mmx: make load8888 take a pointer to data instead of the data itself

Allows us to tune how we load data into the vector registers.

Signed-off-by: Matt Turner <mattst88@gmail.com>

And squashed in:

mmx: define and use load8888u function

For unaligned loads.

Signed-off-by: Matt Turner <mattst88@gmail.com>

1 files changed, 148 insertions, 129 deletions

diff --git a/pixman/pixman-mmx.c b/pixman/pixman-mmx.c
index 6ffeb4c..bd44f63 100644
--- a/pixman/pixman-mmx.c
+++ b/pixman/pixman-mmx.c
@@ -347,5 +347,12 @@ static __inline__ uint32_t ldl_u(uint32_t *p)
 static force_inline __m64
-load8888 (uint32_t v)
+load8888 (const uint32_t *v)
 {
-    return _mm_unpacklo_pi8 (_mm_cvtsi32_si64 (v), _mm_setzero_si64 ());
+    return _mm_unpacklo_pi8 (_mm_cvtsi32_si64 (*v), _mm_setzero_si64 ());
+}
+
+static force_inline __m64
+load8888u (const uint32_t *v)
+{
+    uint32_t l = ldl_u(v);
+    return load8888(&l);
 }
@@ -473,4 +480,4 @@ combine (const uint32_t *src, const uint32_t *mask)
     {
-	__m64 m = load8888 (*mask);
-	__m64 s = load8888 (ssrc);
+	__m64 m = load8888 (mask);
+	__m64 s = load8888 (&ssrc);
 
@@ -507,5 +514,5 @@ mmx_combine_over_u (pixman_implementation_t *imp,
 	    __m64 s, sa;
-	    s = load8888 (ssrc);
+	    s = load8888 (&ssrc);
 	    sa = expand_alpha (s);
-	    store8888 (dest, over (s, sa, load8888 (*dest)));
+	    store8888 (dest, over (s, sa, load8888 (dest)));
 	}
@@ -535,5 +542,5 @@ mmx_combine_over_reverse_u (pixman_implementation_t *imp,
 
-	d = load8888 (*dest);
+	d = load8888 (dest);
 	da = expand_alpha (d);
-	store8888 (dest, over (d, da, load8888 (s)));
+	store8888 (dest, over (d, da, load8888 (&s)));
 
@@ -560,5 +567,6 @@ mmx_combine_in_u (pixman_implementation_t *imp,
 	__m64 x, a;
+	uint32_t ssrc = combine (src, mask);
 
-	x = load8888 (combine (src, mask));
-	a = load8888 (*dest);
+	x = load8888 (&ssrc);
+	a = load8888 (dest);
 	a = expand_alpha (a);
@@ -589,5 +597,6 @@ mmx_combine_in_reverse_u (pixman_implementation_t *imp,
 	__m64 x, a;
+	uint32_t ssrc = combine (src, mask);
 
-	x = load8888 (*dest);
-	a = load8888 (combine (src, mask));
+	x = load8888 (dest);
+	a = load8888 (&ssrc);
 	a = expand_alpha (a);
@@ -617,5 +626,6 @@ mmx_combine_out_u (pixman_implementation_t *imp,
 	__m64 x, a;
+	uint32_t ssrc = combine (src, mask);
 
-	x = load8888 (combine (src, mask));
-	a = load8888 (*dest);
+	x = load8888 (&ssrc);
+	a = load8888 (dest);
 	a = expand_alpha (a);
@@ -646,5 +656,6 @@ mmx_combine_out_reverse_u (pixman_implementation_t *imp,
 	__m64 x, a;
+	uint32_t ssrc = combine (src, mask);
 
-	x = load8888 (*dest);
-	a = load8888 (combine (src, mask));
+	x = load8888 (dest);
+	a = load8888 (&ssrc);
 	a = expand_alpha (a);
@@ -676,5 +687,6 @@ mmx_combine_atop_u (pixman_implementation_t *imp,
 	__m64 s, da, d, sia;
+	uint32_t ssrc = combine (src, mask);
 
