| 1 |
/* |
| 2 |
* Copyright (c) 2019, Intel Corporation |
| 3 |
* |
| 4 |
* Permission is hereby granted, free of charge, to any person obtaining a |
| 5 |
* copy of this software and associated documentation files (the "Software"), |
| 6 |
* to deal in the Software without restriction, including without limitation |
| 7 |
* the rights to use, copy, modify, merge, publish, distribute, sublicense, |
| 8 |
* and/or sell copies of the Software, and to permit persons to whom the |
| 9 |
* Software is furnished to do so, subject to the following conditions: |
| 10 |
* |
| 11 |
* The above copyright notice and this permission notice shall be included |
| 12 |
* in all copies or substantial portions of the Software. |
| 13 |
* |
| 14 |
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS |
| 15 |
* OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
| 16 |
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL |
| 17 |
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR |
| 18 |
* OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, |
| 19 |
* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR |
| 20 |
* OTHER DEALINGS IN THE SOFTWARE. |
| 21 |
*/ |
| 22 |
|
| 23 |
#include <cm/cm.h> |
| 24 |
|
| 25 |
#define CURBEDATACTR_SIZE 10 |
| 26 |
|
| 27 |
const uint Stat_offsets[16] = {0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15}; |
| 28 |
static const ushort masktab[] = {0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31}; |
| 29 |
static const uint offset_idx[] = {0,1,2,3,4,5,6,7}; |
| 30 |
|
| 31 |
#define NV12 0 |
| 32 |
#define P010 1 |
| 33 |
#define P210 2 |
| 34 |
#define YUY2 3 |
| 35 |
#define Y210 4 |
| 36 |
#define ARGB 5 |
| 37 |
#define NV12_Linear 6 |
| 38 |
|
| 39 |
#define HEVC_HISTORY_BUFFER_ENTRY_SIZE 32 |
| 40 |
#define HEVC_MT_TASKDATA_BUFFER_ENTRY_SIZE 96 |
| 41 |
|
| 42 |
////////////////////CURBE and VME////////////////////////////////////// |
| 43 |
|
| 44 |
extern "C" _GENX_MAIN_ void DS_Convert( |
| 45 |
vector<uint, CURBEDATACTR_SIZE> CURBEData, |
| 46 |
SurfaceIndex PakSurfIndex, |
| 47 |
SurfaceIndex EncSurfIndex, |
| 48 |
SurfaceIndex DS4XSurf, |
| 49 |
SurfaceIndex MBStats, |
| 50 |
SurfaceIndex DS2XSurf, |
| 51 |
SurfaceIndex Hevc_History_Surface_index, |
| 52 |
SurfaceIndex Hevc_History_Sum_Surface_index, |
| 53 |
SurfaceIndex Scratch_Surface_index |
| 54 |
) |
| 55 |
{ |
| 56 |
ushort h_pos = get_thread_origin_x(); |
| 57 |
ushort v_pos = get_thread_origin_y(); |
| 58 |
|
| 59 |
vector_ref<uchar, 4 * CURBEDATACTR_SIZE> CURBEData_U8 = CURBEData.format<uchar>(); |
| 60 |
vector_ref<ushort, 2 * CURBEDATACTR_SIZE> CURBEData_U16 = CURBEData.format<ushort>(); |
| 61 |
|
| 62 |
uchar ENCChromaDepth = CURBEData_U8(2); |
| 63 |
uchar ENCLumaDepth = CURBEData_U8(3) & 0x7F; |
| 64 |
uchar colorformat = CURBEData_U8(4); |
| 65 |
vector<ushort, 1>uWidth = CURBEData_U8.format<ushort>().select<1, 1>(4); |
| 66 |
vector<ushort, 1>uHeight = CURBEData_U8.format<ushort>().select<1, 1>(5); |
| 67 |
|
| 68 |
uint height, width; |
| 69 |
|
| 70 |
uchar bflagConvert = CURBEData_U8(5) & 0x01; |
| 71 |
uchar stage = (CURBEData_U8(5) >> 1) & 0x7; |
| 72 |
uchar MBStatflag = (CURBEData_U8(5) >> 4) & 0x1; |
| 73 |
|
| 74 |
if(stage == 3) |
| 75 |
{ |
| 76 |
height = (((uHeight(0) >> 2) + 31) / 32) * 32; |
| 77 |
width = (((uWidth(0) >> 2) + 31) / 32) * 32; |
| 78 |
} |
| 79 |
else |
| 80 |
{ |
| 81 |
height = uHeight(0); |
| 82 |
width = uWidth(0); |
| 83 |
} |
| 84 |
|
| 85 |
uint Flatness_Threshold = CURBEData(3); |
| 86 |
|
| 87 |
// CSC Coefficients |
| 88 |
ushort CSC0 = CURBEData_U16(8); |
| 89 |
ushort CSC1 = CURBEData_U16(9); |
| 90 |
ushort CSC2 = CURBEData_U16(10); |
| 91 |
ushort CSC3 = CURBEData_U16(11); |
| 92 |
ushort CSC4 = CURBEData_U16(12); |
| 93 |
ushort CSC5 = CURBEData_U16(13); |
| 94 |
ushort CSC6 = CURBEData_U16(14); |
| 95 |
ushort CSC7 = CURBEData_U16(15); |
| 96 |
ushort CSC8 = CURBEData_U16(16); |
| 97 |
ushort CSC9 = CURBEData_U16(17); |
| 98 |
ushort CSC10 = CURBEData_U16(18); |
| 99 |
ushort CSC11 = CURBEData_U16(19); |
| 100 |
|
| 101 |
// Chroma Siting Location |
| 102 |
uchar chroma_siting_location = CURBEData_U8(6); |
| 103 |
|
| 104 |
vector<uint, 1> numLCUs; |
| 105 |
matrix<uchar, 32, 32> YPix; |
| 106 |
|
| 107 |
vector<ushort, 1> replicate; |
| 108 |
replicate[0] = 0; |
| 109 |
ushort h_pos_write = h_pos << 3; |
| 110 |
ushort v_pos_write = v_pos << 3; |
| 111 |
|
| 112 |
ushort h_pos_read = h_pos_write; |
| 113 |
vector<ushort, 1> v_pos_read = v_pos_write; |
| 114 |
vector<ushort, 1> max_dst_row = v_pos_write + 8; |
| 115 |
vector<ushort, 1> max_real_dst_row = (height >> 2); |
| 116 |
|
| 117 |
if (stage == 3) |
| 118 |
{ |
| 119 |
max_real_dst_row = uHeight(0) >> 4; |
| 120 |
} |
| 121 |
|
| 122 |
replicate.merge((max_dst_row - max_real_dst_row), (max_dst_row > max_real_dst_row)); |
| 123 |
|
| 124 |
v_pos_read.merge(((max_real_dst_row << 2) - 4) >> 2, v_pos_write >= max_real_dst_row); |
