Updating to trunk

(bzr r8254.1.54)

6 files changed, 66 insertions, 690 deletions

diff --git a/src/libnr/Makefile_insert b/src/libnr/Makefile_insert
index 4b19028f9..8dd3c46e3 100644
--- a/src/libnr/Makefile_insert
+++ b/src/libnr/Makefile_insert
@@ -1,13 +1,5 @@
 ## Makefile.am fragment sourced by src/Makefile.am.
 
-if USE_MMX
-libnr_mmx_sources =	\
-	libnr/have_mmx.S	\
-	libnr/nr_mmx_R8G8B8A8_P_EMPTY_A8_RGBAP.S	\
-	libnr/nr_mmx_R8G8B8A8_P_R8G8B8A8_P_A8_RGBAP.S	\
-	libnr/nr_mmx_R8G8B8_R8G8B8_R8G8B8A8_P.S
-endif
-
 ink_common_sources +=	\
 	libnr/in-svg-plane.h	\
 	libnr/nr-blit.cpp	\
diff --git a/src/libnr/have_mmx.S b/src/libnr/have_mmx.S
deleted file mode 100644
index d6428191e..000000000
--- a/src/libnr/have_mmx.S
+++ /dev/null
@@ -1,47 +0,0 @@
-	.file "have_mmx.S"
-
-# Ensure Inkscape is execshield protected
-	.section .note.GNU-stack
-	.previous
-	
-	.version "01.01"
-gcc2_compiled.:
-.text
-	.align 16
-.globl nr_have_mmx
-	.type	 nr_have_mmx,@function
-
-nr_have_mmx:
-	push	%ebx
-
-# Check if bit 21 in flags word is writeable
-
-	pushfl	
-	popl	%eax
-	movl	%eax,%ebx
-	xorl	$0x00200000, %eax
-	pushl   %eax
-	popfl
-	pushfl
-	popl	%eax
-
-	cmpl	%eax, %ebx
-
-	je .notfound
-
-# OK, we have CPUID
-
-	movl	$1, %eax
-	cpuid
-	
-	test	$0x00800000, %edx
-	jz	.notfound
-
-	movl	$1, %eax
-	jmp	.out
-
-.notfound:
-	movl  	$0, %eax
-.out:	
-	popl	%ebx
-	ret
diff --git a/src/libnr/nr-compose-transform.cpp b/src/libnr/nr-compose-transform.cpp
index 6e03faf2f..e7c286569 100644
--- a/src/libnr/nr-compose-transform.cpp
+++ b/src/libnr/nr-compose-transform.cpp
@@ -66,18 +66,25 @@ nr_R8G8B8A8_N_R8G8B8A8_N_R8G8B8A8_N_TRANSFORM (unsigned char *px, int w, int h,
 
 	if (alpha == 0) return;
 
+    // Both alpha and color components are stored temporarily with a range of [0,255^2], so more supersampling and we get an overflow
+    if (xd+yd>16) {
+        xd = 8;
+        yd = 8;
+    }
+
 	xsize = (1 << xd);
 	ysize = (1 << yd);
 	size = xsize * ysize;
 	dbits = xd + yd;
+    unsigned int rounding_fix = size/2;
 
 	/* Set up fixed point matrix */
-	FFs_x_x = (long) (d2s[0] * (1 << FBITS) + 0.5);
-	FFs_x_y = (long) (d2s[1] * (1 << FBITS) + 0.5);
-	FFs_y_x = (long) (d2s[2] * (1 << FBITS) + 0.5);
