From b4e226714d0e73aba69e6bfbfa91993a27213e05 Mon Sep 17 00:00:00 2001
From: Alexander <homm86@gmail.com>
Date: Tue, 15 Aug 2017 23:05:12 +0300
Subject: [PATCH 01/11] Faster ImagingUnpackRGB

---
 libImaging/Imaging.h |  1 -
 libImaging/Unpack.c  | 27 +++++++++++++++++++--------
 2 files changed, 19 insertions(+), 9 deletions(-)

diff --git a/libImaging/Imaging.h b/libImaging/Imaging.h
index c2c640f0b..6bcd2401b 100644
--- a/libImaging/Imaging.h
+++ b/libImaging/Imaging.h
@@ -467,7 +467,6 @@ typedef void (*ImagingShuffler)(UINT8* out, const UINT8* in, int pixels);
 /* Public shufflers */
 extern void ImagingPackRGB(UINT8* out, const UINT8* in, int pixels);
 extern void ImagingPackBGR(UINT8* out, const UINT8* in, int pixels);
-extern void ImagingUnpackRGB(UINT8* out, const UINT8* in, int pixels);
 extern void ImagingUnpackBGR(UINT8* out, const UINT8* in, int pixels);
 extern void ImagingUnpackYCC(UINT8* out, const UINT8* in, int pixels);
 extern void ImagingUnpackYCCA(UINT8* out, const UINT8* in, int pixels);
diff --git a/libImaging/Unpack.c b/libImaging/Unpack.c
index dfc8a4d7e..86a2d9737 100644
--- a/libImaging/Unpack.c
+++ b/libImaging/Unpack.c
@@ -463,19 +463,30 @@ unpackP4L(UINT8* out, const UINT8* in, int pixels)
     }
 }
 
+
+#ifdef WORDS_BIGENDIAN
+    #define MAKE_UINT32(u0, u1, u2, u3) (u3 | (u2<<8) | (u1<<16) | (u0<<24))
+    #define CHANNEL_4_MASK 0x000000ff
+#else
+    #define MAKE_UINT32(u0, u1, u2, u3) (u0 | (u1<<8) | (u2<<16) | (u3<<24))
+    #define CHANNEL_4_MASK 0xff000000
+#endif
+
 /* Unpack to "RGB" image */
 
 void
-ImagingUnpackRGB(UINT8* out, const UINT8* in, int pixels)
+ImagingUnpackRGB(UINT8* _out, const UINT8* in, int pixels)
 {
-    int i;
+    int i = 0;
+    UINT32* out = (UINT32*) _out;
     /* RGB triplets */
-    for (i = 0; i < pixels; i++) {
-        out[R] = in[0];
-        out[G] = in[1];
-        out[B] = in[2];
-        out[A] = 255;
-        out += 4; in += 3;
+    for (; i < pixels-1; i++) {
+        out[i] = CHANNEL_4_MASK | *(UINT32*)&in[0];
+        in += 3;
+    }
+    for (; i < pixels; i++) {
+        out[i] = MAKE_UINT32(in[0], in[1], in[2], 255);
+        in += 3;
     }
 }
 

From 762b09cdb5b5ee24a38ae7e60cf2c7ff2863041f Mon Sep 17 00:00:00 2001
From: Alexander <homm86@gmail.com>
Date: Fri, 18 Aug 2017 01:13:50 +0300
Subject: [PATCH 02/11] unroll ImagingPackRGB

---
 libImaging/Pack.c | 16 ++++++++++------
 1 file changed, 10 insertions(+), 6 deletions(-)

diff --git a/libImaging/Pack.c b/libImaging/Pack.c
index 621936351..b624f1956 100644
--- a/libImaging/Pack.c
+++ b/libImaging/Pack.c
@@ -227,13 +227,17 @@ packLAL(UINT8* out, const UINT8* in, int pixels)
 void
 ImagingPackRGB(UINT8* out, const UINT8* in, int pixels)
 {
-    int i;
+    int i = 0;
     /* RGB triplets */
-    for (i = 0; i < pixels; i++) {
-        out[0] = in[R];
-        out[1] = in[G];
-        out[2] = in[B];
-        out += 3; in += 4;
+    for (; i < pixels-1; i++) {
+        ((UINT32*)out)[0] = ((UINT32*)in)[i];
+        out += 3;
+    }
+    for (; i < pixels; i++) {
+        out[0] = in[i*4+R];
+        out[1] = in[i*4+G];
+        out[2] = in[i*4+B];
+        out += 3;
     }
 }
 

From 00b65ce1c3f9578d8067518eacc38a0b32c0c120 Mon Sep 17 00:00:00 2001
From: Alexander <homm86@gmail.com>
Date: Sat, 19 Aug 2017 15:27:39 +0300
Subject: [PATCH 03/11] unused

---
 libImaging/Imaging.h | 3 ---
 1 file changed, 3 deletions(-)

diff --git a/libImaging/Imaging.h b/libImaging/Imaging.h
index 6bcd2401b..0a8bff32c 100644
--- a/libImaging/Imaging.h
+++ b/libImaging/Imaging.h
@@ -465,12 +465,9 @@ extern int ImagingZipEncodeCleanup(ImagingCodecState state);
 typedef void (*ImagingShuffler)(UINT8* out, const UINT8* in, int pixels);
 
 /* Public shufflers */
-extern void ImagingPackRGB(UINT8* out, const UINT8* in, int pixels);
 extern void ImagingPackBGR(UINT8* out, const UINT8* in, int pixels);
-extern void ImagingUnpackBGR(UINT8* out, const UINT8* in, int pixels);
 extern void ImagingUnpackYCC(UINT8* out, const UINT8* in, int pixels);
 extern void ImagingUnpackYCCA(UINT8* out, const UINT8* in, int pixels);
-extern void ImagingUnpackYCbCr(UINT8* out, const UINT8* in, int pixels);
 
 extern void ImagingConvertRGB2YCbCr(UINT8* out, const UINT8* in, int pixels);
 extern void ImagingConvertYCbCr2RGB(UINT8* out, const UINT8* in, int pixels);

From 58542fdfb90883d551b847fe110acef74a8c4526 Mon Sep 17 00:00:00 2001
From: Alexander <homm86@gmail.com>
Date: Sat, 19 Aug 2017 15:30:41 +0300
Subject: [PATCH 04/11] Move WORDS_BIGENDIAN to ImagingUtils.h

---
 libImaging/Bands.c        |  7 -------
 libImaging/Imaging.h      |  4 ++++
 libImaging/ImagingUtils.h | 13 +++++++++++++
 libImaging/Resample.c     |  6 ------
 libImaging/Unpack.c       | 10 +---------
 5 files changed, 18 insertions(+), 22 deletions(-)
 create mode 100644 libImaging/ImagingUtils.h

diff --git a/libImaging/Bands.c b/libImaging/Bands.c
index 758734b52..89cfd33db 100644
--- a/libImaging/Bands.c
+++ b/libImaging/Bands.c
@@ -22,13 +22,6 @@
 #define CLIP(x) ((x) <= 0 ? 0 : (x) < 256 ? (x) : 255)
 