-	s = load8888 (combine (src, mask));
-	d = load8888 (*dest);
+	s = load8888 (&ssrc);
+	d = load8888 (dest);
 	sia = expand_alpha (s);
@@ -708,5 +720,6 @@ mmx_combine_atop_reverse_u (pixman_implementation_t *imp,
 	__m64 s, dia, d, sa;
+	uint32_t ssrc = combine (src, mask);
 
-	s = load8888 (combine (src, mask));
-	d = load8888 (*dest);
+	s = load8888 (&ssrc);
+	d = load8888 (dest);
 	sa = expand_alpha (s);
@@ -738,5 +751,6 @@ mmx_combine_xor_u (pixman_implementation_t *imp,
 	__m64 s, dia, d, sia;
+	uint32_t ssrc = combine (src, mask);
 
-	s = load8888 (combine (src, mask));
-	d = load8888 (*dest);
+	s = load8888 (&ssrc);
+	d = load8888 (dest);
 	sia = expand_alpha (s);
@@ -769,5 +783,6 @@ mmx_combine_add_u (pixman_implementation_t *imp,
 	__m64 s, d;
+	uint32_t ssrc = combine (src, mask);
 
-	s = load8888 (combine (src, mask));
-	d = load8888 (*dest);
+	s = load8888 (&ssrc);
+	d = load8888 (dest);
 	s = pix_add (s, d);
@@ -797,4 +812,4 @@ mmx_combine_saturate_u (pixman_implementation_t *imp,
 	uint32_t d = *dest;
-	__m64 ms = load8888 (s);
-	__m64 md = load8888 (d);
+	__m64 ms = load8888 (&s);
+	__m64 md = load8888 (&d);
 	uint32_t sa = s >> 24;
@@ -804,3 +819,4 @@ mmx_combine_saturate_u (pixman_implementation_t *imp,
 	{
-	    __m64 msa = load8888 (DIV_UN8 (da, sa) << 24);
+	    uint32_t quot = DIV_UN8 (da, sa) << 24;
+	    __m64 msa = load8888 (&quot);
 	    msa = expand_alpha (msa);
@@ -832,4 +848,4 @@ mmx_combine_src_ca (pixman_implementation_t *imp,
     {
-	__m64 a = load8888 (*mask);
-	__m64 s = load8888 (*src);
+	__m64 a = load8888 (mask);
+	__m64 s = load8888 (src);
 
