wip commit to remove load8888/store8888 from combine funcsload-store

1 files changed, 34 insertions, 46 deletions

diff --git a/pixman/pixman-mmx.c b/pixman/pixman-mmx.c
index e6335ce..484741e 100644
--- a/pixman/pixman-mmx.c
+++ b/pixman/pixman-mmx.c
@@ -517,23 +517,20 @@ pix_add_mul (__m64 x, __m64 a, __m64 y, __m64 b)
 
 /* --------------- MMX code patch for fbcompose.c --------------------- */
 
-static force_inline uint32_t
+static force_inline __m64
 combine (const uint32_t *src, const uint32_t *mask)
 {
-    uint32_t ssrc = *src;
+    __m64 vsrc = load8888 (src);
 
     if (mask)
     {
 	__m64 m = load8888 (mask);
-	__m64 s = load8888 (&ssrc);
 
 	m = expand_alpha (m);
-	s = pix_multiply (s, m);
-
-	store8888 (&ssrc, s);
+	vsrc = pix_multiply (vsrc, m);
     }
 
-    return ssrc;
+    return vsrc;
 }
 
 static void
@@ -548,20 +545,19 @@ mmx_combine_over_u (pixman_implementation_t *imp,
 
     while (dest < end)
     {
-	uint32_t ssrc = combine (src, mask);
-	uint32_t a = ssrc >> 24;
+	__m64 vsrc = combine (src, mask);
+/*	uint32_t a = ssrc >> 24;
 
 	if (a == 0xff)
 	{
 	    *dest = ssrc;
 	}
 	else if (ssrc)
-	{
-	    __m64 s, sa;
-	    s = load8888 (&ssrc);
-	    sa = expand_alpha (s);
-	    store8888 (dest, over (s, sa, load8888 (dest)));
-	}
+	{*/
+	    __m64 sa;
+	    sa = expand_alpha (vsrc);
+	    store8888 (dest, over (vsrc, sa, load8888 (dest)));
+//	}
 
 	++dest;
 	++src;
@@ -584,11 +580,11 @@ mmx_combine_over_reverse_u (pixman_implementation_t *imp,
     while (dest < end)
     {
 	__m64 d, da;
-	uint32_t s = combine (src, mask);
+	__m64 vsrc = combine (src, mask);
 
 	d = load8888 (dest);
 	da = expand_alpha (d);
-	store8888 (dest, over (d, da, load8888 (&s)));
+	store8888 (dest, over (d, da, vsrc));
 
 	++dest;
 	++src;
@@ -611,12 +607,11 @@ mmx_combine_in_u (pixman_implementation_t *imp,
     while (dest < end)
     {
 	__m64 x, a;
-	uint32_t ssrc = combine (src, mask);
+	__m64 vsrc = combine (src, mask);
 
-	x = load8888 (&ssrc);
 	a = load8888 (dest);
 	a = expand_alpha (a);
-	x = pix_multiply (x, a);
+	x = pix_multiply (vsrc, a);
 
 	store8888 (dest, x);
 
@@ -641,11 +636,10 @@ mmx_combine_in_reverse_u (pixman_implementation_t *imp,
     while (dest < end)
     {
 	__m64 x, a;
-	uint32_t ssrc = combine (src, mask);
+	__m64 vsrc = combine (src, mask);
 
 	x = load8888 (dest);
-	a = load8888 (&ssrc);
-	a = expand_alpha (a);
+	a = expand_alpha (vsrc);
 	x = pix_multiply (x, a);
 	store8888 (dest, x);
 
@@ -670,13 +664,12 @@ mmx_combine_out_u (pixman_implementation_t *imp,
     while (dest < end)
     {
 	__m64 x, a;
-	uint32_t ssrc = combine (src, mask);
+	__m64 vsrc = combine (src, mask);
 
-	x = load8888 (&ssrc);
 	a = load8888 (dest);
 	a = expand_alpha (a);
 	a = negate (a);
-	x = pix_multiply (x, a);
+	x = pix_multiply (vsrc, a);
 	store8888 (dest, x);
 
