Initial vendor packages

Signed-off-by: Valentin Popov <valentin@popov.link>

3 files changed, 555 insertions, 0 deletions

diff --git a/vendor/jpeg-decoder/src/arch/mod.rs b/vendor/jpeg-decoder/src/arch/mod.rs
new file mode 100644
index 0000000..15b46c5
--- /dev/null
+++ b/vendor/jpeg-decoder/src/arch/mod.rs
@@ -0,0 +1,46 @@
+#![allow(unsafe_code)]
+
+mod neon;
+mod ssse3;
+
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+use std::is_x86_feature_detected;
+
+/// Arch-specific implementation of YCbCr conversion. Returns the number of pixels that were
+/// converted.
+pub fn get_color_convert_line_ycbcr() -> Option<unsafe fn(&[u8], &[u8], &[u8], &mut [u8]) -> usize>
+{
+    #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+    #[allow(unsafe_code)]
+    {
+        if is_x86_feature_detected!("ssse3") {
+            return Some(ssse3::color_convert_line_ycbcr);
+        }
+    }
+    // Runtime detection is not needed on aarch64.
+    #[cfg(all(feature = "nightly_aarch64_neon", target_arch = "aarch64"))]
+    {
+        return Some(neon::color_convert_line_ycbcr);
+    }
+    #[allow(unreachable_code)]
+    None
+}
+
+/// Arch-specific implementation of 8x8 IDCT.
+pub fn get_dequantize_and_idct_block_8x8(
+) -> Option<unsafe fn(&[i16; 64], &[u16; 64], usize, &mut [u8])> {
+    #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+    #[allow(unsafe_code)]
+    {
+        if is_x86_feature_detected!("ssse3") {
+            return Some(ssse3::dequantize_and_idct_block_8x8);
+        }
+    }
+    // Runtime detection is not needed on aarch64.
+    #[cfg(all(feature = "nightly_aarch64_neon", target_arch = "aarch64"))]
+    {
+        return Some(neon::dequantize_and_idct_block_8x8);
+    }
+    #[allow(unreachable_code)]
+    None
+}
diff --git a/vendor/jpeg-decoder/src/arch/neon.rs b/vendor/jpeg-decoder/src/arch/neon.rs
new file mode 100644
index 0000000..4843578
--- /dev/null
+++ b/vendor/jpeg-decoder/src/arch/neon.rs
@@ -0,0 +1,221 @@
+#[cfg(all(feature = "nightly_aarch64_neon", target_arch = "aarch64"))]
+use core::arch::aarch64::*;
+
+#[cfg(all(feature = "nightly_aarch64_neon", target_arch = "aarch64"))]
+#[target_feature(enable = "neon")]
+unsafe fn idct8(data: &mut [int16x8_t; 8]) {
+    // The fixed-point constants here are obtained by taking the fractional part of the constants
+    // from the non-SIMD implementation and scaling them up by 1<<15. This is because
+    // vqrdmulhq_n_s16(a, b) is effectively equivalent to (a*b)>>15 (except for possibly some
+    // slight differences in rounding).
+
+    // The code here is effectively equivalent to the calls to "kernel" in idct.rs, except that it
+    // doesn't apply any further scaling and fixed point constants have a different precision.
