yading@10: /* yading@10: * Copyright (C) 2003 David S. Miller yading@10: * yading@10: * This file is part of FFmpeg. yading@10: * yading@10: * FFmpeg is free software; you can redistribute it and/or yading@10: * modify it under the terms of the GNU Lesser General Public yading@10: * License as published by the Free Software Foundation; either yading@10: * version 2.1 of the License, or (at your option) any later version. yading@10: * yading@10: * FFmpeg is distributed in the hope that it will be useful, yading@10: * but WITHOUT ANY WARRANTY; without even the implied warranty of yading@10: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU yading@10: * Lesser General Public License for more details. yading@10: * yading@10: * You should have received a copy of the GNU Lesser General Public yading@10: * License along with FFmpeg; if not, write to the Free Software yading@10: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA yading@10: */ yading@10: yading@10: /* The *no_round* functions have been added by James A. Morrison, 2003,2004. yading@10: The vis code from libmpeg2 was adapted for libavcodec by James A. Morrison. yading@10: */ yading@10: yading@10: #include yading@10: #include yading@10: yading@10: #include "libavutil/attributes.h" yading@10: #include "libavutil/mem.h" yading@10: #include "libavcodec/hpeldsp.h" yading@10: #include "vis.h" yading@10: yading@10: /* The trick used in some of this file is the formula from the MMX yading@10: * motion comp code, which is: yading@10: * yading@10: * (x+y+1)>>1 == (x|y)-((x^y)>>1) yading@10: * yading@10: * This allows us to average 8 bytes at a time in a 64-bit FPU reg. yading@10: * We avoid overflows by masking before we do the shift, and we yading@10: * implement the shift by multiplying by 1/2 using mul8x16. So in yading@10: * VIS this is (assume 'x' is in f0, 'y' is in f2, a repeating mask yading@10: * of '0xfe' is in f4, a repeating mask of '0x7f' is in f6, and yading@10: * the value 0x80808080 is in f8): yading@10: * yading@10: * fxor f0, f2, f10 yading@10: * fand f10, f4, f10 yading@10: * fmul8x16 f8, f10, f10 yading@10: * fand f10, f6, f10 yading@10: * for f0, f2, f12 yading@10: * fpsub16 f12, f10, f10 yading@10: */ yading@10: yading@10: #define DUP4(x) {x, x, x, x} yading@10: #define DUP8(x) {x, x, x, x, x, x, x, x} yading@10: DECLARE_ALIGNED(8, static const int16_t, constants1)[] = DUP4 (1); yading@10: DECLARE_ALIGNED(8, static const int16_t, constants2)[] = DUP4 (2); yading@10: DECLARE_ALIGNED(8, static const int16_t, constants3)[] = DUP4 (3); yading@10: DECLARE_ALIGNED(8, static const int16_t, constants6)[] = DUP4 (6); yading@10: DECLARE_ALIGNED(8, static const int8_t, constants_fe)[] = DUP8 (0xfe); yading@10: DECLARE_ALIGNED(8, static const int8_t, constants_7f)[] = DUP8 (0x7f); yading@10: DECLARE_ALIGNED(8, static const int8_t, constants128)[] = DUP8 (128); yading@10: DECLARE_ALIGNED(8, static const int16_t, constants256_512)[] = yading@10: {256, 512, 256, 512}; yading@10: DECLARE_ALIGNED(8, static const int16_t, constants256_1024)[] = yading@10: {256, 1024, 256, 1024}; yading@10: yading@10: #define REF_0 0 yading@10: #define REF_0_1 1 yading@10: #define REF_2 2 yading@10: #define REF_2_1 3 yading@10: #define REF_4 4 yading@10: #define REF_4_1 5 yading@10: #define REF_6 6 yading@10: #define REF_6_1 7 yading@10: #define REF_S0 8 yading@10: #define REF_S0_1 9 yading@10: #define REF_S2 10 yading@10: #define REF_S2_1 11 yading@10: #define REF_S4 12 yading@10: #define REF_S4_1 13 yading@10: #define REF_S6 14 yading@10: #define REF_S6_1 15 yading@10: #define DST_0 16 yading@10: #define DST_1 17 yading@10: #define DST_2 18 yading@10: #define DST_3 19 yading@10: #define CONST_1 20 yading@10: #define CONST_2 20 yading@10: #define CONST_3 20 yading@10: #define CONST_6 20 yading@10: #define MASK_fe 20 yading@10: #define CONST_128 22 yading@10: #define CONST_256 22 yading@10: #define CONST_512 22 yading@10: #define CONST_1024 22 yading@10: #define TMP0 24 yading@10: #define TMP1 25 yading@10: #define TMP2 26 yading@10: #define TMP3 27 yading@10: #define TMP4 28 yading@10: #define TMP5 29 yading@10: #define ZERO 30 yading@10: #define MASK_7f 30 yading@10: yading@10: #define TMP6 32 yading@10: #define TMP8 34 yading@10: #define TMP10 36 yading@10: #define TMP12 38 yading@10: #define TMP14 40 yading@10: #define TMP16 42 yading@10: #define TMP18 44 yading@10: #define TMP20 46 yading@10: #define TMP22 48 yading@10: #define TMP24 50 yading@10: #define TMP26 52 yading@10: #define TMP28 54 yading@10: #define TMP30 56 yading@10: #define TMP32 58 yading@10: yading@10: static void MC_put_o_16_vis (uint8_t * dest, const uint8_t * ref, yading@10: const ptrdiff_t stride, int height) yading@10: { yading@10: ref = vis_alignaddr(ref); yading@10: do { /* 5 cycles */ yading@10: vis_ld64(ref[0], TMP0); yading@10: yading@10: vis_ld64_2(ref, 8, TMP2); yading@10: yading@10: vis_ld64_2(ref, 16, TMP4); yading@10: ref += stride; yading@10: yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: vis_st64(REF_0, dest[0]); yading@10: yading@10: vis_faligndata(TMP2, TMP4, REF_2); yading@10: vis_st64_2(REF_2, dest, 8); yading@10: dest += stride; yading@10: } while (--height); yading@10: } yading@10: yading@10: static void MC_put_o_8_vis (uint8_t * dest, const uint8_t * ref, yading@10: const ptrdiff_t stride, int height) yading@10: { yading@10: ref = vis_alignaddr(ref); yading@10: do { /* 4 cycles */ yading@10: vis_ld64(ref[0], TMP0); yading@10: yading@10: vis_ld64(ref[8], TMP2); yading@10: ref += stride; yading@10: yading@10: /* stall */ yading@10: yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: vis_st64(REF_0, dest[0]); yading@10: dest += stride; yading@10: } while (--height); yading@10: } yading@10: yading@10: yading@10: static void MC_avg_o_16_vis (uint8_t * dest, const uint8_t * ref, yading@10: const ptrdiff_t stride, int height) yading@10: { yading@10: int stride_8 = stride + 8; yading@10: yading@10: ref = vis_alignaddr(ref); yading@10: yading@10: vis_ld64(ref[0], TMP0); yading@10: yading@10: vis_ld64(ref[8], TMP2); yading@10: yading@10: vis_ld64(ref[16], TMP4); yading@10: yading@10: vis_ld64(dest[0], DST_0); yading@10: yading@10: vis_ld64(dest[8], DST_2); yading@10: yading@10: vis_ld64(constants_fe[0], MASK_fe); yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_ld64(constants_7f[0], MASK_7f); yading@10: vis_faligndata(TMP2, TMP4, REF_2); yading@10: yading@10: vis_ld64(constants128[0], CONST_128); yading@10: yading@10: ref += stride; yading@10: height = (height >> 1) - 1; yading@10: yading@10: do { /* 24 cycles */ yading@10: vis_ld64(ref[0], TMP0); yading@10: vis_xor(DST_0, REF_0, TMP6); yading@10: yading@10: vis_ld64_2(ref, 8, TMP2); yading@10: vis_and(TMP6, MASK_fe, TMP6); yading@10: yading@10: vis_ld64_2(ref, 16, TMP4); yading@10: ref += stride; yading@10: vis_mul8x16(CONST_128, TMP6, TMP6); yading@10: vis_xor(DST_2, REF_2, TMP8); yading@10: yading@10: vis_and(TMP8, MASK_fe, TMP8); yading@10: yading@10: vis_or(DST_0, REF_0, TMP10); yading@10: vis_ld64_2(dest, stride, DST_0); yading@10: vis_mul8x16(CONST_128, TMP8, TMP8); yading@10: yading@10: vis_or(DST_2, REF_2, TMP12); yading@10: vis_ld64_2(dest, stride_8, DST_2); yading@10: yading@10: vis_ld64(ref[0], TMP14); yading@10: vis_and(TMP6, MASK_7f, TMP6); yading@10: yading@10: vis_and(TMP8, MASK_7f, TMP8); yading@10: yading@10: vis_psub16(TMP10, TMP6, TMP6); yading@10: vis_st64(TMP6, dest[0]); yading@10: yading@10: vis_psub16(TMP12, TMP8, TMP8); yading@10: vis_st64_2(TMP8, dest, 8); yading@10: yading@10: dest += stride; yading@10: vis_ld64_2(ref, 8, TMP16); yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_ld64_2(ref, 16, TMP18); yading@10: vis_faligndata(TMP2, TMP4, REF_2); yading@10: ref += stride; yading@10: yading@10: vis_xor(DST_0, REF_0, TMP20); yading@10: yading@10: vis_and(TMP20, MASK_fe, TMP20); yading@10: yading@10: vis_xor(DST_2, REF_2, TMP22); yading@10: vis_mul8x16(CONST_128, TMP20, TMP20); yading@10: yading@10: