add_ps(l, _mm256_set1_ps(16f32)),
+        _mm256_set1_ps(1f32 / 116f32),
+    );
+    let y_high = _mm256_mul_ps(_mm256_mul_ps(l_h, l_h), l_h);
+    let y_low = _mm256_mul_ps(l, _mm256_set1_ps(LUV_MULTIPLIER_INVERSE_Y));
+    let y = _mm256_select_ps(
+        zero_mask,
+        zeros,
+        _mm256_select_ps(
+            _mm256_cmp_ps::<_CMP_GT_OS>(l, _mm256_set1_ps(8f32)),
+            y_high,
+            y_low,
+        ),
+    );
+    let zero_mask_2 = _mm256_cmp_ps::<_CMP_EQ_OS>(v, zeros);
+    let den = _mm256_rcp_ps(_mm256_mul_ps(v, _mm256_set1_ps(4f32)));
+    let mut x = _mm256_mul_ps(
+        _mm256_mul_ps(_mm256_mul_ps(y, u), den),
+        _mm256_set1_ps(9f32),
+    );
+    x = _mm256_select_ps(zero_mask, zeros, x);
+    x = _mm256_select_ps(zero_mask_2, zeros, x);
+    let mut z = _mm256_mul_ps(
+        _mm256_mul_ps(
+            _mm256_prefer_fma_ps(
+                _mm256_prefer_fma_ps(_mm256_set1_ps(12f32), _mm256_set1_ps(-3f32), u),
+                v,
+                _mm256_set1_ps(-20f32),
+            ),
+            y,
+        ),
+        den,
+    );
+    z = _mm256_select_ps(zero_mask, zeros, z);
+    z = _mm256_select_ps(zero_mask_2, zeros, z);
+    (x, y, z)
+}
+
+#[inline(always)]
+pub(crate) unsafe fn avx_lch_to_xyz(l: __m256, c: __m256, h: __m256) -> (__m256, __m256, __m256) {
+    let u = _mm256_mul_ps(c, _mm256_cos_ps(h));
+    let v = _mm256_mul_ps(c, _mm256_sin_ps(h));
+    avx_luv_to_xyz(l, u, v)
+}
+
+#[inline(always)]
+pub(crate) unsafe fn avx2_triple_to_xyz(
+    r: __m256i,
+    g: __m256i,
+    b: __m256i,
+    c1: __m256,
+    c2: __m256,
+    c3: __m256,
+    c4: __m256,
+    c5: __m256,
+    c6: __m256,
+    c7: __m256,
+    c8: __m256,
+    c9: __m256,
+    transfer: &unsafe fn(__m256) -> __m256,
+) -> (__m256, __m256, __m256) {
+    let u8_scale = _mm256_set1_ps(1f32 / 255f32);
+    let r_f = _mm256_mul_ps(_mm256_cvtepi32_ps(r), u8_scale);
+    let g_f = _mm256_mul_ps(_mm256_cvtepi32_ps(g), u8_scale);
+    let b_f = _mm256_mul_ps(_mm256_cvtepi32_ps(b), u8_scale);
+    let r_linear = transfer(r_f);
+    let g_linear = transfer(g_f);
+    let b_linear = transfer(b_f);
+
+    let (x, y, z) = _mm256_color_matrix_ps(
+        r_linear, g_linear, b_linear, c1, c2, c3, c4, c5, c6, c7, c8, c9,
+    );
+    (x, y, z)
+}
+
+#[inline(always)]
+pub(crate) unsafe fn avx2_triple_to_luv(
+    x: __m256,
+    y: __m256,
+    z: __m256,
+) -> (__m256, __m256, __m256) {
+    let zeros = _mm256_setzero_ps();
+    let den = _mm256_prefer_fma_ps(
+        _mm256_prefer_fma_ps(x, z, _mm256_set1_ps(3f32)),
+        y,
+        _mm256_set1_ps(15f32),