| 125 |
replicate.merge(7, v_pos_write >= max_real_dst_row); |
| 126 |
|
| 127 |
int final_h_pos = h_pos_read << 2; |
| 128 |
int final_h_pos_plus_32 = 2 * final_h_pos + 32; |
| 129 |
|
| 130 |
int final_v_pos = v_pos_read(0) << 2; |
| 131 |
int final_v_pos_c = final_v_pos >> 1 ; |
| 132 |
|
| 133 |
vector<ushort, 1> replicate2; |
| 134 |
replicate2[0] = 0; |
| 135 |
|
| 136 |
ushort h_pos_write2 = h_pos << 4; |
| 137 |
ushort v_pos_write2 = v_pos << 4; |
| 138 |
ushort h_pos_read2 = h_pos_write2; |
| 139 |
|
| 140 |
vector<ushort, 1> v_pos_read2 = v_pos_write2; |
| 141 |
vector<ushort, 1> max_dst_row2 = v_pos_write2 + 16; |
| 142 |
|
| 143 |
vector<ushort, 1> max_real_dst_row2 = (height >> 1); |
| 144 |
|
| 145 |
if(stage==3) |
| 146 |
{ |
| 147 |
max_real_dst_row2 = uHeight(0) >> 5; |
| 148 |
} |
| 149 |
|
| 150 |
replicate2.merge((max_dst_row2 - max_real_dst_row2), (max_dst_row2 > max_real_dst_row2)); |
| 151 |
|
| 152 |
v_pos_read2.merge(max_real_dst_row2 - 1, v_pos_write2 >= max_real_dst_row2); |
| 153 |
replicate2.merge(15, v_pos_write2 >= max_real_dst_row2); |
| 154 |
|
| 155 |
v_pos_read2[0] = v_pos_read2[0] << 1; |
| 156 |
|
| 157 |
uint final_h_pos_argb = 0; |
| 158 |
uchar quot = 0, first_index_val = 0, first_index = 0; |
| 159 |
|
| 160 |
if(bflagConvert) |
| 161 |
{ |
| 162 |
final_h_pos_argb = final_h_pos; |
| 163 |
if(final_h_pos == width) |
| 164 |
{ |
| 165 |
final_h_pos_argb = final_h_pos - 32; |
| 166 |
} |
| 167 |
|
| 168 |
quot = (final_h_pos + 31) / width; |
| 169 |
first_index_val = (final_h_pos + 31) % width; |
| 170 |
first_index = first_index_val; |
| 171 |
|
| 172 |
if((first_index_val > 32) && (quot > 0)) |
| 173 |
{ |
| 174 |
final_h_pos_argb = final_h_pos - (first_index_val / 32) * 32; |
| 175 |
first_index = first_index_val - (first_index_val / 32) * 32; |
| 176 |
|
| 177 |
if(final_h_pos_argb == width) |
| 178 |
{ |
| 179 |
final_h_pos_argb = final_h_pos_argb - 32; |
| 180 |
} |
| 181 |
} |
| 182 |
|
| 183 |
if(colorformat == P010) |
| 184 |
{ |
| 185 |
matrix<ushort, 32, 32> YPix_16; |
| 186 |
vector<ushort, 1> max_col = h_pos_write; |
| 187 |
vector<ushort, 1> max_real_col = width >> 2; |
| 188 |
vector<ushort, 1> replicate_col(0); |
| 189 |
|
| 190 |
replicate_col.merge(1, (max_col >= max_real_col)); |
| 191 |
if(replicate_col[0]) |
| 192 |
{ |
| 193 |
read_plane(PakSurfIndex, GENX_SURFACE_Y_PLANE, 2 * width - 2, final_v_pos, YPix_16.select<32, 1, 4, 1>(0, 0)); |
| 194 |
YPix_16.select<32, 1, 4, 1>(0, 4) = YPix_16.select<32, 1, 4, 1>(0, 0); |
| 195 |
YPix_16.select<32, 1, 8, 1>(0, 8) = YPix_16.select<32, 1, 8, 1>(0, 0); |
| 196 |
YPix_16.select<32, 1, 16, 1>(0, 16) = YPix_16.select<32, 1, 16,1>(0, 0); |
| 197 |
} |
| 198 |
else |
| 199 |
{ |
| 200 |
read_plane(PakSurfIndex, GENX_SURFACE_Y_PLANE, 2 * final_h_pos, final_v_pos, YPix_16.select<8, 1, 16, 1>(0, 0)); |
| 201 |
read_plane(PakSurfIndex, GENX_SURFACE_Y_PLANE, final_h_pos_plus_32, final_v_pos, YPix_16.select<8, 1, 16, 1>(0, 16)); |
| 202 |
read_plane(PakSurfIndex, GENX_SURFACE_Y_PLANE, 2 * final_h_pos, final_v_pos + 8, YPix_16.select<8, 1, 16, 1>(8, 0)); |
| 203 |
read_plane(PakSurfIndex, GENX_SURFACE_Y_PLANE, final_h_pos_plus_32, final_v_pos + 8, YPix_16.select<8, 1, 16, 1>(8, 16)); |
| 204 |
read_plane(PakSurfIndex, GENX_SURFACE_Y_PLANE, 2 * final_h_pos, final_v_pos + 16, YPix_16.select<8, 1, 16, 1>(16, 0)); |
| 205 |
read_plane(PakSurfIndex, GENX_SURFACE_Y_PLANE, final_h_pos_plus_32, final_v_pos + 16, YPix_16.select<8, 1, 16, 1>(16, 16)); |
| 206 |
read_plane(PakSurfIndex, GENX_SURFACE_Y_PLANE, 2 * final_h_pos, final_v_pos + 24, YPix_16.select<8, 1, 16, 1>(24, 0)); |
| 207 |
read_plane(PakSurfIndex, GENX_SURFACE_Y_PLANE, final_h_pos_plus_32, final_v_pos + 24, YPix_16.select<8, 1, 16, 1>(24, 16)); |
| 208 |
} |
| 209 |
|
| 210 |
uint shiftval = 16 - ENCLumaDepth; |
| 211 |
uint tempval = 1 << (shiftval - 1); |
| 212 |
|
| 213 |
vector<ushort, 32> roundval; |
| 214 |
vector<uchar, 32> roundingenable = 1; |
| 215 |
|
| 216 |
roundval.merge(tempval, 0, roundingenable); |
| 217 |
|
| 218 |
vector<ushort, 32> temptemp; |
| 219 |
int cnt; |
| 220 |
for(cnt=0; cnt < 32; cnt++) |
| 221 |
{ |
| 222 |
temptemp = YPix_16.row(cnt); |
| 223 |
temptemp.merge(temptemp, 0xFF00, temptemp < 0xFF00); |
| 224 |
YPix.row(cnt) = cm_shr<uchar>(temptemp + tempval, shiftval); |
| 225 |
} |
| 226 |
|
| 227 |
matrix_ref<ushort, 16, 32> UVPix_16 = YPix_16.select<16, 1, 32, 1>(0, 0); |
| 228 |
matrix<uchar, 16, 32> UVPix; |
| 229 |
|
| 230 |
read_plane(PakSurfIndex, GENX_SURFACE_UV_PLANE, 2 * final_h_pos, final_v_pos_c, UVPix_16.select<8, 1, 16, 1>(0, 0)); |
| 231 |
read_plane(PakSurfIndex, GENX_SURFACE_UV_PLANE, final_h_pos_plus_32, final_v_pos_c, UVPix_16.select<8, 1, 16, 1>(0, 16)); |
| 232 |
read_plane(PakSurfIndex, GENX_SURFACE_UV_PLANE, 2 * final_h_pos, final_v_pos_c + 8, UVPix_16.select<8, 1, 16, 1>(8, 0)); |
