diff --git a/ChangeLog b/ChangeLog
index 4a55fa3a7d..996500474f 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,20 @@
+2003-09-15  Helvetix Victorinox  <helvetix@gimp.org>
+
+	* gimp-composite-mmx-{test,installer}.c
+	* gimp-composite-sse-{test,installer}.c
+	* gimp-composite-sse2-{test,installer}.c
+	* make-installer.py: sort test output by mode name
+
+	* gimp-composite-sse2.[ch]: New compositing mode implementations
+	  darken
+	  difference
+	  grain_extract
+	  lighten
+
+	* gimp-composite-x86.h: added
+
+	* gimp-composite-{sse,mmx,sse2}.[ch]: Tightening declarations of clobbered registers.
+                                                                                                                                               
 2003-09-16  Manish Singh  <yosh@gimp.org>
 
 	* app/Makefile.am: use -u to prevent garbage collection of symbols
diff --git a/app/composite/gimp-composite-3dnow-test.c b/app/composite/gimp-composite-3dnow-test.c
index d3963f8eb8..8fd48e755d 100644
--- a/app/composite/gimp-composite-3dnow-test.c
+++ b/app/composite/gimp-composite-3dnow-test.c
@@ -71,7 +71,7 @@ main (int argc, char *argv[])
   putenv ("GIMP_COMPOSITE=0x1");
 
   iterations = 1;
-  n_pixels = 1048577;
+  n_pixels = 163921;
 
   argv++, argc--;
   while (argc >= 2) {
diff --git a/app/composite/gimp-composite-altivec-test.c b/app/composite/gimp-composite-altivec-test.c
index d8c4def20e..c9739b313e 100644
--- a/app/composite/gimp-composite-altivec-test.c
+++ b/app/composite/gimp-composite-altivec-test.c
@@ -71,7 +71,7 @@ main (int argc, char *argv[])
   putenv ("GIMP_COMPOSITE=0x1");
 
   iterations = 1;
-  n_pixels = 1048577;
+  n_pixels = 163921;
 
   argv++, argc--;
   while (argc >= 2) {
diff --git a/app/composite/gimp-composite-mmx-test.c b/app/composite/gimp-composite-mmx-test.c
index 953cd2f6d2..5465affe3c 100644
--- a/app/composite/gimp-composite-mmx-test.c
+++ b/app/composite/gimp-composite-mmx-test.c
@@ -57,74 +57,52 @@ gimp_composite_mmx_test (int iterations, int n_pixels)
   }
 
 
-  gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_MULTIPLY, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
-  gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_MULTIPLY, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
-  ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
-  ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_multiply_rgba8_rgba8_rgba8_mmx, &special_ctx);
-  if (gimp_composite_regression_compare_contexts ("multiply", &generic_ctx, &special_ctx)) {
-    return (1);
-  }
-  gimp_composite_regression_timer_report ("multiply", ft0, ft1);
-
-  gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_SCREEN, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
-  gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_SCREEN, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
-  ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
-  ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_screen_rgba8_rgba8_rgba8_mmx, &special_ctx);
-  if (gimp_composite_regression_compare_contexts ("screen", &generic_ctx, &special_ctx)) {
-    return (1);
-  }
-  gimp_composite_regression_timer_report ("screen", ft0, ft1);
-
-  gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_DIFFERENCE, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
-  gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_DIFFERENCE, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
-  ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
-  ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_difference_rgba8_rgba8_rgba8_mmx, &special_ctx);
-  if (gimp_composite_regression_compare_contexts ("difference", &generic_ctx, &special_ctx)) {
-    return (1);
-  }
-  gimp_composite_regression_timer_report ("difference", ft0, ft1);
-
   gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_ADDITION, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
   gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_ADDITION, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
   ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
   ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_addition_rgba8_rgba8_rgba8_mmx, &special_ctx);
   if (gimp_composite_regression_compare_contexts ("addition", &generic_ctx, &special_ctx)) {
+    printf("addition failed\n");
     return (1);
   }
   gimp_composite_regression_timer_report ("addition", ft0, ft1);
 
-  gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_SUBTRACT, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
-  gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_SUBTRACT, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
+  gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_BURN, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
+  gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_BURN, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
   ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
-  ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_subtract_rgba8_rgba8_rgba8_mmx, &special_ctx);
-  if (gimp_composite_regression_compare_contexts ("subtract", &generic_ctx, &special_ctx)) {
+  ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_burn_rgba8_rgba8_rgba8_mmx, &special_ctx);
+  if (gimp_composite_regression_compare_contexts ("burn", &generic_ctx, &special_ctx)) {
+    printf("burn failed\n");
     return (1);
   }
-  gimp_composite_regression_timer_report ("subtract", ft0, ft1);
+  gimp_composite_regression_timer_report ("burn", ft0, ft1);
 
   gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_DARKEN, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
   gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_DARKEN, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
   ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
   ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_darken_rgba8_rgba8_rgba8_mmx, &special_ctx);
   if (gimp_composite_regression_compare_contexts ("darken", &generic_ctx, &special_ctx)) {
+    printf("darken failed\n");
     return (1);
   }
   gimp_composite_regression_timer_report ("darken", ft0, ft1);
 
-  gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_LIGHTEN, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
-  gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_LIGHTEN, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
+  gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_DIFFERENCE, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
+  gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_DIFFERENCE, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
   ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
-  ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_lighten_rgba8_rgba8_rgba8_mmx, &special_ctx);
-  if (gimp_composite_regression_compare_contexts ("lighten", &generic_ctx, &special_ctx)) {
+  ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_difference_rgba8_rgba8_rgba8_mmx, &special_ctx);
+  if (gimp_composite_regression_compare_contexts ("difference", &generic_ctx, &special_ctx)) {
+    printf("difference failed\n");
     return (1);
   }
-  gimp_composite_regression_timer_report ("lighten", ft0, ft1);
+  gimp_composite_regression_timer_report ("difference", ft0, ft1);
 
   gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_DIVIDE, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
   gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_DIVIDE, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
   ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
   ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_divide_rgba8_rgba8_rgba8_mmx, &special_ctx);
   if (gimp_composite_regression_compare_contexts ("divide", &generic_ctx, &special_ctx)) {
+    printf("divide failed\n");
     return (1);
   }
   gimp_composite_regression_timer_report ("divide", ft0, ft1);
@@ -134,24 +112,17 @@ gimp_composite_mmx_test (int iterations, int n_pixels)
   ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
   ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_dodge_rgba8_rgba8_rgba8_mmx, &special_ctx);
   if (gimp_composite_regression_compare_contexts ("dodge", &generic_ctx, &special_ctx)) {
+    printf("dodge failed\n");
     return (1);
   }
   gimp_composite_regression_timer_report ("dodge", ft0, ft1);
 
-  gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_BURN, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
-  gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_BURN, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
-  ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
-  ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_burn_rgba8_rgba8_rgba8_mmx, &special_ctx);
-  if (gimp_composite_regression_compare_contexts ("burn", &generic_ctx, &special_ctx)) {
-    return (1);
-  }
-  gimp_composite_regression_timer_report ("burn", ft0, ft1);
-
   gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_GRAIN_EXTRACT, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
   gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_GRAIN_EXTRACT, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
   ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
   ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_grain_extract_rgba8_rgba8_rgba8_mmx, &special_ctx);
   if (gimp_composite_regression_compare_contexts ("grain_extract", &generic_ctx, &special_ctx)) {
+    printf("grain_extract failed\n");
     return (1);
   }
   gimp_composite_regression_timer_report ("grain_extract", ft0, ft1);
@@ -161,27 +132,70 @@ gimp_composite_mmx_test (int iterations, int n_pixels)
   ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
   ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_grain_merge_rgba8_rgba8_rgba8_mmx, &special_ctx);
   if (gimp_composite_regression_compare_contexts ("grain_merge", &generic_ctx, &special_ctx)) {
+    printf("grain_merge failed\n");
     return (1);
   }
   gimp_composite_regression_timer_report ("grain_merge", ft0, ft1);
 
-  gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_SWAP, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
-  gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_SWAP, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
+  gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_LIGHTEN, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
+  gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_LIGHTEN, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
   ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
-  ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_swap_rgba8_rgba8_rgba8_mmx, &special_ctx);
-  if (gimp_composite_regression_compare_contexts ("swap", &generic_ctx, &special_ctx)) {
+  ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_lighten_rgba8_rgba8_rgba8_mmx, &special_ctx);
+  if (gimp_composite_regression_compare_contexts ("lighten", &generic_ctx, &special_ctx)) {
+    printf("lighten failed\n");
     return (1);
   }
-  gimp_composite_regression_timer_report ("swap", ft0, ft1);
+  gimp_composite_regression_timer_report ("lighten", ft0, ft1);
+
+  gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_MULTIPLY, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
+  gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_MULTIPLY, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
+  ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
+  ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_multiply_rgba8_rgba8_rgba8_mmx, &special_ctx);
+  if (gimp_composite_regression_compare_contexts ("multiply", &generic_ctx, &special_ctx)) {
+    printf("multiply failed\n");
+    return (1);
+  }
+  gimp_composite_regression_timer_report ("multiply", ft0, ft1);
 
   gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_SCALE, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
   gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_SCALE, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
   ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
   ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_scale_rgba8_rgba8_rgba8_mmx, &special_ctx);
   if (gimp_composite_regression_compare_contexts ("scale", &generic_ctx, &special_ctx)) {
+    printf("scale failed\n");
     return (1);
   }
   gimp_composite_regression_timer_report ("scale", ft0, ft1);
+
+  gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_SCREEN, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
+  gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_SCREEN, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
+  ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
+  ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_screen_rgba8_rgba8_rgba8_mmx, &special_ctx);
+  if (gimp_composite_regression_compare_contexts ("screen", &generic_ctx, &special_ctx)) {
+    printf("screen failed\n");
+    return (1);
+  }
+  gimp_composite_regression_timer_report ("screen", ft0, ft1);
+
+  gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_SUBTRACT, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
+  gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_SUBTRACT, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
+  ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
+  ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_subtract_rgba8_rgba8_rgba8_mmx, &special_ctx);
+  if (gimp_composite_regression_compare_contexts ("subtract", &generic_ctx, &special_ctx)) {
+    printf("subtract failed\n");
+    return (1);
+  }
+  gimp_composite_regression_timer_report ("subtract", ft0, ft1);
+
+  gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_SWAP, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
+  gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_SWAP, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
+  ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
+  ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_swap_rgba8_rgba8_rgba8_mmx, &special_ctx);
+  if (gimp_composite_regression_compare_contexts ("swap", &generic_ctx, &special_ctx)) {
+    printf("swap failed\n");
+    return (1);
+  }
+  gimp_composite_regression_timer_report ("swap", ft0, ft1);
 #endif
   return (0);
 }
@@ -197,7 +211,7 @@ main (int argc, char *argv[])
   putenv ("GIMP_COMPOSITE=0x1");
 
   iterations = 1;
-  n_pixels = 1048577;
+  n_pixels = 163921;
 
   argv++, argc--;
   while (argc >= 2) {
diff --git a/app/composite/gimp-composite-mmx.c b/app/composite/gimp-composite-mmx.c
index 7006e9ded4..81f587bea3 100644
--- a/app/composite/gimp-composite-mmx.c
+++ b/app/composite/gimp-composite-mmx.c
@@ -1,7 +1,8 @@
-/* The GIMP -- an image manipulation program
+/* -*- mode: c tab-width: 2; c-basic-indent: 2; indent-tabs-mode: nil -*-
+	*
+	* The GIMP -- an image manipulation program
  * Copyright (C) 1995 Spencer Kimball and Peter Mattis
  *
- * -*- mode: c tab-width: 2; c-basic-indent: 2; indent-tabs-mode: nil -*-
  *
  * This program is free software; you can redistribute it and/or modify
  * it under the terms of the GNU General Public License as published by
@@ -42,109 +43,12 @@
 
 #include "gimp-composite.h"
 #include "gimp-composite-mmx.h"
+#include "gimp-composite-x86.h"
 
 #define pminub(src,dst,tmp)  "\tmovq %%" #dst ", %%" #tmp ";" "psubusb %%" #src ", %%" #tmp ";" "psubb %%" #tmp ", %%" #dst "\n"
 #define pmaxub(a,b,tmp)      "\tmovq %%" #a ", %%" #tmp ";" "psubusb %%" #b ", %%" #tmp ";" "paddb %%" #tmp ", %%" #b "\n"
 
 
-/*
- * Clobbers eax, ecx edx
- */
-/*
- * Double-word divide.  Adjusted for subsequent unsigned packing
- * (high-order bit of each word is cleared)
- */
-#define pdivwX(dividend,divisor,quotient) "movd %%" #dividend ",%%eax; " \
-                                          "movd %%" #divisor  ",%%ecx; " \
-                                          "xorl %%edx,%%edx; "           \
-                                          "divw %%cx; "                  \
-                                          "roll $16, %%eax; "            \
-                                          "roll $16, %%ecx; "            \
-                                          "xorl %%edx,%%edx; "           \
-                                          "divw %%cx; "                  \
-                                          "btr $15, %%eax; "             \
-                                          "roll $16, %%eax; "            \
-                                          "btr $15, %%eax; "             \
-                                          "movd %%eax,%%" #quotient ";"
-
-/*
- * Quadword divide.  No adjustment for subsequent unsigned packing
- * (high-order bit of each word is left alone)
- */
-#define pdivwqX(dividend,divisor,quotient) "movd   %%" #dividend ",%%eax; " \
-                                          "movd   %%" #divisor  ",%%ecx; " \
-                                          "xorl   %%edx,%%edx; "           \
-                                          "divw   %%cx; "                  \
-                                          "roll   $16, %%eax; "            \
-                                          "roll   $16, %%ecx; "            \
-                                          "xorl   %%edx,%%edx; "           \
-                                          "divw   %%cx; "                  \
-                                          "roll   $16, %%eax; "            \
-                                          "movd   %%eax,%%" #quotient "; " \
-                                          "psrlq $32,%%" #dividend ";"     \
-                                          "psrlq $32,%%" #divisor ";"      \
-                                          "movd   %%" #dividend ",%%eax; " \
-                                          "movd   %%" #divisor  ",%%ecx; " \
-                                          "xorl   %%edx,%%edx; "           \
-                                          "divw   %%cx; "                  \
-                                          "roll   $16, %%eax; "            \
-                                          "roll   $16, %%ecx; "            \
-                                          "xorl   %%edx,%%edx; "           \
-                                          "divw   %%cx; "                  \
-                                          "roll   $16, %%eax; "            \
-                                          "movd   %%eax,%%" #divisor ";"   \
-                                          "psllq  $32,%%" #divisor ";"     \
-                                          "por    %%" #divisor ",%%" #quotient ";"
-   
-/*
- * Quadword divide.  Adjusted for subsequent unsigned packing
- * (high-order bit of each word is cleared)
- */
-#define pdivwuqX(dividend,divisor,quotient) \
-                                          pdivwX(dividend,divisor,quotient) \
-                                            "psrlq  $32,%%" #dividend ";"   \
-                                            "psrlq  $32,%%" #divisor ";"    \
-                                          pdivwX(dividend,divisor,quotient) \
-                                          "movd   %%eax,%%" #divisor ";"    \
-                                            "psllq  $32,%%" #divisor ";"    \
-                                            "por    %%" #divisor ",%%" #quotient ";"
-
-/* equivalent to the INT_MULT() macro in gimp-composite-generic.c */
-/*
- * opr2 = INT_MULT(opr1, opr2, t)
- *
- * Operates across quad-words using x86 word (16bit) value.
- * Result is left in opr2
- *
- * opr1 = opr1 * opr2 + w128
- * opr2 = opr1
- * opr2 = ((opr2 >> 8) + opr1) >> 8
- */
-#define pmulwX(opr1,opr2,w128) \
-                  "\tpmullw    %%"#opr2", %%"#opr1"; " \
-                  "\tpaddw     %%"#w128", %%"#opr1"; " \
-                  "\tmovq      %%"#opr1", %%"#opr2"; " \
-                  "\tpsrlw     $8,        %%"#opr2"; " \
-                  "\tpaddw     %%"#opr1", %%"#opr2"; " \
-                  "\tpsrlw     $8,        %%"#opr2"\n"
- 
-/* a = INT_MULT(a,b) */
-#define mmx_int_mult(a,b,w128) \
-                  "\tpmullw    %%"#b",    %%"#a"; " \
-                  "\tpaddw     %%"#w128", %%"#a"; " \
-                  "\tmovq      %%"#a",    %%"#b"; " \
-                  "\tpsrlw     $8,        %%"#b"; " \
-                  "\tpaddw     %%"#a",    %%"#b"; " \
-                  "\tpsrlw     $8,        %%"#b"\n"
-
-#define mmx_low_bytes_to_words(src,dst,zero) \
-         "\tmovq      %%"#src", %%"#dst"; " \
-         "\tpunpcklbw %%"#zero", %%"#dst"\n"
-
-#define mmx_high_bytes_to_words(src,dst,zero) \
-         "\tmovq      %%"#src", %%"#dst"; " \
-         "\tpunpckhbw %%"#zero", %%"#dst"\n"
-
  
 void
 debug_display_mmx(void)
@@ -188,40 +92,40 @@ gimp_composite_addition_rgba8_rgba8_rgba8_mmx (GimpCompositeContext *_op)
 
   for (; op.n_pixels >= 2; op.n_pixels -= 2)
     {
-						asm ("  movq    %0, %%mm2\n"
-											"\tmovq    %1, %%mm3\n"
-											"\tmovq    %%mm2, %%mm4\n"
-											"\tpaddusb %%mm3, %%mm4\n"
-											"\tmovq    %%mm0, %%mm1\n"
-											"\tpandn   %%mm4, %%mm1\n"
-											"\t" pminub(mm3, mm2, mm4) "\n"
-											"\tpand    %%mm0, %%mm2\n"
-											"\tpor     %%mm2, %%mm1\n"
-											"\tmovq    %%mm1, %2\n"
-											: /* empty */
-											: "m" (*op.A), "m" (*op.B), "m" (*op.D)
-											: "0", "1", "2", "%mm0", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5", "%mm6", "%mm7");
-						op.A += 8;
-						op.B += 8;
-						op.D += 8;
-				}
+      asm ("  movq    %0, %%mm2\n"
+           "\tmovq    %1, %%mm3\n"
+           "\tmovq    %%mm2, %%mm4\n"
+           "\tpaddusb %%mm3, %%mm4\n"
+           "\tmovq    %%mm0, %%mm1\n"
+           "\tpandn   %%mm4, %%mm1\n"
+           "\t" pminub(mm3, mm2, mm4) "\n"
+           "\tpand    %%mm0, %%mm2\n"
+           "\tpor     %%mm2, %%mm1\n"
+           "\tmovq    %%mm1, %2\n"
+           : /* empty */
+           : "m" (*op.A), "m" (*op.B), "m" (*op.D)
+           : "0", "1", "2", "%mm0", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5", "%mm6", "%mm7");
+      op.A += 8;
+      op.B += 8;
+      op.D += 8;
+    }
 
   if (op.n_pixels)
     {
       asm volatile ("  movd    %0, %%mm2\n"
                     "\tmovd    %1, %%mm3\n"
-																				"\tmovq    %%mm2, %%mm4\n"
-																				"\tpaddusb %%mm3, %%mm4\n"
-																				"\tmovq    %%mm0, %%mm1\n"
-																				"\tpandn   %%mm4, %%mm1\n"
-																				"\t" pminub(mm3, mm2, mm4) "\n"
-																				"\tpand    %%mm0, %%mm2\n"
-																				"\tpor     %%mm2, %%mm1\n"
-																				"\tmovd    %%mm1, %2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D)
-																				: "0", "1", "2", "%mm0", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5", "%mm6", "%mm7");
-				}
+                    "\tmovq    %%mm2, %%mm4\n"
+                    "\tpaddusb %%mm3, %%mm4\n"
+                    "\tmovq    %%mm0, %%mm1\n"
+                    "\tpandn   %%mm4, %%mm1\n"
+                    "\t" pminub(mm3, mm2, mm4) "\n"
+                    "\tpand    %%mm0, %%mm2\n"
+                    "\tpor     %%mm2, %%mm1\n"
+                    "\tmovd    %%mm1, %2\n"
+                    : /* empty */
+                    : "m" (*op.A), "m" (*op.B), "m" (*op.D)
+                    : "0", "1", "2", "%mm0", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5", "%mm6", "%mm7");
+    }
 
   asm("emms");
 }
@@ -232,109 +136,109 @@ gimp_composite_burn_rgba8_rgba8_rgba8_mmx (GimpCompositeContext *_op)
   GimpCompositeContext op = *_op;
 
   for (; op.n_pixels >= 2; op.n_pixels -= 2)
-				{
-						asm ("  movq         %0,%%mm0\n"
-											"\tmovq         %1,%%mm1\n"
+    {
+      asm ("  movq         %0,%%mm0\n"
+           "\tmovq         %1,%%mm1\n"
          
-											"\tmovq         %3,%%mm2\n"
-											"\tpsubb     %%mm0,%%mm2\n" /* mm2 = 255 - A */
-											"\tpxor      %%mm4,%%mm4\n"
-											"\tpunpcklbw %%mm2,%%mm4\n" /* mm4 = (255- A) * 256  */
+           "\tmovq         %3,%%mm2\n"
+           "\tpsubb     %%mm0,%%mm2\n" /* mm2 = 255 - A */
+           "\tpxor      %%mm4,%%mm4\n"
+           "\tpunpcklbw %%mm2,%%mm4\n" /* mm4 = (255- A) * 256  */
 
-											"\tmovq      %%mm1,%%mm3\n"
-											"\tpxor      %%mm5,%%mm5\n"
-											"\tpunpcklbw %%mm5,%%mm3\n"
-											"\tmovq         %4,%%mm5\n"
-											"\tpaddusw   %%mm3,%%mm5\n" /* mm5 = B + 1 */
+           "\tmovq      %%mm1,%%mm3\n"
+           "\tpxor      %%mm5,%%mm5\n"
+           "\tpunpcklbw %%mm5,%%mm3\n"
+           "\tmovq         %4,%%mm5\n"
+           "\tpaddusw   %%mm3,%%mm5\n" /* mm5 = B + 1 */
 
-											"\t" pdivwqX(mm4,mm5,mm7) "\n"
+           "\t" pdivwqX(mm4,mm5,mm7) "\n"
 
-											"\tmovq         %3,%%mm2\n"
-											"\tpsubb     %%mm0,%%mm2\n" /* mm2 = 255 - A */
-											"\tpxor      %%mm4,%%mm4\n"
-											"\tpunpckhbw %%mm2,%%mm4\n" /* mm4 = (255- A) * 256  */
+           "\tmovq         %3,%%mm2\n"
+           "\tpsubb     %%mm0,%%mm2\n" /* mm2 = 255 - A */
+           "\tpxor      %%mm4,%%mm4\n"
+           "\tpunpckhbw %%mm2,%%mm4\n" /* mm4 = (255- A) * 256  */
 
-											"\tmovq      %%mm1,%%mm3\n"
-											"\tpxor      %%mm5,%%mm5\n"
-											"\tpunpckhbw %%mm5,%%mm3\n"
-											"\tmovq         %4,%%mm5\n"
-											"\tpaddusw   %%mm3,%%mm5\n" /* mm5 = B + 1 */
-											"\t" pdivwqX(mm4,mm5,mm6) "\n"
+           "\tmovq      %%mm1,%%mm3\n"
+           "\tpxor      %%mm5,%%mm5\n"
+           "\tpunpckhbw %%mm5,%%mm3\n"
+           "\tmovq         %4,%%mm5\n"
+           "\tpaddusw   %%mm3,%%mm5\n" /* mm5 = B + 1 */
+           "\t" pdivwqX(mm4,mm5,mm6) "\n"
 
-											"\tmovq      %5,%%mm4\n"
-											"\tmovq      %%mm4,%%mm5\n"
-											"\tpsubusw   %%mm6,%%mm4\n"
-											"\tpsubusw   %%mm7,%%mm5\n"
+           "\tmovq      %5,%%mm4\n"
+           "\tmovq      %%mm4,%%mm5\n"
+           "\tpsubusw   %%mm6,%%mm4\n"
+           "\tpsubusw   %%mm7,%%mm5\n"
                   
-											"\tpackuswb  %%mm4,%%mm5\n"
+           "\tpackuswb  %%mm4,%%mm5\n"
 
-											"\t" pminub(mm0,mm1,mm3) "\n" /* mm1 = min(mm0,mm1) clobber mm3 */
+           "\t" pminub(mm0,mm1,mm3) "\n" /* mm1 = min(mm0,mm1) clobber mm3 */
 
-											"\tmovq         %6,%%mm7\n"
-											"\tpand      %%mm7,%%mm1\n" /* mm1 = mm7 & alpha_mask */
+           "\tmovq         %6,%%mm7\n"
+           "\tpand      %%mm7,%%mm1\n" /* mm1 = mm7 & alpha_mask */
 
-											"\tpandn     %%mm5,%%mm7\n" /* mm7 = ~mm7 & mm5 */
-											"\tpor       %%mm1,%%mm7\n" /* mm7 = mm7 | mm1 */
+           "\tpandn     %%mm5,%%mm7\n" /* mm7 = ~mm7 & mm5 */
+           "\tpor       %%mm1,%%mm7\n" /* mm7 = mm7 | mm1 */
 
-											"\tmovq      %%mm7,%2\n"
-											: /* empty */
-											: "m" (*op.A), "m" (*op.B), "m" (*op.D), "m" (*rgba8_b255), "m" (*rgba8_w1), "m" (*rgba8_w255), "m" (*rgba8_alpha_mask)
-											: "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
-						op.A += 8;
-						op.B += 8;
-						op.D += 8;
-				}
+           "\tmovq      %%mm7,%2\n"
+           : /* empty */
+           : "m" (*op.A), "m" (*op.B), "m" (*op.D), "m" (*rgba8_b255), "m" (*rgba8_w1), "m" (*rgba8_w255), "m" (*rgba8_alpha_mask)
+           : "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
+      op.A += 8;
+      op.B += 8;
+      op.D += 8;
+    }
 
   if (op.n_pixels)
-				{
-						asm volatile ("  movd      %0,%%mm0\n"
-																				"\tmovd      %1,%%mm1\n"
+    {
+      asm volatile ("  movd      %0,%%mm0\n"
+                    "\tmovd      %1,%%mm1\n"
 
-																				"\tmovq      %3,%%mm2\n"
-																				"\tpsubb     %%mm0,%%mm2\n" /* mm2 = 255 - A */
-																				"\tpxor      %%mm4,%%mm4\n"
-																				"\tpunpcklbw %%mm2,%%mm4\n" /* mm4 = (255- A) * 256  */
+                    "\tmovq      %3,%%mm2\n"
+                    "\tpsubb     %%mm0,%%mm2\n" /* mm2 = 255 - A */
+                    "\tpxor      %%mm4,%%mm4\n"
+                    "\tpunpcklbw %%mm2,%%mm4\n" /* mm4 = (255- A) * 256  */
 
-																				"\tmovq      %%mm1,%%mm3\n"
-																				"\tpxor      %%mm5,%%mm5\n"
-																				"\tpunpcklbw %%mm5,%%mm3\n"
-																				"\tmovq      %4,%%mm5\n"
-																				"\tpaddusw   %%mm3,%%mm5\n" /* mm5 = B + 1 */
+                    "\tmovq      %%mm1,%%mm3\n"
+                    "\tpxor      %%mm5,%%mm5\n"
+                    "\tpunpcklbw %%mm5,%%mm3\n"
+                    "\tmovq      %4,%%mm5\n"
+                    "\tpaddusw   %%mm3,%%mm5\n" /* mm5 = B + 1 */
 
-																				"\t" pdivwqX(mm4,mm5,mm7) "\n"
+                    "\t" pdivwqX(mm4,mm5,mm7) "\n"
 
-																				"\tmovq      %3,%%mm2\n"
-																				"\tpsubb   %%mm0,%%mm2\n" /* mm2 = 255 - A */
-																				"\tpxor      %%mm4,%%mm4\n"
-																				"\tpunpckhbw %%mm2,%%mm4\n" /* mm4 = (255- A) * 256  */
+                    "\tmovq      %3,%%mm2\n"
+                    "\tpsubb   %%mm0,%%mm2\n" /* mm2 = 255 - A */
+                    "\tpxor      %%mm4,%%mm4\n"
+                    "\tpunpckhbw %%mm2,%%mm4\n" /* mm4 = (255- A) * 256  */
 
-																				"\tmovq      %%mm1,%%mm3\n"
-																				"\tpxor      %%mm5,%%mm5\n"
-																				"\tpunpckhbw %%mm5,%%mm3\n"
-																				"\tmovq      %4,%%mm5\n"
-																				"\tpaddusw   %%mm3,%%mm5\n" /* mm5 = B + 1 */
-																				"\t" pdivwqX(mm4,mm5,mm6) "\n"
+                    "\tmovq      %%mm1,%%mm3\n"
+                    "\tpxor      %%mm5,%%mm5\n"
+                    "\tpunpckhbw %%mm5,%%mm3\n"
+                    "\tmovq      %4,%%mm5\n"
+                    "\tpaddusw   %%mm3,%%mm5\n" /* mm5 = B + 1 */
+                    "\t" pdivwqX(mm4,mm5,mm6) "\n"
 
-																				"\tmovq      %5,%%mm4\n"
-																				"\tmovq      %%mm4,%%mm5\n"
-																				"\tpsubusw     %%mm6,%%mm4\n"
-																				"\tpsubusw     %%mm7,%%mm5\n"
+                    "\tmovq      %5,%%mm4\n"
+                    "\tmovq      %%mm4,%%mm5\n"
+                    "\tpsubusw     %%mm6,%%mm4\n"
+                    "\tpsubusw     %%mm7,%%mm5\n"
                   
-																				"\tpackuswb  %%mm4,%%mm5\n"
+                    "\tpackuswb  %%mm4,%%mm5\n"
 
-																				"\t" pminub(mm0,mm1,mm3) "\n" /* mm1 = min(mm0,mm1) clobber mm3 */
+                    "\t" pminub(mm0,mm1,mm3) "\n" /* mm1 = min(mm0,mm1) clobber mm3 */
 
-																				"\tmovq      %6,%%mm7\n"
-																				"\tpand      %%mm7,%%mm1\n" /* mm1 = mm7 & alpha_mask */
+                    "\tmovq      %6,%%mm7\n"
+                    "\tpand      %%mm7,%%mm1\n" /* mm1 = mm7 & alpha_mask */
 
-																				"\tpandn     %%mm5,%%mm7\n" /* mm7 = ~mm7 & mm5 */
-																				"\tpor       %%mm1,%%mm7\n" /* mm7 = mm7 | mm1 */
+                    "\tpandn     %%mm5,%%mm7\n" /* mm7 = ~mm7 & mm5 */
+                    "\tpor       %%mm1,%%mm7\n" /* mm7 = mm7 | mm1 */
 
-																				"\tmovd      %%mm7,%2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D), "m" (*rgba8_b255), "m" (*rgba8_w1), "m" (*rgba8_w255), "m" (*rgba8_alpha_mask)
-																				: "0", "1", "2", "%mm0", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5", "%mm6", "%mm7");
-				}
+                    "\tmovd      %%mm7,%2\n"
+                    : /* empty */
+                    : "m" (*op.A), "m" (*op.B), "m" (*op.D), "m" (*rgba8_b255), "m" (*rgba8_w1), "m" (*rgba8_w255), "m" (*rgba8_alpha_mask)
+                    : "0", "1", "2", "%mm0", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5", "%mm6", "%mm7");
+    }
 
   asm("emms");
 }
@@ -345,30 +249,30 @@ gimp_composite_darken_rgba8_rgba8_rgba8_mmx (GimpCompositeContext *_op)
 {
   GimpCompositeContext op = *_op;
 
