Merge pull request #2015 from uploadcare/paste-speedup

Speedup paste with masks up to 80%

Tests/helper.py

@@ -238,3 +238,12 @@ if sys.platform == 'win32':
         IMCONVERT = os.path.join(IMCONVERT, 'convert.exe')
 else:
     IMCONVERT = 'convert'
+
+
+class cached_property(object):
+    def __init__(self, func):
+        self.func = func
+
+    def __get__(self, instance, cls=None):
+        result = instance.__dict__[self.func.__name__] = self.func(instance)
+        return result

Tests/test_image_paste.py

@@ -0,0 +1,252 @@
+from helper import PillowTestCase, cached_property
+
+from PIL import Image
+
+
+class TestImagingPaste(PillowTestCase):
+    masks = {}
+    size = 128
+
+    def assert_9points_image(self, im, expected):
+        expected = [
+                point[0]
+            if im.mode == 'L' else
+                point[:len(im.mode)]
+            for point in expected
+        ]
+        px = im.load()
+        actual = [
+            px[0, 0],
+            px[self.size // 2, 0],
+            px[self.size - 1, 0],
+            px[0, self.size // 2],
+            px[self.size // 2, self.size // 2],
+            px[self.size - 1, self.size // 2],
+            px[0, self.size - 1],
+            px[self.size // 2, self.size - 1],
+            px[self.size - 1, self.size - 1],
+        ]
+        self.assertEqual(actual, expected)
+
+    @cached_property
+    def mask_1(self):
+        mask = Image.new('1', (self.size, self.size))
+        px = mask.load()
+        for y in range(mask.height):
+            for x in range(mask.width):
+                px[y, x] = (x + y) % 2
+        return mask
+
+    @cached_property
+    def mask_L(self):
+        return self.gradient_L.transpose(Image.ROTATE_270)
+
+    @cached_property
+    def gradient_L(self):
+        gradient = Image.new('L', (self.size, self.size))
+        px = gradient.load()
+        for y in range(gradient.height):
+            for x in range(gradient.width):
+                px[y, x] = (x + y) % 255
+        return gradient
+
+    @cached_property
+    def gradient_RGB(self):
+        return Image.merge('RGB', [
+            self.gradient_L,
+            self.gradient_L.transpose(Image.ROTATE_90),
+            self.gradient_L.transpose(Image.ROTATE_180),
+        ])
+
+    @cached_property
+    def gradient_RGBA(self):
+        return Image.merge('RGBA', [
+            self.gradient_L,
+            self.gradient_L.transpose(Image.ROTATE_90),
+            self.gradient_L.transpose(Image.ROTATE_180),
+            self.gradient_L.transpose(Image.ROTATE_270),
+        ])
+
+    @cached_property
+    def gradient_RGBa(self):
+        return Image.merge('RGBa', [
+            self.gradient_L,
+            self.gradient_L.transpose(Image.ROTATE_90),
+            self.gradient_L.transpose(Image.ROTATE_180),
+            self.gradient_L.transpose(Image.ROTATE_270),
+        ])
+
+    def test_image_solid(self):
+        for mode in ('RGBA', 'RGB', 'L'):
+            im = Image.new(mode, (200, 200), 'red')
+            im2 = getattr(self, 'gradient_' + mode)
+
+            im.paste(im2, (12, 23))
+
+            im = im.crop((12, 23, im2.width + 12, im2.height + 23))
+            self.assert_image_equal(im, im2)
+
+    def test_image_mask_1(self):
+        for mode in ('RGBA', 'RGB', 'L'):
+            im = Image.new(mode, (200, 200), 'white')
+            im2 = getattr(self, 'gradient_' + mode)
+
+            im.paste(im2, (0, 0), self.mask_1)
+
+            self.assert_9points_image(im, [
+                (255, 255, 255, 255),
+                (255, 255, 255, 255),
+                (127, 254, 127, 0),
+                (255, 255, 255, 255),
+                (255, 255, 255, 255),
+                (191, 190, 63, 64),
+                (127, 0, 127, 254),
+                (191, 64, 63, 190),
+                (255, 255, 255, 255),
+            ])
+
+    def test_image_mask_L(self):
+        for mode in ('RGBA', 'RGB', 'L'):
+            im = Image.new(mode, (200, 200), 'white')
+            im2 = getattr(self, 'gradient_' + mode)
+
+            im.paste(im2, (0, 0), self.mask_L)
+
+            self.assert_9points_image(im, [
+                (128, 191, 255, 191),
+                (208, 239, 239, 208),
+                (255, 255, 255, 255),
+                (112, 111, 206, 207),
+                (192, 191, 191, 191),
+                (239, 239, 207, 207),
