/* ------------------------------------------------------------------ * Copyright (C) 1998-2009 PacketVideo * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at * * http://www.apache.org/licenses/LICENSE-2.0 * * Unless required by applicable law or agreed to in writing, software * distributed under the License is distributed on an "AS IS" BASIS, * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either * express or implied. * See the License for the specific language governing permissions * and limitations under the License. * ------------------------------------------------------------------- */ #include "avcenc_lib.h" #include "avcenc_int.h" #define CLIP_RESULT(x) if((uint)x > 0xFF){ \ x = 0xFF & (~(x>>31));} /* (blkwidth << 2) + (dy << 1) + dx */ static void (*const eChromaMC_SIMD[8])(uint8 *, int , int , int , uint8 *, int, int , int) = { &eChromaFullMC_SIMD, &eChromaHorizontalMC_SIMD, &eChromaVerticalMC_SIMD, &eChromaDiagonalMC_SIMD, &eChromaFullMC_SIMD, &eChromaHorizontalMC2_SIMD, &eChromaVerticalMC2_SIMD, &eChromaDiagonalMC2_SIMD }; /* Perform motion prediction and compensation with residue if exist. */ void AVCMBMotionComp(AVCEncObject *encvid, AVCCommonObj *video) { (void)(encvid); AVCMacroblock *currMB = video->currMB; AVCPictureData *currPic = video->currPic; int mbPartIdx, subMbPartIdx; int ref_idx; int offset_MbPart_indx = 0; int16 *mv; uint32 x_pos, y_pos; uint8 *curL, *curCb, *curCr; uint8 *ref_l, *ref_Cb, *ref_Cr; uint8 *predBlock, *predCb, *predCr; int block_x, block_y, offset_x, offset_y, offsetP, offset; int x_position = (video->mb_x << 4); int y_position = (video->mb_y << 4); int MbHeight, MbWidth, mbPartIdx_X, mbPartIdx_Y, offset_indx; int picWidth = currPic->width; int picPitch = currPic->pitch; int picHeight = currPic->height; uint32 tmp_word; tmp_word = y_position * picPitch; curL = currPic->Sl + tmp_word + x_position; offset = (tmp_word >> 2) + (x_position >> 1); curCb = currPic->Scb + offset; curCr = currPic->Scr + offset; predBlock = curL; predCb = curCb; predCr = curCr; GetMotionVectorPredictor(video, 1); for (mbPartIdx = 0; mbPartIdx < currMB->NumMbPart; mbPartIdx++) { MbHeight = currMB->SubMbPartHeight[mbPartIdx]; MbWidth = currMB->SubMbPartWidth[mbPartIdx]; mbPartIdx_X = ((mbPartIdx + offset_MbPart_indx) & 1); mbPartIdx_Y = (mbPartIdx + offset_MbPart_indx) >> 1; ref_idx = currMB->ref_idx_L0[(mbPartIdx_Y << 1) + mbPartIdx_X]; offset_indx = 0; ref_l = video->RefPicList0[ref_idx]->Sl; ref_Cb = video->RefPicList0[ref_idx]->Scb; ref_Cr = video->RefPicList0[ref_idx]->Scr; for (subMbPartIdx = 0; subMbPartIdx < currMB->NumSubMbPart[mbPartIdx]; subMbPartIdx++) { block_x = (mbPartIdx_X << 1) + ((subMbPartIdx + offset_indx) & 1); block_y = (mbPartIdx_Y << 1) + (((subMbPartIdx + offset_indx) >> 1) & 1); mv = (int16*)(currMB->mvL0 + block_x + (block_y << 2)); offset_x = x_position + (block_x << 2); offset_y = y_position + (block_y << 2); x_pos = (offset_x << 2) + *mv++; /*quarter pel */ y_pos = (offset_y << 2) + *mv; /*quarter pel */ //offset = offset_y * currPic->width; //offsetC = (offset >> 2) + (offset_x >> 1); offsetP = (block_y << 2) * picPitch + (block_x << 2); eLumaMotionComp(ref_l, picPitch, picHeight, x_pos, y_pos, /*comp_Sl + offset + offset_x,*/ predBlock + offsetP, picPitch, MbWidth, MbHeight); offsetP = (block_y * picWidth) + (block_x << 1); eChromaMotionComp(ref_Cb, picWidth >> 1, picHeight >> 1, x_pos, y_pos, /*comp_Scb + offsetC,*/ predCb + offsetP, picPitch >> 1, MbWidth >> 1, MbHeight >> 1); eChromaMotionComp(ref_Cr, picWidth >> 1, picHeight >> 1, x_pos, y_pos, /*comp_Scr + offsetC,*/ predCr + offsetP, picPitch >> 1, MbWidth >> 1, MbHeight >> 1); offset_indx = currMB->SubMbPartWidth[mbPartIdx] >> 3; } offset_MbPart_indx = currMB->MbPartWidth >> 4; } return ; } /* preform the actual motion comp here */ void eLumaMotionComp(uint8 *ref, int picpitch, int picheight, int x_pos, int y_pos, uint8 *pred, int pred_pitch, int blkwidth, int blkheight) { (void)(picheight); int dx, dy; int temp2[21][21]; /* for intermediate results */ uint8 *ref2; dx = x_pos & 3; dy = y_pos & 3; x_pos = x_pos >> 2; /* round it to full-pel resolution */ y_pos = y_pos >> 2; /* perform actual motion compensation */ if (dx == 0 && dy == 0) { /* fullpel position *//* G */ ref += y_pos * picpitch + x_pos; eFullPelMC(ref, picpitch, pred, pred_pitch, blkwidth, blkheight); } /* other positions */ else if (dy == 0) { /* no vertical interpolation *//* a,b,c*/ ref += y_pos * picpitch + x_pos; eHorzInterp1MC(ref, picpitch, pred, pred_pitch, blkwidth, blkheight, dx); } else if (dx == 0) { /*no horizontal interpolation *//* d,h,n */ ref += y_pos * picpitch + x_pos; eVertInterp1MC(ref, picpitch, pred, pred_pitch, blkwidth, blkheight, dy); } else if (dy == 2) { /* horizontal cross *//* i, j, k */ ref += y_pos * picpitch + x_pos - 2; /* move to the left 2 pixels */ eVertInterp2MC(ref, picpitch, &temp2[0][0], 21, blkwidth + 5, blkheight); eHorzInterp2MC(&temp2[0][2], 21, pred, pred_pitch, blkwidth, blkheight, dx); } else if (dx == 2) { /* vertical cross */ /* f,q */ ref += (y_pos - 2) * picpitch + x_pos; /* move to up 2 lines */ eHorzInterp3MC(ref, picpitch, &temp2[0][0], 21, blkwidth, blkheight + 5); eVertInterp3MC(&temp2[2][0], 21, pred, pred_pitch, blkwidth, blkheight, dy); } else { /* diagonal *//* e,g,p,r */ ref2 = ref + (y_pos + (dy / 2)) * picpitch + x_pos; ref += (y_pos * picpitch) + x_pos + (dx / 2); eDiagonalInterpMC(ref2, ref, picpitch, pred, pred_pitch, blkwidth, blkheight); } return ; } void eCreateAlign(uint8 *ref, int picpitch, int y_pos, uint8 *out, int blkwidth, int blkheight) { int i, j; int offset, out_offset; uint32 prev_pix, result, pix1, pix2, pix4; ref += y_pos * picpitch;// + x_pos; out_offset = 24 - blkwidth; //switch(x_pos&0x3){ switch (((intptr_t)ref)&0x3) { case 1: offset = picpitch - blkwidth - 3; for (j = 0; j < blkheight; j++) { pix1 = *ref++; pix2 = *((uint16*)ref); ref += 2; result = (pix2 << 8) | pix1; for (i = 3; i < blkwidth; i += 4) { pix4 = *((uint32*)ref); ref += 4; prev_pix = (pix4 << 24) & 0xFF000000; /* mask out byte belong to previous word */ result |= prev_pix; *((uint32*)out) = result; /* write 4 bytes */ out += 4; result = pix4 >> 8; /* for the next loop */ } ref += offset; out += out_offset; } break; case 2: offset = picpitch - blkwidth - 2; for (j = 0; j < blkheight; j++) { result = *((uint16*)ref); ref += 2; for (i = 2; i < blkwidth; i += 4) { pix4 = *((uint32*)ref); ref += 4; prev_pix = (pix4 << 16) & 0xFFFF0000; /* mask out byte belong to previous word */ result |= prev_pix; *((uint32*)out) = result; /* write 4 bytes */ out += 4; result = pix4 >> 16; /* for the next loop */ } ref += offset; out += out_offset; } break; case 3: offset = picpitch - blkwidth - 1; for (j = 0; j < blkheight; j++) { result = *ref++; for (i = 1; i < blkwidth; i += 4) { pix4 = *((uint32*)ref); ref += 4; prev_pix = (pix4 << 8) & 0xFFFFFF00; /* mask out byte belong to previous word */ result |= prev_pix; *((uint32*)out) = result; /* write 4 bytes */ out += 4; result = pix4 >> 24; /* for the next loop */ } ref += offset; out += out_offset; } break; } } void eHorzInterp1MC(uint8 *in, int inpitch, uint8 *out, int outpitch, int blkwidth, int blkheight, int dx) { uint8 *p_ref, *tmp; uint32 *p_cur; uint32 pkres; int result, curr_offset, ref_offset; int j; int32 r0, r1, r2, r3, r4, r5; int32 r13, r6; p_cur = (uint32*)out; /* assume it's word aligned */ curr_offset = (outpitch - blkwidth) >> 2; p_ref = in; ref_offset = inpitch - blkwidth; if (dx&1) { dx = ((dx >> 1) ? -3 : -4); /* use in 3/4 pel */ p_ref -= 2; r13 = 0; for (j = blkheight; j > 0; j--) { tmp = p_ref + blkwidth; r0 = p_ref[0]; r1 = p_ref[2]; r0 |= (r1 << 16); /* 0,c,0,a */ r1 = p_ref[1]; r2 = p_ref[3]; r1 |= (r2 << 16); /* 0,d,0,b */ while (p_ref < tmp) { r2 = *(p_ref += 4); /* move pointer to e */ r3 = p_ref[2]; r2 |= (r3 << 16); /* 0,g,0,e */ r3 = p_ref[1]; r4 = p_ref[3]; r3 |= (r4 << 16); /* 0,h,0,f */ r4 = r0 + r3; /* c+h, a+f */ r5 = r0 + r1; /* c+d, a+b */ r6 = r2 + r3; /* g+h, e+f */ r5 >>= 16; r5 |= (r6 << 16); /* e+f, c+d */ r4 += r5 * 20; /* c+20*e+20*f+h, a+20*c+20*d+f */ r4 += 0x100010; /* +16, +16 */ r5 = r1 + r2; /* d+g, b+e */ r4 -= r5 * 5; /* c-5*d+20*e+20*f-5*g+h, a-5*b+20*c+20*d-5*e+f */ r4 >>= 5; r13 |= r4; /* check clipping */ r5 = p_ref[dx+2]; r6 = p_ref[dx+4]; r5 |= (r6 << 16); r4 += r5; r4 += 0x10001; r4 = (r4 >> 1) & 0xFF00FF; r5 = p_ref[4]; /* i */ r6 = (r5 << 16); r5 = r6 | (r2 >> 16);/* 0,i,0,g */ r5 += r1; /* d+i, b+g */ /* r5 not free */ r1 >>= 16; r1 |= (r3 << 16); /* 0,f,0,d */ /* r1 has changed */ r1 += r2; /* f+g, d+e */ r5 += 20 * r1; /* d+20f+20g+i, b+20d+20e+g */ r0 >>= 16; r0 |= (r2 << 16); /* 0,e,0,c */ /* r0 has changed */ r0 += r3; /* e+h, c+f */ r5 += 0x100010; /* 16,16 */ r5 -= r0 * 5; /* d-5e+20f+20g-5h+i, b-5c+20d+20e-5f+g */ r5 >>= 5; r13 |= r5; /* check clipping */ r0 = p_ref[dx+3]; r1 = p_ref[dx+5]; r0 |= (r1 << 16); r5 += r0; r5 += 0x10001; r5 = (r5 >> 1) & 0xFF00FF; r4 |= (r5 << 8); /* pack them together */ *p_cur++ = r4; r1 = r3; r0 = r2; } p_cur += curr_offset; /* move to the next line */ p_ref += ref_offset; /* ref_offset = inpitch-blkwidth; */ if (r13&0xFF000700) /* need clipping */ { /* move back to the beginning of the line */ p_ref -= (ref_offset + blkwidth); /* input */ p_cur -= (outpitch >> 2); tmp = p_ref + blkwidth; for (; p_ref < tmp;) { r0 = *p_ref++; r1 = *p_ref++; r2 = *p_ref++; r3 = *p_ref++; r4 = *p_ref++; /* first pixel */ r5 = *p_ref++; result = (r0 + r5); r0 = (r1 + r4); result -= (r0 * 5);//result -= r0; result -= (r0<<2); r0 = (r2 + r3); result += (r0 * 20);//result += (r0<<4); result += (r0<<2); result = (result + 16) >> 5; CLIP_RESULT(result) /* 3/4 pel, no need to clip */ result = (result + p_ref[dx] + 1); pkres = (result >> 1) ; /* second pixel */ r0 = *p_ref++; result = (r1 + r0); r1 = (r2 + r5); result -= (r1 * 5);//result -= r1; result -= (r1<<2); r1 = (r3 + r4); result += (r1 * 20);//result += (r1<<4); result += (r1<<2); result = (result + 16) >> 5; CLIP_RESULT(result) /* 3/4 pel, no need to clip */ result = (result + p_ref[dx] + 1); result = (result >> 1); pkres |= (result << 8); /* third pixel */ r1 = *p_ref++; result = (r2 + r1); r2 = (r3 + r0); result -= (r2 * 5);//result -= r2; result -= (r2<<2); r2 = (r4 + r5); result += (r2 * 20);//result += (r2<<4); result += (r2<<2); result = (result + 16) >> 5; CLIP_RESULT(result) /* 3/4 pel, no need to clip */ result = (result + p_ref[dx] + 1); result = (result >> 1); pkres |= (result << 16); /* fourth pixel */ r2 = *p_ref++; result = (r3 + r2); r3 = (r4 + r1); result -= (r3 * 5);//result -= r3; result -= (r3<<2); r3 = (r5 + r0); result += (r3 * 20);//result += (r3<<4); result += (r3<<2); result = (result + 16) >> 5; CLIP_RESULT(result) /* 3/4 pel, no need to clip */ result = (result + p_ref[dx] + 1); result = (result >> 1); pkres |= (result << 24); *p_cur++ = pkres; /* write 4 pixels */ p_ref -= 5; /* offset back to the middle of filter */ } p_cur += curr_offset; /* move to the next line */ p_ref += ref_offset; /* move to the next line */ } } } else { p_ref -= 2; r13 = 0; for (j = blkheight; j > 0; j--) { tmp = p_ref + blkwidth; r0 = p_ref[0]; r1 = p_ref[2]; r0 |= (r1 << 16); /* 0,c,0,a */ r1 = p_ref[1]; r2 = p_ref[3]; r1 |= (r2 << 16); /* 0,d,0,b */ while (p_ref < tmp) { r2 = *(p_ref += 4); /* move pointer to e */ r3 = p_ref[2]; r2 |= (r3 << 16); /* 0,g,0,e */ r3 = p_ref[1]; r4 = p_ref[3]; r3 |= (r4 << 16); /* 0,h,0,f */ r4 = r0 + r3; /* c+h, a+f */ r5 = r0 + r1; /* c+d, a+b */ r6 = r2 + r3; /* g+h, e+f */ r5 >>= 16; r5 |= (r6 << 16); /* e+f, c+d */ r4 += r5 * 20; /* c+20*e+20*f+h, a+20*c+20*d+f */ r4 += 0x100010; /* +16, +16 */ r5 = r1 + r2; /* d+g, b+e */ r4 -= r5 * 5; /* c-5*d+20*e+20*f-5*g+h, a-5*b+20*c+20*d-5*e+f */ r4 >>= 5; r13 |= r4; /* check clipping */ r4 &= 0xFF00FF; /* mask */ r5 = p_ref[4]; /* i */ r6 = (r5 << 16); r5 = r6 | (r2 >> 16);/* 0,i,0,g */ r5 += r1; /* d+i, b+g */ /* r5 not free */ r1 >>= 16; r1 |= (r3 << 16); /* 0,f,0,d */ /* r1 has changed */ r1 += r2; /* f+g, d+e */ r5 += 20 * r1; /* d+20f+20g+i, b+20d+20e+g */ r0 >>= 16; r0 |= (r2 << 16); /* 0,e,0,c */ /* r0 has changed */ r0 += r3; /* e+h, c+f */ r5 += 0x100010; /* 16,16 */ r5 -= r0 * 5; /* d-5e+20f+20g-5h+i, b-5c+20d+20e-5f+g */ r5 >>= 5; r13 |= r5; /* check clipping */ r5 &= 0xFF00FF; /* mask */ r4 |= (r5 << 8); /* pack them together */ *p_cur++ = r4; r1 = r3; r0 = r2; } p_cur += curr_offset; /* move to the next line */ p_ref += ref_offset; /* ref_offset = inpitch-blkwidth; */ if (r13&0xFF000700) /* need clipping */ { /* move back to the beginning of the line */ p_ref -= (ref_offset + blkwidth); /* input */ p_cur -= (outpitch >> 2); tmp = p_ref + blkwidth; for (; p_ref < tmp;) { r0 = *p_ref++; r1 = *p_ref++; r2 = *p_ref++; r3 = *p_ref++; r4 = *p_ref++; /* first pixel */ r5 = *p_ref++; result = (r0 + r5); r0 = (r1 + r4); result -= (r0 * 5);//result -= r0; result -= (r0<<2); r0 = (r2 + r3); result += (r0 * 20);//result += (r0<<4); result += (r0<<2); result = (result + 16) >> 5; CLIP_RESULT(result) pkres = result; /* second pixel */ r0 = *p_ref++; result = (r1 + r0); r1 = (r2 + r5); result -= (r1 * 5);//result -= r1; result -= (r1<<2); r1 = (r3 + r4); result += (r1 * 20);//result += (r1<<4); result += (r1<<2); result = (result + 16) >> 5; CLIP_RESULT(result) pkres |= (result << 8); /* third pixel */ r1 = *p_ref++; result = (r2 + r1); r2 = (r3 + r0); result -= (r2 * 5);//result -= r2; result -= (r2<<2); r2 = (r4 + r5); result += (r2 * 20);//result += (r2<<4); result += (r2<<2); result = (result + 16) >> 5; CLIP_RESULT(result) pkres |= (result << 16); /* fourth pixel */ r2 = *p_ref++; result = (r3 + r2); r3 = (r4 + r1); result -= (r3 * 5);//result -= r3; result -= (r3<<2); r3 = (r5 + r0); result += (r3 * 20);//result += (r3<<4); result += (r3<<2); result = (result + 16) >> 5; CLIP_RESULT(result) pkres |= (result << 24); *p_cur++ = pkres; /* write 4 pixels */ p_ref -= 5; } p_cur += curr_offset; /* move to the next line */ p_ref += ref_offset; } } } return ; } void eHorzInterp2MC(int *in, int inpitch, uint8 *out, int outpitch, int blkwidth, int blkheight, int dx) { int *p_ref, *tmp; uint32 *p_cur; uint32 pkres; int result, result2, curr_offset, ref_offset; int j, r0, r1, r2, r3, r4, r5; p_cur = (uint32*)out; /* assume it's word aligned */ curr_offset = (outpitch - blkwidth) >> 2; p_ref = in; ref_offset = inpitch - blkwidth; if (dx&1) { dx = ((dx >> 1) ? -3 : -4); /* use in 3/4 pel */ for (j = blkheight; j > 0 ; j--) { tmp = p_ref + blkwidth; for (; p_ref < tmp;) { r0 = p_ref[-2]; r1 = p_ref[-1]; r2 = *p_ref++; r3 = *p_ref++; r4 = *p_ref++; /* first pixel */ r5 = *p_ref++; result = (r0 + r5); r0 = (r1 + r4); result -= (r0 * 5);//result -= r0; result -= (r0<<2); r0 = (r2 + r3); result += (r0 * 20);//result += (r0<<4); result += (r0<<2); result = (result + 512) >> 10; CLIP_RESULT(result) result2 = ((p_ref[dx] + 16) >> 5); CLIP_RESULT(result2) /* 3/4 pel, no need to clip */ result = (result + result2 + 1); pkres = (result >> 1); /* second pixel */ r0 = *p_ref++; result = (r1 + r0); r1 = (r2 + r5); result -= (r1 * 5);//result -= r1; result -= (r1<<2); r1 = (r3 + r4); result += (r1 * 20);//result += (r1<<4); result += (r1<<2); result = (result + 512) >> 10; CLIP_RESULT(result) result2 = ((p_ref[dx] + 16) >> 5); CLIP_RESULT(result2) /* 3/4 pel, no need to clip */ result = (result + result2 + 1); result = (result >> 1); pkres |= (result << 8); /* third pixel */ r1 = *p_ref++; result = (r2 + r1); r2 = (r3 + r0); result -= (r2 * 5);//result -= r2; result -= (r2<<2); r2 = (r4 + r5); result += (r2 * 20);//result += (r2<<4); result += (r2<<2); result = (result + 512) >> 10; CLIP_RESULT(result) result2 = ((p_ref[dx] + 16) >> 5); CLIP_RESULT(result2) /* 3/4 pel, no need to clip */ result = (result + result2 + 1); result = (result >> 1); pkres |= (result << 16); /* fourth pixel */ r2 = *p_ref++; result = (r3 + r2); r3 = (r4 + r1); result -= (r3 * 5);//result -= r3; result -= (r3<<2); r3 = (r5 + r0); result += (r3 * 20);//result += (r3<<4); result += (r3<<2); result = (result + 512) >> 10; CLIP_RESULT(result) result2 = ((p_ref[dx] + 16) >> 5); CLIP_RESULT(result2) /* 3/4 pel, no need to clip */ result = (result + result2 + 1); result = (result >> 1); pkres |= (result << 24); *p_cur++ = pkres; /* write 4 pixels */ p_ref -= 3; /* offset back to the middle of filter */ } p_cur += curr_offset; /* move to the next line */ p_ref += ref_offset; /* move to the next line */ } } else { for (j = blkheight; j > 0 ; j--) { tmp = p_ref + blkwidth; for (; p_ref < tmp;) { r0 = p_ref[-2]; r1 = p_ref[-1]; r2 = *p_ref++; r3 = *p_ref++; r4 = *p_ref++; /* first pixel */ r5 = *p_ref++; result = (r0 + r5); r0 = (r1 + r4); result -= (r0 * 5);//result -= r0; result -= (r0<<2); r0 = (r2 + r3); result += (r0 * 20);//result += (r0<<4); result += (r0<<2); result = (result + 512) >> 10; CLIP_RESULT(result) pkres = result; /* second pixel */ r0 = *p_ref++; result = (r1 + r0); r1 = (r2 + r5); result -= (r1 * 5);//result -= r1; result -= (r1<<2); r1 = (r3 + r4); result += (r1 * 20);//result += (r1<<4); result += (r1<<2); result = (result + 512) >> 10; CLIP_RESULT(result) pkres |= (result << 8); /* third pixel */ r1 = *p_ref++; result = (r2 + r1); r2 = (r3 + r0); result -= (r2 * 5);//result -= r2; result -= (r2<<2); r2 = (r4 + r5); result += (r2 * 20);//result += (r2<<4); result += (r2<<2); result = (result + 512) >> 10; CLIP_RESULT(result) pkres |= (result << 16); /* fourth pixel */ r2 = *p_ref++; result = (r3 + r2); r3 = (r4 + r1); result -= (r3 * 5);//result -= r3; result -= (r3<<2); r3 = (r5 + r0); result += (r3 * 20);//result += (r3<<4); result += (r3<<2); result = (result + 512) >> 10; CLIP_RESULT(result) pkres |= (result << 24); *p_cur++ = pkres; /* write 4 pixels */ p_ref -= 3; /* offset back to the middle of filter */ } p_cur += curr_offset; /* move to the next line */ p_ref += ref_offset; /* move to the next line */ } } return ; } void eHorzInterp3MC(uint8 *in, int inpitch, int *out, int outpitch, int blkwidth, int blkheight) { uint8 *p_ref, *tmp; int *p_cur; int result, curr_offset, ref_offset; int j, r0, r1, r2, r3, r4, r5; p_cur = out; curr_offset = (outpitch - blkwidth); p_ref = in; ref_offset = inpitch - blkwidth; for (j = blkheight; j > 0 ; j--) { tmp = p_ref + blkwidth; for (; p_ref < tmp;) { r0 = p_ref[-2]; r1 = p_ref[-1]; r2 = *p_ref++; r3 = *p_ref++; r4 = *p_ref++; /* first pixel */ r5 = *p_ref++; result = (r0 + r5); r0 = (r1 + r4); result -= (r0 * 5);//result -= r0; result -= (r0<<2); r0 = (r2 + r3); result += (r0 * 20);//result += (r0<<4); result += (r0<<2); *p_cur++ = result; /* second pixel */ r0 = *p_ref++; result = (r1 + r0); r1 = (r2 + r5); result -= (r1 * 5);//result -= r1; result -= (r1<<2); r1 = (r3 + r4); result += (r1 * 20);//result += (r1<<4); result += (r1<<2); *p_cur++ = result; /* third pixel */ r1 = *p_ref++; result = (r2 + r1); r2 = (r3 + r0); result -= (r2 * 5);//result -= r2; result -= (r2<<2); r2 = (r4 + r5); result += (r2 * 20);//result += (r2<<4); result += (r2<<2); *p_cur++ = result; /* fourth pixel */ r2 = *p_ref++; result = (r3 + r2); r3 = (r4 + r1); result -= (r3 * 5);//result -= r3; result -= (r3<<2); r3 = (r5 + r0); result += (r3 * 20);//result += (r3<<4); result += (r3<<2); *p_cur++ = result; p_ref -= 3; /* move back to the middle of the filter */ } p_cur += curr_offset; /* move to the next line */ p_ref += ref_offset; } return ; } void eVertInterp1MC(uint8 *in, int inpitch, uint8 *out, int outpitch, int blkwidth, int blkheight, int dy) { uint8 *p_cur, *p_ref, *tmp; int result, curr_offset, ref_offset; int j, i; int32 r0, r1, r2, r3, r4, r5, r6, r7, r8, r13; uint8 tmp_in[24][24]; /* not word-aligned */ if (((intptr_t)in)&0x3) { eCreateAlign(in, inpitch, -2, &tmp_in[0][0], blkwidth, blkheight + 5); in = &tmp_in[2][0]; inpitch = 24; } p_cur = out; curr_offset = 1 - outpitch * (blkheight - 1); /* offset vertically back up and one pixel to right */ ref_offset = blkheight * inpitch; /* for limit */ curr_offset += 3; if (dy&1) { dy = (dy >> 1) ? 