| Matthieu Delahaye | 60498fe | 2014-02-18 13:21:06 -0600 | [diff] [blame] | 1 | #include "rsCpuIntrinsicInterPred.h" |
| 2 | |
| 3 | void RsdCpuScriptIntrinsicInterPred::setGlobalObj(uint32_t slot, |
| 4 | ObjectBase *data) { |
| 5 | Allocation *alloc = static_cast<Allocation *>(data); |
| 6 | if (slot == 0) mRef = (uint8_t *)alloc->mHal.state.userProvidedPtr; |
| 7 | if (slot == 1) mParam = (uint8_t *)alloc->mHal.state.userProvidedPtr; |
| 8 | } |
| 9 | |
| 10 | void RsdCpuScriptIntrinsicInterPred::setGlobalVar(uint32_t slot, |
| 11 | const void *data, |
| 12 | size_t dataLength) { |
| 13 | mFriParamCount = ((int32_t *)data)[0]; |
| 14 | mSecParamCount = ((int32_t *)data)[1]; |
| 15 | mParamOffset = ((int32_t *)data)[2]; |
| 16 | } |
| 17 | |
| 18 | void RsdCpuScriptIntrinsicInterPred::kernel(const RsForEachStubParamStruct *p, |
| 19 | uint32_t xstart, uint32_t xend, |
| 20 | uint32_t instep, uint32_t outstep) { |
| 21 | RsdCpuScriptIntrinsicInterPred *cp = (RsdCpuScriptIntrinsicInterPred *)p->usr; |
| 22 | cp->mCount++; |
| 23 | const int vp9_convolve_mode[2][2] = {{24, 16}, {8, 0}}; |
| 24 | uint8_t *ref_base = cp->mRef; |
| 25 | INTER_PRED_PARAM *fri_param = (INTER_PRED_PARAM *)cp->mParam; |
| 26 | INTER_PRED_PARAM *sec_param = (INTER_PRED_PARAM *)(cp->mParam + cp->mParamOffset); |
| 27 | int32_t fri_count = cp->mFriParamCount; |
| 28 | int32_t sec_count = cp->mSecParamCount; |
| 29 | int mode_num; |
| 30 | uint8_t *src; |
| 31 | uint8_t *dst; |
| 32 | const int16_t *filter_x; |
| 33 | const int16_t *filter_y; |
| 34 | for (int i = 0; i < fri_count; i++) { |
| 35 | |
| 36 | mode_num = vp9_convolve_mode[(fri_param[i].x_step_q4 == 16)] |
| 37 | [(fri_param[i].y_step_q4 == 16)]; |
| 38 | src = ref_base + fri_param[i].src_mv; |
| 39 | dst = ref_base + fri_param[i].dst_mv; |
| 40 | |
| 41 | filter_x = inter_pred_filters + fri_param[i].filter_x_mv; |
| 42 | filter_y = inter_pred_filters + fri_param[i].filter_y_mv; |
| 43 | |
| 44 | cp->mSwitchConvolve[fri_param[i].pred_mode + mode_num]( |
| 45 | src, fri_param[i].src_stride, |
| 46 | dst, fri_param[i].dst_stride, |
| 47 | filter_x, fri_param[i].x_step_q4, |
| 48 | filter_y, fri_param[i].y_step_q4, |
| 49 | fri_param[i].w, fri_param[i].h |
| 50 | ); |
| 51 | } |
| 52 | |
| 53 | for (int i = 0; i < sec_count; i++) { |
| 54 | mode_num = vp9_convolve_mode[(sec_param[i].x_step_q4 == 16)] |
| 55 | [(sec_param[i].y_step_q4 == 16)]; |
| 56 | src = ref_base + sec_param[i].src_mv; |
| 57 | dst = ref_base + sec_param[i].dst_mv; |
| 58 | |