+
+    let p2 = data[2];
+    let p3 = data[6];
+    let p1 = vqrdmulhq_n_s16(vqaddq_s16(p2, p3), 17734); // 0.5411961
+    let t2 = vqsubq_s16(
+        vqsubq_s16(p1, p3),
+        vqrdmulhq_n_s16(p3, 27779), // 0.847759065
+    );
+    let t3 = vqaddq_s16(p1, vqrdmulhq_n_s16(p2, 25079)); // 0.765366865
+
+    let p2 = data[0];
+    let p3 = data[4];
+    let t0 = vqaddq_s16(p2, p3);
+    let t1 = vqsubq_s16(p2, p3);
+
+    let x0 = vqaddq_s16(t0, t3);
+    let x3 = vqsubq_s16(t0, t3);
+    let x1 = vqaddq_s16(t1, t2);
+    let x2 = vqsubq_s16(t1, t2);
+
+    let t0 = data[7];
+    let t1 = data[5];
+    let t2 = data[3];
+    let t3 = data[1];
+
+    let p3 = vqaddq_s16(t0, t2);
+    let p4 = vqaddq_s16(t1, t3);
+    let p1 = vqaddq_s16(t0, t3);
+    let p2 = vqaddq_s16(t1, t2);
+    let p5 = vqaddq_s16(p3, p4);
+    let p5 = vqaddq_s16(p5, vqrdmulhq_n_s16(p5, 5763)); // 0.175875602
+
+    let t0 = vqrdmulhq_n_s16(t0, 9786); // 0.298631336
+    let t1 = vqaddq_s16(
+        vqaddq_s16(t1, t1),
+        vqrdmulhq_n_s16(t1, 1741), // 0.053119869
+    );
+    let t2 = vqaddq_s16(
+        vqaddq_s16(t2, vqaddq_s16(t2, t2)),
+        vqrdmulhq_n_s16(t2, 2383), // 0.072711026
+    );
+    let t3 = vqaddq_s16(t3, vqrdmulhq_n_s16(t3, 16427)); // 0.501321110
+
+    let p1 = vqsubq_s16(p5, vqrdmulhq_n_s16(p1, 29490)); // 0.899976223
+    let p2 = vqsubq_s16(
+        vqsubq_s16(vqsubq_s16(p5, p2), p2),
+        vqrdmulhq_n_s16(p2, 18446), // 0.562915447
+    );
+
+    let p3 = vqsubq_s16(
+        vqrdmulhq_n_s16(p3, -31509), // -0.961570560
+        p3,
+    );
+    let p4 = vqrdmulhq_n_s16(p4, -12785); // -0.390180644
+
+    let t3 = vqaddq_s16(vqaddq_s16(p1, p4), t3);
+    let t2 = vqaddq_s16(vqaddq_s16(p2, p3), t2);
+    let t1 = vqaddq_s16(vqaddq_s16(p2, p4), t1);
+    let t0 = vqaddq_s16(vqaddq_s16(p1, p3), t0);
+
+    data[0] = vqaddq_s16(x0, t3);
+    data[7] = vqsubq_s16(x0, t3);
+    data[1] = vqaddq_s16(x1, t2);
+    data[6] = vqsubq_s16(x1, t2);
+    data[2] = vqaddq_s16(x2, t1);
+    data[5] = vqsubq_s16(x2, t1);
+    data[3] = vqaddq_s16(x3, t0);
+    data[4] = vqsubq_s16(x3, t0);
+}
+
+#[cfg(all(feature = "nightly_aarch64_neon", target_arch = "aarch64"))]
+#[target_feature(enable = "neon")]
+unsafe fn transpose8(data: &mut [int16x8_t; 8]) {
+    // Use NEON's 2x2 matrix transposes (vtrn) to do the transposition in each 4x4 block, then
+    // combine the 4x4 blocks.
+    let a01 = vtrnq_s16(data[0], data[1]);
+    let a23 = vtrnq_s16(data[2], data[3]);
+
+    let four0 = vtrnq_s32(vreinterpretq_s32_s16(a01.0), vreinterpretq_s32_s16(a23.0));
+    let four1 = vtrnq_s32(vreinterpretq_s32_s16(a01.1), vreinterpretq_s32_s16(a23.1));
+
+    let a45 = vtrnq_s16(data[4], data[5]);
+    let a67 = vtrnq_s16(data[6], data[7]);
+
+    let four2 = vtrnq_s32(vreinterpretq_s32_s16(a45.0), vreinterpretq_s32_s16(a67.0));
+    let four3 = vtrnq_s32(vreinterpretq_s32_s16(a45.1), vreinterpretq_s32_s16(a67.1));
+
+    data[0] = vreinterpretq_s16_s32(vcombine_s32(vget_low_s32(four0.0), vget_low_s32(four2.0)));
+    data[1] = vreinterpretq_s16_s32(vcombine_s32(vget_low_s32(four1.0), vget_low_s32(four3.0)));
+    data[2] = vreinterpretq_s16_s32(vcombine_s32(vget_low_s32(four0.1), vget_low_s32(four2.1)));
+    data[3] = vreinterpretq_s16_s32(vcombine_s32(vget_low_s32(four1.1), vget_low_s32(four3.1)));
+    data[4] = vreinterpretq_s16_s32(vcombine_s32(vget_high_s32(four0.0), vget_high_s32(four2.0)));
+    data[5] = vreinterpretq_s16_s32(vcombine_s32(vget_high_s32(four1.0), vget_high_s32(four3.0)));
+    data[6] = vreinterpretq_s16_s32(vcombine_s32(vget_high_s32(four0.1), vget_high_s32(four2.1)));
+    data[7] = vreinterpretq_s16_s32(vcombine_s32(vget_high_s32(four1.1), vget_high_s32(four3.1)));
+}
+
+#[cfg(all(feature = "nightly_aarch64_neon", target_arch = "aarch64"))]
+#[target_feature(enable = "neon")]
+pub unsafe fn dequantize_and_idct_block_8x8(
+    coefficients: &[i16; 64],
+    quantization_table: &[u16; 64],
+    output_linestride: usize,
+    output: &mut [u8],
+) {
+    // The loop below will write to positions [output_linestride * i, output_linestride * i + 8)
+    // for 0<=i<8. Thus, the last accessed position is at an offset of output_linestrade * 7 + 7,
+    // and if that position is in-bounds, so are all other accesses.