 	++dest;
@@ -700,11 +693,10 @@ mmx_combine_out_reverse_u (pixman_implementation_t *imp,
     while (dest < end)
     {
 	__m64 x, a;
-	uint32_t ssrc = combine (src, mask);
+	__m64 vsrc = combine (src, mask);
 
 	x = load8888 (dest);
-	a = load8888 (&ssrc);
-	a = expand_alpha (a);
+	a = expand_alpha (vsrc);
 	a = negate (a);
 	x = pix_multiply (x, a);
 
@@ -731,14 +723,13 @@ mmx_combine_atop_u (pixman_implementation_t *imp,
     while (dest < end)
     {
 	__m64 s, da, d, sia;
-	uint32_t ssrc = combine (src, mask);
+	__m64 vsrc = combine (src, mask);
 
-	s = load8888 (&ssrc);
 	d = load8888 (dest);
-	sia = expand_alpha (s);
+	sia = expand_alpha (vsrc);
 	sia = negate (sia);
 	da = expand_alpha (d);
-	s = pix_add_mul (s, da, d, sia);
+	s = pix_add_mul (vsrc, da, d, sia);
 	store8888 (dest, s);
 
 	++dest;
@@ -764,14 +755,13 @@ mmx_combine_atop_reverse_u (pixman_implementation_t *imp,
     while (dest < end)
     {
 	__m64 s, dia, d, sa;
-	uint32_t ssrc = combine (src, mask);
+	__m64 vsrc = combine (src, mask);
 
-	s = load8888 (&ssrc);
 	d = load8888 (dest);
-	sa = expand_alpha (s);
+	sa = expand_alpha (vsrc);
 	dia = expand_alpha (d);
 	dia = negate (dia);
-	s = pix_add_mul (s, dia, d, sa);
+	s = pix_add_mul (vsrc, dia, d, sa);
 	store8888 (dest, s);
 
 	++dest;
@@ -795,15 +785,14 @@ mmx_combine_xor_u (pixman_implementation_t *imp,
     while (dest < end)
     {
 	__m64 s, dia, d, sia;
-	uint32_t ssrc = combine (src, mask);
+	__m64 vsrc = combine (src, mask);
 
-	s = load8888 (&ssrc);
 	d = load8888 (dest);
-	sia = expand_alpha (s);
+	sia = expand_alpha (vsrc);
 	dia = expand_alpha (d);
 	sia = negate (sia);
 	dia = negate (dia);
-	s = pix_add_mul (s, dia, d, sia);
+	s = pix_add_mul (vsrc, dia, d, sia);
 	store8888 (dest, s);
 
 	++dest;
@@ -827,12 +816,11 @@ mmx_combine_add_u (pixman_implementation_t *imp,
     while (dest < end)
     {
 	__m64 s, d;
-	uint32_t ssrc = combine (src, mask);
+	__m64 vsrc = combine (src, mask);
 
-	s = load8888 (&ssrc);
 	d = load8888 (dest);
-	s = pix_add (s, d);
-	store8888 (dest, s);
+	s = pix_add (vsrc, d);
+	store (dest, s);
 
 	++dest;
 	++src;
@@ -3271,7 +3259,7 @@ _pixman_implementation_create_mmx (pixman_implementation_t *fallback)
     imp->combine_32[PIXMAN_OP_ATOP_REVERSE] = mmx_combine_atop_reverse_u;
     imp->combine_32[PIXMAN_OP_XOR] = mmx_combine_xor_u;
     imp->combine_32[PIXMAN_OP_ADD] = mmx_combine_add_u;
-    imp->combine_32[PIXMAN_OP_SATURATE] = mmx_combine_saturate_u;
+//    imp->combine_32[PIXMAN_OP_SATURATE] = mmx_combine_saturate_u;
 
     imp->combine_32_ca[PIXMAN_OP_SRC] = mmx_combine_src_ca;
     imp->combine_32_ca[PIXMAN_OP_OVER] = mmx_combine_over_ca;