 
-#ifdef WORDS_BIGENDIAN
-    #define MAKE_UINT32(u0, u1, u2, u3) (u3 | (u2<<8) | (u1<<16) | (u0<<24))
-#else
-    #define MAKE_UINT32(u0, u1, u2, u3) (u0 | (u1<<8) | (u2<<16) | (u3<<24))
-#endif
-
-
 Imaging
 ImagingGetBand(Imaging imIn, int band)
 {
diff --git a/libImaging/Imaging.h b/libImaging/Imaging.h
index 0a8bff32c..2031442b2 100644
--- a/libImaging/Imaging.h
+++ b/libImaging/Imaging.h
@@ -509,6 +509,10 @@ extern Py_ssize_t _imaging_tell_pyFd(PyObject *fd);
 #define	IMAGING_CODEC_CONFIG	-8
 #define	IMAGING_CODEC_MEMORY	-9
 
+
+
+#include "ImagingUtils.h"
+
 #if defined(__cplusplus)
 }
 #endif
diff --git a/libImaging/ImagingUtils.h b/libImaging/ImagingUtils.h
new file mode 100644
index 000000000..d2a8f98a6
--- /dev/null
+++ b/libImaging/ImagingUtils.h
@@ -0,0 +1,13 @@
+#ifdef WORDS_BIGENDIAN
+    #define MAKE_UINT32(u0, u1, u2, u3) (u3 | (u2<<8) | (u1<<16) | (u0<<24))
+    #define MASK_UINT32_CHANNEL_0 0xff000000
+    #define MASK_UINT32_CHANNEL_1 0x00ff0000
+    #define MASK_UINT32_CHANNEL_2 0x0000ff00
+    #define MASK_UINT32_CHANNEL_3 0x000000ff
+#else
+    #define MAKE_UINT32(u0, u1, u2, u3) (u0 | (u1<<8) | (u2<<16) | (u3<<24))
+    #define MASK_UINT32_CHANNEL_0 0x000000ff
+    #define MASK_UINT32_CHANNEL_1 0x0000ff00
+    #define MASK_UINT32_CHANNEL_2 0x00ff0000
+    #define MASK_UINT32_CHANNEL_3 0xff000000
+#endif
\ No newline at end of file
diff --git a/libImaging/Resample.c b/libImaging/Resample.c
index afd045f7a..ff56c535b 100644
--- a/libImaging/Resample.c
+++ b/libImaging/Resample.c
@@ -5,12 +5,6 @@
 
 #define ROUND_UP(f) ((int) ((f) >= 0.0 ? (f) + 0.5F : (f) - 0.5F))
 
-#ifdef WORDS_BIGENDIAN
-    #define MAKE_UINT32(u0, u1, u2, u3) (u3 | (u2<<8) | (u1<<16) | (u0<<24))
-#else
-    #define MAKE_UINT32(u0, u1, u2, u3) (u0 | (u1<<8) | (u2<<16) | (u3<<24))
-#endif
-
 
 struct filter {
     double (*filter)(double x);
diff --git a/libImaging/Unpack.c b/libImaging/Unpack.c
index 86a2d9737..c294287a6 100644
--- a/libImaging/Unpack.c
+++ b/libImaging/Unpack.c
@@ -464,14 +464,6 @@ unpackP4L(UINT8* out, const UINT8* in, int pixels)
 }
 
 
-#ifdef WORDS_BIGENDIAN
-    #define MAKE_UINT32(u0, u1, u2, u3) (u3 | (u2<<8) | (u1<<16) | (u0<<24))
-    #define CHANNEL_4_MASK 0x000000ff
-#else
-    #define MAKE_UINT32(u0, u1, u2, u3) (u0 | (u1<<8) | (u2<<16) | (u3<<24))
-    #define CHANNEL_4_MASK 0xff000000
-#endif
-
 /* Unpack to "RGB" image */
 
 void
@@ -481,7 +473,7 @@ ImagingUnpackRGB(UINT8* _out, const UINT8* in, int pixels)
     UINT32* out = (UINT32*) _out;
     /* RGB triplets */
     for (; i < pixels-1; i++) {
-        out[i] = CHANNEL_4_MASK | *(UINT32*)&in[0];
+        out[i] = MASK_UINT32_CHANNEL_3 | *(UINT32*)&in[0];
         in += 3;
     }
     for (; i < pixels; i++) {

From d64f163760c0790c1a468cbb80087dd42dbc5458 Mon Sep 17 00:00:00 2001
From: Alexander <homm86@gmail.com>
Date: Sat, 19 Aug 2017 15:47:04 +0300
Subject: [PATCH 05/11] Move MULDIV255 and SHIFTFORDIV255 to ImagingUtils.h

---
 libImaging/AlphaComposite.c |  3 ---
 libImaging/Convert.c        |  4 ----
 libImaging/Draw.c           |  4 ----
 libImaging/ImagingUtils.h   | 10 +++++++++-
 libImaging/Pack.c           |  3 ---
 libImaging/Paste.c          | 13 ++-----------
 6 files changed, 11 insertions(+), 26 deletions(-)

diff --git a/libImaging/AlphaComposite.c b/libImaging/AlphaComposite.c
index 7e99c59ce..00d28f956 100644
--- a/libImaging/AlphaComposite.c
+++ b/libImaging/AlphaComposite.c
@@ -77,9 +77,6 @@ ImagingAlphaComposite(Imaging imDst, Imaging imSrc)
                 UINT16 coef1 = src->a * 255 * 255 * 128 / outa255;
                 UINT16 coef2 = 255 * 128 - coef1;
 
-                #define SHIFTFORDIV255(a)\
-                    ((((a) >> 8) + a) >> 8)
-
                 tmpr = src->r * coef1 + dst->r * coef2 + (0x80 << 7);
                 out->r = SHIFTFORDIV255(tmpr) >> 7;
                 tmpg = src->g * coef1 + dst->g * coef2 + (0x80 << 7);
diff --git a/libImaging/Convert.c b/libImaging/Convert.c
index f4253bda4..f820b1f6d 100644
--- a/libImaging/Convert.c
+++ b/libImaging/Convert.c
@@ -41,10 +41,6 @@
 #define CLIP(v) ((v) <= 0 ? 0 : (v) >= 255 ? 255 : (v))
 #define CLIP16(v) ((v) <= -32768 ? -32768 : (v) >= 32767 ? 32767 : (v))
 
-/* like (a * b + 127) / 255), but much faster on most platforms */
-#define MULDIV255(a, b, tmp)\
-        (tmp = (a) * (b) + 128, ((((tmp) >> 8) + (tmp)) >> 8))
-
 /* ITU-R Recommendation 601-2 (assuming nonlinear RGB) */
 #define L(rgb)\
     ((INT32) (rgb)[0]*299 + (INT32) (rgb)[1]*587 + (INT32) (rgb)[2]*114)
diff --git a/libImaging/Draw.c b/libImaging/Draw.c
index 65ab34a66..a4d24133c 100644
--- a/libImaging/Draw.c
+++ b/libImaging/Draw.c
@@ -42,10 +42,6 @@
 #define INK8(ink) (*(UINT8*)ink)
 #define INK32(ink) (*(INT32*)ink)
 