| 233 |
read_plane(PakSurfIndex, GENX_SURFACE_UV_PLANE, final_h_pos_plus_32, final_v_pos_c + 8, UVPix_16.select<8, 1, 16, 1>(8, 16)); |
| 234 |
|
| 235 |
shiftval = 16 - ENCChromaDepth; |
| 236 |
tempval = 1 << (shiftval - 1); |
| 237 |
roundval.merge(tempval, 0, roundingenable); |
| 238 |
|
| 239 |
for(cnt = 0; cnt < 16; cnt++) |
| 240 |
{ |
| 241 |
temptemp = UVPix_16.row(cnt); |
| 242 |
temptemp.merge(temptemp, 0xFF00, temptemp < 0xFF00); |
| 243 |
UVPix.row(cnt) = cm_shr<uchar>(temptemp + tempval, shiftval); |
| 244 |
} |
| 245 |
|
| 246 |
write_plane(EncSurfIndex, GENX_SURFACE_UV_PLANE, final_h_pos, final_v_pos_c, UVPix.select<8, 1, 32, 1>(0, 0)); |
| 247 |
write_plane(EncSurfIndex, GENX_SURFACE_UV_PLANE, final_h_pos, final_v_pos_c + 8, UVPix.select<8, 1, 32, 1>(8, 0)); |
| 248 |
|
| 249 |
write_plane(EncSurfIndex, GENX_SURFACE_Y_PLANE, final_h_pos, final_v_pos, YPix.select<8, 1, 32, 1>(0, 0)); |
| 250 |
write_plane(EncSurfIndex, GENX_SURFACE_Y_PLANE, final_h_pos, final_v_pos + 8, YPix.select<8, 1, 32, 1>(8, 0)); |
| 251 |
write_plane(EncSurfIndex, GENX_SURFACE_Y_PLANE, final_h_pos, final_v_pos + 16, YPix.select<8, 1, 32, 1>(16,0)); |
| 252 |
write_plane(EncSurfIndex, GENX_SURFACE_Y_PLANE, final_h_pos, final_v_pos + 24, YPix.select<8, 1, 32, 1>(24,0)); |
| 253 |
|
| 254 |
if (!stage) |
| 255 |
{ |
| 256 |
cm_fence(); |
| 257 |
return; |
| 258 |
} |
| 259 |
} |
| 260 |
else if (colorformat == Y210) |
| 261 |
{ |
| 262 |
matrix<ushort, 8, 32> YUV_Data16_10bit; |
| 263 |
matrix<uchar, 8, 32> UVPix_422; |
| 264 |
|
| 265 |
read (PakSurfIndex, 4 * final_h_pos_argb, final_v_pos, YUV_Data16_10bit.select<8, 1, 16, 1>(0, 0)); |
| 266 |
read (PakSurfIndex, 4 * final_h_pos_argb + 32, final_v_pos, YUV_Data16_10bit.select<8, 1, 16, 1>(0, 16)); |
| 267 |
|
| 268 |
YUV_Data16_10bit.merge(YUV_Data16_10bit, 0xFF00, YUV_Data16_10bit < 0xFF00); |
| 269 |
|
| 270 |
YPix.select<8, 1, 16, 1>(0, 0) = cm_shr<uchar>(YUV_Data16_10bit.select<8, 1, 16, 2>(0, 0) + 0x80, 8); |
| 271 |
UVPix_422.select<8, 1, 16, 1>(0, 0) = cm_shr<uchar>(YUV_Data16_10bit.select<8, 1, 16, 2>(0, 1) + 0x80, 8); |
| 272 |
|
| 273 |
read (PakSurfIndex, 4 * final_h_pos_argb+64, final_v_pos, YUV_Data16_10bit.select<8, 1, 16, 1>(0, 0)); |
| 274 |
read (PakSurfIndex, 4 * final_h_pos_argb+96, final_v_pos, YUV_Data16_10bit.select<8, 1, 16, 1>(0, 16)); |
| 275 |
|
| 276 |
YUV_Data16_10bit.merge(YUV_Data16_10bit, 0xFF00, YUV_Data16_10bit < 0xFF00); |
| 277 |
YPix.select<8, 1, 16, 1>(0, 16) = cm_shr<uchar>(YUV_Data16_10bit.select<8, 1, 16, 2>(0, 0) + 0x80, 8); |
| 278 |
UVPix_422.select<8,1,16,1>(0,16) = cm_shr<uchar>(YUV_Data16_10bit.select<8, 1, 16, 2>(0, 1) + 0x80, 8); |
| 279 |
|
| 280 |
if((final_v_pos + 8) <= height) |
| 281 |
{ |
| 282 |
write(EncSurfIndex, final_h_pos, final_v_pos, YPix.select<8, 1, 32, 1>(0, 0)); |
| 283 |
} |
| 284 |
write(EncSurfIndex, final_h_pos, height + final_v_pos, UVPix_422 ); |
| 285 |
|
| 286 |
read (PakSurfIndex, 4 * final_h_pos_argb, final_v_pos + 8, YUV_Data16_10bit.select<8, 1, 16, 1>(0, 0)); |
| 287 |
read (PakSurfIndex, 4 * final_h_pos_argb + 32, final_v_pos + 8, YUV_Data16_10bit.select<8, 1, 16, 1>(0, 16)); |
| 288 |
|
| 289 |
YUV_Data16_10bit.merge(YUV_Data16_10bit, 0xFF00, YUV_Data16_10bit < 0xFF00); |
| 290 |
YPix.select<8, 1, 16, 1>(8, 0) = cm_shr<uchar>(YUV_Data16_10bit.select<8, 1, 16, 2>(0, 0) + 0x80, 8); |
| 291 |
UVPix_422.select<8,1,16,1>(0, 0) = cm_shr<uchar>(YUV_Data16_10bit.select<8, 1, 16, 2>(0, 1) + 0x80, 8); |
| 292 |
|
| 293 |
read (PakSurfIndex, 4 * final_h_pos_argb + 64, final_v_pos + 8, YUV_Data16_10bit.select<8, 1, 16, 1>(0, 0)); |
| 294 |
read (PakSurfIndex, 4 * final_h_pos_argb + 96, final_v_pos + 8, YUV_Data16_10bit.select<8, 1, 16, 1>(0, 16)); |
| 295 |
|
| 296 |
YUV_Data16_10bit.merge(YUV_Data16_10bit, 0xFF00, YUV_Data16_10bit < 0xFF00); |
| 297 |
YPix.select<8, 1, 16, 1>(8, 16) = cm_shr<uchar>(YUV_Data16_10bit.select<8, 1, 16, 2>(0, 0) + 0x80, 8); |
| 298 |
UVPix_422.select<8,1,16,1>(0, 16) = cm_shr<uchar>(YUV_Data16_10bit.select<8, 1, 16, 2>(0, 1) + 0x80, 8); |
| 299 |
|
| 300 |
if( (final_v_pos + 16) <= height) |
| 301 |
{ |
| 302 |
write(EncSurfIndex, final_h_pos, final_v_pos + 8, YPix.select<8, 1, 32, 1>(8, 0) ); |
| 303 |
} |
| 304 |
write(EncSurfIndex, final_h_pos, height + final_v_pos + 8, UVPix_422 ); |
| 305 |
|
| 306 |
read (PakSurfIndex, 4 * final_h_pos_argb, final_v_pos + 16, YUV_Data16_10bit.select<8, 1, 16, 1>(0, 0)); |
| 307 |
read (PakSurfIndex, 4 * final_h_pos_argb + 32, final_v_pos + 16, YUV_Data16_10bit.select<8, 1, 16, 1>(0, 16)); |
| 308 |
|
| 309 |
YUV_Data16_10bit.merge(YUV_Data16_10bit, 0xFF00, YUV_Data16_10bit < 0xFF00); |
| 310 |
YPix.select<8, 1, 16, 1>(16, 0) = cm_shr<uchar>(YUV_Data16_10bit.select<8, 1, 16, 2>(0, 0) + 0x80, 8); |
| 311 |
UVPix_422.select<8, 1, 16, 1>(0, 0) = cm_shr<uchar>(YUV_Data16_10bit.select<8, 1, 16, 2>(0, 1) + 0x80, 8); |
| 312 |
|
| 313 |