@@ -857,5 +873,5 @@ mmx_combine_over_ca (pixman_implementation_t *imp,
     {
-	__m64 a = load8888 (*mask);
-	__m64 s = load8888 (*src);
-	__m64 d = load8888 (*dest);
+	__m64 a = load8888 (mask);
+	__m64 s = load8888 (src);
+	__m64 d = load8888 (dest);
 	__m64 sa = expand_alpha (s);
@@ -883,5 +899,5 @@ mmx_combine_over_reverse_ca (pixman_implementation_t *imp,
     {
-	__m64 a = load8888 (*mask);
-	__m64 s = load8888 (*src);
-	__m64 d = load8888 (*dest);
+	__m64 a = load8888 (mask);
+	__m64 s = load8888 (src);
+	__m64 d = load8888 (dest);
 	__m64 da = expand_alpha (d);
@@ -909,5 +925,5 @@ mmx_combine_in_ca (pixman_implementation_t *imp,
     {
-	__m64 a = load8888 (*mask);
-	__m64 s = load8888 (*src);
-	__m64 d = load8888 (*dest);
+	__m64 a = load8888 (mask);
+	__m64 s = load8888 (src);
+	__m64 d = load8888 (dest);
 	__m64 da = expand_alpha (d);
@@ -937,5 +953,5 @@ mmx_combine_in_reverse_ca (pixman_implementation_t *imp,
     {
-	__m64 a = load8888 (*mask);
-	__m64 s = load8888 (*src);
-	__m64 d = load8888 (*dest);
+	__m64 a = load8888 (mask);
+	__m64 s = load8888 (src);
+	__m64 d = load8888 (dest);
 	__m64 sa = expand_alpha (s);
@@ -965,5 +981,5 @@ mmx_combine_out_ca (pixman_implementation_t *imp,
     {
-	__m64 a = load8888 (*mask);
-	__m64 s = load8888 (*src);
-	__m64 d = load8888 (*dest);
+	__m64 a = load8888 (mask);
+	__m64 s = load8888 (src);
+	__m64 d = load8888 (dest);
 	__m64 da = expand_alpha (d);
@@ -994,5 +1010,5 @@ mmx_combine_out_reverse_ca (pixman_implementation_t *imp,
     {
-	__m64 a = load8888 (*mask);
-	__m64 s = load8888 (*src);
-	__m64 d = load8888 (*dest);
+	__m64 a = load8888 (mask);
+	__m64 s = load8888 (src);
+	__m64 d = load8888 (dest);
 	__m64 sa = expand_alpha (s);
@@ -1023,5 +1039,5 @@ mmx_combine_atop_ca (pixman_implementation_t *imp,
     {
-	__m64 a = load8888 (*mask);
-	__m64 s = load8888 (*src);
-	__m64 d = load8888 (*dest);
+	__m64 a = load8888 (mask);
+	__m64 s = load8888 (src);
+	__m64 d = load8888 (dest);
 	__m64 da = expand_alpha (d);
@@ -1054,5 +1070,5 @@ mmx_combine_atop_reverse_ca (pixman_implementation_t *imp,
     {
-	__m64 a = load8888 (*mask);
-	__m64 s = load8888 (*src);
-	__m64 d = load8888 (*dest);
+	__m64 a = load8888 (mask);
+	__m64 s = load8888 (src);
+	__m64 d = load8888 (dest);
 	__m64 da = expand_alpha (d);
@@ -1085,5 +1101,5 @@ mmx_combine_xor_ca (pixman_implementation_t *imp,
     {
-	__m64 a = load8888 (*mask);
-	__m64 s = load8888 (*src);
-	__m64 d = load8888 (*dest);
+	__m64 a = load8888 (mask);
+	__m64 s = load8888 (src);
+	__m64 d = load8888 (dest);
 	__m64 da = expand_alpha (d);
@@ -1117,5 +1133,5 @@ mmx_combine_add_ca (pixman_implementation_t *imp,
     {
-	__m64 a = load8888 (*mask);
-	__m64 s = load8888 (*src);
-	__m64 d = load8888 (*dest);
+	__m64 a = load8888 (mask);
+	__m64 s = load8888 (src);
+	__m64 d = load8888 (dest);
 
@@ -1154,3 +1170,3 @@ mmx_composite_over_n_8888 (pixman_implementation_t *imp,
 
-    vsrc = load8888 (src);
+    vsrc = load8888 (&src);
     vsrca = expand_alpha (vsrc);
@@ -1167,3 +1183,3 @@ mmx_composite_over_n_8888 (pixman_implementation_t *imp,
 	{
-	    store8888 (dst, over (vsrc, vsrca, load8888 (*dst)));
+	    store8888 (dst, over (vsrc, vsrca, load8888 (dst)));
 
@@ -1193,3 +1209,3 @@ mmx_composite_over_n_8888 (pixman_implementation_t *imp,
 	{
-	    store8888 (dst, over (vsrc, vsrca, load8888 (*dst)));
+	    store8888 (dst, over (vsrc, vsrca, load8888 (dst)));
 	}
@@ -1220,3 +1236,3 @@ mmx_composite_over_n_0565 (pixman_implementation_t *imp,
 
-    vsrc = load8888 (src);
+    vsrc = load8888 (&src);
     vsrca = expand_alpha (vsrc);
@@ -1299,3 +1315,3 @@ mmx_composite_over_n_8888_8888_ca (pixman_implementation_t *imp,
 