-		for (; op.n_pixels >= 2; op.n_pixels -= 2)
+  for (; op.n_pixels >= 2; op.n_pixels -= 2)
     {
-						asm volatile ("  movq    %0, %%mm2\n"
-																				"\tmovq    %1, %%mm3\n"
-																				"\t" pminub(mm3, mm2, mm4) "\n"
-																				"\tmovq    %%mm2, %2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D)
-																				: "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
-						op.A += 8;
-						op.B += 8;
-						op.D += 8;
-				}
+      asm volatile ("  movq    %1, %%mm2\n"
+                    "\tmovq    %2, %%mm3\n"
+                    "\t" pminub(mm3, mm2, mm4) "\n"
+                    "\tmovq    %%mm2, %0\n"
+                    : "=m" (*op.D)
+                    : "m" (*op.A), "m" (*op.B)
+                    : "%mm1", "%mm2", "%mm3", "%mm4");
+      op.A += 8;
+      op.B += 8;
+      op.D += 8;
+    }
 
   if (op.n_pixels)
     {
-						asm volatile ("  movd    %0, %%mm2\n"
-																				"\tmovd    %1, %%mm3\n"
-																				"\t" pminub(mm3, mm2, mm4) "\n"
-																				"\tmovd    %%mm2, %2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D)
-																				: "0", "1", "2", "%mm2", "%mm3", "%mm4");
-				}
+      asm volatile ("  movd    %0, %%mm2\n"
+                    "\tmovd    %1, %%mm3\n"
+                    "\t" pminub(mm3, mm2, mm4) "\n"
+                    "\tmovd    %%mm2, %2\n"
+                    : /* empty */
+                    : "m" (*op.A), "m" (*op.B), "m" (*op.D)
+                    : "0", "1", "2", "%mm2", "%mm3", "%mm4");
+    }
         
   asm("emms");
 }
@@ -381,47 +285,47 @@ gimp_composite_difference_rgba8_rgba8_rgba8_mmx (GimpCompositeContext *_op)
   asm volatile ("movq    %0,%%mm0"     :  : "m" (*rgba8_alpha_mask) : "%mm0");
 
   for (; op.n_pixels >= 2; op.n_pixels -= 2)
-				{
-						asm volatile ("  movq       %0, %%mm2\n"
-																				"\tmovq       %1, %%mm3\n"
-																				"\tmovq    %%mm2, %%mm4\n"
-																				"\tmovq    %%mm3, %%mm5\n"
-																				"\tpsubusb %%mm3, %%mm4\n"
-																				"\tpsubusb %%mm2, %%mm5\n"
-																				"\tpaddb   %%mm5, %%mm4\n"
-																				"\tmovq    %%mm0, %%mm1\n"
-																				"\tpandn   %%mm4, %%mm1\n"
-																				"\tpminub  %%mm3, %%mm2\n"
-																				"\tpand    %%mm0, %%mm2\n"
-																				"\tpor     %%mm2, %%mm1\n"
-																				"\tmovq    %%mm1, %2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D)
-																				: "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
-						op.A += 8;
-						op.B += 8;
-						op.D += 8;
-				}
+    {
+      asm volatile ("  movq       %0, %%mm2\n"
+                    "\tmovq       %1, %%mm3\n"
+                    "\tmovq    %%mm2, %%mm4\n"
+                    "\tmovq    %%mm3, %%mm5\n"
+                    "\tpsubusb %%mm3, %%mm4\n"
+                    "\tpsubusb %%mm2, %%mm5\n"
+                    "\tpaddb   %%mm5, %%mm4\n"
+                    "\tmovq    %%mm0, %%mm1\n"
+                    "\tpandn   %%mm4, %%mm1\n"
+                    "\tpminub  %%mm3, %%mm2\n"
+                    "\tpand    %%mm0, %%mm2\n"
+                    "\tpor     %%mm2, %%mm1\n"
+                    "\tmovq    %%mm1, %2\n"
+                    : /* empty */
+                    : "m" (*op.A), "m" (*op.B), "m" (*op.D)
+                    : "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
+      op.A += 8;
+      op.B += 8;
+      op.D += 8;
+    }
   
   if (op.n_pixels)
-				{
-						asm volatile ("  movd       %0, %%mm2\n"
-																				"\tmovd       %1, %%mm3\n"
-																				"\tmovq    %%mm2, %%mm4\n"
-																				"\tmovq    %%mm3, %%mm5\n"
-																				"\tpsubusb %%mm3, %%mm4\n"
-																				"\tpsubusb %%mm2, %%mm5\n"
-																				"\tpaddb   %%mm5, %%mm4\n"
-																				"\tmovq    %%mm0, %%mm1\n"
-																				"\tpandn   %%mm4, %%mm1\n"
-																				"\tpminub  %%mm3, %%mm2\n"
-																				"\tpand    %%mm0, %%mm2\n"
-																				"\tpor     %%mm2, %%mm1\n"
-																				"\tmovd    %%mm1, %2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D)
-																				: "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
-				}
+    {
+      asm volatile ("  movd       %0, %%mm2\n"
+                    "\tmovd       %1, %%mm3\n"
+                    "\tmovq    %%mm2, %%mm4\n"
+                    "\tmovq    %%mm3, %%mm5\n"
+                    "\tpsubusb %%mm3, %%mm4\n"
+                    "\tpsubusb %%mm2, %%mm5\n"
+                    "\tpaddb   %%mm5, %%mm4\n"
+                    "\tmovq    %%mm0, %%mm1\n"
+                    "\tpandn   %%mm4, %%mm1\n"
+                    "\tpminub  %%mm3, %%mm2\n"
+                    "\tpand    %%mm0, %%mm2\n"
+                    "\tpor     %%mm2, %%mm1\n"
+                    "\tmovd    %%mm1, %2\n"
+                    : /* empty */
+                    : "m" (*op.A), "m" (*op.B), "m" (*op.D)
+                    : "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
+    }
 
   asm("emms");
 }
@@ -432,95 +336,95 @@ gimp_composite_divide_rgba8_rgba8_rgba8_mmx (GimpCompositeContext *_op)
   GimpCompositeContext op = *_op;
 
   asm volatile ("movq    %0, %%mm0\n"
-																"\tmovq    %1, %%mm7\n"
-																:
-																: "m" (*rgba8_alpha_mask), "m" (*rgba8_w1)
-																: "%mm0", "%mm7");
+                "\tmovq    %1, %%mm7\n"
+                :
+                : "m" (*rgba8_alpha_mask), "m" (*rgba8_w1)
+                : "%mm0", "%mm7");
   
   for (; op.n_pixels >= 2; op.n_pixels -= 2)
-				{
-						asm volatile ("  movq         %0,%%mm0\n"
-																				"\tmovq         %1,%%mm1\n"
-																				"\tpxor      %%mm2,%%mm2\n"
-																				"\tpunpcklbw %%mm0,%%mm2\n" /* mm2 = A*256 */
+    {
+      asm volatile ("  movq         %0,%%mm0\n"
+                    "\tmovq         %1,%%mm1\n"
+                    "\tpxor      %%mm2,%%mm2\n"
+                    "\tpunpcklbw %%mm0,%%mm2\n" /* mm2 = A*256 */
 
-																				"\tmovq      %%mm1,%%mm3\n"
-																				"\tpxor      %%mm5,%%mm5\n"
-																				"\tpunpcklbw %%mm5,%%mm3\n"
-																				"\tpaddw     %%mm7,%%mm3\n" /* mm3 = B+1 */
+                    "\tmovq      %%mm1,%%mm3\n"
+                    "\tpxor      %%mm5,%%mm5\n"
+                    "\tpunpcklbw %%mm5,%%mm3\n"
+                    "\tpaddw     %%mm7,%%mm3\n" /* mm3 = B+1 */
 
-																				"\t" pdivwuqX(mm2,mm3,mm5) "\n" /* mm5 = (A*256)/(B+1) */
+                    "\t" pdivwuqX(mm2,mm3,mm5) "\n" /* mm5 = (A*256)/(B+1) */
 
-																				"\tpxor      %%mm2,%%mm2\n"
-																				"\tpunpckhbw %%mm0,%%mm2\n" /* mm2 = A*256 */
+                    "\tpxor      %%mm2,%%mm2\n"
+                    "\tpunpckhbw %%mm0,%%mm2\n" /* mm2 = A*256 */
 
-																				"\tmovq      %%mm1,%%mm3\n"
-																				"\tpxor      %%mm6,%%mm6\n"
-																				"\tpunpckhbw %%mm6,%%mm3\n"
-																				"\tpaddw     %%mm7,%%mm3\n" /* mm3 = B+1 */
+                    "\tmovq      %%mm1,%%mm3\n"
+                    "\tpxor      %%mm6,%%mm6\n"
+                    "\tpunpckhbw %%mm6,%%mm3\n"
+                    "\tpaddw     %%mm7,%%mm3\n" /* mm3 = B+1 */
 
-																				"\t" pdivwuqX(mm2,mm3,mm4) "\n" /* mm4 = (A*256)/(B+1) */
+                    "\t" pdivwuqX(mm2,mm3,mm4) "\n" /* mm4 = (A*256)/(B+1) */
                   
-																				"\tpackuswb  %%mm4,%%mm5\n" /* expects mm4 and mm5 to be signed values */
+                    "\tpackuswb  %%mm4,%%mm5\n" /* expects mm4 and mm5 to be signed values */
 
-																				"\t" pminub(mm0,mm1,mm3) "\n"
-																				"\tmovq      %3,%%mm3\n"
-																				"\tmovq      %%mm3,%%mm2\n"
+                    "\t" pminub(mm0,mm1,mm3) "\n"
+                    "\tmovq      %3,%%mm3\n"
+                    "\tmovq      %%mm3,%%mm2\n"
 
-																				"\tpandn     %%mm5,%%mm3\n"
+                    "\tpandn     %%mm5,%%mm3\n"
 
-																				"\tpand      %%mm2,%%mm1\n"
-																				"\tpor       %%mm1,%%mm3\n"
+                    "\tpand      %%mm2,%%mm1\n"
+                    "\tpor       %%mm1,%%mm3\n"
 
-																				"\tmovq      %%mm3,%2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D), "m" (*rgba8_alpha_mask)
-																				: "%eax", "%ecx", "%edx", "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
-						op.A += 8;
-						op.B += 8;
-						op.D += 8;
-				}
+                    "\tmovq      %%mm3,%2\n"
+                    : /* empty */
+                    : "m" (*op.A), "m" (*op.B), "m" (*op.D), "m" (*rgba8_alpha_mask)
+                    : "%eax", "%ecx", "%edx", "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
+      op.A += 8;
+      op.B += 8;
+      op.D += 8;
+    }
 
   if (op.n_pixels)
-				{
-						asm volatile ("  movd         %0,%%mm0\n"
-																				"\tmovd         %1,%%mm1\n"
-																				"\tpxor      %%mm2,%%mm2\n"
-																				"\tpunpcklbw %%mm0,%%mm2\n" /* mm2 = A*256 */
+    {
+      asm volatile ("  movd         %0,%%mm0\n"
+                    "\tmovd         %1,%%mm1\n"
+                    "\tpxor      %%mm2,%%mm2\n"
+                    "\tpunpcklbw %%mm0,%%mm2\n" /* mm2 = A*256 */
 
-																				"\tmovq      %%mm1,%%mm3\n"
-																				"\tpxor      %%mm5,%%mm5\n"
-																				"\tpunpcklbw %%mm5,%%mm3\n"
-																				"\tpaddw     %%mm7,%%mm3\n" /* mm3 = B+1 */
+                    "\tmovq      %%mm1,%%mm3\n"
+                    "\tpxor      %%mm5,%%mm5\n"
+                    "\tpunpcklbw %%mm5,%%mm3\n"
+                    "\tpaddw     %%mm7,%%mm3\n" /* mm3 = B+1 */
 
-																				"\t" pdivwuqX(mm2,mm3,mm5) "\n" /* mm5 = (A*256)/(B+1) */
+                    "\t" pdivwuqX(mm2,mm3,mm5) "\n" /* mm5 = (A*256)/(B+1) */
 
-																				"\tpxor      %%mm2,%%mm2\n"
-																				"\tpunpckhbw %%mm0,%%mm2\n" /* mm2 = A*256 */
+                    "\tpxor      %%mm2,%%mm2\n"
+                    "\tpunpckhbw %%mm0,%%mm2\n" /* mm2 = A*256 */
 
-																				"\tmovq      %%mm1,%%mm3\n"
-																				"\tpxor      %%mm6,%%mm6\n"
-																				"\tpunpckhbw %%mm6,%%mm3\n"
-																				"\tpaddw     %%mm7,%%mm3\n" /* mm3 = B+1 */
+                    "\tmovq      %%mm1,%%mm3\n"
+                    "\tpxor      %%mm6,%%mm6\n"
+                    "\tpunpckhbw %%mm6,%%mm3\n"
+                    "\tpaddw     %%mm7,%%mm3\n" /* mm3 = B+1 */
 
-																				"\t" pdivwuqX(mm2,mm3,mm4) "\n" /* mm4 = (A*256)/(B+1) */
+                    "\t" pdivwuqX(mm2,mm3,mm4) "\n" /* mm4 = (A*256)/(B+1) */
                   
-																				"\tpackuswb  %%mm4,%%mm5\n" /* expects mm4 and mm5 to be signed values */
+                    "\tpackuswb  %%mm4,%%mm5\n" /* expects mm4 and mm5 to be signed values */
 
-																				"\t" pminub(mm0,mm1,mm3) "\n"
-																				"\tmovq         %3,%%mm3\n"
-																				"\tmovq      %%mm3,%%mm2\n"
+                    "\t" pminub(mm0,mm1,mm3) "\n"
+                    "\tmovq         %3,%%mm3\n"
+                    "\tmovq      %%mm3,%%mm2\n"
 
-																				"\tpandn     %%mm5,%%mm3\n"
+                    "\tpandn     %%mm5,%%mm3\n"
 
-																				"\tpand      %%mm2,%%mm1\n"
-																				"\tpor       %%mm1,%%mm3\n"
+                    "\tpand      %%mm2,%%mm1\n"
+                    "\tpor       %%mm1,%%mm3\n"
 
-																				"\tmovd      %%mm3,%2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D), "m" (*rgba8_alpha_mask)
-																				: "%eax", "%ecx", "%edx", "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
-				}
+                    "\tmovd      %%mm3,%2\n"
+                    : /* empty */
+                    : "m" (*op.A), "m" (*op.B), "m" (*op.D), "m" (*rgba8_alpha_mask)
+                    : "%eax", "%ecx", "%edx", "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
+    }
 
   asm("emms");
 }
@@ -531,87 +435,87 @@ gimp_composite_dodge_rgba8_rgba8_rgba8_mmx (GimpCompositeContext *_op)
   GimpCompositeContext op = *_op;
 
   for (; op.n_pixels >= 2; op.n_pixels -= 2)
-				{
-						asm volatile ("  movq         %0,%%mm0\n"
-																				"\tmovq         %1,%%mm1\n"
-																				"\tmovq      %%mm1,%%mm3\n"
-																				"\tpxor      %%mm2,%%mm2\n"
-																				"\tpunpcklbw %%mm2,%%mm3\n"
-																				"\tpunpcklbw %%mm0,%%mm2\n"
+    {
+      asm volatile ("  movq         %0,%%mm0\n"
+                    "\tmovq         %1,%%mm1\n"
+                    "\tmovq      %%mm1,%%mm3\n"
+                    "\tpxor      %%mm2,%%mm2\n"
+                    "\tpunpcklbw %%mm2,%%mm3\n"
+                    "\tpunpcklbw %%mm0,%%mm2\n"
 
-																				"\tmovq         %3,%%mm4\n"
-																				"\tpsubw     %%mm3,%%mm4\n"
+                    "\tmovq         %3,%%mm4\n"
+                    "\tpsubw     %%mm3,%%mm4\n"
 
-																				"\t" pdivwuqX(mm2,mm4,mm5) "\n"
+                    "\t" pdivwuqX(mm2,mm4,mm5) "\n"
 
-																				"\tmovq      %%mm1,%%mm3\n"
-																				"\tpxor      %%mm2,%%mm2\n"
-																				"\tpunpckhbw %%mm2,%%mm3\n"
-																				"\tpunpckhbw %%mm0,%%mm2\n"
+                    "\tmovq      %%mm1,%%mm3\n"
+                    "\tpxor      %%mm2,%%mm2\n"
+                    "\tpunpckhbw %%mm2,%%mm3\n"
+                    "\tpunpckhbw %%mm0,%%mm2\n"
 
-																				"\tmovq         %3,%%mm4\n"
-																				"\tpsubw     %%mm3,%%mm4\n"
+                    "\tmovq         %3,%%mm4\n"
+                    "\tpsubw     %%mm3,%%mm4\n"
 
-																				"\t" pdivwuqX(mm2,mm4,mm6) "\n"
+                    "\t" pdivwuqX(mm2,mm4,mm6) "\n"
 
-																				"\tpackuswb  %%mm6,%%mm5\n"
+                    "\tpackuswb  %%mm6,%%mm5\n"
 
-																				"\tmovq         %4,%%mm6\n"
-																				"\tmovq      %%mm1,%%mm7\n"
-																				"\t" pminub(mm0,mm7,mm2) "\n"
-																				"\tpand      %%mm6,%%mm7\n"
-																				"\tpandn     %%mm5,%%mm6\n"
+                    "\tmovq         %4,%%mm6\n"
+                    "\tmovq      %%mm1,%%mm7\n"
+                    "\t" pminub(mm0,mm7,mm2) "\n"
+                    "\tpand      %%mm6,%%mm7\n"
+                    "\tpandn     %%mm5,%%mm6\n"
 
-																				"\tpor       %%mm6,%%mm7\n"
+                    "\tpor       %%mm6,%%mm7\n"
 
-																				"\tmovq      %%mm7,%2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D), "m" (*rgba8_w256), "m" (*rgba8_alpha_mask)
-																				: "0", "1", "2", "%mm0", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5", "%mm6", "%mm7");
-						op.A += 8;
-						op.B += 8;
-						op.D += 8;
-				}
+                    "\tmovq      %%mm7,%2\n"
+                    : /* empty */
+                    : "m" (*op.A), "m" (*op.B), "m" (*op.D), "m" (*rgba8_w256), "m" (*rgba8_alpha_mask)
+                    : "0", "1", "2", "%mm0", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5", "%mm6", "%mm7");
+      op.A += 8;
+      op.B += 8;
+      op.D += 8;
+    }
 
   if (op.n_pixels)
-				{
-						asm volatile ("  movd         %0,%%mm0\n"
-																				"\tmovq         %1,%%mm1\n"
-																				"\tmovq      %%mm1,%%mm3\n"
-																				"\tpxor      %%mm2,%%mm2\n"
-																				"\tpunpcklbw %%mm2,%%mm3\n"
-																				"\tpunpcklbw %%mm0,%%mm2\n"
+    {
+      asm volatile ("  movd         %0,%%mm0\n"
+                    "\tmovq         %1,%%mm1\n"
+                    "\tmovq      %%mm1,%%mm3\n"
+                    "\tpxor      %%mm2,%%mm2\n"
+                    "\tpunpcklbw %%mm2,%%mm3\n"
+                    "\tpunpcklbw %%mm0,%%mm2\n"
 
-																				"\tmovq         %3,%%mm4\n"
-																				"\tpsubw     %%mm3,%%mm4\n"
+                    "\tmovq         %3,%%mm4\n"
+                    "\tpsubw     %%mm3,%%mm4\n"
 
-																				"\t" pdivwuqX(mm2,mm4,mm5) "\n"
+                    "\t" pdivwuqX(mm2,mm4,mm5) "\n"
 
-																				"\tmovq      %%mm1,%%mm3\n"
-																				"\tpxor      %%mm2,%%mm2\n"
-																				"\tpunpckhbw %%mm2,%%mm3\n"
-																				"\tpunpckhbw %%mm0,%%mm2\n"
+                    "\tmovq      %%mm1,%%mm3\n"
+                    "\tpxor      %%mm2,%%mm2\n"
+                    "\tpunpckhbw %%mm2,%%mm3\n"
+                    "\tpunpckhbw %%mm0,%%mm2\n"
 
-																				"\tmovq         %3,%%mm4\n"
-																				"\tpsubw     %%mm3,%%mm4\n"
+                    "\tmovq         %3,%%mm4\n"
+                    "\tpsubw     %%mm3,%%mm4\n"
 
-																				"\t" pdivwuqX(mm2,mm4,mm6) "\n"
+                    "\t" pdivwuqX(mm2,mm4,mm6) "\n"
 
-																				"\tpackuswb  %%mm6,%%mm5\n"
+                    "\tpackuswb  %%mm6,%%mm5\n"
 
-																				"\tmovq         %4,%%mm6\n"
-																				"\tmovq      %%mm1,%%mm7\n"
-																				"\t" pminub(mm0,mm7,mm2) "\n"
-																				"\tpand      %%mm6,%%mm7\n"
-																				"\tpandn     %%mm5,%%mm6\n"
+                    "\tmovq         %4,%%mm6\n"
+                    "\tmovq      %%mm1,%%mm7\n"
+                    "\t" pminub(mm0,mm7,mm2) "\n"
+                    "\tpand      %%mm6,%%mm7\n"
+                    "\tpandn     %%mm5,%%mm6\n"
 
-																				"\tpor       %%mm6,%%mm7\n"
+                    "\tpor       %%mm6,%%mm7\n"
 
-																				"\tmovd      %%mm7,%2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D), "m" (*rgba8_w256), "m" (*rgba8_alpha_mask)
-																				: "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
-				}
+                    "\tmovd      %%mm7,%2\n"
+                    : /* empty */
+                    : "m" (*op.A), "m" (*op.B), "m" (*op.D), "m" (*rgba8_w256), "m" (*rgba8_alpha_mask)
+                    : "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
+    }
 
   asm("emms");
 }
@@ -621,72 +525,72 @@ gimp_composite_grain_extract_rgba8_rgba8_rgba8_mmx (GimpCompositeContext *_op)
 {
   GimpCompositeContext op = *_op;
 
-		asm volatile ("movq    %0,%%mm0"     :  : "m" (*rgba8_alpha_mask) : "%mm0");
+  asm volatile ("movq    %0,%%mm0"     :  : "m" (*rgba8_alpha_mask) : "%mm0");
   asm volatile ("pxor    %%mm6,%%mm6"  :  :                        : "%mm6");
   asm volatile ("movq    %0,%%mm7"     :  : "m" (*rgba8_w128)       : "%mm7");
 
   for (; op.n_pixels >= 2; op.n_pixels -= 2)
-				{
-						asm volatile ("  movq         %0,%%mm2\n"
-																				"\tmovq         %1,%%mm3\n"
-																				mmx_low_bytes_to_words(mm2,mm4,mm6)
-																				mmx_low_bytes_to_words(mm3,mm5,mm6)
-																				"\tpsubw     %%mm5,%%mm4\n"
-																				"\tpaddw     %%mm7,%%mm4\n"
-																				"\tmovq      %%mm4,%%mm1\n"
+    {
+      asm volatile ("  movq         %0,%%mm2\n"
+                    "\tmovq         %1,%%mm3\n"
+                    mmx_low_bytes_to_words(mm2,mm4,mm6)
+                    mmx_low_bytes_to_words(mm3,mm5,mm6)
+                    "\tpsubw     %%mm5,%%mm4\n"
+                    "\tpaddw     %%mm7,%%mm4\n"
+                    "\tmovq      %%mm4,%%mm1\n"
 
-																				mmx_high_bytes_to_words(mm2,mm4,mm6)
-																				mmx_high_bytes_to_words(mm3,mm5,mm6)
+                    mmx_high_bytes_to_words(mm2,mm4,mm6)
+                    mmx_high_bytes_to_words(mm3,mm5,mm6)
 
-																				"\tpsubw     %%mm5,%%mm4\n"
-																				"\tpaddw     %%mm7,%%mm4\n"
+                    "\tpsubw     %%mm5,%%mm4\n"
+                    "\tpaddw     %%mm7,%%mm4\n"
 
-																				"\tpackuswb  %%mm4,%%mm1\n"
-																				"\tmovq      %%mm1,%%mm4\n"
+                    "\tpackuswb  %%mm4,%%mm1\n"
+                    "\tmovq      %%mm1,%%mm4\n"
 
-																				"\tmovq      %%mm0,%%mm1\n"
-																				"\tpandn     %%mm4,%%mm1\n"
-											
-																				"\t" pminub(mm3,mm2,mm4) "\n"
-																				"\tpand      %%mm0,%%mm2\n"
+                    "\tmovq      %%mm0,%%mm1\n"
+                    "\tpandn     %%mm4,%%mm1\n"
+           
+                    "\t" pminub(mm3,mm2,mm4) "\n"
+                    "\tpand      %%mm0,%%mm2\n"
 
-																				"\tpor       %%mm2,%%mm1\n"
-																				"\tmovq      %%mm1,%2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D)
-																				: "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
-						op.A += 8;
-						op.B += 8;
-						op.D += 8;
-				}
+                    "\tpor       %%mm2,%%mm1\n"
+                    "\tmovq      %%mm1,%2\n"
+                    : /* empty */
+                    : "m" (*op.A), "m" (*op.B), "m" (*op.D)
+                    : "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
+      op.A += 8;
+      op.B += 8;
+      op.D += 8;
+    }
 
   if (op.n_pixels)
-				{
-						asm volatile ("  movd    %0, %%mm2\n"
-																				"\tmovd    %1, %%mm3\n"
+    {
+      asm volatile ("  movd    %0, %%mm2\n"
+                    "\tmovd    %1, %%mm3\n"
 
-																				mmx_low_bytes_to_words(mm2,mm4,mm6)
-																				mmx_low_bytes_to_words(mm3,mm5,mm6)
+                    mmx_low_bytes_to_words(mm2,mm4,mm6)
+                    mmx_low_bytes_to_words(mm3,mm5,mm6)
 
-																				"\tpsubw     %%mm5, %%mm4\n"
-																				"\tpaddw     %%mm7, %%mm4\n"
-																				"\tmovq      %%mm4, %%mm1\n"
+                    "\tpsubw     %%mm5, %%mm4\n"
+                    "\tpaddw     %%mm7, %%mm4\n"
+                    "\tmovq      %%mm4, %%mm1\n"
 
-																				"\tpackuswb  %%mm6, %%mm1\n"
+                    "\tpackuswb  %%mm6, %%mm1\n"
 
-																				"\tmovq      %%mm1, %%mm4\n"
+                    "\tmovq      %%mm1, %%mm4\n"
 
-																				"\tmovq      %%mm0, %%mm1; pandn     %%mm4, %%mm1\n"
+                    "\tmovq      %%mm0, %%mm1; pandn     %%mm4, %%mm1\n"
 
-																				"\t" pminub(mm3,mm2,mm4) "\n"
-																				"\tpand      %%mm0, %%mm2\n"
+                    "\t" pminub(mm3,mm2,mm4) "\n"
+                    "\tpand      %%mm0, %%mm2\n"
 
-																				"\tpor       %%mm2, %%mm1\n"
-																				"\tmovd      %%mm1, %2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D)
-																				: "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
-				}
+                    "\tpor       %%mm2, %%mm1\n"
+                    "\tmovd      %%mm1, %2\n"
+                    : /* empty */
+                    : "m" (*op.A), "m" (*op.B), "m" (*op.D)
+                    : "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
+    }
 
   asm("emms");
 }
@@ -704,63 +608,63 @@ gimp_composite_grain_merge_rgba8_rgba8_rgba8_mmx (GimpCompositeContext *_op)
                 : "%mm0", "%mm6", "%mm7");
 
   for (; op.n_pixels >= 2; op.n_pixels -= 2)
-				{
-						asm volatile ("  movq    %0, %%mm2\n"
-																				"\tmovq    %1, %%mm3\n"
+    {
+      asm volatile ("  movq    %0, %%mm2\n"
+                    "\tmovq    %1, %%mm3\n"
 
-																				mmx_low_bytes_to_words(mm2,mm4,mm6)
-																				mmx_low_bytes_to_words(mm3,mm5,mm6)
-																				"\tpaddw     %%mm5, %%mm4\n"
-																				"\tpsubw     %%mm7, %%mm4\n"
+                    mmx_low_bytes_to_words(mm2,mm4,mm6)
+                    mmx_low_bytes_to_words(mm3,mm5,mm6)
+                    "\tpaddw     %%mm5, %%mm4\n"
+                    "\tpsubw     %%mm7, %%mm4\n"
 
-																				mmx_high_bytes_to_words(mm2,mm1,mm6)
-																				mmx_high_bytes_to_words(mm3,mm5,mm6)
-																				"\tpaddw     %%mm5, %%mm1\n"
-																				"\tpsubw     %%mm7, %%mm1\n"
+                    mmx_high_bytes_to_words(mm2,mm1,mm6)
+                    mmx_high_bytes_to_words(mm3,mm5,mm6)
+                    "\tpaddw     %%mm5, %%mm1\n"
+                    "\tpsubw     %%mm7, %%mm1\n"
 
-																				"\tpackuswb  %%mm1, %%mm4\n"
+                    "\tpackuswb  %%mm1, %%mm4\n"
          
-																				"\t" pminub(mm3,mm2,mm5) "\n"
-																				"\tpand      %%mm0, %%mm2\n"
+                    "\t" pminub(mm3,mm2,mm5) "\n"
+                    "\tpand      %%mm0, %%mm2\n"
 
-																				"\tmovq      %%mm0, %%mm1\n"
-																				"\tpandn     %%mm4, %%mm1\n"
-																				"\tpor       %%mm2, %%mm1\n"
-																				"\tmovq      %%mm1, %2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D)
-																				: "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
-						op.A += 8;
-						op.B += 8;
-						op.D += 8;
-				}
+                    "\tmovq      %%mm0, %%mm1\n"
+                    "\tpandn     %%mm4, %%mm1\n"
+                    "\tpor       %%mm2, %%mm1\n"
+                    "\tmovq      %%mm1, %2\n"
+                    : /* empty */
+                    : "m" (*op.A), "m" (*op.B), "m" (*op.D)
+                    : "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
+      op.A += 8;
+      op.B += 8;
+      op.D += 8;
+    }
 
   if (op.n_pixels)
-				{
-						asm volatile ("  movd    %0, %%mm2\n"
-																				"\tmovd    %1, %%mm3\n"
+    {
+      asm volatile ("  movd    %0, %%mm2\n"
+                    "\tmovd    %1, %%mm3\n"
 
-																				mmx_low_bytes_to_words(mm2,mm4,mm6)
-																				mmx_low_bytes_to_words(mm3,mm5,mm6)
-																				
-																				"\tpaddw     %%mm5, %%mm4\n"
-																				"\tpsubw     %%mm7, %%mm4\n"
-																				"\tmovq      %%mm4, %%mm1\n"
-																				"\tpackuswb  %%mm6, %%mm1\n"
-																				
-																				"\tmovq      %%mm1, %%mm4\n"
-																				
-																				"\tmovq      %%mm0, %%mm1; pandn     %%mm4, %%mm1\n"
-																				
-																				"\t" pminub(mm3,mm2,mm4) "\n"
-																				"\tpand      %%mm0, %%mm2\n"
-																				
-																				"\tpor       %%mm2, %%mm1\n"
-																				"\tmovd      %%mm1, %2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D)
-																				: "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
-				}
+                    mmx_low_bytes_to_words(mm2,mm4,mm6)
+                    mmx_low_bytes_to_words(mm3,mm5,mm6)
+                    
+                    "\tpaddw     %%mm5, %%mm4\n"
+                    "\tpsubw     %%mm7, %%mm4\n"
+                    "\tmovq      %%mm4, %%mm1\n"
+                    "\tpackuswb  %%mm6, %%mm1\n"
+                    
+                    "\tmovq      %%mm1, %%mm4\n"
+                    
+                    "\tmovq      %%mm0, %%mm1; pandn     %%mm4, %%mm1\n"
+                    
+                    "\t" pminub(mm3,mm2,mm4) "\n"
+                    "\tpand      %%mm0, %%mm2\n"
+                    
+                    "\tpor       %%mm2, %%mm1\n"
+                    "\tmovd      %%mm1, %2\n"
+                    : /* empty */
+                    : "m" (*op.A), "m" (*op.B), "m" (*op.D)
+                    : "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
+    }
 
   asm("emms");
 }
@@ -773,27 +677,27 @@ gimp_composite_lighten_rgba8_rgba8_rgba8_mmx (GimpCompositeContext *_op)
   asm volatile ("movq    %0,%%mm0"     :  : "m" (*rgba8_alpha_mask) : "%mm0");
 
   for (; op.n_pixels >= 2; op.n_pixels -= 2)
-				{
-						asm volatile ("  movq       %0, %%mm2\n"
-																				"\tmovq       %1, %%mm3\n"
-																				"\tmovq    %%mm2, %%mm4\n"
-																				"\t" pmaxub(mm3,mm4,mm5) "\n"
-																				"\tmovq    %%mm0, %%mm1\n"
-																				"\tpandn   %%mm4, %%mm1\n"
-																				"\t" pminub(mm2,mm3,mm4) "\n"
-																				"\tpand    %%mm0, %%mm3\n"
-																				"\tpor     %%mm3, %%mm1\n"
-																				"\tmovq    %%mm1, %2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D)
-																				: "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
-						op.A += 8;
-						op.B += 8;
-						op.D += 8;
-				}
+    {
+      asm volatile ("  movq       %0, %%mm2\n"
+                    "\tmovq       %1, %%mm3\n"
+                    "\tmovq    %%mm2, %%mm4\n"
+                    "\t" pmaxub(mm3,mm4,mm5) "\n"
+                    "\tmovq    %%mm0, %%mm1\n"
+                    "\tpandn   %%mm4, %%mm1\n"
+                    "\t" pminub(mm2,mm3,mm4) "\n"
+                    "\tpand    %%mm0, %%mm3\n"
+                    "\tpor     %%mm3, %%mm1\n"
+                    "\tmovq    %%mm1, %2\n"
+                    : /* empty */
+                    : "m" (*op.A), "m" (*op.B), "m" (*op.D)
+                    : "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
+      op.A += 8;
+      op.B += 8;
+      op.D += 8;
+    }
 
   if (op.n_pixels)
-				{
+    {
     asm volatile ("  movd    %0, %%mm2\n"
                   "\tmovd    %1, %%mm3\n"
                   "\tmovq    %%mm2, %%mm4\n"
@@ -825,38 +729,38 @@ gimp_composite_multiply_rgba8_rgba8_rgba8_mmx (GimpCompositeContext *_op)
   asm volatile ("pxor    %%mm6,%%mm6"  :  :  : "%mm6");
 
   for (; op.n_pixels >= 2; op.n_pixels -= 2)
-				{
-						asm volatile ("  movq     %0, %%mm2\n"
-																				"\tmovq     %1, %%mm3\n"
+    {
+      asm volatile ("  movq     %0, %%mm2\n"
+                    "\tmovq     %1, %%mm3\n"
 
-																				mmx_low_bytes_to_words(mm2,mm1,mm6)
-																				mmx_low_bytes_to_words(mm3,mm5,mm6)
-																				mmx_int_mult(mm5,mm1,mm7)
+                    mmx_low_bytes_to_words(mm2,mm1,mm6)
+                    mmx_low_bytes_to_words(mm3,mm5,mm6)
+                    mmx_int_mult(mm5,mm1,mm7)
          
-																				mmx_high_bytes_to_words(mm2,mm4,mm6)
-																				mmx_high_bytes_to_words(mm3,mm5,mm6)
-																				mmx_int_mult(mm5,mm4,mm7)
+                    mmx_high_bytes_to_words(mm2,mm4,mm6)
+                    mmx_high_bytes_to_words(mm3,mm5,mm6)
+                    mmx_int_mult(mm5,mm4,mm7)
          