read (PakSurfIndex, 4 * final_h_pos_argb + 64, final_v_pos+16, YUV_Data16_10bit.select<8, 1, 16, 1>(0, 0)); |
| 314 |
read (PakSurfIndex, 4 * final_h_pos_argb + 96, final_v_pos+16, YUV_Data16_10bit.select<8, 1, 16, 1>(0, 16)); |
| 315 |
|
| 316 |
YUV_Data16_10bit.merge(YUV_Data16_10bit, 0xFF00, YUV_Data16_10bit < 0xFF00); |
| 317 |
YPix.select<8, 1, 16, 1>(16, 16) = cm_shr<uchar>(YUV_Data16_10bit.select<8, 1, 16, 2>(0, 0) + 0x80, 8); |
| 318 |
UVPix_422.select<8,1,16,1>(0, 16) = cm_shr<uchar>(YUV_Data16_10bit.select<8, 1, 16, 2>(0, 1) + 0x80, 8); |
| 319 |
|
| 320 |
if( (final_v_pos + 24) <= height) |
| 321 |
{ |
| 322 |
write(EncSurfIndex, final_h_pos, final_v_pos+16, YPix.select<8, 1, 32, 1>(16, 0) ); |
| 323 |
} |
| 324 |
write(EncSurfIndex, final_h_pos, height + final_v_pos + 16, UVPix_422 ); |
| 325 |
|
| 326 |
read (PakSurfIndex, 4 * final_h_pos_argb, final_v_pos + 24, YUV_Data16_10bit.select<8, 1, 16, 1>(0, 0)); |
| 327 |
read (PakSurfIndex, 4 * final_h_pos_argb + 32, final_v_pos + 24, YUV_Data16_10bit.select<8, 1, 16, 1>(0, 16)); |
| 328 |
|
| 329 |
YUV_Data16_10bit.merge(YUV_Data16_10bit, 0xFF00, YUV_Data16_10bit < 0xFF00); |
| 330 |
YPix.select<8, 1, 16, 1>(24, 0) = cm_shr<uchar>(YUV_Data16_10bit.select<8, 1, 16, 2>(0, 0) + 0x80, 8); |
| 331 |
UVPix_422.select<8, 1, 16, 1>(0, 0) = cm_shr<uchar>(YUV_Data16_10bit.select<8, 1, 16, 2>(0, 1) + 0x80, 8); |
| 332 |
|
| 333 |
read (PakSurfIndex, 4 * final_h_pos_argb + 64, final_v_pos + 24, YUV_Data16_10bit.select<8, 1, 16, 1>(0, 0)); |
| 334 |
read (PakSurfIndex, 4 * final_h_pos_argb + 96, final_v_pos + 24, YUV_Data16_10bit.select<8, 1, 16, 1>(0, 16)); |
| 335 |
|
| 336 |
YUV_Data16_10bit.merge(YUV_Data16_10bit, 0xFF00, YUV_Data16_10bit < 0xFF00); |
| 337 |
YPix.select<8, 1, 16, 1>(24, 16) = cm_shr<uchar>(YUV_Data16_10bit.select<8, 1, 16, 2>(0, 0) + 0x80, 8); |
| 338 |
UVPix_422.select<8, 1, 16, 1>(0, 16) = cm_shr<uchar>(YUV_Data16_10bit.select<8, 1, 16, 2>(0, 1) + 0x80, 8); |
| 339 |
|
| 340 |
if( (final_v_pos + 32) <= height) |
| 341 |
{ |
| 342 |
write(EncSurfIndex, final_h_pos, final_v_pos + 24, YPix.select<8, 1, 32, 1>(24, 0) ); |
| 343 |
} |
| 344 |
write(EncSurfIndex, final_h_pos, height + final_v_pos+24, UVPix_422 ); |
| 345 |
} |
| 346 |
else if(colorformat == ARGB) |
| 347 |
{ |
| 348 |
matrix<uchar, 16, 32> ARGB_Data; |
| 349 |
matrix_ref<uchar, 8, 32> c1 = ARGB_Data.select<8,1,32,1>(0, 0); |
| 350 |
matrix<uchar, 16, 32> r; |
| 351 |
matrix<uchar, 16, 32> g; |
| 352 |
matrix<uchar, 16, 32> b; |
| 353 |
|
| 354 |
for(uint i = 0; i < 2; i++) |
| 355 |
{ |
| 356 |
#pragma unroll |
| 357 |
for( int j = 0; j < 4;j++) |
| 358 |
{ |
| 359 |
read(PakSurfIndex, 4 * final_h_pos_argb + 32 * j, final_v_pos + i * 16, c1); |
| 360 |
r.select<8, 1, 8, 1>(0, 8 * j) = c1.select<8, 1, 8, 4>(0, 2); |
| 361 |
g.select<8, 1, 8, 1>(0, 8 * j) = c1.select<8, 1, 8, 4>(0, 1); |
| 362 |
b.select<8, 1, 8, 1>(0, 8 * j) = c1.select<8, 1, 8, 4>(0, 0); |
| 363 |
} |
| 364 |
|
| 365 |
#pragma unroll |
| 366 |
for( int j = 0; j < 4;j++) |
| 367 |
{ |
| 368 |
read(PakSurfIndex, 4 * final_h_pos_argb + 32 * j, final_v_pos + i * 16 + 8, c1); |
| 369 |
r.select<8,1,8,1>(8,8*j) = c1.select<8, 1, 8, 4>(0, 2); |
| 370 |
g.select<8,1,8,1>(8,8*j) = c1.select<8, 1, 8, 4>(0, 1); |
| 371 |
b.select<8,1,8,1>(8,8*j) = c1.select<8, 1, 8, 4>(0, 0); |
| 372 |
} |
| 373 |
|
| 374 |
if(quot > 0) |
| 375 |
{ |
| 376 |
vector<uchar, 32> replicants; |
| 377 |
vector<ushort, 32> masktable = 1; |
| 378 |
|
| 379 |
vector<ushort, 32> final_h_pos_mask(masktab); |
| 380 |
final_h_pos_mask = final_h_pos_mask + final_h_pos; |
| 381 |
|
| 382 |
masktable.merge(masktable, 0, final_h_pos_mask < width); |
| 383 |
|
| 384 |
for(uint mcount=0;mcount<16;mcount++) |
| 385 |
{ |
| 386 |
replicants = r(mcount,first_index); |
| 387 |
r.row(mcount).merge(r.row(mcount), replicants, masktable); |
| 388 |
|
| 389 |
replicants = g(mcount,first_index); |
| 390 |
g.row(mcount).merge(g.row(mcount), replicants, masktable); |
| 391 |
|
| 392 |
replicants = b(mcount,first_index); |
| 393 |
b.row(mcount).merge(b.row(mcount), replicants, masktable); |
| 394 |
} |
| 395 |
} |
| 396 |
|
| 397 |
vector<uint, 32> temp; |
| 398 |
matrix_ref<uchar, 16, 32> v = YPix.select<16, 1, 32, 1>(16, 0).format<uchar, 16, 32>(); |
| 399 |
|
| 400 |
#pragma unroll |
| 401 |
for(uint k = 0; k < 16; k++) |
| 402 |
{ |
| 403 |
temp = g.row(k) * CSC8; |
| 404 |
temp = temp + r.row(k) * CSC10; |
| 405 |
|
| 406 |
temp = (temp + b.row(k) * CSC9) >> 7; |
| 407 |
ARGB_Data.row(k) = temp.select<32,1>(0) + CSC11; |
| 408 |
} |
| 409 |
|
| 410 |
#pragma unroll |
| 411 |
for(uint k=0; k<16; k++) |
| 412 |
{ |
| 413 |
temp = r.row(k) * CSC2; |
| 414 |
temp = temp + g.row(k) * CSC0; |
| 415 |
temp = (temp + b.row(k) * CSC1) >> 7; |
| 416 |
v.row(k) = temp.select<32,1>(0) + CSC3; |
| 417 |
} |
| 418 |
|
| 419 |
matrix<ushort, 4, 16> first_sum; |
| 420 |
if (chroma_siting_location == 1) |