-    vsrc = load8888 (src);
+    vsrc = load8888 (&src);
     vsrca = expand_alpha (vsrc);
@@ -1314,4 +1330,4 @@ mmx_composite_over_n_8888_8888_ca (pixman_implementation_t *imp,
 	    {
-		__m64 vdest = load8888 (*q);
-		vdest = in_over (vsrc, vsrca, load8888 (m), vdest);
+		__m64 vdest = load8888 (q);
+		vdest = in_over (vsrc, vsrca, load8888 (&m), vdest);
 		store8888 (q, vdest);
@@ -1335,5 +1351,5 @@ mmx_composite_over_n_8888_8888_ca (pixman_implementation_t *imp,
 
-		dest0 = in_over (vsrc, vsrca, load8888 (m0),
+		dest0 = in_over (vsrc, vsrca, load8888 (&m0),
 		                 expand8888 (vdest, 0));
-		dest1 = in_over (vsrc, vsrca, load8888 (m1),
+		dest1 = in_over (vsrc, vsrca, load8888 (&m1),
 		                 expand8888 (vdest, 1));
@@ -1354,4 +1370,4 @@ mmx_composite_over_n_8888_8888_ca (pixman_implementation_t *imp,
 	    {
-		__m64 vdest = load8888 (*q);
-		vdest = in_over (vsrc, vsrca, load8888 (m), vdest);
+		__m64 vdest = load8888 (q);
+		vdest = in_over (vsrc, vsrca, load8888 (&m), vdest);
 		store8888 (q, vdest);
@@ -1391,3 +1407,3 @@ mmx_composite_over_8888_n_8888 (pixman_implementation_t *imp,
     mask = mask | mask >> 8 | mask >> 16 | mask >> 24;
-    vmask = load8888 (mask);
+    vmask = load8888 (&mask);
 
@@ -1403,4 +1419,4 @@ mmx_composite_over_8888_n_8888 (pixman_implementation_t *imp,
 	{
-	    __m64 s = load8888 (*src);
-	    __m64 d = load8888 (*dst);
+	    __m64 s = load8888 (src);
+	    __m64 d = load8888 (dst);
 
@@ -1431,4 +1447,4 @@ mmx_composite_over_8888_n_8888 (pixman_implementation_t *imp,
 	{
-	    __m64 s = load8888 (*src);
-	    __m64 d = load8888 (*dst);
+	    __m64 s = load8888 (src);
+	    __m64 d = load8888 (dst);
 
@@ -1462,3 +1478,3 @@ mmx_composite_over_x888_n_8888 (pixman_implementation_t *imp,
     mask = mask | mask >> 8 | mask >> 16 | mask >> 24;
-    vmask = load8888 (mask);
+    vmask = load8888 (&mask);
     srca = MC (4x00ff);
@@ -1475,4 +1491,5 @@ mmx_composite_over_x888_n_8888 (pixman_implementation_t *imp,
 	{
-	    __m64 s = load8888 (*src | 0xff000000);
-	    __m64 d = load8888 (*dst);
+	    uint32_t ssrc = *src | 0xff000000;
+	    __m64 s = load8888 (&ssrc);
+	    __m64 d = load8888 (dst);
 
@@ -1553,4 +1570,5 @@ mmx_composite_over_x888_n_8888 (pixman_implementation_t *imp,
 	{
-	    __m64 s = load8888 (*src | 0xff000000);
-	    __m64 d = load8888 (*dst);
+	    uint32_t ssrc = *src | 0xff000000;
+	    __m64 s = load8888 (&ssrc);
+	    __m64 d = load8888 (dst);
 
