2 * downscale lowres luma: full-res buffer to down scale image, and to packed hpel image
6 * fenc_img is an output image (area of memory referenced through a texture
7 * cache). A read of any pixel location (x,y) returns four pixel values:
14 * This is a 4x replication of the lowres pixels, a trade-off between memory
15 * size and read latency.
19 * hpel_planes is an output image that contains the four HPEL planes used for
20 * subpel refinement. A read of any pixel location (x,y) returns a UInt32 with
21 * the four planar values C | V | H | F
23 * launch dimensions: [lowres-width, lowres-height]
25 kernel void downscale_hpel( const global pixel *fenc,
26 write_only image2d_t fenc_img,
27 write_only image2d_t hpel_planes,
30 int x = get_global_id( 0 );
31 int y = get_global_id( 1 );
34 fenc += y * stride * 2;
35 const global pixel *src1 = fenc + stride;
36 const global pixel *src2 = (y == get_global_size( 1 )-1) ? src1 : src1 + stride;
37 int2 pos = (int2)(x, y);
40 right = rhadd( fenc[x*2], src1[x*2] );
41 left = rhadd( fenc[x*2+1], src1[x*2+1] );
42 values.s0 = rhadd( right, left ); // F
44 right = rhadd( fenc[2*x+1], src1[2*x+1] );
45 left = rhadd( fenc[2*x+2], src1[2*x+2] );
46 values.s1 = rhadd( right, left ); // H
48 right = rhadd( src1[2*x], src2[2*x] );
49 left = rhadd( src1[2*x+1], src2[2*x+1] );
50 values.s2 = rhadd( right, left ); // V
52 right = rhadd( src1[2*x+1], src2[2*x+1] );
53 left = rhadd( src1[2*x+2], src2[2*x+2] );
54 values.s3 = rhadd( right, left ); // C
56 uint4 val = (uint4) ((values.s3 & 0xff) << 24) | ((values.s2 & 0xff) << 16) | ((values.s1 & 0xff) << 8) | (values.s0 & 0xff);
57 write_imageui( hpel_planes, pos, val );
59 x = select( x, x+1, x+1 < get_global_size( 0 ) );
60 right = rhadd( fenc[x*2], src1[x*2] );
61 left = rhadd( fenc[x*2+1], src1[x*2+1] );
62 values.s1 = rhadd( right, left );
64 x = select( x, x+1, x+1 < get_global_size( 0 ) );
65 right = rhadd( fenc[x*2], src1[x*2] );
66 left = rhadd( fenc[x*2+1], src1[x*2+1] );
67 values.s2 = rhadd( right, left );
69 x = select( x, x+1, x+1 < get_global_size( 0 ) );
70 right = rhadd( fenc[x*2], src1[x*2] );
71 left = rhadd( fenc[x*2+1], src1[x*2+1] );
72 values.s3 = rhadd( right, left );
74 write_imageui( fenc_img, pos, values );
78 * downscale lowres hierarchical motion search image, copy from one image to
79 * another decimated image. This kernel is called iteratively to generate all
82 * launch dimensions: [lower_res width, lower_res height]
84 kernel void downscale1( read_only image2d_t higher_res, write_only image2d_t lower_res )
86 int x = get_global_id( 0 );
87 int y = get_global_id( 1 );
88 int2 pos = (int2)(x, y);
89 int gs = get_global_size( 0 );
90 uint4 top, bot, values;
91 top = read_imageui( higher_res, sampler, (int2)(x*2, 2*y) );
92 bot = read_imageui( higher_res, sampler, (int2)(x*2, 2*y+1) );
93 values.s0 = rhadd( rhadd( top.s0, bot.s0 ), rhadd( top.s1, bot.s1 ) );
95 /* these select statements appear redundant, and they should be, but tests break when
96 * they are not here. I believe this was caused by a driver bug
98 values.s1 = select( values.s0, rhadd( rhadd( top.s2, bot.s2 ), rhadd( top.s3, bot.s3 ) ), ( x + 1 < gs) );
99 top = read_imageui( higher_res, sampler, (int2)(x*2+4, 2*y) );
100 bot = read_imageui( higher_res, sampler, (int2)(x*2+4, 2*y+1) );
101 values.s2 = select( values.s1, rhadd( rhadd( top.s0, bot.s0 ), rhadd( top.s1, bot.s1 ) ), ( x + 2 < gs ) );
102 values.s3 = select( values.s2, rhadd( rhadd( top.s2, bot.s2 ), rhadd( top.s3, bot.s3 ) ), ( x + 3 < gs ) );
103 write_imageui( lower_res, pos, (uint4)(values) );
107 * Second copy of downscale kernel, no differences. This is a (no perf loss)
108 * workaround for a scheduling bug in current Tahiti drivers. This bug has
109 * theoretically been fixed in the July 2012 driver release from AMD.
111 kernel void downscale2( read_only image2d_t higher_res, write_only image2d_t lower_res )
113 int x = get_global_id( 0 );
114 int y = get_global_id( 1 );
115 int2 pos = (int2)(x, y);
116 int gs = get_global_size( 0 );
117 uint4 top, bot, values;
118 top = read_imageui( higher_res, sampler, (int2)(x*2, 2*y) );
119 bot = read_imageui( higher_res, sampler, (int2)(x*2, 2*y+1) );
120 values.s0 = rhadd( rhadd( top.s0, bot.s0 ), rhadd( top.s1, bot.s1 ) );
122 // see comment in above function copy
123 values.s1 = select( values.s0, rhadd( rhadd( top.s2, bot.s2 ), rhadd( top.s3, bot.s3 ) ), ( x + 1 < gs) );
124 top = read_imageui( higher_res, sampler, (int2)(x*2+4, 2*y) );
125 bot = read_imageui( higher_res, sampler, (int2)(x*2+4, 2*y+1) );
126 values.s2 = select( values.s1, rhadd( rhadd( top.s0, bot.s0 ), rhadd( top.s1, bot.s1 ) ), ( x + 2 < gs ) );
127 values.s3 = select( values.s2, rhadd( rhadd( top.s2, bot.s2 ), rhadd( top.s3, bot.s3 ) ), ( x + 3 < gs ) );
128 write_imageui( lower_res, pos, (uint4)(values) );
131 /* OpenCL 1.2 finally added a memset command, but we're not targeting 1.2 */
132 kernel void memset_int16( global int16_t *buf, int16_t value )
134 buf[get_global_id( 0 )] = value;