-/* like (a * b + 127) / 255), but much faster on most platforms */
-#define MULDIV255(a, b, tmp)\
-        (tmp = (a) * (b) + 128, ((((tmp) >> 8) + (tmp)) >> 8))
-
 #define BLEND(mask, in1, in2, tmp1, tmp2)\
         (MULDIV255(in1, 255 - mask, tmp1) + MULDIV255(in2, mask, tmp2))
 
diff --git a/libImaging/ImagingUtils.h b/libImaging/ImagingUtils.h
index d2a8f98a6..bb4ada311 100644
--- a/libImaging/ImagingUtils.h
+++ b/libImaging/ImagingUtils.h
@@ -10,4 +10,12 @@
     #define MASK_UINT32_CHANNEL_1 0x0000ff00
     #define MASK_UINT32_CHANNEL_2 0x00ff0000
     #define MASK_UINT32_CHANNEL_3 0xff000000
-#endif
\ No newline at end of file
+#endif
+
+
+#define SHIFTFORDIV255(a)\
+    ((((a) >> 8) + a) >> 8)
+
+/* like (a * b + 127) / 255), but much faster on most platforms */
+#define MULDIV255(a, b, tmp)\
+    (tmp = (a) * (b) + 128, SHIFTFORDIV255(tmp))
diff --git a/libImaging/Pack.c b/libImaging/Pack.c
index b624f1956..47086e091 100644
--- a/libImaging/Pack.c
+++ b/libImaging/Pack.c
@@ -72,9 +72,6 @@
 #define C64L C64N
 #endif
 
-/* like (a * b + 127) / 255), but much faster on most platforms */
-#define MULDIV255(a, b, tmp)\
-        (tmp = (a) * (b) + 128, ((((tmp) >> 8) + (tmp)) >> 8))
 
 static void
 pack1(UINT8* out, const UINT8* in, int pixels)
diff --git a/libImaging/Paste.c b/libImaging/Paste.c
index a18e5d6f2..402beff92 100644
--- a/libImaging/Paste.c
+++ b/libImaging/Paste.c
@@ -23,20 +23,11 @@
 
 #include "Imaging.h"
 
-/* like (a * b + 127) / 255), but much faster on most platforms */
-#define MULDIV255NEW(a, tmp)\
-        (tmp = (a) + 128, ((((tmp) >> 8) + (tmp)) >> 8))
-
-#define MULDIV255OLD(a, tmp)\
-        (((a) + 127) / 255)
-
-#define MULDIV255 MULDIV255NEW
-
 #define BLEND(mask, in1, in2, tmp1)\
-        MULDIV255(in1 * (255 - mask) + in2 * mask, tmp1)
+        (MULDIV255(in1, 255 - mask, tmp1) + MULDIV255(in2, mask, tmp1))
 
 #define PREBLEND(mask, in1, in2, tmp1)\
-        (MULDIV255(in1 * (255 - mask), tmp1) + in2)
+        (MULDIV255(in1, (255 - mask), tmp1) + in2)
 
 static inline void
 paste(Imaging imOut, Imaging imIn, int dx, int dy, int sx, int sy,

From fc40ef0c1ea3018fc8c72d5aaf52446537add385 Mon Sep 17 00:00:00 2001
From: Alexander <homm86@gmail.com>
Date: Sat, 19 Aug 2017 15:58:23 +0300
Subject: [PATCH 06/11] Move PREBLEND and BLEND to ImagingUtils.h

---
 libImaging/Draw.c         |  3 ---
 libImaging/ImagingUtils.h |  6 ++++++
 libImaging/Paste.c        | 41 +++++++++++++++++----------------------
 3 files changed, 24 insertions(+), 26 deletions(-)

diff --git a/libImaging/Draw.c b/libImaging/Draw.c
index a4d24133c..ef8a4db64 100644
--- a/libImaging/Draw.c
+++ b/libImaging/Draw.c
@@ -42,9 +42,6 @@
 #define INK8(ink) (*(UINT8*)ink)
 #define INK32(ink) (*(INT32*)ink)
 
-#define BLEND(mask, in1, in2, tmp1, tmp2)\
-        (MULDIV255(in1, 255 - mask, tmp1) + MULDIV255(in2, mask, tmp2))
-
 /*
  * Rounds around zero (up=away from zero, down=torwards zero)
  * This guarantees that ROUND_UP|DOWN(f) == -ROUND_UP|DOWN(-f)
diff --git a/libImaging/ImagingUtils.h b/libImaging/ImagingUtils.h
index bb4ada311..231142735 100644
--- a/libImaging/ImagingUtils.h
+++ b/libImaging/ImagingUtils.h
@@ -19,3 +19,9 @@
 /* like (a * b + 127) / 255), but much faster on most platforms */
 #define MULDIV255(a, b, tmp)\
     (tmp = (a) * (b) + 128, SHIFTFORDIV255(tmp))
+
+#define BLEND(mask, in1, in2, tmp1, tmp2)\
+        (MULDIV255(in1, 255 - mask, tmp1) + MULDIV255(in2, mask, tmp2))
+
+#define PREBLEND(mask, in1, in2, tmp1)\
+        (MULDIV255(in1, (255 - mask), tmp1) + in2)
diff --git a/libImaging/Paste.c b/libImaging/Paste.c
index 402beff92..8c62f9ed3 100644
--- a/libImaging/Paste.c
+++ b/libImaging/Paste.c
@@ -23,11 +23,6 @@
 
 #include "Imaging.h"
 
-#define BLEND(mask, in1, in2, tmp1)\
-        (MULDIV255(in1, 255 - mask, tmp1) + MULDIV255(in2, mask, tmp1))
-
-#define PREBLEND(mask, in1, in2, tmp1)\
-        (MULDIV255(in1, (255 - mask), tmp1) + in2)
 
 static inline void
 paste(Imaging imOut, Imaging imIn, int dx, int dy, int sx, int sy,
@@ -91,7 +86,7 @@ paste_mask_L(Imaging imOut, Imaging imIn, Imaging imMask,
     /* paste with mode "L" matte */
 
     int x, y;
-    unsigned int tmp1;
+    unsigned int tmp1, tmp2;
 
     if (imOut->image8) {
 
@@ -100,7 +95,7 @@ paste_mask_L(Imaging imOut, Imaging imIn, Imaging imMask,
             UINT8* in = imIn->image8[y+sy]+sx;
             UINT8* mask = imMask->image8[y+sy]+sx;
             for (x = 0; x < xsize; x++) {
-                *out = BLEND(*mask, *out, *in, tmp1);
+                *out = BLEND(*mask, *out, *in, tmp1, tmp2);
                 out++, in++, mask++;
             }
         }
@@ -113,10 +108,10 @@ paste_mask_L(Imaging imOut, Imaging imIn, Imaging imMask,
             UINT8* mask = (UINT8*) (imMask->image8[y+sy] + sx);
             for (x = 0; x < xsize; x++) {
                 UINT8 a = mask[0];
-                out[0] = BLEND(a, out[0], in[0], tmp1);
-                out[1] = BLEND(a, out[1], in[1], tmp1);
-                out[2] = BLEND(a, out[2], in[2], tmp1);
-                out[3] = BLEND(a, out[3], in[3], tmp1);
+                out[0] = BLEND(a, out[0], in[0], tmp1, tmp2);
+                out[1] = BLEND(a, out[1], in[1], tmp1, tmp2);
+                out[2] = BLEND(a, out[2], in[2], tmp1, tmp2);
+                out[3] = BLEND(a, out[3], in[3], tmp1, tmp2);
                 out += 4; in += 4; mask ++;
             }
         }
@@ -131,7 +126,7 @@ paste_mask_RGBA(Imaging imOut, Imaging imIn, Imaging imMask,
     /* paste with mode "RGBA" matte */
 