@@ -1604,5 +1622,5 @@ mmx_composite_over_8888_8888 (pixman_implementation_t *imp,
 		__m64 ms, sa;
-		ms = load8888 (s);
+		ms = load8888 (&s);
 		sa = expand_alpha (ms);
-		store8888 (dst, over (ms, sa, load8888 (*dst)));
+		store8888 (dst, over (ms, sa, load8888 (dst)));
 	    }
@@ -1647,3 +1665,3 @@ mmx_composite_over_8888_0565 (pixman_implementation_t *imp,
 	{
-	    __m64 vsrc = load8888 (*src);
+	    __m64 vsrc = load8888 (src);
 	    uint64_t d = *dst;
@@ -1668,6 +1686,6 @@ mmx_composite_over_8888_0565 (pixman_implementation_t *imp,
 
-	    vsrc0 = load8888 (*(src + 0));
-	    vsrc1 = load8888 (*(src + 1));
-	    vsrc2 = load8888 (*(src + 2));
-	    vsrc3 = load8888 (*(src + 3));
+	    vsrc0 = load8888 ((src + 0));
+	    vsrc1 = load8888 ((src + 1));
+	    vsrc2 = load8888 ((src + 2));
+	    vsrc3 = load8888 ((src + 3));
 
@@ -1691,3 +1709,3 @@ mmx_composite_over_8888_0565 (pixman_implementation_t *imp,
 	{
-	    __m64 vsrc = load8888 (*src);
+	    __m64 vsrc = load8888 (src);
 	    uint64_t d = *dst;
@@ -1734,3 +1752,3 @@ mmx_composite_over_n_8_8888 (pixman_implementation_t *imp,
 
-    vsrc = load8888 (src);
+    vsrc = load8888 (&src);
     vsrca = expand_alpha (vsrc);
@@ -1755,3 +1773,3 @@ mmx_composite_over_n_8_8888 (pixman_implementation_t *imp,
 				       expand_alpha_rev (to_m64 (m)),
-				       load8888 (*dst));
+				       load8888 (dst));
 
@@ -1806,3 +1824,3 @@ mmx_composite_over_n_8_8888 (pixman_implementation_t *imp,
 	    {
-		__m64 vdest = load8888 (*dst);
+		__m64 vdest = load8888 (dst);
 
@@ -1999,3 +2017,3 @@ mmx_composite_src_n_8_8888 (pixman_implementation_t *imp,
 
-    vsrc = load8888 (src);
+    vsrc = load8888 (&src);
 
@@ -2070,3 +2088,3 @@ mmx_composite_src_n_8_8888 (pixman_implementation_t *imp,
 	    {
-		__m64 vdest = load8888 (*dst);
+		__m64 vdest = load8888 (dst);
 
@@ -2109,3 +2127,3 @@ mmx_composite_over_n_8_0565 (pixman_implementation_t *imp,
 
-    vsrc = load8888 (src);
+    vsrc = load8888 (&src);
     vsrca = expand_alpha (vsrc);
@@ -2248,3 +2266,3 @@ mmx_composite_over_pixbuf_0565 (pixman_implementation_t *imp,
 	{
-	    __m64 vsrc = load8888 (*src);
+	    __m64 vsrc = load8888 (src);
 	    uint64_t d = *dst;
@@ -2281,6 +2299,6 @@ mmx_composite_over_pixbuf_0565 (pixman_implementation_t *imp,
 		__m64 vdest;
-		vdest = pack_565 (invert_colors (load8888 (s0)), _mm_setzero_si64 (), 0);
-		vdest = pack_565 (invert_colors (load8888 (s1)), vdest, 1);
-		vdest = pack_565 (invert_colors (load8888 (s2)), vdest, 2);
-		vdest = pack_565 (invert_colors (load8888 (s3)), vdest, 3);
+		vdest = pack_565 (invert_colors (load8888 (&s0)), _mm_setzero_si64 (), 0);
+		vdest = pack_565 (invert_colors (load8888 (&s1)), vdest, 1);
+		vdest = pack_565 (invert_colors (load8888 (&s2)), vdest, 2);
+		vdest = pack_565 (invert_colors (load8888 (&s3)), vdest, 3);
 