+    assert!(
+        output.len()
+            > output_linestride
+                .checked_mul(7)
+                .unwrap()
+                .checked_add(7)
+                .unwrap()
+    );
+
+    const SHIFT: i32 = 3;
+
+    // Read the DCT coefficients, scale them up and dequantize them.
+    let mut data = [vdupq_n_s16(0); 8];
+    for i in 0..8 {
+        data[i] = vshlq_n_s16(
+            vmulq_s16(
+                vld1q_s16(coefficients.as_ptr().wrapping_add(i * 8)),
+                vreinterpretq_s16_u16(vld1q_u16(quantization_table.as_ptr().wrapping_add(i * 8))),
+            ),
+            SHIFT,
+        );
+    }
+
+    // Usual column IDCT - transpose - column IDCT - transpose approach.
+    idct8(&mut data);
+    transpose8(&mut data);
+    idct8(&mut data);
+    transpose8(&mut data);
+
+    for i in 0..8 {
+        // The two passes of the IDCT algorithm give us a factor of 8, so the shift here is
+        // increased by 3.
+        // As values will be stored in a u8, they need to be 128-centered and not 0-centered.
+        // We add 128 with the appropriate shift for that purpose.
+        const OFFSET: i16 = 128 << (SHIFT + 3);
+        // We want rounding right shift, so we should add (1/2) << (SHIFT+3) before shifting.
+        const ROUNDING_BIAS: i16 = (1 << (SHIFT + 3)) >> 1;
+
+        let data_with_offset = vqaddq_s16(data[i], vdupq_n_s16(OFFSET + ROUNDING_BIAS));
+
+        vst1_u8(
+            output.as_mut_ptr().wrapping_add(output_linestride * i),
+            vqshrun_n_s16(data_with_offset, SHIFT + 3),
+        );
+    }
+}
+
+#[cfg(all(feature = "nightly_aarch64_neon", target_arch = "aarch64"))]
+#[target_feature(enable = "neon")]
+pub unsafe fn color_convert_line_ycbcr(y: &[u8], cb: &[u8], cr: &[u8], output: &mut [u8]) -> usize {
+    assert!(output.len() % 3 == 0);
+    let num = output.len() / 3;
+    assert!(num <= y.len());
+    assert!(num <= cb.len());
+    assert!(num <= cr.len());
+    let num_vecs = num / 8;
+
+    for i in 0..num_vecs {
+        const SHIFT: i32 = 6;
+        // Load.
+        let y = vld1_u8(y.as_ptr().wrapping_add(i * 8));
+        let cb = vld1_u8(cb.as_ptr().wrapping_add(i * 8));
+        let cr = vld1_u8(cr.as_ptr().wrapping_add(i * 8));
+
+        // Convert to 16 bit and shift.
+        let y = vreinterpretq_s16_u16(vshll_n_u8(y, SHIFT));
+        let cb = vreinterpretq_s16_u16(vshll_n_u8(cb, SHIFT));
+        let cr = vreinterpretq_s16_u16(vshll_n_u8(cr, SHIFT));
+
+        // Add offsets
+        let y = vqaddq_s16(y, vdupq_n_s16((1 << SHIFT) >> 1));
+        let c128 = vdupq_n_s16(128 << SHIFT);
+        let cb = vqsubq_s16(cb, c128);
+        let cr = vqsubq_s16(cr, c128);
+
+        // Compute cr * 1.402, cb * 0.34414, cr * 0.71414, cb * 1.772
+        let cr_140200 = vqaddq_s16(vqrdmulhq_n_s16(cr, 13173), cr);
+        let cb_034414 = vqrdmulhq_n_s16(cb, 11276);
+        let cr_071414 = vqrdmulhq_n_s16(cr, 23401);
+        let cb_177200 = vqaddq_s16(vqrdmulhq_n_s16(cb, 25297), cb);
+
+        // Last conversion step.