-																				"\tpackuswb  %%mm4, %%mm1\n"
+                    "\tpackuswb  %%mm4, %%mm1\n"
          
-																				"\tmovq      %%mm0, %%mm4\n"
-																				"\tpandn     %%mm1, %%mm4\n"
-																				"\tmovq      %%mm4, %%mm1\n"
-																				"\t" pminub(mm3,mm2,mm4) "\n"
-																				"\tpand      %%mm0, %%mm2\n"
-																				"\tpor       %%mm2, %%mm1\n"
-																				
-																				"\tmovq    %%mm1, %2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D)
-																				: "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
-						op.A += 8;
-						op.B += 8;
-						op.D += 8;
+                    "\tmovq      %%mm0, %%mm4\n"
+                    "\tpandn     %%mm1, %%mm4\n"
+                    "\tmovq      %%mm4, %%mm1\n"
+                    "\t" pminub(mm3,mm2,mm4) "\n"
+                    "\tpand      %%mm0, %%mm2\n"
+                    "\tpor       %%mm2, %%mm1\n"
+                    
+                    "\tmovq    %%mm1, %2\n"
+                    : /* empty */
+                    : "m" (*op.A), "m" (*op.B), "m" (*op.D)
+                    : "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
+      op.A += 8;
+      op.B += 8;
+      op.D += 8;
   }
 
   if (op.n_pixels)
-				{
+    {
     asm volatile ("  movd     %0, %%mm2\n"
                   "\tmovd     %1, %%mm3\n"
 
@@ -940,114 +844,114 @@ xxxgimp_composite_overlay_rgba8_rgba8_rgba8_mmx (GimpCompositeContext *_op)
   asm volatile ("pxor    %%mm0,%%mm0\n"
                 "movq       %0,%%mm7"
                 : /* empty */
-																: "m" (*rgba8_w128) : "%mm0");
+                : "m" (*rgba8_w128) : "%mm0");
 
   for (; op.n_pixels >= 2; op.n_pixels -= 2)
-				{
-						asm volatile ("  movq         %0,%%mm2\n"
-																				"\tmovq         %1,%%mm3\n"
-																				
-																				/* low bytes */
-																				mmx_low_bytes_to_words(mm3,mm5,mm0)
-																				"\tpcmpeqb   %%mm4,%%mm4\n"
-																				"\tpsubb     %%mm2,%%mm4\n" /* mm4 = 255 - A */
-																				"\tpunpcklbw %%mm0,%%mm4\n" /* mm4 = (low bytes as word) mm4 */
-																				"\tmovq         %3,%%mm6\n"  /* mm6 = words of value 2 */
-																				"\tpmullw    %%mm5,%%mm6\n" /* mm6 = 2 * low bytes of B */
-																				mmx_int_mult(mm6,mm4,mm7)    /* mm4 = INT_MULT(mm6, mm4) */
-																				
-																				/* high bytes */
-																				mmx_high_bytes_to_words(mm3,mm5,mm0)
-																				"\tpcmpeqb   %%mm1,%%mm1\n"
-																				"\tpsubb     %%mm2,%%mm1\n" /* mm1 = 255 - A */
-																				"\tpunpckhbw %%mm0,%%mm1\n" /* mm1 = (high bytes as word) mm1 */
-																				"\tmovq         %3,%%mm6\n"  /* mm6 = words of value 2 */
-																				"\tpmullw    %%mm5,%%mm6\n" /* mm6 = 2 * high bytes of B */
-																				mmx_int_mult(mm6,mm1,mm7)    /* mm1 = INT_MULT(mm6, mm1) */
-																				
-																				"\tpackuswb  %%mm1,%%mm4\n"  /* mm4 = intermediate value */
-																				
-																				mmx_low_bytes_to_words(mm4,mm5,mm0)
-																				mmx_low_bytes_to_words(mm2,mm6,mm0)
-																				"\tpaddw     %%mm6,%%mm5\n"
-																				mmx_int_mult(mm6,mm5,mm7)   /* mm5 = INT_MULT(mm6, mm5) low bytes */
-																				
-																				mmx_high_bytes_to_words(mm4,mm1,mm0)
-																				mmx_high_bytes_to_words(mm2,mm6,mm0)
-																				"\tpaddw     %%mm6,%%mm1\n"
-																				mmx_int_mult(mm6,mm1,mm7)   /* mm1 = INT_MULT(mm6, mm1) high bytes */
-																				
-																				"\tpackuswb  %%mm1,%%mm5\n"
-																				
-																				"\tmovq         %4,%%mm0\n"
-																				"\tmovq      %%mm0,%%mm1\n"
-																				"\tpandn     %%mm5,%%mm1\n"
-																				
-																				"\t" pminub(mm2,mm3,mm4) "\n"
-																				"\tpand      %%mm0,%%mm3\n"
-																				
-																				"\tpor       %%mm3,%%mm1\n"
-																				
-																				"\tmovq      %%mm1,%2\n"
-																				: "+m" (*op.A), "+m" (*op.B), "+m" (*op.D)
-																				: "m" (*rgba8_w2), "m" (*rgba8_alpha_mask)
-																				: "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
-						op.A += 8;
-						op.B += 8;
-						op.D += 8;
+    {
+      asm volatile ("  movq         %0,%%mm2\n"
+                    "\tmovq         %1,%%mm3\n"
+                    
+                    /* low bytes */
+                    mmx_low_bytes_to_words(mm3,mm5,mm0)
+                    "\tpcmpeqb   %%mm4,%%mm4\n"
+                    "\tpsubb     %%mm2,%%mm4\n" /* mm4 = 255 - A */
+                    "\tpunpcklbw %%mm0,%%mm4\n" /* mm4 = (low bytes as word) mm4 */
+                    "\tmovq         %3,%%mm6\n"  /* mm6 = words of value 2 */
+                    "\tpmullw    %%mm5,%%mm6\n" /* mm6 = 2 * low bytes of B */
+                    mmx_int_mult(mm6,mm4,mm7)    /* mm4 = INT_MULT(mm6, mm4) */
+                    
+                    /* high bytes */
+                    mmx_high_bytes_to_words(mm3,mm5,mm0)
+                    "\tpcmpeqb   %%mm1,%%mm1\n"
+                    "\tpsubb     %%mm2,%%mm1\n" /* mm1 = 255 - A */
+                    "\tpunpckhbw %%mm0,%%mm1\n" /* mm1 = (high bytes as word) mm1 */
+                    "\tmovq         %3,%%mm6\n"  /* mm6 = words of value 2 */
+                    "\tpmullw    %%mm5,%%mm6\n" /* mm6 = 2 * high bytes of B */
+                    mmx_int_mult(mm6,mm1,mm7)    /* mm1 = INT_MULT(mm6, mm1) */
+                    
+                    "\tpackuswb  %%mm1,%%mm4\n"  /* mm4 = intermediate value */
+                    
+                    mmx_low_bytes_to_words(mm4,mm5,mm0)
+                    mmx_low_bytes_to_words(mm2,mm6,mm0)
+                    "\tpaddw     %%mm6,%%mm5\n"
+                    mmx_int_mult(mm6,mm5,mm7)   /* mm5 = INT_MULT(mm6, mm5) low bytes */
+                    
+                    mmx_high_bytes_to_words(mm4,mm1,mm0)
+                    mmx_high_bytes_to_words(mm2,mm6,mm0)
+                    "\tpaddw     %%mm6,%%mm1\n"
+                    mmx_int_mult(mm6,mm1,mm7)   /* mm1 = INT_MULT(mm6, mm1) high bytes */
+                    
+                    "\tpackuswb  %%mm1,%%mm5\n"
+                    
+                    "\tmovq         %4,%%mm0\n"
+                    "\tmovq      %%mm0,%%mm1\n"
+                    "\tpandn     %%mm5,%%mm1\n"
+                    
+                    "\t" pminub(mm2,mm3,mm4) "\n"
+                    "\tpand      %%mm0,%%mm3\n"
+                    
+                    "\tpor       %%mm3,%%mm1\n"
+                    
+                    "\tmovq      %%mm1,%2\n"
+                    : "+m" (*op.A), "+m" (*op.B), "+m" (*op.D)
+                    : "m" (*rgba8_w2), "m" (*rgba8_alpha_mask)
+                    : "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
+      op.A += 8;
+      op.B += 8;
+      op.D += 8;
   }
 
   if (op.n_pixels)
-				{
-						asm volatile ("  movd         %0,%%mm2\n"
-																				"\tmovd         %1,%%mm3\n"
+    {
+      asm volatile ("  movd         %0,%%mm2\n"
+                    "\tmovd         %1,%%mm3\n"
 
-																				/* low bytes */
-																				mmx_low_bytes_to_words(mm3,mm5,mm0)
-																				"\tpcmpeqb   %%mm4,%%mm4\n"
-																				"\tpsubb     %%mm2,%%mm4\n" /* mm4 = 255 - A */
-																				"\tpunpcklbw %%mm0,%%mm4\n" /* mm4 = (low bytes as word) mm4 */
-																				"\tmovq         %3,%%mm6\n"  /* mm6 = words of value 2 */
-																				"\tpmullw    %%mm5,%%mm6\n" /* mm6 = 2 * low bytes of B */
-																				mmx_int_mult(mm6,mm4,mm7)    /* mm4 = INT_MULT(mm6, mm4) */
+                    /* low bytes */
+                    mmx_low_bytes_to_words(mm3,mm5,mm0)
+                    "\tpcmpeqb   %%mm4,%%mm4\n"
+                    "\tpsubb     %%mm2,%%mm4\n" /* mm4 = 255 - A */
+                    "\tpunpcklbw %%mm0,%%mm4\n" /* mm4 = (low bytes as word) mm4 */
+                    "\tmovq         %3,%%mm6\n"  /* mm6 = words of value 2 */
+                    "\tpmullw    %%mm5,%%mm6\n" /* mm6 = 2 * low bytes of B */
+                    mmx_int_mult(mm6,mm4,mm7)    /* mm4 = INT_MULT(mm6, mm4) */
 
-																				/* high bytes */
-																				mmx_high_bytes_to_words(mm3,mm5,mm0)
-																				"\tpcmpeqb   %%mm1,%%mm1\n"
-																				"\tpsubb     %%mm2,%%mm1\n" /* mm1 = 255 - A */
-																				"\tpunpckhbw %%mm0,%%mm1\n" /* mm1 = (high bytes as word) mm1 */
-																				"\tmovq         %3,%%mm6\n"  /* mm6 = words of value 2 */
-																				"\tpmullw    %%mm5,%%mm6\n" /* mm6 = 2 * high bytes of B */
-																				mmx_int_mult(mm6,mm1,mm7)    /* mm1 = INT_MULT(mm6, mm1) */
+                    /* high bytes */
+                    mmx_high_bytes_to_words(mm3,mm5,mm0)
+                    "\tpcmpeqb   %%mm1,%%mm1\n"
+                    "\tpsubb     %%mm2,%%mm1\n" /* mm1 = 255 - A */
+                    "\tpunpckhbw %%mm0,%%mm1\n" /* mm1 = (high bytes as word) mm1 */
+                    "\tmovq         %3,%%mm6\n"  /* mm6 = words of value 2 */
+                    "\tpmullw    %%mm5,%%mm6\n" /* mm6 = 2 * high bytes of B */
+                    mmx_int_mult(mm6,mm1,mm7)    /* mm1 = INT_MULT(mm6, mm1) */
 
-																				"\tpackuswb  %%mm1,%%mm4\n"  /* mm4 = intermediate value */
+                    "\tpackuswb  %%mm1,%%mm4\n"  /* mm4 = intermediate value */
 
-																				mmx_low_bytes_to_words(mm4,mm5,mm0)
-																				mmx_low_bytes_to_words(mm2,mm6,mm0)
-																				"\tpaddw     %%mm6,%%mm5\n"
-																				mmx_int_mult(mm6,mm5,mm7)   /* mm5 = INT_MULT(mm6, mm5) low bytes */
+                    mmx_low_bytes_to_words(mm4,mm5,mm0)
+                    mmx_low_bytes_to_words(mm2,mm6,mm0)
+                    "\tpaddw     %%mm6,%%mm5\n"
+                    mmx_int_mult(mm6,mm5,mm7)   /* mm5 = INT_MULT(mm6, mm5) low bytes */
 
-																				mmx_high_bytes_to_words(mm4,mm1,mm0)
-																				mmx_high_bytes_to_words(mm2,mm6,mm0)
-																				"\tpaddw     %%mm6,%%mm1\n"
-																				mmx_int_mult(mm6,mm1,mm7)   /* mm1 = INT_MULT(mm6, mm1) high bytes */
+                    mmx_high_bytes_to_words(mm4,mm1,mm0)
+                    mmx_high_bytes_to_words(mm2,mm6,mm0)
+                    "\tpaddw     %%mm6,%%mm1\n"
+                    mmx_int_mult(mm6,mm1,mm7)   /* mm1 = INT_MULT(mm6, mm1) high bytes */
 
-																				"\tpackuswb  %%mm1,%%mm5\n"
+                    "\tpackuswb  %%mm1,%%mm5\n"
 
-																				"\tmovq         %4,%%mm0\n"
-																				"\tmovq      %%mm0,%%mm1\n"
-																				"\tpandn     %%mm5,%%mm1\n"
+                    "\tmovq         %4,%%mm0\n"
+                    "\tmovq      %%mm0,%%mm1\n"
+                    "\tpandn     %%mm5,%%mm1\n"
 
-																				"\t" pminub(mm2,mm3,mm4) "\n"
-																				"\tpand      %%mm0,%%mm3\n"
+                    "\t" pminub(mm2,mm3,mm4) "\n"
+                    "\tpand      %%mm0,%%mm3\n"
 
-																				"\tpor       %%mm3,%%mm1\n"
+                    "\tpor       %%mm3,%%mm1\n"
 
-																				"\tmovd      %%mm1,%2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D), "m" (*rgba8_w2), "m" (*rgba8_alpha_mask)
-																				: "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
-				}
+                    "\tmovd      %%mm1,%2\n"
+                    : /* empty */
+                    : "m" (*op.A), "m" (*op.B), "m" (*op.D), "m" (*rgba8_w2), "m" (*rgba8_alpha_mask)
+                    : "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
+    }
 
   asm("emms");
 }
@@ -1073,32 +977,32 @@ gimp_composite_scale_rgba8_rgba8_rgba8_mmx (GimpCompositeContext *_op)
                 : "%eax", "%mm0", "%mm5", "%mm6", "%mm7");
 
   for (; op.n_pixels >= 2; op.n_pixels -= 2)
-				{
-						asm volatile ("movq           %0,%%mm2\n"
-																				"\tmovq      %%mm2,%%mm1\n"
-																				"\tpunpcklbw %%mm0,%%mm1\n"
-																				"\tmovq      %%mm3,%%mm5\n"
+    {
+      asm volatile ("movq           %0,%%mm2\n"
+                    "\tmovq      %%mm2,%%mm1\n"
+                    "\tpunpcklbw %%mm0,%%mm1\n"
+                    "\tmovq      %%mm3,%%mm5\n"
 
-																				"\t" pmulwX(mm5,mm1,mm7) "\n"
+                    "\t" pmulwX(mm5,mm1,mm7) "\n"
 
-																				"\tmovq      %%mm2,%%mm4\n"
-																				"\tpunpckhbw %%mm0,%%mm4\n"
-																				"\tmovq      %%mm3,%%mm5\n"
+                    "\tmovq      %%mm2,%%mm4\n"
+                    "\tpunpckhbw %%mm0,%%mm4\n"
+                    "\tmovq      %%mm3,%%mm5\n"
 
-																				"\t" pmulwX(mm5,mm4,mm7) "\n"
+                    "\t" pmulwX(mm5,mm4,mm7) "\n"
 
-																				"\tpackuswb  %%mm4,%%mm1\n"
-											
-																				"\tmovq    %%mm1,%1\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.D)
-																				: "0", "1", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5", "%mm6", "%mm7");
-						op.A += 8;
-						op.D += 8;
-				}
+                    "\tpackuswb  %%mm4,%%mm1\n"
+           
+                    "\tmovq    %%mm1,%1\n"
+                    : /* empty */
+                    : "m" (*op.A), "m" (*op.D)
+                    : "0", "1", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5", "%mm6", "%mm7");
+      op.A += 8;
+      op.D += 8;
+    }
 
   if (op.n_pixels)
-				{
+    {
     asm volatile ("movd           %0,%%mm2\n"
                   "\tmovq      %%mm2,%%mm1\n"
                   "\tpunpcklbw %%mm0,%%mm1\n"
@@ -1126,62 +1030,62 @@ gimp_composite_screen_rgba8_rgba8_rgba8_mmx (GimpCompositeContext *_op)
   asm volatile ("pxor    %mm6, %mm6");
 
   for (; op.n_pixels >= 2; op.n_pixels -= 2)
-				{
-						asm volatile ("  movq         %0,%%mm2\n"
-																				"\tmovq         %1,%%mm3\n"
+    {
+      asm volatile ("  movq         %0,%%mm2\n"
+                    "\tmovq         %1,%%mm3\n"
 
-																				"\tpcmpeqb   %%mm4,%%mm4\n"
-																				"\tpsubb     %%mm2,%%mm4\n"
-																				"\tpcmpeqb   %%mm5,%%mm5\n"
-																				"\tpsubb     %%mm3,%%mm5\n"
+                    "\tpcmpeqb   %%mm4,%%mm4\n"
+                    "\tpsubb     %%mm2,%%mm4\n"
+                    "\tpcmpeqb   %%mm5,%%mm5\n"
+                    "\tpsubb     %%mm3,%%mm5\n"
 
-																				"\tpunpcklbw %%mm6,%%mm4\n"
-																				"\tpunpcklbw %%mm6,%%mm5\n"
-																				"\tpmullw    %%mm4,%%mm5\n"
-																				"\tpaddw     %%mm7,%%mm5\n"
-																				"\tmovq      %%mm5,%%mm1\n"
-																				"\tpsrlw       $ 8,%%mm1\n"
-																				"\tpaddw     %%mm5,%%mm1\n"
-																				"\tpsrlw       $ 8,%%mm1\n"
+                    "\tpunpcklbw %%mm6,%%mm4\n"
+                    "\tpunpcklbw %%mm6,%%mm5\n"
+                    "\tpmullw    %%mm4,%%mm5\n"
+                    "\tpaddw     %%mm7,%%mm5\n"
+                    "\tmovq      %%mm5,%%mm1\n"
+                    "\tpsrlw       $ 8,%%mm1\n"
+                    "\tpaddw     %%mm5,%%mm1\n"
+                    "\tpsrlw       $ 8,%%mm1\n"
 
-																				"\tpcmpeqb   %%mm4,%%mm4\n"
-																				"\tpsubb     %%mm2,%%mm4\n"
-																				"\tpcmpeqb   %%mm5,%%mm5\n"
-																				"\tpsubb     %%mm3,%%mm5\n"
-											
-																				"\tpunpckhbw %%mm6,%%mm4\n"
-																				"\tpunpckhbw %%mm6,%%mm5\n"
-																				"\tpmullw    %%mm4,%%mm5\n"
-																				"\tpaddw     %%mm7,%%mm5\n"
-																				"\tmovq      %%mm5,%%mm4\n"
-																				"\tpsrlw       $ 8,%%mm4\n"
-																				"\tpaddw     %%mm5,%%mm4\n"
-																				"\tpsrlw       $ 8,%%mm4\n"
+                    "\tpcmpeqb   %%mm4,%%mm4\n"
+                    "\tpsubb     %%mm2,%%mm4\n"
+                    "\tpcmpeqb   %%mm5,%%mm5\n"
+                    "\tpsubb     %%mm3,%%mm5\n"
+           
+                    "\tpunpckhbw %%mm6,%%mm4\n"
+                    "\tpunpckhbw %%mm6,%%mm5\n"
+                    "\tpmullw    %%mm4,%%mm5\n"
+                    "\tpaddw     %%mm7,%%mm5\n"
+                    "\tmovq      %%mm5,%%mm4\n"
+                    "\tpsrlw       $ 8,%%mm4\n"
+                    "\tpaddw     %%mm5,%%mm4\n"
+                    "\tpsrlw       $ 8,%%mm4\n"
 
-																				"\tpackuswb  %%mm4,%%mm1\n"
+                    "\tpackuswb  %%mm4,%%mm1\n"
 
-																				"\tpcmpeqb   %%mm4,%%mm4\n"
-																				"\tpsubb     %%mm1,%%mm4\n"
+                    "\tpcmpeqb   %%mm4,%%mm4\n"
+                    "\tpsubb     %%mm1,%%mm4\n"
 
-																				"\tmovq      %%mm0,%%mm1\n"
-																				"\tpandn     %%mm4,%%mm1\n"
-											
-																				"\t" pminub(mm2,mm3,mm4) "\n"
-																				"\tpand      %%mm0,%%mm3\n"
+                    "\tmovq      %%mm0,%%mm1\n"
+                    "\tpandn     %%mm4,%%mm1\n"
+           
+                    "\t" pminub(mm2,mm3,mm4) "\n"
+                    "\tpand      %%mm0,%%mm3\n"
 
-																				"\tpor       %%mm3,%%mm1\n"
-											
-																				"\tmovq      %%mm1,%2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D)
-																				: "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
-						op.A += 8;
-						op.B += 8;
-						op.D += 8;
+                    "\tpor       %%mm3,%%mm1\n"
+           
+                    "\tmovq      %%mm1,%2\n"
+                    : /* empty */
+                    : "m" (*op.A), "m" (*op.B), "m" (*op.D)
+                    : "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
+      op.A += 8;
+      op.B += 8;
+      op.D += 8;
   }
 
   if (op.n_pixels)
-				{
+    {
     asm volatile ("  movd         %0,%%mm2\n"
                   "\tmovd         %1,%%mm3\n"
 
@@ -1244,31 +1148,31 @@ gimp_composite_subtract_rgba8_rgba8_rgba8_mmx (GimpCompositeContext *_op)
   asm volatile ("movq    %0,%%mm0"     :  : "m" (*rgba8_alpha_mask) : "%mm0");
 
   for (; op.n_pixels >= 2; op.n_pixels -= 2)
-				{
-						asm volatile ("  movq       %0,%%mm2\n"
-																				"\tmovq       %1,%%mm3\n"
+    {
+      asm volatile ("  movq       %0,%%mm2\n"
+                    "\tmovq       %1,%%mm3\n"
 
-																				"\tmovq    %%mm2,%%mm4\n"
-																				"\tpsubusb %%mm3,%%mm4\n"
+                    "\tmovq    %%mm2,%%mm4\n"
+                    "\tpsubusb %%mm3,%%mm4\n"
                   
-																				"\tmovq    %%mm0,%%mm1\n"
-																				"\tpandn   %%mm4,%%mm1\n"
-																				
-																				"\t" pminub(mm3,mm2,mm4) "\n"
-																				
-																				"\tpand    %%mm0,%%mm2\n"
-																				"\tpor     %%mm2,%%mm1\n"
-																				"\tmovq    %%mm1,%2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D)
-																				: "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
-						op.A += 8;
-						op.B += 8;
-						op.D += 8;
-				}
+                    "\tmovq    %%mm0,%%mm1\n"
+                    "\tpandn   %%mm4,%%mm1\n"
+                    
+                    "\t" pminub(mm3,mm2,mm4) "\n"
+                    
+                    "\tpand    %%mm0,%%mm2\n"
+                    "\tpor     %%mm2,%%mm1\n"
+                    "\tmovq    %%mm1,%2\n"
+                    : /* empty */
+                    : "m" (*op.A), "m" (*op.B), "m" (*op.D)
+                    : "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
+      op.A += 8;
+      op.B += 8;
+      op.D += 8;
+    }
 
   if (op.n_pixels)
-				{
+    {
     asm volatile ("  movd       %0,%%mm2\n"
                   "\tmovd       %1,%%mm3\n"
 
@@ -1297,20 +1201,20 @@ gimp_composite_swap_rgba8_rgba8_rgba8_mmx (GimpCompositeContext *_op)
   GimpCompositeContext op = *_op;
 
   for (; op.n_pixels >= 2; op.n_pixels -= 2)
-				{
-						asm volatile ("  movq       %0,%%mm2\n"
-																				"\tmovq       %1,%%mm3\n"
-																				"\tmovq    %%mm3,%0\n"
-																				"\tmovq    %%mm2,%1\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B)
-																				: "0", "1", "%mm1", "%mm2", "%mm3", "%mm4");
-						op.A += 8;
-						op.B += 8;
-				}
+    {
+      asm volatile ("  movq       %0,%%mm2\n"
+                    "\tmovq       %1,%%mm3\n"
+                    "\tmovq    %%mm3,%0\n"
+                    "\tmovq    %%mm2,%1\n"
+                    : /* empty */
+                    : "m" (*op.A), "m" (*op.B)
+                    : "0", "1", "%mm1", "%mm2", "%mm3", "%mm4");
+      op.A += 8;
+      op.B += 8;
+    }
 
   if (op.n_pixels)
-				{
+    {
     asm volatile ("  movd       %0,%%mm2\n"
                   "\tmovd       %1,%%mm3\n"
                   "\tmovd    %%mm3,%0\n"
@@ -1421,7 +1325,7 @@ gimp_composite_burn_va8_va8_va8_mmx (GimpCompositeContext *_op)
       : "%mm1");
 
   for (; op.n_pixels >= 4; op.n_pixels -= 4)
-				{
+    {
     asm volatile ("  movq         %0,%%mm0\n"
                   "\tmovq         %1,%%mm1\n"
 
@@ -1466,16 +1370,16 @@ gimp_composite_burn_va8_va8_va8_mmx (GimpCompositeContext *_op)
                   "\tpor       %%mm1,%%mm7\n" /* mm7 = mm7 | mm1 */
 
                   "\tmovq      %%mm7,%2\n"
-																		: /* empty */
+                  : /* empty */
                   : "+m" (*op.A), "+m" (*op.B), "+m" (*op.D), "m" (*va8_b255), "m" (*va8_w1), "m" (*va8_w255), "m" (*va8_alpha_mask)
                   : "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
-						op.A += 8;
-						op.B += 8;
-						op.D += 8;
+      op.A += 8;
+      op.B += 8;
+      op.D += 8;
   }
 
   if (op.n_pixels)
-				{
+    {
     asm volatile ("  movd         %0,%%mm0\n"
                   "\tmovd         %1,%%mm1\n"
                   "\tmovq         %3,%%mm2\n"
@@ -2315,4 +2219,3 @@ gimp_composite_mmx_init (void)
 {
 
 }
-
diff --git a/app/composite/gimp-composite-regression.c b/app/composite/gimp-composite-regression.c
index 8e49a25b05..b3835950cd 100644
--- a/app/composite/gimp-composite-regression.c
+++ b/app/composite/gimp-composite-regression.c
@@ -127,7 +127,6 @@ gimp_composite_regression_compare_contexts (char *operation, GimpCompositeContex
 #endif
   }
 
-
   return (0);
 }
 
diff --git a/app/composite/gimp-composite-sse-test.c b/app/composite/gimp-composite-sse-test.c
index 88fba8945b..637f34a88f 100644
--- a/app/composite/gimp-composite-sse-test.c
+++ b/app/composite/gimp-composite-sse-test.c
@@ -57,74 +57,52 @@ gimp_composite_sse_test (int iterations, int n_pixels)
   }
 
 
-  gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_MULTIPLY, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
-  gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_MULTIPLY, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
-  ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
-  ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_multiply_rgba8_rgba8_rgba8_sse, &special_ctx);
-  if (gimp_composite_regression_compare_contexts ("multiply", &generic_ctx, &special_ctx)) {
-    return (1);
-  }
-  gimp_composite_regression_timer_report ("multiply", ft0, ft1);
-
-  gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_SCREEN, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
-  gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_SCREEN, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
-  ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
-  ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_screen_rgba8_rgba8_rgba8_sse, &special_ctx);
-  if (gimp_composite_regression_compare_contexts ("screen", &generic_ctx, &special_ctx)) {
-    return (1);
-  }
-  gimp_composite_regression_timer_report ("screen", ft0, ft1);
-
-  gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_DIFFERENCE, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
-  gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_DIFFERENCE, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
-  ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
-  ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_difference_rgba8_rgba8_rgba8_sse, &special_ctx);
-  if (gimp_composite_regression_compare_contexts ("difference", &generic_ctx, &special_ctx)) {
-    return (1);
-  }
-  gimp_composite_regression_timer_report ("difference", ft0, ft1);
-
   gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_ADDITION, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
   gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_ADDITION, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
   ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
   ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_addition_rgba8_rgba8_rgba8_sse, &special_ctx);
   if (gimp_composite_regression_compare_contexts ("addition", &generic_ctx, &special_ctx)) {
+    printf("addition failed\n");
     return (1);
   }
   gimp_composite_regression_timer_report ("addition", ft0, ft1);
 
-  gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_SUBTRACT, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
-  gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_SUBTRACT, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
+  gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_BURN, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
+  gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_BURN, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
   ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
-  ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_subtract_rgba8_rgba8_rgba8_sse, &special_ctx);
-  if (gimp_composite_regression_compare_contexts ("subtract", &generic_ctx, &special_ctx)) {
+  ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_burn_rgba8_rgba8_rgba8_sse, &special_ctx);
+  if (gimp_composite_regression_compare_contexts ("burn", &generic_ctx, &special_ctx)) {
+    printf("burn failed\n");
     return (1);
   }
-  gimp_composite_regression_timer_report ("subtract", ft0, ft1);
+  gimp_composite_regression_timer_report ("burn", ft0, ft1);
 
   gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_DARKEN, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
   gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_DARKEN, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
   ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
   ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_darken_rgba8_rgba8_rgba8_sse, &special_ctx);
   if (gimp_composite_regression_compare_contexts ("darken", &generic_ctx, &special_ctx)) {
+    printf("darken failed\n");
     return (1);
   }
   gimp_composite_regression_timer_report ("darken", ft0, ft1);
 
-  gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_LIGHTEN, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
-  gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_LIGHTEN, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
+  gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_DIFFERENCE, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
+  gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_DIFFERENCE, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
   ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
-  ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_lighten_rgba8_rgba8_rgba8_sse, &special_ctx);
-  if (gimp_composite_regression_compare_contexts ("lighten", &generic_ctx, &special_ctx)) {
+  ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_difference_rgba8_rgba8_rgba8_sse, &special_ctx);
+  if (gimp_composite_regression_compare_contexts ("difference", &generic_ctx, &special_ctx)) {
+    printf("difference failed\n");
     return (1);
   }
-  gimp_composite_regression_timer_report ("lighten", ft0, ft1);
+  gimp_composite_regression_timer_report ("difference", ft0, ft1);
 
   gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_DIVIDE, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
   gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_DIVIDE, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
   ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
   ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_divide_rgba8_rgba8_rgba8_sse, &special_ctx);
   if (gimp_composite_regression_compare_contexts ("divide", &generic_ctx, &special_ctx)) {
+    printf("divide failed\n");
     return (1);
   }
   gimp_composite_regression_timer_report ("divide", ft0, ft1);
@@ -134,24 +112,17 @@ gimp_composite_sse_test (int iterations, int n_pixels)
   ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
   ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_dodge_rgba8_rgba8_rgba8_sse, &special_ctx);
   if (gimp_composite_regression_compare_contexts ("dodge", &generic_ctx, &special_ctx)) {
+    printf("dodge failed\n");
     return (1);
   }
   gimp_composite_regression_timer_report ("dodge", ft0, ft1);
 
-  gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_BURN, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
-  gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_BURN, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
-  ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
-  ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_burn_rgba8_rgba8_rgba8_sse, &special_ctx);
-  if (gimp_composite_regression_compare_contexts ("burn", &generic_ctx, &special_ctx)) {
-    return (1);
-  }
-  gimp_composite_regression_timer_report ("burn", ft0, ft1);
-
   gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_GRAIN_EXTRACT, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
   gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_GRAIN_EXTRACT, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
   ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
   ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_grain_extract_rgba8_rgba8_rgba8_sse, &special_ctx);
   if (gimp_composite_regression_compare_contexts ("grain_extract", &generic_ctx, &special_ctx)) {
+    printf("grain_extract failed\n");
     return (1);
   }
   gimp_composite_regression_timer_report ("grain_extract", ft0, ft1);
@@ -161,27 +132,70 @@ gimp_composite_sse_test (int iterations, int n_pixels)
   ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
   ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_grain_merge_rgba8_rgba8_rgba8_sse, &special_ctx);
   if (gimp_composite_regression_compare_contexts ("grain_merge", &generic_ctx, &special_ctx)) {
+    printf("grain_merge failed\n");
     return (1);
   }
   gimp_composite_regression_timer_report ("grain_merge", ft0, ft1);
 