     int x, y;
-    unsigned int tmp1;
+    unsigned int tmp1, tmp2;
 
     if (imOut->image8) {
 
@@ -140,7 +135,7 @@ paste_mask_RGBA(Imaging imOut, Imaging imIn, Imaging imMask,
             UINT8* in = imIn->image8[y+sy]+sx;
             UINT8* mask = (UINT8*) imMask->image[y+sy]+sx*4+3;
             for (x = 0; x < xsize; x++) {
-                *out = BLEND(*mask, *out, *in, tmp1);
+                *out = BLEND(*mask, *out, *in, tmp1, tmp2);
                 out++, in++, mask += 4;
             }
         }
@@ -153,10 +148,10 @@ paste_mask_RGBA(Imaging imOut, Imaging imIn, Imaging imMask,
             UINT8* mask = (UINT8*) (imMask->image32[y+sy] + sx);
             for (x = 0; x < xsize; x++) {
                 UINT8 a = mask[3];
-                out[0] = BLEND(a, out[0], in[0], tmp1);
-                out[1] = BLEND(a, out[1], in[1], tmp1);
-                out[2] = BLEND(a, out[2], in[2], tmp1);
-                out[3] = BLEND(a, out[3], in[3], tmp1);
+                out[0] = BLEND(a, out[0], in[0], tmp1, tmp2);
+                out[1] = BLEND(a, out[1], in[1], tmp1, tmp2);
+                out[2] = BLEND(a, out[2], in[2], tmp1, tmp2);
+                out[3] = BLEND(a, out[3], in[3], tmp1, tmp2);
                 out += 4; in += 4; mask += 4;
             }
         }
@@ -364,7 +359,7 @@ fill_mask_L(Imaging imOut, const UINT8* ink, Imaging imMask,
     /* fill with mode "L" matte */
 
     int x, y, i;
-    unsigned int tmp1;
+    unsigned int tmp1, tmp2;
 
     if (imOut->image8) {
 
@@ -372,7 +367,7 @@ fill_mask_L(Imaging imOut, const UINT8* ink, Imaging imMask,
             UINT8* out = imOut->image8[y+dy]+dx;
             UINT8* mask = imMask->image8[y+sy]+sx;
             for (x = 0; x < xsize; x++) {
-                *out = BLEND(*mask, *out, ink[0], tmp1);
+                *out = BLEND(*mask, *out, ink[0], tmp1, tmp2);
                 out++, mask++;
             }
         }
@@ -384,7 +379,7 @@ fill_mask_L(Imaging imOut, const UINT8* ink, Imaging imMask,
             UINT8* mask = (UINT8*) imMask->image[y+sy]+sx;
             for (x = 0; x < xsize; x++) {
                 for (i = 0; i < pixelsize; i++) {
-                    *out = BLEND(*mask, *out, ink[i], tmp1);
+                    *out = BLEND(*mask, *out, ink[i], tmp1, tmp2);
                     out++;
                 }
                 mask++;
@@ -401,7 +396,7 @@ fill_mask_RGBA(Imaging imOut, const UINT8* ink, Imaging imMask,
     /* fill with mode "RGBA" matte */
 
     int x, y, i;
-    unsigned int tmp1;
+    unsigned int tmp1, tmp2;
 
     if (imOut->image8) {
 
@@ -410,7 +405,7 @@ fill_mask_RGBA(Imaging imOut, const UINT8* ink, Imaging imMask,
             UINT8* out = imOut->image8[y+dy]+dx;
             UINT8* mask = (UINT8*) imMask->image[y+sy]+sx;
             for (x = 0; x < xsize; x++) {
-                *out = BLEND(*mask, *out, ink[0], tmp1);
+                *out = BLEND(*mask, *out, ink[0], tmp1, tmp2);
                 out++, mask += 4;
             }
         }
@@ -424,7 +419,7 @@ fill_mask_RGBA(Imaging imOut, const UINT8* ink, Imaging imMask,
             UINT8* mask = (UINT8*) imMask->image[y+sy]+sx;
             for (x = 0; x < xsize; x++) {
                 for (i = 0; i < pixelsize; i++) {
-                    *out = BLEND(*mask, *out, ink[i], tmp1);
+                    *out = BLEND(*mask, *out, ink[i], tmp1, tmp2);
                     out++;
                 }
                 mask += 4;

From 40d30f678bdfe22677404b2c6f9f5e1cbcfd5010 Mon Sep 17 00:00:00 2001
From: Alexander <homm86@gmail.com>
Date: Tue, 22 Aug 2017 00:00:17 +0300
Subject: [PATCH 07/11] Use MAKE_UINT32 when possible for unpacking

---
 libImaging/Unpack.c | 203 +++++++++++++++++++-------------------------
 1 file changed, 85 insertions(+), 118 deletions(-)

diff --git a/libImaging/Unpack.c b/libImaging/Unpack.c
index c294287a6..53b9d3d04 100644
--- a/libImaging/Unpack.c
+++ b/libImaging/Unpack.c
@@ -313,26 +313,25 @@ unpackL4IR(UINT8* out, const UINT8* in, int pixels)
 }
 
 static void
-unpackLA(UINT8* out, const UINT8* in, int pixels)
+unpackLA(UINT8* _out, const UINT8* in, int pixels)
 {
     int i;
+    UINT32* out = (UINT32*) _out;
     /* LA, pixel interleaved */
     for (i = 0; i < pixels; i++) {
-        out[R] = out[G] = out[B] = in[0];
-        out[A] = in[1];
-        in += 2; out += 4;
+        out[i] = MAKE_UINT32(in[0], in[0], in[0], in[1]);
+        in += 2;
     }
 }
 
 static void
-unpackLAL(UINT8* out, const UINT8* in, int pixels)
+unpackLAL(UINT8* _out, const UINT8* in, int pixels)
 {
     int i;
+    UINT32* out = (UINT32*) _out;
     /* LA, line interleaved */
     for (i = 0; i < pixels; i++) {
-        out[R] = out[G] = out[B] = in[i];
-        out[A] = in[i+pixels];
-        out += 4;
+        out[i] = MAKE_UINT32(in[i], in[i], in[i], in[i+pixels]);
     }
 }
 
@@ -483,58 +482,50 @@ ImagingUnpackRGB(UINT8* _out, const UINT8* in, int pixels)
 }
 
 void
-unpackRGB16B(UINT8* out, const UINT8* in, int pixels)
+unpackRGB16B(UINT8* _out, const UINT8* in, int pixels)
 {
     int i;
+    UINT32* out = (UINT32*) _out;
     /* 16-bit RGB triplets, big-endian order */
     for (i = 0; i < pixels; i++) {
-        out[R] = in[0];
-        out[G] = in[2];
-        out[B] = in[4];
-        out[A] = 255;
-        out += 4; in += 6;
+        out[i] = MAKE_UINT32(in[0], in[2], in[4], 255);
+        in += 6;
     }
 }
 
 static void
-unpackRGBL(UINT8* out, const UINT8* in, int pixels)
+unpackRGBL(UINT8* _out, const UINT8* in, int pixels)
 {
     int i;
+    UINT32* out = (UINT32*) _out;
     /* RGB, line interleaved */
     for (i = 0; i < pixels; i++) {
-        out[R] = in[i];
-        out[G] = in[i+pixels];
-        out[B] = in[i+pixels+pixels];
-        out[A] = 255;
-        out += 4;
+        out[i] = MAKE_UINT32(in[i], in[i+pixels], in[i+pixels+pixels], 255);
     }
 }
 