| 421 |
{ |
| 422 |
|
| 423 |
#pragma unroll |
| 424 |
for(uint k = 0; k < 2; k++) |
| 425 |
{ |
| 426 |
first_sum.select<4, 1, 16, 1>(0, 0) = ARGB_Data.select<4, 2, 16, 2>(8 * k, 0) + ARGB_Data.select<4, 2, 16, 2>(8 * k, 1); |
| 427 |
first_sum.select<4, 1, 16, 1>(0, 0) += 1; |
| 428 |
ARGB_Data.select<4, 1, 16, 2>(4 * k, 0) = first_sum.select<4, 1, 16, 1>(0, 0) >> 1; |
| 429 |
} |
| 430 |
|
| 431 |
#pragma unroll |
| 432 |
for(uint k=0; k<2; k++) |
| 433 |
{ |
| 434 |
first_sum.select<4, 1, 16, 1>(0, 0) = v.select<4, 2, 16, 2>(8 * k, 0) + v.select<4, 2, 16, 2>(8 * k, 1); |
| 435 |
first_sum.select<4, 1, 16, 1>(0, 0) += 1; |
| 436 |
ARGB_Data.select<4, 1, 16, 2>(4 * k, 1) = first_sum.select<4, 1, 16, 1>(0, 0) >> 1; |
| 437 |
} |
| 438 |
} |
| 439 |
else if (chroma_siting_location == 2) |
| 440 |
{ |
| 441 |
#pragma unroll |
| 442 |
for(uint k = 0; k < 2; k++) |
| 443 |
{ |
| 444 |
first_sum.select<4, 1, 16, 1>(0, 0) = ARGB_Data.select<4, 2, 16, 2>(8 * k + 1, 0) + ARGB_Data.select<4, 2, 16, 2>(8 * k + 1, 1); |
| 445 |
first_sum.select<4, 1, 16, 1>(0, 0) += 1; |
| 446 |
ARGB_Data.select<4, 1, 16, 2>(4 * k, 0) = first_sum.select<4, 1, 16, 1>(0, 0) >> 1; |
| 447 |
} |
| 448 |
|
| 449 |
#pragma unroll |
| 450 |
for(uint k=0; k<2; k++) |
| 451 |
{ |
| 452 |
first_sum.select<4, 1, 16, 1>(0, 0) = v.select<4, 2, 16, 2>(8 * k + 1, 0) + v.select<4, 2, 16, 2>(8 * k + 1, 1); |
| 453 |
first_sum.select<4, 1, 16, 1>(0, 0) += 1; |
| 454 |
ARGB_Data.select<4, 1, 16, 2>(4 * k, 1) = first_sum.select<4, 1, 16, 1>(0, 0) >> 1; |
| 455 |
} |
| 456 |
} |
| 457 |
else if (chroma_siting_location == 3) |
| 458 |
{ |
| 459 |
ARGB_Data.select<8, 1, 16, 2>(0, 0) = ARGB_Data.select<8, 2, 16, 2>(0, 0); |
| 460 |
ARGB_Data.select<8, 1, 16, 2>(0, 1) = v.select<8, 2, 16, 2>(0, 0); |
| 461 |
} |
| 462 |
else if (chroma_siting_location == 4) |
| 463 |
{ |
| 464 |
#pragma unroll |
| 465 |
for(uint k=0; k<2; k++) |
| 466 |
{ |
| 467 |
first_sum.select<4, 1, 16, 1>(0, 0) = ARGB_Data.select<4, 2, 16, 2>(8 * k, 0) + ARGB_Data.select<4, 2, 16, 2>(8 * k + 1, 0); |
| 468 |
first_sum.select<4, 1, 16, 1>(0, 0) += 1; |
| 469 |
ARGB_Data.select<4, 1, 16, 2>(4 * k, 0) = first_sum.select<4, 1, 16, 1>(0, 0) >> 1; |
| 470 |
} |
| 471 |
|
| 472 |
#pragma unroll |
| 473 |
for(uint k=0; k<2; k++) |
| 474 |
{ |
| 475 |
first_sum.select<4, 1, 16, 1>(0, 0) = v.select<4, 2, 16, 2>(8 * k, 0) + v.select<4, 2, 16, 2>(8 * k + 1, 0); |
| 476 |
first_sum.select<4, 1, 16, 1>(0, 0) += 1; |
| 477 |
ARGB_Data.select<4, 1, 16, 2>(4 * k, 1) = first_sum.select<4, 1, 16, 1>(0, 0) >> 1; |
| 478 |
} |
| 479 |
} |
| 480 |
else if (chroma_siting_location == 5) |
| 481 |
{ |
| 482 |
ARGB_Data.select<8, 1, 16, 2>(0, 0) = ARGB_Data.select<8, 2, 16, 2>(1, 0); |
| 483 |
ARGB_Data.select<8, 1, 16, 2>(0, 1) = v.select<8, 2, 16, 2>(1, 0); |
| 484 |
} |
| 485 |
else |
| 486 |
{ |
| 487 |
|
| 488 |
#pragma unroll |
| 489 |
for(uint k = 0; k < 4; k++) |
| 490 |
{ |
| 491 |
first_sum = ARGB_Data.select<4, 1, 16, 2>(4 * k, 0) + ARGB_Data.select<4, 1, 16, 2>(4 * k, 1); |
| 492 |
|
| 493 |
first_sum.select<2, 1, 16, 1>(0, 0) = first_sum.select<2, 2, 16, 1>(0, 0) + first_sum.select<2, 2, 16, 1>(1, 0); |
| 494 |
first_sum.select<2, 1, 16, 1>(0, 0) += 2; |
| 495 |
ARGB_Data.select<2, 1, 16, 2>(2*k, 0) = first_sum.select<2, 1, 16, 1>(0, 0) >> 2; |
| 496 |
} |
| 497 |
|
| 498 |
#pragma unroll |
| 499 |
for(uint k=0; k<4; k++) |
| 500 |
{ |
| 501 |
first_sum = v.select<4, 1, 16, 2>(4 * k, 0) + v.select<4, 1, 16, 2>(4 * k, 1); |
| 502 |
|
| 503 |
first_sum.select<2, 1, 16, 1>(0, 0) = first_sum.select<2, 2, 16, 1>(0, 0) + first_sum.select<2, 2, 16, 1>(1, 0); |
| 504 |
first_sum.select<2, 1, 16, 1>(0, 0) += 2; |
| 505 |
ARGB_Data.select<2, 1, 16, 2>(2 * k, 1) = first_sum.select<2, 1, 16, 1>(0, 0) >> 2; |
| 506 |
} |
| 507 |
} |
| 508 |
|
| 509 |
write_plane(EncSurfIndex, GENX_SURFACE_UV_PLANE, final_h_pos, (final_v_pos >> 1) + i * 8, ARGB_Data.select<8, 1, 32, 1>(0, 0)); |
| 510 |
|
| 511 |
#pragma unroll |
| 512 |
for(uint k=0; k < 16; k++) |
| 513 |
{ |
| 514 |
temp = r.row(k) * CSC6; |
| 515 |
temp = temp + g.row(k) * CSC4; |
| 516 |
temp = (temp + b.row(k) * CSC5) >> 7; |
| 517 |
ARGB_Data.row(k) = temp.select<32,1>(0) + CSC7; |
| 518 |
} |
| 519 |
|
| 520 |
write_plane(EncSurfIndex, GENX_SURFACE_Y_PLANE, final_h_pos, final_v_pos + i*16, ARGB_Data.select<8, 1, 32, 1>(0, 0)); |
| 521 |
write_plane(EncSurfIndex, GENX_SURFACE_Y_PLANE, final_h_pos, final_v_pos + i*16 + 8, ARGB_Data.select<8, 1, 32, 1>(8, 0)); |
| 522 |
YPix.select<16,1,32,1>(0+i*16,0) = ARGB_Data.select<16,1,32,1>(0,0); |
| 523 |
} |
| 524 |
|
| 525 |
if (!stage) |
| 526 |
{ |
| 527 |
cm_fence(); |
| 528 |
return; |
| 529 |
} |
| 530 |
} |
| 531 |
else if (colorformat == YUY2) |
| 532 |
{ |
| 533 |
matrix<uchar, 8, 64> YUV_Data8_8bit; |
| 534 |
matrix<ushort, 4, 32> temp = 0; |
| 535 |
|
| 536 |
uchar YUY2convertflag = (CURBEData_U8(5) >> 5) & 0x1; |