-  gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_SWAP, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
-  gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_SWAP, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
+  gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_LIGHTEN, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
+  gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_LIGHTEN, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
   ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
-  ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_swap_rgba8_rgba8_rgba8_sse, &special_ctx);
-  if (gimp_composite_regression_compare_contexts ("swap", &generic_ctx, &special_ctx)) {
+  ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_lighten_rgba8_rgba8_rgba8_sse, &special_ctx);
+  if (gimp_composite_regression_compare_contexts ("lighten", &generic_ctx, &special_ctx)) {
+    printf("lighten failed\n");
     return (1);
   }
-  gimp_composite_regression_timer_report ("swap", ft0, ft1);
+  gimp_composite_regression_timer_report ("lighten", ft0, ft1);
+
+  gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_MULTIPLY, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
+  gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_MULTIPLY, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
+  ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
+  ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_multiply_rgba8_rgba8_rgba8_sse, &special_ctx);
+  if (gimp_composite_regression_compare_contexts ("multiply", &generic_ctx, &special_ctx)) {
+    printf("multiply failed\n");
+    return (1);
+  }
+  gimp_composite_regression_timer_report ("multiply", ft0, ft1);
 
   gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_SCALE, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
   gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_SCALE, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
   ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
   ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_scale_rgba8_rgba8_rgba8_sse, &special_ctx);
   if (gimp_composite_regression_compare_contexts ("scale", &generic_ctx, &special_ctx)) {
+    printf("scale failed\n");
     return (1);
   }
   gimp_composite_regression_timer_report ("scale", ft0, ft1);
+
+  gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_SCREEN, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
+  gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_SCREEN, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
+  ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
+  ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_screen_rgba8_rgba8_rgba8_sse, &special_ctx);
+  if (gimp_composite_regression_compare_contexts ("screen", &generic_ctx, &special_ctx)) {
+    printf("screen failed\n");
+    return (1);
+  }
+  gimp_composite_regression_timer_report ("screen", ft0, ft1);
+
+  gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_SUBTRACT, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
+  gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_SUBTRACT, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
+  ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
+  ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_subtract_rgba8_rgba8_rgba8_sse, &special_ctx);
+  if (gimp_composite_regression_compare_contexts ("subtract", &generic_ctx, &special_ctx)) {
+    printf("subtract failed\n");
+    return (1);
+  }
+  gimp_composite_regression_timer_report ("subtract", ft0, ft1);
+
+  gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_SWAP, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
+  gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_SWAP, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
+  ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
+  ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_swap_rgba8_rgba8_rgba8_sse, &special_ctx);
+  if (gimp_composite_regression_compare_contexts ("swap", &generic_ctx, &special_ctx)) {
+    printf("swap failed\n");
+    return (1);
+  }
+  gimp_composite_regression_timer_report ("swap", ft0, ft1);
 #endif
   return (0);
 }
@@ -197,7 +211,7 @@ main (int argc, char *argv[])
   putenv ("GIMP_COMPOSITE=0x1");
 
   iterations = 1;
-  n_pixels = 1048577;
+  n_pixels = 163921;
 
   argv++, argc--;
   while (argc >= 2) {
diff --git a/app/composite/gimp-composite-sse.c b/app/composite/gimp-composite-sse.c
index 2633bfed1b..a95eacb0b1 100644
--- a/app/composite/gimp-composite-sse.c
+++ b/app/composite/gimp-composite-sse.c
@@ -1,7 +1,9 @@
-/* The GIMP -- an image manipulation program
+/* -*- mode: c tab-width: 2; c-basic-indent: 2; indent-tabs-mode: nil -*-
+ *
+ * The GIMP -- an image manipulation program
  * Copyright (C) 1995 Spencer Kimball and Peter Mattis
  *
- * -*- mode: c tab-width: 2; c-basic-indent: 2; indent-tabs-mode: nil -*-
+ * 
  *
  * This program is free software; you can redistribute it and/or modify
  * it under the terms of the GNU General Public License as published by
@@ -42,17 +44,15 @@
 
 #include "gimp-composite.h"
 #include "gimp-composite-sse.h"
+#include "gimp-composite-x86.h"
 
 #define pminub(src,dst,tmp)  "pminub " "%%" #src ", %%" #dst
 #define pmaxub(src,dst,tmp)  "pmaxub " "%%" #src ", %%" #dst
 
-
-/*
- * Clobbers eax, ecx edx
- */
 /*
  * Double-word divide.  Adjusted for subsequent unsigned packing
  * (high-order bit of each word is cleared)
+ * Clobbers eax, ecx edx
  */
 #define pdivwX(dividend,divisor,quotient) "movd %%" #dividend ",%%eax; " \
                                           "movd %%" #divisor  ",%%ecx; " \
@@ -137,15 +137,6 @@
                   "\tpaddw     %%"#a",    %%"#b"; " \
                   "\tpsrlw     $8,        %%"#b"\n"
 
-#define mmx_low_bytes_to_words(src,dst,zero) \
-         "\tmovq      %%"#src", %%"#dst"; " \
-         "\tpunpcklbw %%"#zero", %%"#dst"\n"
-
-#define mmx_high_bytes_to_words(src,dst,zero) \
-         "\tmovq      %%"#src", %%"#dst"; " \
-         "\tpunpckhbw %%"#zero", %%"#dst"\n"
-
-
 const static guint32 rgba8_alpha_mask_64[2] = { 0xFF000000, 0xFF000000 };
 const static guint32 rgba8_b1_64[2] =         { 0x01010101, 0x01010101 };
 const static guint32 rgba8_b255_64[2] =       { 0xFFFFFFFF, 0xFFFFFFFF };
@@ -174,155 +165,156 @@ gimp_composite_addition_rgba8_rgba8_rgba8_sse (GimpCompositeContext *_op)
                 : "%mm0");
 
   for (; op.n_pixels >= 2; op.n_pixels -= 2)
-				{
-						asm ("  movq       %0, %%mm2\n"
-											"\tmovq       %1, %%mm3\n"
-											"\tmovq    %%mm2, %%mm4\n"
-											"\tpaddusb %%mm3, %%mm4\n"
-											"\tmovq    %%mm0, %%mm1\n"
-											"\tpandn   %%mm4, %%mm1\n"
-											"\t" pminub(mm3, mm2, mm4) "\n"
-											"\tpand    %%mm0, %%mm2\n"
-											"\tpor     %%mm2, %%mm1\n"
-											"\tmovq    %%mm1, %2\n"
-											: /* empty */
-											: "m" (*op.A), "m" (*op.B), "m" (*op.D)
-											: "0", "1", "2", "%mm0", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5", "%mm6", "%mm7");
-						op.A += 8;
-						op.B += 8;
-						op.D += 8;
-				}
-
+    {
+      asm ("  movq       %1, %%mm2\n"
+           "\tmovq       %2, %%mm3\n"
+           "\tmovq    %%mm2, %%mm4\n"
+           "\tpaddusb %%mm3, %%mm4\n"
+           "\tmovq    %%mm0, %%mm1\n"
+           "\tpandn   %%mm4, %%mm1\n"
+           "\tpminub  %%mm3, %%mm2\n"
+           "\tpand    %%mm0, %%mm2\n"
+           "\tpor     %%mm2, %%mm1\n"
+           "\tmovq    %%mm1, %0\n"
+           : "=m" (*op.D)
+           : "m" (*op.A), "m" (*op.B)
+           : "%mm0", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5", "%mm6", "%mm7");
+      op.A += 8;
+      op.B += 8;
+      op.D += 8;
+    }
+  
   if (op.n_pixels)
-				{
-						asm volatile ("  movd       %0, %%mm2\n"
-																				"\tmovd       %1, %%mm3\n"
-																				"\tmovq    %%mm2, %%mm4\n"
-																				"\tpaddusb %%mm3, %%mm4\n"
-																				"\tmovq    %%mm0, %%mm1\n"
-																				"\tpandn   %%mm4, %%mm1\n"
-																				"\t" pminub(mm3, mm2, mm4) "\n"
-																				"\tpand    %%mm0, %%mm2\n"
-																				"\tpor     %%mm2, %%mm1\n"
-																				"\tmovd    %%mm1, %2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D)
-																				: "0", "1", "2", "%mm0", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5", "%mm6", "%mm7");
-				}
-
+    {
+      asm volatile ("  movd       %1, %%mm2\n"
+                    "\tmovd       %2, %%mm3\n"
+                    "\tmovq    %%mm2, %%mm4\n"
+                    "\tpaddusb %%mm3, %%mm4\n"
+                    "\tmovq    %%mm0, %%mm1\n"
+                    "\tpandn   %%mm4, %%mm1\n"
+                    "\tpminub  %%mm3, %%mm2\n"
+                    "\tpand    %%mm0, %%mm2\n"
+                    "\tpor     %%mm2, %%mm1\n"
+                    "\tmovd    %%mm1, %0\n"
+                    : "=m" (*op.D)
+                    : "m" (*op.A), "m" (*op.B)
+                    : "%mm0", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5", "%mm6", "%mm7");
+    }
+  
   asm("emms");
 }
 
+
 void
 gimp_composite_burn_rgba8_rgba8_rgba8_sse (GimpCompositeContext *_op)
 {
   GimpCompositeContext op = *_op;
-
+  
   for (; op.n_pixels >= 2; op.n_pixels -= 2)
-				{
-						asm ("  movq         %0,%%mm0\n"
-											"\tmovq         %1,%%mm1\n"
-         
-											"\tmovq         %3,%%mm2\n"
-											"\tpsubb     %%mm0,%%mm2\n" /* mm2 = 255 - A */
-											"\tpxor      %%mm4,%%mm4\n"
-											"\tpunpcklbw %%mm2,%%mm4\n" /* mm4 = (255- A) * 256  */
-
-											"\tmovq      %%mm1,%%mm3\n"
-											"\tpxor      %%mm5,%%mm5\n"
-											"\tpunpcklbw %%mm5,%%mm3\n"
-											"\tmovq         %4,%%mm5\n"
-											"\tpaddusw   %%mm3,%%mm5\n" /* mm5 = B + 1 */
-
-											"\t" pdivwqX(mm4,mm5,mm7) "\n"
-
-											"\tmovq         %3,%%mm2\n"
-											"\tpsubb     %%mm0,%%mm2\n" /* mm2 = 255 - A */
-											"\tpxor      %%mm4,%%mm4\n"
-											"\tpunpckhbw %%mm2,%%mm4\n" /* mm4 = (255- A) * 256  */
-
-											"\tmovq      %%mm1,%%mm3\n"
-											"\tpxor      %%mm5,%%mm5\n"
-											"\tpunpckhbw %%mm5,%%mm3\n"
-											"\tmovq         %4,%%mm5\n"
-											"\tpaddusw   %%mm3,%%mm5\n" /* mm5 = B + 1 */
-											"\t" pdivwqX(mm4,mm5,mm6) "\n"
-
-											"\tmovq         %5,%%mm4\n"
-											"\tmovq      %%mm4,%%mm5\n"
-											"\tpsubusw   %%mm6,%%mm4\n"
-											"\tpsubusw   %%mm7,%%mm5\n"
-                  
-											"\tpackuswb  %%mm4,%%mm5\n"
-
-											"\t" pminub(mm0,mm1,mm3) "\n" /* mm1 = min(mm0,mm1) clobber mm3 */
-
-											"\tmovq         %6,%%mm7\n"
-											"\tpand      %%mm7,%%mm1\n" /* mm1 = mm7 & alpha_mask */
-
-											"\tpandn     %%mm5,%%mm7\n" /* mm7 = ~mm7 & mm5 */
-											"\tpor       %%mm1,%%mm7\n" /* mm7 = mm7 | mm1 */
-
-											"\tmovq      %%mm7,%2\n"
-											: /* empty */
-											: "m" (*op.A), "m" (*op.B), "m" (*op.D), "m" (*rgba8_b255_64), "m" (*rgba8_w1_64), "m" (*rgba8_w255_64), "m" (*rgba8_alpha_mask_64)
-											: "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
-						op.A += 8;
-						op.B += 8;
-						op.D += 8;
-				}
-
+    {
+      asm ("  movq         %1,%%mm0\n"
+           "\tmovq         %2,%%mm1\n"
+           
+           "\tmovq         %3,%%mm2\n"
+           "\tpsubb     %%mm0,%%mm2\n" /* mm2 = 255 - A */
+           "\tpxor      %%mm4,%%mm4\n"
+           "\tpunpcklbw %%mm2,%%mm4\n" /* mm4 = (255- A) * 256  */
+           
+           "\tmovq      %%mm1,%%mm3\n"
+           "\tpxor      %%mm5,%%mm5\n"
+           "\tpunpcklbw %%mm5,%%mm3\n"
+           "\tmovq         %4,%%mm5\n"
+           "\tpaddusw   %%mm3,%%mm5\n" /* mm5 = B + 1 */
+           
+           "\t" pdivwqX(mm4,mm5,mm7) "\n"
+           
+           "\tmovq         %3,%%mm2\n"
+           "\tpsubb     %%mm0,%%mm2\n" /* mm2 = 255 - A */
+           "\tpxor      %%mm4,%%mm4\n"
+           "\tpunpckhbw %%mm2,%%mm4\n" /* mm4 = (255- A) * 256  */
+           
+           "\tmovq      %%mm1,%%mm3\n"
+           "\tpxor      %%mm5,%%mm5\n"
+           "\tpunpckhbw %%mm5,%%mm3\n"
+           "\tmovq         %4,%%mm5\n"
+           "\tpaddusw   %%mm3,%%mm5\n" /* mm5 = B + 1 */
+           "\t" pdivwqX(mm4,mm5,mm6) "\n"
+           
+           "\tmovq         %5,%%mm4\n"
+           "\tmovq      %%mm4,%%mm5\n"
+           "\tpsubusw   %%mm6,%%mm4\n"
+           "\tpsubusw   %%mm7,%%mm5\n"
+           
+           "\tpackuswb  %%mm4,%%mm5\n"
+           
+           "\tpminub    %%mm0,%%mm1\n" /* mm1 = min(mm0,mm1) clobber mm3 */
+           
+           "\tmovq         %6,%%mm7\n"
+           "\tpand      %%mm7,%%mm1\n" /* mm1 = mm7 & alpha_mask */
+           
+           "\tpandn     %%mm5,%%mm7\n" /* mm7 = ~mm7 & mm5 */
+           "\tpor       %%mm1,%%mm7\n" /* mm7 = mm7 | mm1 */
+           
+           "\tmovq      %%mm7,%0\n"
+           : "=m" (*op.D)
+           : "m" (*op.A), "m" (*op.B), "m" (*rgba8_b255_64), "m" (*rgba8_w1_64), "m" (*rgba8_w255_64), "m" (*rgba8_alpha_mask_64)
+           : "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5", "%mm6", "%mm7");
+      op.A += 8;
+      op.B += 8;
+      op.D += 8;
+    }
+  
   if (op.n_pixels)
-				{
-						asm volatile ("  movd         %0,%%mm0\n"
-																				"\tmovd         %1,%%mm1\n"
-
-																				"\tmovq         %3,%%mm2\n"
-																				"\tpsubb     %%mm0,%%mm2\n" /* mm2 = 255 - A */
-																				"\tpxor      %%mm4,%%mm4\n"
-																				"\tpunpcklbw %%mm2,%%mm4\n" /* mm4 = (255- A) * 256  */
-
-																				"\tmovq      %%mm1,%%mm3\n"
-																				"\tpxor      %%mm5,%%mm5\n"
-																				"\tpunpcklbw %%mm5,%%mm3\n"
-																				"\tmovq         %4,%%mm5\n"
-																				"\tpaddusw   %%mm3,%%mm5\n" /* mm5 = B + 1 */
-
-																				"\t" pdivwqX(mm4,mm5,mm7) "\n"
-
-																				"\tmovq         %3,%%mm2\n"
-																				"\tpsubb     %%mm0,%%mm2\n" /* mm2 = 255 - A */
-																				"\tpxor      %%mm4,%%mm4\n"
-																				"\tpunpckhbw %%mm2,%%mm4\n" /* mm4 = (255- A) * 256  */
-
-																				"\tmovq      %%mm1,%%mm3\n"
-																				"\tpxor      %%mm5,%%mm5\n"
-																				"\tpunpckhbw %%mm5,%%mm3\n"
-																				"\tmovq         %4,%%mm5\n"
-																				"\tpaddusw   %%mm3,%%mm5\n" /* mm5 = B + 1 */
-																				"\t" pdivwqX(mm4,mm5,mm6) "\n"
-
-																				"\tmovq         %5,%%mm4\n"
-																				"\tmovq      %%mm4,%%mm5\n"
-																				"\tpsubusw   %%mm6,%%mm4\n"
-																				"\tpsubusw   %%mm7,%%mm5\n"
-                  
-																				"\tpackuswb  %%mm4,%%mm5\n"
-
-																				"\t" pminub(mm0,mm1,mm3) "\n" /* mm1 = min(mm0,mm1) clobber mm3 */
-
-																				"\tmovq         %6,%%mm7\n"
-																				"\tpand      %%mm7,%%mm1\n" /* mm1 = mm7 & alpha_mask */
-
-																				"\tpandn     %%mm5,%%mm7\n" /* mm7 = ~mm7 & mm5 */
-																				"\tpor       %%mm1,%%mm7\n" /* mm7 = mm7 | mm1 */
-
-																				"\tmovd      %%mm7,%2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D), "m" (*rgba8_b255_64), "m" (*rgba8_w1_64), "m" (*rgba8_w255_64), "m" (*rgba8_alpha_mask_64)
-																				: "0", "1", "2", "%mm0", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5", "%mm6", "%mm7");
-				}
-
+    {
+      asm volatile ("  movd         %1,%%mm0\n"
+                    "\tmovd         %2,%%mm1\n"
+                    
+                    "\tmovq         %3,%%mm2\n"
+                    "\tpsubb     %%mm0,%%mm2\n" /* mm2 = 255 - A */
+                    "\tpxor      %%mm4,%%mm4\n"
+                    "\tpunpcklbw %%mm2,%%mm4\n" /* mm4 = (255- A) * 256  */
+                    
+                    "\tmovq      %%mm1,%%mm3\n"
+                    "\tpxor      %%mm5,%%mm5\n"
+                    "\tpunpcklbw %%mm5,%%mm3\n"
+                    "\tmovq         %4,%%mm5\n"
+                    "\tpaddusw   %%mm3,%%mm5\n" /* mm5 = B + 1 */
+                    
+                    "\t" pdivwqX(mm4,mm5,mm7) "\n"
+                    
+                    "\tmovq         %3,%%mm2\n"
+                    "\tpsubb     %%mm0,%%mm2\n" /* mm2 = 255 - A */
+                    "\tpxor      %%mm4,%%mm4\n"
+                    "\tpunpckhbw %%mm2,%%mm4\n" /* mm4 = (255- A) * 256  */
+                    
+                    "\tmovq      %%mm1,%%mm3\n"
+                    "\tpxor      %%mm5,%%mm5\n"
+                    "\tpunpckhbw %%mm5,%%mm3\n"
+                    "\tmovq         %4,%%mm5\n"
+                    "\tpaddusw   %%mm3,%%mm5\n" /* mm5 = B + 1 */
+                    "\t" pdivwqX(mm4,mm5,mm6) "\n"
+                    
+                    "\tmovq         %5,%%mm4\n"
+                    "\tmovq      %%mm4,%%mm5\n"
+                    "\tpsubusw   %%mm6,%%mm4\n"
+                    "\tpsubusw   %%mm7,%%mm5\n"
+                    
+                    "\tpackuswb  %%mm4,%%mm5\n"
+                    
+                    "\tpminub    %%mm0,%%mm1\n" /* mm1 = min(mm0,mm1) clobber mm3 */
+                    
+                    "\tmovq         %6,%%mm7\n"
+                    "\tpand      %%mm7,%%mm1\n" /* mm1 = mm7 & alpha_mask */
+                    
+                    "\tpandn     %%mm5,%%mm7\n" /* mm7 = ~mm7 & mm5 */
+                    "\tpor       %%mm1,%%mm7\n" /* mm7 = mm7 | mm1 */
+                    
+                    "\tmovd      %%mm7,%0\n"
+                    : "=m" (*op.D)
+                    : "m" (*op.A), "m" (*op.B), "m" (*rgba8_b255_64), "m" (*rgba8_w1_64), "m" (*rgba8_w255_64), "m" (*rgba8_alpha_mask_64)
+                    : "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5", "%mm6", "%mm7");
+    }
+  
   asm("emms");
 }
 
@@ -331,32 +323,31 @@ void
 gimp_composite_darken_rgba8_rgba8_rgba8_sse (GimpCompositeContext *_op)
 {
   GimpCompositeContext op = *_op;
-
+  
   for (; op.n_pixels >= 2; op.n_pixels -= 2)
-				{
-						asm volatile ("  movq       %0, %%mm2\n"
-																				"\tmovq       %1, %%mm3\n"
-																				"\t" pminub(mm3, mm2, mm4) "\n"
-																				"\tmovq    %%mm2, %2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D)
-																				: "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
-						op.A += 8;
-						op.B += 8;
-						op.D += 8;
-				}
-
+    {
+      asm volatile ("  movq       %1, %%mm2\n"
+                    "\tpminub     %2, %%mm2\n"
+                    "\tmovq    %%mm2, %0\n"
+                    : "=m" (*op.D)
+                    : "m" (*op.A), "m" (*op.B)
+                    : "%mm1", "%mm2", "%mm3", "%mm4");
+      op.A += 8;
+      op.B += 8;
+      op.D += 8;
+    }
+  
   if (op.n_pixels)
-				{
-						asm volatile ("  movd       %0, %%mm2\n"
-																				"\tmovd       %1, %%mm3\n"
-																				"\t" pminub(mm3, mm2, mm4) "\n"
-																				"\tmovd    %%mm2, %2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D)
-																				: "0", "1", "2", "%mm2", "%mm3", "%mm4");
-				}
-        
+    {
+      asm volatile ("  movd       %1, %%mm2\n"
+                    "\tmovd       %2, %%mm3\n"
+                    "\tpminub  %%mm3, %%mm2\n"
+                    "\tmovd    %%mm2, %0\n"
+                    : "=m" (*op.D)
+                    : "m" (*op.A), "m" (*op.B)
+                    : "%mm2", "%mm3", "%mm4");
+    }
+  
   asm("emms");
 }
 
@@ -364,52 +355,52 @@ void
 gimp_composite_difference_rgba8_rgba8_rgba8_sse (GimpCompositeContext *_op)
 {
   GimpCompositeContext op = *_op;
-
-		asm volatile ("movq    %0,%%mm0"     :  : "m" (*rgba8_alpha_mask_64) : "%mm0");
-
+  
+  asm volatile ("movq    %0,%%mm0"     :  : "m" (*rgba8_alpha_mask_64) : "%mm0");
+  
   for (; op.n_pixels >= 2; op.n_pixels -= 2)
-				{
-						asm volatile ("  movq       %0, %%mm2\n"
-																				"\tmovq       %1, %%mm3\n"
-																				"\tmovq    %%mm2, %%mm4\n"
-																				"\tmovq    %%mm3, %%mm5\n"
-																				"\tpsubusb %%mm3, %%mm4\n"
-																				"\tpsubusb %%mm2, %%mm5\n"
-																				"\tpaddb   %%mm5, %%mm4\n"
-																				"\tmovq    %%mm0, %%mm1\n"
-																				"\tpandn   %%mm4, %%mm1\n"
-																				"\tpminub  %%mm3, %%mm2\n"
-																				"\tpand    %%mm0, %%mm2\n"
-																				"\tpor     %%mm2, %%mm1\n"
-																				"\tmovq    %%mm1, %2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D)
-																				: "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
-						op.A += 8;
-						op.B += 8;
-						op.D += 8;
-				}
+    {
+      asm volatile ("  movq       %1, %%mm2\n"
+                    "\tmovq       %2, %%mm3\n"
+                    "\tmovq    %%mm2, %%mm4\n"
+                    "\tmovq    %%mm3, %%mm5\n"
+                    "\tpsubusb %%mm3, %%mm4\n"
+                    "\tpsubusb %%mm2, %%mm5\n"
+                    "\tpaddb   %%mm5, %%mm4\n"
+                    "\tmovq    %%mm0, %%mm1\n"
+                    "\tpandn   %%mm4, %%mm1\n"
+                    "\tpminub  %%mm3, %%mm2\n"
+                    "\tpand    %%mm0, %%mm2\n"
+                    "\tpor     %%mm2, %%mm1\n"
+                    "\tmovq    %%mm1, %0\n"
+                    : "=m" (*op.D)
+                    : "m" (*op.A), "m" (*op.B)
+                    : "%mm1", "%mm2", "%mm3", "%mm4");
+      op.A += 8;
+      op.B += 8;
+      op.D += 8;
+    }
   
   if (op.n_pixels)
-				{
-						asm volatile ("  movd       %0, %%mm2\n"
-																				"\tmovd       %1, %%mm3\n"
-																				"\tmovq    %%mm2, %%mm4\n"
-																				"\tmovq    %%mm3, %%mm5\n"
-																				"\tpsubusb %%mm3, %%mm4\n"
-																				"\tpsubusb %%mm2, %%mm5\n"
-																				"\tpaddb   %%mm5, %%mm4\n"
-																				"\tmovq    %%mm0, %%mm1\n"
-																				"\tpandn   %%mm4, %%mm1\n"
-																				"\tpminub  %%mm3, %%mm2\n"
-																				"\tpand    %%mm0, %%mm2\n"
-																				"\tpor     %%mm2, %%mm1\n"
-																				"\tmovd    %%mm1, %2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D)
-																				: "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
-				}
-
+    {
+      asm volatile ("  movd       %1, %%mm2\n"
+                    "\tmovd       %2, %%mm3\n"
+                    "\tmovq    %%mm2, %%mm4\n"
+                    "\tmovq    %%mm3, %%mm5\n"
+                    "\tpsubusb %%mm3, %%mm4\n"
+                    "\tpsubusb %%mm2, %%mm5\n"
+                    "\tpaddb   %%mm5, %%mm4\n"
+                    "\tmovq    %%mm0, %%mm1\n"
+                    "\tpandn   %%mm4, %%mm1\n"
+                    "\tpminub  %%mm3, %%mm2\n"
+                    "\tpand    %%mm0, %%mm2\n"
+                    "\tpor     %%mm2, %%mm1\n"
+                    "\tmovd    %%mm1, %0\n"
+                    : "=m" (*op.D)
+                    : "m" (*op.A), "m" (*op.B), "m" (*op.D)
+                    : "%mm1", "%mm2", "%mm3", "%mm4");
+    }
+  
   asm("emms");
 }
 
@@ -417,99 +408,99 @@ void
 gimp_composite_divide_rgba8_rgba8_rgba8_sse (GimpCompositeContext *_op)
 {
   GimpCompositeContext op = *_op;
-
-  asm volatile ("movq    %0, %%mm0\n"
-																"\tmovq    %1, %%mm7\n"
-																:
-																: "m" (*rgba8_alpha_mask_64), "m" (*rgba8_w1_64)
-																: "%mm0", "%mm7");
+  
+  asm volatile ("  movq    %0, %%mm0\n"
+                "\tmovq    %1, %%mm7\n"
+                :
+                : "m" (*rgba8_alpha_mask_64), "m" (*rgba8_w1_64)
+                : "%mm0", "%mm7");
   
   for (; op.n_pixels >= 2; op.n_pixels -= 2)
-				{
-						asm volatile ("  movq         %0,%%mm0\n"
-																				"\tmovq         %1,%%mm1\n"
-																				"\tpxor      %%mm2,%%mm2\n"
-																				"\tpunpcklbw %%mm0,%%mm2\n" /* mm2 = A*256 */
-
-																				"\tmovq      %%mm1,%%mm3\n"
-																				"\tpxor      %%mm5,%%mm5\n"
-																				"\tpunpcklbw %%mm5,%%mm3\n"
-																				"\tpaddw     %%mm7,%%mm3\n" /* mm3 = B+1 */
-
-																				"\t" pdivwuqX(mm2,mm3,mm5) "\n" /* mm5 = (A*256)/(B+1) */
-
-																				"\tpxor      %%mm2,%%mm2\n"
-																				"\tpunpckhbw %%mm0,%%mm2\n" /* mm2 = A*256 */
-
-																				"\tmovq      %%mm1,%%mm3\n"
-																				"\tpxor      %%mm6,%%mm6\n"
-																				"\tpunpckhbw %%mm6,%%mm3\n"
-																				"\tpaddw     %%mm7,%%mm3\n" /* mm3 = B+1 */
-
-																				"\t" pdivwuqX(mm2,mm3,mm4) "\n" /* mm4 = (A*256)/(B+1) */
-                  
-																				"\tpackuswb  %%mm4,%%mm5\n" /* expects mm4 and mm5 to be signed values */
-
-																				"\t" pminub(mm0,mm1,mm3) "\n"
-																				"\tmovq         %3,%%mm3\n"
-																				"\tmovq      %%mm3,%%mm2\n"
-
-																				"\tpandn     %%mm5,%%mm3\n"
-
-																				"\tpand      %%mm2,%%mm1\n"
-																				"\tpor       %%mm1,%%mm3\n"
-
-																				"\tmovq      %%mm3,%2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D), "m" (*rgba8_alpha_mask_64)
-																				: "%eax", "%ecx", "%edx", "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
-						op.A += 8;
-						op.B += 8;
-						op.D += 8;
-				}
-
+    {
+      asm volatile ("  movq         %0,%%mm0\n"
+                    "\tmovq         %1,%%mm1\n"
+                    "\tpxor      %%mm2,%%mm2\n"
+                    "\tpunpcklbw %%mm0,%%mm2\n" /* mm2 = A*256 */
+                    
+                    "\tmovq      %%mm1,%%mm3\n"
+                    "\tpxor      %%mm5,%%mm5\n"
+                    "\tpunpcklbw %%mm5,%%mm3\n"
+                    "\tpaddw     %%mm7,%%mm3\n" /* mm3 = B+1 */
+                    
+                    "\t" pdivwuqX(mm2,mm3,mm5) "\n" /* mm5 = (A*256)/(B+1) */
+                    
+                    "\tpxor      %%mm2,%%mm2\n"
+                    "\tpunpckhbw %%mm0,%%mm2\n" /* mm2 = A*256 */
+                    
+                    "\tmovq      %%mm1,%%mm3\n"
+                    "\tpxor      %%mm6,%%mm6\n"
+                    "\tpunpckhbw %%mm6,%%mm3\n"
+                    "\tpaddw     %%mm7,%%mm3\n" /* mm3 = B+1 */
+                    
+                    "\t" pdivwuqX(mm2,mm3,mm4) "\n" /* mm4 = (A*256)/(B+1) */
+                    
+                    "\tpackuswb  %%mm4,%%mm5\n" /* expects mm4 and mm5 to be signed values */
+                    
+                    "\tpminub    %%mm0,%%mm1\n"
+                    "\tmovq         %3,%%mm3\n"
+                    "\tmovq      %%mm3,%%mm2\n"
+                    
+                    "\tpandn     %%mm5,%%mm3\n"
+                    
+                    "\tpand      %%mm2,%%mm1\n"
+                    "\tpor       %%mm1,%%mm3\n"
+                    
+                    "\tmovq      %%mm3,%2\n"
+                    : /* empty */
+                    : "m" (*op.A), "m" (*op.B), "m" (*op.D), "m" (*rgba8_alpha_mask_64)
+                    : "%eax", "%ecx", "%edx", "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
+      op.A += 8;
+      op.B += 8;
+      op.D += 8;
+    }
+  
   if (op.n_pixels)
-				{
-						asm volatile ("  movd         %0,%%mm0\n"
-																				"\tmovd         %1,%%mm1\n"
-
-																				"\tpxor      %%mm2,%%mm2\n"
-																				"\tpunpcklbw %%mm0,%%mm2\n" /* mm2 = A*256 */
-
-																				"\tmovq      %%mm1,%%mm3\n"
-																				"\tpxor      %%mm5,%%mm5\n"
-																				"\tpunpcklbw %%mm5,%%mm3\n"
-																				"\tpaddw     %%mm7,%%mm3\n" /* mm3 = B+1 */
-
-																				"\t" pdivwuqX(mm2,mm3,mm5) "\n" /* mm5 = (A*256)/(B+1) */
-
-																				"\tpxor      %%mm2,%%mm2\n"
-																				"\tpunpckhbw %%mm0,%%mm2\n" /* mm2 = A*256 */
-
-																				"\tmovq      %%mm1,%%mm3\n"
-																				"\tpxor      %%mm6,%%mm6\n"
-																				"\tpunpckhbw %%mm6,%%mm3\n"
-																				"\tpaddw     %%mm7,%%mm3\n" /* mm3 = B+1 */
-
-																				"\t" pdivwuqX(mm2,mm3,mm4) "\n" /* mm4 = (A*256)/(B+1) */
-                  
-																				"\tpackuswb  %%mm4,%%mm5\n" /* expects mm4 and mm5 to be signed values */
-
-																				"\t" pminub(mm0,mm1,mm3) "\n"
-																				"\tmovq         %3,%%mm3\n"
-																				"\tmovq      %%mm3,%%mm2\n"
-
-																				"\tpandn     %%mm5,%%mm3\n"
-
-																				"\tpand      %%mm2,%%mm1\n"
-																				"\tpor       %%mm1,%%mm3\n"
-
-																				"\tmovd      %%mm3,%2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D), "m" (*rgba8_alpha_mask_64)
-																				: "%eax", "%ecx", "%edx", "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
-				}
-
+    {
+      asm volatile ("  movd         %0,%%mm0\n"
+                    "\tmovd         %1,%%mm1\n"
+                    
+                    "\tpxor      %%mm2,%%mm2\n"
+                    "\tpunpcklbw %%mm0,%%mm2\n" /* mm2 = A*256 */
+                    
+                    "\tmovq      %%mm1,%%mm3\n"
+                    "\tpxor      %%mm5,%%mm5\n"
+                    "\tpunpcklbw %%mm5,%%mm3\n"
+                    "\tpaddw     %%mm7,%%mm3\n" /* mm3 = B+1 */
+                    
+                    "\t" pdivwuqX(mm2,mm3,mm5) "\n" /* mm5 = (A*256)/(B+1) */
+                    
+                    "\tpxor      %%mm2,%%mm2\n"
+                    "\tpunpckhbw %%mm0,%%mm2\n" /* mm2 = A*256 */
+                    
+                    "\tmovq      %%mm1,%%mm3\n"
+                    "\tpxor      %%mm6,%%mm6\n"
+                    "\tpunpckhbw %%mm6,%%mm3\n"
+                    "\tpaddw     %%mm7,%%mm3\n" /* mm3 = B+1 */
+                    
+                    "\t" pdivwuqX(mm2,mm3,mm4) "\n" /* mm4 = (A*256)/(B+1) */
+                    
+                    "\tpackuswb  %%mm4,%%mm5\n" /* expects mm4 and mm5 to be signed values */
+                    
+                    "\tpminub    %%mm0,%%mm1\n"
+                    "\tmovq         %3,%%mm3\n"
+                    "\tmovq      %%mm3,%%mm2\n"
+                    
+                    "\tpandn     %%mm5,%%mm3\n"
+                    
+                    "\tpand      %%mm2,%%mm1\n"
+                    "\tpor       %%mm1,%%mm3\n"
+                    
+                    "\tmovd      %%mm3,%2\n"
+                    : /* empty */
+                    : "m" (*op.A), "m" (*op.B), "m" (*op.D), "m" (*rgba8_alpha_mask_64)
+                    : "%eax", "%ecx", "%edx", "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
+    }
+  
   asm("emms");
 }
 