-	FFs_y_y = (long) (d2s[3] * (1 << FBITS) + 0.5);
-	FFs__x = (long) (d2s[4] * (1 << FBITS) + 0.5);
-	FFs__y = (long) (d2s[5] * (1 << FBITS) + 0.5);
+	FFs_x_x = (long) floor(d2s[0] * (1 << FBITS) + 0.5);
+	FFs_x_y = (long) floor(d2s[1] * (1 << FBITS) + 0.5);
+	FFs_y_x = (long) floor(d2s[2] * (1 << FBITS) + 0.5);
+	FFs_y_y = (long) floor(d2s[3] * (1 << FBITS) + 0.5);
+	FFs__x = (long) floor(d2s[4] * (1 << FBITS) + 0.5);
+	FFs__y = (long) floor(d2s[5] * (1 << FBITS) + 0.5);
 
 	FFs_x_x_S = FFs_x_x >> xd;
 	FFs_x_y_S = FFs_x_y >> xd;
@@ -114,35 +121,40 @@ nr_R8G8B8A8_N_R8G8B8A8_N_R8G8B8A8_N_TRANSFORM (unsigned char *px, int w, int h,
 					sy = (FFsy + FF_sy_S[i]) >> FBITS;
 					if ((sy >= 0) && (sy < sh)) {
 						const unsigned char *s;
-						unsigned int ca;
 						s = spx + sy * srs + sx * 4;
-						ca = NR_PREMUL_112 (s[3], alpha);
-						r += NR_PREMUL_121 (s[0], ca);
-						g += NR_PREMUL_121 (s[1], ca);
-						b += NR_PREMUL_121 (s[2], ca);
-						a += NR_NORMALIZE_21(ca);
+						r += NR_PREMUL_112 (s[0], s[3]);
+						g += NR_PREMUL_112 (s[1], s[3]);
+						b += NR_PREMUL_112 (s[2], s[3]);
+						a += s[3];
 					}
 				}
 			}
-			a >>= dbits;
+			a = (a*alpha + rounding_fix) >> dbits;
+            // Compare to nr_R8G8B8A8_N_R8G8B8A8_N_R8G8B8A8_P
 			if (a != 0) {
-				r = r >> dbits;
-				g = g >> dbits;
-				b = b >> dbits;
-				if (a == 255) {
-					/* Transparent BG, premul src */
-					d[0] = r;
-					d[1] = g;
-					d[2] = b;
-					d[3] = a;
+				r = (r + rounding_fix) >> dbits;
+				g = (g + rounding_fix) >> dbits;
+				b = (b + rounding_fix) >> dbits;
+                if (a == 255*255) {
+					/* Full coverage, demul src */
+					d[0] = NR_NORMALIZE_21(r);
+					d[1] = NR_NORMALIZE_21(g);
+					d[2] = NR_NORMALIZE_21(b);
+					d[3] = NR_NORMALIZE_21(a);
+                } else if (d[3] == 0) {
+                    /* Only foreground, demul src */
+                    d[0] = NR_DEMUL_221(r,a);
+                    d[1] = NR_DEMUL_221(g,a);
+                    d[2] = NR_DEMUL_221(b,a);
+                    d[3] = NR_NORMALIZE_21(a);
 				} else {
 					unsigned int ca;
 					/* Full composition */
-					ca = NR_COMPOSEA_112(a, d[3]);
