diff --git a/libvips/foreign/cairo.c b/libvips/foreign/cairo.c
index b242ab49..e922cbbe 100644
--- a/libvips/foreign/cairo.c
+++ b/libvips/foreign/cairo.c
@@ -35,7 +35,7 @@
 #include <vips/vips.h>
 #include <vips/internal.h>
 
-/* Convert from ARGB to RGBA and undo premultiplication. 
+/* Convert from Cairo's BGRA to RGBA and undo premultiplication. 
  *
  * See also openslide's argb2rgba().
  */
@@ -45,22 +45,26 @@ vips__cairo2rgba( guint32 * restrict buf, int n )
 	int i;
 
 	for( i = 0; i < n; i++ ) {
-		guint32 * restrict p = buf + i;
-		guint32 x = *p;
-		guint8 a = x >> 24;
-		VipsPel * restrict out = (VipsPel *) p;
+		guint32 bgra = GUINT32_FROM_BE( buf[i] );
+		guint8 a = bgra & 0xff;
 
-		if( a == 255 ) 
-			*p = GUINT32_TO_BE( (x << 8) | 255 );
-		else if( a == 0 ) 
-			*p = GUINT32_TO_BE( x << 8 );
-		else {
-			/* Undo premultiplication.
-			 */
-			out[0] = 255 * ((x >> 16) & 255) / a;
-			out[1] = 255 * ((x >> 8) & 255) / a;
-			out[2] = 255 * (x & 255) / a;
-			out[3] = a;
-		}
+                guint32 rgba;
+
+                if( a == 0 || 
+                        a == 255 )
+			rgba = 
+				(bgra & 0x00ff00ff) |
+			        (bgra & 0x0000ff00) << 16 |
+			        (bgra & 0xff000000) >> 16;
+                else
+                        /* Undo premultiplication.
+                         */
+                        rgba = 
+                                ((255 * ((bgra >> 8) & 0xff) / a) << 24) |
+                                ((255 * ((bgra >> 16) & 0xff) / a) << 16) |
+                                ((255 * ((bgra >> 24) & 0xff) / a) << 8) |
+                                a;
+
+                buf[i] = GUINT32_TO_BE( rgba );
 	}
 }
diff --git a/libvips/foreign/pdfiumload.c b/libvips/foreign/pdfiumload.c
index d0b46463..038552bf 100644
--- a/libvips/foreign/pdfiumload.c
+++ b/libvips/foreign/pdfiumload.c
@@ -43,7 +43,6 @@
 
 /* TODO 
  *
- * - speed up BRGA -> RGBA conversion
  * - what about filename encodings?
  * - need to test on Windows
  */
@@ -569,19 +568,25 @@ vips_foreign_load_pdf_generate( VipsRegion *or,
 		i += 1;
 	}
 
-	/* PDFium writes BRGA, we must swap.
-	 *
-	 * FIXME ... this is slow, try using vips__cairo2rgba()? Do we need to
-	 * unpremultiply as well?
+	/* PDFium writes BGRA, we must swap.
 	 */
 	for( y = 0; y < r->height; y++ ) {
-		VipsPel *p;
+		guint32 * restrict p = 
+			(guint32 *) VIPS_REGION_ADDR( or, r->left, r->top + y );
+
 		int x;
 
-		p = VIPS_REGION_ADDR( or, r->left, r->top + y );
 		for( x = 0; x < r->width; x++ ) { 
-			VIPS_SWAP( VipsPel, p[0], p[2] );
-			p += 4;
+			guint32 bgra = GUINT32_FROM_BE( p[x] );
+
+			guint rgba;
+
+			rgba = 
+				(bgra & 0x00ff00ff) |
+			        (bgra & 0x0000ff00) << 16 |
+			        (bgra & 0xff000000) >> 16;
+
+			p[x] = GUINT32_TO_BE( rgba );
 		}
 	}