+                (128, 1, 128, 254),
+                (207, 113, 112, 207),
+                (255, 191, 128, 191),
+            ])
+
+    def test_image_mask_RGBA(self):
+        for mode in ('RGBA', 'RGB', 'L'):
+            im = Image.new(mode, (200, 200), 'white')
+            im2 = getattr(self, 'gradient_' + mode)
+
+            im.paste(im2, (0, 0), self.gradient_RGBA)
+
+            self.assert_9points_image(im, [
+                (128, 191, 255, 191),
+                (208, 239, 239, 208),
+                (255, 255, 255, 255),
+                (112, 111, 206, 207),
+                (192, 191, 191, 191),
+                (239, 239, 207, 207),
+                (128, 1, 128, 254),
+                (207, 113, 112, 207),
+                (255, 191, 128, 191),
+            ])
+
+    def test_image_mask_RGBa(self):
+        for mode in ('RGBA', 'RGB', 'L'):
+            im = Image.new(mode, (200, 200), 'white')
+            im2 = getattr(self, 'gradient_' + mode)
+
+            im.paste(im2, (0, 0), self.gradient_RGBa)
+
+            self.assert_9points_image(im, [
+                (128, 255, 126, 255),
+                (0, 127, 126, 255),
+                (126, 253, 126, 255),
+                (128, 127, 254, 255),
+                (0, 255, 254, 255),
+                (126, 125, 254, 255),
+                (128, 1, 128, 255),
+                (0, 129, 128, 255),
+                (126, 255, 128, 255),
+            ])
+
+    def test_color_solid(self):
+        for mode in ('RGBA', 'RGB', 'L'):
+            im = Image.new(mode, (200, 200), 'black')
+
+            rect = (12, 23, 128 + 12, 128 + 23)
+            im.paste('white', rect)
+
+            hist = im.crop(rect).histogram()
+            while hist:
+                head, hist = hist[:256], hist[256:]
+                self.assertEqual(head[255], 128 * 128)
+                self.assertEqual(sum(head[:255]), 0)
+
+    def test_color_mask_1(self):
+        for mode in ('RGBA', 'RGB', 'L'):
+            im = Image.new(mode, (200, 200), (50, 60, 70, 80)[:len(mode)])
+            color = (10, 20, 30, 40)[:len(mode)]
+
+            im.paste(color, (0, 0), self.mask_1)
+
+            self.assert_9points_image(im, [
+                (50, 60, 70, 80),
+                (50, 60, 70, 80),
+                (10, 20, 30, 40),
+                (50, 60, 70, 80),
+                (50, 60, 70, 80),
+                (10, 20, 30, 40),
+                (10, 20, 30, 40),
+                (10, 20, 30, 40),
+                (50, 60, 70, 80),
+            ])
+
+    def test_color_mask_L(self):
+        for mode in ('RGBA', 'RGB', 'L'):
+            im = getattr(self, 'gradient_' + mode).copy()
+            color = 'white'
+
+            im.paste(color, (0, 0), self.mask_L)
+
+            self.assert_9points_image(im, [
+                (127, 191, 254, 191),
+                (111, 207, 206, 110),
+                (127, 254, 127, 0),
+                (207, 207, 239, 239),
+                (191, 191, 190, 191),
+                (207, 206, 111, 112),
+                (254, 254, 254, 255),
+                (239, 206, 206, 238),
+                (254, 191, 127, 191),
+            ])
+
+    def test_color_mask_RGBA(self):
+        for mode in ('RGBA', 'RGB', 'L'):
+            im = getattr(self, 'gradient_' + mode).copy()
+            color = 'white'
+
+            im.paste(color, (0, 0), self.gradient_RGBA)
+
+            self.assert_9points_image(im, [
+                (127, 191, 254, 191),
+                (111, 207, 206, 110),
+                (127, 254, 127, 0),
+                (207, 207, 239, 239),
+                (191, 191, 190, 191),
+                (207, 206, 111, 112),
+                (254, 254, 254, 255),
+                (239, 206, 206, 238),
+                (254, 191, 127, 191),
+            ])
+
+    def test_color_mask_RGBa(self):
+        for mode in ('RGBA', 'RGB', 'L'):
+            im = getattr(self, 'gradient_' + mode).copy()
+            color = 'white'
+
+            im.paste(color, (0, 0), self.gradient_RGBa)
+
+            self.assert_9points_image(im, [
+                (255, 63, 126, 63),
+                (47, 143, 142, 46),
+                (126, 253, 126, 255),
+                (15, 15, 47, 47),
+                (63, 63, 62, 63),
+                (142, 141, 46, 47),
+                (255, 255, 255, 0),
+                (48, 15, 15, 47),
+                (126, 63, 255, 63)
+            ])