-					d[0] = NR_COMPOSENNN_111121 (r, a, d[0], d[3], ca);
-					d[1] = NR_COMPOSENNN_111121 (g, a, d[1], d[3], ca);
-					d[2] = NR_COMPOSENNN_111121 (b, a, d[2], d[3], ca);
-					d[3] = NR_NORMALIZE_21(ca);
+					ca = NR_COMPOSEA_213(a, d[3]);
+					d[0] = NR_COMPOSEPNN_221131 (r, a, d[0], d[3], ca);
+					d[1] = NR_COMPOSEPNN_221131 (g, a, d[1], d[3], ca);
+					d[2] = NR_COMPOSEPNN_221131 (b, a, d[2], d[3], ca);
+					d[3] = NR_NORMALIZE_31(ca);
 				}
 			}
 			/* Advance pointers */
@@ -227,9 +239,7 @@ nr_R8G8B8A8_P_R8G8B8A8_P_R8G8B8A8_N_TRANSFORM_n (unsigned char *px, int w, int h
 	int x, y;
 
 	size = (1 << dbits);
-    unsigned alpha_rounding_fix = size * 255;
-    unsigned rgb_rounding_fix = size * (255 * 256);
-    if (alpha > 127) ++alpha;
+    unsigned int rounding_fix = size/2;
 
 	d0 = px;
 	FFsx0 = FFd2s[4];
@@ -252,32 +262,30 @@ nr_R8G8B8A8_P_R8G8B8A8_P_R8G8B8A8_N_TRANSFORM_n (unsigned char *px, int w, int h
 					sy = (long (FFsy >> (FBITS_HP - FBITS)) + FF_S[2 * i + 1]) >> FBITS;
 					if ((sy >= 0) && (sy < sh)) {
 						const unsigned char *s;
-						unsigned int ca;
 						s = spx + sy * srs + sx * 4;
-						ca = NR_PREMUL_112(s[3], alpha);
-						r += NR_PREMUL_123(s[0], ca);
-						g += NR_PREMUL_123(s[1], ca);
-						b += NR_PREMUL_123(s[2], ca);
-						a += ca;
+						r += NR_PREMUL_112(s[0], s[3]);
+						g += NR_PREMUL_112(s[1], s[3]);
+						b += NR_PREMUL_112(s[2], s[3]);
+						a += s[3];
 					}
 				}
 			}
-			a = (a + alpha_rounding_fix) >> (8 + dbits);
+			a = (a*alpha + rounding_fix) >> dbits;
 			if (a != 0) {
-				r = (r + rgb_rounding_fix) >> (16 + dbits);
-				g = (g + rgb_rounding_fix) >> (16 + dbits);
-				b = (b + rgb_rounding_fix) >> (16 + dbits);
-				if ((a == 255) || (d[3] == 0)) {
+				r = (r + rounding_fix) >> dbits;
+				g = (g + rounding_fix) >> dbits;
+				b = (b + rounding_fix) >> dbits;
+				if ((a == 255*255) || (d[3] == 0)) {
 					/* Transparent BG, premul src */
-					d[0] = r;
-					d[1] = g;
-					d[2] = b;
-					d[3] = a;
+					d[0] = NR_NORMALIZE_21(r);
+					d[1] = NR_NORMALIZE_21(g);
+					d[2] = NR_NORMALIZE_21(b);
+					d[3] = NR_NORMALIZE_21(a);
 				} else {
-					d[0] = NR_COMPOSEPPP_1111 (r, a, d[0]);
-					d[1] = NR_COMPOSEPPP_1111 (g, a, d[1]);