| 537 |
|
| 538 |
for(int i=0; i<2; i++) |
| 539 |
{ |
| 540 |
read (PakSurfIndex, 2 * final_h_pos_argb, final_v_pos + i * 16, YUV_Data8_8bit.select<8, 1, 32, 1>(0, 0)); |
| 541 |
read (PakSurfIndex, 2 * final_h_pos_argb + 32, final_v_pos + i * 16, YUV_Data8_8bit.select<8, 1, 32, 1>(0, 32)); |
| 542 |
|
| 543 |
YPix.select<8, 1, 32, 1>(i * 16, 0) = YUV_Data8_8bit.select<8, 1, 32, 2>(0, 0); |
| 544 |
|
| 545 |
if(YUY2convertflag) |
| 546 |
{ |
| 547 |
matrix_ref<uchar, 8, 32> uv_channel = YUV_Data8_8bit.select<8, 1, 32, 2>(0, 1); |
| 548 |
|
| 549 |
temp.select<4, 1, 32, 1>(0, 0) = uv_channel.select<4, 2, 32, 1>(0, 0) + uv_channel.select<4, 2, 32, 1>(1, 0); |
| 550 |
temp.select<4, 1, 32, 1>(0, 0) += 1; |
| 551 |
temp.select<4, 1, 32, 1>(0, 0) = temp.select<4, 1, 32, 1>(0, 0) >> 1; |
| 552 |
|
| 553 |
matrix_ref <uchar, 4, 64> uv_temp = temp.format<uchar, 4, 64>(); |
| 554 |
matrix_ref <uchar, 4, 32> uv = uv_temp.select<4, 1, 32, 2>(0, 0); |
| 555 |
|
| 556 |
if( (final_v_pos+8 + i * 16) <= height) |
| 557 |
{ |
| 558 |
write_plane(EncSurfIndex, GENX_SURFACE_Y_PLANE, final_h_pos, final_v_pos + i * 16, YUV_Data8_8bit.select<8, 1, 32, 2>(0, 0)); |
| 559 |
} |
| 560 |
|
| 561 |
write_plane(EncSurfIndex, GENX_SURFACE_UV_PLANE, final_h_pos, (final_v_pos >> 1) + i * 8, uv.select<4, 1, 32, 1>(0, 0)); |
| 562 |
} |
| 563 |
else |
| 564 |
{ |
| 565 |
if( (final_v_pos + 8 + i * 16) <= height) |
| 566 |
{ |
| 567 |
write(EncSurfIndex, final_h_pos, final_v_pos + i * 16, YUV_Data8_8bit.select<8, 1, 32, 2>(0, 0)); |
| 568 |
} |
| 569 |
write(EncSurfIndex, final_h_pos, height + final_v_pos + i * 16, YUV_Data8_8bit.select<8, 1, 32, 2>(0, 1)); |
| 570 |
} |
| 571 |
|
| 572 |
|
| 573 |
read (PakSurfIndex, 2 * final_h_pos_argb, final_v_pos + 8 + i * 16, YUV_Data8_8bit.select<8, 1, 32, 1>(0, 0)); |
| 574 |
read (PakSurfIndex, 2 * final_h_pos_argb + 32, final_v_pos + 8 + i * 16, YUV_Data8_8bit.select<8, 1, 32, 1>(0, 32)); |
| 575 |
YPix.select<8, 1, 32, 1>(8 + i * 16, 0) = YUV_Data8_8bit.select<8, 1, 32, 2>(0, 0); |
| 576 |
|
| 577 |
if(YUY2convertflag) |
| 578 |
{ |
| 579 |
matrix_ref<uchar, 8, 32> uv_channel = YUV_Data8_8bit.select<8, 1, 32, 2>(0, 1); |
| 580 |
temp.select<4, 1, 32, 1>(0, 0) = uv_channel.select<4, 2, 32, 1>(0, 0) + uv_channel.select<4, 2, 32, 1>(1, 0); |
| 581 |
temp.select<4, 1, 32, 1>(0, 0) += 1; |
| 582 |
temp.select<4, 1, 32, 1>(0, 0) = temp.select<4, 1, 32, 1>(0, 0) >> 1; |
| 583 |
|
| 584 |
matrix_ref<uchar, 4, 64> uv_temp = temp.format<uchar, 4, 64>(); |
| 585 |
matrix_ref<uchar, 4, 32> uv = uv_temp.select<4, 1, 32, 2>(0, 0); |
| 586 |
|
| 587 |
if ((final_v_pos+16 + i * 16) <= height) |
| 588 |
{ |
| 589 |
write_plane(EncSurfIndex, GENX_SURFACE_Y_PLANE, final_h_pos, final_v_pos + 8 + i * 16, YUV_Data8_8bit.select<8, 1, 32, 2>(0, 0) ); |
| 590 |
} |
| 591 |
write_plane(EncSurfIndex, GENX_SURFACE_UV_PLANE, final_h_pos, (final_v_pos >> 1) + 4 + i * 8, uv.select<4, 1, 32, 1>(0, 0)); |
| 592 |
} |
| 593 |
else |
| 594 |
{ |
| 595 |
if ((final_v_pos + 16 + i * 16) <= height) |
| 596 |
{ |
| 597 |
write(EncSurfIndex, final_h_pos, final_v_pos + 8 + i * 16, YUV_Data8_8bit.select<8, 1, 32, 2>(0, 0)); |
| 598 |
} |
| 599 |
write(EncSurfIndex, final_h_pos, height + final_v_pos + 8 + i * 16, YUV_Data8_8bit.select<8, 1, 32, 2>(0, 1)); |
| 600 |
} |
| 601 |
} |
| 602 |
|
| 603 |
if(!stage) |
| 604 |
{ |
| 605 |
cm_fence(); |
| 606 |
return; |
| 607 |
} |
| 608 |
} |
| 609 |
else if (colorformat == NV12_Linear) |
| 610 |
{ |
| 611 |
read(PakSurfIndex, final_h_pos, height + (final_v_pos >> 1), YPix.select<8,1,32,1>(0, 0)); |
| 612 |
read(PakSurfIndex, final_h_pos, height + (final_v_pos >> 1) + 8, YPix.select<8,1,32,1>(8, 0)); |
| 613 |
|
| 614 |
write_plane(EncSurfIndex, GENX_SURFACE_UV_PLANE, final_h_pos, final_v_pos >> 1, YPix.select<8,1,32,1>(0, 0)); |
| 615 |
write_plane(EncSurfIndex, GENX_SURFACE_UV_PLANE, final_h_pos, (final_v_pos >> 1) + 8, YPix.select<8,1,32,1>(8, 0)); |
| 616 |
|
| 617 |
read(PakSurfIndex, final_h_pos, final_v_pos, YPix.select<8, 1, 32, 1>(0, 0)); |
| 618 |
read(PakSurfIndex, final_h_pos, final_v_pos + 8, YPix.select<8, 1, 32, 1>(8, 0)); |
| 619 |
read(PakSurfIndex, final_h_pos, final_v_pos + 16, YPix.select<8, 1, 32, 1>(16,0)); |
| 620 |
read(PakSurfIndex, final_h_pos, final_v_pos + 24, YPix.select<8, 1, 32, 1>(24,0)); |
| 621 |
|
| 622 |
write_plane(EncSurfIndex, GENX_SURFACE_Y_PLANE, final_h_pos, final_v_pos, YPix.select<8, 1, 32, 1>(0, 0)); |
| 623 |
write_plane(EncSurfIndex, GENX_SURFACE_Y_PLANE, final_h_pos, final_v_pos + 8, YPix.select<8, 1, 32, 1>(8, 0)); |
| 624 |
write_plane(EncSurfIndex, GENX_SURFACE_Y_PLANE, final_h_pos, final_v_pos + 16, YPix.select<8, 1, 32, 1>(16,0)); |
| 625 |
write_plane(EncSurfIndex, GENX_SURFACE_Y_PLANE, final_h_pos, final_v_pos + 24, YPix.select<8, 1, 32, 1>(24,0)); |
| 626 |
} |
| 627 |
|
| 628 |
if( (colorformat == YUY2) || (colorformat == Y210) ) |