libImaging/Paste.c

@@ -24,19 +24,19 @@
 #include "Imaging.h"
 
 /* like (a * b + 127) / 255), but much faster on most platforms */
-#define	MULDIV255NEW(a, b, tmp)\
+#define MULDIV255NEW(a, tmp)\
-     	(tmp = (a) * (b) + 128, ((((tmp) >> 8) + (tmp)) >> 8))
+        (tmp = (a) + 128, ((((tmp) >> 8) + (tmp)) >> 8))
 
-#define	MULDIV255OLD(a, b, tmp)\
+#define MULDIV255OLD(a, tmp)\
-        (((a) * (b) + 127) / 255)
+        (((a) + 127) / 255)
 
 #define MULDIV255 MULDIV255NEW
 
-#define	BLEND(mask, in1, in2, tmp1, tmp2)\
+#define BLEND(mask, in1, in2, tmp1)\
-	(MULDIV255(in1, 255 - mask, tmp1) + MULDIV255(in2, mask, tmp2))
+        MULDIV255(in1 * (255 - mask) + 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,
@@ -99,8 +99,8 @@ paste_mask_L(Imaging imOut, Imaging imIn, Imaging imMask,
 {
     /* paste with mode "L" matte */
 
-    int x, y, i;
+    int x, y;
-    unsigned int tmp1, tmp2;
+    unsigned int tmp1;
 
     if (imOut->image8) {
 
@@ -109,7 +109,7 @@ 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++;
             }
         }
@@ -117,15 +117,16 @@ paste_mask_L(Imaging imOut, Imaging imIn, Imaging imMask,
     } else {
 
         for (y = 0; y < ysize; y++) {
-            UINT8* out = (UINT8*) imOut->image[y+dy]+dx*pixelsize;
+            UINT8* out = (UINT8*) (imOut->image32[y + dy] + dx);
-            UINT8* in = (UINT8*) imIn->image[y+sy]+sx*pixelsize;
+            UINT8* in = (UINT8*) (imIn->image32[y + sy] + sx);
-            UINT8* mask = (UINT8*) imMask->image[y+sy]+sx;
+            UINT8* mask = (UINT8*) (imMask->image8[y+sy] + sx);
             for (x = 0; x < xsize; x++) {
-                for (i = 0; i < pixelsize; i++) {
+                UINT8 a = mask[0];
-                    *out = BLEND(*mask, *out, *in, tmp1, tmp2);
+                out[0] = BLEND(a, out[0], in[0], tmp1);
-                    out++, in++;
+                out[1] = BLEND(a, out[1], in[1], tmp1);
-                }
+                out[2] = BLEND(a, out[2], in[2], tmp1);
-                mask++;
+                out[3] = BLEND(a, out[3], in[3], tmp1);
+                out += 4; in += 4; mask ++;
             }
         }
     }
@@ -138,8 +139,8 @@ paste_mask_RGBA(Imaging imOut, Imaging imIn, Imaging imMask,
 {
     /* paste with mode "RGBA" matte */
 