+        let r = vqaddq_s16(y, cr_140200);
+        let g = vqsubq_s16(y, vqaddq_s16(cb_034414, cr_071414));
+        let b = vqaddq_s16(y, cb_177200);
+
+        // Shift back and convert to u8.
+        let r = vqshrun_n_s16(r, SHIFT);
+        let g = vqshrun_n_s16(g, SHIFT);
+        let b = vqshrun_n_s16(b, SHIFT);
+
+        // Shuffle + store.
+        vst3_u8(
+            output.as_mut_ptr().wrapping_add(24 * i),
+            uint8x8x3_t(r, g, b),
+        );
+    }
+
+    num_vecs * 8
+}
diff --git a/vendor/jpeg-decoder/src/arch/ssse3.rs b/vendor/jpeg-decoder/src/arch/ssse3.rs
new file mode 100644
index 0000000..374a70c
--- /dev/null
+++ b/vendor/jpeg-decoder/src/arch/ssse3.rs
@@ -0,0 +1,288 @@
+#[cfg(target_arch = "x86")]
+use std::arch::x86::*;
+#[cfg(target_arch = "x86_64")]
+use std::arch::x86_64::*;
+
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+#[target_feature(enable = "ssse3")]
+unsafe fn idct8(data: &mut [__m128i; 8]) {
+    // The fixed-point constants here are obtained by taking the fractional part of the constants
+    // from the non-SIMD implementation and scaling them up by 1<<15. This is because
+    // _mm_mulhrs_epi16(a, b) is effectively equivalent to (a*b)>>15 (except for possibly some
+    // slight differences in rounding).
+
+    // The code here is effectively equivalent to the calls to "kernel" in idct.rs, except that it
+    // doesn't apply any further scaling and fixed point constants have a different precision.
+
+    let p2 = data[2];
+    let p3 = data[6];
+    let p1 = _mm_mulhrs_epi16(_mm_adds_epi16(p2, p3), _mm_set1_epi16(17734)); // 0.5411961
+    let t2 = _mm_subs_epi16(
+        _mm_subs_epi16(p1, p3),
+        _mm_mulhrs_epi16(p3, _mm_set1_epi16(27779)), // 0.847759065
+    );
+    let t3 = _mm_adds_epi16(p1, _mm_mulhrs_epi16(p2, _mm_set1_epi16(25079))); // 0.765366865
+
+    let p2 = data[0];
+    let p3 = data[4];
+    let t0 = _mm_adds_epi16(p2, p3);
+    let t1 = _mm_subs_epi16(p2, p3);
+
+    let x0 = _mm_adds_epi16(t0, t3);
+    let x3 = _mm_subs_epi16(t0, t3);
+    let x1 = _mm_adds_epi16(t1, t2);
+    let x2 = _mm_subs_epi16(t1, t2);
+
+    let t0 = data[7];
+    let t1 = data[5];
+    let t2 = data[3];
+    let t3 = data[1];
+
+    let p3 = _mm_adds_epi16(t0, t2);
+    let p4 = _mm_adds_epi16(t1, t3);
+    let p1 = _mm_adds_epi16(t0, t3);
+    let p2 = _mm_adds_epi16(t1, t2);
+    let p5 = _mm_adds_epi16(p3, p4);
+    let p5 = _mm_adds_epi16(p5, _mm_mulhrs_epi16(p5, _mm_set1_epi16(5763))); // 0.175875602
+
+    let t0 = _mm_mulhrs_epi16(t0, _mm_set1_epi16(9786)); // 0.298631336
+    let t1 = _mm_adds_epi16(
+        _mm_adds_epi16(t1, t1),
+        _mm_mulhrs_epi16(t1, _mm_set1_epi16(1741)), // 0.053119869
+    );
+    let t2 = _mm_adds_epi16(
+        _mm_adds_epi16(t2, _mm_adds_epi16(t2, t2)),
+        _mm_mulhrs_epi16(t2, _mm_set1_epi16(2383)), // 0.072711026
+    );
+    let t3 = _mm_adds_epi16(t3, _mm_mulhrs_epi16(t3, _mm_set1_epi16(16427))); // 0.501321110
+
+    let p1 = _mm_subs_epi16(p5, _mm_mulhrs_epi16(p1, _mm_set1_epi16(29490))); // 0.899976223
+    let p2 = _mm_subs_epi16(
+        _mm_subs_epi16(_mm_subs_epi16(p5, p2), p2),
+        _mm_mulhrs_epi16(p2, _mm_set1_epi16(18446)), // 0.562915447
