2 * Copyright © <2010>, Intel Corporation.
4 * This program is licensed under the terms and conditions of the
5 * Eclipse Public License (EPL), version 1.0. The full text of the EPL is at
6 * http://www.opensource.org/licenses/eclipse-1.0.php.
9 // Modual name: IntraFrame.asm
11 // Make intra predition estimation for Intra frame
15 // Now, begin source code....
22 mov (16) tmp_reg0.0<1>:UD 0x0:UD {align1};
23 mov (16) tmp_reg2.0<1>:UD 0x0:UD {align1};
24 mov (16) tmp_reg3.0<1>:UD 0x0:UD {align1};
26 shl (2) read0_header.0<1>:D orig_xy_ub<2,2,1>:UB 4:UW {align1}; /* (x, y) * 16 */
27 add (1) read0_header.0<1>:D read0_header.0<0,1,0>:D -8:W {align1}; /* X offset */
28 add (1) read0_header.4<1>:D read0_header.4<0,1,0>:D -1:W {align1}; /* Y offset */
29 mov (1) read0_header.8<1>:UD BLOCK_32X1 {align1};
30 mov (1) read0_header.20<1>:UB thread_id_ub {align1}; /* dispatch id */
32 shl (2) read1_header.0<1>:D orig_xy_ub<2,2,1>:UB 4:UW {align1}; /* (x, y) * 16 */
33 add (1) read1_header.0<1>:D read1_header.0<0,1,0>:D -4:W {align1}; /* X offset */
34 mov (1) read1_header.8<1>:UD BLOCK_4X16 {align1};
35 mov (1) read1_header.20<1>:UB thread_id_ub {align1}; /* dispatch id */
37 shl (2) vme_m0.8<1>:UW orig_xy_ub<2,2,1>:UB 4:UW {align1}; /* Source = (x, y) * 16 */
40 shl (2) vme_m0.0<1>:UW orig_xy_ub<2,2,1>:UB 4:UW {align1};
41 add (1) vme_m0.0<1>:W vme_m0.0<0,1,0>:W -16:W {align1}; /* Reference = (x-16,y-12)-(x+32,y+24) */
42 add (1) vme_m0.2<1>:W vme_m0.2<0,1,0>:W -12:W {align1};
44 mov (1) vme_m0.0<1>:W -16:W {align1} ; /* Reference = (x-16,y-12)-(x+32,y+24) */
45 mov (1) vme_m0.2<1>:W -12:W {align1} ;
48 mov (1) vme_m0.12<1>:UD SEARCH_CTRL_SINGLE + INTER_PART_MASK + INTER_SAD_HAAR + SUB_PEL_MODE_QUARTER:UD {align1}; /* 16x16 Source, 1/4 pixel, harr */
49 mov (1) vme_m0.20<1>:UB thread_id_ub {align1}; /* dispatch id */
50 mov (1) vme_m0.22<1>:UW REF_REGION_SIZE {align1}; /* Reference Width&Height, 32x32 */
52 mov (1) vme_m1.0<1>:UD ADAPTIVE_SEARCH_ENABLE:ud {align1} ;
53 mov (1) vme_m1.4<1>:UD FB_PRUNING_ENABLE:UD {align1};
54 /* MV num is passed by constant buffer. R4.28 */
55 mov (1) vme_m1.4<1>:UB r4.28<0,1,0>:UB {align1};
56 mov (1) vme_m1.8<1>:UD START_CENTER + SEARCH_PATH_LEN:UD {align1};
58 mul (1) obw_m0.8<1>:UD w_in_mb_uw<0,1,0>:UW orig_y_ub<0,1,0>:UB {align1};
59 add (1) obw_m0.8<1>:UD obw_m0.8<0,1,0>:UD orig_x_ub<0,1,0>:UB {align1};
60 mul (1) obw_m0.8<1>:UD obw_m0.8<0,1,0>:UD INTER_VME_OUTPUT_IN_OWS:UD {align1};
61 mov (1) obw_m0.20<1>:UB thread_id_ub {align1}; /* dispatch id */
66 * Media Read Message -- fetch neighbor edge pixels
69 mov (8) msg_reg0.0<1>:UD read0_header.0<8,8,1>:UD {align1};
70 send (8) msg_ind INEP_ROW<1>:UB null read(BIND_IDX_INEP, 0, 0, 4) mlen 1 rlen 1 {align1};
73 mov (8) msg_reg0.0<1>:UD read1_header.0<8,8,1>:UD {align1};
74 send (8) msg_ind INEP_COL0<1>:UB null read(BIND_IDX_INEP, 0, 0, 4) mlen 1 rlen 2 {align1};