-					d[2] = NR_COMPOSEPPP_1111 (b, a, d[2]);
-					d[3] = NR_COMPOSEA_111(a, d[3]);
+					d[0] = NR_COMPOSEPPP_2211 (r, a, d[0]);
+					d[1] = NR_COMPOSEPPP_2211 (g, a, d[1]);
+					d[2] = NR_COMPOSEPPP_2211 (b, a, d[2]);
+					d[3] = NR_COMPOSEA_211(a, d[3]);
 				}
 			}
 			/* Advance pointers */
@@ -302,11 +310,17 @@ void nr_R8G8B8A8_P_R8G8B8A8_P_R8G8B8A8_N_TRANSFORM (unsigned char *px, int w, in
 
 	if (alpha == 0) return;
 
-	dbits = xd + yd;
+    // Both alpha and color components are stored temporarily with a range of [0,255^2], so more supersampling and we get an overflow
+    if (xd+yd>16) {
+        xd = 8;
+        yd = 8;
+    }
+
+    dbits = xd + yd;
 
 	for (i = 0; i < 6; i++) {
-		FFd2s[i] = (long) (d2s[i] * (1 << FBITS) + 0.5);
-		FFd2s_HP[i] = (long long) (d2s[i] * (1 << FBITS_HP) + 0.5);;
+		FFd2s[i] = (long) floor(d2s[i] * (1 << FBITS) + 0.5);
+		FFd2s_HP[i] = (long long) floor(d2s[i] * (1 << FBITS_HP) + 0.5);;
 	}
 
 	if (dbits == 0) {
diff --git a/src/libnr/nr_mmx_R8G8B8A8_P_EMPTY_A8_RGBAP.S b/src/libnr/nr_mmx_R8G8B8A8_P_EMPTY_A8_RGBAP.S
deleted file mode 100644
index db2cbec5a..000000000
--- a/src/libnr/nr_mmx_R8G8B8A8_P_EMPTY_A8_RGBAP.S
+++ /dev/null
@@ -1,125 +0,0 @@
-	.file	"nr-compose.c"
-	
-# Ensure Inkscape is execshield protected
-	.section .note.GNU-stack
-	.previous
-	
-	.text
-	.align 2
-.globl nr_mmx_R8G8B8A8_P_EMPTY_A8_RGBAP
-	.type	nr_mmx_R8G8B8A8_P_EMPTY_A8_RGBAP,@function
-
-/*
- * This code is in public domain
- *
- * c	 32(%ebp)
- * srs	 28(%ebp)
- * spx	 24(%ebp)
- * rs	 20(%ebp)
- * h	 16(%ebp)
- * w	 12(%ebp)
- * px	 8(%ebp)
- * r	-8(%ebp)
- * g	-12(%ebp)
- * b	-16(%ebp)
- * a	-20(%ebp)
- * s	-24(%ebp) -> %esi
- * d	-28(%ebp) -> %edi
- * x	-32(%ebp) -> %ebx
- * y	-36(%ebp)
- * ca	-40(%ebp)
- *
- * mm0 Fg
- * mm1 FgA
- * mm2 FgPre
- * mm3
- * mm4
- * mm5
- * mm6 128
- * mm7 0
- *
-*/
-
-nr_mmx_R8G8B8A8_P_EMPTY_A8_RGBAP:
-	pushl	%ebp
-	movl	%esp, %ebp
-	pushl	%ebx
-	subl	$36, %esp
-	pushl	%edi
-	pushl	%esi
-
-/* Load %mm7 with [0 0 0 0] */
-	movl	$0, %eax
-	movd	%eax, %mm7
-
-/* Load %mm6 with [128 128 128 128] */
-	movl	$0x80808080, %eax
-	movd	%eax, %mm6
-	punpcklbw %mm7, %mm6
-
-/* FgC -> %mm0 */
-	movl	32(%ebp), %eax