+    );
+
+    let p3 = _mm_subs_epi16(
+        _mm_mulhrs_epi16(p3, _mm_set1_epi16(-31509)), // -0.961570560
+        p3,
+    );
+    let p4 = _mm_mulhrs_epi16(p4, _mm_set1_epi16(-12785)); // -0.390180644
+
+    let t3 = _mm_adds_epi16(_mm_adds_epi16(p1, p4), t3);
+    let t2 = _mm_adds_epi16(_mm_adds_epi16(p2, p3), t2);
+    let t1 = _mm_adds_epi16(_mm_adds_epi16(p2, p4), t1);
+    let t0 = _mm_adds_epi16(_mm_adds_epi16(p1, p3), t0);
+
+    data[0] = _mm_adds_epi16(x0, t3);
+    data[7] = _mm_subs_epi16(x0, t3);
+    data[1] = _mm_adds_epi16(x1, t2);
+    data[6] = _mm_subs_epi16(x1, t2);
+    data[2] = _mm_adds_epi16(x2, t1);
+    data[5] = _mm_subs_epi16(x2, t1);
+    data[3] = _mm_adds_epi16(x3, t0);
+    data[4] = _mm_subs_epi16(x3, t0);
+}
+
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+#[target_feature(enable = "ssse3")]
+unsafe fn transpose8(data: &mut [__m128i; 8]) {
+    // Transpose a 8x8 matrix with a sequence of interleaving operations.
+    // Naming: dABl contains elements from the *l*ower halves of vectors A and B, interleaved, i.e.
+    // A0 B0 A1 B1 ...
+    // dABCDll contains elements from the lower quarter (ll) of vectors A, B, C, D, interleaved -
+    // A0 B0 C0 D0 A1 B1 C1 D1 ...
+    let d01l = _mm_unpacklo_epi16(data[0], data[1]);
+    let d23l = _mm_unpacklo_epi16(data[2], data[3]);
+    let d45l = _mm_unpacklo_epi16(data[4], data[5]);
+    let d67l = _mm_unpacklo_epi16(data[6], data[7]);
+    let d01h = _mm_unpackhi_epi16(data[0], data[1]);
+    let d23h = _mm_unpackhi_epi16(data[2], data[3]);
+    let d45h = _mm_unpackhi_epi16(data[4], data[5]);
+    let d67h = _mm_unpackhi_epi16(data[6], data[7]);
+    // Operating on 32-bits will interleave *consecutive pairs* of 16-bit integers.
+    let d0123ll = _mm_unpacklo_epi32(d01l, d23l);
+    let d0123lh = _mm_unpackhi_epi32(d01l, d23l);
+    let d4567ll = _mm_unpacklo_epi32(d45l, d67l);
+    let d4567lh = _mm_unpackhi_epi32(d45l, d67l);
+    let d0123hl = _mm_unpacklo_epi32(d01h, d23h);
+    let d0123hh = _mm_unpackhi_epi32(d01h, d23h);
+    let d4567hl = _mm_unpacklo_epi32(d45h, d67h);
+    let d4567hh = _mm_unpackhi_epi32(d45h, d67h);
+    // Operating on 64-bits will interleave *consecutive quadruples* of 16-bit integers.
+    data[0] = _mm_unpacklo_epi64(d0123ll, d4567ll);
+    data[1] = _mm_unpackhi_epi64(d0123ll, d4567ll);
+    data[2] = _mm_unpacklo_epi64(d0123lh, d4567lh);
+    data[3] = _mm_unpackhi_epi64(d0123lh, d4567lh);
+    data[4] = _mm_unpacklo_epi64(d0123hl, d4567hl);
+    data[5] = _mm_unpackhi_epi64(d0123hl, d4567hl);
+    data[6] = _mm_unpacklo_epi64(d0123hh, d4567hh);
+    data[7] = _mm_unpackhi_epi64(d0123hh, d4567hh);
+}
+
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+#[target_feature(enable = "ssse3")]
+pub unsafe fn dequantize_and_idct_block_8x8(
+    coefficients: &[i16; 64],
+    quantization_table: &[u16; 64],
+    output_linestride: usize,
+    output: &mut [u8],
+) {
+    // The loop below will write to positions [output_linestride * i, output_linestride * i + 8)
+    // for 0<=i<8. Thus, the last accessed position is at an offset of output_linestrade * 7 + 7,
+    // and if that position is in-bounds, so are all other accesses.