 static void
-unpackRGBR(UINT8* out, const UINT8* in, int pixels)
+unpackRGBR(UINT8* _out, const UINT8* in, int pixels)
 {
     int i;
+    UINT32* out = (UINT32*) _out;
     /* RGB, bit reversed */
     for (i = 0; i < pixels; i++) {
-        out[R] = BITFLIP[in[0]];
-        out[G] = BITFLIP[in[1]];
-        out[B] = BITFLIP[in[2]];
-        out[A] = 255;
-        out += 4; in += 3;
+        out[i] = MAKE_UINT32(BITFLIP[in[0]], BITFLIP[in[1]],
+                             BITFLIP[in[2]], 255);
+        in += 3;
     }
 }
 
 void
-ImagingUnpackBGR(UINT8* out, const UINT8* in, int pixels)
+ImagingUnpackBGR(UINT8* _out, const UINT8* in, int pixels)
 {
     int i;
+    UINT32* out = (UINT32*) _out;
     /* RGB, reversed bytes */
     for (i = 0; i < pixels; i++) {
-        out[R] = in[2];
-        out[G] = in[1];
-        out[B] = in[0];
-        out[A] = 255;
-        out += 4; in += 3;
+        out[i] = MAKE_UINT32(in[2], in[1], in[0], 255);
+        in += 3;
     }
 }
 
@@ -659,118 +650,106 @@ ImagingUnpackRGBA4B(UINT8* out, const UINT8* in, int pixels)
 }
 
 static void
-ImagingUnpackBGRX(UINT8* out, const UINT8* in, int pixels)
+ImagingUnpackBGRX(UINT8* _out, const UINT8* in, int pixels)
 {
     int i;
+    UINT32* out = (UINT32*) _out;
     /* RGB, reversed bytes with padding */
     for (i = 0; i < pixels; i++) {
-        out[R] = in[2];
-        out[G] = in[1];
-        out[B] = in[0];
-        out[A] = 255;
-        out += 4; in += 4;
+        out[i] = MAKE_UINT32(in[2], in[1], in[0], 255);
+        in += 4;
     }
 }
 
 static void
-ImagingUnpackXRGB(UINT8* out, const UINT8* in, int pixels)
+ImagingUnpackXRGB(UINT8* _out, const UINT8* in, int pixels)
 {
     int i;
+    UINT32* out = (UINT32*) _out;
     /* RGB, leading pad */
     for (i = 0; i < pixels; i++) {
-        out[R] = in[1];
-        out[G] = in[2];
-        out[B] = in[3];
-        out[A] = 255;
-        out += 4; in += 4;
+        out[i] = MAKE_UINT32(in[1], in[2], in[3], 255);
+        in += 4;
     }
 }
 
 static void
-ImagingUnpackXBGR(UINT8* out, const UINT8* in, int pixels)
+ImagingUnpackXBGR(UINT8* _out, const UINT8* in, int pixels)
 {
     int i;
+    UINT32* out = (UINT32*) _out;
     /* RGB, reversed bytes, leading pad */
     for (i = 0; i < pixels; i++) {
-        out[R] = in[3];
-        out[G] = in[2];
-        out[B] = in[1];
-        out[A] = 255;
-        out += 4; in += 4;
+        out[i] = MAKE_UINT32(in[3], in[2], in[1], 255);
+        in += 4;
     }
 }
 
 /* Unpack to "RGBA" image */
 
 static void
-unpackRGBALA(UINT8* out, const UINT8* in, int pixels)
+unpackRGBALA(UINT8* _out, const UINT8* in, int pixels)
 {
     int i;
+    UINT32* out = (UINT32*) _out;
     /* greyscale with alpha */
     for (i = 0; i < pixels; i++) {
-        out[R] = out[G] = out[B] = in[0];
-        out[A] = in[1];
-        out += 4; in += 2;
+        out[i] = MAKE_UINT32(in[0], in[0], in[0], in[1]);
+        in += 2;
     }
 }
 
 static void
-unpackRGBALA16B(UINT8* out, const UINT8* in, int pixels)
+unpackRGBALA16B(UINT8* _out, const UINT8* in, int pixels)
 {
     int i;
+    UINT32* out = (UINT32*) _out;
     /* 16-bit greyscale with alpha, big-endian */
     for (i = 0; i < pixels; i++) {
-        out[R] = out[G] = out[B] = in[0];
-        out[A] = in[2];
-        out += 4; in += 4;
+        out[i] = MAKE_UINT32(in[0], in[0], in[0], in[2]);
+        in += 4;
     }
 }
 
 static void
-unpackRGBa(UINT8* out, const UINT8* in, int pixels)
+unpackRGBa(UINT8* _out, const UINT8* in, int pixels)
 {
     int i;
+    UINT32* out = (UINT32*) _out;
     /* premultiplied RGBA */
     for (i = 0; i < pixels; i++) {
         int a = in[3];
-        if (!a)
-            out[R] = out[G] = out[B] = out[A] = 0;
+        if ( ! a)
+            out[i] = 0;
         else if (a == 255) {
-            out[R] = in[0];
-            out[G] = in[1];
-            out[B] = in[2];
-            out[A] = a;
+            out[i] = MAKE_UINT32(in[0], in[1], in[2], a);
         } else {
-            out[R] = CLIP(in[0] * 255 / a);
-            out[G] = CLIP(in[1] * 255 / a);
-            out[B] = CLIP(in[2] * 255 / a);
-            out[A] = a;
+            out[i] = MAKE_UINT32(CLIP(in[0] * 255 / a),
+                                 CLIP(in[1] * 255 / a),
+                                 CLIP(in[2] * 255 / a), a);
         }
-        out += 4; in += 4;
+        in += 4;
     }
 }
 
 static void
-unpackBGRa(UINT8* out, const UINT8* in, int pixels)
+unpackBGRa(UINT8* _out, const UINT8* in, int pixels)
 {
     int i;
+    UINT32* out = (UINT32*) _out;
     /* premultiplied BGRA */
     for (i = 0; i < pixels; i++) {
         int a = in[3];
         if (!a)
-            out[R] = out[G] = out[B] = out[A] = 0;
+            out[i] = 0;
         else if (a == 255) {
-            out[R] = in[2];
-            out[G] = in[1];
-            out[B] = in[0];
-            out[A] = a;
+            out[i] = MAKE_UINT32(in[2], in[1], in[0], a);
         } else {
-            out[R] = CLIP(in[2] * 255 / a);
-            out[G] = CLIP(in[1] * 255 / a);
-            out[B] = CLIP(in[0] * 255 / a);
-            out[A] = a;
+            out[i] = MAKE_UINT32(CLIP(in[2] * 255 / a),
+                                 CLIP(in[1] * 255 / a),
+                                 CLIP(in[0] * 255 / a), a);
         }
-        out += 4; in += 4;
+        in += 4;
     }
 }
 