| 59 | filter_x = inter_pred_filters + sec_param[i].filter_x_mv; |
| 60 | filter_y = inter_pred_filters + sec_param[i].filter_y_mv; |
| 61 | |
| 62 | cp->mSwitchConvolve[sec_param[i].pred_mode + mode_num + 1]( |
| 63 | src, sec_param[i].src_stride, |
| 64 | dst, sec_param[i].dst_stride, |
| 65 | filter_x, sec_param[i].x_step_q4, |
| 66 | filter_y, sec_param[i].y_step_q4, |
| 67 | sec_param[i].w, sec_param[i].h |
| 68 | ); |
| 69 | } |
| 70 | |
| 71 | } |
| 72 | |
| 73 | RsdCpuScriptIntrinsicInterPred::RsdCpuScriptIntrinsicInterPred(RsdCpuReferenceImpl *ctx, |
| 74 | const Script *s, const Element *e) |
| 75 | : RsdCpuScriptIntrinsic(ctx, s, e, RS_SCRIPT_INTRINSIC_ID_INTER_PRED) { |
| 76 | mRootPtr = &kernel; |
| 77 | mCount = 0; |
| 78 | mParamOffset = 0; |
| 79 | mFriParamCount = 0; |
| 80 | mSecParamCount = 0; |
| 81 | mRef = NULL; |
| 82 | mParam = NULL; |
| 83 | |
| 84 | #if defined(ARCH_ARM_HAVE_VFP) |
| 85 | mSwitchConvolve[0] = vp9_convolve_copy_neon; |
| 86 | mSwitchConvolve[1] = vp9_convolve_avg_neon; |
| 87 | mSwitchConvolve[2] = vp9_convolve8_vert_neon; |
| 88 | mSwitchConvolve[3] = vp9_convolve8_avg_vert_neon; |
| 89 | mSwitchConvolve[4] = vp9_convolve8_horiz_neon; |
| 90 | mSwitchConvolve[5] = vp9_convolve8_avg_horiz_neon; |
| 91 | mSwitchConvolve[6] = vp9_convolve8_neon; |
| 92 | mSwitchConvolve[7] = vp9_convolve8_avg_neon; |
| 93 | |
| 94 | mSwitchConvolve[8] = vp9_convolve8_vert_neon; |
| 95 | mSwitchConvolve[9] = vp9_convolve8_avg_vert_neon; |
| 96 | mSwitchConvolve[10] = vp9_convolve8_vert_neon; |
| 97 | mSwitchConvolve[11] = vp9_convolve8_avg_vert_neon; |
| 98 | mSwitchConvolve[12] = vp9_convolve8_neon; |
| 99 | mSwitchConvolve[13] = vp9_convolve8_avg_neon; |
| 100 | mSwitchConvolve[14] = vp9_convolve8_neon; |
| 101 | mSwitchConvolve[15] = vp9_convolve8_avg_neon; |
| 102 | |
| 103 | mSwitchConvolve[16] = vp9_convolve8_horiz_neon; |
| 104 | mSwitchConvolve[17] = vp9_convolve8_avg_horiz_neon; |
| 105 | mSwitchConvolve[18] = vp9_convolve8_neon; |
| 106 | mSwitchConvolve[19] = vp9_convolve8_avg_neon; |
| 107 | mSwitchConvolve[20] = vp9_convolve8_horiz_neon; |
| 108 | mSwitchConvolve[21] = vp9_convolve8_avg_horiz_neon; |
| 109 | mSwitchConvolve[22] = vp9_convolve8_neon; |
| 110 | mSwitchConvolve[23] = vp9_convolve8_avg_neon; |
| 111 | |
| 112 | mSwitchConvolve[24] = vp9_convolve8_neon; |
| 113 | mSwitchConvolve[25] = vp9_convolve8_avg_neon; |
| 114 | mSwitchConvolve[26] = vp9_convolve8_neon; |
| 115 | mSwitchConvolve[27] = vp9_convolve8_avg_neon; |