+    assert!(
+        output.len()
+            > output_linestride
+                .checked_mul(7)
+                .unwrap()
+                .checked_add(7)
+                .unwrap()
+    );
+
+    #[cfg(target_arch = "x86")]
+    use std::arch::x86::*;
+    #[cfg(target_arch = "x86_64")]
+    use std::arch::x86_64::*;
+
+    const SHIFT: i32 = 3;
+
+    // Read the DCT coefficients, scale them up and dequantize them.
+    let mut data = [_mm_setzero_si128(); 8];
+    for i in 0..8 {
+        data[i] = _mm_slli_epi16(
+            _mm_mullo_epi16(
+                _mm_loadu_si128(coefficients.as_ptr().wrapping_add(i * 8) as *const _),
+                _mm_loadu_si128(quantization_table.as_ptr().wrapping_add(i * 8) as *const _),
+            ),
+            SHIFT,
+        );
+    }
+
+    // Usual column IDCT - transpose - column IDCT - transpose approach.
+    idct8(&mut data);
+    transpose8(&mut data);
+    idct8(&mut data);
+    transpose8(&mut data);
+
+    for i in 0..8 {
+        let mut buf = [0u8; 16];
+        // The two passes of the IDCT algorithm give us a factor of 8, so the shift here is
+        // increased by 3.
+        // As values will be stored in a u8, they need to be 128-centered and not 0-centered.
+        // We add 128 with the appropriate shift for that purpose.
+        const OFFSET: i16 = 128 << (SHIFT + 3);
+        // We want rounding right shift, so we should add (1/2) << (SHIFT+3) before shifting.
+        const ROUNDING_BIAS: i16 = (1 << (SHIFT + 3)) >> 1;
+
+        let data_with_offset = _mm_adds_epi16(data[i], _mm_set1_epi16(OFFSET + ROUNDING_BIAS));
+
+        _mm_storeu_si128(
+            buf.as_mut_ptr() as *mut _,
+            _mm_packus_epi16(
+                _mm_srai_epi16(data_with_offset, SHIFT + 3),
+                _mm_setzero_si128(),
+            ),
+        );
+        std::ptr::copy_nonoverlapping::<u8>(
+            buf.as_ptr(),
+            output.as_mut_ptr().wrapping_add(output_linestride * i) as *mut _,
+            8,
+        );
+    }
+}
+
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+#[target_feature(enable = "ssse3")]
+pub unsafe fn color_convert_line_ycbcr(y: &[u8], cb: &[u8], cr: &[u8], output: &mut [u8]) -> usize {
+    assert!(output.len() % 3 == 0);
+    let num = output.len() / 3;
+    assert!(num <= y.len());
+    assert!(num <= cb.len());
+    assert!(num <= cr.len());
+    // _mm_loadu_si64 generates incorrect code for Rust <1.58. To circumvent this, we use a full
+    // 128-bit load, but that requires leaving an extra vector of border to the scalar code.
+    // From Rust 1.58 on, the _mm_loadu_si128 can be replaced with _mm_loadu_si64 and this
+    // .saturating_sub() can be removed.
+    let num_vecs = (num / 8).saturating_sub(1);
+
+    for i in 0..num_vecs {
+        const SHIFT: i32 = 6;
+        // Load.
+        let y = _mm_loadu_si128(y.as_ptr().wrapping_add(i * 8) as *const _);
+        let cb = _mm_loadu_si128(cb.as_ptr().wrapping_add(i * 8) as *const _);
+        let cr = _mm_loadu_si128(cr.as_ptr().wrapping_add(i * 8) as *const _);
+
+        // Convert to 16 bit.
