1/* 2 * Copyright © 2009 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 (including the next 12 * paragraph) shall be included in all copies or substantial portions of the 13 * Software. 14 * 15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 20 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 21 * SOFTWARE. 22 * 23 * Author: 24 * Zou Nan hai <nanhai.zou@intel.com> 25 * Yan Li <li.l.yan@intel.com> 26 * Liu Xi bin<xibin.liu@intel.com> 27 */ 28/* GRF allocation: 29 g1~g30: constant buffer 30 g1~g2:intra IQ matrix 31 g3~g4:non intra IQ matrix 32 g5~g20:IDCT table 33 g31: thread payload 34 g58~g81:reference data 35 g82: thread payload backup 36 g83~g106:IDCT data 37 g115: message descriptor for reading reference data */ 38mov (1) g115.8<1>UD 0x01FUD {align1}; 39send (16) 0 g40.0<1>UW g115<8,8,1>UW read(surface,2,0,2) mlen 1 rlen 1 {align1}; 40add (1) g115.4<1>UD g115.4<1,1,1>UD 2UD {align1}; 41send (16) 0 g42.0<1>UW g115<8,8,1>UW read(surface,2,0,2) mlen 1 rlen 1 {align1}; 42add (1) g115.4<1>UD g115.4<1,1,1>UD 2UD {align1}; 43send (16) 0 g44.0<1>UW g115<8,8,1>UW read(surface,2,0,2) mlen 1 rlen 1 {align1}; 44add (1) g115.4<1>UD g115.4<1,1,1>UD 2UD {align1}; 45send (16) 0 g46.0<1>UW g115<8,8,1>UW read(surface,2,0,2) mlen 1 rlen 1 {align1}; 46add (1) g115.4<1>UD g115.4<1,1,1>UD 2UD {align1}; 47mov (1) g115.8<1>UD 0x07001FUD {align1}; 48send (16) 0 g48.0<1>UW g115<8,8,1>UW read(surface,2,0,2) mlen 1 rlen 8 {align1}; 49add (1) g115.4<1>UD g115.4<1,1,1>UD 8UD {align1}; 50mov (1) g115.8<1>UD 0x1FUD {align1}; 51send (16) 0 g56.0<1>UW g115<8,8,1>UW read(surface,2,0,2) mlen 1 rlen 1 {align1}; 52 53add (16) g32.0<1>UW g40.0<16,16,1>UB g42.0<16,16,1>UB {align1}; 54add (16) g33.0<1>UW g42.0<16,16,1>UB g44.0<16,16,1>UB {align1}; 55add (16) g34.0<1>UW g44.0<16,16,1>UB g46.0<16,16,1>UB {align1}; 56add (16) g35.0<1>UW g46.0<16,16,1>UB g48.0<16,16,1>UB {align1}; 57add (16) g36.0<1>UW g48.0<16,16,1>UB g50.0<16,16,1>UB {align1}; 58add (16) g37.0<1>UW g50.0<16,16,1>UB g52.0<16,16,1>UB {align1}; 59add (16) g38.0<1>UW g52.0<16,16,1>UB g54.0<16,16,1>UB {align1}; 60add (16) g39.0<1>UW g54.0<16,16,1>UB g56.0<16,16,1>UB {align1}; 61 62add (16) g32.0<1>UW g32.0<16,16,1>UW g40.1<16,16,1>UB {align1}; 63add (16) g33.0<1>UW g33.0<16,16,1>UW g42.1<16,16,1>UB {align1}; 64add (16) g34.0<1>UW g34.0<16,16,1>UW g44.1<16,16,1>UB {align1}; 65add (16) g35.0<1>UW g35.0<16,16,1>UW g46.1<16,16,1>UB {align1}; 66add (16) g36.0<1>UW g36.0<16,16,1>UW g48.1<16,16,1>UB {align1}; 67add (16) g37.0<1>UW g37.0<16,16,1>UW g50.1<16,16,1>UB {align1}; 68add (16) g38.0<1>UW g38.0<16,16,1>UW g52.1<16,16,1>UB {align1}; 69add (16) g39.0<1>UW g39.0<16,16,1>UW g54.1<16,16,1>UB {align1}; 70 71add (16) g32.0<1>UW g32.0<16,16,1>UW g42.1<16,16,1>UB {align1}; 72add (16) g33.0<1>UW g33.0<16,16,1>UW g44.1<16,16,1>UB {align1}; 73add (16) g34.0<1>UW g34.0<16,16,1>UW g46.1<16,16,1>UB {align1}; 74add (16) g35.0<1>UW g35.0<16,16,1>UW g48.1<16,16,1>UB {align1}; 75add (16) g36.0<1>UW g36.0<16,16,1>UW g50.1<16,16,1>UB {align1}; 76add (16) g37.0<1>UW g37.0<16,16,1>UW g52.1<16,16,1>UB {align1}; 77add (16) g38.0<1>UW g38.0<16,16,1>UW g54.1<16,16,1>UB {align1}; 78add (16) g39.0<1>UW g39.0<16,16,1>UW g56.1<16,16,1>UB {align1}; 79 80shr (16) g32.0<1>UW g32.0<16,16,1>UW 2UW {align1}; 81shr (16) g33.0<1>UW g33.0<16,16,1>UW 2UW {align1}; 82shr (16) g34.0<1>UW g34.0<16,16,1>UW 2UW {align1}; 83shr (16) g35.0<1>UW g35.0<16,16,1>UW 2UW {align1}; 84shr (16) g36.0<1>UW g36.0<16,16,1>UW 2UW {align1}; 85shr (16) g37.0<1>UW g37.0<16,16,1>UW 2UW {align1}; 86shr (16) g38.0<1>UW g38.0<16,16,1>UW 2UW {align1}; 87shr (16) g39.0<1>UW g39.0<16,16,1>UW 2UW {align1}; 88