@@ -517,165 +508,170 @@ void
 gimp_composite_dodge_rgba8_rgba8_rgba8_sse (GimpCompositeContext *_op)
 {
   GimpCompositeContext op = *_op;
-
+  
   for (; op.n_pixels >= 2; op.n_pixels -= 2)
-				{
-						asm volatile ("  movq         %0,%%mm0\n"
-																				"\tmovq         %1,%%mm1\n"
-																				"\tmovq      %%mm1,%%mm3\n"
-																				"\tpxor      %%mm2,%%mm2\n"
-																				"\tpunpcklbw %%mm2,%%mm3\n"
-																				"\tpunpcklbw %%mm0,%%mm2\n"
-
-																				"\tmovq         %3,%%mm4\n"
-																				"\tpsubw     %%mm3,%%mm4\n"
-
-																				"\t" pdivwuqX(mm2,mm4,mm5) "\n"
-
-																				"\tmovq      %%mm1,%%mm3\n"
-																				"\tpxor      %%mm2,%%mm2\n"
-																				"\tpunpckhbw %%mm2,%%mm3\n"
-																				"\tpunpckhbw %%mm0,%%mm2\n"
-
-																				"\tmovq         %3,%%mm4\n"
-																				"\tpsubw     %%mm3,%%mm4\n"
-
-																				"\t" pdivwuqX(mm2,mm4,mm6) "\n"
-
-																				"\tpackuswb  %%mm6,%%mm5\n"
-
-																				"\tmovq         %4,%%mm6\n"
-																				"\tmovq      %%mm1,%%mm7\n"
-																				"\t" pminub(mm0,mm7,mm2) "\n"
-																				"\tpand      %%mm6,%%mm7\n"
-																				"\tpandn     %%mm5,%%mm6\n"
-
-																				"\tpor       %%mm6,%%mm7\n"
-
-																				"\tmovq      %%mm7,%2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D), "m" (*rgba8_w256_64), "m" (*rgba8_alpha_mask_64)
-																				: "0", "1", "2", "%mm0", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5", "%mm6", "%mm7");
-						op.A += 8;
-						op.B += 8;
-						op.D += 8;
-				}
-
+    {
+      asm volatile ("  movq         %0,%%mm0\n"
+                    "\tmovq         %1,%%mm1\n"
+                    "\tmovq      %%mm1,%%mm3\n"
+                    "\tpxor      %%mm2,%%mm2\n"
+                    "\tpunpcklbw %%mm2,%%mm3\n"
+                    "\tpunpcklbw %%mm0,%%mm2\n"
+                    
+                    "\tmovq         %3,%%mm4\n"
+                    "\tpsubw     %%mm3,%%mm4\n"
+                    
+                    "\t" pdivwuqX(mm2,mm4,mm5) "\n"
+                    
+                    "\tmovq      %%mm1,%%mm3\n"
+                    "\tpxor      %%mm2,%%mm2\n"
+                    "\tpunpckhbw %%mm2,%%mm3\n"
+                    "\tpunpckhbw %%mm0,%%mm2\n"
+                    
+                    "\tmovq         %3,%%mm4\n"
+                    "\tpsubw     %%mm3,%%mm4\n"
+                    
+                    "\t" pdivwuqX(mm2,mm4,mm6) "\n"
+                    
+                    "\tpackuswb  %%mm6,%%mm5\n"
+                    
+                    "\tmovq         %4,%%mm6\n"
+                    "\tmovq      %%mm1,%%mm7\n"
+                    "\t" pminub(mm0,mm7,mm2) "\n"
+                    "\tpand      %%mm6,%%mm7\n"
+                    "\tpandn     %%mm5,%%mm6\n"
+                    
+                    "\tpor       %%mm6,%%mm7\n"
+                    
+                    "\tmovq      %%mm7,%2\n"
+                    : /* empty */
+                    : "m" (*op.A), "m" (*op.B), "m" (*op.D), "m" (*rgba8_w256_64), "m" (*rgba8_alpha_mask_64)
+                    : "0", "1", "2", "%eax", "%ecx", "%edx", "%mm0", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5", "%mm6", "%mm7");
+      op.A += 8;
+      op.B += 8;
+      op.D += 8;
+    }
+  
   if (op.n_pixels)
-				{
-						asm volatile ("  movd         %0,%%mm0\n"
-																				"\tmovq         %1,%%mm1\n"
-																				"\tmovq      %%mm1,%%mm3\n"
-																				"\tpxor      %%mm2,%%mm2\n"
-																				"\tpunpcklbw %%mm2,%%mm3\n"
-																				"\tpunpcklbw %%mm0,%%mm2\n"
-
-																				"\tmovq         %3,%%mm4\n"
-																				"\tpsubw     %%mm3,%%mm4\n"
-
-																				"\t" pdivwuqX(mm2,mm4,mm5) "\n"
-
-																				"\tmovq      %%mm1,%%mm3\n"
-																				"\tpxor      %%mm2,%%mm2\n"
-																				"\tpunpckhbw %%mm2,%%mm3\n"
-																				"\tpunpckhbw %%mm0,%%mm2\n"
-
-																				"\tmovq         %3,%%mm4\n"
-																				"\tpsubw     %%mm3,%%mm4\n"
-
-																				"\t" pdivwuqX(mm2,mm4,mm6) "\n"
-
-																				"\tpackuswb  %%mm6,%%mm5\n"
-
-																				"\tmovq         %4,%%mm6\n"
-																				"\tmovq      %%mm1,%%mm7\n"
-																				"\t" pminub(mm0,mm7,mm2) "\n"
-																				"\tpand      %%mm6,%%mm7\n"
-																				"\tpandn     %%mm5,%%mm6\n"
-
-																				"\tpor       %%mm6,%%mm7\n"
-
-																				"\tmovd      %%mm7,%2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D), "m" (*rgba8_w256_64), "m" (*rgba8_alpha_mask_64)
-																				: "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
-				}
-
+    {
+      asm volatile ("  movd         %0,%%mm0\n"
+                    "\tmovq         %1,%%mm1\n"
+                    "\tmovq      %%mm1,%%mm3\n"
+                    "\tpxor      %%mm2,%%mm2\n"
+                    "\tpunpcklbw %%mm2,%%mm3\n"
+                    "\tpunpcklbw %%mm0,%%mm2\n"
+                    
+                    "\tmovq         %3,%%mm4\n"
+                    "\tpsubw     %%mm3,%%mm4\n"
+                    
+                    "\t" pdivwuqX(mm2,mm4,mm5) "\n"
+                    
+                    "\tmovq      %%mm1,%%mm3\n"
+                    "\tpxor      %%mm2,%%mm2\n"
+                    "\tpunpckhbw %%mm2,%%mm3\n"
+                    "\tpunpckhbw %%mm0,%%mm2\n"
+                    
+                    "\tmovq         %3,%%mm4\n"
+                    "\tpsubw     %%mm3,%%mm4\n"
+                    
+                    "\t" pdivwuqX(mm2,mm4,mm6) "\n"
+                    
+                    "\tpackuswb  %%mm6,%%mm5\n"
+                    
+                    "\tmovq         %4,%%mm6\n"
+                    "\tmovq      %%mm1,%%mm7\n"
+                    "\tpminub    %%mm0,%%mm7\n"
+                    "\tpand      %%mm6,%%mm7\n"
+                    "\tpandn     %%mm5,%%mm6\n"
+                    
+                    "\tpor       %%mm6,%%mm7\n"
+                    
+                    "\tmovd      %%mm7,%2\n"
+                    : /* empty */
+                    : "m" (*op.A), "m" (*op.B), "m" (*op.D), "m" (*rgba8_w256_64), "m" (*rgba8_alpha_mask_64)
+                    : "0", "1", "2", "%eax", "%ecx", "%edx", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
+    }
+  
   asm("emms");
 }
 
 void
 gimp_composite_grain_extract_rgba8_rgba8_rgba8_sse (GimpCompositeContext *_op)
 {
-		GimpCompositeContext op = *_op;
-
-		asm volatile ("movq    %0,%%mm0"     :  : "m" (*rgba8_alpha_mask_64) : "%mm0");
-		asm volatile ("pxor    %%mm6,%%mm6"  :  :                         : "%mm6");
-  asm volatile ("movq    %0,%%mm7"     :  : "m" (*rgba8_w128_64)       : "%mm7");
+  GimpCompositeContext op = *_op;
+  
+  asm volatile ("  movq       %0,%%mm0\n"
+                "\tpxor    %%mm6,%%mm6\n"
+                "\tmovq       %1,%%mm7\n"
+                : /* empty */
+                : "m" (*rgba8_alpha_mask_64), "m" (*rgba8_w128_64)
+                : "%mm0", "%mm6", "%mm7");
 
+  
   for (; op.n_pixels >= 2; op.n_pixels -= 2)
-				{
-						asm volatile ("  movq         %0,%%mm2\n"
-																				"\tmovq         %1,%%mm3\n"
-																				mmx_low_bytes_to_words(mm2,mm4,mm6)
-																				mmx_low_bytes_to_words(mm3,mm5,mm6)
-																				"\tpsubw     %%mm5,%%mm4\n"
-																				"\tpaddw     %%mm7,%%mm4\n"
-																				"\tmovq      %%mm4,%%mm1\n"
-
-																				mmx_high_bytes_to_words(mm2,mm4,mm6)
-																				mmx_high_bytes_to_words(mm3,mm5,mm6)
-
-																				"\tpsubw     %%mm5,%%mm4\n"
-																				"\tpaddw     %%mm7,%%mm4\n"
-
-																				"\tpackuswb  %%mm4,%%mm1\n"
-																				"\tmovq      %%mm1,%%mm4\n"
-
-																				"\tmovq      %%mm0,%%mm1\n"
-																				"\tpandn     %%mm4,%%mm1\n"
-											
-																				"\t" pminub(mm3,mm2,mm4) "\n"
-																				"\tpand      %%mm0,%%mm2\n"
-
-																				"\tpor       %%mm2,%%mm1\n"
-																				"\tmovq      %%mm1,%2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D)
-																				: "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
-						op.A += 8;
-						op.B += 8;
-						op.D += 8;
-				}
-
+    {
+      asm volatile ("  movq         %1,%%mm2\n"
+                    "\tmovq         %2,%%mm3\n"
+                    mmx_low_bytes_to_words(mm2,mm4,mm6)
+                    mmx_low_bytes_to_words(mm3,mm5,mm6)
+                    "\tpsubw     %%mm5,%%mm4\n"
+                    "\tpaddw     %%mm7,%%mm4\n"
+                    "\tmovq      %%mm4,%%mm1\n"
+                    
+                    mmx_high_bytes_to_words(mm2,mm4,mm6)
+                    mmx_high_bytes_to_words(mm3,mm5,mm6)
+                    
+                    "\tpsubw     %%mm5,%%mm4\n"
+                    "\tpaddw     %%mm7,%%mm4\n"
+                    
+                    "\tpackuswb  %%mm4,%%mm1\n"
+                    "\tmovq      %%mm1,%%mm4\n"
+                    
+                    "\tmovq      %%mm0,%%mm1\n"
+                    "\tpandn     %%mm4,%%mm1\n"
+                    
+                    "\tpminub    %%mm3,%%mm2\n"
+                    "\tpand      %%mm0,%%mm2\n"
+                    
+                    "\tpor       %%mm2,%%mm1\n"
+                    "\tmovq      %%mm1,%0\n"
+                    : "+m" (*op.D)
+                    : "m" (*op.A), "m" (*op.B)
+                    : "%mm1", "%mm2", "%mm3", "%mm4");
+      op.A += 8;
+      op.B += 8;
+      op.D += 8;
+    }
+  
   if (op.n_pixels)
-				{
-						asm volatile ("  movd         %0, %%mm2\n"
-																				"\tmovd         %1, %%mm3\n"
-
-																				mmx_low_bytes_to_words(mm2,mm4,mm6)
-																				mmx_low_bytes_to_words(mm3,mm5,mm6)
-
-																				"\tpsubw     %%mm5, %%mm4\n"
-																				"\tpaddw     %%mm7, %%mm4\n"
-																				"\tmovq      %%mm4, %%mm1\n"
-
-																				"\tpackuswb  %%mm6, %%mm1\n"
-
-																				"\tmovq      %%mm1, %%mm4\n"
-
-																				"\tmovq      %%mm0, %%mm1; pandn     %%mm4, %%mm1\n"
-
-																				"\t" pminub(mm3,mm2,mm4) "\n"
-																				"\tpand      %%mm0, %%mm2\n"
-
-																				"\tpor       %%mm2, %%mm1\n"
-																				"\tmovd      %%mm1, %2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D)
-																				: "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
-  }
-
+    {
+      asm volatile ("  movd         %1, %%mm2\n"
+                    "\tmovd         %2, %%mm3\n"
+                    
+                    mmx_low_bytes_to_words(mm2,mm4,mm6)
+                    mmx_low_bytes_to_words(mm3,mm5,mm6)
+                    
+                    "\tpsubw     %%mm5, %%mm4\n"
+                    "\tpaddw     %%mm7, %%mm4\n"
+                    "\tmovq      %%mm4, %%mm1\n"
+                    
+                    "\tpackuswb  %%mm6, %%mm1\n"
+                    
+                    "\tmovq      %%mm1, %%mm4\n"
+                    
+                    "\tmovq      %%mm0, %%mm1\n"
+                    "\tpandn     %%mm4, %%mm1\n"
+                    
+                    "\tpminub    %%mm3, %%mm2\n"
+                    "\tpand      %%mm0, %%mm2\n"
+                    
+                    "\tpor       %%mm2, %%mm1\n"
+                    "\tmovd      %%mm1, %0\n"
+                    : "+m" (*op.D)
+                    : "m" (*op.A), "m" (*op.B)
+                    : "%mm1", "%mm2", "%mm3", "%mm4");
+    }
+  
   asm("emms");
 }
 
@@ -683,73 +679,74 @@ void
 gimp_composite_grain_merge_rgba8_rgba8_rgba8_sse (GimpCompositeContext *_op)
 {
   GimpCompositeContext op = *_op;
-
+  
   asm volatile ("movq    %0, %%mm0\n"
                 "pxor    %%mm6, %%mm6\n"
                 "movq    %1, %%mm7\n"
                 : /* empty */
                 : "m" (*rgba8_alpha_mask_64), "m" (*rgba8_w128_64)
                 : "%mm0", "%mm6", "%mm7");
-
+  
   for (; op.n_pixels >= 2; op.n_pixels -= 2)
-				{
-						asm volatile ("  movq         %0, %%mm2\n"
-																				"\tmovq         %1, %%mm3\n"
-
-																				mmx_low_bytes_to_words(mm2,mm4,mm6)
-																				mmx_low_bytes_to_words(mm3,mm5,mm6)
-																				"\tpaddw     %%mm5, %%mm4\n"
-																				"\tpsubw     %%mm7, %%mm4\n"
-
-																				mmx_high_bytes_to_words(mm2,mm1,mm6)
-																				mmx_high_bytes_to_words(mm3,mm5,mm6)
-																				"\tpaddw     %%mm5, %%mm1\n"
-																				"\tpsubw     %%mm7, %%mm1\n"
-
-																				"\tpackuswb  %%mm1, %%mm4\n"
-         
-																				"\t" pminub(mm3,mm2,mm5) "\n"
-																				"\tpand      %%mm0, %%mm2\n"
-
-																				"\tmovq      %%mm0, %%mm1\n"
-																				"\tpandn     %%mm4, %%mm1\n"
-																				"\tpor       %%mm2, %%mm1\n"
-																				"\tmovq      %%mm1, %2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D)
-																				: "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
-						op.A += 8;
-						op.B += 8;
-						op.D += 8;
-  }
-
+    {
+      asm volatile ("  movq         %1, %%mm2\n"
+                    "\tmovq         %2, %%mm3\n"
+                    
+                    mmx_low_bytes_to_words(mm2,mm4,mm6)
+                    mmx_low_bytes_to_words(mm3,mm5,mm6)
+                    "\tpaddw     %%mm5, %%mm4\n"
+                    "\tpsubw     %%mm7, %%mm4\n"
+                    
+                    mmx_high_bytes_to_words(mm2,mm1,mm6)
+                    mmx_high_bytes_to_words(mm3,mm5,mm6)
+                    "\tpaddw     %%mm5, %%mm1\n"
+                    "\tpsubw     %%mm7, %%mm1\n"
+                    
+                    "\tpackuswb  %%mm1, %%mm4\n"
+                    
+                    "\t" pminub(mm3,mm2,mm5) "\n"
+                    "\tpand      %%mm0, %%mm2\n"
+                    
+                    "\tmovq      %%mm0, %%mm1\n"
+                    "\tpandn     %%mm4, %%mm1\n"
+                    "\tpor       %%mm2, %%mm1\n"
+                    "\tmovq      %%mm1, %0\n"
+                    : "+m" (*op.D)
+                    : "m" (*op.A), "m" (*op.B)
+                    : "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
+      op.A += 8;
+      op.B += 8;
+      op.D += 8;
+    }
+  
   if (op.n_pixels)
-				{
-						asm volatile ("  movd         %0, %%mm2\n"
-																				"\tmovd         %1, %%mm3\n"
-
-																				mmx_low_bytes_to_words(mm2,mm4,mm6)
-																				mmx_low_bytes_to_words(mm3,mm5,mm6)
-																				
-																				"\tpaddw     %%mm5, %%mm4\n"
-																				"\tpsubw     %%mm7, %%mm4\n"
-																				"\tmovq      %%mm4, %%mm1\n"
-																				"\tpackuswb  %%mm6, %%mm1\n"
-																				
-																				"\tmovq      %%mm1, %%mm4\n"
-																				
-																				"\tmovq      %%mm0, %%mm1; pandn     %%mm4, %%mm1\n"
-																				
-																				"\t" pminub(mm3,mm2,mm4) "\n"
-																				"\tpand      %%mm0, %%mm2\n"
-																				
-																				"\tpor       %%mm2, %%mm1\n"
-																				"\tmovd      %%mm1, %2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D)
-																				: "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
-  }
-
+    {
+      asm volatile ("  movd         %1, %%mm2\n"
+                    "\tmovd         %2, %%mm3\n"
+                    
+                    mmx_low_bytes_to_words(mm2,mm4,mm6)
+                    mmx_low_bytes_to_words(mm3,mm5,mm6)
+                    
+                    "\tpaddw     %%mm5, %%mm4\n"
+                    "\tpsubw     %%mm7, %%mm4\n"
+                    "\tmovq      %%mm4, %%mm1\n"
+                    "\tpackuswb  %%mm6, %%mm1\n"
+                    
+                    "\tmovq      %%mm1, %%mm4\n"
+                    
+                    "\tmovq      %%mm0, %%mm1\n"
+                    "\tpandn     %%mm4, %%mm1\n"
+                    
+                    "\tpminub    %%mm3, %%mm2\n"
+                    "\tpand      %%mm0, %%mm2\n"
+                    
+                    "\tpor       %%mm2, %%mm1\n"
+                    "\tmovd      %%mm1, %0\n"
+                    : "+m" (*op.D)
+                    : "m" (*op.A), "m" (*op.B)
+                    : "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
+    }
+  
   asm("emms");
 }
 
@@ -757,49 +754,49 @@ void
 gimp_composite_lighten_rgba8_rgba8_rgba8_sse (GimpCompositeContext *_op)
 {
   GimpCompositeContext op = *_op;
-
+  
   asm volatile ("movq    %0,%%mm0"     :  : "m" (*rgba8_alpha_mask_64) : "%mm0");
-
+  
   for (; op.n_pixels >= 2; op.n_pixels -= 2)
-				{
-    asm volatile ("  movq       %0, %%mm2\n"
-                  "\tmovq       %1, %%mm3\n"
-                  "\tmovq    %%mm2, %%mm4\n"
-                  "\t" pmaxub(mm3,mm4,mm5) "\n"
-                  "\tmovq    %%mm0, %%mm1\n"
-                  "\tpandn   %%mm4, %%mm1\n"
-                  "\t" pminub(mm2,mm3,mm4) "\n"
-                  "\tpand    %%mm0, %%mm3\n"
-                  "\tpor     %%mm3, %%mm1\n"
-                  "\tmovq    %%mm1, %2\n"
-                  : /* empty */
-                  : "m" (*op.A), "m" (*op.B), "m" (*op.D)
-                  : "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
-						op.A += 8;
-						op.B += 8;
-						op.D += 8;
-  }
-
+    {
+      asm volatile ("  movq       %1, %%mm2\n"
+                    "\tmovq       %2, %%mm3\n"
+                    "\tmovq    %%mm2, %%mm4\n"
+                    "\tpmaxub  %%mm3, %%mm4\n"
+                    "\tmovq    %%mm0, %%mm1\n"
+                    "\tpandn   %%mm4, %%mm1\n"
+                    "\tpminub  %%mm2, %%mm3\n"
+                    "\tpand    %%mm0, %%mm3\n"
+                    "\tpor     %%mm3, %%mm1\n"
+                    "\tmovq    %%mm1, %0\n"
+                    : "=m" (*op.D)
+                    : "m" (*op.A), "m" (*op.B)
+                    : "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
+      op.A += 8;
+      op.B += 8;
+      op.D += 8;
+    }
+  
   if (op.n_pixels)
-				{
-    asm volatile ("  movd       %0, %%mm2\n"
-                  "\tmovd       %1, %%mm3\n"
-                  "\tmovq    %%mm2, %%mm4\n"
-                  "\t" pmaxub(mm3,mm4,mm5) "\n"
-
-                  "\tmovq    %%mm0, %%mm1\n"
-                  "\tpandn   %%mm4, %%mm1\n"
-
-                  "\t" pminub(mm2,mm3,mm4) "\n"
-
-                  "\tpand    %%mm0, %%mm3\n"
-                  "\tpor     %%mm3, %%mm1\n"
-                  "\tmovd    %%mm1, %2\n"
-                  : /* empty */
-                  : "m" (*op.A), "m" (*op.B), "m" (*op.D)
-                  : "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
-  }
-
+    {
+      asm volatile ("  movd       %1, %%mm2\n"
+                    "\tmovd       %2, %%mm3\n"
+                    "\tmovq    %%mm2, %%mm4\n"
+                    "\tpmaxub  %%mm3, %%mm4\n"
+                    
+                    "\tmovq    %%mm0, %%mm1\n"
+                    "\tpandn   %%mm4, %%mm1\n"
+                    
+                    "\tpminub  %%mm2,%%mm3\n"
+                    
+                    "\tpand    %%mm0, %%mm3\n"
+                    "\tpor     %%mm3, %%mm1\n"
+                    "\tmovd    %%mm1, %0\n"
+                    : "=m" (*op.D)
+                    : "m" (*op.A), "m" (*op.B)
+                    : "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
+    }
+  
   asm("emms");
 }
 
@@ -807,66 +804,66 @@ void
 gimp_composite_multiply_rgba8_rgba8_rgba8_sse (GimpCompositeContext *_op)
 {
   GimpCompositeContext op = *_op;
-
+  
   asm volatile ("movq    %0,%%mm0"     :  : "m" (*rgba8_alpha_mask_64) : "%mm0");
   asm volatile ("movq    %0,%%mm7"     :  : "m" (*rgba8_w128_64) : "%mm7");
   asm volatile ("pxor    %%mm6,%%mm6"  :  :  : "%mm6");
-
+  
   for (; op.n_pixels >= 2; op.n_pixels -= 2)
-				{
-						asm volatile ("  movq         %0, %%mm2\n"
-																				"\tmovq         %1, %%mm3\n"
-
-																				mmx_low_bytes_to_words(mm2,mm1,mm6)
-																				mmx_low_bytes_to_words(mm3,mm5,mm6)
-																				mmx_int_mult(mm5,mm1,mm7)
-         
-																				mmx_high_bytes_to_words(mm2,mm4,mm6)
-																				mmx_high_bytes_to_words(mm3,mm5,mm6)
-																				mmx_int_mult(mm5,mm4,mm7)
-         
-																				"\tpackuswb  %%mm4, %%mm1\n"
-         
-																				"\tmovq      %%mm0, %%mm4\n"
-																				"\tpandn     %%mm1, %%mm4\n"
-																				"\tmovq      %%mm4, %%mm1\n"
-																				"\t" pminub(mm3,mm2,mm4) "\n"
-																				"\tpand      %%mm0, %%mm2\n"
-																				"\tpor       %%mm2, %%mm1\n"
-																				
-																				"\tmovq      %%mm1, %2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D)
-																				: "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
-						op.A += 8;
-						op.B += 8;
-						op.D += 8;
-  }
-
+    {
+      asm volatile ("  movq         %0, %%mm2\n"
+                    "\tmovq         %1, %%mm3\n"
+                    
+                    mmx_low_bytes_to_words(mm2,mm1,mm6)
+                    mmx_low_bytes_to_words(mm3,mm5,mm6)
+                    mmx_int_mult(mm5,mm1,mm7)
+                    
+                    mmx_high_bytes_to_words(mm2,mm4,mm6)
+                    mmx_high_bytes_to_words(mm3,mm5,mm6)
+                    mmx_int_mult(mm5,mm4,mm7)
+                    
+                    "\tpackuswb  %%mm4, %%mm1\n"
+                    
+                    "\tmovq      %%mm0, %%mm4\n"
+                    "\tpandn     %%mm1, %%mm4\n"
+                    "\tmovq      %%mm4, %%mm1\n"
+                    "\t" pminub(mm3,mm2,mm4) "\n"
+                    "\tpand      %%mm0, %%mm2\n"
+                    "\tpor       %%mm2, %%mm1\n"
+                    
+                    "\tmovq      %%mm1, %2\n"
+                    : /* empty */
+                    : "m" (*op.A), "m" (*op.B), "m" (*op.D)
+                    : "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
+      op.A += 8;
+      op.B += 8;
+      op.D += 8;
+    }
+  
   if (op.n_pixels)
-				{
-    asm volatile ("  movd         %0, %%mm2\n"
-                  "\tmovd         %1, %%mm3\n"
-
-                  mmx_low_bytes_to_words(mm2,mm1,mm6)
-                  mmx_low_bytes_to_words(mm3,mm5,mm6)
-                  pmulwX(mm5,mm1,mm7)
-
-                  "\tpackuswb  %%mm6, %%mm1\n"
-
-                  "\tmovq      %%mm0, %%mm4\n"
-                  "\tpandn     %%mm1, %%mm4\n"
-                  "\tmovq      %%mm4, %%mm1\n"
-                  "\t" pminub(mm3,mm2,mm4) "\n"
-                  "\tpand      %%mm0, %%mm2\n"
-                  "\tpor       %%mm2, %%mm1\n"
-                  
-                  "\tmovd      %%mm1, %2\n"
-                  : /* empty */
-                  : "m" (*op.A), "m" (*op.B), "m" (*op.D)
-                  : "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
-  }
-
+    {
+      asm volatile ("  movd         %0, %%mm2\n"
+                    "\tmovd         %1, %%mm3\n"
+                    
+                    mmx_low_bytes_to_words(mm2,mm1,mm6)
+                    mmx_low_bytes_to_words(mm3,mm5,mm6)
+                    pmulwX(mm5,mm1,mm7)
+                    
+                    "\tpackuswb  %%mm6, %%mm1\n"
+                    
+                    "\tmovq      %%mm0, %%mm4\n"
+                    "\tpandn     %%mm1, %%mm4\n"
+                    "\tmovq      %%mm4, %%mm1\n"
+                    "\t" pminub(mm3,mm2,mm4) "\n"
+                    "\tpand      %%mm0, %%mm2\n"
+                    "\tpor       %%mm2, %%mm1\n"
+                    
+                    "\tmovd      %%mm1, %2\n"
+                    : /* empty */
+                    : "m" (*op.A), "m" (*op.B), "m" (*op.D)
+                    : "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
+    }
+  
   asm("emms");
 }
 
@@ -882,7 +879,7 @@ sse_op_overlay(void)
                 "\tmovq         %0,%%mm6\n"  /* mm6 = words of value 2 */
                 "\tpmullw    %%mm5,%%mm6\n" /* mm6 = 2 * low bytes of B */
                 mmx_int_mult(mm6,mm4,mm7)    /* mm4 = INT_MULT(mm6, mm4) */
-
+                
                 /* high bytes */
                 mmx_high_bytes_to_words(mm3,mm5,mm0)
                 "\tpcmpeqb   %%mm1,%%mm1\n"
@@ -891,9 +888,9 @@ sse_op_overlay(void)
                 "\tmovq         %0,%%mm6\n"  /* mm6 = words of value 2 */
                 "\tpmullw    %%mm5,%%mm6\n" /* mm6 = 2 * high bytes of B */
                 mmx_int_mult(mm6,mm1,mm7)    /* mm1 = INT_MULT(mm6, mm1) */
-
+                
                 "\tpackuswb  %%mm1,%%mm4\n"  /* mm4 = intermediate value */
-
+                
                 mmx_low_bytes_to_words(mm4,mm5,mm0)
                 mmx_low_bytes_to_words(mm2,mm6,mm0)
                 "\tpaddw     %%mm6,%%mm5\n"
@@ -909,12 +906,12 @@ sse_op_overlay(void)
                 "\tmovq         %1,%%mm0\n"
                 "\tmovq      %%mm0,%%mm1\n"
                 "\tpandn     %%mm5,%%mm1\n"
-
+                
                 "\t" pminub(mm2,mm3,mm4) "\n"
                 "\tpand      %%mm0,%%mm3\n"
-
+                
                 "\tpor       %%mm3,%%mm1\n"
-
+                
                 : /* empty */
                 : "m" (*rgba8_w2_64), "m" (*rgba8_alpha_mask_64)
                 );
@@ -924,119 +921,116 @@ void
 xxxgimp_composite_overlay_rgba8_rgba8_rgba8_sse (GimpCompositeContext *_op)
 {
   GimpCompositeContext op = *_op;
-
+  
   asm volatile ("pxor    %%mm0,%%mm0\n"
                 "movq       %0,%%mm7"
                 : /* empty */
-																: "m" (*rgba8_w128_64) : "%mm0");
-
+                : "m" (*rgba8_w128_64) : "%mm0");
+  
   for (; op.n_pixels >= 2; op.n_pixels -= 2)
-				{
-						asm volatile ("  movq         %0,%%mm2\n"
-																				"\tmovq         %1,%%mm3\n"
+    {
+      asm volatile ("  movq         %0,%%mm2\n"
+                    "\tmovq         %1,%%mm3\n"
+                    
+                    /* low bytes */
+                    mmx_low_bytes_to_words(mm3,mm5,mm0)
+                    "\tpcmpeqb   %%mm4,%%mm4\n"
+                    "\tpsubb     %%mm2,%%mm4\n" /* mm4 = 255 - A */
+                    "\tpunpcklbw %%mm0,%%mm4\n" /* mm4 = (low bytes as word) mm4 */
+                    "\tmovq         %3,%%mm6\n"  /* mm6 = words of value 2 */
+                    "\tpmullw    %%mm5,%%mm6\n" /* mm6 = 2 * low bytes of B */
+                    mmx_int_mult(mm6,mm4,mm7)    /* mm4 = INT_MULT(mm6, mm4) */
+                    
+                    /* high bytes */
+                    mmx_high_bytes_to_words(mm3,mm5,mm0)
+                    "\tpcmpeqb   %%mm1,%%mm1\n"
+                    "\tpsubb     %%mm2,%%mm1\n" /* mm1 = 255 - A */
+                    "\tpunpckhbw %%mm0,%%mm1\n" /* mm1 = (high bytes as word) mm1 */
+                    "\tmovq         %3,%%mm6\n"  /* mm6 = words of value 2 */
+                    "\tpmullw    %%mm5,%%mm6\n" /* mm6 = 2 * high bytes of B */
+                    mmx_int_mult(mm6,mm1,mm7)    /* mm1 = INT_MULT(mm6, mm1) */
+                    "\tpackuswb  %%mm1,%%mm4\n"  /* mm4 = intermediate value */
+                    mmx_low_bytes_to_words(mm4,mm5,mm0)
+                    mmx_low_bytes_to_words(mm2,mm6,mm0)
+                    "\tpaddw     %%mm6,%%mm5\n"
+                    mmx_int_mult(mm6,mm5,mm7)   /* mm5 = INT_MULT(mm6, mm5) low bytes */
+                    mmx_high_bytes_to_words(mm4,mm1,mm0)
+                    mmx_high_bytes_to_words(mm2,mm6,mm0)
+                    "\tpaddw     %%mm6,%%mm1\n"
+                    mmx_int_mult(mm6,mm1,mm7)   /* mm1 = INT_MULT(mm6, mm1) high bytes */
 