| 629 |
{ |
| 630 |
if(quot > 0) |
| 631 |
{ |
| 632 |
vector<ushort, 32> maskgb = masktab; |
| 633 |
first_index = first_index >> 1; |
| 634 |
|
| 635 |
vector<ushort, 16> replicants = 0; |
| 636 |
vector<ushort, 16> masktable = 0; |
| 637 |
|
| 638 |
vector<ushort, 16> final_h_pos_mask = 0; |
| 639 |
final_h_pos_mask = final_h_pos + 2 * maskgb.select<16, 1>(); |
| 640 |
|
| 641 |
matrix_ref<ushort, 32, 16>YPix_2byte = YPix.format<ushort, 32, 16>(); |
| 642 |
|
| 643 |
masktable = 1; |
| 644 |
masktable.merge(masktable, 0, final_h_pos_mask < width); |
| 645 |
|
| 646 |
for(uint mcount = 0; mcount < 32; mcount++) |
| 647 |
{ |
| 648 |
replicants = YPix_2byte(mcount, first_index); |
| 649 |
YPix_2byte.row(mcount).merge(YPix_2byte.row(mcount), replicants, masktable); |
| 650 |
|
| 651 |
mcount++; |
| 652 |
replicants = YPix_2byte(mcount, first_index); |
| 653 |
YPix_2byte.row(mcount).merge(YPix_2byte.row(mcount), replicants, masktable); |
| 654 |
|
| 655 |
mcount++; |
| 656 |
replicants = YPix_2byte(mcount, first_index); |
| 657 |
YPix_2byte.row(mcount).merge(YPix_2byte.row(mcount), replicants, masktable); |
| 658 |
|
| 659 |
mcount++; |
| 660 |
replicants = YPix_2byte(mcount,first_index); |
| 661 |
YPix_2byte.row(mcount).merge(YPix_2byte.row(mcount), replicants, masktable); |
| 662 |
} |
| 663 |
} |
| 664 |
} |
| 665 |
} |
| 666 |
else |
| 667 |
{ |
| 668 |
read_plane(PakSurfIndex, GENX_SURFACE_Y_PLANE, final_h_pos, final_v_pos, YPix.select<8, 1, 32, 1>(0, 0)); |
| 669 |
read_plane(PakSurfIndex, GENX_SURFACE_Y_PLANE, final_h_pos, final_v_pos + 8, YPix.select<8, 1, 32, 1>(8, 0)); |
| 670 |
read_plane(PakSurfIndex, GENX_SURFACE_Y_PLANE, final_h_pos, final_v_pos + 16, YPix.select<8, 1, 32, 1>(16,0)); |
| 671 |
read_plane(PakSurfIndex, GENX_SURFACE_Y_PLANE, final_h_pos, final_v_pos + 24, YPix.select<8, 1, 32, 1>(24,0)); |
| 672 |
} |
| 673 |
|
| 674 |
matrix<uchar, 16, 16> YDown2 = 0; |
| 675 |
matrix<ushort, 8, 16> temp1 = 0; |
| 676 |
matrix<ushort, 8, 8> temp2 = 0; |
| 677 |
matrix<ushort, 4, 4> temp3 = 0; |
| 678 |
matrix<uchar, 8, 8> out = 0; |
| 679 |
|
| 680 |
#pragma unroll |
| 681 |
for(uint i = 0; i < 2; i++) { |
| 682 |
#pragma unroll |
| 683 |
for(uint j = 0; j< 2; j++) |
| 684 |
{ |
| 685 |
temp1.select<8, 1, 16, 1>(0, 0) = YPix.select<8, 2, 16, 1>(i << 4, j << 4) + YPix.select<8, 2, 16, 1>((i << 4) + 1, j << 4 ); |
| 686 |
temp2.select<8, 1, 8, 1>(0, 0) = temp1.select<8, 1, 8, 2>(0, 0) + temp1.select<8, 1, 8, 2>(0, 1); |
| 687 |
|
| 688 |
temp2.select<8, 1, 8, 1>(0, 0) += 2; |
| 689 |
temp2.select<8, 1, 8, 1>(0, 0) = temp2.select<8, 1, 8, 1>(0, 0) >> 2; |
| 690 |
|
| 691 |
if((stage == 1) || (stage == 4)) |
| 692 |
{ |
| 693 |
YDown2.select<8, 1, 8, 1>(i << 3, j << 3) = temp2.select<8, 1, 8, 1>(0, 0); |
| 694 |
} |
| 695 |
|
| 696 |
temp2.select<4, 2, 8, 1>(0, 0) = temp2.select<4, 2, 8, 1>(0, 0) + temp2.select<4, 2, 8, 1>(1, 0); |
| 697 |
temp3.select<4, 1, 4, 1>(0, 0) = temp2.select<4, 2, 4, 2>(0, 0) + temp2.select<4, 2, 4, 2>(0, 1); |
| 698 |
|
| 699 |
temp3.select<4, 1, 4, 1>(0, 0) += 2; |
| 700 |
temp3.select<4, 1, 4, 1>(0, 0) = temp3.select<4, 1, 4, 1>(0, 0) >> 2; |
| 701 |
|
| 702 |
out.select<4, 1, 4, 1>(i << 2, j << 2) = temp3.select<4, 1, 4, 1>(0, 0); |
| 703 |
} |
| 704 |
} |
| 705 |
|
| 706 |
if ((!replicate[0]) || (!replicate2[0])) |
| 707 |
{ |
| 708 |
if(stage>1) |
| 709 |
{ |
| 710 |
write(DS4XSurf, h_pos_write, v_pos_write, out); |
| 711 |
} |
| 712 |
|
| 713 |
if(stage ==1 || stage == 4) |
| 714 |
{ |
| 715 |
write(DS2XSurf, h_pos_write2, v_pos_write2, YDown2); |
| 716 |
} |
| 717 |
} |
| 718 |
else |
| 719 |
{ |
| 720 |
matrix<uchar, 1, 8>last_line = out.select<1, 1, 8, 1>(7 - replicate[0], 0); |
| 721 |
vector<int, 16>mask2; |
| 722 |
vector_ref<int, 8>mask = mask2.select<8, 1>(); |
| 723 |
|
| 724 |
#pragma unroll |
| 725 |
for(uint i = 1; i < 8; i++) |
| 726 |
{ |
| 727 |
mask = i > 7 - replicate[0]; |
| 728 |
out.select<1, 1, 8, 1>(i, 0).merge(last_line, mask); |
| 729 |
} |
| 730 |
|
| 731 |
if ((stage == 4) || (stage == 1)) |
| 732 |
{ |
| 733 |
if(replicate2[0] >= 15) |
| 734 |
{ |
| 735 |
matrix<uchar, 1, 16> lastline2 = YDown2.row(1); |
| 736 |
|
| 737 |
for(uint i=0; i<16; i++) |
| 738 |
{ |
| 739 |
YDown2.row(i) = lastline2; |
| 740 |
} |
| 741 |
} |
| 742 |
else |
| 743 |
{ |
| 744 |
matrix<uchar, 1, 16> last_line2 = YDown2.select<1, 1, 16, 1>(15 - replicate2[0], 0); |
| 745 |
|
| 746 |
#pragma unroll |
| 747 |
for(uint i = 1; i < 16; i++) |
| 748 |
{ |
| 749 |
mask2 = i > 15 - replicate2[0]; |
| 750 |
YDown2.select<1, 1, 16, 1>(i, 0).merge(last_line2, mask2 ); |
| 751 |
} |
| 752 |
} |
| 753 |
write(DS2XSurf, h_pos_write2, v_pos_write2, YDown2); |
| 754 |
} |
| 755 |
|
| 756 |
if(stage>1) |
| 757 |
write(DS4XSurf, h_pos_write, v_pos_write, out); |
| 758 |
} |
| 759 |
|
| 760 |
if(MBStatflag) |
| 761 |
{ |
| 762 |
width = uWidth(0); |
| 763 |
height = uHeight(0); |
| 764 |