@@ -2292,6 +2310,6 @@ mmx_composite_over_pixbuf_0565 (pixman_implementation_t *imp,
 
-		vdest = pack_565 (over_rev_non_pre (load8888 (s0), expand565 (vdest, 0)), vdest, 0);
-		vdest = pack_565 (over_rev_non_pre (load8888 (s1), expand565 (vdest, 1)), vdest, 1);
-		vdest = pack_565 (over_rev_non_pre (load8888 (s2), expand565 (vdest, 2)), vdest, 2);
-		vdest = pack_565 (over_rev_non_pre (load8888 (s3), expand565 (vdest, 3)), vdest, 3);
+		vdest = pack_565 (over_rev_non_pre (load8888 (&s0), expand565 (vdest, 0)), vdest, 0);
+		vdest = pack_565 (over_rev_non_pre (load8888 (&s1), expand565 (vdest, 1)), vdest, 1);
+		vdest = pack_565 (over_rev_non_pre (load8888 (&s2), expand565 (vdest, 2)), vdest, 2);
+		vdest = pack_565 (over_rev_non_pre (load8888 (&s3), expand565 (vdest, 3)), vdest, 3);
 
@@ -2309,3 +2327,3 @@ mmx_composite_over_pixbuf_0565 (pixman_implementation_t *imp,
 	{
-	    __m64 vsrc = load8888 (*src);
+	    __m64 vsrc = load8888 (src);
 	    uint64_t d = *dst;
@@ -2356,4 +2374,4 @@ mmx_composite_over_pixbuf_8888 (pixman_implementation_t *imp,
 	{
-	    __m64 s = load8888 (*src);
-	    __m64 d = load8888 (*dst);
+	    __m64 s = load8888 (src);
+	    __m64 d = load8888 (dst);
 
@@ -2368,3 +2386,3 @@ mmx_composite_over_pixbuf_8888 (pixman_implementation_t *imp,
 	{
-	    uint64_t s0, s1;
+	    uint32_t s0, s1;
 	    unsigned char a0, a1;
@@ -2380,4 +2398,4 @@ mmx_composite_over_pixbuf_8888 (pixman_implementation_t *imp,
 	    {
-		d0 = invert_colors (load8888 (s0));
-		d1 = invert_colors (load8888 (s1));
+		d0 = invert_colors (load8888 (&s0));
+		d1 = invert_colors (load8888 (&s1));
 
@@ -2389,4 +2407,4 @@ mmx_composite_over_pixbuf_8888 (pixman_implementation_t *imp,
 
-		d0 = over_rev_non_pre (load8888 (s0), expand8888 (vdest, 0));
-		d1 = over_rev_non_pre (load8888 (s1), expand8888 (vdest, 1));
+		d0 = over_rev_non_pre (load8888 (&s0), expand8888 (vdest, 0));
+		d1 = over_rev_non_pre (load8888 (&s1), expand8888 (vdest, 1));
 
@@ -2402,4 +2420,4 @@ mmx_composite_over_pixbuf_8888 (pixman_implementation_t *imp,
 	{
-	    __m64 s = load8888 (*src);
-	    __m64 d = load8888 (*dst);
+	    __m64 s = load8888 (src);
+	    __m64 d = load8888 (dst);
 
@@ -2433,3 +2451,3 @@ mmx_composite_over_n_8888_0565_ca (pixman_implementation_t *imp,
 
-    vsrc = load8888 (src);
+    vsrc = load8888 (&src);
     vsrca = expand_alpha (vsrc);
@@ -2450,3 +2468,3 @@ mmx_composite_over_n_8888_0565_ca (pixman_implementation_t *imp,
 		__m64 vdest = expand565 (to_m64 (d), 0);
-		vdest = pack_565 (in_over (vsrc, vsrca, load8888 (m), vdest), vdest, 0);
+		vdest = pack_565 (in_over (vsrc, vsrca, load8888 (&m), vdest), vdest, 0);
 		*q = to_uint64 (vdest);
@@ -2472,6 +2490,6 @@ mmx_composite_over_n_8888_0565_ca (pixman_implementation_t *imp,
 