+        let shuf16 = _mm_setr_epi8(
+            0, -0x7F, 1, -0x7F, 2, -0x7F, 3, -0x7F, 4, -0x7F, 5, -0x7F, 6, -0x7F, 7, -0x7F,
+        );
+        let y = _mm_slli_epi16(_mm_shuffle_epi8(y, shuf16), SHIFT);
+        let cb = _mm_slli_epi16(_mm_shuffle_epi8(cb, shuf16), SHIFT);
+        let cr = _mm_slli_epi16(_mm_shuffle_epi8(cr, shuf16), SHIFT);
+
+        // Add offsets
+        let c128 = _mm_set1_epi16(128 << SHIFT);
+        let y = _mm_adds_epi16(y, _mm_set1_epi16((1 << SHIFT) >> 1));
+        let cb = _mm_subs_epi16(cb, c128);
+        let cr = _mm_subs_epi16(cr, c128);
+
+        // Compute cr * 1.402, cb * 0.34414, cr * 0.71414, cb * 1.772
+        let cr_140200 = _mm_adds_epi16(_mm_mulhrs_epi16(cr, _mm_set1_epi16(13173)), cr);
+        let cb_034414 = _mm_mulhrs_epi16(cb, _mm_set1_epi16(11276));
+        let cr_071414 = _mm_mulhrs_epi16(cr, _mm_set1_epi16(23401));
+        let cb_177200 = _mm_adds_epi16(_mm_mulhrs_epi16(cb, _mm_set1_epi16(25297)), cb);
+
+        // Last conversion step.
+        let r = _mm_adds_epi16(y, cr_140200);
+        let g = _mm_subs_epi16(y, _mm_adds_epi16(cb_034414, cr_071414));
+        let b = _mm_adds_epi16(y, cb_177200);
+
+        // Shift back and convert to u8.
+        let zero = _mm_setzero_si128();
+        let r = _mm_packus_epi16(_mm_srai_epi16(r, SHIFT), zero);
+        let g = _mm_packus_epi16(_mm_srai_epi16(g, SHIFT), zero);
+        let b = _mm_packus_epi16(_mm_srai_epi16(b, SHIFT), zero);
+
+        // Shuffle rrrrrrrrggggggggbbbbbbbb to rgbrgbrgb...
+
+        // Control vectors for _mm_shuffle_epi8. -0x7F is selected so that the resulting position
+        // after _mm_shuffle_epi8 will be filled with 0, so that the r, g, and b vectors can then
+        // be OR-ed together.
+        let shufr = _mm_setr_epi8(
+            0, -0x7F, -0x7F, 1, -0x7F, -0x7F, 2, -0x7F, -0x7F, 3, -0x7F, -0x7F, 4, -0x7F, -0x7F, 5,
+        );
+        let shufg = _mm_setr_epi8(
+            -0x7F, 0, -0x7F, -0x7F, 1, -0x7F, -0x7F, 2, -0x7F, -0x7F, 3, -0x7F, -0x7F, 4, -0x7F,
+            -0x7F,
+        );
+        let shufb = _mm_alignr_epi8(shufg, shufg, 15);
+
+        let rgb_low = _mm_or_si128(
+            _mm_shuffle_epi8(r, shufr),
+            _mm_or_si128(_mm_shuffle_epi8(g, shufg), _mm_shuffle_epi8(b, shufb)),
+        );
+
+        // For the next part of the rgb vectors, we need to select R values from 6 up, G and B from
+        // 5 up. The highest bit of -0x7F + 6 is still set, so the corresponding location will
+        // still be 0.
+        let shufr1 = _mm_add_epi8(shufb, _mm_set1_epi8(6));
+        let shufg1 = _mm_add_epi8(shufr, _mm_set1_epi8(5));
+        let shufb1 = _mm_add_epi8(shufg, _mm_set1_epi8(5));
+
+        let rgb_hi = _mm_or_si128(
+            _mm_shuffle_epi8(r, shufr1),
+            _mm_or_si128(_mm_shuffle_epi8(g, shufg1), _mm_shuffle_epi8(b, shufb1)),
+        );
+
+        let mut data = [0u8; 32];
+        _mm_storeu_si128(data.as_mut_ptr() as *mut _, rgb_low);
+        _mm_storeu_si128(data.as_mut_ptr().wrapping_add(16) as *mut _, rgb_hi);
+        std::ptr::copy_nonoverlapping::<u8>(
+            data.as_ptr(),
+            output.as_mut_ptr().wrapping_add(24 * i),
+            24,
+        );
+    }
+
+    num_vecs * 8
+}