-	movd	(%eax), %mm0
-	punpcklbw %mm7, %mm0
-
-/* for (y = ...) */
-	movl    16(%ebp), %ecx
-.fory:
-
-/* d = px */
-/* s = spx */
-	movl	8(%ebp), %edi
-	movl	24(%ebp), %esi
-
-/* for (x = ...) */
-	movl	12(%ebp), %ebx
-.forx:
-
-/* [m m m m] -> %mm1 */
-	movzbl	(%esi), %eax
-	testb	$0xff, %al
-	jz	.clip
-	movd	%eax, %mm1
-	punpcklwd %mm1, %mm1
-	punpckldq %mm1, %mm1
-
-/* Fg -> mm2 */
-	movq	%mm0, %mm2
-	pmullw	%mm1, %mm2
-	paddw	%mm6, %mm2
-	movq	%mm2, %mm3
-	psrlw	$8, %mm3
-	paddw	%mm3, %mm2
-	psrlw	$8, %mm2
-
-/* Store pixel */
-	packuswb %mm2, %mm2
-	movd	%mm2, (%edi)
-
-.clip:
-	addl	$4, %edi
-	incl	%esi
-
-	decl	%ebx
-	jnz	.forx
-
-	movl	20(%ebp), %eax
-	addl	%eax, 8(%ebp)
-	movl	28(%ebp), %eax
-	addl	%eax, 24(%ebp)
-
-	decl	%ecx
-	jnz	.fory
-
-.exit:
-	emms
-	popl	%esi
-	popl	%edi
-	addl	$36, %esp
-	popl	%ebx
-	popl	%ebp
-	ret
-
-.Lfe1:
-	.size	nr_mmx_R8G8B8A8_P_EMPTY_A8_RGBAP,.Lfe1-nr_mmx_R8G8B8A8_P_EMPTY_A8_RGBAP
-	.ident	"GCC: (GNU) 3.2"
diff --git a/src/libnr/nr_mmx_R8G8B8A8_P_R8G8B8A8_P_A8_RGBAP.S b/src/libnr/nr_mmx_R8G8B8A8_P_R8G8B8A8_P_A8_RGBAP.S
deleted file mode 100644
index fe1d9be57..000000000
--- a/src/libnr/nr_mmx_R8G8B8A8_P_R8G8B8A8_P_A8_RGBAP.S
+++ /dev/null
@@ -1,231 +0,0 @@
-	.file	"nr-compose.c"
-
-# Ensure Inkscape is execshield protected
-	.section .note.GNU-stack
-	.previous
-	
-	.text
-	.align 2
-.globl nr_mmx_R8G8B8A8_P_R8G8B8A8_P_A8_RGBAP
-	.type	nr_mmx_R8G8B8A8_P_R8G8B8A8_P_A8_RGBAP,@function
-
-/*
- * This code is in public domain
- *
- * c	 32(%ebp)
- * srs	 28(%ebp)
- * spx	 24(%ebp)
- * rs	 20(%ebp)
- * h	 16(%ebp)
- * w	 12(%ebp)
- * px	 8(%ebp)
- * r	-8(%ebp)
- * g	-12(%ebp)
- * b	-16(%ebp)
- * a	-20(%ebp)
- * s	-24(%ebp) -> %esi
- * d	-28(%ebp) -> %edi
- * x	-32(%ebp) -> %ebx
- * y	-36(%ebp)
- * ca	-40(%ebp)
- *
- * mm0 Fg
- * mm1 MMMM
- * mm2 FgM
- * mm3
- * mm4
- * mm5 255
- * mm6 128
- * mm7 0
- *
-*/
-
-nr_mmx_R8G8B8A8_P_R8G8B8A8_P_A8_RGBAP:
-	pushl	%ebp
-	movl	%esp, %ebp
-	pushl	%ebx
-	subl	$36, %esp
-	pushl	%edi
-	pushl	%esi
-
-/* Load %mm7 with [0 0 0 0] */
-	movl	$0, %eax
-	movd	%eax, %mm7
-
-/* Load %mm6 with [128 128 128 128] */
-	movl	$0x80808080, %eax
-	movd	%eax, %mm6
-	punpcklbw %mm7, %mm6
-
-/* Load %mm5 with [255 255 255 255] */