80 mov (8) vme_msg_0.0<1>:UD vme_m0.0<8,8,1>:UD {align1};
83 mov (1) intra_flag<1>:UW 0x0:UW {align1} ;
84 and.z.f0.0 (1) null<1>:UW transform_8x8_ub<0,1,0>:UB 1:UW {align1};
85 (f0.0) mov (1) intra_part_mask_ub<1>:UB LUMA_INTRA_8x8_DISABLE {align1};
87 cmp.nz.f0.0 (1) null<1>:UW orig_x_ub<0,1,0>:UB 0:UW {align1}; /* X != 0 */
88 (f0.0) add (1) mb_intra_struct_ub<1>:UB mb_intra_struct_ub<0,1,0>:UB INTRA_PRED_AVAIL_FLAG_AE {align1}; /* A */
90 cmp.nz.f0.0 (1) null<1>:UW orig_y_ub<0,1,0>:UB 0:UW {align1}; /* Y != 0 */
91 (f0.0) add (1) mb_intra_struct_ub<1>:UB mb_intra_struct_ub<0,1,0>:UB INTRA_PRED_AVAIL_FLAG_B {align1}; /* B */
93 mul.nz.f0.0 (1) null<1>:UW orig_x_ub<0,1,0>:UB orig_y_ub<0,1,0>:UB {align1}; /* X * Y != 0 */
94 (f0.0) add (1) mb_intra_struct_ub<1>:UB mb_intra_struct_ub<0,1,0>:UB INTRA_PRED_AVAIL_FLAG_D {align1}; /* D */
96 add (1) tmp_x_w<1>:W orig_x_ub<0,1,0>:UB 1:UW {align1}; /* X + 1 */
97 add (1) tmp_x_w<1>:W w_in_mb_uw<0,1,0>:UW -tmp_x_w<0,1,0>:W {align1}; /* width - (X + 1) */
98 mul.nz.f0.0 (1) null<1>:UD tmp_x_w<0,1,0>:W orig_y_ub<0,1,0>:UB {align1}; /* (width - (X + 1)) * Y != 0 */
99 (f0.0) add (1) mb_intra_struct_ub<1>:UB mb_intra_struct_ub<0,1,0>:UB INTRA_PRED_AVAIL_FLAG_C {align1}; /* C */
101 and.nz.f0.0 (1) null<1>:UW slice_edge_ub<0,1,0>:UB 2:UW {align1};
102 (f0.0) and (1) mb_intra_struct_ub<1>:UB mb_intra_struct_ub<0,1,0>:UB 0xE0 {align1}; /* slice edge disable B,C,D*/
104 mov (8) vme_msg_1<1>:UD vme_m1.0<8,8,1>:UD {align1};
107 mov (8) vme_msg_2<1>:UD 0x0:UD {align1};
110 mov (1) INEP_ROW.0<1>:UD 0x0:UD {align1};
111 and (1) INEP_ROW.4<1>:UD INEP_ROW.4<0,1,0>:UD 0xFF000000:UD {align1};
112 mov (8) vme_msg_3<1>:UD INEP_ROW.0<8,8,1>:UD {align1};
115 mov (8) vme_msg_4<1>:UD 0x0 {align1};
116 mov (16) vme_msg_4.0<1>:UB INEP_COL0.3<32,8,4>:UB {align1};
117 mov (1) vme_msg_4.16<1>:UD INTRA_PREDICTORE_MODE {align1};
127 VME_MESSAGE_TYPE_MIXED
130 rlen vme_inter_wb_length
133 * Oword Block Write message
137 mov (8) msg_reg0.0<1>:UD obw_m0.0<8,8,1>:UD {align1};
140 mov (16) obw_m1.0<1>:UW vme_wb1.0<16,16,1>:UB {align1};
141 add (8) obw_m1.0<2>:W obw_m1.0<16,8,2>:W -64:W {align1};
142 add (8) obw_m1.2<2>:W obw_m1.2<16,8,2>:W -48:W {align1};
143 mov (16) obw_m2.0<1>:UW vme_wb1.16<16,16,1>:UB {align1};
144 add (8) obw_m2.0<2>:W obw_m2.0<16,8,2>:W -64:W {align1};
145 add (8) obw_m2.2<2>:W obw_m2.2<16,8,2>:W -48:W {align1};
146 mov (16) obw_m3.0<1>:UW vme_wb2.0<16,16,1>:UB {align1};
147 add (8) obw_m3.0<2>:W obw_m3.0<16,8,2>:W -64:W {align1};
148 add (8) obw_m3.2<2>:W obw_m3.2<16,8,2>:W -48:W {align1};
149 mov (16) obw_m4.0<1>:UW vme_wb2.16<16,16,1>:UB {align1};
150 add (8) obw_m4.0<2>:W obw_m4.0<16,8,2>:W -64:W {align1};
151 add (8) obw_m4.2<2>:W obw_m4.2<16,8,2>:W -48:W {align1};
153 mov (8) obw_m1.0<1>:ud vme_wb1.0<8,8,1>:ud {align1};