@@ -789,73 +768,62 @@ unpackRGBAI(UINT8* out, const UINT8* in, int pixels)
 }
 
 static void
-unpackRGBAL(UINT8* out, const UINT8* in, int pixels)
+unpackRGBAL(UINT8* _out, const UINT8* in, int pixels)
 {
     int i;
-
+    UINT32* out = (UINT32*) _out;
     /* RGBA, line interleaved */
     for (i = 0; i < pixels; i++) {
-        out[R] = in[i];
-        out[G] = in[i+pixels];
-        out[B] = in[i+pixels+pixels];
-        out[A] = in[i+pixels+pixels+pixels];
-        out += 4;
+        out[i] = MAKE_UINT32(in[i], in[i+pixels], in[i+pixels+pixels],
+                             in[i+pixels+pixels+pixels]);
     }
 }
 
 void
-unpackRGBA16B(UINT8* out, const UINT8* in, int pixels)
+unpackRGBA16B(UINT8* _out, const UINT8* in, int pixels)
 {
     int i;
+    UINT32* out = (UINT32*) _out;
     /* 16-bit RGBA, big-endian order */
     for (i = 0; i < pixels; i++) {
-        out[R] = in[0];
-        out[G] = in[2];
-        out[B] = in[4];
-        out[A] = in[6];
-        out += 4; in += 8;
+        out[i] = MAKE_UINT32(in[0], in[2], in[4], in[6]);
+        in += 8;
     }
 }
 
 static void
-unpackARGB(UINT8* out, const UINT8* in, int pixels)
+unpackARGB(UINT8* _out, const UINT8* in, int pixels)
 {
     int i;
+    UINT32* out = (UINT32*) _out;
     /* RGBA, leading pad */
     for (i = 0; i < pixels; i++) {
-        out[R] = in[1];
-        out[G] = in[2];
-        out[B] = in[3];
-        out[A] = in[0];
-        out += 4; in += 4;
+        out[i] = MAKE_UINT32(in[1], in[2], in[3], in[0]);
+        in += 4;
     }
 }
 
 static void
-unpackABGR(UINT8* out, const UINT8* in, int pixels)
+unpackABGR(UINT8* _out, const UINT8* in, int pixels)
 {
     int i;
+    UINT32* out = (UINT32*) _out;
     /* RGBA, reversed bytes */
     for (i = 0; i < pixels; i++) {
-        out[R] = in[3];
-        out[G] = in[2];
-        out[B] = in[1];
-        out[A] = in[0];
-        out += 4; in += 4;
+        out[i] = MAKE_UINT32(in[3], in[2], in[1], in[0]);
+        in += 4;
     }
 }
 
 static void
-unpackBGRA(UINT8* out, const UINT8* in, int pixels)
+unpackBGRA(UINT8* _out, const UINT8* in, int pixels)
 {
     int i;
+    UINT32* out = (UINT32*) _out;
     /* RGBA, reversed bytes */
     for (i = 0; i < pixels; i++) {
-        out[R] = in[2];
-        out[G] = in[1];
-        out[B] = in[0];
-        out[A] = in[3];
-        out += 4; in += 4;
+        out[i] = MAKE_UINT32(in[2], in[1], in[0], in[3]);
+        in += 4;
     }
 }
 
@@ -863,16 +831,14 @@ unpackBGRA(UINT8* out, const UINT8* in, int pixels)
 /* Unpack to "CMYK" image */
 
 static void
-unpackCMYKI(UINT8* out, const UINT8* in, int pixels)
+unpackCMYKI(UINT8* _out, const UINT8* in, int pixels)
 {
     int i;
+    UINT32* out = (UINT32*) _out;
     /* CMYK, inverted bytes (Photoshop 2.5) */
     for (i = 0; i < pixels; i++) {
-        out[C] = ~in[0];
-        out[M] = ~in[1];
-        out[Y] = ~in[2];
-        out[K] = ~in[3];
-        out += 4; in += 4;
+        out[i] = ~MAKE_UINT32(in[0], in[1], in[2], in[3]);
+        in += 4;
     }
 }
 
@@ -893,6 +859,7 @@ ImagingUnpackLAB(UINT8* out, const UINT8* in, int pixels)
     int i;
     /* LAB triplets */
     for (i = 0; i < pixels; i++) {
+        /* signed in outside world */
         out[0] = in[0];
         out[1] = in[1] ^ 128; /* signed in outside world */
         out[2] = in[2] ^ 128;

From 6d65123c10533289f2206d54addff1b31dc2b20c Mon Sep 17 00:00:00 2001
From: Alexander <homm86@gmail.com>
Date: Tue, 22 Aug 2017 00:50:42 +0300
Subject: [PATCH 08/11] Add some tests

---
 Tests/test_lib_pack.py | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/Tests/test_lib_pack.py b/Tests/test_lib_pack.py
index 83872c5d1..dbd5501fc 100644
--- a/Tests/test_lib_pack.py
+++ b/Tests/test_lib_pack.py
@@ -122,6 +122,7 @@ class TestLibPack(PillowTestCase):
         self.assertEqual(unpack("RGB", "BGRX", 4), (3, 2, 1))
         self.assertEqual(unpack("RGB", "XRGB", 4), (2, 3, 4))
         self.assertEqual(unpack("RGB", "XBGR", 4), (4, 3, 2))
+        self.assertEqual(unpack("RGB", "RGBX;L", 4), (1, 2, 3))
 
         self.assertEqual(unpack("RGBA", "RGBA", 4), (1, 2, 3, 4))
         self.assertEqual(unpack("RGBA", "RGBa", 4), (63, 127, 191, 4))
@@ -132,6 +133,8 @@ class TestLibPack(PillowTestCase):
         self.assertEqual(unpack("RGBA", "RGBA;15", 2), (8, 131, 0, 0))
         self.assertEqual(unpack("RGBA", "BGRA;15", 2), (0, 131, 8, 0))
         self.assertEqual(unpack("RGBA", "RGBA;4B", 2), (17, 0, 34, 0))
+        self.assertEqual(unpack("RGBA", "LA;16B", 4), (1, 1, 1, 3))
+        self.assertEqual(unpack("RGBA", "RGBA;16B", 8), (1, 3, 5, 7))
 
         self.assertEqual(unpack("RGBa", "RGBa", 4), (1, 2, 3, 4))
         self.assertEqual(unpack("RGBa", "BGRa", 4), (3, 2, 1, 4))

From ad2c9526976d64c1172e3938753be06afac7f13b Mon Sep 17 00:00:00 2001
From: Alexander <homm86@gmail.com>
Date: Tue, 22 Aug 2017 00:52:42 +0300
Subject: [PATCH 09/11] Add more tests

---
 Tests/test_lib_pack.py | 18 ++++++++++++++++++
 libImaging/Unpack.c    |  1 -
 2 files changed, 18 insertions(+), 1 deletion(-)

diff --git a/Tests/test_lib_pack.py b/Tests/test_lib_pack.py
index dbd5501fc..a1118a6d9 100644
--- a/Tests/test_lib_pack.py
+++ b/Tests/test_lib_pack.py
@@ -96,6 +96,17 @@ class TestLibPack(PillowTestCase):
         self.assertEqual(unpack_1("1", "1;R", 170), (0, X, 0, X, 0, X, 0, X))
         self.assertEqual(unpack_1("1", "1;IR", 170), (X, 0, X, 0, X, 0, X, 0))
 