-																				/* low bytes */
-																				mmx_low_bytes_to_words(mm3,mm5,mm0)
-																				"\tpcmpeqb   %%mm4,%%mm4\n"
-																				"\tpsubb     %%mm2,%%mm4\n" /* mm4 = 255 - A */
-																				"\tpunpcklbw %%mm0,%%mm4\n" /* mm4 = (low bytes as word) mm4 */
-																				"\tmovq         %3,%%mm6\n"  /* mm6 = words of value 2 */
-																				"\tpmullw    %%mm5,%%mm6\n" /* mm6 = 2 * low bytes of B */
-																				mmx_int_mult(mm6,mm4,mm7)    /* mm4 = INT_MULT(mm6, mm4) */
-																				
-																				/* high bytes */
-																				mmx_high_bytes_to_words(mm3,mm5,mm0)
-																				"\tpcmpeqb   %%mm1,%%mm1\n"
-																				"\tpsubb     %%mm2,%%mm1\n" /* mm1 = 255 - A */
-																				"\tpunpckhbw %%mm0,%%mm1\n" /* mm1 = (high bytes as word) mm1 */
-																				"\tmovq         %3,%%mm6\n"  /* mm6 = words of value 2 */
-																				"\tpmullw    %%mm5,%%mm6\n" /* mm6 = 2 * high bytes of B */
-																				mmx_int_mult(mm6,mm1,mm7)    /* mm1 = INT_MULT(mm6, mm1) */
-																				
-																				"\tpackuswb  %%mm1,%%mm4\n"  /* mm4 = intermediate value */
-																				
-																				mmx_low_bytes_to_words(mm4,mm5,mm0)
-																				mmx_low_bytes_to_words(mm2,mm6,mm0)
-																				"\tpaddw     %%mm6,%%mm5\n"
-																				mmx_int_mult(mm6,mm5,mm7)   /* mm5 = INT_MULT(mm6, mm5) low bytes */
-																				
-																				mmx_high_bytes_to_words(mm4,mm1,mm0)
-																				mmx_high_bytes_to_words(mm2,mm6,mm0)
-																				"\tpaddw     %%mm6,%%mm1\n"
-																				mmx_int_mult(mm6,mm1,mm7)   /* mm1 = INT_MULT(mm6, mm1) high bytes */
-																				
-																				"\tpackuswb  %%mm1,%%mm5\n"
-																				
-																				"\tmovq         %4,%%mm0\n"
-																				"\tmovq      %%mm0,%%mm1\n"
-																				"\tpandn     %%mm5,%%mm1\n"
-																				
-																				"\t" pminub(mm2,mm3,mm4) "\n"
-																				"\tpand      %%mm0,%%mm3\n"
-																				
-																				"\tpor       %%mm3,%%mm1\n"
-																				
-																				"\tmovq      %%mm1,%2\n"
-																				: "+m" (*op.A), "+m" (*op.B), "+m" (*op.D)
-																				: "m" (*rgba8_w2_64), "m" (*rgba8_alpha_mask_64)
-																				: "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
-						op.A += 8;
-						op.B += 8;
-						op.D += 8;
-				}
+                    "\tpackuswb  %%mm1,%%mm5\n"
+
+                    "\tmovq         %4,%%mm0\n"
+                    "\tmovq      %%mm0,%%mm1\n"
+                    "\tpandn     %%mm5,%%mm1\n"
+
+                    "\t" pminub(mm2,mm3,mm4) "\n"
+                    "\tpand      %%mm0,%%mm3\n"
+
+                    "\tpor       %%mm3,%%mm1\n"
+
+                    "\tmovq      %%mm1,%2\n"
+                    : "+m" (*op.A), "+m" (*op.B), "+m" (*op.D)
+                    : "m" (*rgba8_w2_64), "m" (*rgba8_alpha_mask_64)
+                    : "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
+      op.A += 8;
+      op.B += 8;
+      op.D += 8;
+    }
 
   if (op.n_pixels)
-				{
-						asm volatile ("  movd         %0,%%mm2\n"
-																				"\tmovd         %1,%%mm3\n"
+    {
+      asm volatile ("  movd         %0,%%mm2\n"
+                    "\tmovd         %1,%%mm3\n"
+                    
+                    /* low bytes */
+                    mmx_low_bytes_to_words(mm3,mm5,mm0)
+                    "\tpcmpeqb   %%mm4,%%mm4\n"
+                    "\tpsubb     %%mm2,%%mm4\n" /* mm4 = 255 - A */
+                    "\tpunpcklbw %%mm0,%%mm4\n" /* mm4 = (low bytes as word) mm4 */
+                    "\tmovq         %3,%%mm6\n"  /* mm6 = words of value 2 */
+                    "\tpmullw    %%mm5,%%mm6\n" /* mm6 = 2 * low bytes of B */
+                    mmx_int_mult(mm6,mm4,mm7)    /* mm4 = INT_MULT(mm6, mm4) */
+                    
+                    /* high bytes */
+                    mmx_high_bytes_to_words(mm3,mm5,mm0)
+                    "\tpcmpeqb   %%mm1,%%mm1\n"
+                    "\tpsubb     %%mm2,%%mm1\n" /* mm1 = 255 - A */
+                    "\tpunpckhbw %%mm0,%%mm1\n" /* mm1 = (high bytes as word) mm1 */
+                    "\tmovq         %3,%%mm6\n"  /* mm6 = words of value 2 */
+                    "\tpmullw    %%mm5,%%mm6\n" /* mm6 = 2 * high bytes of B */
+                    mmx_int_mult(mm6,mm1,mm7)    /* mm1 = INT_MULT(mm6, mm1) */
+                    
+                    "\tpackuswb  %%mm1,%%mm4\n"  /* mm4 = intermediate value */
+                    
+                    mmx_low_bytes_to_words(mm4,mm5,mm0)
+                    mmx_low_bytes_to_words(mm2,mm6,mm0)
+                    "\tpaddw     %%mm6,%%mm5\n"
+                    mmx_int_mult(mm6,mm5,mm7)   /* mm5 = INT_MULT(mm6, mm5) low bytes */
+                    
+                    mmx_high_bytes_to_words(mm4,mm1,mm0)
+                    mmx_high_bytes_to_words(mm2,mm6,mm0)
+                    "\tpaddw     %%mm6,%%mm1\n"
+                    mmx_int_mult(mm6,mm1,mm7)   /* mm1 = INT_MULT(mm6, mm1) high bytes */
+                    
+                    "\tpackuswb  %%mm1,%%mm5\n"
+                    
+                    "\tmovq         %4,%%mm0\n"
+                    "\tmovq      %%mm0,%%mm1\n"
+                    "\tpandn     %%mm5,%%mm1\n"
+                    
+                    "\t" pminub(mm2,mm3,mm4) "\n"
+                    "\tpand      %%mm0,%%mm3\n"
+                    
+                    "\tpor       %%mm3,%%mm1\n"
+                    
+                    "\tmovd      %%mm1,%2\n"
+                    : /* empty */
+                    : "m" (*op.A), "m" (*op.B), "m" (*op.D), "m" (*rgba8_w2_64), "m" (*rgba8_alpha_mask_64)
+                    : "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
+    }
 
-																				/* low bytes */
-																				mmx_low_bytes_to_words(mm3,mm5,mm0)
-																				"\tpcmpeqb   %%mm4,%%mm4\n"
-																				"\tpsubb     %%mm2,%%mm4\n" /* mm4 = 255 - A */
-																				"\tpunpcklbw %%mm0,%%mm4\n" /* mm4 = (low bytes as word) mm4 */
-																				"\tmovq         %3,%%mm6\n"  /* mm6 = words of value 2 */
-																				"\tpmullw    %%mm5,%%mm6\n" /* mm6 = 2 * low bytes of B */
-																				mmx_int_mult(mm6,mm4,mm7)    /* mm4 = INT_MULT(mm6, mm4) */
-																				
-																				/* high bytes */
-																				mmx_high_bytes_to_words(mm3,mm5,mm0)
-																				"\tpcmpeqb   %%mm1,%%mm1\n"
-																				"\tpsubb     %%mm2,%%mm1\n" /* mm1 = 255 - A */
-																				"\tpunpckhbw %%mm0,%%mm1\n" /* mm1 = (high bytes as word) mm1 */
-																				"\tmovq         %3,%%mm6\n"  /* mm6 = words of value 2 */
-																				"\tpmullw    %%mm5,%%mm6\n" /* mm6 = 2 * high bytes of B */
-																				mmx_int_mult(mm6,mm1,mm7)    /* mm1 = INT_MULT(mm6, mm1) */
-																				
-																				"\tpackuswb  %%mm1,%%mm4\n"  /* mm4 = intermediate value */
-																				
-																				mmx_low_bytes_to_words(mm4,mm5,mm0)
-																				mmx_low_bytes_to_words(mm2,mm6,mm0)
-																				"\tpaddw     %%mm6,%%mm5\n"
-																				mmx_int_mult(mm6,mm5,mm7)   /* mm5 = INT_MULT(mm6, mm5) low bytes */
-																				
-																				mmx_high_bytes_to_words(mm4,mm1,mm0)
-																				mmx_high_bytes_to_words(mm2,mm6,mm0)
-																				"\tpaddw     %%mm6,%%mm1\n"
-																				mmx_int_mult(mm6,mm1,mm7)   /* mm1 = INT_MULT(mm6, mm1) high bytes */
-																				
-																				"\tpackuswb  %%mm1,%%mm5\n"
-																				
-																				"\tmovq         %4,%%mm0\n"
-																				"\tmovq      %%mm0,%%mm1\n"
-																				"\tpandn     %%mm5,%%mm1\n"
-																				
-																				"\t" pminub(mm2,mm3,mm4) "\n"
-																				"\tpand      %%mm0,%%mm3\n"
-																				
-																				"\tpor       %%mm3,%%mm1\n"
-																				
-																				"\tmovd      %%mm1,%2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D), "m" (*rgba8_w2_64), "m" (*rgba8_alpha_mask_64)
-																				: "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
-				}
-		
   asm("emms");
 }
 
@@ -1058,48 +1052,48 @@ gimp_composite_scale_rgba8_rgba8_rgba8_sse (GimpCompositeContext *_op)
                 "\tmovq     %1,%%mm7\n"
                 : /* empty */
                 : "m" (op.scale.scale), "m" (*rgba8_w128_64)
-                : "%eax", "%mm0", "%mm5", "%mm6", "%mm7");
+                : "%eax", "%ebx", "%mm0", "%mm3", "%mm5", "%mm6", "%mm7");
 
   for (; op.n_pixels >= 2; op.n_pixels -= 2)
-				{
-    asm volatile ("movq           %0,%%mm2\n"
-                  "\tmovq      %%mm2,%%mm1\n"
-                  "\tpunpcklbw %%mm0,%%mm1\n"
-                  "\tmovq      %%mm3,%%mm5\n"
-
-                  "\t" pmulwX(mm5,mm1,mm7) "\n"
-
-                  "\tmovq      %%mm2,%%mm4\n"
-                  "\tpunpckhbw %%mm0,%%mm4\n"
-                  "\tmovq      %%mm3,%%mm5\n"
-
-                  "\t" pmulwX(mm5,mm4,mm7) "\n"
-
-                  "\tpackuswb  %%mm4,%%mm1\n"
-
-                  "\tmovq      %%mm1,%1\n"
-                  : /* empty */
-                  : "m" (*op.A), "m" (*op.D)
-                  : "0", "1", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5", "%mm6", "%mm7");
-						op.A += 8;
-						op.D += 8;
-  }
-
+    {
+      asm volatile ("movq           %1,%%mm2\n"
+                    "\tmovq      %%mm2,%%mm1\n"
+                    "\tpunpcklbw %%mm0,%%mm1\n"
+                    "\tmovq      %%mm3,%%mm5\n"
+                    
+                    "\t" pmulwX(mm5,mm1,mm7) "\n"
+                    
+                    "\tmovq      %%mm2,%%mm4\n"
+                    "\tpunpckhbw %%mm0,%%mm4\n"
+                    "\tmovq      %%mm3,%%mm5\n"
+                    
+                    "\t" pmulwX(mm5,mm4,mm7) "\n"
+                    
+                    "\tpackuswb  %%mm4,%%mm1\n"
+                    
+                    "\tmovq      %%mm1,%0\n"
+                    : "+m" (*op.D)
+                    : "m" (*op.A)
+                    : "%mm1", "%mm2", "%mm4", "%mm5", "%mm7");
+      op.A += 8;
+      op.D += 8;
+    }
+  
   if (op.n_pixels)
-				{
-						asm volatile ("movd           %0,%%mm2\n"
-																				"\tmovq      %%mm2,%%mm1\n"
-																				"\tpunpcklbw %%mm0,%%mm1\n"
-																				"\tmovq      %%mm3,%%mm5\n"
+    {
+      asm volatile ("  movd         %1,%%mm2\n"
+                    "\tmovq      %%mm2,%%mm1\n"
+                    "\tpunpcklbw %%mm0,%%mm1\n"
+                    "\tmovq      %%mm3,%%mm5\n"
 
-																				"\t" pmulwX(mm5,mm1,mm7) "\n"
+                    "\t" pmulwX(mm5,mm1,mm7) "\n"
 
-																				"\tpackuswb  %%mm0,%%mm1\n"
-																				"\tmovd      %%mm1,%1\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.D)
-																				: "0", "1", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5", "%mm6", "%mm7");
-				}
+                    "\tpackuswb  %%mm0,%%mm1\n"
+                    "\tmovd      %%mm1,%0\n"
+                    : "+m" (*op.D)
+                    : "m" (*op.A)
+                    : "%mm1", "%mm2", "%mm4", "%mm5", "%mm6", "%mm7");
+    }
 
   asm("emms");
 }
@@ -1108,118 +1102,118 @@ void
 gimp_composite_screen_rgba8_rgba8_rgba8_sse (GimpCompositeContext *_op)
 {
   GimpCompositeContext op = *_op;
-
+  
   asm volatile ("movq    %0,%%mm0"     :  : "m" (*rgba8_alpha_mask_64) : "%mm0");
   asm volatile ("movq    %0,%%mm7"     :  : "m" (*rgba8_w128_64)  : "%mm7");
   asm volatile ("pxor    %mm6, %mm6");
-
+  
   for (; op.n_pixels >= 2; op.n_pixels -= 2)
-				{
-						asm volatile ("  movq         %0,%%mm2\n"
-																				"\tmovq         %1,%%mm3\n"
-																				
-																				"\tpcmpeqb   %%mm4,%%mm4\n"
-																				"\tpsubb     %%mm2,%%mm4\n"
-																				"\tpcmpeqb   %%mm5,%%mm5\n"
-																				"\tpsubb     %%mm3,%%mm5\n"
-
-																				"\tpunpcklbw %%mm6,%%mm4\n"
-																				"\tpunpcklbw %%mm6,%%mm5\n"
-																				"\tpmullw    %%mm4,%%mm5\n"
-																				"\tpaddw     %%mm7,%%mm5\n"
-																				"\tmovq      %%mm5,%%mm1\n"
-																				"\tpsrlw       $ 8,%%mm1\n"
-																				"\tpaddw     %%mm5,%%mm1\n"
-																				"\tpsrlw       $ 8,%%mm1\n"
-																				
-																				"\tpcmpeqb   %%mm4,%%mm4\n"
-																				"\tpsubb     %%mm2,%%mm4\n"
-																				"\tpcmpeqb   %%mm5,%%mm5\n"
-																				"\tpsubb     %%mm3,%%mm5\n"
-																				
-																				"\tpunpckhbw %%mm6,%%mm4\n"
-																				"\tpunpckhbw %%mm6,%%mm5\n"
-																				"\tpmullw    %%mm4,%%mm5\n"
-																				"\tpaddw     %%mm7,%%mm5\n"
-																				"\tmovq      %%mm5,%%mm4\n"
-																				"\tpsrlw       $ 8,%%mm4\n"
-																				"\tpaddw     %%mm5,%%mm4\n"
-																				"\tpsrlw       $ 8,%%mm4\n"
-																				
-																				"\tpackuswb  %%mm4,%%mm1\n"
-																				
-																				"\tpcmpeqb   %%mm4,%%mm4\n"
-																				"\tpsubb     %%mm1,%%mm4\n"
-																				
-																				"\tmovq      %%mm0,%%mm1\n"
-																				"\tpandn     %%mm4,%%mm1\n"
-																				
-																				"\t" pminub(mm2,mm3,mm4) "\n"
-																				"\tpand      %%mm0,%%mm3\n"
-																				
-																				"\tpor       %%mm3,%%mm1\n"
-																				
-																				"\tmovq      %%mm1,%2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D)
-																				: "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
-						op.A += 8;
-						op.B += 8;
-						op.D += 8;
-  }
+    {
+      asm volatile ("  movq         %0,%%mm2\n"
+                    "\tmovq         %1,%%mm3\n"
+                    
+                    "\tpcmpeqb   %%mm4,%%mm4\n"
+                    "\tpsubb     %%mm2,%%mm4\n"
+                    "\tpcmpeqb   %%mm5,%%mm5\n"
+                    "\tpsubb     %%mm3,%%mm5\n"
+                    
+                    "\tpunpcklbw %%mm6,%%mm4\n"
+                    "\tpunpcklbw %%mm6,%%mm5\n"
+                    "\tpmullw    %%mm4,%%mm5\n"
+                    "\tpaddw     %%mm7,%%mm5\n"
+                    "\tmovq      %%mm5,%%mm1\n"
+                    "\tpsrlw       $ 8,%%mm1\n"
+                    "\tpaddw     %%mm5,%%mm1\n"
+                    "\tpsrlw       $ 8,%%mm1\n"
+                    
+                    "\tpcmpeqb   %%mm4,%%mm4\n"
+                    "\tpsubb     %%mm2,%%mm4\n"
+                    "\tpcmpeqb   %%mm5,%%mm5\n"
+                    "\tpsubb     %%mm3,%%mm5\n"
+                    
+                    "\tpunpckhbw %%mm6,%%mm4\n"
+                    "\tpunpckhbw %%mm6,%%mm5\n"
+                    "\tpmullw    %%mm4,%%mm5\n"
+                    "\tpaddw     %%mm7,%%mm5\n"
+                    "\tmovq      %%mm5,%%mm4\n"
+                    "\tpsrlw       $ 8,%%mm4\n"
+                    "\tpaddw     %%mm5,%%mm4\n"
+                    "\tpsrlw       $ 8,%%mm4\n"
+                    
+                    "\tpackuswb  %%mm4,%%mm1\n"
+                    
+                    "\tpcmpeqb   %%mm4,%%mm4\n"
+                    "\tpsubb     %%mm1,%%mm4\n"
+                    
+                    "\tmovq      %%mm0,%%mm1\n"
+                    "\tpandn     %%mm4,%%mm1\n"
+                    
+                    "\t" pminub(mm2,mm3,mm4) "\n"
+                    "\tpand      %%mm0,%%mm3\n"
+                    
+                    "\tpor       %%mm3,%%mm1\n"
+                    
+                    "\tmovq      %%mm1,%2\n"
+                    : /* empty */
+                    : "m" (*op.A), "m" (*op.B), "m" (*op.D)
+                    : "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
+      op.A += 8;
+      op.B += 8;
+      op.D += 8;
+    }
 
   if (op.n_pixels)
-				{
-						asm volatile ("  movd         %0,%%mm2\n"
-																				"\tmovd         %1,%%mm3\n"
-
-																				"\tpcmpeqb   %%mm4,%%mm4\n"
-																				"\tpsubb     %%mm2,%%mm4\n"
-																				"\tpcmpeqb   %%mm5,%%mm5\n"
-																				"\tpsubb     %%mm3,%%mm5\n"
-
-																				"\tpunpcklbw %%mm6,%%mm4\n"
-																				"\tpunpcklbw %%mm6,%%mm5\n"
-																				"\tpmullw    %%mm4,%%mm5\n"
-																				"\tpaddw     %%mm7,%%mm5\n"
-																				"\tmovq      %%mm5,%%mm1\n"
-																				"\tpsrlw       $ 8,%%mm1\n"
-																				"\tpaddw     %%mm5,%%mm1\n"
-																				"\tpsrlw       $ 8,%%mm1\n"
-
-																				"\tpcmpeqb   %%mm4,%%mm4\n"
-																				"\tpsubb     %%mm2,%%mm4\n"
-																				"\tpcmpeqb   %%mm5,%%mm5\n"
-																				"\tpsubb     %%mm3,%%mm5\n"
-
-																				"\tpunpckhbw %%mm6,%%mm4\n"
-																				"\tpunpckhbw %%mm6,%%mm5\n"
-																				"\tpmullw    %%mm4,%%mm5\n"
-																				"\tpaddw     %%mm7,%%mm5\n"
-																				"\tmovq      %%mm5,%%mm4\n"
-																				"\tpsrlw       $ 8,%%mm4\n"
-																				"\tpaddw     %%mm5,%%mm4\n"
-																				"\tpsrlw       $ 8,%%mm4\n"
-
-																				"\tpackuswb  %%mm4,%%mm1\n"
-
-																				"\tpcmpeqb   %%mm4,%%mm4\n"
-																				"\tpsubb     %%mm1,%%mm4\n"
-
-																				"\tmovq      %%mm0,%%mm1\n"
-																				"\tpandn     %%mm4,%%mm1\n"
-
-																				"\t" pminub(mm2,mm3,mm4) "\n"
-																				"\tpand      %%mm0,%%mm3\n"
-
-																				"\tpor       %%mm3,%%mm1\n"
-
-																				"\tmovd      %%mm1,%2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D)
-																				: "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
-				}
+    {
+      asm volatile ("  movd         %0,%%mm2\n"
+                    "\tmovd         %1,%%mm3\n"
 
+                    "\tpcmpeqb   %%mm4,%%mm4\n"
+                    "\tpsubb     %%mm2,%%mm4\n"
+                    "\tpcmpeqb   %%mm5,%%mm5\n"
+                    "\tpsubb     %%mm3,%%mm5\n"
+                    
+                    "\tpunpcklbw %%mm6,%%mm4\n"
+                    "\tpunpcklbw %%mm6,%%mm5\n"
+                    "\tpmullw    %%mm4,%%mm5\n"
+                    "\tpaddw     %%mm7,%%mm5\n"
+                    "\tmovq      %%mm5,%%mm1\n"
+                    "\tpsrlw       $ 8,%%mm1\n"
+                    "\tpaddw     %%mm5,%%mm1\n"
+                    "\tpsrlw       $ 8,%%mm1\n"
+                    
+                    "\tpcmpeqb   %%mm4,%%mm4\n"
+                    "\tpsubb     %%mm2,%%mm4\n"
+                    "\tpcmpeqb   %%mm5,%%mm5\n"
+                    "\tpsubb     %%mm3,%%mm5\n"
+                    
+                    "\tpunpckhbw %%mm6,%%mm4\n"
+                    "\tpunpckhbw %%mm6,%%mm5\n"
+                    "\tpmullw    %%mm4,%%mm5\n"
+                    "\tpaddw     %%mm7,%%mm5\n"
+                    "\tmovq      %%mm5,%%mm4\n"
+                    "\tpsrlw       $ 8,%%mm4\n"
+                    "\tpaddw     %%mm5,%%mm4\n"
+                    "\tpsrlw       $ 8,%%mm4\n"
+                    
+                    "\tpackuswb  %%mm4,%%mm1\n"
+                    
+                    "\tpcmpeqb   %%mm4,%%mm4\n"
+                    "\tpsubb     %%mm1,%%mm4\n"
+                    
+                    "\tmovq      %%mm0,%%mm1\n"
+                    "\tpandn     %%mm4,%%mm1\n"
+                    
+                    "\t" pminub(mm2,mm3,mm4) "\n"
+                    "\tpand      %%mm0,%%mm3\n"
+                    
+                    "\tpor       %%mm3,%%mm1\n"
+                    
+                    "\tmovd      %%mm1,%2\n"
+                    : /* empty */
+                    : "m" (*op.A), "m" (*op.B), "m" (*op.D)
+                    : "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
+    }
+  
   asm("emms");
 }
 
@@ -1228,54 +1222,46 @@ void
 gimp_composite_subtract_rgba8_rgba8_rgba8_sse (GimpCompositeContext *_op)
 {
   GimpCompositeContext op = *_op;
-
+  
   asm volatile ("movq    %0,%%mm0"     :  : "m" (*rgba8_alpha_mask_64) : "%mm0");
-
+  
   for (; op.n_pixels >= 2; op.n_pixels -= 2)
-				{
-						asm volatile ("  movq       %0,%%mm2\n"
-																				"\tmovq       %1,%%mm3\n"
-
-																				"\tmovq    %%mm2,%%mm4\n"
-																				"\tpsubusb %%mm3,%%mm4\n"
-                  
-																				"\tmovq    %%mm0,%%mm1\n"
-																				"\tpandn   %%mm4,%%mm1\n"
-																				
-																				"\t" pminub(mm3,mm2,mm4) "\n"
-																				
-																				"\tpand    %%mm0,%%mm2\n"
-																				"\tpor     %%mm2,%%mm1\n"
-																				"\tmovq    %%mm1,%2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D)
-																				: "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
-						op.A += 8;
-						op.B += 8;
-						op.D += 8;
-  }
-
+    {
+      asm volatile ("  movq       %1,%%mm2\n"
+                    "\tmovq       %2,%%mm3\n"
+                    "\tmovq    %%mm2,%%mm4\n"
+                    "\tpsubusb %%mm3,%%mm4\n"
+                    "\tmovq    %%mm0,%%mm1\n"
+                    "\tpandn   %%mm4,%%mm1\n"
+                    "\tpminub  %%mm3,%%mm2\n"
+                    "\tpand    %%mm0,%%mm2\n"
+                    "\tpor     %%mm2,%%mm1\n"
+                    "\tmovq    %%mm1,%0\n"
+                    : "+m" (*op.D)
+                    : "m" (*op.A), "m" (*op.B)
+                    : "%mm1", "%mm2", "%mm3", "%mm4");
+      op.A += 8;
+      op.B += 8;
+      op.D += 8;
+    }
+  
   if (op.n_pixels)
-				{
-						asm volatile ("  movd       %0,%%mm2\n"
-																				"\tmovd       %1,%%mm3\n"
-
-																				"\tmovq    %%mm2,%%mm4\n"
-																				"\tpsubusb %%mm3,%%mm4\n"
-                  
-																				"\tmovq    %%mm0,%%mm1\n"
-																				"\tpandn   %%mm4,%%mm1\n"
-                  
-																				"\t" pminub(mm3,mm2,mm4) "\n"
-
-																				"\tpand    %%mm0,%%mm2\n"
-																				"\tpor     %%mm2,%%mm1\n"
-																				"\tmovd    %%mm1,%2\n"
-																				: /* empty */
-																				: "m" (*op.A), "m" (*op.B), "m" (*op.D)
-																				: "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
-  }
-
+    {
+      asm volatile ("  movd       %1,%%mm2\n"
+                    "\tmovd       %2,%%mm3\n"
+                    "\tmovq    %%mm2,%%mm4\n"
+                    "\tpsubusb %%mm3,%%mm4\n"
+                    "\tmovq    %%mm0,%%mm1\n"
+                    "\tpandn   %%mm4,%%mm1\n"
+                    "\tpminub  %%mm3,%%mm2\n"
+                    "\tpand    %%mm0,%%mm2\n"
+                    "\tpor     %%mm2,%%mm1\n"
+                    "\tmovd    %%mm1,%0\n"
+                    : "+m" (*op.D)
+                    : "m" (*op.A), "m" (*op.B)
+                    : "%mm1", "%mm2", "%mm3", "%mm4");
+    }
+  
   asm("emms");
 }
 
@@ -1283,30 +1269,30 @@ void
 gimp_composite_swap_rgba8_rgba8_rgba8_sse (GimpCompositeContext *_op)
 {
   GimpCompositeContext op = *_op;
-
+  
   for (; op.n_pixels >= 2; op.n_pixels -= 2)
-				{
-    asm volatile ("  movq       %0,%%mm2\n"
-                  "\tmovq       %1,%%mm3\n"
-                  "\tmovq    %%mm3,%0\n"
-                  "\tmovq    %%mm2,%1\n"
-                  : /* empty */
-                  : "m" (*op.A), "m" (*op.B)
-                  : "0", "1", "%mm1", "%mm2", "%mm3", "%mm4");
-						op.A += 8;
-						op.B += 8;
-  }
-
+    {
+      asm volatile ("  movq       %0,%%mm2\n"
+                    "\tmovq       %1,%%mm3\n"
+                    "\tmovq    %%mm3,%0\n"
+                    "\tmovq    %%mm2,%1\n"
+                    : "+m" (*op.A), "+m" (*op.B)
+                    : 
+                    : "%mm1", "%mm2", "%mm3", "%mm4");
+      op.A += 8;
+      op.B += 8;
+    }
+  
   if (op.n_pixels)
-				{
-    asm volatile ("  movd       %0,%%mm2\n"
-                  "\tmovd       %1,%%mm3\n"
-                  "\tmovd    %%mm3,%0\n"
-                  "\tmovd    %%mm2,%1\n"                  
-                  : /* empty */
-                  : "m" (*op.A), "m" (*op.B)
-                  : "0", "1", "%mm1", "%mm2", "%mm3", "%mm4");
-  }
+    {
+      asm volatile ("  movd       %0,%%mm2\n"
+                    "\tmovd       %1,%%mm3\n"
+                    "\tmovd    %%mm3,%0\n"
+                    "\tmovd    %%mm2,%1\n"                  
+                    : "+m" (*op.A), "+m" (*op.B)
+                    : /* empty */
+                    : "%mm1", "%mm2", "%mm3", "%mm4");
+    }
   
   asm("emms");
 }
@@ -1320,20 +1306,20 @@ void
 xxxgimp_composite_addition_va8_va8_va8_sse (GimpCompositeContext *_op)
 {
   GimpCompositeContext op = *_op;
-
+  
   asm("pushl %edi");
   asm("pushl %ebx");
   asm("movl 12(%esp), %edi");
   asm("movq v8_alpha_mask, %mm0");
-
+  
   asm("subl $ 4, %ecx");
   asm("jl .add_pixels_1a_1a_last3");
   asm("movl $ 8, %ebx");
   asm(".add_pixels_1a_1a_loop:");
-
+  
   asm("movq (%eax), %mm2");
   asm("movq (%edx), %mm3");
-
+  
   asm("movq %mm2, %mm4");
   asm("paddusb %mm3, %mm4");
   asm("movq %mm0, %mm1");
@@ -1349,13 +1335,13 @@ xxxgimp_composite_addition_va8_va8_va8_sse (GimpCompositeContext *_op)
   asm("addl %ebx, %edi");
   asm("subl $ 4, %ecx");
   asm("jge .add_pixels_1a_1a_loop");
-
+  
   asm(".add_pixels_1a_1a_last3:");
   asm("test $ 2, %ecx");
   asm("jz .add_pixels_1a_1a_last1");
   asm("movd (%eax), %mm2");
   asm("movd (%edx), %mm3");
-
+  
   asm("movq %mm2, %mm4");
   asm("paddusb %mm3, %mm4");
   asm("movq %mm0, %mm1");
@@ -1368,7 +1354,7 @@ xxxgimp_composite_addition_va8_va8_va8_sse (GimpCompositeContext *_op)
   asm("addl $ 4, %eax");
   asm("addl $ 4, %edx");
   asm("addl $ 4, %edi");
-
+  
   asm(".add_pixels_1a_1a_last1:");
   asm("test $ 1, %ecx");
   asm("jz .add_pixels_1a_1a_end");
@@ -1408,107 +1394,107 @@ xxxgimp_composite_burn_va8_va8_va8_sse (GimpCompositeContext *_op)
       : "%mm1");
 
   for (; op.n_pixels >= 4; op.n_pixels -= 4)
-				{
-    asm volatile ("  movq      (%0),%%mm0; addl $8,%0\n"
-                  "\tmovq      (%1),%%mm1; addl $8,%1\n"
+    {
+      asm volatile ("  movq      (%0),%%mm0; addl $8,%0\n"
+                    "\tmovq      (%1),%%mm1; addl $8,%1\n"
+                    
+                    "\tmovq      %3,%%mm2\n"
+                    "\tpsubb     %%mm0,%%mm2\n" /* mm2 = 255 - A */
+                    "\tpxor      %%mm4,%%mm4\n"
+                    "\tpunpcklbw %%mm2,%%mm4\n" /* mm4 = (255- A) * 256  */
 
-                  "\tmovq      %3,%%mm2\n"
-                  "\tpsubb     %%mm0,%%mm2\n" /* mm2 = 255 - A */
-                  "\tpxor      %%mm4,%%mm4\n"
-                  "\tpunpcklbw %%mm2,%%mm4\n" /* mm4 = (255- A) * 256  */
+                    "\tmovq      %%mm1,%%mm3\n"
+                    "\tpxor      %%mm5,%%mm5\n"
+                    "\tpunpcklbw %%mm5,%%mm3\n"
+                    "\tmovq      %4,%%mm5\n"
+                    "\tpaddusw   %%mm3,%%mm5\n" /* mm5 = B + 1 */
 
-                  "\tmovq      %%mm1,%%mm3\n"
-                  "\tpxor      %%mm5,%%mm5\n"
-                  "\tpunpcklbw %%mm5,%%mm3\n"
-                  "\tmovq      %4,%%mm5\n"
-                  "\tpaddusw   %%mm3,%%mm5\n" /* mm5 = B + 1 */
+                    "\t" pdivwqX(mm4,mm5,mm7) "\n"
 
-                  "\t" pdivwqX(mm4,mm5,mm7) "\n"
+                    "\tmovq      %3,%%mm2\n"
+                    "\tpsubb     %%mm0,%%mm2\n" /* mm2 = 255 - A */
+                    "\tpxor      %%mm4,%%mm4\n"
+                    "\tpunpckhbw %%mm2,%%mm4\n" /* mm4 = (255- A) * 256  */
 
-                  "\tmovq      %3,%%mm2\n"
-                  "\tpsubb     %%mm0,%%mm2\n" /* mm2 = 255 - A */
-                  "\tpxor      %%mm4,%%mm4\n"
-                  "\tpunpckhbw %%mm2,%%mm4\n" /* mm4 = (255- A) * 256  */
+                    "\tmovq      %%mm1,%%mm3\n"
+                    "\tpxor      %%mm5,%%mm5\n"
+                    "\tpunpckhbw %%mm5,%%mm3\n"
+                    "\tmovq      %4,%%mm5\n"
+                    "\tpaddusw   %%mm3,%%mm5\n" /* mm5 = B + 1 */
+                    "\t" pdivwqX(mm4,mm5,mm6) "\n"
 