-	movl	$0xffffffff, %eax
-	movd	%eax, %mm5
-	punpcklbw %mm7, %mm5
-
-/* FgC -> %mm0 */
-	movl	32(%ebp), %eax
-	movd	(%eax), %mm0
-	punpcklbw %mm7, %mm0
-
-/* Check full opacity */
-	cmpb	$0xff, %al
-	jz	.opaque
-
-/* for (y = ...) */
-	movl    16(%ebp), %ecx
-.fory:
-
-/* d = px */
-/* s = spx */
-	movl	8(%ebp), %edi
-	movl	24(%ebp), %esi
-
-/* for (x = ...) */
-	movl	12(%ebp), %ebx
-.forx:
-
-/* [m m m m] -> %mm1 */
-	movzbl	(%esi), %eax
-	testb	$0xff, %al
-	jz	.clip
-	movd	%eax, %mm1
-	punpcklwd %mm1, %mm1
-	punpckldq %mm1, %mm1
-
-/* Fg -> mm2 */
-	movq	%mm0, %mm2
-	pmullw	%mm1, %mm2
-	paddw	%mm6, %mm2
-	movq	%mm2, %mm3
-	psrlw	$8, %mm3
-	paddw	%mm3, %mm2
-	psrlw	$8, %mm2
-
-/* [255 - FgA] -> mm1 */
-	movq	%mm2, %mm1
-	punpckhwd %mm1, %mm1
-	punpckhdq %mm1, %mm1
-	pxor	%mm5, %mm1
-
-/* Bg -> mm3 */
-	movd	(%edi), %mm3
-	punpcklbw %mm7, %mm3
-
-/* Fg + ((255 - FgA) * Bg) / 255 */
-	pmullw	%mm1, %mm3
-	paddw	%mm6, %mm3
-	movq	%mm3, %mm4
-	psrlw	$8, %mm4
-	paddw	%mm4, %mm3
-	psrlw	$8, %mm3
-	paddw	%mm2, %mm3
-
-/* Store pixel */
-	packuswb %mm3, %mm3
-	movd	%mm3, (%edi)
-
-.clip:
-	addl	$4, %edi
-	incl	%esi
-
-	decl	%ebx
-	jnz	.forx
-
-	movl	20(%ebp), %eax
-	addl	%eax, 8(%ebp)
-	movl	28(%ebp), %eax
-	addl	%eax, 24(%ebp)
-
-	decl	%ecx
-	jnz	.fory
-
-.exit:
-	emms
-	popl	%esi
-	popl	%edi
-	addl	$36, %esp
-	popl	%ebx
-	popl	%ebp
-	ret
-
-.opaque:
-/* for (y = ...) */
-	movl    16(%ebp), %ecx
-.o_fory:
-
-/* d = px */
-/* s = spx */
-	movl	8(%ebp), %edi
-	movl	24(%ebp), %esi
-
-/* for (x = ...) */
-	movl	12(%ebp), %ebx
-.o_forx:
-
-/* [m m m m] -> %mm1 */
-	movzbl	(%esi), %eax
-	testb	$0xff, %al
-	jz	.o_clip
-	cmpb	$0xff, %al
-	jz	.o_full
-	movd	%eax, %mm1
-	punpcklwd %mm1, %mm1
-	punpckldq %mm1, %mm1
-
-/* Fg -> mm2 */
-	movq	%mm0, %mm2
-	pmullw	%mm1, %mm2
-	paddw	%mm6, %mm2
-	movq	%mm2, %mm3
-	psrlw	$8, %mm3
-	paddw	%mm3, %mm2
-	psrlw	$8, %mm2
-
-/* [255 - FgA] -> mm1 */
-	movq	%mm2, %mm1
-	punpckhwd %mm1, %mm1
-	punpckhdq %mm1, %mm1
-	pxor	%mm5, %mm1
-
-/* Bg -> mm3 */
-	movd	(%edi), %mm3
-	punpcklbw %mm7, %mm3
-
-/* Fg + ((255 - FgA) * Bg) / 255 */
-	pmullw	%mm1, %mm3
-	paddw	%mm6, %mm3
-	movq	%mm3, %mm4
-	psrlw	$8, %mm4
-	paddw	%mm4, %mm3
-	psrlw	$8, %mm3
-	paddw	%mm2, %mm3