| 116 | mSwitchConvolve[28] = vp9_convolve8_neon; |
| 117 | mSwitchConvolve[29] = vp9_convolve8_avg_neon; |
| 118 | mSwitchConvolve[30] = vp9_convolve8_neon; |
| 119 | mSwitchConvolve[31] = vp9_convolve8_avg_neon; |
| 120 | #else |
| 121 | mSwitchConvolve[0] = vp9_convolve_copy_c; |
| 122 | mSwitchConvolve[1] = vp9_convolve_avg_c; |
| 123 | mSwitchConvolve[2] = vp9_convolve8_vert_c; |
| 124 | mSwitchConvolve[3] = vp9_convolve8_avg_vert_c; |
| 125 | mSwitchConvolve[4] = vp9_convolve8_horiz_c; |
| 126 | mSwitchConvolve[5] = vp9_convolve8_avg_horiz_c; |
| 127 | mSwitchConvolve[6] = vp9_convolve8_c; |
| 128 | mSwitchConvolve[7] = vp9_convolve8_avg_c; |
| 129 | |
| 130 | mSwitchConvolve[8] = vp9_convolve8_vert_c; |
| 131 | mSwitchConvolve[9] = vp9_convolve8_avg_vert_c; |
| 132 | mSwitchConvolve[10] = vp9_convolve8_vert_c; |
| 133 | mSwitchConvolve[11] = vp9_convolve8_avg_vert_c; |
| 134 | mSwitchConvolve[12] = vp9_convolve8_c; |
| 135 | mSwitchConvolve[13] = vp9_convolve8_avg_c; |
| 136 | mSwitchConvolve[14] = vp9_convolve8_c; |
| 137 | mSwitchConvolve[15] = vp9_convolve8_avg_c; |
| 138 | |
| 139 | mSwitchConvolve[16] = vp9_convolve8_horiz_c; |
| 140 | mSwitchConvolve[17] = vp9_convolve8_avg_horiz_c; |
| 141 | mSwitchConvolve[18] = vp9_convolve8_c; |
| 142 | mSwitchConvolve[19] = vp9_convolve8_avg_c; |
| 143 | mSwitchConvolve[20] = vp9_convolve8_horiz_c; |
| 144 | mSwitchConvolve[21] = vp9_convolve8_avg_horiz_c; |
| 145 | mSwitchConvolve[22] = vp9_convolve8_c; |
| 146 | mSwitchConvolve[23] = vp9_convolve8_avg_c; |
| 147 | |
| 148 | mSwitchConvolve[24] = vp9_convolve8_c; |
| 149 | mSwitchConvolve[25] = vp9_convolve8_avg_c; |
| 150 | mSwitchConvolve[26] = vp9_convolve8_c; |
| 151 | mSwitchConvolve[27] = vp9_convolve8_avg_c; |
| 152 | mSwitchConvolve[28] = vp9_convolve8_c; |
| 153 | mSwitchConvolve[29] = vp9_convolve8_avg_c; |
| 154 | mSwitchConvolve[30] = vp9_convolve8_c; |
| 155 | mSwitchConvolve[31] = vp9_convolve8_avg_c; |
| 156 | #endif |
| 157 | } |
| 158 | |
| 159 | RsdCpuScriptIntrinsicInterPred::~RsdCpuScriptIntrinsicInterPred() { |
| 160 | } |
| 161 | |
| 162 | void RsdCpuScriptIntrinsicInterPred::populateScript(Script *s) { |
| 163 | s->mHal.info.exportedVariableCount = 3; |
| 164 | } |
| 165 | |
| 166 | void RsdCpuScriptIntrinsicInterPred::invokeFreeChildren() { |
| 167 | } |
| 168 | |
| 169 | |
| 170 | RsdCpuScriptImpl * rsdIntrinsic_InterPred(RsdCpuReferenceImpl *ctx, |
| 171 | const Script *s, const Element *e) { |
| 172 | return new RsdCpuScriptIntrinsicInterPred(ctx, s, e); |
| 173 | } |