-		vdest = pack_565 (in_over (vsrc, vsrca, load8888 (m0), expand565 (vdest, 0)), vdest, 0);
-		vdest = pack_565 (in_over (vsrc, vsrca, load8888 (m1), expand565 (vdest, 1)), vdest, 1);
-		vdest = pack_565 (in_over (vsrc, vsrca, load8888 (m2), expand565 (vdest, 2)), vdest, 2);
-		vdest = pack_565 (in_over (vsrc, vsrca, load8888 (m3), expand565 (vdest, 3)), vdest, 3);
+		vdest = pack_565 (in_over (vsrc, vsrca, load8888 (&m0), expand565 (vdest, 0)), vdest, 0);
+		vdest = pack_565 (in_over (vsrc, vsrca, load8888 (&m1), expand565 (vdest, 1)), vdest, 1);
+		vdest = pack_565 (in_over (vsrc, vsrca, load8888 (&m2), expand565 (vdest, 2)), vdest, 2);
+		vdest = pack_565 (in_over (vsrc, vsrca, load8888 (&m3), expand565 (vdest, 3)), vdest, 3);
 
@@ -2493,3 +2511,3 @@ mmx_composite_over_n_8888_0565_ca (pixman_implementation_t *imp,
 		__m64 vdest = expand565 (to_m64 (d), 0);
-		vdest = pack_565 (in_over (vsrc, vsrca, load8888 (m), vdest), vdest, 0);
+		vdest = pack_565 (in_over (vsrc, vsrca, load8888 (&m), vdest), vdest, 0);
 		*q = to_uint64 (vdest);
@@ -2529,3 +2547,3 @@ mmx_composite_in_n_8_8 (pixman_implementation_t *imp,
 
-    vsrc = load8888 (src);
+    vsrc = load8888 (&src);
     vsrca = expand_alpha (vsrc);
@@ -2561,4 +2579,4 @@ mmx_composite_in_n_8_8 (pixman_implementation_t *imp,
 
-	    vmask = load8888 (ldl_u((uint32_t *)mask));
-	    vdest = load8888 (*(uint32_t *)dst);
+	    vmask = load8888u ((uint32_t *)mask);
+	    vdest = load8888 ((uint32_t *)dst);
 
@@ -2631,3 +2649,3 @@ mmx_composite_in_8_8 (pixman_implementation_t *imp,
 
-	    store8888 (d, in (load8888 (ldl_u((uint32_t *)s)), load8888 (*d)));
+	    store8888 (d, in (load8888u (s), load8888 (d)));
 
@@ -2679,3 +2697,3 @@ mmx_composite_add_n_8_8 (pixman_implementation_t *imp,
 
-    vsrc = load8888 (src);
+    vsrc = load8888 (&src);
     vsrca = expand_alpha (vsrc);
@@ -2712,4 +2730,4 @@ mmx_composite_add_n_8_8 (pixman_implementation_t *imp,
 
-	    vmask = load8888 (ldl_u((uint32_t *)mask));
-	    vdest = load8888 (*(uint32_t *)dst);
+	    vmask = load8888u ((uint32_t *)mask);
+	    vdest = load8888 ((uint32_t *)dst);
 
@@ -3056,3 +3074,4 @@ mmx_composite_over_x888_8_8888 (pixman_implementation_t *imp,
 	    {
-		__m64 s = load8888 (*src | 0xff000000);
+		uint32_t ssrc = *src | 0xff000000;
+		__m64 s = load8888 (&ssrc);
 
@@ -3066,3 +3085,3 @@ mmx_composite_over_x888_8_8888 (pixman_implementation_t *imp,
 		    __m64 vm = expand_alpha_rev (to_m64 (m));
-		    __m64 vdest = in_over (s, sa, vm, load8888 (*dst));
+		    __m64 vdest = in_over (s, sa, vm, load8888 (dst));