+    );
+    let nan_mask = _mm256_cmp_ps::<_CMP_LT_OS>(den, _mm256_set1_ps(0f32));
+    let l_low_mask = _mm256_cmp_ps::<_CMP_LT_OS>(y, _mm256_set1_ps(LUV_CUTOFF_FORWARD_Y));
+    let y_cbrt = _mm256_cbrt_ps(y);
+    let l = _mm256_select_ps(
+        l_low_mask,
+        _mm256_mul_ps(y, _mm256_set1_ps(LUV_MULTIPLIER_FORWARD_Y)),
+        _mm256_prefer_fma_ps(_mm256_set1_ps(-16f32), y_cbrt, _mm256_set1_ps(116f32)),
+    );
+    let u_prime = _mm256_div_ps(_mm256_mul_ps(x, _mm256_set1_ps(4f32)), den);
+    let v_prime = _mm256_div_ps(_mm256_mul_ps(y, _mm256_set1_ps(9f32)), den);
+    let sub_u_prime = _mm256_sub_ps(u_prime, _mm256_set1_ps(crate::luv::LUV_WHITE_U_PRIME));
+    let sub_v_prime = _mm256_sub_ps(v_prime, _mm256_set1_ps(crate::luv::LUV_WHITE_V_PRIME));
+    let l13 = _mm256_mul_ps(l, _mm256_set1_ps(13f32));
+    let u = _mm256_select_ps(nan_mask, zeros, _mm256_mul_ps(l13, sub_u_prime));
+    let v = _mm256_select_ps(nan_mask, zeros, _mm256_mul_ps(l13, sub_v_prime));
+    (l, u, v)
+}
+
+#[inline(always)]
+pub(crate) unsafe fn avx2_triple_to_lab(
+    x: __m256,
+    y: __m256,
+    z: __m256,
+) -> (__m256, __m256, __m256) {
+    let x = _mm256_mul_ps(x, _mm256_set1_ps(100f32 / 95.047f32));
+    let y = _mm256_mul_ps(y, _mm256_set1_ps(100f32 / 100f32));
+    let z = _mm256_mul_ps(z, _mm256_set1_ps(100f32 / 108.883f32));
+    let cbrt_x = _mm256_cbrt_ps(x);
+    let cbrt_y = _mm256_cbrt_ps(y);
+    let cbrt_z = _mm256_cbrt_ps(z);
+    let s_1 = _mm256_set1_ps(16.0 / 116.0);
+    let s_2 = _mm256_set1_ps(7.787);
+    let lower_x = _mm256_prefer_fma_ps(s_1, s_2, x);
+    let lower_y = _mm256_prefer_fma_ps(s_1, s_2, y);
+    let lower_z = _mm256_prefer_fma_ps(s_1, s_2, z);
+    let cutoff = _mm256_set1_ps(0.008856f32);
+    let x = _mm256_select_ps(_mm256_cmp_ps::<_CMP_GT_OS>(x, cutoff), cbrt_x, lower_x);
+    let y = _mm256_select_ps(_mm256_cmp_ps::<_CMP_GT_OS>(y, cutoff), cbrt_y, lower_y);
+    let z = _mm256_select_ps(_mm256_cmp_ps::<_CMP_GT_OS>(z, cutoff), cbrt_z, lower_z);
+    let l = _mm256_prefer_fma_ps(_mm256_set1_ps(-16.0f32), y, _mm256_set1_ps(116.0f32));
+    let a = _mm256_mul_ps(_mm256_sub_ps(x, y), _mm256_set1_ps(500f32));
+    let b = _mm256_mul_ps(_mm256_sub_ps(y, z), _mm256_set1_ps(200f32));
+    (l, a, b)
+}
+
+#[inline(always)]
+pub(crate) unsafe fn avx_triple_to_lch(
+    x: __m256,
+    y: __m256,
+    z: __m256,
+) -> (__m256, __m256, __m256) {
+    let (luv_l, luv_u, luv_v) = avx2_triple_to_luv(x, y, z);
+    let lch_c = _mm256_hypot_ps(luv_u, luv_v);
+    let lch_h = _mm256_atan2_ps(luv_v, luv_u);
+    (luv_l, lch_c, lch_h)
+}