-                  "\tmovq      %%mm1,%%mm3\n"
-                  "\tpxor      %%mm5,%%mm5\n"
-                  "\tpunpckhbw %%mm5,%%mm3\n"
-                  "\tmovq      %4,%%mm5\n"
-                  "\tpaddusw   %%mm3,%%mm5\n" /* mm5 = B + 1 */
-                  "\t" pdivwqX(mm4,mm5,mm6) "\n"
-
-                  "\tmovq      %5,%%mm4\n"
-                  "\tmovq      %%mm4,%%mm5\n"
-                  "\tpsubusw     %%mm6,%%mm4\n"
-                  "\tpsubusw     %%mm7,%%mm5\n"
+                    "\tmovq      %5,%%mm4\n"
+                    "\tmovq      %%mm4,%%mm5\n"
+                    "\tpsubusw     %%mm6,%%mm4\n"
+                    "\tpsubusw     %%mm7,%%mm5\n"
                   
-                  "\tpackuswb  %%mm4,%%mm5\n"
+                    "\tpackuswb  %%mm4,%%mm5\n"
 
-                  "\t" pminub(mm0,mm1,mm3) "\n" /* mm1 = min(mm0,mm1) clobber mm3 */
-
-                  "\tmovq      %6,%%mm7\n"
-                  "\tpand      %%mm7,%%mm1\n" /* mm1 = mm7 & alpha_mask */
-
-                  "\tpandn     %%mm5,%%mm7\n" /* mm7 = ~mm7 & mm5 */
-                  "\tpor       %%mm1,%%mm7\n" /* mm7 = mm7 | mm1 */
-
-                  "\tmovq      %%mm7,(%2); addl $8,%2\n"
-                  : "+r" (op.A), "+r" (op.B), "+r" (op.D)
-                  : "m" (*va8_b255), "m" (*va8_w1), "m" (*va8_w255), "m" (*va8_alpha_mask)
-                  : "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
-  }
+                    "\t" pminub(mm0,mm1,mm3) "\n" /* mm1 = min(mm0,mm1) clobber mm3 */
+                    
+                    "\tmovq      %6,%%mm7\n"
+                    "\tpand      %%mm7,%%mm1\n" /* mm1 = mm7 & alpha_mask */
+                    
+                    "\tpandn     %%mm5,%%mm7\n" /* mm7 = ~mm7 & mm5 */
+                    "\tpor       %%mm1,%%mm7\n" /* mm7 = mm7 | mm1 */
+                    
+                    "\tmovq      %%mm7,(%2); addl $8,%2\n"
+                    : "+r" (op.A), "+r" (op.B), "+r" (op.D)
+                    : "m" (*va8_b255), "m" (*va8_w1), "m" (*va8_w255), "m" (*va8_alpha_mask)
+                    : "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
+    }
 
   if (op.n_pixels)
-				{
-    asm volatile ("  movd      (%0),%%mm0\n"
-                  "\tmovd      (%1),%%mm1\n"
-
-                  "\tmovq      %3,%%mm2\n"
-                  "\tpsubb     %%mm0,%%mm2\n" /* mm2 = 255 - A */
-                  "\tpxor      %%mm4,%%mm4\n"
-                  "\tpunpcklbw %%mm2,%%mm4\n" /* mm4 = (255- A) * 256  */
-
-                  "\tmovq      %%mm1,%%mm3\n"
-                  "\tpxor      %%mm5,%%mm5\n"
-                  "\tpunpcklbw %%mm5,%%mm3\n"
-                  "\tmovq      %4,%%mm5\n"
-                  "\tpaddusw   %%mm3,%%mm5\n" /* mm5 = B + 1 */
-
-                  "\t" pdivwqX(mm4,mm5,mm7) "\n"
-
-                  "\tmovq      %3,%%mm2\n"
-                  "\tpsubb   %%mm0,%%mm2\n" /* mm2 = 255 - A */
-                  "\tpxor      %%mm4,%%mm4\n"
-                  "\tpunpckhbw %%mm2,%%mm4\n" /* mm4 = (255- A) * 256  */
-
-                  "\tmovq      %%mm1,%%mm3\n"
-                  "\tpxor      %%mm5,%%mm5\n"
-                  "\tpunpckhbw %%mm5,%%mm3\n"
-                  "\tmovq      %4,%%mm5\n"
-                  "\tpaddusw   %%mm3,%%mm5\n" /* mm5 = B + 1 */
-                  "\t" pdivwqX(mm4,mm5,mm6) "\n"
-
-                  "\tmovq      %5,%%mm4\n"
-                  "\tmovq      %%mm4,%%mm5\n"
-                  "\tpsubusw     %%mm6,%%mm4\n"
-                  "\tpsubusw     %%mm7,%%mm5\n"
-                  
-                  "\tpackuswb  %%mm4,%%mm5\n"
-
-                  "\t" pminub(mm0,mm1,mm3) "\n" /* mm1 = min(mm0,mm1) clobber mm3 */
-
-                  "\tmovq      %6,%%mm7\n"
-                  "\tpand      %%mm7,%%mm1\n" /* mm1 = mm7 & alpha_mask */
-
-                  "\tpandn     %%mm5,%%mm7\n" /* mm7 = ~mm7 & mm5 */
-                  "\tpor       %%mm1,%%mm7\n" /* mm7 = mm7 | mm1 */
-
-                  "\tmovd      %%mm7,(%2)\n"
-                  : /* empty */
-                  : "r" (op.A), "r" (op.B), "r" (op.D), "m" (*va8_b255), "m" (*va8_w1), "m" (*va8_w255), "m" (*va8_alpha_mask)
-                  : "0", "1", "2", "%mm0", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5", "%mm6", "%mm7");
-  }
-
+    {
+      asm volatile ("  movd      (%0),%%mm0\n"
+                    "\tmovd      (%1),%%mm1\n"
+                    
+                    "\tmovq      %3,%%mm2\n"
+                    "\tpsubb     %%mm0,%%mm2\n" /* mm2 = 255 - A */
+                    "\tpxor      %%mm4,%%mm4\n"
+                    "\tpunpcklbw %%mm2,%%mm4\n" /* mm4 = (255- A) * 256  */
+                    
+                    "\tmovq      %%mm1,%%mm3\n"
+                    "\tpxor      %%mm5,%%mm5\n"
+                    "\tpunpcklbw %%mm5,%%mm3\n"
+                    "\tmovq      %4,%%mm5\n"
+                    "\tpaddusw   %%mm3,%%mm5\n" /* mm5 = B + 1 */
+                    
+                    "\t" pdivwqX(mm4,mm5,mm7) "\n"
+                    
+                    "\tmovq      %3,%%mm2\n"
+                    "\tpsubb   %%mm0,%%mm2\n" /* mm2 = 255 - A */
+                    "\tpxor      %%mm4,%%mm4\n"
+                    "\tpunpckhbw %%mm2,%%mm4\n" /* mm4 = (255- A) * 256  */
+                    
+                    "\tmovq      %%mm1,%%mm3\n"
+                    "\tpxor      %%mm5,%%mm5\n"
+                    "\tpunpckhbw %%mm5,%%mm3\n"
+                    "\tmovq      %4,%%mm5\n"
+                    "\tpaddusw   %%mm3,%%mm5\n" /* mm5 = B + 1 */
+                    "\t" pdivwqX(mm4,mm5,mm6) "\n"
+                    
+                    "\tmovq      %5,%%mm4\n"
+                    "\tmovq      %%mm4,%%mm5\n"
+                    "\tpsubusw     %%mm6,%%mm4\n"
+                    "\tpsubusw     %%mm7,%%mm5\n"
+                    
+                    "\tpackuswb  %%mm4,%%mm5\n"
+                    
+                    "\t" pminub(mm0,mm1,mm3) "\n" /* mm1 = min(mm0,mm1) clobber mm3 */
+                    
+                    "\tmovq      %6,%%mm7\n"
+                    "\tpand      %%mm7,%%mm1\n" /* mm1 = mm7 & alpha_mask */
+                    
+                    "\tpandn     %%mm5,%%mm7\n" /* mm7 = ~mm7 & mm5 */
+                    "\tpor       %%mm1,%%mm7\n" /* mm7 = mm7 | mm1 */
+                    
+                    "\tmovd      %%mm7,(%2)\n"
+                    : /* empty */
+                    : "r" (op.A), "r" (op.B), "r" (op.D), "m" (*va8_b255), "m" (*va8_w1), "m" (*va8_w255), "m" (*va8_alpha_mask)
+                    : "0", "1", "2", "%mm0", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5", "%mm6", "%mm7");
+    }
+  
   asm("emms");
 }
 
diff --git a/app/composite/gimp-composite-sse2-installer.c b/app/composite/gimp-composite-sse2-installer.c
index 9ce5757eed..a0de35cfde 100644
--- a/app/composite/gimp-composite-sse2-installer.c
+++ b/app/composite/gimp-composite-sse2-installer.c
@@ -20,6 +20,10 @@ static struct install_table {
  { GIMP_COMPOSITE_DIFFERENCE, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, gimp_composite_difference_rgba8_rgba8_rgba8_sse2 }, 
  { GIMP_COMPOSITE_ADDITION, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, gimp_composite_addition_rgba8_rgba8_rgba8_sse2 }, 
  { GIMP_COMPOSITE_SUBTRACT, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, gimp_composite_subtract_rgba8_rgba8_rgba8_sse2 }, 
+ { GIMP_COMPOSITE_DARKEN, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, gimp_composite_darken_rgba8_rgba8_rgba8_sse2 }, 
+ { GIMP_COMPOSITE_LIGHTEN, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, gimp_composite_lighten_rgba8_rgba8_rgba8_sse2 }, 
+ { GIMP_COMPOSITE_DODGE, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, gimp_composite_dodge_rgba8_rgba8_rgba8_sse2 }, 
+ { GIMP_COMPOSITE_GRAIN_EXTRACT, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, gimp_composite_grain_extract_rgba8_rgba8_rgba8_sse2 }, 
  { GIMP_COMPOSITE_SWAP, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, gimp_composite_swap_rgba8_rgba8_rgba8_sse2 }, 
 #endif
  { 0, 0, 0, 0, NULL }
diff --git a/app/composite/gimp-composite-sse2-test.c b/app/composite/gimp-composite-sse2-test.c
index 762f4676d7..fda9fb3bdd 100644
--- a/app/composite/gimp-composite-sse2-test.c
+++ b/app/composite/gimp-composite-sse2-test.c
@@ -57,29 +57,72 @@ gimp_composite_sse2_test (int iterations, int n_pixels)
   }
 
 
-  gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_DIFFERENCE, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
-  gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_DIFFERENCE, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
-  ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
-  ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_difference_rgba8_rgba8_rgba8_sse2, &special_ctx);
-  if (gimp_composite_regression_compare_contexts ("difference", &generic_ctx, &special_ctx)) {
-    return (1);
-  }
-  gimp_composite_regression_timer_report ("difference", ft0, ft1);
-
   gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_ADDITION, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
   gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_ADDITION, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
   ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
   ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_addition_rgba8_rgba8_rgba8_sse2, &special_ctx);
   if (gimp_composite_regression_compare_contexts ("addition", &generic_ctx, &special_ctx)) {
+    printf("addition failed\n");
     return (1);
   }
   gimp_composite_regression_timer_report ("addition", ft0, ft1);
 
+  gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_DARKEN, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
+  gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_DARKEN, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
+  ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
+  ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_darken_rgba8_rgba8_rgba8_sse2, &special_ctx);
+  if (gimp_composite_regression_compare_contexts ("darken", &generic_ctx, &special_ctx)) {
+    printf("darken failed\n");
+    return (1);
+  }
+  gimp_composite_regression_timer_report ("darken", ft0, ft1);
+
+  gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_DIFFERENCE, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
+  gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_DIFFERENCE, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
+  ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
+  ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_difference_rgba8_rgba8_rgba8_sse2, &special_ctx);
+  if (gimp_composite_regression_compare_contexts ("difference", &generic_ctx, &special_ctx)) {
+    printf("difference failed\n");
+    return (1);
+  }
+  gimp_composite_regression_timer_report ("difference", ft0, ft1);
+
+  gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_DODGE, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
+  gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_DODGE, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
+  ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
+  ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_dodge_rgba8_rgba8_rgba8_sse2, &special_ctx);
+  if (gimp_composite_regression_compare_contexts ("dodge", &generic_ctx, &special_ctx)) {
+    printf("dodge failed\n");
+    return (1);
+  }
+  gimp_composite_regression_timer_report ("dodge", ft0, ft1);
+
+  gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_GRAIN_EXTRACT, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
+  gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_GRAIN_EXTRACT, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
+  ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
+  ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_grain_extract_rgba8_rgba8_rgba8_sse2, &special_ctx);
+  if (gimp_composite_regression_compare_contexts ("grain_extract", &generic_ctx, &special_ctx)) {
+    printf("grain_extract failed\n");
+    return (1);
+  }
+  gimp_composite_regression_timer_report ("grain_extract", ft0, ft1);
+
+  gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_LIGHTEN, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
+  gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_LIGHTEN, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
+  ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
+  ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_lighten_rgba8_rgba8_rgba8_sse2, &special_ctx);
+  if (gimp_composite_regression_compare_contexts ("lighten", &generic_ctx, &special_ctx)) {
+    printf("lighten failed\n");
+    return (1);
+  }
+  gimp_composite_regression_timer_report ("lighten", ft0, ft1);
+
   gimp_composite_context_init (&special_ctx, GIMP_COMPOSITE_SUBTRACT, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D2);
   gimp_composite_context_init (&generic_ctx, GIMP_COMPOSITE_SUBTRACT, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, GIMP_PIXELFORMAT_RGBA8, n_pixels, (unsigned char *) rgba8A, (unsigned char *) rgba8B, (unsigned char *) rgba8B, (unsigned char *) rgba8D1);
   ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
   ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_subtract_rgba8_rgba8_rgba8_sse2, &special_ctx);
   if (gimp_composite_regression_compare_contexts ("subtract", &generic_ctx, &special_ctx)) {
+    printf("subtract failed\n");
     return (1);
   }
   gimp_composite_regression_timer_report ("subtract", ft0, ft1);
@@ -89,6 +132,7 @@ gimp_composite_sse2_test (int iterations, int n_pixels)
   ft0 = gimp_composite_regression_time_function (iterations, gimp_composite_dispatch, &generic_ctx);
   ft1 = gimp_composite_regression_time_function (iterations, gimp_composite_swap_rgba8_rgba8_rgba8_sse2, &special_ctx);
   if (gimp_composite_regression_compare_contexts ("swap", &generic_ctx, &special_ctx)) {
+    printf("swap failed\n");
     return (1);
   }
   gimp_composite_regression_timer_report ("swap", ft0, ft1);
@@ -107,7 +151,7 @@ main (int argc, char *argv[])
   putenv ("GIMP_COMPOSITE=0x1");
 
   iterations = 1;
-  n_pixels = 1048577;
+  n_pixels = 163921;
 
   argv++, argc--;
   while (argc >= 2) {
diff --git a/app/composite/gimp-composite-sse2.c b/app/composite/gimp-composite-sse2.c
index 07003dde96..aed14ebf11 100644
--- a/app/composite/gimp-composite-sse2.c
+++ b/app/composite/gimp-composite-sse2.c
@@ -1,4 +1,6 @@
-/* The GIMP -- an image manipulation program
+/* -*- mode: c tab-width: 2; c-basic-indent: 2; indent-tabs-mode: nil -*-
+ *
+ * The GIMP -- an image manipulation program
  * Copyright (C) 1995 Spencer Kimball and Peter Mattis
  *
  * -*- mode: c tab-width: 2; c-basic-indent: 2; indent-tabs-mode: nil -*-
@@ -34,12 +36,10 @@
 
 #include "gimp-composite.h"
 #include "gimp-composite-sse2.h"
+#include "gimp-composite-x86.h"
 
 #if __GNUC__ >= 3
 
-#define pminub(src,dst,tmp)  "pminub " "%%" #src ", %%" #dst
-#define pmaxub(src,dst,tmp)  "pmaxub " "%%" #src ", %%" #dst
-
 const static guint32 rgba8_alpha_mask_128[4] = { 0xFF000000, 0xFF000000, 0xFF000000, 0xFF000000 };
 const static guint32 rgba8_b1_128[4] =         { 0x01010101, 0x01010101, 0x01010101, 0x01010101 };
 const static guint32 rgba8_b255_128[4] =       { 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF };
@@ -90,75 +90,74 @@ void
 gimp_composite_addition_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *_op)
 {
   GimpCompositeContext op = *_op;
-
-  asm volatile ("movdqu    %0,%%xmm0"
+  
+  asm volatile ("  movdqu    %0,%%xmm0\n"
+                "\tmovq      %1,%%mm0"    
                 : /* empty */
-                : "m" (*rgba8_alpha_mask_128)
-                : "%xmm0");
-
+                : "m" (*rgba8_alpha_mask_128), "m" (*rgba8_alpha_mask_64)
+                : "%xmm0", "%mm0");
+  
   for (; op.n_pixels >= 4; op.n_pixels -= 4)
     {
-      asm ("  movdqu  %0, %%xmm2\n"
-           "\tmovdqu  %1, %%xmm3\n"
-           "\tmovdqu  %%xmm2, %%xmm4\n"
-           "\tpaddusb %%xmm3, %%xmm4\n"
-
-           "\tmovdqu  %%xmm0, %%xmm1\n"
-           "\tpandn   %%xmm4, %%xmm1\n"
-           "\t" pminub(xmm3, xmm2, xmm4) "\n"
-           "\tpand    %%xmm0, %%xmm2\n"
-           "\tpor     %%xmm2, %%xmm1\n"
-           "\tmovdqu  %%xmm1, %2\n"
-           : /* empty */
-           : "m" (*op.A), "m" (*op.B), "m" (*op.D)
-           : "0", "1", "2", "%xmm0", "%xmm1", "%xmm2", "%xmm3", "%xmm4", "%xmm5", "%xmm6", "%xmm7");
+      asm ("  movdqu      %1,%%xmm2\n"
+           "\tmovdqu      %2,%%xmm3\n"
+           "\tmovdqu  %%xmm2,%%xmm4\n"
+           "\tpaddusb %%xmm3,%%xmm4\n"
+           
+           "\tmovdqu  %%xmm0,%%xmm1\n"
+           "\tpandn   %%xmm4,%%xmm1\n"
+           "\tpminub  %%xmm3,%%xmm2\n"
+           "\tpand    %%xmm0,%%xmm2\n"
+           "\tpor     %%xmm2,%%xmm1\n"
+           "\tmovdqu  %%xmm1,%0\n"
+           : "+m" (*op.D)
+           : "m" (*op.A), "m" (*op.B)
+           : "%xmm0", "%xmm1", "%xmm2", "%xmm3", "%xmm4", "%xmm5", "%xmm6", "%xmm7");
       op.A += 16;
       op.B += 16;
       op.D += 16;
     }
-
-  asm volatile ("movq    %0,%%mm0"    
-                : /* empty */
-                : "m" (*rgba8_alpha_mask_64)
-                : "%mm0");
-
+  
   for (; op.n_pixels >= 2; op.n_pixels -= 2)
     {
-      asm ("  movq       %0, %%mm2\n"
-           "\tmovq       %1, %%mm3\n"
-           "\tmovq    %%mm2, %%mm4\n"
-           "\tpaddusb %%mm3, %%mm4\n"
-           "\tmovq    %%mm0, %%mm1\n"
-           "\tpandn   %%mm4, %%mm1\n"
-           "\t" pminub(mm3, mm2, mm4) "\n"
-           "\tpand    %%mm0, %%mm2\n"
-           "\tpor     %%mm2, %%mm1\n"
-           "\tmovq    %%mm1, %2\n"
-           : /* empty */
-           : "m" (*op.A), "m" (*op.B), "m" (*op.D)
-           : "0", "1", "2", "%mm0", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5", "%mm6", "%mm7");
+      asm ("  movq       %1,%%mm2\n"
+           "\tmovq       %2,%%mm3\n"
+           "\tmovq    %%mm2,%%mm4\n"
+           "\tpaddusb %%mm3,%%mm4\n"
+           "\tmovq    %%mm0,%%mm1\n"
+           "\tpandn   %%mm4,%%mm1\n"
+           "\tpminub  %%mm3,%%mm2\n"
+           "\tpand    %%mm0,%%mm2\n"
+           "\tpor     %%mm2,%%mm1\n"
+           "\tmovq    %%mm1,%0\n"
+           : "+m" (*op.D)
+           : "m" (*op.A), "m" (*op.B)
+           : "%mm0", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5", "%mm6", "%mm7");
       op.A += 8;
       op.B += 8;
       op.D += 8;
     }
-
-  if (op.n_pixels)
+  
+  for (; op.n_pixels >= 1; op.n_pixels -= 1)
     {
-      asm volatile ("  movd    (%0), %%mm2;\n"
-                    "\tmovd    (%1), %%mm3;\n"
-                    "\tmovq    %%mm2, %%mm4\n"
-                    "\tpaddusb %%mm3, %%mm4\n"
-                    "\tmovq    %%mm0, %%mm1\n"
-                    "\tpandn   %%mm4, %%mm1\n"
-                    "\t" pminub(mm3, mm2, mm4) "\n"
-                    "\tpand    %%mm0, %%mm2\n"
-                    "\tpor     %%mm2, %%mm1\n"
-                    "\tmovd    %%mm1, (%2);\n"
-                    : /* empty */
-                    : "r" (op.A), "r" (op.B), "r" (op.D)
-                    : "0", "1", "2", "%mm0", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5", "%mm6", "%mm7");
+      asm volatile ("  movd       %1,%%mm2\n"
+                    "\tmovd       %2,%%mm3\n"
+                    "\tmovq    %%mm2,%%mm4\n"
+                    "\tpaddusb %%mm3,%%mm4\n"
+                    "\tmovq    %%mm0,%%mm1\n"
+                    "\tpandn   %%mm4,%%mm1\n"
+                    "\tpminub  %%mm3,%%mm2\n"
+                    "\tpand    %%mm0,%%mm2\n"
+                    "\tpor     %%mm2,%%mm1\n"
+                    "\tmovd    %%mm1,%0\n"
+                    : "+m" (*op.D)
+                    : "m" (*op.A), "m" (*op.B)
+                    : "%mm0", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5", "%mm6", "%mm7");
+      op.A += 4;
+      op.B += 4;
+      op.D += 4;
     }
-
+  
   asm("emms");
 }
 
@@ -166,31 +165,70 @@ gimp_composite_addition_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *_op)
 void
 xxxgimp_composite_burn_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *_op)
 {
-
+  
 }
 
-
 void
-xxxgimp_composite_darken_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *_op)
+gimp_composite_darken_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *_op)
 {
-
+  GimpCompositeContext op = *_op;
+  
+  for (; op.n_pixels >= 4; op.n_pixels -= 4)
+    {
+      asm volatile ("  movdqu          %1,%%xmm2\n"
+                    "\tmovdqu          %2,%%xmm3\n"
+                    "\tpminub      %%xmm3,%%xmm2\n"
+                    "\tmovdqu      %%xmm2,%0\n"
+                    : "=m" (*op.D)
+                    : "m" (*op.A), "m" (*op.B)
+                    : "%xmm1", "%xmm2", "%xmm3", "%xmm4");
+      op.A += 16;
+      op.B += 16;
+      op.D += 16;
+    }
+  
+  for (; op.n_pixels >= 2; op.n_pixels -= 2)
+    {
+      asm volatile ("  movq       %1, %%mm2\n"
+                    "\tpminub     %2, %%mm2\n"
+                    "\tmovq    %%mm2, %0\n"
+                    : "=m" (*op.D)
+                    : "m" (*op.A), "m" (*op.B)
+                    : "%mm1", "%mm2", "%mm3", "%mm4");
+      op.A += 8;
+      op.B += 8;
+      op.D += 8;
+    }
+  
+  if (op.n_pixels)
+    {
+      asm volatile ("  movd       %1, %%mm2\n"
+                    "\tmovd       %2, %%mm3\n"
+                    "\tpminub  %%mm3, %%mm2\n"
+                    "\tmovd    %%mm2, %0\n"
+                    : "=m" (*op.D)
+                    : "m" (*op.A), "m" (*op.B)
+                    : "%mm2", "%mm3", "%mm4");
+    }
+  
+  asm("emms");
 }
 
 void
 gimp_composite_difference_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *_op)
 {
   GimpCompositeContext op = *_op;
-
+  
   asm volatile ("  movq   %0,%%mm0\n"
                 "\tmovdqu %1,%%xmm0"
                 :
                 : "m" (*rgba8_alpha_mask_64), "m" (*rgba8_alpha_mask_128)
-                : "%mm0");
-
+                : "%mm0", "%xmm0");
+  
   for (; op.n_pixels >= 4; op.n_pixels -= 4)
     {
-      asm volatile ("  movdqu        %0,%%xmm2\n"
-                    "\tmovdqu        %1,%%xmm3\n"
+      asm volatile ("  movdqu        %1,%%xmm2\n"
+                    "\tmovdqu        %2,%%xmm3\n"
                     "\tmovdqu    %%xmm2,%%xmm4\n"
                     "\tmovdqu    %%xmm3,%%xmm5\n"
                     "\tpsubusb   %%xmm3,%%xmm4\n"
@@ -201,10 +239,10 @@ gimp_composite_difference_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *_op)
                     "\tpminub    %%xmm3,%%xmm2\n"
                     "\tpand      %%xmm0,%%xmm2\n"
                     "\tpor       %%xmm2,%%xmm1\n"
-                    "\tmovdqu    %%xmm1,%2\n"
-                    : /* empty */
-                    : "m" (*op.A), "m" (*op.B), "m" (*op.D)
-                    : "0", "1", "2", "%xmm1", "%xmm2", "%xmm3", "%xmm4", "%xmm5");
+                    "\tmovdqu    %%xmm1,%0\n"
+                    : "+m" (*op.D)
+                    : "m" (*op.A), "m" (*op.B)
+                    : "%xmm1", "%xmm2", "%xmm3", "%xmm4", "%xmm5");
       op.A += 16;
       op.B += 16;
       op.D += 16;
@@ -212,8 +250,8 @@ gimp_composite_difference_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *_op)
   
   for (; op.n_pixels >= 2; op.n_pixels -= 2)
     {
-      asm volatile ("  movq       %0, %%mm2\n"
-                    "\tmovq       %1, %%mm3\n"
+      asm volatile ("  movq       %1, %%mm2\n"
+                    "\tmovq       %2, %%mm3\n"
                     "\tmovq    %%mm2, %%mm4\n"
                     "\tmovq    %%mm3, %%mm5\n"
                     "\tpsubusb %%mm3, %%mm4\n"
@@ -224,10 +262,10 @@ gimp_composite_difference_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *_op)
                     "\tpminub  %%mm3, %%mm2\n"
                     "\tpand    %%mm0, %%mm2\n"
                     "\tpor     %%mm2, %%mm1\n"
-                    "\tmovq    %%mm1, %2\n"
-                    : /* empty */
-                    : "m" (*op.A), "m" (*op.B), "m" (*op.D)
-                    : "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
+                    "\tmovq    %%mm1, %0\n"
+                    : "+m" (*op.D)
+                    : "m" (*op.A), "m" (*op.B)
+                    : "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
       op.A += 8;
       op.B += 8;
       op.D += 8;
@@ -235,8 +273,8 @@ gimp_composite_difference_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *_op)
   
   if (op.n_pixels)
     {
-      asm volatile ("  movd       %0, %%mm2\n"
-                    "\tmovd       %1, %%mm3\n"
+      asm volatile ("  movd       %1, %%mm2\n"
+                    "\tmovd       %2, %%mm3\n"
                     "\tmovq    %%mm2, %%mm4\n"
                     "\tmovq    %%mm3, %%mm5\n"
                     "\tpsubusb %%mm3, %%mm4\n"
@@ -247,141 +285,398 @@ gimp_composite_difference_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *_op)
                     "\tpminub  %%mm3, %%mm2\n"
                     "\tpand    %%mm0, %%mm2\n"
                     "\tpor     %%mm2, %%mm1\n"
-                    "\tmovd    %%mm1, %2\n"
-                    : /* empty */
-                    : "m" (*op.A), "m" (*op.B), "m" (*op.D)
-                    : "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4");
+                    "\tmovd    %%mm1, %0\n"
+                    : "+m" (*op.D)
+                    : "m" (*op.A), "m" (*op.B)
+                    : "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
     }
+  
+  asm("emms");
+}
 
+
+void
+gimp_composite_dodge_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *_op)
+{
+  GimpCompositeContext op = *_op;
+
+#if 0  
+  for (; op.n_pixels >= 4; op.n_pixels -= 4)
+    {
+      asm volatile ("  movdqu        %0,%%xmm0\n"
+                    "\tmovdqu        %1,%%xmm1\n"
+                    "\tmovdqu    %%xmm1,%%xmm3\n"
+                    "\tpxor      %%xmm2,%%xmm2\n"
+                    "\tpunpcklbw %%xmm2,%%xmm3\n"
+                    "\tpunpcklbw %%xmm0,%%xmm2\n"
+                    
+                    "\tmovdqu        %3,%%xmm4\n"
+                    "\tpsubw     %%xmm3,%%xmm4\n"
+                    
+                    "\t" xmm_pdivwuqX(xmm2,xmm4,xmm5,xmm6) "\n"
+                    
+                    "\tmovdqu    %%xmm1,%%xmm3\n"
+                    "\tpxor      %%xmm2,%%xmm2\n"
+                    "\tpunpckhbw %%xmm2,%%xmm3\n"
+                    "\tpunpckhbw %%xmm0,%%xmm2\n"
+                    
+                    "\tmovdqu        %3,%%xmm4\n"
+                    "\tpsubw     %%xmm3,%%xmm4\n"
+                    
+                    "\t" xmm_pdivwuqX(xmm2,xmm4,xmm6,xmm7) "\n"
+                    
+                    "\tpackuswb  %%xmm6,%%xmm5\n"
+                    
+                    "\tmovdqu        %4,%%xmm6\n"
+                    "\tmovdqu    %%xmm1,%%xmm7\n"
+                    "\tpminub    %%xmm0,%%xmm7\n"
+                    "\tpand      %%xmm6,%%xmm7\n"
+                    "\tpandn     %%xmm5,%%xmm6\n"
+                    
+                    "\tpor       %%xmm6,%%xmm7\n"
+                    
+                    "\tmovdqu    %%xmm7,%2\n"
+                    : /* empty */
+                    : "m" (*op.A), "m" (*op.B), "m" (*op.D), "m" (*rgba8_w256_128), "m" (*rgba8_alpha_mask_128)
+                    : "0", "1", "2", "%eax", "%ecx", "%edx", "%xmm0", "%xmm1", "%xmm2", "%xmm3", "%xmm4", "%xmm5", "%xmm6", "%xmm7");
+      op.A += 16;
+      op.B += 16;
+      op.D += 16;
+    }
+#endif
+  
+  for (; op.n_pixels >= 2; op.n_pixels -= 2)
+    {
+      asm volatile ("  movq         %0,%%mm0\n"
+                    "\tmovq         %1,%%mm1\n"
+                    "\tmovq      %%mm1,%%mm3\n"
+                    "\tpxor      %%mm2,%%mm2\n"
+                    "\tpunpcklbw %%mm2,%%mm3\n"
+                    "\tpunpcklbw %%mm0,%%mm2\n"
+                    
+                    "\tmovq         %3,%%mm4\n"
+                    "\tpsubw     %%mm3,%%mm4\n"
+                    
+                    "\t" pdivwuqX(mm2,mm4,mm5) "\n"
+                    
+                    "\tmovq      %%mm1,%%mm3\n"
+                    "\tpxor      %%mm2,%%mm2\n"
+                    "\tpunpckhbw %%mm2,%%mm3\n"
+                    "\tpunpckhbw %%mm0,%%mm2\n"
+                    
+                    "\tmovq         %3,%%mm4\n"
+                    "\tpsubw     %%mm3,%%mm4\n"
+                    
+                    "\t" pdivwuqX(mm2,mm4,mm6) "\n"
+                    
+                    "\tpackuswb  %%mm6,%%mm5\n"
+                    
+                    "\tmovq         %4,%%mm6\n"
+                    "\tmovq      %%mm1,%%mm7\n"
+                    "\tpminub    %%mm0,%%mm7\n"
+                    "\tpand      %%mm6,%%mm7\n"
+                    "\tpandn     %%mm5,%%mm6\n"
+                    
+                    "\tpor       %%mm6,%%mm7\n"
+                    
+                    "\tmovq      %%mm7,%2\n"
+                    : /* empty */
+                    : "m" (*op.A), "m" (*op.B), "m" (*op.D), "m" (*rgba8_w256_64), "m" (*rgba8_alpha_mask_64)
+                    : "0", "1", "2", "%eax", "%ecx", "%edx", "%mm0", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5", "%mm6", "%mm7");
+      op.A += 8;
+      op.B += 8;
+      op.D += 8;
+    }
+  
+  if (op.n_pixels)
+    {
+      asm volatile ("  movd         %0,%%mm0\n"
+                    "\tmovq         %1,%%mm1\n"
+                    "\tmovq      %%mm1,%%mm3\n"
+                    "\tpxor      %%mm2,%%mm2\n"
+                    "\tpunpcklbw %%mm2,%%mm3\n"
+                    "\tpunpcklbw %%mm0,%%mm2\n"
+                    
+                    "\tmovq         %3,%%mm4\n"
+                    "\tpsubw     %%mm3,%%mm4\n"
+                    
+                    "\t" pdivwuqX(mm2,mm4,mm5) "\n"
+                    
+                    "\tmovq      %%mm1,%%mm3\n"
+                    "\tpxor      %%mm2,%%mm2\n"
+                    "\tpunpckhbw %%mm2,%%mm3\n"
+                    "\tpunpckhbw %%mm0,%%mm2\n"
+                    
+                    "\tmovq         %3,%%mm4\n"
+                    "\tpsubw     %%mm3,%%mm4\n"
+                    
+                    "\t" pdivwuqX(mm2,mm4,mm6) "\n"
+                    
+                    "\tpackuswb  %%mm6,%%mm5\n"
+                    
+                    "\tmovq         %4,%%mm6\n"
+                    "\tmovq      %%mm1,%%mm7\n"
+                    "\tpminub    %%mm0,%%mm7\n"
+                    "\tpand      %%mm6,%%mm7\n"
+                    "\tpandn     %%mm5,%%mm6\n"
+                    
+                    "\tpor       %%mm6,%%mm7\n"
+                    
+                    "\tmovd      %%mm7,%2\n"
+                    : /* empty */
+                    : "m" (*op.A), "m" (*op.B), "m" (*op.D), "m" (*rgba8_w256_64), "m" (*rgba8_alpha_mask_64)
+                    : "0", "1", "2", "%eax", "%ecx", "%edx", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
+    }
+  
   asm("emms");
 }
 