0 : -inpitch; for (j = 0; j < blkwidth; j += 4, in += 4) { r13 = 0; p_ref = in; p_cur -= outpitch; /* compensate for the first offset */ tmp = p_ref + ref_offset; /* limit */ while (p_ref < tmp) /* the loop un-rolled */ { r0 = *((uint32*)(p_ref - (inpitch << 1))); /* load 4 bytes */ p_ref += inpitch; r6 = (r0 >> 8) & 0xFF00FF; /* second and fourth byte */ r0 &= 0xFF00FF; r1 = *((uint32*)(p_ref + (inpitch << 1))); /* r1, r7, ref[3] */ r7 = (r1 >> 8) & 0xFF00FF; r1 &= 0xFF00FF; r0 += r1; r6 += r7; r2 = *((uint32*)p_ref); /* r2, r8, ref[1] */ r8 = (r2 >> 8) & 0xFF00FF; r2 &= 0xFF00FF; r1 = *((uint32*)(p_ref - inpitch)); /* r1, r7, ref[0] */ r7 = (r1 >> 8) & 0xFF00FF; r1 &= 0xFF00FF; r1 += r2; r7 += r8; r0 += 20 * r1; r6 += 20 * r7; r0 += 0x100010; r6 += 0x100010; r2 = *((uint32*)(p_ref - (inpitch << 1))); /* r2, r8, ref[-1] */ r8 = (r2 >> 8) & 0xFF00FF; r2 &= 0xFF00FF; r1 = *((uint32*)(p_ref + inpitch)); /* r1, r7, ref[2] */ r7 = (r1 >> 8) & 0xFF00FF; r1 &= 0xFF00FF; r1 += r2; r7 += r8; r0 -= 5 * r1; r6 -= 5 * r7; r0 >>= 5; r6 >>= 5; /* clip */ r13 |= r6; r13 |= r0; //CLIPPACK(r6,result) r1 = *((uint32*)(p_ref + dy)); r2 = (r1 >> 8) & 0xFF00FF; r1 &= 0xFF00FF; r0 += r1; r6 += r2; r0 += 0x10001; r6 += 0x10001; r0 = (r0 >> 1) & 0xFF00FF; r6 = (r6 >> 1) & 0xFF00FF; r0 |= (r6 << 8); /* pack it back */ *((uint32*)(p_cur += outpitch)) = r0; } p_cur += curr_offset; /* offset to the next pixel */ if (r13 & 0xFF000700) /* this column need clipping */ { p_cur -= 4; for (i = 0; i < 4; i++) { p_ref = in + i; p_cur -= outpitch; /* compensate for the first offset */ tmp = p_ref + ref_offset; /* limit */ while (p_ref < tmp) { /* loop un-rolled */ r0 = *(p_ref - (inpitch << 1)); r1 = *(p_ref - inpitch); r2 = *p_ref; r3 = *(p_ref += inpitch); /* modify pointer before loading */ r4 = *(p_ref += inpitch); /* first pixel */ r5 = *(p_ref += inpitch); result = (r0 + r5); r0 = (r1 + r4); result -= (r0 * 5);//result -= r0; result -= (r0<<2); r0 = (r2 + r3); result += (r0 * 20);//result += (r0<<4); result += (r0<<2); result = (result + 16) >> 5; CLIP_RESULT(result) /* 3/4 pel, no need to clip */ result = (result + p_ref[dy-(inpitch<<1)] + 1); result = (result >> 1); *(p_cur += outpitch) = result; /* second pixel */ r0 = *(p_ref += inpitch); result = (r1 + r0); r1 = (r2 + r5); result -= (r1 * 5);//result -= r1; result -= (r1<<2); r1 = (r3 + r4); result += (r1 * 20);//result += (r1<<4); result += (r1<<2); result = (result + 16) >> 5; CLIP_RESULT(result) /* 3/4 pel, no need to clip */ result = (result + p_ref[dy-(inpitch<<1)] + 1); result = (result >> 1); *(p_cur += outpitch) = result; /* third pixel */ r1 = *(p_ref += inpitch); result = (r2 + r1); r2 = (r3 + r0); result -= (r2 * 5);//result -= r2; result -= (r2<<2); r2 = (r4 + r5); result += (r2 * 20);//result += (r2<<4); result += (r2<<2); result = (result + 16) >> 5; CLIP_RESULT(result) /* 3/4 pel, no need to clip */ result = (result + p_ref[dy-(inpitch<<1)] + 1); result = (result >> 1); *(p_cur += outpitch) = result; /* fourth pixel */ r2 = *(p_ref += inpitch); result = (r3 + r2); r3 = (r4 + r1); result -= (r3 * 5);//result -= r3; result -= (r3<<2); r3 = (r5 + r0); result += (r3 * 20);//result += (r3<<4); result += (r3<<2); result = (result + 16) >> 5; CLIP_RESULT(result) /* 3/4 pel, no need to clip */ result = (result + p_ref[dy-(inpitch<<1)] + 1); result = (result >> 1); *(p_cur += outpitch) = result; p_ref -= (inpitch << 1); /* move back to center of the filter of the next one */ } p_cur += (curr_offset - 3); } } } } else { for (j = 0; j < blkwidth; j += 4, in += 4) { r13 = 0; p_ref = in; p_cur -= outpitch; /* compensate for the first offset */ tmp = p_ref + ref_offset; /* limit */ while (p_ref < tmp) /* the loop un-rolled */ { r0 = *((uint32*)(p_ref - (inpitch << 1))); /* load 4 bytes */ p_ref += inpitch; r6 = (r0 >> 8) & 0xFF00FF; /* second and fourth byte */ r0 &= 0xFF00FF; r1 = *((uint32*)(p_ref + (inpitch << 1))); /* r1, r7, ref[3] */ r7 = (r1 >> 8) & 0xFF00FF; r1 &= 0xFF00FF; r0 += r1; r6 += r7; r2 = *((uint32*)p_ref); /* r2, r8, ref[1] */ r8 = (r2 >> 8) & 0xFF00FF; r2 &= 0xFF00FF; r1 = *((uint32*)(p_ref - inpitch)); /* r1, r7, ref[0] */ r7 = (r1 >> 8) & 0xFF00FF; r1 &= 0xFF00FF; r1 += r2; r7 += r8; r0 += 20 * r1; r6 += 20 * r7; r0 += 0x100010; r6 += 0x100010; r2 = *((uint32*)(p_ref - (inpitch << 1))); /* r2, r8, ref[-1] */ r8 = (r2 >> 8) & 0xFF00FF; r2 &= 0xFF00FF; r1 = *((uint32*)(p_ref + inpitch)); /* r1, r7, ref[2] */ r7 = (r1 >> 8) & 0xFF00FF; r1 &= 0xFF00FF; r1 += r2; r7 += r8; r0 -= 5 * r1; r6 -= 5 * r7; r0 >>= 5; r6 >>= 5; /* clip */ r13 |= r6; r13 |= r0; //CLIPPACK(r6,result) r0 &= 0xFF00FF; r6 &= 0xFF00FF; r0 |= (r6 << 8); /* pack it back */ *((uint32*)(p_cur += outpitch)) = r0; } p_cur += curr_offset; /* offset to the next pixel */ if (r13 & 0xFF000700) /* this column need clipping */ { p_cur -= 4; for (i = 0; i < 4; i++) { p_ref = in + i; p_cur -= outpitch; /* compensate for the first offset */ tmp = p_ref + ref_offset; /* limit */ while (p_ref < tmp) { /* loop un-rolled */ r0 = *(p_ref - (inpitch << 1)); r1 = *(p_ref - inpitch); r2 = *p_ref; r3 = *(p_ref += inpitch); /* modify pointer before loading */ r4 = *(p_ref += inpitch); /* first pixel */ r5 = *(p_ref += inpitch); result = (r0 + r5); r0 = (r1 + r4); result -= (r0 * 5);//result -= r0; result -= (r0<<2); r0 = (r2 + r3); result += (r0 * 20);//result += (r0<<4); result += (r0<<2); result = (result + 16) >> 5; CLIP_RESULT(result) *(p_cur += outpitch) = result; /* second pixel */ r0 = *(p_ref += inpitch); result = (r1 + r0); r1 = (r2 + r5); result -= (r1 * 5);//result -= r1; result -= (r1<<2); r1 = (r3 + r4); result += (r1 * 20);//result += (r1<<4); result += (r1<<2); result = (result + 16) >> 5; CLIP_RESULT(result) *(p_cur += outpitch) = result; /* third pixel */ r1 = *(p_ref += inpitch); result = (r2 + r1); r2 = (r3 + r0); result -= (r2 * 5);//result -= r2; result -= (r2<<2); r2 = (r4 + r5); result += (r2 * 20);//result += (r2<<4); result += (r2<<2); result = (result + 16) >> 5; CLIP_RESULT(result) *(p_cur += outpitch) = result; /* fourth pixel */ r2 = *(p_ref += inpitch); result = (r3 + r2); r3 = (r4 + r1); result -= (r3 * 5);//result -= r3; result -= (r3<<2); r3 = (r5 + r0); result += (r3 * 20);//result += (r3<<4); result += (r3<<2); result = (result + 16) >> 5; CLIP_RESULT(result) *(p_cur += outpitch) = result; p_ref -= (inpitch << 1); /* move back to center of the filter of the next one */ } p_cur += (curr_offset - 3); } } } } return ; } void eVertInterp2MC(uint8 *in, int inpitch, int *out, int outpitch, int blkwidth, int blkheight) { int *p_cur; uint8 *p_ref, *tmp; int result, curr_offset, ref_offset; int j, r0, r1, r2, r3, r4, r5; p_cur = out; curr_offset = 1 - outpitch * (blkheight - 1); /* offset vertically back up and one pixel to right */ ref_offset = blkheight * inpitch; /* for limit */ for (j = 0; j < blkwidth; j++) { p_cur -= outpitch; /* compensate for the first offset */ p_ref = in++; tmp = p_ref + ref_offset; /* limit */ while (p_ref < tmp) { /* loop un-rolled */ r0 = *(p_ref - (inpitch << 1)); r1 = *(p_ref - inpitch); r2 = *p_ref; r3 = *(p_ref += inpitch); /* modify pointer before loading */ r4 = *(p_ref += inpitch); /* first pixel */ r5 = *(p_ref += inpitch); result = (r0 + r5); r0 = (r1 + r4); result -= (r0 * 5);//result -= r0; result -= (r0<<2); r0 = (r2 + r3); result += (r0 * 20);//result += (r0<<4); result += (r0<<2); *(p_cur += outpitch) = result; /* second pixel */ r0 = *(p_ref += inpitch); result = (r1 + r0); r1 = (r2 + r5); result -= (r1 * 5);//result -= r1; result -= (r1<<2); r1 = (r3 + r4); result += (r1 * 20);//result += (r1<<4); result += (r1<<2); *(p_cur += outpitch) = result; /* third pixel */ r1 = *(p_ref += inpitch); result = (r2 + r1); r2 = (r3 + r0); result -= (r2 * 5);//result -= r2; result -= (r2<<2); r2 = (r4 + r5); result += (r2 * 20);//result += (r2<<4); result += (r2<<2); *(p_cur += outpitch) = result; /* fourth pixel */ r2 = *(p_ref += inpitch); result = (r3 + r2); r3 = (r4 + r1); result -= (r3 * 5);//result -= r3; result -= (r3<<2); r3 = (r5 + r0); result += (r3 * 20);//result += (r3<<4); result += (r3<<2); *(p_cur += outpitch) = result; p_ref -= (inpitch << 1); /* move back to center of the filter of the next one */ } p_cur += curr_offset; } return ; } void eVertInterp3MC(int *in, int inpitch, uint8 *out, int outpitch, int blkwidth, int blkheight, int dy) { uint8 *p_cur; int *p_ref, *tmp; int result, result2, curr_offset, ref_offset; int j, r0, r1, r2, r3, r4, r5; p_cur = out; curr_offset = 1 - outpitch * (blkheight - 1); /* offset vertically back up and one pixel to right */ ref_offset = blkheight * inpitch; /* for limit */ if (dy&1) { dy = (dy >> 1) ? -(inpitch << 1) : -(inpitch << 1) - inpitch; for (j = 0; j < blkwidth; j++) { p_cur -= outpitch; /* compensate for the first offset */ p_ref = in++; tmp = p_ref + ref_offset; /* limit */ while (p_ref < tmp) { /* loop un-rolled */ r0 = *(p_ref - (inpitch << 1)); r1 = *(p_ref - inpitch); r2 = *p_ref; r3 = *(p_ref += inpitch); /* modify pointer before loading */ r4 = *(p_ref += inpitch); /* first pixel */ r5 = *(p_ref += inpitch); result = (r0 + r5); r0 = (r1 + r4); result -= (r0 * 5);//result -= r0; result -= (r0<<2); r0 = (r2 + r3); result += (r0 * 20);//result += (r0<<4); result += (r0<<2); result = (result + 512) >> 10; CLIP_RESULT(result) result2 = ((p_ref[dy] + 16) >> 5); CLIP_RESULT(result2) /* 3/4 pel, no need to clip */ result = (result + result2 + 1); result = (result >> 1); *(p_cur += outpitch) = result; /* second pixel */ r0 = *(p_ref += inpitch); result = (r1 + r0); r1 = (r2 + r5); result -= (r1 * 5);//result -= r1; result -= (r1<<2); r1 = (r3 + r4); result += (r1 * 20);//result += (r1<<4); result += (r1<<2); result = (result + 512) >> 10; CLIP_RESULT(result) result2 = ((p_ref[dy] + 16) >> 5); CLIP_RESULT(result2) /* 3/4 pel, no need to clip */ result = (result + result2 + 1); result = (result >> 1); *(p_cur += outpitch) = result; /* third pixel */ r1 = *(p_ref += inpitch); result = (r2 + r1); r2 = (r3 + r0); result -= (r2 * 5);//result -= r2; result -= (r2<<2); r2 = (r4 + r5); result += (r2 * 20);//result += (r2<<4); result += (r2<<2); result = (result + 512) >> 10; CLIP_RESULT(result) result2 = ((p_ref[dy] + 16) >> 5); CLIP_RESULT(result2) /* 3/4 pel, no need to clip */ result = (result + result2 + 1); result = (result >> 1); *(p_cur += outpitch) = result; /* fourth pixel */ r2 = *(p_ref += inpitch); result = (r3 + r2); r3 = (r4 + r1); result -= (r3 * 5);//result -= r3; result -= (r3<<2); r3 = (r5 + r0); result += (r3 * 20);//result += (r3<<4); result += (r3<<2); result = (result + 512) >> 10; CLIP_RESULT(result) result2 = ((p_ref[dy] + 16) >> 5); CLIP_RESULT(result2) /* 3/4 pel, no need to clip */ result = (result + result2 + 1); result = (result >> 1); *(p_cur += outpitch) = result; p_ref -= (inpitch << 1); /* move back to center of the filter of the next one */ } p_cur += curr_offset; } } else { for (j = 0; j < blkwidth; j++) { p_cur -= outpitch; /* compensate for the first offset */ p_ref = in++; tmp = p_ref + ref_offset; /* limit */ while (p_ref < tmp) { /* loop un-rolled */ r0 = *(p_ref - (inpitch << 1)); r1 = *(p_ref - inpitch); r2 = *p_ref; r3 = *(p_ref += inpitch); /* modify pointer before loading */ r4 = *(p_ref += inpitch); /* first pixel */ r5 = *(p_ref += inpitch); result = (r0 + r5); r0 = (r1 + r4); result -= (r0 * 5);//result -= r0; result -= (r0<<2); r0 = (r2 + r3); result += (r0 * 20);//result += (r0<<4); result += (r0<<2); result = (result + 512) >> 10; CLIP_RESULT(result) *(p_cur += outpitch) = result; /* second pixel */ r0 = *(p_ref += inpitch); result = (r1 + r0); r1 = (r2 + r5); result -= (r1 * 5);//result -= r1; result -= (r1<<2); r1 = (r3 + r4); result += (r1 * 20);//result += (r1<<4); result += (r1<<2); result = (result + 512) >> 10; CLIP_RESULT(result) *(p_cur += outpitch) = result; /* third pixel */ r1 = *(p_ref += inpitch); result = (r2 + r1); r2 = (r3 + r0); result -= (r2 * 5);//result -= r2; result -= (r2<<2); r2 = (r4 + r5); result += (r2 * 20);//result += (r2<<4); result += (r2<<2); result = (result + 512) >> 