-
-	jmp	.o_store
-
-.o_full:
-	movq	%mm0, %mm3
-
-.o_store:	
-/* Store pixel */
-	packuswb %mm3, %mm3
-	movd	%mm3, (%edi)
-
-.o_clip:
-	addl	$4, %edi
-	incl	%esi
-
-	decl	%ebx
-	jnz	.o_forx
-
-	movl	20(%ebp), %eax
-	addl	%eax, 8(%ebp)
-	movl	28(%ebp), %eax
-	addl	%eax, 24(%ebp)
-
-	decl	%ecx
-	jnz	.o_fory
-	jmp	.exit
-
-.Lfe1:
-	.size	nr_mmx_R8G8B8A8_P_R8G8B8A8_P_A8_RGBAP,.Lfe1-nr_mmx_R8G8B8A8_P_R8G8B8A8_P_A8_RGBAP
-	.ident	"GCC: (GNU) 3.2"
diff --git a/src/libnr/nr_mmx_R8G8B8_R8G8B8_R8G8B8A8_P.S b/src/libnr/nr_mmx_R8G8B8_R8G8B8_R8G8B8A8_P.S
deleted file mode 100644
index 37261e572..000000000
--- a/src/libnr/nr_mmx_R8G8B8_R8G8B8_R8G8B8A8_P.S
+++ /dev/null
@@ -1,227 +0,0 @@
-	.file	"nr-compose.c"
-
-# Ensure Inkscape is execshield protected
-	.section .note.GNU-stack
-	.previous
-	
-	.text
-	.align 2
-.globl nr_mmx_R8G8B8_R8G8B8_R8G8B8A8_P
-	.type	nr_mmx_R8G8B8_R8G8B8_R8G8B8A8_P,@function
-
-/*
- * This code is in public domain
- *
- * alpha 32(%ebp)
- * srs	 28(%ebp)
- * spx	 24(%ebp)
- * rs	 20(%ebp)
- * h	 16(%ebp)
- * w	 12(%ebp)
- * px	 8(%ebp)
- * r	-8(%ebp)
- * g	-12(%ebp)
- * b	-16(%ebp)
- * a	-20(%ebp)
- * s	-24(%ebp) -> %esi
- * d	-28(%ebp) -> %edi
- * x	-32(%ebp) -> %ebx
- * y	-36(%ebp)
- * ca	-40(%ebp)
- *
- * mm0 A
- * mm1 FgA
- * mm2 FgPre
- * mm3
- * mm4
- * mm5 255
- * mm6 128
- * mm7 0
- *
-*/
-
-nr_mmx_R8G8B8_R8G8B8_R8G8B8A8_P:
-	pushl	%ebp
-	movl	%esp, %ebp
-	pushl	%ebx
-	subl	$36, %esp
-	pushl	%edi
-	pushl	%esi
-
-/* Load %mm7 with [0 0 0 0] */
-	movl	$0, %eax
-	movd	%eax, %mm7
-
-/* Load %mm6 with [128 128 128 128] */
-	movl	$0x80808080, %eax
-	movd	%eax, %mm6
-	punpcklbw %mm7, %mm6
-
-/* Load %mm5 with [255 255 255 255] */
-	movl	$0xffffffff, %eax
-	movd	%eax, %mm5
-	punpcklbw %mm7, %mm5
-
-/* Load %mm0 with [a a a a] */
-/* Check full opacity */
-	movzbl	32(%ebp), %eax
-	cmpb	$0xff, %al
-	jz	.opaque
-	movd	%eax, %mm0
-	punpcklwd %mm0, %mm0
-	punpckldq %mm0, %mm0
-
-/* for (y = ...) */
-	movl    16(%ebp), %ecx
-.fory:
-
-/* d = px */
-/* s = spx */
-	movl	8(%ebp), %edi
-	movl	24(%ebp), %esi
-
-/* for (x = ...) */
-	movl	12(%ebp), %ebx
-.forx:
-
-/* Fg -> %mm1 */
-/* fixme: Do we have to bother about alignment here? (Lauris) */
-	movl	(%esi), %eax
