[Pixman] [PATCH 1/4] vmx: optimize scaled_nearest_scanline_vmx_8888_8888_OVER

[Oded Gabbay]

This patch optimizes scaled_nearest_scanline_vmx_8888_8888_OVER and all
the functions it calls (combine1, combine4 and
core_combine_over_u_pixel_vmx).

The optimization is done by removing use of expand_alpha_1x128 and
expand_alpha_2x128 in favor of splat_alpha 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 = 24847.3MB/s (6211.8MP/s for 32bpp fills)

                Before          After           Change
              --------------------------------------------
L1              182.05          210.22         +15.47%
L2              180.6           208.92         +15.68%
M               180.52          208.22         +15.34%
HT              130.17          178.97         +37.49%
VT              145.82          184.22         +26.33%
R               104.51          129.38         +23.80%
RT              48.3            61.54          +27.41%
Kops/s          430             504            +17.21%

Signed-off-by: Oded Gabbay <oded.gabbay at gmail.com>
---
 pixman/pixman-vmx.c | 80 ++++++++++++-----------------------------------------
 1 file changed, 18 insertions(+), 62 deletions(-)

diff --git a/pixman/pixman-vmx.c b/pixman/pixman-vmx.c
index a9bd024..d9fc5d6 100644
--- a/pixman/pixman-vmx.c
+++ b/pixman/pixman-vmx.c
@@ -623,10 +623,9 @@ in_over_2x128 (vector unsigned int* src_lo,
 static force_inline uint32_t
 core_combine_over_u_pixel_vmx (uint32_t src, uint32_t dst)
 {
-    uint8_t a;
-    vector unsigned int vmxs;
+    uint32_t a;
 
-    a = src >> 24;
+    a = ALPHA_8(src);
 
     if (a == 0xff)
     {
@@ -634,9 +633,7 @@ core_combine_over_u_pixel_vmx (uint32_t src, uint32_t dst)
     }
     else if (src)
     {
-	vmxs = unpack_32_1x128 (src);
-	return pack_1x128_32(
-		over(vmxs, expand_alpha_1x128 (vmxs), unpack_32_1x128 (dst)));
+	UN8x4_MUL_UN8_ADD_UN8x4(dst, (~a & MASK), src);
     }
 
     return dst;
@@ -646,19 +643,10 @@ static force_inline uint32_t
 combine1 (const uint32_t *ps, const uint32_t *pm)
 {
     uint32_t s = *ps;
+    uint32_t a = ALPHA_8(*pm);
 
     if (pm)
-    {
-	vector unsigned int ms, mm;
-
-	mm = unpack_32_1x128 (*pm);
-	mm = expand_alpha_1x128 (mm);
-
-	ms = unpack_32_1x128 (s);
-	ms = pix_multiply (ms, mm);
-
-	s = pack_1x128_32 (ms);
-    }
+	UN8x4_MUL_UN8(s, a);
 
     return s;
 }
@@ -666,38 +654,22 @@ combine1 (const uint32_t *ps, const uint32_t *pm)
 static force_inline vector unsigned int
 combine4 (const uint32_t* ps, const uint32_t* pm)
 {
-    vector unsigned int vmx_src_lo, vmx_src_hi;
-    vector unsigned int vmx_msk_lo, vmx_msk_hi;
-    vector unsigned int s;
+    vector unsigned int src, msk;
 
     if (pm)
     {
-	vmx_msk_lo = load_128_unaligned(pm);
+	msk = load_128_unaligned(pm);
 
-	if (is_transparent(vmx_msk_lo))
+	if (is_transparent(msk))
 	    return (vector unsigned int) AVV(0);
     }
 
-    s = load_128_unaligned(ps);
+    src = load_128_unaligned(ps);
 
     if (pm)
-    {
-	unpack_128_2x128(s, (vector unsigned int) AVV(0),
-			    &vmx_src_lo, &vmx_src_hi);
-
-	unpack_128_2x128(vmx_msk_lo, (vector unsigned int) AVV(0),
-			    &vmx_msk_lo, &vmx_msk_hi);
-
-	expand_alpha_2x128(vmx_msk_lo, vmx_msk_hi, &vmx_msk_lo, &vmx_msk_hi);
+	src = pix_multiply(src, msk);
 
-	pix_multiply_2x128(&vmx_src_lo, &vmx_src_hi,
-			   &vmx_msk_lo, &vmx_msk_hi,
-			   &vmx_src_lo, &vmx_src_hi);
-
-	s = pack_2x128_128(vmx_src_lo, vmx_src_hi);
-    }
-
-    return s;
+    return src;
 }
 
 static void
@@ -2966,9 +2938,7 @@ scaled_nearest_scanline_vmx_8888_8888_OVER (uint32_t*       pd,
     uint32_t s, d;
     const uint32_t* pm = NULL;
 
-    vector unsigned int vmx_dst_lo, vmx_dst_hi;
-    vector unsigned int vmx_src_lo, vmx_src_hi;
-    vector unsigned int vmx_alpha_lo, vmx_alpha_hi;
+    vector unsigned int vsrc, vdst;
 
     if (fully_transparent_src)
 	return;
@@ -3015,31 +2985,17 @@ scaled_nearest_scanline_vmx_8888_8888_OVER (uint32_t*       pd,
 	tmp[2] = tmp3;
 	tmp[3] = tmp4;
 
-	vmx_src_hi = combine4 ((const uint32_t *) &tmp, pm);
+	vsrc = combine4 ((const uint32_t *) &tmp, pm);
 
-	if (is_opaque (vmx_src_hi))
+	if (is_opaque (vsrc))
 	{
-	    save_128_aligned (pd, vmx_src_hi);
+	    save_128_aligned (pd, vsrc);
 	}
-	else if (!is_zero (vmx_src_hi))
+	else if (!is_zero (vsrc))
 	{
-	    vmx_dst_hi = load_128_aligned (pd);
-
-	    unpack_128_2x128 (vmx_src_hi, (vector unsigned int) AVV(0),
-				&vmx_src_lo, &vmx_src_hi);
-
-	    unpack_128_2x128 (vmx_dst_hi, (vector unsigned int) AVV(0),
-				&vmx_dst_lo, &vmx_dst_hi);
-
-	    expand_alpha_2x128 (
-		vmx_src_lo, vmx_src_hi, &vmx_alpha_lo, &vmx_alpha_hi);
-
-	    over_2x128 (&vmx_src_lo, &vmx_src_hi,
-			&vmx_alpha_lo, &vmx_alpha_hi,
-			&vmx_dst_lo, &vmx_dst_hi);
+	    vdst = over(vsrc, splat_alpha(vsrc), load_128_aligned (pd));
 
-	    /* rebuid the 4 pixel data and save*/
-	    save_128_aligned (pd, pack_2x128_128 (vmx_dst_lo, vmx_dst_hi));
+	    save_128_aligned (pd, vdst);
 	}
 
 	w -= 4;
-- 
2.4.3