vmx: optimize vmx_composite_over_n_8888_8888_ca

This patch optimizes vmx_composite_over_n_8888_8888_ca by removing use
of expand_alpha_1x128, unpack/pack and in_over_2x128 in favor of
splat_alpha, in_over and MUL/ADD macros from pixman_combine32.h.

Running "lowlevel-blt-bench -n over_8888_8888" on POWER8, 8 cores,
3.4GHz, RHEL 7.2 ppc64le gave the following results:

reference memcpy speed = 23475.4MB/s (5868.8MP/s for 32bpp fills)

                Before          After           Change
              --------------------------------------------
L1              244.97          474.05         +93.51%
L2              243.74          473.05         +94.08%
M               243.29          467.16         +92.02%
HT              144.03          252.79         +75.51%
VT              174.24          279.03         +60.14%
R               109.86          149.98         +36.52%
RT              47.96           53.18          +10.88%
Kops/s          524             576            +9.92%

Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Acked-by: Pekka Paalanen <pekka.paalanen@collabora.co.uk>
Acked-by: Siarhei Siamashka <siarhei.siamashka@gmail.com>

1 files changed, 21 insertions, 31 deletions

diff --git a/pixman/pixman-vmx.c b/pixman/pixman-vmx.c
index 4105dee1..a5c5db98 100644
--- a/pixman/pixman-vmx.c
+++ b/pixman/pixman-vmx.c
@@ -2746,7 +2746,7 @@ vmx_composite_over_n_8888_8888_ca (pixman_implementation_t *imp,
                                     pixman_composite_info_t *info)
 {
     PIXMAN_COMPOSITE_ARGS (info);
-    uint32_t src;
+    uint32_t src, ia;
     uint32_t    *dst_line, d;
     uint32_t    *mask_line, m;
     uint32_t pack_cmp;
@@ -2754,9 +2754,6 @@ vmx_composite_over_n_8888_8888_ca (pixman_implementation_t *imp,
 
     vector unsigned int vsrc, valpha, vmask, vdest;
 
-    vector unsigned int vmx_dst, vmx_dst_lo, vmx_dst_hi;
-    vector unsigned int vmx_mask, vmx_mask_lo, vmx_mask_hi;
-
     src = _pixman_image_get_solid (imp, src_image, dest_image->bits.format);
 
     if (src == 0)
@@ -2767,31 +2764,33 @@ vmx_composite_over_n_8888_8888_ca (pixman_implementation_t *imp,
     PIXMAN_IMAGE_GET_LINE (
 	mask_image, mask_x, mask_y, uint32_t, mask_stride, mask_line, 1);
 
-    vsrc = unpacklo_128_16x8(create_mask_1x32_128 (&src),
-			    (vector unsigned int) AVV(0));
-
-    valpha = expand_alpha_1x128(vsrc);
+    vsrc = (vector unsigned int) {src, src, src, src};
+    valpha = splat_alpha(vsrc);
+    ia = ALPHA_8 (src);
 
     while (height--)
     {
 	int w = width;
 	const uint32_t *pm = (uint32_t *)mask_line;
 	uint32_t *pd = (uint32_t *)dst_line;
+	uint32_t s;
 
 	dst_line += dst_stride;
 	mask_line += mask_stride;
 
 	while (w && (uintptr_t)pd & 15)
 	{
+	    s = src;
 	    m = *pm++;
 
 	    if (m)
 	    {
 		d = *pd;
-		vmask = unpack_32_1x128(m);
-		vdest = unpack_32_1x128(d);
-
-		*pd = pack_1x128_32(in_over (vsrc, valpha, vmask, vdest));
+		UN8x4_MUL_UN8x4 (s, m);
+		UN8x4_MUL_UN8 (m, ia);
+		m = ~m;
+		UN8x4_MUL_UN8x4_ADD_UN8x4 (d, m, s);
+		*pd = d;
 	    }
 
 	    pd++;
@@ -2801,28 +2800,17 @@ vmx_composite_over_n_8888_8888_ca (pixman_implementation_t *imp,
 	while (w >= 4)
 	{
 	    /* pm is NOT necessarily 16-byte aligned */
-	    vmx_mask = load_128_unaligned (pm);
+	    vmask = load_128_unaligned (pm);
 
-	    pack_cmp = vec_all_eq(vmx_mask, (vector unsigned int) AVV(0));
+	    pack_cmp = vec_all_eq(vmask, (vector unsigned int) AVV(0));
 
 	    /* if all bits in mask are zero, pack_cmp is not 0 */
 	    if (pack_cmp == 0)
 	    {
 		/* pd is 16-byte aligned */
-		vmx_dst = load_128_aligned (pd);
-
-		unpack_128_2x128 (vmx_mask, (vector unsigned int) AVV(0),
-				    &vmx_mask_lo, &vmx_mask_hi);
+		vdest = in_over (vsrc, valpha, vmask, load_128_aligned (pd));
 
-		unpack_128_2x128 (vmx_dst, (vector unsigned int) AVV(0),
-				    &vmx_dst_lo, &vmx_dst_hi);
-
-		in_over_2x128 (&vsrc, &vsrc,
-			       &valpha, &valpha,
-			       &vmx_mask_lo, &vmx_mask_hi,
-			       &vmx_dst_lo, &vmx_dst_hi);
-
-		save_128_aligned(pd, pack_2x128_128(vmx_dst_lo, vmx_dst_hi));
+		save_128_aligned(pd, vdest);
 	    }
 
 	    pd += 4;
@@ -2832,15 +2820,17 @@ vmx_composite_over_n_8888_8888_ca (pixman_implementation_t *imp,
 
 	while (w)
 	{
+	    s = src;
 	    m = *pm++;
 
 	    if (m)
 	    {
 		d = *pd;
-		vmask = unpack_32_1x128(m);
-		vdest = unpack_32_1x128(d);
-
-		*pd = pack_1x128_32(in_over (vsrc, valpha, vmask, vdest));
+		UN8x4_MUL_UN8x4 (s, m);
+		UN8x4_MUL_UN8 (m, ia);
+		m = ~m;
+		UN8x4_MUL_UN8x4_ADD_UN8x4 (d, m, s);
+		*pd = d;
 	    }
 
 	    pd++;