-	testl	$0xff000000, %eax
-	jz	.clip
-	movd	%eax, %mm1
-	punpcklbw %mm7, %mm1
-
-/* [Fg * a] -> mm1 */
-	pmullw	%mm0, %mm1
-	paddw	%mm6, %mm1
-	movq	%mm1, %mm2
-	psrlw	$8, %mm2
-	paddw	%mm2, %mm1
-	psrlw	$8, %mm1
-
-/* [255 - FgA] -> mm2 */
-	movq	%mm1, %mm2
-	punpckhwd %mm2, %mm2
-	punpckhdq %mm2, %mm2
-	pxor	%mm5, %mm2
-
-/* Bg -> mm3 */
-	movd	(%edi), %mm3
-	punpcklbw %mm7, %mm3
-
-/* Fg + ((255 - FgA) * Bg) / 255 */
-	pmullw	%mm2, %mm3
-	paddw	%mm6, %mm3
-	movq	%mm3, %mm4
-	psrlw	$8, %mm4
-	paddw	%mm4, %mm3
-	psrlw	$8, %mm3
-	paddw	%mm1, %mm3
-
-/* Store pixel */
-	packuswb %mm3, %mm3
-	movd	%mm3, %eax
-	movb	%al, 0(%edi)
-	shrl	$8, %eax
-	movb	%al, 1(%edi)
-	shrl	$8, %eax
-	movb	%al, 2(%edi)
-
-.clip:
-	addl	$3, %edi
-	addl	$4, %esi
-
-	decl	%ebx
-	jnz	.forx
-
-	movl	20(%ebp), %eax
-	addl	%eax, 8(%ebp)
-	movl	28(%ebp), %eax
-	addl	%eax, 24(%ebp)
-
-	decl	%ecx
-	jnz	.fory
-
-.exit:
-	emms
-	popl	%esi
-	popl	%edi
-	addl	$36, %esp
-	popl	%ebx
-	popl	%ebp
-	ret
-
-.opaque:
-/* for (y = ...) */
-	movl    16(%ebp), %ecx
-.o_fory:
-
-/* d = px */
-/* s = spx */
-	movl	8(%ebp), %edi
-	movl	24(%ebp), %esi
-
-/* for (x = ...) */
-	movl	12(%ebp), %ebx
-.o_forx:
-
-/* Fg -> %mm1 */
-/* fixme: Do we have to bother about alignment here? (Lauris) */
-	movl	(%esi), %eax
-	testl	$0xff000000, %eax
-	jz	.o_clip
-	cmpl	$0xff000000, %eax
-	jnb	.o_store
-	movd	%eax, %mm1
-	punpcklbw %mm7, %mm1
-
-/* [255 - FgA] -> mm2 */
-	movq	%mm1, %mm2
-	punpckhwd %mm2, %mm2
-	punpckhdq %mm2, %mm2
-	pxor	%mm5, %mm2
-
-/* Bg -> mm3 */
-	movd	(%edi), %mm3
-	punpcklbw %mm7, %mm3
-
-/* Fg + ((255 - FgA) * Bg) / 255 */
-	pmullw	%mm2, %mm3
-	paddw	%mm6, %mm3
-	movq	%mm3, %mm4
-	psrlw	$8, %mm4
-	paddw	%mm4, %mm3
-	psrlw	$8, %mm3
-	paddw	%mm1, %mm3
-
-/* Store pixel */
-	packuswb %mm3, %mm3
-	movd	%mm3, %eax
-.o_store:
-	movb	%al, 0(%edi)
-	shrl	$8, %eax
-	movb	%al, 1(%edi)
-	shrl	$8, %eax
-	movb	%al, 2(%edi)
-
-.o_clip:
-	addl	$3, %edi
-	addl	$4, %esi
-
-	decl	%ebx
-	jnz	.o_forx
-
-	movl	20(%ebp), %eax
-	addl	%eax, 8(%ebp)
-	movl	28(%ebp), %eax
-	addl	%eax, 24(%ebp)
-
-	decl	%ecx
-	jnz	.o_fory
-
-	jmp .exit
-
-.Lfe1:
-	.size	nr_mmx_R8G8B8_R8G8B8_R8G8B8A8_P,.Lfe1-nr_mmx_R8G8B8_R8G8B8_R8G8B8A8_P
-	.ident	"GCC: (GNU) 3.2"