-    int x, y, i;
+    int x, y;
-    unsigned int tmp1, tmp2;
+    unsigned int tmp1;
 
     if (imOut->image8) {
 
@@ -148,7 +149,7 @@ 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;
             }
         }
@@ -156,15 +157,16 @@ paste_mask_RGBA(Imaging imOut, Imaging imIn, Imaging imMask,
     } else {
 
         for (y = 0; y < ysize; y++) {
-            UINT8* out = (UINT8*) imOut->image[y+dy]+dx*pixelsize;
+            UINT8* out = (UINT8*) (imOut->image32[y + dy] + dx);
-            UINT8* in = (UINT8*) imIn->image[y+sy]+sx*pixelsize;
+            UINT8* in = (UINT8*) (imIn->image32[y + sy] + sx);
-            UINT8* mask = (UINT8*) imMask->image[y+sy]+sx*4+3;
+            UINT8* mask = (UINT8*) (imMask->image32[y+sy] + sx);
             for (x = 0; x < xsize; x++) {
-                for (i = 0; i < pixelsize; i++) {
+                UINT8 a = mask[3];
-                    *out = BLEND(*mask, *out, *in, tmp1, tmp2);
+                out[0] = BLEND(a, out[0], in[0], tmp1);
-                    out++, in++;
+                out[1] = BLEND(a, out[1], in[1], tmp1);
-                }
+                out[2] = BLEND(a, out[2], in[2], tmp1);
-                mask += 4;
+                out[3] = BLEND(a, out[3], in[3], tmp1);
+                out += 4; in += 4; mask += 4;
             }
         }
     }
@@ -178,7 +180,7 @@ paste_mask_RGBa(Imaging imOut, Imaging imIn, Imaging imMask,
 {
     /* paste with mode "RGBa" matte */
 
-    int x, y, i;
+    int x, y;
     unsigned int tmp1;
 
     if (imOut->image8) {
@@ -196,15 +198,16 @@ paste_mask_RGBa(Imaging imOut, Imaging imIn, Imaging imMask,
     } else {
 
         for (y = 0; y < ysize; y++) {
-            UINT8* out = (UINT8*) imOut->image[y+dy]+dx*pixelsize;
+            UINT8* out = (UINT8*) (imOut->image32[y + dy] + dx);
-            UINT8* in = (UINT8*) imIn->image[y+sy]+sx*pixelsize;
+            UINT8* in = (UINT8*) (imIn->image32[y + sy] + sx);
-            UINT8* mask = (UINT8*) imMask->image[y+sy]+sx*4+3;
+            UINT8* mask = (UINT8*) (imMask->image32[y+sy] + sx);
             for (x = 0; x < xsize; x++) {
-                for (i = 0; i < pixelsize; i++) {
+                UINT8 a = mask[3];
-                    *out = PREBLEND(*mask, *out, *in, tmp1);
+                out[0] = PREBLEND(a, out[0], in[0], tmp1);
-                    out++, in++;
+                out[1] = PREBLEND(a, out[1], in[1], tmp1);
-                }
+                out[2] = PREBLEND(a, out[2], in[2], tmp1);
-                mask += 4;
+                out[3] = PREBLEND(a, out[3], in[3], tmp1);
+                out += 4; in += 4; mask += 4;
             }
         }
     }
@@ -370,7 +373,7 @@ 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) {
 
@@ -378,7 +381,7 @@ 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++;
             }
         }
@@ -390,7 +393,7 @@ 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++;
@@ -407,7 +410,7 @@ 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) {
 
@@ -416,7 +419,7 @@ 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;
             }
         }
@@ -430,7 +433,7 @@ 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;