154 mov (8) obw_m2.0<1>:ud vme_wb2.0<8,8,1>:ud {align1};
155 mov (8) obw_m3.0<1>:ud vme_wb3.0<8,8,1>:ud {align1};
156 mov (8) obw_m4.0<1>:ud vme_wb4.0<8,8,1>:ud {align1};
159 mov (8) msg_reg1.0<1>:UD obw_m1.0<8,8,1>:UD {align1};
161 mov (8) msg_reg2.0<1>:UD obw_m2.0<8,8,1>:UD {align1};
163 mov (8) msg_reg3.0<1>:UD obw_m3.0<8,8,1>:UD {align1};
165 mov (8) msg_reg4.0<1>:UD obw_m4.0<8,8,1>:UD {align1};
167 /* bind index 3, write 8 oword, msg type: 8(OWord Block Write) */
177 OBW_WRITE_COMMIT_CATEGORY,
185 add (1) msg_reg0.8<1>:UD obw_m0.8<0,1,0>:UD INTER_VME_OUTPUT_MV_IN_OWS:UD {align1} ;
187 and.z.f0.0 (1) null<1>:ud vme_wb0.0<0,1,0>:ud INTRAMBFLAG_MASK:ud {align1} ;
189 (-f0.0)jmpi (1) __INTRA_INFO ;
192 mov (1) tmp_uw1<1>:uw 0:uw {align1} ;
193 mov (1) tmp_ud1<1>:ud 0:ud {align1} ;
194 (f0.0)and (1) tmp_uw1<1>:uw vme_wb0.2<0,1,0>:uw MV32_BIT_MASK:uw {align1} ;
195 (f0.0)shr (1) tmp_uw1<1>:uw tmp_uw1<1>:uw MV32_BIT_SHIFT:uw {align1} ;
196 (f0.0)mul (1) tmp_ud1<1>:ud tmp_uw1<0,1,0>:uw 96:uw {align1} ;
197 (f0.0)add (1) tmp_ud1<1>:ud tmp_ud1<0,1,0>:ud 32:uw {align1} ;
198 (f0.0)shl (1) tmp_uw1<1>:uw tmp_uw1<0,1,0>:uw MFC_MV32_BIT_SHIFT:uw {align1} ;
199 (f0.0)add (1) tmp_uw1<1>:uw tmp_uw1<0,1,0>:uw MVSIZE_UW_BASE:uw {align1} ;
200 add (1) tmp_uw1<1>:uw tmp_uw1<0,1,0>:uw CBP_DC_YUV_UW:uw {align1} ;
202 mov (1) msg_reg1.0<1>:uw vme_wb0.0<0,1,0>:uw {align1} ;
203 mov (1) msg_reg1.2<1>:uw tmp_uw1<0,1,0>:uw {align1} ;
204 mov (1) msg_reg1.4<1>:UD vme_wb0.28<0,1,0>:UD {align1};
205 mov (1) msg_reg1.8<1>:ud tmp_ud1<0,1,0>:ud {align1} ;
207 jmpi (1) __OUTPUT_INFO ;
210 mov (1) msg_reg1.0<1>:UD vme_wb.0<0,1,0>:UD {align1};
211 mov (1) msg_reg1.4<1>:UD vme_wb.16<0,1,0>:UD {align1};
212 mov (1) msg_reg1.8<1>:UD vme_wb.20<0,1,0>:UD {align1};
213 mov (1) msg_reg1.12<1>:UD vme_wb.24<0,1,0>:UD {align1};
216 /* bind index 3, write 1 oword, msg type: 8(OWord Block Write) */
226 OBW_WRITE_COMMIT_CATEGORY,
233 add (1) orig_x_ub<1>:ub orig_x_ub<0,1,0>:ub 1:uw {align1} ;
234 add (1) vme_m0.8<1>:UW vme_m0.8<0,1,0>:UW 16:UW {align1}; /* X += 16 */
236 add (1) vme_m0.0<1>:W vme_m0.0<0,1,0>:W 16:W {align1}; /* X += 16 */
239 cmp.e.f0.0 (1) null<1>:uw w_in_mb_uw<0,1,0>:uw orig_x_ub<0,1,0>:ub {align1};
241 (f0.0)mov (1) orig_x_ub<1>:ub 0:uw {align1} ;
242 (f0.0)mov (1) vme_m0.8<1>:uw 0:uw {align1} ;
243 (f0.0)add (1) vme_m0.10<1>:uw vme_m0.10<0,1,0>:uw 16:uw {align1} ;
245 (f0.0)mov (1) vme_m0.0<1>:w -16:W {align1}; /* Reference = (x-16,y-12)-(x+32,y+24) */
246 (f0.0)add (1) vme_m0.2<1>:w vme_m0.2<0,1,0>:w 16:w {align1};
249 add (1) obw_m0.8<1>:UD obw_m0.8<0,1,0>:UD INTER_VME_OUTPUT_IN_OWS:UW {align1} ;
251 add.z.f0.1 (1) num_macroblocks<1>:w num_macroblocks<0,1,0>:w -1:w {align1} ;
252 (-f0.1)jmpi (1) __VME_LOOP ;
259 mov (8) msg_reg0<1>:UD r0<8,8,1>:UD {align1};
260 send (16) msg_ind acc0<1>UW null thread_spawner(0, 0, 1) mlen 1 rlen 0 {align1 EOT};