+        self.assertEqual(unpack("P", "P;1", 1), 0)
+        self.assertEqual(unpack("P", "P;2", 2), 0)
+        self.assertEqual(unpack("P", "P;2L", 2), 0)
+        self.assertEqual(unpack("P", "P;4", 4), 0)
+        self.assertEqual(unpack("P", "P;4L", 4), 0)
+        self.assertEqual(unpack("P", "P",  8), 1)
+        self.assertEqual(unpack("P", "P;R", 8), 128)
+
+        self.assertEqual(unpack("PA", "PA",  16), (1, 2))
+        self.assertEqual(unpack("PA", "PA;L", 16), (1, 2))
+
         self.assertEqual(unpack("L", "L;2", 1), 0)
         self.assertEqual(unpack("L", "L;4", 1), 0)
         self.assertEqual(unpack("L", "L", 1), 1)
@@ -133,6 +144,8 @@ class TestLibPack(PillowTestCase):
         self.assertEqual(unpack("RGBA", "RGBA;15", 2), (8, 131, 0, 0))
         self.assertEqual(unpack("RGBA", "BGRA;15", 2), (0, 131, 8, 0))
         self.assertEqual(unpack("RGBA", "RGBA;4B", 2), (17, 0, 34, 0))
+        self.assertEqual(unpack("RGBA", "RGBA;L", 4), (1, 2, 3, 4))
+        self.assertEqual(unpack("RGBA", "LA", 2), (1, 1, 1, 2))
         self.assertEqual(unpack("RGBA", "LA;16B", 4), (1, 1, 1, 3))
         self.assertEqual(unpack("RGBA", "RGBA;16B", 8), (1, 3, 5, 7))
 
@@ -152,6 +165,11 @@ class TestLibPack(PillowTestCase):
         self.assertEqual(unpack("CMYK", "CMYK", 4), (1, 2, 3, 4))
         self.assertEqual(unpack("CMYK", "CMYK;I", 4), (254, 253, 252, 251))
 
+        self.assertEqual(unpack("CMYK", "C;I", 1), (254, 0, 0, 0))
+        self.assertEqual(unpack("CMYK", "M;I", 1), (0, 254, 0, 0))
+        self.assertEqual(unpack("CMYK", "Y;I", 1), (0, 0, 254, 0))
+        self.assertEqual(unpack("CMYK", "K;I", 1), (0, 0, 0, 254))
+
         self.assertRaises(ValueError, lambda: unpack("L", "L", 0))
         self.assertRaises(ValueError, lambda: unpack("RGB", "RGB", 2))
         self.assertRaises(ValueError, lambda: unpack("CMYK", "CMYK", 2))
diff --git a/libImaging/Unpack.c b/libImaging/Unpack.c
index 53b9d3d04..30740d259 100644
--- a/libImaging/Unpack.c
+++ b/libImaging/Unpack.c
@@ -859,7 +859,6 @@ ImagingUnpackLAB(UINT8* out, const UINT8* in, int pixels)
     int i;
     /* LAB triplets */
     for (i = 0; i < pixels; i++) {
-        /* signed in outside world */
         out[0] = in[0];
         out[1] = in[1] ^ 128; /* signed in outside world */
         out[2] = in[2] ^ 128;

From eeeafa2651df4323ae38874e26257a061ca2cc54 Mon Sep 17 00:00:00 2001
From: Alexander <homm86@gmail.com>
Date: Tue, 22 Aug 2017 22:30:20 +0300
Subject: [PATCH 10/11] One more tests

---
 Tests/test_lib_pack.py | 1 +
 1 file changed, 1 insertion(+)

diff --git a/Tests/test_lib_pack.py b/Tests/test_lib_pack.py
index a1118a6d9..bc4c75ddf 100644
--- a/Tests/test_lib_pack.py
+++ b/Tests/test_lib_pack.py
@@ -145,6 +145,7 @@ class TestLibPack(PillowTestCase):
         self.assertEqual(unpack("RGBA", "BGRA;15", 2), (0, 131, 8, 0))
         self.assertEqual(unpack("RGBA", "RGBA;4B", 2), (17, 0, 34, 0))
         self.assertEqual(unpack("RGBA", "RGBA;L", 4), (1, 2, 3, 4))
+        self.assertEqual(unpack("RGBA", "RGBA;I", 4), (254, 253, 252, 4))
         self.assertEqual(unpack("RGBA", "LA", 2), (1, 1, 1, 2))
         self.assertEqual(unpack("RGBA", "LA;16B", 4), (1, 1, 1, 3))
         self.assertEqual(unpack("RGBA", "RGBA;16B", 8), (1, 3, 5, 7))

From b705df3b648f63119958796685d5ac0274d0c53f Mon Sep 17 00:00:00 2001
From: Alexander <homm86@gmail.com>
Date: Sun, 27 Aug 2017 12:40:53 +0300
Subject: [PATCH 11/11] one div for blending

---
 libImaging/Draw.c         | 16 ++++++++--------
 libImaging/ImagingUtils.h |  9 ++++++---
 libImaging/Paste.c        | 36 ++++++++++++++++++------------------
 3 files changed, 32 insertions(+), 29 deletions(-)

diff --git a/libImaging/Draw.c b/libImaging/Draw.c
index ef8a4db64..3fae3d931 100644
--- a/libImaging/Draw.c
+++ b/libImaging/Draw.c
@@ -81,14 +81,14 @@ point32(Imaging im, int x, int y, int ink)
 static inline void
 point32rgba(Imaging im, int x, int y, int ink)
 {
-    unsigned int tmp1, tmp2;
+    unsigned int tmp1;
 
     if (x >= 0 && x < im->xsize && y >= 0 && y < im->ysize) {
         UINT8* out = (UINT8*) im->image[y]+x*4;
         UINT8* in = (UINT8*) &ink;
-        out[0] = BLEND(in[3], out[0], in[0], tmp1, tmp2);
-        out[1] = BLEND(in[3], out[1], in[1], tmp1, tmp2);
-        out[2] = BLEND(in[3], out[2], in[2], tmp1, tmp2);
+        out[0] = BLEND(in[3], out[0], in[0], tmp1);
+        out[1] = BLEND(in[3], out[1], in[1], tmp1);
+        out[2] = BLEND(in[3], out[2], in[2], tmp1);
     }
 }
 
@@ -140,7 +140,7 @@ static inline void
 hline32rgba(Imaging im, int x0, int y0, int x1, int ink)
 {
     int tmp;
-    unsigned int tmp1, tmp2;
+    unsigned int tmp1;
 
     if (y0 >= 0 && y0 < im->ysize) {
         if (x0 > x1)
@@ -157,9 +157,9 @@ hline32rgba(Imaging im, int x0, int y0, int x1, int ink)
             UINT8* out = (UINT8*) im->image[y0]+x0*4;
             UINT8* in = (UINT8*) &ink;
             while (x0 <= x1) {
-                out[0] = BLEND(in[3], out[0], in[0], tmp1, tmp2);
-                out[1] = BLEND(in[3], out[1], in[1], tmp1, tmp2);
-                out[2] = BLEND(in[3], out[2], in[2], tmp1, tmp2);
+                out[0] = BLEND(in[3], out[0], in[0], tmp1);
+                out[1] = BLEND(in[3], out[1], in[1], tmp1);
+                out[2] = BLEND(in[3], out[2], in[2], tmp1);
                 x0++; out += 4;
             }
         }
diff --git a/libImaging/ImagingUtils.h b/libImaging/ImagingUtils.h
index 231142735..5541fec30 100644
--- a/libImaging/ImagingUtils.h
+++ b/libImaging/ImagingUtils.h
@@ -20,8 +20,11 @@
 #define MULDIV255(a, b, tmp)\
     (tmp = (a) * (b) + 128, SHIFTFORDIV255(tmp))
 
