--- mozilla/gfx/cairo/libpixman/src/pixman-access.c.original	2009-04-14 07:26:21.000000000 -0400
+++ mozilla/gfx/cairo/libpixman/src/pixman-access.c	2009-04-14 10:32:23.000000000 -0400
@@ -403,10 +403,55 @@
 fbFetch_a8 (bits_image_t *pict, int x, int y, int width, uint32_t *buffer)
 {
     const uint32_t *bits = pict->bits + y*pict->rowstride;
     const uint8_t *pixel = (const uint8_t *)bits + x;
     const uint8_t *end = pixel + width;
+
+#ifdef _USE_ALPHA_COPY_SSE2_
+    // align dest pointer
+    while ( width && ((unsigned)buffer & 15) )
+    {
+        width--;
+        *buffer++ = READ(pict, pixel++) << 24;
+    }
+
+    // bulk read-shift-write
+    if ( (width >= 16) && !((unsigned)buffer & 15) )
+    {
+        __m128i xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6;
+        const __m128i xmm7 = _mm_setzero_si128();
+
+        while (width >= 16)
+        {
+            xmm0 = _mm_loadu_si128((__m128i *)pixel);
+            width  -= 16;
+            pixel  += 16;
+
+            xmm1 = _mm_load_si128(&xmm7);
+            xmm2 = _mm_load_si128(&xmm7);
+            xmm1 = _mm_unpacklo_epi8(xmm1, xmm0);
+            xmm2 = _mm_unpackhi_epi8(xmm2, xmm0);
+
+            xmm3 = _mm_load_si128(&xmm7);
+            xmm4 = _mm_load_si128(&xmm7);
+            xmm5 = _mm_load_si128(&xmm7);
+            xmm6 = _mm_load_si128(&xmm7);
+
+            xmm3 = _mm_unpacklo_epi16(xmm3, xmm1);
+            xmm4 = _mm_unpackhi_epi16(xmm4, xmm1);
+            xmm5 = _mm_unpacklo_epi16(xmm5, xmm2);
+            xmm6 = _mm_unpackhi_epi16(xmm6, xmm2);
+
+            _mm_store_si128( ((__m128i *)&buffer[ 0]), xmm3 );
+            _mm_store_si128( ((__m128i *)&buffer[ 4]), xmm4 );
+            _mm_store_si128( ((__m128i *)&buffer[ 8]), xmm5 );
+            _mm_store_si128( ((__m128i *)&buffer[12]), xmm6 );
+            buffer += 16;
+        }
+    }
+#endif // _USE_ALPHA_COPY_SSE2_
+
     while (pixel < end) {
 	*buffer++ = READ(pict, pixel++) << 24;
     }
 }