|
| 765 |
matrix<ushort, 32, 32> squared; |
| 766 |
matrix<uint, 2, 2> sum; |
| 767 |
matrix<uint, 2, 2> pixAvg; |
| 768 |
matrix<uint, 2, 2> SumOfSquared; |
| 769 |
matrix_ref<uint,2, 2> VarianceLuma = SumOfSquared; |
| 770 |
matrix_ref<uint,2, 2> flatness_check = sum; |
| 771 |
|
| 772 |
squared = YPix * YPix; |
| 773 |
|
| 774 |
sum[0][0] = cm_sum<uint>(YPix.select<16, 1, 16, 1>(0, 0), SAT); |
| 775 |
SumOfSquared[0][0] = cm_sum<uint>(squared.select<16, 1, 16, 1>(0, 0), SAT); |
| 776 |
|
| 777 |
sum[0][1] = cm_sum<uint>(YPix.select<16, 1, 16, 1>(0, 16), SAT); |
| 778 |
SumOfSquared[0][1] = cm_sum<uint>(squared.select<16, 1, 16, 1>(0, 16), SAT); |
| 779 |
|
| 780 |
sum[1][0] = cm_sum<uint>(YPix.select<16, 1, 16, 1>(16, 0), SAT); |
| 781 |
SumOfSquared[1][0] = cm_sum<uint>(squared.select<16, 1, 16, 1>(16, 0), SAT); |
| 782 |
|
| 783 |
sum[1][1] = cm_sum<uint>(YPix.select<16, 1, 16, 1>(16, 16), SAT); |
| 784 |
SumOfSquared[1][1] = cm_sum<uint>(squared.select<16, 1, 16, 1>(16, 16), SAT); |
| 785 |
|
| 786 |
pixAvg = sum >> 8; |
| 787 |
sum *= sum; |
| 788 |
sum >>= 8; |
| 789 |
|
| 790 |
VarianceLuma = SumOfSquared - sum; |
| 791 |
VarianceLuma >>= 8; |
| 792 |
|
| 793 |
flatness_check = (VarianceLuma < Flatness_Threshold); |
| 794 |
|
| 795 |
unsigned short NumMBperRow = width / 16; |
| 796 |
NumMBperRow += ((width % 16)> 0); |
| 797 |
|
| 798 |
unsigned int offset_vp = (2 * 16 * 4 * (v_pos * NumMBperRow + h_pos)); |
| 799 |
unsigned int offset2_vp = offset_vp + NumMBperRow * 64; |
| 800 |
vector<uint, 64> writeVProc(0); |
| 801 |
vector<uint, 16> Element_Offset(Stat_offsets); |
| 802 |
|
| 803 |
offset_vp >>= 2; |
| 804 |
offset2_vp >>= 2; |
| 805 |
|
| 806 |
writeVProc.select<2,16>(5) = flatness_check.row(0); |
| 807 |
writeVProc.select<2,16>(37) = flatness_check.row(1); |
| 808 |
|
| 809 |
writeVProc.select<2,16>(6) = VarianceLuma.row(0); |
| 810 |
writeVProc.select<2,16>(38) = VarianceLuma.row(1); |
| 811 |
|
| 812 |
writeVProc.select<2,16>(11) = pixAvg.row(0); |
| 813 |
writeVProc.select<2,16>(43) = pixAvg.row(1); |
| 814 |
|
| 815 |
if((h_pos * 32 < width) && (v_pos * 32 < height)) |
| 816 |
{ |
| 817 |
if((h_pos * 32)+ 16 >= width) |
| 818 |
{ |
| 819 |
write(MBStats, offset_vp, Element_Offset, writeVProc.select<16,1>(0)); |
| 820 |
} |
| 821 |
else |
| 822 |
{ |
| 823 |
write(MBStats, offset_vp, Element_Offset, writeVProc.select<16,1>(0)); |
| 824 |
write(MBStats, offset_vp + 16, Element_Offset, writeVProc.select<16,1>(16)); |
| 825 |
} |
| 826 |
|
| 827 |
if((v_pos * 32)+16 < height) |
| 828 |
{ |
| 829 |
if((h_pos * 32)+ 16 >= width) |
| 830 |
{ |
| 831 |
write(MBStats, offset2_vp, Element_Offset, writeVProc.select<16,1>(32)); |
| 832 |
} |
| 833 |
else |
| 834 |
{ |
| 835 |
write(MBStats, offset2_vp, Element_Offset, writeVProc.select<16,1>(32)); |
| 836 |
write(MBStats, offset2_vp + 16, Element_Offset, writeVProc.select<16,1>(48)); |
| 837 |
} |
| 838 |
} |
| 839 |
} |
| 840 |
} |
| 841 |
|
| 842 |
uchar hevc_enc_history = (CURBEData_U8(5) >> 6) & 0x1; |
| 843 |
|
| 844 |
if(hevc_enc_history) |
| 845 |
{ |
| 846 |
uchar LCUtype = (CURBEData_U8(5) >> 7) & 0x1; |
| 847 |
|
| 848 |
vector<ushort, 32> clear_buffer; |
| 849 |
vector_ref<ushort, 16> clear_buffer1 = clear_buffer.select<16, 1>(0); |
| 850 |
vector<ushort, 16> History_perLCU; |
| 851 |
|
| 852 |
uint offset = 0; |
| 853 |
uchar shift_factor = LCUtype == 0 ? 6 : 5; |
| 854 |
|
| 855 |
vector_ref<ushort, 1> WidthLCU = History_perLCU.select<1, 1>(0); |
| 856 |
vector_ref<ushort, 1> HeightLCU = History_perLCU.select<1, 1>(1); |
| 857 |
|
| 858 |
WidthLCU(0) = (uWidth(0) + ((1 << shift_factor) - 1)) >> shift_factor; |
| 859 |
HeightLCU(0) = (uHeight(0) + ((1 << shift_factor) - 1)) >> shift_factor; |
| 860 |
|
| 861 |
if(h_pos == 0 && v_pos == 0) |
| 862 |
{ |
| 863 |
clear_buffer = 0; |
| 864 |
numLCUs = WidthLCU * HeightLCU; |
| 865 |
|
| 866 |
for(uint i = 0; i < numLCUs(0); i++) |
| 867 |
{ |
| 868 |
read(Hevc_History_Surface_index, offset, History_perLCU); |
| 869 |
clear_buffer.select<16, 1>(0) += History_perLCU; |
| 870 |
offset += HEVC_HISTORY_BUFFER_ENTRY_SIZE; |
| 871 |
} |
| 872 |
|
| 873 |
clear_buffer.select<4,1>(4) = History_perLCU.select<4,1>(4); |
| 874 |
clear_buffer.select<8,1>(8) = History_perLCU.select<8,1>(8); |
| 875 |
|
| 876 |
write(Hevc_History_Sum_Surface_index, 0, clear_buffer); |
| 877 |
} |
| 878 |
|
| 879 |
if((h_pos == 1) && (v_pos == 0)) |
| 880 |
{ |
| 881 |
offset = 0; |
| 882 |
clear_buffer1 = 0; |
| 883 |
numLCUs = WidthLCU * HeightLCU; |
| 884 |
|
| 885 |
for(uint i = 0; i < numLCUs(0); i++) |
| 886 |
{ |
| 887 |
write(Scratch_Surface_index, offset, clear_buffer1); |
| 888 |
write(Scratch_Surface_index, offset + 32, clear_buffer1); |
| 889 |
write(Scratch_Surface_index, offset + 64, clear_buffer1); |
| 890 |
|
| 891 |
offset += 96; |
| 892 |
} |
| 893 |
} |
| 894 |
} |
| 895 |
|
| 896 |
cm_fence(); |
| 897 |
} |