From 9940519e3bcf486c0ba50fad443a64d1a2c722d7 Mon Sep 17 00:00:00 2001
From: Alexander <homm86@gmail.com>
Date: Sun, 10 Sep 2017 02:37:04 +0300
Subject: [PATCH] =?UTF-8?q?RGB=20=E2=86=92=20L=202.2=20times=20faster?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 libImaging/Convert.c | 42 +++++++++++++++++++++++++++++++++++++-----
 1 file changed, 37 insertions(+), 5 deletions(-)

diff --git a/libImaging/Convert.c b/libImaging/Convert.c
index b360c548da1..ab6b12c3a2c 100644
--- a/libImaging/Convert.c
+++ b/libImaging/Convert.c
@@ -217,10 +217,25 @@ rgb2bit(UINT8* out, const UINT8* in, int xsize)
 static void
 rgb2l(UINT8* out, const UINT8* in, int xsize)
 {
-    int x;
-    for (x = 0; x < xsize; x++, in += 4)
+    int x = 0;
+    __m128i coeff = _mm_set_epi16(
+        0, 3735, 19235, 9798, 0, 3735, 19235, 9798);
+    for (; x < xsize - 3; x += 4, in += 16) {
+        __m128i pix0, pix1;
+        __m128i source = _mm_loadu_si128((__m128i*)in);
+        pix0 = _mm_unpacklo_epi8(source, _mm_setzero_si128());
+        pix1 = _mm_unpackhi_epi8(source, _mm_setzero_si128());
+        pix0 = _mm_madd_epi16(pix0, coeff);
+        pix1 = _mm_madd_epi16(pix1, coeff);
+        pix0 = _mm_hadd_epi32(pix0, pix1);
+        pix0 = _mm_srli_epi32(pix0, 15);
+        pix0 = _mm_packus_epi32(pix0, pix0);
+        pix0 = _mm_packus_epi16(pix0, pix0);
+        *(UINT32*)&out[x] = _mm_cvtsi128_si32(pix0);
+    }
+    for (; x < xsize; x++, in += 4)
         /* ITU-R Recommendation 601-2 (assuming nonlinear RGB) */
-        *out++ = L24(in) >> 16;
+        out[x] = L24(in) >> 16;
 }
 
 static void
@@ -426,8 +441,25 @@ rgb2rgba(UINT8* out, const UINT8* in, int xsize)
 static void
 rgba2la(UINT8* out, const UINT8* in, int xsize)
 {
-    int x;
-    for (x = 0; x < xsize; x++, in += 4, out += 4) {
+    int x = 0;
+    __m128i coeff = _mm_set_epi16(
+        0, 3735, 19235, 9798, 0, 3735, 19235, 9798);
+    for (; x < xsize - 3; x += 4, in += 16, out += 16) {
+        __m128i pix0, pix1;
+        __m128i source = _mm_loadu_si128((__m128i*)in);
+        __m128i alpha = _mm_and_si128(source, _mm_set1_epi32(0xff000000));
+        pix0 = _mm_unpacklo_epi8(source, _mm_setzero_si128());
+        pix1 = _mm_unpackhi_epi8(source, _mm_setzero_si128());
+        pix0 = _mm_madd_epi16(pix0, coeff);
+        pix1 = _mm_madd_epi16(pix1, coeff);
+        pix0 = _mm_hadd_epi32(pix0, pix1);
+        pix0 = _mm_srli_epi32(pix0, 15);
+        pix0 = _mm_shuffle_epi8(pix0, _mm_set_epi8(
+            -1,12,12,12, -1,8,8,8, -1,4,4,4, -1,0,0,0));
+        pix0 = _mm_or_si128(pix0, alpha);
+        _mm_storeu_si128((__m128i*)out, pix0);
+    }
+    for (; x < xsize; x++, in += 4, out += 4) {
         /* ITU-R Recommendation 601-2 (assuming nonlinear RGB) */
         out[0] = out[1] = out[2] = L24(in) >> 16;
         out[3] = in[3];