-#define BLEND(mask, in1, in2, tmp1, tmp2)\
-        (MULDIV255(in1, 255 - mask, tmp1) + MULDIV255(in2, mask, tmp2))
+#define DIV255(a, tmp)\
+    (tmp = (a) + 128, SHIFTFORDIV255(tmp))
+
+#define BLEND(mask, in1, in2, tmp1)\
+    DIV255(in1 * (255 - mask) + in2 * mask, tmp1)
 
 #define PREBLEND(mask, in1, in2, tmp1)\
-        (MULDIV255(in1, (255 - mask), tmp1) + in2)
+    (MULDIV255(in1, (255 - mask), tmp1) + in2)
diff --git a/libImaging/Paste.c b/libImaging/Paste.c
index 8c62f9ed3..0bda25739 100644
--- a/libImaging/Paste.c
+++ b/libImaging/Paste.c
@@ -86,7 +86,7 @@ paste_mask_L(Imaging imOut, Imaging imIn, Imaging imMask,
     /* paste with mode "L" matte */
 
     int x, y;
-    unsigned int tmp1, tmp2;
+    unsigned int tmp1;
 
     if (imOut->image8) {
 
@@ -95,7 +95,7 @@ paste_mask_L(Imaging imOut, Imaging imIn, Imaging imMask,
             UINT8* in = imIn->image8[y+sy]+sx;
             UINT8* mask = imMask->image8[y+sy]+sx;
             for (x = 0; x < xsize; x++) {
-                *out = BLEND(*mask, *out, *in, tmp1, tmp2);
+                *out = BLEND(*mask, *out, *in, tmp1);
                 out++, in++, mask++;
             }
         }
@@ -108,10 +108,10 @@ paste_mask_L(Imaging imOut, Imaging imIn, Imaging imMask,
             UINT8* mask = (UINT8*) (imMask->image8[y+sy] + sx);
             for (x = 0; x < xsize; x++) {
                 UINT8 a = mask[0];
-                out[0] = BLEND(a, out[0], in[0], tmp1, tmp2);
-                out[1] = BLEND(a, out[1], in[1], tmp1, tmp2);
-                out[2] = BLEND(a, out[2], in[2], tmp1, tmp2);
-                out[3] = BLEND(a, out[3], in[3], tmp1, tmp2);
+                out[0] = BLEND(a, out[0], in[0], tmp1);
+                out[1] = BLEND(a, out[1], in[1], tmp1);
+                out[2] = BLEND(a, out[2], in[2], tmp1);
+                out[3] = BLEND(a, out[3], in[3], tmp1);
                 out += 4; in += 4; mask ++;
             }
         }
@@ -126,7 +126,7 @@ paste_mask_RGBA(Imaging imOut, Imaging imIn, Imaging imMask,
     /* paste with mode "RGBA" matte */
 
     int x, y;
-    unsigned int tmp1, tmp2;
+    unsigned int tmp1;
 
     if (imOut->image8) {
 
@@ -135,7 +135,7 @@ paste_mask_RGBA(Imaging imOut, Imaging imIn, Imaging imMask,
             UINT8* in = imIn->image8[y+sy]+sx;
             UINT8* mask = (UINT8*) imMask->image[y+sy]+sx*4+3;
             for (x = 0; x < xsize; x++) {
-                *out = BLEND(*mask, *out, *in, tmp1, tmp2);
+                *out = BLEND(*mask, *out, *in, tmp1);
                 out++, in++, mask += 4;
             }
         }
@@ -148,10 +148,10 @@ paste_mask_RGBA(Imaging imOut, Imaging imIn, Imaging imMask,
             UINT8* mask = (UINT8*) (imMask->image32[y+sy] + sx);
             for (x = 0; x < xsize; x++) {
                 UINT8 a = mask[3];
-                out[0] = BLEND(a, out[0], in[0], tmp1, tmp2);
-                out[1] = BLEND(a, out[1], in[1], tmp1, tmp2);
-                out[2] = BLEND(a, out[2], in[2], tmp1, tmp2);
-                out[3] = BLEND(a, out[3], in[3], tmp1, tmp2);
+                out[0] = BLEND(a, out[0], in[0], tmp1);
+                out[1] = BLEND(a, out[1], in[1], tmp1);
+                out[2] = BLEND(a, out[2], in[2], tmp1);
+                out[3] = BLEND(a, out[3], in[3], tmp1);
                 out += 4; in += 4; mask += 4;
             }
         }
@@ -359,7 +359,7 @@ fill_mask_L(Imaging imOut, const UINT8* ink, Imaging imMask,
     /* fill with mode "L" matte */
 
     int x, y, i;
-    unsigned int tmp1, tmp2;
+    unsigned int tmp1;
 
     if (imOut->image8) {
 
@@ -367,7 +367,7 @@ fill_mask_L(Imaging imOut, const UINT8* ink, Imaging imMask,
             UINT8* out = imOut->image8[y+dy]+dx;
             UINT8* mask = imMask->image8[y+sy]+sx;
             for (x = 0; x < xsize; x++) {
-                *out = BLEND(*mask, *out, ink[0], tmp1, tmp2);
+                *out = BLEND(*mask, *out, ink[0], tmp1);
                 out++, mask++;
             }
         }
@@ -379,7 +379,7 @@ fill_mask_L(Imaging imOut, const UINT8* ink, Imaging imMask,
             UINT8* mask = (UINT8*) imMask->image[y+sy]+sx;
             for (x = 0; x < xsize; x++) {
                 for (i = 0; i < pixelsize; i++) {
-                    *out = BLEND(*mask, *out, ink[i], tmp1, tmp2);
+                    *out = BLEND(*mask, *out, ink[i], tmp1);
                     out++;
                 }
                 mask++;
@@ -396,7 +396,7 @@ fill_mask_RGBA(Imaging imOut, const UINT8* ink, Imaging imMask,
     /* fill with mode "RGBA" matte */
 
     int x, y, i;
-    unsigned int tmp1, tmp2;
+    unsigned int tmp1;
 
     if (imOut->image8) {
 
@@ -405,7 +405,7 @@ fill_mask_RGBA(Imaging imOut, const UINT8* ink, Imaging imMask,
             UINT8* out = imOut->image8[y+dy]+dx;
             UINT8* mask = (UINT8*) imMask->image[y+sy]+sx;
             for (x = 0; x < xsize; x++) {
-                *out = BLEND(*mask, *out, ink[0], tmp1, tmp2);
+                *out = BLEND(*mask, *out, ink[0], tmp1);
                 out++, mask += 4;
             }
         }
@@ -419,7 +419,7 @@ fill_mask_RGBA(Imaging imOut, const UINT8* ink, Imaging imMask,
             UINT8* mask = (UINT8*) imMask->image[y+sy]+sx;
             for (x = 0; x < xsize; x++) {
                 for (i = 0; i < pixelsize; i++) {
-                    *out = BLEND(*mask, *out, ink[i], tmp1, tmp2);
+                    *out = BLEND(*mask, *out, ink[i], tmp1);
                     out++;
                 }
                 mask += 4;