 void
-xxxgimp_composite_divide_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *_op)
+gimp_composite_grain_extract_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *_op)
 {
+  GimpCompositeContext op = *_op;
+  
+  asm volatile ("  movq       %0,%%mm0\n"
+                "\tpxor    %%mm6,%%mm6\n"
+                "\tmovq       %1,%%mm7\n"
+																"\tmovdqu     %2,%%xmm0\n"
+                "\tpxor   %%xmm6,%%xmm6\n"
+                "\tmovdqu     %3,%%xmm7\n"
+                : /* empty */
+                : "m" (*rgba8_alpha_mask_64), "m" (*rgba8_w128_64), "m" (*rgba8_alpha_mask_128), "m" (*rgba8_w128_128)
+                : "%mm0", "%mm6", "%mm7", "%xmm0", "%xmm6", "%xmm7");
+
+  for (; op.n_pixels >= 4; op.n_pixels -= 4)
+    {
+      asm volatile ("  movdqu       %1,%%xmm2\n"
+                    "\tmovdqu       %2,%%xmm3\n"
+                    xmm_low_bytes_to_words(xmm2,xmm4,xmm6)
+                    xmm_low_bytes_to_words(xmm3,xmm5,xmm6)
+                    "\tpsubw     %%xmm5,%%xmm4\n"
+                    "\tpaddw     %%xmm7,%%xmm4\n"
+                    "\tmovdqu    %%xmm4,%%xmm1\n"
+                    
+                    xmm_high_bytes_to_words(xmm2,xmm4,xmm6)
+                    xmm_high_bytes_to_words(xmm3,xmm5,xmm6)
+                    
+                    "\tpsubw     %%xmm5,%%xmm4\n"
+                    "\tpaddw     %%xmm7,%%xmm4\n"
+                    
+                    "\tpackuswb  %%xmm4,%%xmm1\n"
+                    "\tmovdqu    %%xmm1,%%xmm4\n"
+                    
+                    "\tmovdqu    %%xmm0,%%xmm1\n"
+                    "\tpandn     %%xmm4,%%xmm1\n"
+                    
+                    "\tpminub    %%xmm3,%%xmm2\n"
+                    "\tpand      %%xmm0,%%xmm2\n"
+                    
+                    "\tpor       %%xmm2,%%xmm1\n"
+                    "\tmovdqu    %%xmm1,%0\n"
+                    : "+m" (*op.D)
+                    : "m" (*op.A), "m" (*op.B)
+                    : "%xmm1", "%xmm2", "%xmm3", "%xmm4");
+      op.A += 16;
+      op.B += 16;
+      op.D += 16;
+    }
+  
+  for (; op.n_pixels >= 2; op.n_pixels -= 2)
+    {
+      asm volatile ("  movq         %1,%%mm2\n"
+                    "\tmovq         %2,%%mm3\n"
+                    mmx_low_bytes_to_words(mm2,mm4,mm6)
+                    mmx_low_bytes_to_words(mm3,mm5,mm6)
+                    "\tpsubw     %%mm5,%%mm4\n"
+                    "\tpaddw     %%mm7,%%mm4\n"
+                    "\tmovq      %%mm4,%%mm1\n"
+                    
+                    mmx_high_bytes_to_words(mm2,mm4,mm6)
+                    mmx_high_bytes_to_words(mm3,mm5,mm6)
+                    
+                    "\tpsubw     %%mm5,%%mm4\n"
+                    "\tpaddw     %%mm7,%%mm4\n"
+                    
+                    "\tpackuswb  %%mm4,%%mm1\n"
+                    "\tmovq      %%mm1,%%mm4\n"
+                    
+                    "\tmovq      %%mm0,%%mm1\n"
+                    "\tpandn     %%mm4,%%mm1\n"
+                    
+                    "\tpminub    %%mm3,%%mm2\n"
+                    "\tpand      %%mm0,%%mm2\n"
+                    
+                    "\tpor       %%mm2,%%mm1\n"
+                    "\tmovq      %%mm1,%0\n"
+                    : "+m" (*op.D)
+                    : "m" (*op.A), "m" (*op.B)
+                    : "%mm1", "%mm2", "%mm3", "%mm4");
+      op.A += 8;
+      op.B += 8;
+      op.D += 8;
+    }
+  
+  if (op.n_pixels)
+    {
+      asm volatile ("  movd         %1, %%mm2\n"
+                    "\tmovd         %2, %%mm3\n"
+                    mmx_low_bytes_to_words(mm2,mm4,mm6)
+                    mmx_low_bytes_to_words(mm3,mm5,mm6)
+                    "\tpsubw     %%mm5, %%mm4\n"
+                    "\tpaddw     %%mm7, %%mm4\n"
+                    "\tmovq      %%mm4, %%mm1\n"
+                    "\tpackuswb  %%mm6, %%mm1\n"
+                    "\tmovq      %%mm1, %%mm4\n"
+                    "\tmovq      %%mm0, %%mm1\n"
+                    "\tpandn     %%mm4, %%mm1\n"
+                    "\tpminub    %%mm3, %%mm2\n"
+                    "\tpand      %%mm0, %%mm2\n"
+                    "\tpor       %%mm2, %%mm1\n"
+                    "\tmovd      %%mm1, %0\n"
+                    : "+m" (*op.D)
+                    : "m" (*op.A), "m" (*op.B)
+                    : "%mm1", "%mm2", "%mm3", "%mm4");
+    }
+  
+  asm("emms");
 }
 
 void
-xxxgimp_composite_dodge_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *_op)
+gimp_composite_lighten_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *_op)
 {
+  GimpCompositeContext op = *_op;
+  
+  asm volatile ("movq    %0,%%mm0"     :  : "m" (*rgba8_alpha_mask_64) : "%mm0");
+  
+  for (; op.n_pixels >= 4; op.n_pixels -= 4)
+    {
+      asm volatile ("  movdqu      %1, %%xmm2\n"
+                    "\tmovdqu      %2, %%xmm3\n"
+                    "\tmovdqu  %%xmm2, %%xmm4\n"
+                    "\tpmaxub  %%xmm3, %%xmm4\n"
+                    "\tmovdqu  %%xmm0, %%xmm1\n"
+                    "\tpandn   %%xmm4, %%xmm1\n"
+                    "\tpminub  %%xmm2, %%xmm3\n"
+                    "\tpand    %%xmm0, %%xmm3\n"
+                    "\tpor     %%xmm3, %%xmm1\n"
+                    "\tmovdqu  %%xmm1, %0\n"
+                    : "=m" (*op.D)
+                    : "m" (*op.A), "m" (*op.B)
+                    : "%xmm1", "%xmm2", "%xmm3", "%xmm4");
+      op.A += 16;
+      op.B += 16;
+      op.D += 16;
+    }
+  
+  for (; op.n_pixels >= 2; op.n_pixels -= 2)
+    {
+      asm volatile ("  movq       %1, %%mm2\n"
+                    "\tmovq       %2, %%mm3\n"
+                    "\tmovq    %%mm2, %%mm4\n"
+                    "\tpmaxub  %%mm3, %%mm4\n"
+                    "\tmovq    %%mm0, %%mm1\n"
+                    "\tpandn   %%mm4, %%mm1\n"
+                    "\tpminub  %%mm2, %%mm3\n"
+                    "\tpand    %%mm0, %%mm3\n"
+                    "\tpor     %%mm3, %%mm1\n"
+                    "\tmovq    %%mm1, %0\n"
+                    : "=m" (*op.D)
+                    : "m" (*op.A), "m" (*op.B)
+                    : "%mm1", "%mm2", "%mm3", "%mm4");
+      op.A += 8;
+      op.B += 8;
+      op.D += 8;
+    }
+  
+  if (op.n_pixels)
+    {
+      asm volatile ("  movd       %1, %%mm2\n"
+                    "\tmovd       %2, %%mm3\n"
+                    "\tmovq    %%mm2, %%mm4\n"
+                    "\tpmaxub  %%mm3, %%mm4\n"
+                    "\tmovq    %%mm0, %%mm1\n"
+                    "\tpandn   %%mm4, %%mm1\n"
+                    "\tpminub  %%mm2,%%mm3\n"
+                    "\tpand    %%mm0, %%mm3\n"
+                    "\tpor     %%mm3, %%mm1\n"
+                    "\tmovd    %%mm1, %0\n"
+                    : "=m" (*op.D)
+                    : "m" (*op.A), "m" (*op.B)
+                    : "%mm1", "%mm2", "%mm3", "%mm4");
+    }
+  
+  asm("emms");
 }
 
-void
-xxxgimp_composite_grain_extract_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *_op)
-{
-
-}
-
-void
-xxxgimp_composite_grain_merge_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *_op)
-{
-
-}
-
-void
-xxxgimp_composite_lighten_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *_op)
-{
-}
-
-void
-xxxgimp_composite_multiply_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *_op)
-{
-}
-
-void
-xxxgimp_composite_overlay_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *_op)
-{
-}
-
-
-void
-xxxgimp_composite_scale_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *_op)
-{
-}
-
-void
-xxxgimp_composite_screen_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *_op)
-{
-}
-
-
 void
 gimp_composite_subtract_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *_op)
 {
   GimpCompositeContext op = *_op;
-
+  
   asm volatile ("  movq    %0,%%mm0\n"
                 "\tmovdqu  %1,%%xmm0\n"
                 : /* empty */
                 : "m" (*rgba8_alpha_mask_64), "m" (*rgba8_alpha_mask_128)
                 : "%mm0", "%xmm0");
-
+  
   for (; op.n_pixels >= 4; op.n_pixels -= 4)
     {
-      asm volatile ("  movdqu       %0,%%xmm2\n"
-                    "\tmovdqu       %1,%%xmm3\n"
+      asm volatile ("  movdqu       %1,%%xmm2\n"
+                    "\tmovdqu       %2,%%xmm3\n"
                     "\tmovdqu   %%xmm2,%%xmm4\n"
                     "\tpsubusb  %%xmm3,%%xmm4\n"
-                  
+                    
                     "\tmovdqu   %%xmm0,%%xmm1\n"
                     "\tpandn    %%xmm4,%%xmm1\n"
-                    "\t" pminub(xmm3,xmm2,xmm4) "\n"
+                    "\tpminub   %%xmm3,%%xmm2\n"
                     "\tpand     %%xmm0,%%xmm2\n"
                     "\tpor      %%xmm2,%%xmm1\n"
-                    "\tmovdqu   %%xmm1,%2\n"
-                    : /* empty */
-                    : "m" (*op.A), "m" (*op.B), "m" (*op.D)
-                    : "0", "1", "2", "%xmm1", "%xmm2", "%xmm3", "%xmm4", "%xmm5");
+                    "\tmovdqu   %%xmm1,%0\n"
+                    : "+m" (*op.D)
+                    : "m" (*op.A), "m" (*op.B)
+                    : "%xmm1", "%xmm2", "%xmm3", "%xmm4");
       op.A += 16;
       op.B += 16;
       op.D += 16;
     }
-
+  
   for (; op.n_pixels >= 2; op.n_pixels -= 2)
     {
-      asm volatile ("  movq       %0,%%mm2\n"
-                    "\tmovq       %1,%%mm3\n"
-
+      asm volatile ("  movq       %1,%%mm2\n"
+                    "\tmovq       %2,%%mm3\n"
                     "\tmovq    %%mm2,%%mm4\n"
                     "\tpsubusb %%mm3,%%mm4\n"
-                  
                     "\tmovq    %%mm0,%%mm1\n"
                     "\tpandn   %%mm4,%%mm1\n"
-                    
-                    "\t" pminub(mm3,mm2,mm4) "\n"
-                    
+                    "\tpminub  %%mm3,%%mm2\n"
                     "\tpand    %%mm0,%%mm2\n"
                     "\tpor     %%mm2,%%mm1\n"
-                    "\tmovq    %%mm1,%2\n"
-                    : /* empty */
-                    : "m" (*op.A), "m" (*op.B), "m" (*op.D)
-                    : "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
+                    "\tmovq    %%mm1,%0\n"
+                    : "+m" (*op.D)
+                    : "m" (*op.A), "m" (*op.B)
+                    : "%mm1", "%mm2", "%mm3", "%mm4");
       op.A += 8;
       op.B += 8;
       op.D += 8;
-  }
-
+    }
+  
   if (op.n_pixels)
     {
-      asm volatile ("  movd       %0,%%mm2\n"
-                    "\tmovd       %1,%%mm3\n"
-
+      asm volatile ("  movd       %1,%%mm2\n"
+                    "\tmovd       %2,%%mm3\n"
                     "\tmovq    %%mm2,%%mm4\n"
                     "\tpsubusb %%mm3,%%mm4\n"
-                  
                     "\tmovq    %%mm0,%%mm1\n"
                     "\tpandn   %%mm4,%%mm1\n"
-                  
-                    "\t" pminub(mm3,mm2,mm4) "\n"
-
+                    "\tpminub  %%mm3,%%mm2\n"
                     "\tpand    %%mm0,%%mm2\n"
                     "\tpor     %%mm2,%%mm1\n"
-                    "\tmovd    %%mm1,%2\n"
-                    : /* empty */
-                    : "m" (*op.A), "m" (*op.B), "m" (*op.D)
-                    : "0", "1", "2", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5");
-  }
-
+                    "\tmovd    %%mm1,%0\n"
+                    : "+m" (*op.D)
+                    : "m" (*op.A), "m" (*op.B)
+                    : "%mm1", "%mm2", "%mm3", "%mm4");
+    }
+  
   asm("emms");
 }
 
@@ -389,7 +684,7 @@ void
 gimp_composite_swap_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *_op)
 {
   GimpCompositeContext op = *_op;
-
+  
   /*
    * Inhale one whole i686 cache line at once. 64 bytes, 16 rgba8 pixels, 4 128 bit xmm registers.
    */
@@ -403,7 +698,7 @@ gimp_composite_swap_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *_op)
                     "\tmovdqu      %5,%%xmm5\n"
                     "\tmovdqu      %6,%%xmm6\n"
                     "\tmovdqu      %7,%%xmm7\n"
-
+                    
                     "\tmovdqu      %%xmm0,%1\n"
                     "\tmovdqu      %%xmm1,%0\n"
                     "\tmovdqu      %%xmm2,%3\n"
@@ -412,52 +707,53 @@ gimp_composite_swap_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *_op)
                     "\tmovdqu      %%xmm5,%4\n"
                     "\tmovdqu      %%xmm6,%7\n"
                     "\tmovdqu      %%xmm7,%6\n"
+                    : "+m" (op.A[0]), "+m" (op.B[0]),
+                      "+m" (op.A[1]), "+m" (op.B[1]),
+                      "+m" (op.A[2]), "+m" (op.B[2]),
+                      "+m" (op.A[3]), "+m" (op.B[3])
                     : /* empty */
-                    : "m" (op.A[0]), "m" (op.B[0]),
-                    "m" (op.A[1]), "m" (op.B[1]),
-                    "m" (op.A[2]), "m" (op.B[2]),
-                    "m" (op.A[3]), "m" (op.B[3])
-																				);
+                    : "%xmm0", "%xmm1", "%xmm2", "%xmm3", "%xmm4", "%xmm5", "%xmm6", "%xmm7"
+                    );
       op.A += 64;
       op.B += 64;
-  }
-
+    }
+  
   for (; op.n_pixels >= 4; op.n_pixels -= 4)
     {
-    asm volatile ("  movdqu      %0,%%xmm2\n"
-                  "\tmovdqu      %1,%%xmm3\n"
-                  "\tmovdqu  %%xmm3,%0\n"
-                  "\tmovdqu  %%xmm2,%1\n"
-                  : /* empty */
-                  : "m" (*op.A), "m" (*op.B)
-                  : "0", "1", "%xmm1", "%xmm2", "%xmm3", "%xmm4");
+      asm volatile ("  movdqu      %0,%%xmm2\n"
+                    "\tmovdqu      %1,%%xmm3\n"
+                    "\tmovdqu  %%xmm3,%0\n"
+                    "\tmovdqu  %%xmm2,%1\n"
+                    : "+m" (*op.A), "+m" (*op.B)
+                    : /* empty */
+                    : "%xmm2", "%xmm3");
       op.A += 16;
       op.B += 16;
-  }
-
+    }
+  
   for (; op.n_pixels >= 2; op.n_pixels -= 2)
     {
-    asm volatile ("  movq      %0,%%mm2\n"
-                  "\tmovq      %1,%%mm3\n"
-                  "\tmovq   %%mm3,%0\n"
-                  "\tmovq   %%mm2,%1\n"
-                  : /* empty */
-                  : "m" (*op.A), "m" (*op.B)
-                  : "0", "1", "%mm1", "%mm2", "%mm3", "%mm4");
+      asm volatile ("  movq      %0,%%mm2\n"
+                    "\tmovq      %1,%%mm3\n"
+                    "\tmovq   %%mm3,%0\n"
+                    "\tmovq   %%mm2,%1\n"
+                    : "+m" (*op.A), "+m" (*op.B)
+                    : /* empty */
+                    : "%mm2", "%mm3");
       op.A += 8;
       op.B += 8;
-  }
-
+    }
+  
   if (op.n_pixels)
     {
-    asm volatile ("  movd      %0,%%mm2\n"
-                  "\tmovd      %1,%%mm3\n"
-                  "\tmovd   %%mm3,%0\n"
-                  "\tmovd   %%mm2,%1\n"                  
-                  : /* empty */
-                  : "m" (*op.A), "m" (*op.B)
-                  : "0", "1", "%mm1", "%mm2", "%mm3", "%mm4");
-  }
+      asm volatile ("  movd      %0,%%mm2\n"
+                    "\tmovd      %1,%%mm3\n"
+                    "\tmovd   %%mm3,%0\n"
+                    "\tmovd   %%mm2,%1\n"                  
+                    : "+m" (*op.A), "+m" (*op.B)
+                    : /* empty */
+                    : "%mm1", "%mm2", "%mm3", "%mm4");
+    }
   
   asm("emms");
 }
diff --git a/app/composite/gimp-composite-sse2.h b/app/composite/gimp-composite-sse2.h
index 8310ee6272..d7fd192539 100644
--- a/app/composite/gimp-composite-sse2.h
+++ b/app/composite/gimp-composite-sse2.h
@@ -10,7 +10,11 @@ extern void gimp_composite_sse2_init (void);
 extern void gimp_composite_sse2_install (void);
 
 extern void gimp_composite_addition_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *);
+extern void gimp_composite_darken_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *);
 extern void gimp_composite_difference_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *);
+extern void gimp_composite_grain_extract_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *);
+extern void gimp_composite_lighten_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *);
 extern void gimp_composite_subtract_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *);
 extern void gimp_composite_swap_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *);
+extern void gimp_composite_dodge_rgba8_rgba8_rgba8_sse2 (GimpCompositeContext *);
 #endif
diff --git a/app/composite/gimp-composite-vis-test.c b/app/composite/gimp-composite-vis-test.c
index c0d15e9d8f..15a4eb932e 100644
--- a/app/composite/gimp-composite-vis-test.c
+++ b/app/composite/gimp-composite-vis-test.c
@@ -71,7 +71,7 @@ main (int argc, char *argv[])
   putenv ("GIMP_COMPOSITE=0x1");
 
   iterations = 1;
-  n_pixels = 1048577;
+  n_pixels = 163921;
 
   argv++, argc--;
   while (argc >= 2) {
diff --git a/app/composite/gimp-composite-x86.h b/app/composite/gimp-composite-x86.h
new file mode 100644
index 0000000000..9d9d179fe5
--- /dev/null
+++ b/app/composite/gimp-composite-x86.h
@@ -0,0 +1,208 @@
+#if __GNUC__ >= 3
+
+#define mmx_low_bytes_to_words(src,dst,zero) \
+         "\tmovq      %%"#src", %%"#dst"; " \
+         "\tpunpcklbw %%"#zero", %%"#dst"\n"
+
+#define mmx_high_bytes_to_words(src,dst,zero) \
+         "\tmovq      %%"#src", %%"#dst"; " \
+         "\tpunpckhbw %%"#zero", %%"#dst"\n"
+
+#define xmm_low_bytes_to_words(src,dst,zero) \
+         "\tmovdqu     %%"#src", %%"#dst"; " \
+         "\tpunpcklbw %%"#zero", %%"#dst"\n"
+
+#define xmm_high_bytes_to_words(src,dst,zero) \
+         "\tmovdqu     %%"#src", %%"#dst"; " \
+         "\tpunpckhbw %%"#zero", %%"#dst"\n"
+
+/* a = INT_MULT(a,b) */
+#define mmx_int_mult(a,b,w128) \
+                  "\tpmullw    %%"#b",    %%"#a"; " \
+                  "\tpaddw     %%"#w128", %%"#a"; " \
+                  "\tmovq      %%"#a",    %%"#b"; " \
+                  "\tpsrlw     $8,        %%"#b"; " \
+                  "\tpaddw     %%"#a",    %%"#b"; " \
+                  "\tpsrlw     $8,        %%"#b"\n"
+
+#define sse2_int_mult(a,b,w128) \
+                  "\tpmullw    %%"#b",    %%"#a"; " \
+                  "\tpaddw     %%"#w128", %%"#a"; " \
+                  "\tmovdqu    %%"#a",    %%"#b"; " \
+                  "\tpsrlw     $8,        %%"#b"; " \
+                  "\tpaddw     %%"#a",    %%"#b"; " \
+                  "\tpsrlw     $8,        %%"#b"\n"
+
+/*
+ * Double-word divide.  Adjusted for subsequent unsigned packing
+ * (high-order bit of each word is cleared)
+ * Clobbers eax, ecx edx
+ */
+#define pdivwX(dividend,divisor,quotient) "movd %%" #dividend ",%%eax; " \
+                                          "movd %%" #divisor  ",%%ecx; " \
+                                          "xorl %%edx,%%edx; "           \
+                                          "divw %%cx; "                  \
+                                          "roll $16, %%eax; "            \
+                                          "roll $16, %%ecx; "            \
+                                          "xorl %%edx,%%edx; "           \
+                                          "divw %%cx; "                  \
+                                          "btr $15, %%eax; "             \
+                                          "roll $16, %%eax; "            \
+                                          "btr $15, %%eax; "             \
+                                          "movd %%eax,%%" #quotient ";"
+
+
+
+/*
+ * Quadword divide.  No adjustment for subsequent unsigned packing
+ * (high-order bit of each word is left alone)
+ */
+#define pdivwqX(dividend,divisor,quotient) "movd   %%" #dividend ",%%eax; " \
+                                          "movd   %%" #divisor  ",%%ecx; " \
+                                          "xorl   %%edx,%%edx; "           \
+                                          "divw   %%cx; "                  \
+                                          "roll   $16, %%eax; "            \
+                                          "roll   $16, %%ecx; "            \
+                                          "xorl   %%edx,%%edx; "           \
+                                          "divw   %%cx; "                  \
+                                          "roll   $16, %%eax; "            \
+                                          "movd   %%eax,%%" #quotient "; " \
+                                          "psrlq $32,%%" #dividend ";"     \
+                                          "psrlq $32,%%" #divisor ";"      \
+                                          "movd   %%" #dividend ",%%eax; " \
+                                          "movd   %%" #divisor  ",%%ecx; " \
+                                          "xorl   %%edx,%%edx; "           \
+                                          "divw   %%cx; "                  \
+                                          "roll   $16, %%eax; "            \
+                                          "roll   $16, %%ecx; "            \
+                                          "xorl   %%edx,%%edx; "           \
+                                          "divw   %%cx; "                  \
+                                          "roll   $16, %%eax; "            \
+                                          "movd   %%eax,%%" #divisor ";"   \
+                                          "psllq  $32,%%" #divisor ";"     \
+                                          "por    %%" #divisor ",%%" #quotient ";"
+   
+/*
+ * Quadword divide.  Adjusted for subsequent unsigned packing
+ * (high-order bit of each word is cleared)
+ */
+#define pdivwuqX(dividend,divisor,quotient) \
+                                          pdivwX(dividend,divisor,quotient) \
+                                            "psrlq  $32,%%" #dividend ";"   \
+                                            "psrlq  $32,%%" #divisor ";"    \
+                                          pdivwX(dividend,divisor,quotient) \
+                                          "movd   %%eax,%%" #divisor ";"    \
+                                            "psllq  $32,%%" #divisor ";"    \
+                                            "por    %%" #divisor ",%%" #quotient ";"
+
+#define xmm_pdivwqX(dividend,divisor,quotient,scratch) "movd   %%" #dividend ",%%eax; " \
+                                                       "movd   %%" #divisor  ",%%ecx; " \
+                                                       "xorl   %%edx,%%edx; "           \
+                                                       "divw   %%cx; "                  \
+                                                       "roll   $16, %%eax; "            \
+                                                       "roll   $16, %%ecx; "            \
+                                                       "xorl   %%edx,%%edx; "           \
+                                                       "divw   %%cx; "                  \
+                                                       "roll   $16, %%eax; "            \
+                                                       "movd   %%eax,%%" #quotient "; " \
+                                                       "psrlq $32,%%" #divisor ";"      \
+                                                       "psrlq $32,%%" #dividend ";"     \
+                                                       "movd   %%" #dividend ",%%eax; " \
+                                                       "movd   %%" #divisor  ",%%ecx; " \
+                                                       "xorl   %%edx,%%edx; "           \
+                                                       "divw   %%cx; "                  \
+                                                       "roll   $16, %%eax; "            \
+                                                       "roll   $16, %%ecx; "            \
+                                                       "xorl   %%edx,%%edx; "           \
+                                                       "divw   %%cx; "                  \
+                                                       "roll   $16, %%eax; "            \
+                                                       "movd   %%eax,%%" #scratch ";"   \
+                                                       "psllq  $32,%%" #scratch ";"     \
+                                                       "psrlq $32,%%" #divisor ";"      \
+                                                       "psrlq $32,%%" #dividend ";"     \
+                                                       "movd   %%" #dividend ",%%eax; " \
+                                                       "movd   %%" #divisor  ",%%ecx; " \
+                                                       "xorl   %%edx,%%edx; "           \
+                                                       "divw   %%cx; "                  \
+                                                       "roll   $16, %%eax; "            \
+                                                       "roll   $16, %%ecx; "            \
+                                                       "xorl   %%edx,%%edx; "           \
+                                                       "divw   %%cx; "                  \
+                                                       "roll   $16, %%eax; "            \
+                                                       "movd   %%eax,%%" #scratch ";"   \
+                                                       "psllq  $64,%%" #scratch ";"     \
+                                                       "psrlq $32,%%" #divisor ";"      \
+                                                       "psrlq $32,%%" #dividend ";"     \
+                                                       "movd   %%" #dividend ",%%eax; " \
+                                                       "movd   %%" #divisor  ",%%ecx; " \
+                                                       "xorl   %%edx,%%edx; "           \
+                                                       "divw   %%cx; "                  \
+                                                       "roll   $16, %%eax; "            \
+                                                       "roll   $16, %%ecx; "            \
+                                                       "xorl   %%edx,%%edx; "           \
+                                                       "divw   %%cx; "                  \
+                                                       "roll   $16, %%eax; "            \
+                                                       "movd   %%eax,%%" #scratch ";"   \
+                                                       "psllq  $96,%%" #scratch ";"     \
+                                                       "por    %%" #scratch ",%%" #quotient ";"
+   
+#define xmm_pdivwX(dividend,divisor,quotient) "movd %%" #dividend ",%%eax; " \
+                                              "movd %%" #divisor  ",%%ecx; " \
+                                              "xorl %%edx,%%edx; "           \
+                                              "divw %%cx; "                  \
+                                              "roll $16, %%eax; "            \
+                                              "roll $16, %%ecx; "            \
+                                              "xorl %%edx,%%edx; "           \
+                                              "divw %%cx; "                  \
+                                              "btr $15, %%eax; "             \
+                                              "roll $16, %%eax; "            \
+                                              "btr $15, %%eax; "             \
+                                              "movd %%eax,%%" #quotient ";"
+
+#define xmm_pdivwuqX(dividend,divisor,quotient,scratch) \
+                                          xmm_pdivwX(dividend,divisor,scratch)      \
+                                            "movd   %%"#scratch ",%%"#quotient  ";" \
+                                          "psrlq  $32,%%"#dividend              ";" \
+                                          "psrlq  $32,%%"#divisor               ";" \
+                                          xmm_pdivwX(dividend,divisor,scratch)      \
+                                            "psllq  $32,%%"#scratch             ";" \
+                                            "por    %%"#scratch ",%%"#quotient  ";" \
+                                          "psrlq  $32,%%"#dividend              ";" \
+                                          "psrlq  $32,%%"#divisor               ";" \
+                                          xmm_pdivwX(dividend,divisor,scratch)      \
+                                            "psllq  $64,%%"#scratch             ";" \
+                                            "por    %%"#scratch ",%%"#quotient  ";" \
+                                          "psrlq  $32,%%"#dividend              ";" \
+                                          "psrlq  $32,%%"#divisor               ";" \
+                                          xmm_pdivwX(dividend,divisor,scratch)      \
+																																										  "psllq  $96,%%"#scratch             ";" \
+                                            "por    %%"#scratch ",%%"#quotient
+
+/* equivalent to the INT_MULT() macro in gimp-composite-generic.c */
+/*
+ * opr2 = INT_MULT(opr1, opr2, t)
+ *
+ * Operates across quad-words using x86 word (16bit) value.
+ * Result is left in opr2
+ *
+ * opr1 = opr1 * opr2 + w128
+ * opr2 = opr1
+ * opr2 = ((opr2 >> 8) + opr1) >> 8
+ */
+#define pmulwX(opr1,opr2,w128) \
+                  "\tpmullw    %%"#opr2", %%"#opr1"; " \
+                  "\tpaddw     %%"#w128", %%"#opr1"; " \
+                  "\tmovq      %%"#opr1", %%"#opr2"; " \
+                  "\tpsrlw     $8,        %%"#opr2"; " \
+                  "\tpaddw     %%"#opr1", %%"#opr2"; " \
+                  "\tpsrlw     $8,        %%"#opr2"\n"
+ 
+#define xmm_pmulwX(opr1,opr2,w128) \
+                  "\tpmullw    %%"#opr2", %%"#opr1"; " \
+                  "\tpaddw     %%"#w128", %%"#opr1"; " \
+                  "\tmovdqu    %%"#opr1", %%"#opr2"; " \
+                  "\tpsrlw     $8,        %%"#opr2"; " \
+                  "\tpaddw     %%"#opr1", %%"#opr2"; " \
+                  "\tpsrlw     $8,        %%"#opr2"\n"
+ 
+#endif
diff --git a/app/composite/make-installer.py b/app/composite/make-installer.py
index 4f5cd8c001..65300165f1 100755
--- a/app/composite/make-installer.py
+++ b/app/composite/make-installer.py
@@ -317,6 +317,8 @@ def gimp_composite_regression(fpout, function_tables, options):
   #pp.pprint(function_tables)
 
   generic_table = function_tables
+
+  composite_modes.sort();
   
   for mode in composite_modes:
     for A in filter(lambda pf: pf != "GIMP_PIXELFORMAT_ANY", pixel_format):
@@ -368,6 +370,8 @@ def gimp_composite_regression(fpout, function_tables, options):
             print >>fpout, '  ft0 = gimp_composite_regression_time_function (iterations, %s, &generic_ctx);' % ("gimp_composite_dispatch")
             print >>fpout, '  ft1 = gimp_composite_regression_time_function (iterations, %s, &special_ctx);' % (generic_table[key][0])
             print >>fpout, '  if (gimp_composite_regression_compare_contexts ("%s", &generic_ctx, &special_ctx)) {' % (mode_name(mode))
+            
+            print >>fpout, '    printf("%s failed\\n");' % (mode_name(mode))
             print >>fpout, '    return (1);'
             print >>fpout, '  }'
             print >>fpout, '  gimp_composite_regression_timer_report ("%s", ft0, ft1);' % (mode_name(mode))
@@ -518,7 +522,7 @@ op.add_option('-t', '--test', action='store_true',                 dest='test',
               help='generate regression testing code')
 op.add_option('-i', '--iterations', action='store', type='int',    dest='iterations', default=1,
               help='number of iterations in regression tests')
-op.add_option('-n', '--n-pixels', action='store',   type="int",    dest='n_pixels',   default=1024*1024+1,
+op.add_option('-n', '--n-pixels', action='store',   type="int",    dest='n_pixels',   default=64*2049+16*2049+1,
               help='number of pixels in each regression test iteration')
 op.add_option('-r', '--requires', action='append',  type='string', dest='requires',   default=[],
               help='cpp #if conditionals')