10; CLIP_RESULT(result) *(p_cur += outpitch) = result; /* fourth pixel */ r2 = *(p_ref += inpitch); result = (r3 + r2); r3 = (r4 + r1); result -= (r3 * 5);//result -= r3; result -= (r3<<2); r3 = (r5 + r0); result += (r3 * 20);//result += (r3<<4); result += (r3<<2); result = (result + 512) >> 10; CLIP_RESULT(result) *(p_cur += outpitch) = result; p_ref -= (inpitch << 1); /* move back to center of the filter of the next one */ } p_cur += curr_offset; } } return ; } void eDiagonalInterpMC(uint8 *in1, uint8 *in2, int inpitch, uint8 *out, int outpitch, int blkwidth, int blkheight) { int j, i; int result; uint8 *p_cur, *p_ref, *p_tmp8, *tmp; int curr_offset, ref_offset; uint8 tmp_res[24][24], tmp_in[24][24]; uint32 *p_tmp; uint32 pkres, tmp_result; int32 r0, r1, r2, r3, r4, r5; int32 r6, r7, r8, r9, r10, r13; ref_offset = inpitch - blkwidth; p_ref = in1 - 2; /* perform horizontal interpolation */ /* not word-aligned */ /* It is faster to read 1 byte at time to avoid calling CreateAlign */ /* if(((uint32)p_ref)&0x3) { CreateAlign(p_ref,inpitch,0,&tmp_in[0][0],blkwidth+8,blkheight); p_ref = &tmp_in[0][0]; ref_offset = 24-blkwidth; }*/ p_tmp = (uint32*) & (tmp_res[0][0]); for (j = blkheight; j > 0; j--) { r13 = 0; tmp = p_ref + blkwidth; //r0 = *((uint32*)p_ref); /* d,c,b,a */ //r1 = (r0>>8)&0xFF00FF; /* 0,d,0,b */ //r0 &= 0xFF00FF; /* 0,c,0,a */ /* It is faster to read 1 byte at a time */ r0 = p_ref[0]; r1 = p_ref[2]; r0 |= (r1 << 16); /* 0,c,0,a */ r1 = p_ref[1]; r2 = p_ref[3]; r1 |= (r2 << 16); /* 0,d,0,b */ while (p_ref < tmp) { //r2 = *((uint32*)(p_ref+=4));/* h,g,f,e */ //r3 = (r2>>8)&0xFF00FF; /* 0,h,0,f */ //r2 &= 0xFF00FF; /* 0,g,0,e */ /* It is faster to read 1 byte at a time */ r2 = *(p_ref += 4); r3 = p_ref[2]; r2 |= (r3 << 16); /* 0,g,0,e */ r3 = p_ref[1]; r4 = p_ref[3]; r3 |= (r4 << 16); /* 0,h,0,f */ r4 = r0 + r3; /* c+h, a+f */ r5 = r0 + r1; /* c+d, a+b */ r6 = r2 + r3; /* g+h, e+f */ r5 >>= 16; r5 |= (r6 << 16); /* e+f, c+d */ r4 += r5 * 20; /* c+20*e+20*f+h, a+20*c+20*d+f */ r4 += 0x100010; /* +16, +16 */ r5 = r1 + r2; /* d+g, b+e */ r4 -= r5 * 5; /* c-5*d+20*e+20*f-5*g+h, a-5*b+20*c+20*d-5*e+f */ r4 >>= 5; r13 |= r4; /* check clipping */ r4 &= 0xFF00FF; /* mask */ r5 = p_ref[4]; /* i */ r6 = (r5 << 16); r5 = r6 | (r2 >> 16);/* 0,i,0,g */ r5 += r1; /* d+i, b+g */ /* r5 not free */ r1 >>= 16; r1 |= (r3 << 16); /* 0,f,0,d */ /* r1 has changed */ r1 += r2; /* f+g, d+e */ r5 += 20 * r1; /* d+20f+20g+i, b+20d+20e+g */ r0 >>= 16; r0 |= (r2 << 16); /* 0,e,0,c */ /* r0 has changed */ r0 += r3; /* e+h, c+f */ r5 += 0x100010; /* 16,16 */ r5 -= r0 * 5; /* d-5e+20f+20g-5h+i, b-5c+20d+20e-5f+g */ r5 >>= 5; r13 |= r5; /* check clipping */ r5 &= 0xFF00FF; /* mask */ r4 |= (r5 << 8); /* pack them together */ *p_tmp++ = r4; r1 = r3; r0 = r2; } p_tmp += ((24 - blkwidth) >> 2); /* move to the next line */ p_ref += ref_offset; /* ref_offset = inpitch-blkwidth; */ if (r13&0xFF000700) /* need clipping */ { /* move back to the beginning of the line */ p_ref -= (ref_offset + blkwidth); /* input */ p_tmp -= 6; /* intermediate output */ tmp = p_ref + blkwidth; while (p_ref < tmp) { r0 = *p_ref++; r1 = *p_ref++; r2 = *p_ref++; r3 = *p_ref++; r4 = *p_ref++; /* first pixel */ r5 = *p_ref++; result = (r0 + r5); r0 = (r1 + r4); result -= (r0 * 5);//result -= r0; result -= (r0<<2); r0 = (r2 + r3); result += (r0 * 20);//result += (r0<<4); result += (r0<<2); result = (result + 16) >> 5; CLIP_RESULT(result) pkres = result; /* second pixel */ r0 = *p_ref++; result = (r1 + r0); r1 = (r2 + r5); result -= (r1 * 5);//result -= r1; result -= (r1<<2); r1 = (r3 + r4); result += (r1 * 20);//result += (r1<<4); result += (r1<<2); result = (result + 16) >> 5; CLIP_RESULT(result) pkres |= (result << 8); /* third pixel */ r1 = *p_ref++; result = (r2 + r1); r2 = (r3 + r0); result -= (r2 * 5);//result -= r2; result -= (r2<<2); r2 = (r4 + r5); result += (r2 * 20);//result += (r2<<4); result += (r2<<2); result = (result + 16) >> 5; CLIP_RESULT(result) pkres |= (result << 16); /* fourth pixel */ r2 = *p_ref++; result = (r3 + r2); r3 = (r4 + r1); result -= (r3 * 5);//result -= r3; result -= (r3<<2); r3 = (r5 + r0); result += (r3 * 20);//result += (r3<<4); result += (r3<<2); result = (result + 16) >> 5; CLIP_RESULT(result) pkres |= (result << 24); *p_tmp++ = pkres; /* write 4 pixel */ p_ref -= 5; } p_tmp += ((24 - blkwidth) >> 2); /* move to the next line */ p_ref += ref_offset; /* ref_offset = inpitch-blkwidth; */ } } /* perform vertical interpolation */ /* not word-aligned */ if (((intptr_t)in2)&0x3) { eCreateAlign(in2, inpitch, -2, &tmp_in[0][0], blkwidth, blkheight + 5); in2 = &tmp_in[2][0]; inpitch = 24; } p_cur = out; curr_offset = 1 - outpitch * (blkheight - 1); /* offset vertically up and one pixel right */ pkres = blkheight * inpitch; /* reuse it for limit */ curr_offset += 3; for (j = 0; j < blkwidth; j += 4, in2 += 4) { r13 = 0; p_ref = in2; p_tmp8 = &(tmp_res[0][j]); /* intermediate result */ p_tmp8 -= 24; /* compensate for the first offset */ p_cur -= outpitch; /* compensate for the first offset */ tmp = p_ref + pkres; /* limit */ while (p_ref < tmp) /* the loop un-rolled */ { /* Read 1 byte at a time is too slow, too many read and pack ops, need to call CreateAlign */ /*p_ref8 = p_ref-(inpitch<<1); r0 = p_ref8[0]; r1 = p_ref8[2]; r0 |= (r1<<16); r6 = p_ref8[1]; r1 = p_ref8[3]; r6 |= (r1<<16); p_ref+=inpitch; */ r0 = *((uint32*)(p_ref - (inpitch << 1))); /* load 4 bytes */ p_ref += inpitch; r6 = (r0 >> 8) & 0xFF00FF; /* second and fourth byte */ r0 &= 0xFF00FF; /*p_ref8 = p_ref+(inpitch<<1); r1 = p_ref8[0]; r7 = p_ref8[2]; r1 |= (r7<<16); r7 = p_ref8[1]; r2 = p_ref8[3]; r7 |= (r2<<16);*/ r1 = *((uint32*)(p_ref + (inpitch << 1))); /* r1, r7, ref[3] */ r7 = (r1 >> 8) & 0xFF00FF; r1 &= 0xFF00FF; r0 += r1; r6 += r7; /*r2 = p_ref[0]; r8 = p_ref[2]; r2 |= (r8<<16); r8 = p_ref[1]; r1 = p_ref[3]; r8 |= (r1<<16);*/ r2 = *((uint32*)p_ref); /* r2, r8, ref[1] */ r8 = (r2 >> 8) & 0xFF00FF; r2 &= 0xFF00FF; /*p_ref8 = p_ref-inpitch; r1 = p_ref8[0]; r7 = p_ref8[2]; r1 |= (r7<<16); r1 += r2; r7 = p_ref8[1]; r2 = p_ref8[3]; r7 |= (r2<<16);*/ r1 = *((uint32*)(p_ref - inpitch)); /* r1, r7, ref[0] */ r7 = (r1 >> 8) & 0xFF00FF; r1 &= 0xFF00FF; r1 += r2; r7 += r8; r0 += 20 * r1; r6 += 20 * r7; r0 += 0x100010; r6 += 0x100010; /*p_ref8 = p_ref-(inpitch<<1); r2 = p_ref8[0]; r8 = p_ref8[2]; r2 |= (r8<<16); r8 = p_ref8[1]; r1 = p_ref8[3]; r8 |= (r1<<16);*/ r2 = *((uint32*)(p_ref - (inpitch << 1))); /* r2, r8, ref[-1] */ r8 = (r2 >> 8) & 0xFF00FF; r2 &= 0xFF00FF; /*p_ref8 = p_ref+inpitch; r1 = p_ref8[0]; r7 = p_ref8[2]; r1 |= (r7<<16); r1 += r2; r7 = p_ref8[1]; r2 = p_ref8[3]; r7 |= (r2<<16);*/ r1 = *((uint32*)(p_ref + inpitch)); /* r1, r7, ref[2] */ r7 = (r1 >> 8) & 0xFF00FF; r1 &= 0xFF00FF; r1 += r2; r7 += r8; r0 -= 5 * r1; r6 -= 5 * r7; r0 >>= 5; r6 >>= 5; /* clip */ r13 |= r6; r13 |= r0; //CLIPPACK(r6,result) /* add with horizontal results */ r10 = *((uint32*)(p_tmp8 += 24)); r9 = (r10 >> 8) & 0xFF00FF; r10 &= 0xFF00FF; r0 += r10; r0 += 0x10001; r0 = (r0 >> 1) & 0xFF00FF; /* mask to 8 bytes */ r6 += r9; r6 += 0x10001; r6 = (r6 >> 1) & 0xFF00FF; /* mask to 8 bytes */ r0 |= (r6 << 8); /* pack it back */ *((uint32*)(p_cur += outpitch)) = r0; } p_cur += curr_offset; /* offset to the next pixel */ if (r13 & 0xFF000700) /* this column need clipping */ { p_cur -= 4; for (i = 0; i < 4; i++) { p_ref = in2 + i; p_tmp8 = &(tmp_res[0][j+i]); /* intermediate result */ p_tmp8 -= 24; /* compensate for the first offset */ p_cur -= outpitch; /* compensate for the first offset */ tmp = p_ref + pkres; /* limit */ while (p_ref < tmp) /* the loop un-rolled */ { r0 = *(p_ref - (inpitch << 1)); r1 = *(p_ref - inpitch); r2 = *p_ref; r3 = *(p_ref += inpitch); /* modify pointer before loading */ r4 = *(p_ref += inpitch); /* first pixel */ r5 = *(p_ref += inpitch); result = (r0 + r5); r0 = (r1 + r4); result -= (r0 * 5);//result -= r0; result -= (r0<<2); r0 = (r2 + r3); result += (r0 * 20);//result += (r0<<4); result += (r0<<2); result = (result + 16) >> 5; CLIP_RESULT(result) tmp_result = *(p_tmp8 += 24); /* modify pointer before loading */ result = (result + tmp_result + 1); /* no clip */ result = (result >> 1); *(p_cur += outpitch) = result; /* second pixel */ r0 = *(p_ref += inpitch); result = (r1 + r0); r1 = (r2 + r5); result -= (r1 * 5);//result -= r1; result -= (r1<<2); r1 = (r3 + r4); result += (r1 * 20);//result += (r1<<4); result += (r1<<2); result = (result + 16) >> 5; CLIP_RESULT(result) tmp_result = *(p_tmp8 += 24); /* intermediate result */ result = (result + tmp_result + 1); /* no clip */ result = (result >> 1); *(p_cur += outpitch) = result; /* third pixel */ r1 = *(p_ref += inpitch); result = (r2 + r1); r2 = (r3 + r0); result -= (r2 * 5);//result -= r2; result -= (r2<<2); r2 = (r4 + r5); result += (r2 * 20);//result += (r2<<4); result += (r2<<2); result = (result + 16) >> 5; CLIP_RESULT(result) tmp_result = *(p_tmp8 += 24); /* intermediate result */ result = (result + tmp_result + 1); /* no clip */ result = (result >> 1); *(p_cur += outpitch) = result; /* fourth pixel */ r2 = *(p_ref += inpitch); result = (r3 + r2); r3 = (r4 + r1); result -= (r3 * 5);//result -= r3; result -= (r3<<2); r3 = (r5 + r0); result += (r3 * 20);//result += (r3<<4); result += (r3<<2); result = (result + 16) >> 5; CLIP_RESULT(result) tmp_result = *(p_tmp8 += 24); /* intermediate result */ result = (result + tmp_result + 1); /* no clip */ result = (result >> 1); *(p_cur += outpitch) = result; p_ref -= (inpitch << 1); /* move back to center of the filter of the next one */ } p_cur += (curr_offset - 3); } } } return ; } /* position G */ void eFullPelMC(uint8 *in, int inpitch, uint8 *out, int outpitch, int blkwidth, int blkheight) { int i, j; int offset_in = inpitch - blkwidth; int offset_out = outpitch - blkwidth; uint32 temp; uint8 byte; if (((intptr_t)in)&3) { for (j = blkheight; j > 0; j--) { for (i = blkwidth; i > 0; i -= 4) { temp = *in++; byte = *in++; temp |= (byte << 8); byte = *in++; temp |= (byte << 16); byte = *in++; temp |= (byte << 24); *((uint32*)out) = temp; /* write 4 bytes */ out += 4; } out += offset_out; in += offset_in; } } else { for (j = blkheight; j > 0; j--) { for (i = blkwidth; i > 0; i -= 4) { temp = *((uint32*)in); *((uint32*)out) = temp; in += 4; out += 4; } out += offset_out; in += offset_in; } } return ; } void ePadChroma(uint8 *ref, int picwidth, int picheight, int picpitch, int x_pos, int y_pos) { int pad_height; int pad_width; uint8 *start; uint32 word1, word2, word3; int offset, j; pad_height = 8 + ((y_pos & 7) ? 1 : 0); pad_width = 8 + ((x_pos & 7) ? 1 : 0); y_pos >>= 3; x_pos >>= 3; // pad vertical first if (y_pos < 0) // need to pad up { if (x_pos < -8) start = ref - 8; else if (x_pos + pad_width > picwidth + 7) start = ref + picwidth + 7 - pad_width; else start = ref + x_pos; /* word-align start */ offset = (intptr_t)start & 0x3; if (offset) start -= offset; word1 = *((uint32*)start); word2 = *((uint32*)(start + 4)); word3 = *((uint32*)(start + 8)); /* pad up N rows */ j = -y_pos; if (j > 8) j = 8; while (j--) { *((uint32*)(start -= picpitch)) = word1; *((uint32*)(start + 4)) = word2; *((uint32*)(start + 8)) = word3; } } else if (y_pos + pad_height >= picheight) /* pad down */ { if (x_pos < -8) start = ref + picpitch * (picheight - 1) - 8; else if (x_pos + pad_width > picwidth + 7) start = ref + picpitch * (picheight - 1) + picwidth + 7 - pad_width; else start = ref + picpitch * (picheight - 1) + x_pos; /* word-align start */ offset = (intptr_t)start & 0x3; if (offset) start -= offset; word1 = *((uint32*)start); word2 = *((uint32*)(start + 4)); word3 = *((uint32*)(start + 8)); /* pad down N rows */ j = y_pos + pad_height - picheight; if (j > 8) j = 8; while (j--) { *((uint32*)(start += picpitch)) = word1; *((uint32*)(start + 4)) = word2; *((uint32*)(start + 8)) = word3; } } /* now pad horizontal */ if (x_pos < 0) // pad left { if (y_pos < -8) start = ref - (picpitch << 3); else if (y_pos + pad_height > picheight + 7) start = ref + (picheight + 7 - pad_height) * picpitch; else start = ref + y_pos * picpitch; // now pad left 8 pixels for pad_height rows */ j = pad_height; start -= picpitch; while (j--) { word1 = *(start += picpitch); word1 |= (word1 << 8); word1 |= (word1 << 16); *((uint32*)(start - 8)) = word1; *((uint32*)(start - 4)) = word1; } } else if (x_pos + pad_width >= picwidth) /* pad right */ { if (y_pos < -8) start = ref - (picpitch << 3) + picwidth - 1; else if (y_pos + pad_height > picheight + 7) start = ref + (picheight + 7 - pad_height) * picpitch + picwidth - 1; else start = ref + y_pos * picpitch + picwidth - 1; // now pad right 8 pixels for pad_height rows */ j = pad_height; start -= picpitch; while (j--) { word1 = *(start += picpitch); word1 |= (word1 << 8); word1 |= (word1 << 16); *((uint32*)(start + 1)) = word1; *((uint32*)(start + 5)) = word1; } } return ; } void eChromaMotionComp(uint8 *ref, int picwidth, int picheight, int x_pos, int y_pos, uint8 *pred, int picpitch, int blkwidth, int blkheight) { int dx, dy; int offset_dx, offset_dy; int index; ePadChroma(ref, picwidth, picheight, picpitch, x_pos, y_pos); dx = x_pos & 7; dy = y_pos & 7; offset_dx = (dx + 7) >> 3; offset_dy = (dy + 7) >> 3; x_pos = x_pos >> 3; /* round it to full-pel resolution */ y_pos = y_pos >> 3; ref += y_pos * picpitch + x_pos; index = offset_dx + (offset_dy << 1) + ((blkwidth << 1) & 0x7); (*(eChromaMC_SIMD[index]))(ref, picpitch , dx, dy, pred, picpitch, blkwidth, blkheight); return ; } /* SIMD routines, unroll the loops in vertical direction, decreasing loops (things to be done) */ void eChromaDiagonalMC_SIMD(uint8 *pRef, int srcPitch, int dx, int dy, uint8 *pOut, int predPitch, int blkwidth, int blkheight) { int32 r0, r1, r2, r3, result0, result1; uint8 temp[288]; uint8 *ref, *out; int i, j; int dx_8 = 8 - dx; int dy_8 = 8 - dy; /* horizontal first */ out = temp; for (i = 0; i < blkheight + 1; i++) { ref = pRef; r0 = ref[0]; for (j = 0; j < blkwidth; j += 4) { r0 |= (ref[2] << 16); result0 = dx_8 * r0; r1 = ref[1] | (ref[3] << 16); result0 += dx * r1; *(int32 *)out = result0; result0 = dx_8 * r1; r2 = ref[4]; r0 = r0 >> 16; r1 = r0 | (r2 << 16); result0 += dx * r1; *(int32 *)(out + 16) = result0; ref += 4; out += 4; r0 = r2; } pRef += srcPitch; out += (32 - blkwidth); } // pRef -= srcPitch*(blkheight+1); ref = temp; for (j = 0; j < blkwidth; j += 4) { r0 = *(int32 *)ref; r1 = *(int32 *)(ref + 16); ref += 32; out = pOut; for (i = 0; i < (blkheight >> 1); i++) { result0 = dy_8 * r0 + 0x00200020; r2 = *(int32 *)ref; result0 += dy * r2; result0 >>= 6; result0 &= 0x00FF00FF; r0 = r2; result1 = dy_8 * r1 + 0x00200020; r3 = *(int32 *)(ref + 16); result1 += dy * r3; result1 >>= 6; result1 &= 0x00FF00FF; r1 = r3; *(int32 *)out = result0 | (result1 << 8); out += predPitch; ref += 32; result0 = dy_8 * r0 + 0x00200020; r2 = *(int32 *)ref; result0 += dy * r2; result0 >>= 6; result0 &= 0x00FF00FF; r0 = r2; result1 = dy_8 * r1 + 0x00200020; r3 = *(int32 *)(ref + 16); result1 += dy * r3; result1 >>= 6; result1 &= 0x00FF00FF; r1 = r3; *(int32 *)out = result0 | (result1 << 8); out += predPitch; ref += 32; } pOut += 4; ref = temp + 4; /* since it can only iterate twice max */ } return; } void eChromaHorizontalMC_SIMD(uint8 *pRef, int srcPitch, int dx, int dy, uint8 *pOut, int predPitch, int blkwidth, int blkheight) { (void)(dy); int32 r0, r1, r2, result0, result1; uint8 *ref, *out; int i, j; int dx_8 = 8 - dx; /* horizontal first */ for (i = 0; i < blkheight; i++) { ref = pRef; out = pOut; r0 = ref[0]; for (j = 0; j < blkwidth; j += 4) { r0 |= (ref[2] << 16); result0 = dx_8 * r0 + 0x00040004; r1 = ref[1] | (ref[3] << 16); result0 += dx * r1; result0 >>= 3; result0 &= 0x00FF00FF; result1 = dx_8 * r1 + 0x00040004; r2 = ref[4]; r0 = r0 >> 16; r1 = r0 | (r2 << 16); result1 += dx * r1; result1 >>= 3; result1 &= 0x00FF00FF; *(int32 *)out = result0 | (result1 << 8); ref += 4; out += 4; r0 = r2; } pRef += srcPitch; pOut += predPitch; } return; } void eChromaVerticalMC_SIMD(uint8 *pRef, int srcPitch, int dx, int dy, uint8 *pOut, int predPitch, int blkwidth, int blkheight) { (void)(dx); int32 r0, r1, r2, r3, result0, result1; int i, j; uint8 *ref, *out; int dy_8 = 8 - dy; /* vertical first */ for (i = 0; i < blkwidth; i += 4) { ref = pRef; out = pOut; r0 = ref[0] | (ref[2] << 16); r1 = ref[1] | (ref[3] << 16); ref += srcPitch; for (j = 0; j < blkheight; j++) { result0 = dy_8 * r0 + 0x00040004; r2 = ref[0] | (ref[2] << 16); result0 += dy * r2; result0 >>= 3; result0 &= 0x00FF00FF; r0 = r2; result1 = dy_8 * r1 + 0x00040004; r3 = ref[1] | (ref[3] << 16); result1 += dy * r3; result1 >>= 3; result1 &= 0x00FF00FF; r1 = r3; *(int32 *)out = result0 | (result1 << 8); ref += srcPitch; out += predPitch; } pOut += 4; pRef += 4; } return; } void eChromaDiagonalMC2_SIMD(uint8 *pRef, int srcPitch, int dx, int dy, uint8 *pOut, int predPitch, int blkwidth, int blkheight) { (void)(blkwidth); int32 r0, r1, temp0, temp1, result; int32 temp[9]; int32 *out; int i, r_temp; int dy_8 = 8 - dy; /* horizontal first */ out = temp; for (i = 0; i < blkheight + 1; i++) { r_temp = pRef[1]; temp0 = (pRef[0] << 3) + dx * (r_temp - pRef[0]); temp1 = (r_temp << 3) + dx * (pRef[2] - r_temp); r0 = temp0 | (temp1 << 16); *out++ = r0; pRef += srcPitch; } pRef -= srcPitch * (blkheight + 1); out = temp; r0 = *out++; for (i = 0; i < blkheight; i++) { result = dy_8 * r0 + 0x00200020; r1 = *out++; result += dy * r1; result >>= 6; result &= 0x00FF00FF; *(int16 *)pOut = (result >> 8) | (result & 0xFF); r0 = r1; pOut += predPitch; } return; } void eChromaHorizontalMC2_SIMD(uint8 *pRef, int srcPitch, int dx, int dy, uint8 *pOut, int predPitch, int blkwidth, int blkheight) { (void)(dy); (void)(blkwidth); int i, temp, temp0, temp1; /* horizontal first */ for (i = 0; i < blkheight; i++) { temp = pRef[1]; temp0 = ((pRef[0] << 3) + dx * (temp - pRef[0]) + 4) >> 3; temp1 = ((temp << 3) + dx * (pRef[2] - temp) + 4) >> 3; *(int16 *)pOut = temp0 | (temp1 << 8); pRef += srcPitch; pOut += predPitch; } return; } void eChromaVerticalMC2_SIMD(uint8 *pRef, int srcPitch, int dx, int dy, uint8 *pOut, int predPitch, int blkwidth, int blkheight) { (void)(dx); (void)(blkwidth); int32 r0, r1, result; int i; int dy_8 = 8 - dy; r0 = pRef[0] | (pRef[1] << 16); pRef += srcPitch; for (i = 0; i < blkheight; i++) { result = dy_8 * r0 + 0x00040004; r1 = pRef[0] | (pRef[1] << 16); result += dy * r1; result >>= 3; result &= 0x00FF00FF; *(int16 *)pOut = (result >> 8) | (result & 0xFF); r0 = r1; pRef += srcPitch; pOut += predPitch; } return; } void eChromaFullMC_SIMD(uint8 *pRef, int srcPitch, int dx, int dy, uint8 *pOut, int predPitch, int blkwidth, int blkheight) { (void)(dx); (void)(dy); int i, j; int offset_in = srcPitch - blkwidth; int offset_out = predPitch - blkwidth; uint16 temp; uint8 byte; if (((intptr_t)pRef)&1) { for (j = blkheight; j > 0; j--) { for (i = blkwidth; i > 0; i -= 2) { temp = *pRef++; byte = *pRef++; temp |= (byte << 8); *((uint16*)pOut) = temp; /* write 2 bytes */ pOut += 2; } pOut += offset_out; pRef += offset_in; } } else { for (j = blkheight; j > 0; j--) { for (i = blkwidth; i > 0; i -= 2) { temp = *((uint16*)pRef); *((uint16*)pOut) = temp; pRef += 2; pOut += 2; } pOut += offset_out; pRef += offset_in; } } return ; }