vis_and(TMP22, MASK_fe, TMP22); yading@10: yading@10: vis_or(DST_0, REF_0, TMP24); yading@10: vis_mul8x16(CONST_128, TMP22, TMP22); yading@10: yading@10: vis_or(DST_2, REF_2, TMP26); yading@10: yading@10: vis_ld64_2(dest, stride, DST_0); yading@10: vis_faligndata(TMP14, TMP16, REF_0); yading@10: yading@10: vis_ld64_2(dest, stride_8, DST_2); yading@10: vis_faligndata(TMP16, TMP18, REF_2); yading@10: yading@10: vis_and(TMP20, MASK_7f, TMP20); yading@10: yading@10: vis_and(TMP22, MASK_7f, TMP22); yading@10: yading@10: vis_psub16(TMP24, TMP20, TMP20); yading@10: vis_st64(TMP20, dest[0]); yading@10: yading@10: vis_psub16(TMP26, TMP22, TMP22); yading@10: vis_st64_2(TMP22, dest, 8); yading@10: dest += stride; yading@10: } while (--height); yading@10: yading@10: vis_ld64(ref[0], TMP0); yading@10: vis_xor(DST_0, REF_0, TMP6); yading@10: yading@10: vis_ld64_2(ref, 8, TMP2); yading@10: vis_and(TMP6, MASK_fe, TMP6); yading@10: yading@10: vis_ld64_2(ref, 16, TMP4); yading@10: vis_mul8x16(CONST_128, TMP6, TMP6); yading@10: vis_xor(DST_2, REF_2, TMP8); yading@10: yading@10: vis_and(TMP8, MASK_fe, TMP8); yading@10: yading@10: vis_or(DST_0, REF_0, TMP10); yading@10: vis_ld64_2(dest, stride, DST_0); yading@10: vis_mul8x16(CONST_128, TMP8, TMP8); yading@10: yading@10: vis_or(DST_2, REF_2, TMP12); yading@10: vis_ld64_2(dest, stride_8, DST_2); yading@10: yading@10: vis_ld64(ref[0], TMP14); yading@10: vis_and(TMP6, MASK_7f, TMP6); yading@10: yading@10: vis_and(TMP8, MASK_7f, TMP8); yading@10: yading@10: vis_psub16(TMP10, TMP6, TMP6); yading@10: vis_st64(TMP6, dest[0]); yading@10: yading@10: vis_psub16(TMP12, TMP8, TMP8); yading@10: vis_st64_2(TMP8, dest, 8); yading@10: yading@10: dest += stride; yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_faligndata(TMP2, TMP4, REF_2); yading@10: yading@10: vis_xor(DST_0, REF_0, TMP20); yading@10: yading@10: vis_and(TMP20, MASK_fe, TMP20); yading@10: yading@10: vis_xor(DST_2, REF_2, TMP22); yading@10: vis_mul8x16(CONST_128, TMP20, TMP20); yading@10: yading@10: vis_and(TMP22, MASK_fe, TMP22); yading@10: yading@10: vis_or(DST_0, REF_0, TMP24); yading@10: vis_mul8x16(CONST_128, TMP22, TMP22); yading@10: yading@10: vis_or(DST_2, REF_2, TMP26); yading@10: yading@10: vis_and(TMP20, MASK_7f, TMP20); yading@10: yading@10: vis_and(TMP22, MASK_7f, TMP22); yading@10: yading@10: vis_psub16(TMP24, TMP20, TMP20); yading@10: vis_st64(TMP20, dest[0]); yading@10: yading@10: vis_psub16(TMP26, TMP22, TMP22); yading@10: vis_st64_2(TMP22, dest, 8); yading@10: } yading@10: yading@10: static void MC_avg_o_8_vis (uint8_t * dest, const uint8_t * ref, yading@10: const ptrdiff_t stride, int height) yading@10: { yading@10: ref = vis_alignaddr(ref); yading@10: yading@10: vis_ld64(ref[0], TMP0); yading@10: yading@10: vis_ld64(ref[8], TMP2); yading@10: yading@10: vis_ld64(dest[0], DST_0); yading@10: yading@10: vis_ld64(constants_fe[0], MASK_fe); yading@10: yading@10: vis_ld64(constants_7f[0], MASK_7f); yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_ld64(constants128[0], CONST_128); yading@10: yading@10: ref += stride; yading@10: height = (height >> 1) - 1; yading@10: yading@10: do { /* 12 cycles */ yading@10: vis_ld64(ref[0], TMP0); yading@10: vis_xor(DST_0, REF_0, TMP4); yading@10: yading@10: vis_ld64(ref[8], TMP2); yading@10: vis_and(TMP4, MASK_fe, TMP4); yading@10: yading@10: vis_or(DST_0, REF_0, TMP6); yading@10: vis_ld64_2(dest, stride, DST_0); yading@10: ref += stride; yading@10: vis_mul8x16(CONST_128, TMP4, TMP4); yading@10: yading@10: vis_ld64(ref[0], TMP12); yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_ld64(ref[8], TMP2); yading@10: vis_xor(DST_0, REF_0, TMP0); yading@10: ref += stride; yading@10: yading@10: vis_and(TMP0, MASK_fe, TMP0); yading@10: yading@10: vis_and(TMP4, MASK_7f, TMP4); yading@10: yading@10: vis_psub16(TMP6, TMP4, TMP4); yading@10: vis_st64(TMP4, dest[0]); yading@10: dest += stride; yading@10: vis_mul8x16(CONST_128, TMP0, TMP0); yading@10: yading@10: vis_or(DST_0, REF_0, TMP6); yading@10: vis_ld64_2(dest, stride, DST_0); yading@10: yading@10: vis_faligndata(TMP12, TMP2, REF_0); yading@10: yading@10: vis_and(TMP0, MASK_7f, TMP0); yading@10: yading@10: vis_psub16(TMP6, TMP0, TMP4); yading@10: vis_st64(TMP4, dest[0]); yading@10: dest += stride; yading@10: } while (--height); yading@10: yading@10: vis_ld64(ref[0], TMP0); yading@10: vis_xor(DST_0, REF_0, TMP4); yading@10: yading@10: vis_ld64(ref[8], TMP2); yading@10: vis_and(TMP4, MASK_fe, TMP4); yading@10: yading@10: vis_or(DST_0, REF_0, TMP6); yading@10: vis_ld64_2(dest, stride, DST_0); yading@10: vis_mul8x16(CONST_128, TMP4, TMP4); yading@10: yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_xor(DST_0, REF_0, TMP0); yading@10: yading@10: vis_and(TMP0, MASK_fe, TMP0); yading@10: yading@10: vis_and(TMP4, MASK_7f, TMP4); yading@10: yading@10: vis_psub16(TMP6, TMP4, TMP4); yading@10: vis_st64(TMP4, dest[0]); yading@10: dest += stride; yading@10: vis_mul8x16(CONST_128, TMP0, TMP0); yading@10: yading@10: vis_or(DST_0, REF_0, TMP6); yading@10: yading@10: vis_and(TMP0, MASK_7f, TMP0); yading@10: yading@10: vis_psub16(TMP6, TMP0, TMP4); yading@10: vis_st64(TMP4, dest[0]); yading@10: } yading@10: yading@10: static void MC_put_x_16_vis (uint8_t * dest, const uint8_t * ref, yading@10: const ptrdiff_t stride, int height) yading@10: { yading@10: unsigned long off = (unsigned long) ref & 0x7; yading@10: unsigned long off_plus_1 = off + 1; yading@10: yading@10: ref = vis_alignaddr(ref); yading@10: yading@10: vis_ld64(ref[0], TMP0); yading@10: yading@10: vis_ld64_2(ref, 8, TMP2); yading@10: yading@10: vis_ld64_2(ref, 16, TMP4); yading@10: yading@10: vis_ld64(constants_fe[0], MASK_fe); yading@10: yading@10: vis_ld64(constants_7f[0], MASK_7f); yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_ld64(constants128[0], CONST_128); yading@10: vis_faligndata(TMP2, TMP4, REF_4); yading@10: yading@10: if (off != 0x7) { yading@10: vis_alignaddr_g0((void *)off_plus_1); yading@10: vis_faligndata(TMP0, TMP2, REF_2); yading@10: vis_faligndata(TMP2, TMP4, REF_6); yading@10: } else { yading@10: vis_src1(TMP2, REF_2); yading@10: vis_src1(TMP4, REF_6); yading@10: } yading@10: yading@10: ref += stride; yading@10: height = (height >> 1) - 1; yading@10: yading@10: do { /* 34 cycles */ yading@10: vis_ld64(ref[0], TMP0); yading@10: vis_xor(REF_0, REF_2, TMP6); yading@10: yading@10: vis_ld64_2(ref, 8, TMP2); yading@10: vis_xor(REF_4, REF_6, TMP8); yading@10: yading@10: vis_ld64_2(ref, 16, TMP4); yading@10: vis_and(TMP6, MASK_fe, TMP6); yading@10: ref += stride; yading@10: yading@10: vis_ld64(ref[0], TMP14); yading@10: vis_mul8x16(CONST_128, TMP6, TMP6); yading@10: vis_and(TMP8, MASK_fe, TMP8); yading@10: yading@10: vis_ld64_2(ref, 8, TMP16); yading@10: vis_mul8x16(CONST_128, TMP8, TMP8); yading@10: vis_or(REF_0, REF_2, TMP10); yading@10: yading@10: vis_ld64_2(ref, 16, TMP18); yading@10: ref += stride; yading@10: vis_or(REF_4, REF_6, TMP12); yading@10: yading@10: vis_alignaddr_g0((void *)off); yading@10: yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_faligndata(TMP2, TMP4, REF_4); yading@10: yading@10: if (off != 0x7) { yading@10: vis_alignaddr_g0((void *)off_plus_1); yading@10: vis_faligndata(TMP0, TMP2, REF_2); yading@10: vis_faligndata(TMP2, TMP4, REF_6); yading@10: } else { yading@10: vis_src1(TMP2, REF_2); yading@10: vis_src1(TMP4, REF_6); yading@10: } yading@10: yading@10: vis_and(TMP6, MASK_7f, TMP6); yading@10: yading@10: vis_and(TMP8, MASK_7f, TMP8); yading@10: yading@10: vis_psub16(TMP10, TMP6, TMP6); yading@10: vis_st64(TMP6, dest[0]); yading@10: yading@10: vis_psub16(TMP12, TMP8, TMP8); yading@10: vis_st64_2(TMP8, dest, 8); yading@10: dest += stride; yading@10: yading@10: vis_xor(REF_0, REF_2, TMP6); yading@10: yading@10: vis_xor(REF_4, REF_6, TMP8); yading@10: yading@10: vis_and(TMP6, MASK_fe, TMP6); yading@10: yading@10: vis_mul8x16(CONST_128, TMP6, TMP6); yading@10: vis_and(TMP8, MASK_fe, TMP8); yading@10: yading@10: vis_mul8x16(CONST_128, TMP8, TMP8); yading@10: vis_or(REF_0, REF_2, TMP10); yading@10: yading@10: vis_or(REF_4, REF_6, TMP12); yading@10: yading@10: vis_alignaddr_g0((void *)off); yading@10: yading@10: vis_faligndata(TMP14, TMP16, REF_0); yading@10: yading@10: vis_faligndata(TMP16, TMP18, REF_4); yading@10: yading@10: if (off != 0x7) { yading@10: vis_alignaddr_g0((void *)off_plus_1); yading@10: vis_faligndata(TMP14, TMP16, REF_2); yading@10: vis_faligndata(TMP16, TMP18, REF_6); yading@10: } else { yading@10: vis_src1(TMP16, REF_2); yading@10: vis_src1(TMP18, REF_6); yading@10: } yading@10: yading@10: vis_and(TMP6, MASK_7f, TMP6); yading@10: yading@10: vis_and(TMP8, MASK_7f, TMP8); yading@10: yading@10: vis_psub16(TMP10, TMP6, TMP6); yading@10: vis_st64(TMP6, dest[0]); yading@10: yading@10: vis_psub16(TMP12, TMP8, TMP8); yading@10: vis_st64_2(TMP8, dest, 8); yading@10: dest += stride; yading@10: } while (--height); yading@10: yading@10: vis_ld64(ref[0], TMP0); yading@10: vis_xor(REF_0, REF_2, TMP6); yading@10: yading@10: vis_ld64_2(ref, 8, TMP2); yading@10: vis_xor(REF_4, REF_6, TMP8); yading@10: yading@10: vis_ld64_2(ref, 16, TMP4); yading@10: vis_and(TMP6, MASK_fe, TMP6); yading@10: yading@10: vis_mul8x16(CONST_128, TMP6, TMP6); yading@10: vis_and(TMP8, MASK_fe, TMP8); yading@10: yading@10: vis_mul8x16(CONST_128, TMP8, TMP8); yading@10: vis_or(REF_0, REF_2, TMP10); yading@10: yading@10: vis_or(REF_4, REF_6, TMP12); yading@10: yading@10: vis_alignaddr_g0((void *)off); yading@10: yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_faligndata(TMP2, TMP4, REF_4); yading@10: yading@10: if (off != 0x7) { yading@10: vis_alignaddr_g0((void *)off_plus_1); yading@10: vis_faligndata(TMP0, TMP2, REF_2); yading@10: vis_faligndata(TMP2, TMP4, REF_6); yading@10: } else { yading@10: vis_src1(TMP2, REF_2); yading@10: vis_src1(TMP4, REF_6); yading@10: } yading@10: yading@10: vis_and(TMP6, MASK_7f, TMP6); yading@10: yading@10: vis_and(TMP8, MASK_7f, TMP8); yading@10: yading@10: vis_psub16(TMP10, TMP6, TMP6); yading@10: vis_st64(TMP6, dest[0]); yading@10: yading@10: vis_psub16(TMP12, TMP8, TMP8); yading@10: vis_st64_2(TMP8, dest, 8); yading@10: dest += stride; yading@10: yading@10: vis_xor(REF_0, REF_2, TMP6); yading@10: yading@10: vis_xor(REF_4, REF_6, TMP8); yading@10: yading@10: vis_and(TMP6, MASK_fe, TMP6); yading@10: yading@10: vis_mul8x16(CONST_128, TMP6, TMP6); yading@10: vis_and(TMP8, MASK_fe, TMP8); yading@10: yading@10: vis_mul8x16(CONST_128, TMP8, TMP8); yading@10: vis_or(REF_0, REF_2, TMP10); yading@10: yading@10: vis_or(REF_4, REF_6, TMP12); yading@10: yading@10: vis_and(TMP6, MASK_7f, TMP6); yading@10: yading@10: vis_and(TMP8, MASK_7f, TMP8); yading@10: yading@10: vis_psub16(TMP10, TMP6, TMP6); yading@10: vis_st64(TMP6, dest[0]); yading@10: yading@10: vis_psub16(TMP12, TMP8, TMP8); yading@10: vis_st64_2(TMP8, dest, 8); yading@10: } yading@10: yading@10: static void MC_put_x_8_vis (uint8_t * dest, const uint8_t * ref, yading@10: const ptrdiff_t stride, int height) yading@10: { yading@10: unsigned long off = (unsigned long) ref & 0x7; yading@10: unsigned long off_plus_1 = off + 1; yading@10: yading@10: ref = vis_alignaddr(ref); yading@10: yading@10: vis_ld64(ref[0], TMP0); yading@10: yading@10: vis_ld64(ref[8], TMP2); yading@10: yading@10: vis_ld64(constants_fe[0], MASK_fe); yading@10: yading@10: vis_ld64(constants_7f[0], MASK_7f); yading@10: yading@10: vis_ld64(constants128[0], CONST_128); yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: if (off != 0x7) { yading@10: vis_alignaddr_g0((void *)off_plus_1); yading@10: vis_faligndata(TMP0, TMP2, REF_2); yading@10: } else { yading@10: vis_src1(TMP2, REF_2); yading@10: } yading@10: yading@10: ref += stride; yading@10: height = (height >> 1) - 1; yading@10: yading@10: do { /* 20 cycles */ yading@10: vis_ld64(ref[0], TMP0); yading@10: vis_xor(REF_0, REF_2, TMP4); yading@10: yading@10: vis_ld64_2(ref, 8, TMP2); yading@10: vis_and(TMP4, MASK_fe, TMP4); yading@10: ref += stride; yading@10: yading@10: vis_ld64(ref[0], TMP8); yading@10: vis_or(REF_0, REF_2, TMP6); yading@10: vis_mul8x16(CONST_128, TMP4, TMP4); yading@10: yading@10: vis_alignaddr_g0((void *)off); yading@10: yading@10: vis_ld64_2(ref, 8, TMP10); yading@10: ref += stride; yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: if (off != 0x7) { yading@10: vis_alignaddr_g0((void *)off_plus_1); yading@10: vis_faligndata(TMP0, TMP2, REF_2); yading@10: } else { yading@10: vis_src1(TMP2, REF_2); yading@10: } yading@10: yading@10: vis_and(TMP4, MASK_7f, TMP4); yading@10: yading@10: vis_psub16(TMP6, TMP4, DST_0); yading@10: vis_st64(DST_0, dest[0]); yading@10: dest += stride; yading@10: yading@10: vis_xor(REF_0, REF_2, TMP12); yading@10: yading@10: vis_and(TMP12, MASK_fe, TMP12); yading@10: yading@10: vis_or(REF_0, REF_2, TMP14); yading@10: vis_mul8x16(CONST_128, TMP12, TMP12); yading@10: yading@10: vis_alignaddr_g0((void *)off); yading@10: vis_faligndata(TMP8, TMP10, REF_0); yading@10: if (off != 0x7) { yading@10: vis_alignaddr_g0((void *)off_plus_1); yading@10: vis_faligndata(TMP8, TMP10, REF_2); yading@10: } else { yading@10: vis_src1(TMP10, REF_2); yading@10: } yading@10: yading@10: vis_and(TMP12, MASK_7f, TMP12); yading@10: yading@10: vis_psub16(TMP14, TMP12, DST_0); yading@10: vis_st64(DST_0, dest[0]); yading@10: dest += stride; yading@10: } while (--height); yading@10: yading@10: vis_ld64(ref[0], TMP0); yading@10: vis_xor(REF_0, REF_2, TMP4); yading@10: yading@10: vis_ld64_2(ref, 8, TMP2); yading@10: vis_and(TMP4, MASK_fe, TMP4); yading@10: yading@10: vis_or(REF_0, REF_2, TMP6); yading@10: vis_mul8x16(CONST_128, TMP4, TMP4); yading@10: yading@10: vis_alignaddr_g0((void *)off); yading@10: yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: if (off != 0x7) { yading@10: vis_alignaddr_g0((void *)off_plus_1); yading@10: vis_faligndata(TMP0, TMP2, REF_2); yading@10: } else { yading@10: vis_src1(TMP2, REF_2); yading@10: } yading@10: yading@10: vis_and(TMP4, MASK_7f, TMP4); yading@10: yading@10: vis_psub16(TMP6, TMP4, DST_0); yading@10: vis_st64(DST_0, dest[0]); yading@10: dest += stride; yading@10: yading@10: vis_xor(REF_0, REF_2, TMP12); yading@10: yading@10: vis_and(TMP12, MASK_fe, TMP12); yading@10: yading@10: vis_or(REF_0, REF_2, TMP14); yading@10: vis_mul8x16(CONST_128, TMP12, TMP12); yading@10: yading@10: vis_and(TMP12, MASK_7f, TMP12); yading@10: yading@10: vis_psub16(TMP14, TMP12, DST_0); yading@10: vis_st64(DST_0, dest[0]); yading@10: dest += stride; yading@10: } yading@10: yading@10: static void MC_avg_x_16_vis (uint8_t * dest, const uint8_t * ref, yading@10: const ptrdiff_t stride, int height) yading@10: { yading@10: unsigned long off = (unsigned long) ref & 0x7; yading@10: unsigned long off_plus_1 = off + 1; yading@10: yading@10: vis_set_gsr(5 << VIS_GSR_SCALEFACT_SHIFT); yading@10: yading@10: vis_ld64(constants3[0], CONST_3); yading@10: vis_fzero(ZERO); yading@10: vis_ld64(constants256_512[0], CONST_256); yading@10: yading@10: ref = vis_alignaddr(ref); yading@10: do { /* 26 cycles */ yading@10: vis_ld64(ref[0], TMP0); yading@10: yading@10: vis_ld64(ref[8], TMP2); yading@10: yading@10: vis_alignaddr_g0((void *)off); yading@10: yading@10: vis_ld64(ref[16], TMP4); yading@10: yading@10: vis_ld64(dest[0], DST_0); yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_ld64(dest[8], DST_2); yading@10: vis_faligndata(TMP2, TMP4, REF_4); yading@10: yading@10: if (off != 0x7) { yading@10: vis_alignaddr_g0((void *)off_plus_1); yading@10: vis_faligndata(TMP0, TMP2, REF_2); yading@10: vis_faligndata(TMP2, TMP4, REF_6); yading@10: } else { yading@10: vis_src1(TMP2, REF_2); yading@10: vis_src1(TMP4, REF_6); yading@10: } yading@10: yading@10: vis_mul8x16au(REF_0, CONST_256, TMP0); yading@10: yading@10: vis_pmerge(ZERO, REF_2, TMP4); yading@10: vis_mul8x16au(REF_0_1, CONST_256, TMP2); yading@10: yading@10: vis_pmerge(ZERO, REF_2_1, TMP6); yading@10: yading@10: vis_padd16(TMP0, TMP4, TMP0); yading@10: yading@10: vis_mul8x16al(DST_0, CONST_512, TMP4); yading@10: vis_padd16(TMP2, TMP6, TMP2); yading@10: yading@10: vis_mul8x16al(DST_1, CONST_512, TMP6); yading@10: yading@10: vis_mul8x16au(REF_6, CONST_256, TMP12); yading@10: yading@10: vis_padd16(TMP0, TMP4, TMP0); yading@10: vis_mul8x16au(REF_6_1, CONST_256, TMP14); yading@10: yading@10: vis_padd16(TMP2, TMP6, TMP2); yading@10: vis_mul8x16au(REF_4, CONST_256, TMP16); yading@10: yading@10: vis_padd16(TMP0, CONST_3, TMP8); yading@10: vis_mul8x16au(REF_4_1, CONST_256, TMP18); yading@10: yading@10: vis_padd16(TMP2, CONST_3, TMP10); yading@10: vis_pack16(TMP8, DST_0); yading@10: yading@10: vis_pack16(TMP10, DST_1); yading@10: vis_padd16(TMP16, TMP12, TMP0); yading@10: yading@10: vis_st64(DST_0, dest[0]); yading@10: vis_mul8x16al(DST_2, CONST_512, TMP4); yading@10: vis_padd16(TMP18, TMP14, TMP2); yading@10: yading@10: vis_mul8x16al(DST_3, CONST_512, TMP6); yading@10: vis_padd16(TMP0, CONST_3, TMP0); yading@10: yading@10: vis_padd16(TMP2, CONST_3, TMP2); yading@10: yading@10: vis_padd16(TMP0, TMP4, TMP0); yading@10: yading@10: vis_padd16(TMP2, TMP6, TMP2); yading@10: vis_pack16(TMP0, DST_2); yading@10: yading@10: vis_pack16(TMP2, DST_3); yading@10: vis_st64(DST_2, dest[8]); yading@10: yading@10: ref += stride; yading@10: dest += stride; yading@10: } while (--height); yading@10: } yading@10: yading@10: static void MC_avg_x_8_vis (uint8_t * dest, const uint8_t * ref, yading@10: const ptrdiff_t stride, int height) yading@10: { yading@10: unsigned long off = (unsigned long) ref & 0x7; yading@10: unsigned long off_plus_1 = off + 1; yading@10: int stride_times_2 = stride << 1; yading@10: yading@10: vis_set_gsr(5 << VIS_GSR_SCALEFACT_SHIFT); yading@10: yading@10: vis_ld64(constants3[0], CONST_3); yading@10: vis_fzero(ZERO); yading@10: vis_ld64(constants256_512[0], CONST_256); yading@10: yading@10: ref = vis_alignaddr(ref); yading@10: height >>= 2; yading@10: do { /* 47 cycles */ yading@10: vis_ld64(ref[0], TMP0); yading@10: yading@10: vis_ld64_2(ref, 8, TMP2); yading@10: ref += stride; yading@10: yading@10: vis_alignaddr_g0((void *)off); yading@10: yading@10: vis_ld64(ref[0], TMP4); yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_ld64_2(ref, 8, TMP6); yading@10: ref += stride; yading@10: yading@10: vis_ld64(ref[0], TMP8); yading@10: yading@10: vis_ld64_2(ref, 8, TMP10); yading@10: ref += stride; yading@10: vis_faligndata(TMP4, TMP6, REF_4); yading@10: yading@10: vis_ld64(ref[0], TMP12); yading@10: yading@10: vis_ld64_2(ref, 8, TMP14); yading@10: ref += stride; yading@10: vis_faligndata(TMP8, TMP10, REF_S0); yading@10: yading@10: vis_faligndata(TMP12, TMP14, REF_S4); yading@10: yading@10: if (off != 0x7) { yading@10: vis_alignaddr_g0((void *)off_plus_1); yading@10: yading@10: vis_ld64(dest[0], DST_0); yading@10: vis_faligndata(TMP0, TMP2, REF_2); yading@10: yading@10: vis_ld64_2(dest, stride, DST_2); yading@10: vis_faligndata(TMP4, TMP6, REF_6); yading@10: yading@10: vis_faligndata(TMP8, TMP10, REF_S2); yading@10: yading@10: vis_faligndata(TMP12, TMP14, REF_S6); yading@10: } else { yading@10: vis_ld64(dest[0], DST_0); yading@10: vis_src1(TMP2, REF_2); yading@10: yading@10: vis_ld64_2(dest, stride, DST_2); yading@10: vis_src1(TMP6, REF_6); yading@10: yading@10: vis_src1(TMP10, REF_S2); yading@10: yading@10: vis_src1(TMP14, REF_S6); yading@10: } yading@10: yading@10: vis_pmerge(ZERO, REF_0, TMP0); yading@10: vis_mul8x16au(REF_0_1, CONST_256, TMP2); yading@10: yading@10: vis_pmerge(ZERO, REF_2, TMP4); yading@10: vis_mul8x16au(REF_2_1, CONST_256, TMP6); yading@10: yading@10: vis_padd16(TMP0, CONST_3, TMP0); yading@10: vis_mul8x16al(DST_0, CONST_512, TMP16); yading@10: yading@10: vis_padd16(TMP2, CONST_3, TMP2); yading@10: vis_mul8x16al(DST_1, CONST_512, TMP18); yading@10: yading@10: vis_padd16(TMP0, TMP4, TMP0); yading@10: vis_mul8x16au(REF_4, CONST_256, TMP8); yading@10: yading@10: vis_padd16(TMP2, TMP6, TMP2); yading@10: vis_mul8x16au(REF_4_1, CONST_256, TMP10); yading@10: yading@10: vis_padd16(TMP0, TMP16, TMP0); yading@10: vis_mul8x16au(REF_6, CONST_256, TMP12); yading@10: yading@10: vis_padd16(TMP2, TMP18, TMP2); yading@10: vis_mul8x16au(REF_6_1, CONST_256, TMP14); yading@10: yading@10: vis_padd16(TMP8, CONST_3, TMP8); yading@10: vis_mul8x16al(DST_2, CONST_512, TMP16); yading@10: yading@10: vis_padd16(TMP8, TMP12, TMP8); yading@10: vis_mul8x16al(DST_3, CONST_512, TMP18); yading@10: yading@10: vis_padd16(TMP10, TMP14, TMP10); yading@10: vis_pack16(TMP0, DST_0); yading@10: yading@10: vis_pack16(TMP2, DST_1); yading@10: vis_st64(DST_0, dest[0]); yading@10: dest += stride; yading@10: vis_padd16(TMP10, CONST_3, TMP10); yading@10: yading@10: vis_ld64_2(dest, stride, DST_0); yading@10: vis_padd16(TMP8, TMP16, TMP8); yading@10: yading@10: vis_ld64_2(dest, stride_times_2, TMP4/*DST_2*/); yading@10: vis_padd16(TMP10, TMP18, TMP10); yading@10: vis_pack16(TMP8, DST_2); yading@10: yading@10: vis_pack16(TMP10, DST_3); yading@10: vis_st64(DST_2, dest[0]); yading@10: dest += stride; yading@10: yading@10: vis_mul8x16au(REF_S0_1, CONST_256, TMP2); yading@10: vis_pmerge(ZERO, REF_S0, TMP0); yading@10: yading@10: vis_pmerge(ZERO, REF_S2, TMP24); yading@10: vis_mul8x16au(REF_S2_1, CONST_256, TMP6); yading@10: yading@10: vis_padd16(TMP0, CONST_3, TMP0); yading@10: vis_mul8x16au(REF_S4, CONST_256, TMP8); yading@10: yading@10: vis_padd16(TMP2, CONST_3, TMP2); yading@10: vis_mul8x16au(REF_S4_1, CONST_256, TMP10); yading@10: yading@10: vis_padd16(TMP0, TMP24, TMP0); yading@10: vis_mul8x16au(REF_S6, CONST_256, TMP12); yading@10: yading@10: vis_padd16(TMP2, TMP6, TMP2); yading@10: vis_mul8x16au(REF_S6_1, CONST_256, TMP14); yading@10: yading@10: vis_padd16(TMP8, CONST_3, TMP8); yading@10: vis_mul8x16al(DST_0, CONST_512, TMP16); yading@10: yading@10: vis_padd16(TMP10, CONST_3, TMP10); yading@10: vis_mul8x16al(DST_1, CONST_512, TMP18); yading@10: yading@10: vis_padd16(TMP8, TMP12, TMP8); yading@10: vis_mul8x16al(TMP4/*DST_2*/, CONST_512, TMP20); yading@10: yading@10: vis_mul8x16al(TMP5/*DST_3*/, CONST_512, TMP22); yading@10: vis_padd16(TMP0, TMP16, TMP0); yading@10: yading@10: vis_padd16(TMP2, TMP18, TMP2); yading@10: vis_pack16(TMP0, DST_0); yading@10: yading@10: vis_padd16(TMP10, TMP14, TMP10); yading@10: vis_pack16(TMP2, DST_1); yading@10: vis_st64(DST_0, dest[0]); yading@10: dest += stride; yading@10: yading@10: vis_padd16(TMP8, TMP20, TMP8); yading@10: yading@10: vis_padd16(TMP10, TMP22, TMP10); yading@10: vis_pack16(TMP8, DST_2); yading@10: yading@10: vis_pack16(TMP10, DST_3); yading@10: vis_st64(DST_2, dest[0]); yading@10: dest += stride; yading@10: } while (--height); yading@10: } yading@10: yading@10: static void MC_put_y_16_vis (uint8_t * dest, const uint8_t * ref, yading@10: const ptrdiff_t stride, int height) yading@10: { yading@10: ref = vis_alignaddr(ref); yading@10: vis_ld64(ref[0], TMP0); yading@10: yading@10: vis_ld64_2(ref, 8, TMP2); yading@10: yading@10: vis_ld64_2(ref, 16, TMP4); yading@10: ref += stride; yading@10: yading@10: vis_ld64(ref[0], TMP6); yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_ld64_2(ref, 8, TMP8); yading@10: vis_faligndata(TMP2, TMP4, REF_4); yading@10: yading@10: vis_ld64_2(ref, 16, TMP10); yading@10: ref += stride; yading@10: yading@10: vis_ld64(constants_fe[0], MASK_fe); yading@10: vis_faligndata(TMP6, TMP8, REF_2); yading@10: yading@10: vis_ld64(constants_7f[0], MASK_7f); yading@10: vis_faligndata(TMP8, TMP10, REF_6); yading@10: yading@10: vis_ld64(constants128[0], CONST_128); yading@10: height = (height >> 1) - 1; yading@10: do { /* 24 cycles */ yading@10: vis_ld64(ref[0], TMP0); yading@10: vis_xor(REF_0, REF_2, TMP12); yading@10: yading@10: vis_ld64_2(ref, 8, TMP2); yading@10: vis_xor(REF_4, REF_6, TMP16); yading@10: yading@10: vis_ld64_2(ref, 16, TMP4); yading@10: ref += stride; yading@10: vis_or(REF_0, REF_2, TMP14); yading@10: yading@10: vis_ld64(ref[0], TMP6); yading@10: vis_or(REF_4, REF_6, TMP18); yading@10: yading@10: vis_ld64_2(ref, 8, TMP8); yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_ld64_2(ref, 16, TMP10); yading@10: ref += stride; yading@10: vis_faligndata(TMP2, TMP4, REF_4); yading@10: yading@10: vis_and(TMP12, MASK_fe, TMP12); yading@10: yading@10: vis_and(TMP16, MASK_fe, TMP16); yading@10: vis_mul8x16(CONST_128, TMP12, TMP12); yading@10: yading@10: vis_mul8x16(CONST_128, TMP16, TMP16); yading@10: vis_xor(REF_0, REF_2, TMP0); yading@10: yading@10: vis_xor(REF_4, REF_6, TMP2); yading@10: yading@10: vis_or(REF_0, REF_2, TMP20); yading@10: yading@10: vis_and(TMP12, MASK_7f, TMP12); yading@10: yading@10: vis_and(TMP16, MASK_7f, TMP16); yading@10: yading@10: vis_psub16(TMP14, TMP12, TMP12); yading@10: vis_st64(TMP12, dest[0]); yading@10: yading@10: vis_psub16(TMP18, TMP16, TMP16); yading@10: vis_st64_2(TMP16, dest, 8); yading@10: dest += stride; yading@10: yading@10: vis_or(REF_4, REF_6, TMP18); yading@10: yading@10: vis_and(TMP0, MASK_fe, TMP0); yading@10: yading@10: vis_and(TMP2, MASK_fe, TMP2); yading@10: vis_mul8x16(CONST_128, TMP0, TMP0); yading@10: yading@10: vis_faligndata(TMP6, TMP8, REF_2); yading@10: vis_mul8x16(CONST_128, TMP2, TMP2); yading@10: yading@10: vis_faligndata(TMP8, TMP10, REF_6); yading@10: yading@10: vis_and(TMP0, MASK_7f, TMP0); yading@10: yading@10: vis_and(TMP2, MASK_7f, TMP2); yading@10: yading@10: vis_psub16(TMP20, TMP0, TMP0); yading@10: vis_st64(TMP0, dest[0]); yading@10: yading@10: vis_psub16(TMP18, TMP2, TMP2); yading@10: vis_st64_2(TMP2, dest, 8); yading@10: dest += stride; yading@10: } while (--height); yading@10: yading@10: vis_ld64(ref[0], TMP0); yading@10: vis_xor(REF_0, REF_2, TMP12); yading@10: yading@10: vis_ld64_2(ref, 8, TMP2); yading@10: vis_xor(REF_4, REF_6, TMP16); yading@10: yading@10: vis_ld64_2(ref, 16, TMP4); yading@10: vis_or(REF_0, REF_2, TMP14); yading@10: yading@10: vis_or(REF_4, REF_6, TMP18); yading@10: yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_faligndata(TMP2, TMP4, REF_4); yading@10: yading@10: vis_and(TMP12, MASK_fe, TMP12); yading@10: yading@10: vis_and(TMP16, MASK_fe, TMP16); yading@10: vis_mul8x16(CONST_128, TMP12, TMP12); yading@10: yading@10: vis_mul8x16(CONST_128, TMP16, TMP16); yading@10: vis_xor(REF_0, REF_2, TMP0); yading@10: yading@10: vis_xor(REF_4, REF_6, TMP2); yading@10: yading@10: vis_or(REF_0, REF_2, TMP20); yading@10: yading@10: vis_and(TMP12, MASK_7f, TMP12); yading@10: yading@10: vis_and(TMP16, MASK_7f, TMP16); yading@10: yading@10: vis_psub16(TMP14, TMP12, TMP12); yading@10: vis_st64(TMP12, dest[0]); yading@10: yading@10: vis_psub16(TMP18, TMP16, TMP16); yading@10: vis_st64_2(TMP16, dest, 8); yading@10: dest += stride; yading@10: yading@10: vis_or(REF_4, REF_6, TMP18); yading@10: yading@10: vis_and(TMP0, MASK_fe, TMP0); yading@10: yading@10: vis_and(TMP2, MASK_fe, TMP2); yading@10: vis_mul8x16(CONST_128, TMP0, TMP0); yading@10: yading@10: vis_mul8x16(CONST_128, TMP2, TMP2); yading@10: yading@10: vis_and(TMP0, MASK_7f, TMP0); yading@10: yading@10: vis_and(TMP2, MASK_7f, TMP2); yading@10: yading@10: vis_psub16(TMP20, TMP0, TMP0); yading@10: vis_st64(TMP0, dest[0]); yading@10: yading@10: vis_psub16(TMP18, TMP2, TMP2); yading@10: vis_st64_2(TMP2, dest, 8); yading@10: } yading@10: yading@10: static void MC_put_y_8_vis (uint8_t * dest, const uint8_t * ref, yading@10: const ptrdiff_t stride, int height) yading@10: { yading@10: ref = vis_alignaddr(ref); yading@10: vis_ld64(ref[0], TMP0); yading@10: yading@10: vis_ld64_2(ref, 8, TMP2); yading@10: ref += stride; yading@10: yading@10: vis_ld64(ref[0], TMP4); yading@10: yading@10: vis_ld64_2(ref, 8, TMP6); yading@10: ref += stride; yading@10: yading@10: vis_ld64(constants_fe[0], MASK_fe); yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_ld64(constants_7f[0], MASK_7f); yading@10: vis_faligndata(TMP4, TMP6, REF_2); yading@10: yading@10: vis_ld64(constants128[0], CONST_128); yading@10: height = (height >> 1) - 1; yading@10: do { /* 12 cycles */ yading@10: vis_ld64(ref[0], TMP0); yading@10: vis_xor(REF_0, REF_2, TMP4); yading@10: yading@10: vis_ld64_2(ref, 8, TMP2); yading@10: ref += stride; yading@10: vis_and(TMP4, MASK_fe, TMP4); yading@10: yading@10: vis_or(REF_0, REF_2, TMP6); yading@10: vis_mul8x16(CONST_128, TMP4, TMP4); yading@10: yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: vis_ld64(ref[0], TMP0); yading@10: yading@10: vis_ld64_2(ref, 8, TMP2); yading@10: ref += stride; yading@10: vis_xor(REF_0, REF_2, TMP12); yading@10: yading@10: vis_and(TMP4, MASK_7f, TMP4); yading@10: yading@10: vis_and(TMP12, MASK_fe, TMP12); yading@10: yading@10: vis_mul8x16(CONST_128, TMP12, TMP12); yading@10: vis_or(REF_0, REF_2, TMP14); yading@10: yading@10: vis_psub16(TMP6, TMP4, DST_0); yading@10: vis_st64(DST_0, dest[0]); yading@10: dest += stride; yading@10: yading@10: vis_faligndata(TMP0, TMP2, REF_2); yading@10: yading@10: vis_and(TMP12, MASK_7f, TMP12); yading@10: yading@10: vis_psub16(TMP14, TMP12, DST_0); yading@10: vis_st64(DST_0, dest[0]); yading@10: dest += stride; yading@10: } while (--height); yading@10: yading@10: vis_ld64(ref[0], TMP0); yading@10: vis_xor(REF_0, REF_2, TMP4); yading@10: yading@10: vis_ld64_2(ref, 8, TMP2); yading@10: vis_and(TMP4, MASK_fe, TMP4); yading@10: yading@10: vis_or(REF_0, REF_2, TMP6); yading@10: vis_mul8x16(CONST_128, TMP4, TMP4); yading@10: yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_xor(REF_0, REF_2, TMP12); yading@10: yading@10: vis_and(TMP4, MASK_7f, TMP4); yading@10: yading@10: vis_and(TMP12, MASK_fe, TMP12); yading@10: yading@10: vis_mul8x16(CONST_128, TMP12, TMP12); yading@10: vis_or(REF_0, REF_2, TMP14); yading@10: yading@10: vis_psub16(TMP6, TMP4, DST_0); yading@10: vis_st64(DST_0, dest[0]); yading@10: dest += stride; yading@10: yading@10: vis_and(TMP12, MASK_7f, TMP12); yading@10: yading@10: vis_psub16(TMP14, TMP12, DST_0); yading@10: vis_st64(DST_0, dest[0]); yading@10: } yading@10: yading@10: static void MC_avg_y_16_vis (uint8_t * dest, const uint8_t * ref, yading@10: const ptrdiff_t stride, int height) yading@10: { yading@10: int stride_8 = stride + 8; yading@10: int stride_16 = stride + 16; yading@10: yading@10: vis_set_gsr(5 << VIS_GSR_SCALEFACT_SHIFT); yading@10: yading@10: ref = vis_alignaddr(ref); yading@10: yading@10: vis_ld64(ref[ 0], TMP0); yading@10: vis_fzero(ZERO); yading@10: yading@10: vis_ld64(ref[ 8], TMP2); yading@10: yading@10: vis_ld64(ref[16], TMP4); yading@10: yading@10: vis_ld64(constants3[0], CONST_3); yading@10: vis_faligndata(TMP0, TMP2, REF_2); yading@10: yading@10: vis_ld64(constants256_512[0], CONST_256); yading@10: vis_faligndata(TMP2, TMP4, REF_6); yading@10: height >>= 1; yading@10: yading@10: do { /* 31 cycles */ yading@10: vis_ld64_2(ref, stride, TMP0); yading@10: vis_pmerge(ZERO, REF_2, TMP12); yading@10: vis_mul8x16au(REF_2_1, CONST_256, TMP14); yading@10: yading@10: vis_ld64_2(ref, stride_8, TMP2); yading@10: vis_pmerge(ZERO, REF_6, TMP16); yading@10: vis_mul8x16au(REF_6_1, CONST_256, TMP18); yading@10: yading@10: vis_ld64_2(ref, stride_16, TMP4); yading@10: ref += stride; yading@10: yading@10: vis_ld64(dest[0], DST_0); yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_ld64_2(dest, 8, DST_2); yading@10: vis_faligndata(TMP2, TMP4, REF_4); yading@10: yading@10: vis_ld64_2(ref, stride, TMP6); yading@10: vis_pmerge(ZERO, REF_0, TMP0); yading@10: vis_mul8x16au(REF_0_1, CONST_256, TMP2); yading@10: yading@10: vis_ld64_2(ref, stride_8, TMP8); yading@10: vis_pmerge(ZERO, REF_4, TMP4); yading@10: yading@10: vis_ld64_2(ref, stride_16, TMP10); yading@10: ref += stride; yading@10: yading@10: vis_ld64_2(dest, stride, REF_S0/*DST_4*/); yading@10: vis_faligndata(TMP6, TMP8, REF_2); yading@10: vis_mul8x16au(REF_4_1, CONST_256, TMP6); yading@10: yading@10: vis_ld64_2(dest, stride_8, REF_S2/*DST_6*/); yading@10: vis_faligndata(TMP8, TMP10, REF_6); yading@10: vis_mul8x16al(DST_0, CONST_512, TMP20); yading@10: yading@10: vis_padd16(TMP0, CONST_3, TMP0); yading@10: vis_mul8x16al(DST_1, CONST_512, TMP22); yading@10: yading@10: vis_padd16(TMP2, CONST_3, TMP2); yading@10: vis_mul8x16al(DST_2, CONST_512, TMP24); yading@10: yading@10: vis_padd16(TMP4, CONST_3, TMP4); yading@10: vis_mul8x16al(DST_3, CONST_512, TMP26); yading@10: yading@10: vis_padd16(TMP6, CONST_3, TMP6); yading@10: yading@10: vis_padd16(TMP12, TMP20, TMP12); yading@10: vis_mul8x16al(REF_S0, CONST_512, TMP20); yading@10: yading@10: vis_padd16(TMP14, TMP22, TMP14); yading@10: vis_mul8x16al(REF_S0_1, CONST_512, TMP22); yading@10: yading@10: vis_padd16(TMP16, TMP24, TMP16); yading@10: vis_mul8x16al(REF_S2, CONST_512, TMP24); yading@10: yading@10: vis_padd16(TMP18, TMP26, TMP18); yading@10: vis_mul8x16al(REF_S2_1, CONST_512, TMP26); yading@10: yading@10: vis_padd16(TMP12, TMP0, TMP12); yading@10: vis_mul8x16au(REF_2, CONST_256, TMP28); yading@10: yading@10: vis_padd16(TMP14, TMP2, TMP14); yading@10: vis_mul8x16au(REF_2_1, CONST_256, TMP30); yading@10: yading@10: vis_padd16(TMP16, TMP4, TMP16); yading@10: vis_mul8x16au(REF_6, CONST_256, REF_S4); yading@10: yading@10: vis_padd16(TMP18, TMP6, TMP18); yading@10: vis_mul8x16au(REF_6_1, CONST_256, REF_S6); yading@10: yading@10: vis_pack16(TMP12, DST_0); yading@10: vis_padd16(TMP28, TMP0, TMP12); yading@10: yading@10: vis_pack16(TMP14, DST_1); yading@10: vis_st64(DST_0, dest[0]); yading@10: vis_padd16(TMP30, TMP2, TMP14); yading@10: yading@10: vis_pack16(TMP16, DST_2); yading@10: vis_padd16(REF_S4, TMP4, TMP16); yading@10: yading@10: vis_pack16(TMP18, DST_3); yading@10: vis_st64_2(DST_2, dest, 8); yading@10: dest += stride; yading@10: vis_padd16(REF_S6, TMP6, TMP18); yading@10: yading@10: vis_padd16(TMP12, TMP20, TMP12); yading@10: yading@10: vis_padd16(TMP14, TMP22, TMP14); yading@10: vis_pack16(TMP12, DST_0); yading@10: yading@10: vis_padd16(TMP16, TMP24, TMP16); yading@10: vis_pack16(TMP14, DST_1); yading@10: vis_st64(DST_0, dest[0]); yading@10: yading@10: vis_padd16(TMP18, TMP26, TMP18); yading@10: vis_pack16(TMP16, DST_2); yading@10: yading@10: vis_pack16(TMP18, DST_3); yading@10: vis_st64_2(DST_2, dest, 8); yading@10: dest += stride; yading@10: } while (--height); yading@10: } yading@10: yading@10: static void MC_avg_y_8_vis (uint8_t * dest, const uint8_t * ref, yading@10: const ptrdiff_t stride, int height) yading@10: { yading@10: int stride_8 = stride + 8; yading@10: yading@10: vis_set_gsr(5 << VIS_GSR_SCALEFACT_SHIFT); yading@10: yading@10: ref = vis_alignaddr(ref); yading@10: yading@10: vis_ld64(ref[ 0], TMP0); yading@10: vis_fzero(ZERO); yading@10: yading@10: vis_ld64(ref[ 8], TMP2); yading@10: yading@10: vis_ld64(constants3[0], CONST_3); yading@10: vis_faligndata(TMP0, TMP2, REF_2); yading@10: yading@10: vis_ld64(constants256_512[0], CONST_256); yading@10: yading@10: height >>= 1; yading@10: do { /* 20 cycles */ yading@10: vis_ld64_2(ref, stride, TMP0); yading@10: vis_pmerge(ZERO, REF_2, TMP8); yading@10: vis_mul8x16au(REF_2_1, CONST_256, TMP10); yading@10: yading@10: vis_ld64_2(ref, stride_8, TMP2); yading@10: ref += stride; yading@10: yading@10: vis_ld64(dest[0], DST_0); yading@10: yading@10: vis_ld64_2(dest, stride, DST_2); yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_ld64_2(ref, stride, TMP4); yading@10: vis_mul8x16al(DST_0, CONST_512, TMP16); yading@10: vis_pmerge(ZERO, REF_0, TMP12); yading@10: yading@10: vis_ld64_2(ref, stride_8, TMP6); yading@10: ref += stride; yading@10: vis_mul8x16al(DST_1, CONST_512, TMP18); yading@10: vis_pmerge(ZERO, REF_0_1, TMP14); yading@10: yading@10: vis_padd16(TMP12, CONST_3, TMP12); yading@10: vis_mul8x16al(DST_2, CONST_512, TMP24); yading@10: yading@10: vis_padd16(TMP14, CONST_3, TMP14); yading@10: vis_mul8x16al(DST_3, CONST_512, TMP26); yading@10: yading@10: vis_faligndata(TMP4, TMP6, REF_2); yading@10: yading@10: vis_padd16(TMP8, TMP12, TMP8); yading@10: yading@10: vis_padd16(TMP10, TMP14, TMP10); yading@10: vis_mul8x16au(REF_2, CONST_256, TMP20); yading@10: yading@10: vis_padd16(TMP8, TMP16, TMP0); yading@10: vis_mul8x16au(REF_2_1, CONST_256, TMP22); yading@10: yading@10: vis_padd16(TMP10, TMP18, TMP2); yading@10: vis_pack16(TMP0, DST_0); yading@10: yading@10: vis_pack16(TMP2, DST_1); yading@10: vis_st64(DST_0, dest[0]); yading@10: dest += stride; yading@10: vis_padd16(TMP12, TMP20, TMP12); yading@10: yading@10: vis_padd16(TMP14, TMP22, TMP14); yading@10: yading@10: vis_padd16(TMP12, TMP24, TMP0); yading@10: yading@10: vis_padd16(TMP14, TMP26, TMP2); yading@10: vis_pack16(TMP0, DST_2); yading@10: yading@10: vis_pack16(TMP2, DST_3); yading@10: vis_st64(DST_2, dest[0]); yading@10: dest += stride; yading@10: } while (--height); yading@10: } yading@10: yading@10: static void MC_put_xy_16_vis (uint8_t * dest, const uint8_t * ref, yading@10: const ptrdiff_t stride, int height) yading@10: { yading@10: unsigned long off = (unsigned long) ref & 0x7; yading@10: unsigned long off_plus_1 = off + 1; yading@10: int stride_8 = stride + 8; yading@10: int stride_16 = stride + 16; yading@10: yading@10: vis_set_gsr(5 << VIS_GSR_SCALEFACT_SHIFT); yading@10: yading@10: ref = vis_alignaddr(ref); yading@10: yading@10: vis_ld64(ref[ 0], TMP0); yading@10: vis_fzero(ZERO); yading@10: yading@10: vis_ld64(ref[ 8], TMP2); yading@10: yading@10: vis_ld64(ref[16], TMP4); yading@10: yading@10: vis_ld64(constants2[0], CONST_2); yading@10: vis_faligndata(TMP0, TMP2, REF_S0); yading@10: yading@10: vis_ld64(constants256_512[0], CONST_256); yading@10: vis_faligndata(TMP2, TMP4, REF_S4); yading@10: yading@10: if (off != 0x7) { yading@10: vis_alignaddr_g0((void *)off_plus_1); yading@10: vis_faligndata(TMP0, TMP2, REF_S2); yading@10: vis_faligndata(TMP2, TMP4, REF_S6); yading@10: } else { yading@10: vis_src1(TMP2, REF_S2); yading@10: vis_src1(TMP4, REF_S6); yading@10: } yading@10: yading@10: height >>= 1; yading@10: do { yading@10: vis_ld64_2(ref, stride, TMP0); yading@10: vis_mul8x16au(REF_S0, CONST_256, TMP12); yading@10: vis_pmerge(ZERO, REF_S0_1, TMP14); yading@10: yading@10: vis_alignaddr_g0((void *)off); yading@10: yading@10: vis_ld64_2(ref, stride_8, TMP2); yading@10: vis_mul8x16au(REF_S2, CONST_256, TMP16); yading@10: vis_pmerge(ZERO, REF_S2_1, TMP18); yading@10: yading@10: vis_ld64_2(ref, stride_16, TMP4); yading@10: ref += stride; yading@10: vis_mul8x16au(REF_S4, CONST_256, TMP20); yading@10: vis_pmerge(ZERO, REF_S4_1, TMP22); yading@10: yading@10: vis_ld64_2(ref, stride, TMP6); yading@10: vis_mul8x16au(REF_S6, CONST_256, TMP24); yading@10: vis_pmerge(ZERO, REF_S6_1, TMP26); yading@10: yading@10: vis_ld64_2(ref, stride_8, TMP8); yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_ld64_2(ref, stride_16, TMP10); yading@10: ref += stride; yading@10: vis_faligndata(TMP2, TMP4, REF_4); yading@10: yading@10: vis_faligndata(TMP6, TMP8, REF_S0); yading@10: yading@10: vis_faligndata(TMP8, TMP10, REF_S4); yading@10: yading@10: if (off != 0x7) { yading@10: vis_alignaddr_g0((void *)off_plus_1); yading@10: vis_faligndata(TMP0, TMP2, REF_2); yading@10: vis_faligndata(TMP2, TMP4, REF_6); yading@10: vis_faligndata(TMP6, TMP8, REF_S2); yading@10: vis_faligndata(TMP8, TMP10, REF_S6); yading@10: } else { yading@10: vis_src1(TMP2, REF_2); yading@10: vis_src1(TMP4, REF_6); yading@10: vis_src1(TMP8, REF_S2); yading@10: vis_src1(TMP10, REF_S6); yading@10: } yading@10: yading@10: vis_mul8x16au(REF_0, CONST_256, TMP0); yading@10: vis_pmerge(ZERO, REF_0_1, TMP2); yading@10: yading@10: vis_mul8x16au(REF_2, CONST_256, TMP4); yading@10: vis_pmerge(ZERO, REF_2_1, TMP6); yading@10: yading@10: vis_padd16(TMP0, CONST_2, TMP8); yading@10: vis_mul8x16au(REF_4, CONST_256, TMP0); yading@10: yading@10: vis_padd16(TMP2, CONST_2, TMP10); yading@10: vis_mul8x16au(REF_4_1, CONST_256, TMP2); yading@10: yading@10: vis_padd16(TMP8, TMP4, TMP8); yading@10: vis_mul8x16au(REF_6, CONST_256, TMP4); yading@10: yading@10: vis_padd16(TMP10, TMP6, TMP10); yading@10: vis_mul8x16au(REF_6_1, CONST_256, TMP6); yading@10: yading@10: vis_padd16(TMP12, TMP8, TMP12); yading@10: yading@10: vis_padd16(TMP14, TMP10, TMP14); yading@10: yading@10: vis_padd16(TMP12, TMP16, TMP12); yading@10: yading@10: vis_padd16(TMP14, TMP18, TMP14); yading@10: vis_pack16(TMP12, DST_0); yading@10: yading@10: vis_pack16(TMP14, DST_1); yading@10: vis_st64(DST_0, dest[0]); yading@10: vis_padd16(TMP0, CONST_2, TMP12); yading@10: yading@10: vis_mul8x16au(REF_S0, CONST_256, TMP0); yading@10: vis_padd16(TMP2, CONST_2, TMP14); yading@10: yading@10: vis_mul8x16au(REF_S0_1, CONST_256, TMP2); yading@10: vis_padd16(TMP12, TMP4, TMP12); yading@10: yading@10: vis_mul8x16au(REF_S2, CONST_256, TMP4); yading@10: vis_padd16(TMP14, TMP6, TMP14); yading@10: yading@10: vis_mul8x16au(REF_S2_1, CONST_256, TMP6); yading@10: vis_padd16(TMP20, TMP12, TMP20); yading@10: yading@10: vis_padd16(TMP22, TMP14, TMP22); yading@10: yading@10: vis_padd16(TMP20, TMP24, TMP20); yading@10: yading@10: vis_padd16(TMP22, TMP26, TMP22); yading@10: vis_pack16(TMP20, DST_2); yading@10: yading@10: vis_pack16(TMP22, DST_3); yading@10: vis_st64_2(DST_2, dest, 8); yading@10: dest += stride; yading@10: vis_padd16(TMP0, TMP4, TMP24); yading@10: yading@10: vis_mul8x16au(REF_S4, CONST_256, TMP0); yading@10: vis_padd16(TMP2, TMP6, TMP26); yading@10: yading@10: vis_mul8x16au(REF_S4_1, CONST_256, TMP2); yading@10: vis_padd16(TMP24, TMP8, TMP24); yading@10: yading@10: vis_padd16(TMP26, TMP10, TMP26); yading@10: vis_pack16(TMP24, DST_0); yading@10: yading@10: vis_pack16(TMP26, DST_1); yading@10: vis_st64(DST_0, dest[0]); yading@10: vis_pmerge(ZERO, REF_S6, TMP4); yading@10: yading@10: vis_pmerge(ZERO, REF_S6_1, TMP6); yading@10: yading@10: vis_padd16(TMP0, TMP4, TMP0); yading@10: yading@10: vis_padd16(TMP2, TMP6, TMP2); yading@10: yading@10: vis_padd16(TMP0, TMP12, TMP0); yading@10: yading@10: vis_padd16(TMP2, TMP14, TMP2); yading@10: vis_pack16(TMP0, DST_2); yading@10: yading@10: vis_pack16(TMP2, DST_3); yading@10: vis_st64_2(DST_2, dest, 8); yading@10: dest += stride; yading@10: } while (--height); yading@10: } yading@10: yading@10: static void MC_put_xy_8_vis (uint8_t * dest, const uint8_t * ref, yading@10: const ptrdiff_t stride, int height) yading@10: { yading@10: unsigned long off = (unsigned long) ref & 0x7; yading@10: unsigned long off_plus_1 = off + 1; yading@10: int stride_8 = stride + 8; yading@10: yading@10: vis_set_gsr(5 << VIS_GSR_SCALEFACT_SHIFT); yading@10: yading@10: ref = vis_alignaddr(ref); yading@10: yading@10: vis_ld64(ref[ 0], TMP0); yading@10: vis_fzero(ZERO); yading@10: yading@10: vis_ld64(ref[ 8], TMP2); yading@10: yading@10: vis_ld64(constants2[0], CONST_2); yading@10: yading@10: vis_ld64(constants256_512[0], CONST_256); yading@10: vis_faligndata(TMP0, TMP2, REF_S0); yading@10: yading@10: if (off != 0x7) { yading@10: vis_alignaddr_g0((void *)off_plus_1); yading@10: vis_faligndata(TMP0, TMP2, REF_S2); yading@10: } else { yading@10: vis_src1(TMP2, REF_S2); yading@10: } yading@10: yading@10: height >>= 1; yading@10: do { /* 26 cycles */ yading@10: vis_ld64_2(ref, stride, TMP0); yading@10: vis_mul8x16au(REF_S0, CONST_256, TMP8); yading@10: vis_pmerge(ZERO, REF_S2, TMP12); yading@10: yading@10: vis_alignaddr_g0((void *)off); yading@10: yading@10: vis_ld64_2(ref, stride_8, TMP2); yading@10: ref += stride; yading@10: vis_mul8x16au(REF_S0_1, CONST_256, TMP10); yading@10: vis_pmerge(ZERO, REF_S2_1, TMP14); yading@10: yading@10: vis_ld64_2(ref, stride, TMP4); yading@10: yading@10: vis_ld64_2(ref, stride_8, TMP6); yading@10: ref += stride; yading@10: vis_faligndata(TMP0, TMP2, REF_S4); yading@10: yading@10: vis_pmerge(ZERO, REF_S4, TMP18); yading@10: yading@10: vis_pmerge(ZERO, REF_S4_1, TMP20); yading@10: yading@10: vis_faligndata(TMP4, TMP6, REF_S0); yading@10: yading@10: if (off != 0x7) { yading@10: vis_alignaddr_g0((void *)off_plus_1); yading@10: vis_faligndata(TMP0, TMP2, REF_S6); yading@10: vis_faligndata(TMP4, TMP6, REF_S2); yading@10: } else { yading@10: vis_src1(TMP2, REF_S6); yading@10: vis_src1(TMP6, REF_S2); yading@10: } yading@10: yading@10: vis_padd16(TMP18, CONST_2, TMP18); yading@10: vis_mul8x16au(REF_S6, CONST_256, TMP22); yading@10: yading@10: vis_padd16(TMP20, CONST_2, TMP20); yading@10: vis_mul8x16au(REF_S6_1, CONST_256, TMP24); yading@10: yading@10: vis_mul8x16au(REF_S0, CONST_256, TMP26); yading@10: vis_pmerge(ZERO, REF_S0_1, TMP28); yading@10: yading@10: vis_mul8x16au(REF_S2, CONST_256, TMP30); yading@10: vis_padd16(TMP18, TMP22, TMP18); yading@10: yading@10: vis_mul8x16au(REF_S2_1, CONST_256, TMP32); yading@10: vis_padd16(TMP20, TMP24, TMP20); yading@10: yading@10: vis_padd16(TMP8, TMP18, TMP8); yading@10: yading@10: vis_padd16(TMP10, TMP20, TMP10); yading@10: yading@10: vis_padd16(TMP8, TMP12, TMP8); yading@10: yading@10: vis_padd16(TMP10, TMP14, TMP10); yading@10: vis_pack16(TMP8, DST_0); yading@10: yading@10: vis_pack16(TMP10, DST_1); yading@10: vis_st64(DST_0, dest[0]); yading@10: dest += stride; yading@10: vis_padd16(TMP18, TMP26, TMP18); yading@10: yading@10: vis_padd16(TMP20, TMP28, TMP20); yading@10: yading@10: vis_padd16(TMP18, TMP30, TMP18); yading@10: yading@10: vis_padd16(TMP20, TMP32, TMP20); yading@10: vis_pack16(TMP18, DST_2); yading@10: yading@10: vis_pack16(TMP20, DST_3); yading@10: vis_st64(DST_2, dest[0]); yading@10: dest += stride; yading@10: } while (--height); yading@10: } yading@10: yading@10: static void MC_avg_xy_16_vis (uint8_t * dest, const uint8_t * ref, yading@10: const ptrdiff_t stride, int height) yading@10: { yading@10: unsigned long off = (unsigned long) ref & 0x7; yading@10: unsigned long off_plus_1 = off + 1; yading@10: int stride_8 = stride + 8; yading@10: int stride_16 = stride + 16; yading@10: yading@10: vis_set_gsr(4 << VIS_GSR_SCALEFACT_SHIFT); yading@10: yading@10: ref = vis_alignaddr(ref); yading@10: yading@10: vis_ld64(ref[ 0], TMP0); yading@10: vis_fzero(ZERO); yading@10: yading@10: vis_ld64(ref[ 8], TMP2); yading@10: yading@10: vis_ld64(ref[16], TMP4); yading@10: yading@10: vis_ld64(constants6[0], CONST_6); yading@10: vis_faligndata(TMP0, TMP2, REF_S0); yading@10: yading@10: vis_ld64(constants256_1024[0], CONST_256); yading@10: vis_faligndata(TMP2, TMP4, REF_S4); yading@10: yading@10: if (off != 0x7) { yading@10: vis_alignaddr_g0((void *)off_plus_1); yading@10: vis_faligndata(TMP0, TMP2, REF_S2); yading@10: vis_faligndata(TMP2, TMP4, REF_S6); yading@10: } else { yading@10: vis_src1(TMP2, REF_S2); yading@10: vis_src1(TMP4, REF_S6); yading@10: } yading@10: yading@10: height >>= 1; yading@10: do { /* 55 cycles */ yading@10: vis_ld64_2(ref, stride, TMP0); yading@10: vis_mul8x16au(REF_S0, CONST_256, TMP12); yading@10: vis_pmerge(ZERO, REF_S0_1, TMP14); yading@10: yading@10: vis_alignaddr_g0((void *)off); yading@10: yading@10: vis_ld64_2(ref, stride_8, TMP2); yading@10: vis_mul8x16au(REF_S2, CONST_256, TMP16); yading@10: vis_pmerge(ZERO, REF_S2_1, TMP18); yading@10: yading@10: vis_ld64_2(ref, stride_16, TMP4); yading@10: ref += stride; yading@10: vis_mul8x16au(REF_S4, CONST_256, TMP20); yading@10: vis_pmerge(ZERO, REF_S4_1, TMP22); yading@10: yading@10: vis_ld64_2(ref, stride, TMP6); yading@10: vis_mul8x16au(REF_S6, CONST_256, TMP24); yading@10: vis_pmerge(ZERO, REF_S6_1, TMP26); yading@10: yading@10: vis_ld64_2(ref, stride_8, TMP8); yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_ld64_2(ref, stride_16, TMP10); yading@10: ref += stride; yading@10: vis_faligndata(TMP2, TMP4, REF_4); yading@10: yading@10: vis_ld64(dest[0], DST_0); yading@10: vis_faligndata(TMP6, TMP8, REF_S0); yading@10: yading@10: vis_ld64_2(dest, 8, DST_2); yading@10: vis_faligndata(TMP8, TMP10, REF_S4); yading@10: yading@10: if (off != 0x7) { yading@10: vis_alignaddr_g0((void *)off_plus_1); yading@10: vis_faligndata(TMP0, TMP2, REF_2); yading@10: vis_faligndata(TMP2, TMP4, REF_6); yading@10: vis_faligndata(TMP6, TMP8, REF_S2); yading@10: vis_faligndata(TMP8, TMP10, REF_S6); yading@10: } else { yading@10: vis_src1(TMP2, REF_2); yading@10: vis_src1(TMP4, REF_6); yading@10: vis_src1(TMP8, REF_S2); yading@10: vis_src1(TMP10, REF_S6); yading@10: } yading@10: yading@10: vis_mul8x16al(DST_0, CONST_1024, TMP30); yading@10: vis_pmerge(ZERO, REF_0, TMP0); yading@10: yading@10: vis_mul8x16al(DST_1, CONST_1024, TMP32); yading@10: vis_pmerge(ZERO, REF_0_1, TMP2); yading@10: yading@10: vis_mul8x16au(REF_2, CONST_256, TMP4); yading@10: vis_pmerge(ZERO, REF_2_1, TMP6); yading@10: yading@10: vis_mul8x16al(DST_2, CONST_1024, REF_0); yading@10: vis_padd16(TMP0, CONST_6, TMP0); yading@10: yading@10: vis_mul8x16al(DST_3, CONST_1024, REF_2); yading@10: vis_padd16(TMP2, CONST_6, TMP2); yading@10: yading@10: vis_padd16(TMP0, TMP4, TMP0); yading@10: vis_mul8x16au(REF_4, CONST_256, TMP4); yading@10: yading@10: vis_padd16(TMP2, TMP6, TMP2); yading@10: vis_mul8x16au(REF_4_1, CONST_256, TMP6); yading@10: yading@10: vis_padd16(TMP12, TMP0, TMP12); yading@10: vis_mul8x16au(REF_6, CONST_256, TMP8); yading@10: yading@10: vis_padd16(TMP14, TMP2, TMP14); yading@10: vis_mul8x16au(REF_6_1, CONST_256, TMP10); yading@10: yading@10: vis_padd16(TMP12, TMP16, TMP12); yading@10: vis_mul8x16au(REF_S0, CONST_256, REF_4); yading@10: yading@10: vis_padd16(TMP14, TMP18, TMP14); yading@10: vis_mul8x16au(REF_S0_1, CONST_256, REF_6); yading@10: yading@10: vis_padd16(TMP12, TMP30, TMP12); yading@10: yading@10: vis_padd16(TMP14, TMP32, TMP14); yading@10: vis_pack16(TMP12, DST_0); yading@10: yading@10: vis_pack16(TMP14, DST_1); yading@10: vis_st64(DST_0, dest[0]); yading@10: vis_padd16(TMP4, CONST_6, TMP4); yading@10: yading@10: vis_ld64_2(dest, stride, DST_0); yading@10: vis_padd16(TMP6, CONST_6, TMP6); yading@10: vis_mul8x16au(REF_S2, CONST_256, TMP12); yading@10: yading@10: vis_padd16(TMP4, TMP8, TMP4); yading@10: vis_mul8x16au(REF_S2_1, CONST_256, TMP14); yading@10: yading@10: vis_padd16(TMP6, TMP10, TMP6); yading@10: yading@10: vis_padd16(TMP20, TMP4, TMP20); yading@10: yading@10: vis_padd16(TMP22, TMP6, TMP22); yading@10: yading@10: vis_padd16(TMP20, TMP24, TMP20); yading@10: yading@10: vis_padd16(TMP22, TMP26, TMP22); yading@10: yading@10: vis_padd16(TMP20, REF_0, TMP20); yading@10: vis_mul8x16au(REF_S4, CONST_256, REF_0); yading@10: yading@10: vis_padd16(TMP22, REF_2, TMP22); yading@10: vis_pack16(TMP20, DST_2); yading@10: yading@10: vis_pack16(TMP22, DST_3); yading@10: vis_st64_2(DST_2, dest, 8); yading@10: dest += stride; yading@10: yading@10: vis_ld64_2(dest, 8, DST_2); yading@10: vis_mul8x16al(DST_0, CONST_1024, TMP30); yading@10: vis_pmerge(ZERO, REF_S4_1, REF_2); yading@10: yading@10: vis_mul8x16al(DST_1, CONST_1024, TMP32); yading@10: vis_padd16(REF_4, TMP0, TMP8); yading@10: yading@10: vis_mul8x16au(REF_S6, CONST_256, REF_4); yading@10: vis_padd16(REF_6, TMP2, TMP10); yading@10: yading@10: vis_mul8x16au(REF_S6_1, CONST_256, REF_6); yading@10: vis_padd16(TMP8, TMP12, TMP8); yading@10: yading@10: vis_padd16(TMP10, TMP14, TMP10); yading@10: yading@10: vis_padd16(TMP8, TMP30, TMP8); yading@10: yading@10: vis_padd16(TMP10, TMP32, TMP10); yading@10: vis_pack16(TMP8, DST_0); yading@10: yading@10: vis_pack16(TMP10, DST_1); yading@10: vis_st64(DST_0, dest[0]); yading@10: yading@10: vis_padd16(REF_0, TMP4, REF_0); yading@10: yading@10: vis_mul8x16al(DST_2, CONST_1024, TMP30); yading@10: vis_padd16(REF_2, TMP6, REF_2); yading@10: yading@10: vis_mul8x16al(DST_3, CONST_1024, TMP32); yading@10: vis_padd16(REF_0, REF_4, REF_0); yading@10: yading@10: vis_padd16(REF_2, REF_6, REF_2); yading@10: yading@10: vis_padd16(REF_0, TMP30, REF_0); yading@10: yading@10: /* stall */ yading@10: yading@10: vis_padd16(REF_2, TMP32, REF_2); yading@10: vis_pack16(REF_0, DST_2); yading@10: yading@10: vis_pack16(REF_2, DST_3); yading@10: vis_st64_2(DST_2, dest, 8); yading@10: dest += stride; yading@10: } while (--height); yading@10: } yading@10: yading@10: static void MC_avg_xy_8_vis (uint8_t * dest, const uint8_t * ref, yading@10: const ptrdiff_t stride, int height) yading@10: { yading@10: unsigned long off = (unsigned long) ref & 0x7; yading@10: unsigned long off_plus_1 = off + 1; yading@10: int stride_8 = stride + 8; yading@10: yading@10: vis_set_gsr(4 << VIS_GSR_SCALEFACT_SHIFT); yading@10: yading@10: ref = vis_alignaddr(ref); yading@10: yading@10: vis_ld64(ref[0], TMP0); yading@10: vis_fzero(ZERO); yading@10: yading@10: vis_ld64_2(ref, 8, TMP2); yading@10: yading@10: vis_ld64(constants6[0], CONST_6); yading@10: yading@10: vis_ld64(constants256_1024[0], CONST_256); yading@10: vis_faligndata(TMP0, TMP2, REF_S0); yading@10: yading@10: if (off != 0x7) { yading@10: vis_alignaddr_g0((void *)off_plus_1); yading@10: vis_faligndata(TMP0, TMP2, REF_S2); yading@10: } else { yading@10: vis_src1(TMP2, REF_S2); yading@10: } yading@10: yading@10: height >>= 1; yading@10: do { /* 31 cycles */ yading@10: vis_ld64_2(ref, stride, TMP0); yading@10: vis_mul8x16au(REF_S0, CONST_256, TMP8); yading@10: vis_pmerge(ZERO, REF_S0_1, TMP10); yading@10: yading@10: vis_ld64_2(ref, stride_8, TMP2); yading@10: ref += stride; yading@10: vis_mul8x16au(REF_S2, CONST_256, TMP12); yading@10: vis_pmerge(ZERO, REF_S2_1, TMP14); yading@10: yading@10: vis_alignaddr_g0((void *)off); yading@10: yading@10: vis_ld64_2(ref, stride, TMP4); yading@10: vis_faligndata(TMP0, TMP2, REF_S4); yading@10: yading@10: vis_ld64_2(ref, stride_8, TMP6); yading@10: ref += stride; yading@10: yading@10: vis_ld64(dest[0], DST_0); yading@10: vis_faligndata(TMP4, TMP6, REF_S0); yading@10: yading@10: vis_ld64_2(dest, stride, DST_2); yading@10: yading@10: if (off != 0x7) { yading@10: vis_alignaddr_g0((void *)off_plus_1); yading@10: vis_faligndata(TMP0, TMP2, REF_S6); yading@10: vis_faligndata(TMP4, TMP6, REF_S2); yading@10: } else { yading@10: vis_src1(TMP2, REF_S6); yading@10: vis_src1(TMP6, REF_S2); yading@10: } yading@10: yading@10: vis_mul8x16al(DST_0, CONST_1024, TMP30); yading@10: vis_pmerge(ZERO, REF_S4, TMP22); yading@10: yading@10: vis_mul8x16al(DST_1, CONST_1024, TMP32); yading@10: vis_pmerge(ZERO, REF_S4_1, TMP24); yading@10: yading@10: vis_mul8x16au(REF_S6, CONST_256, TMP26); yading@10: vis_pmerge(ZERO, REF_S6_1, TMP28); yading@10: yading@10: vis_mul8x16au(REF_S0, CONST_256, REF_S4); yading@10: vis_padd16(TMP22, CONST_6, TMP22); yading@10: yading@10: vis_mul8x16au(REF_S0_1, CONST_256, REF_S6); yading@10: vis_padd16(TMP24, CONST_6, TMP24); yading@10: yading@10: vis_mul8x16al(DST_2, CONST_1024, REF_0); yading@10: vis_padd16(TMP22, TMP26, TMP22); yading@10: yading@10: vis_mul8x16al(DST_3, CONST_1024, REF_2); yading@10: vis_padd16(TMP24, TMP28, TMP24); yading@10: yading@10: vis_mul8x16au(REF_S2, CONST_256, TMP26); yading@10: vis_padd16(TMP8, TMP22, TMP8); yading@10: yading@10: vis_mul8x16au(REF_S2_1, CONST_256, TMP28); yading@10: vis_padd16(TMP10, TMP24, TMP10); yading@10: yading@10: vis_padd16(TMP8, TMP12, TMP8); yading@10: yading@10: vis_padd16(TMP10, TMP14, TMP10); yading@10: yading@10: vis_padd16(TMP8, TMP30, TMP8); yading@10: yading@10: vis_padd16(TMP10, TMP32, TMP10); yading@10: vis_pack16(TMP8, DST_0); yading@10: yading@10: vis_pack16(TMP10, DST_1); yading@10: vis_st64(DST_0, dest[0]); yading@10: dest += stride; yading@10: yading@10: vis_padd16(REF_S4, TMP22, TMP12); yading@10: yading@10: vis_padd16(REF_S6, TMP24, TMP14); yading@10: yading@10: vis_padd16(TMP12, TMP26, TMP12); yading@10: yading@10: vis_padd16(TMP14, TMP28, TMP14); yading@10: yading@10: vis_padd16(TMP12, REF_0, TMP12); yading@10: yading@10: vis_padd16(TMP14, REF_2, TMP14); yading@10: vis_pack16(TMP12, DST_2); yading@10: yading@10: vis_pack16(TMP14, DST_3); yading@10: vis_st64(DST_2, dest[0]); yading@10: dest += stride; yading@10: } while (--height); yading@10: } yading@10: yading@10: /* End of rounding code */ yading@10: yading@10: /* Start of no rounding code */ yading@10: /* The trick used in some of this file is the formula from the MMX yading@10: * motion comp code, which is: yading@10: * yading@10: * (x+y)>>1 == (x&y)+((x^y)>>1) yading@10: * yading@10: * This allows us to average 8 bytes at a time in a 64-bit FPU reg. yading@10: * We avoid overflows by masking before we do the shift, and we yading@10: * implement the shift by multiplying by 1/2 using mul8x16. So in yading@10: * VIS this is (assume 'x' is in f0, 'y' is in f2, a repeating mask yading@10: * of '0xfe' is in f4, a repeating mask of '0x7f' is in f6, and yading@10: * the value 0x80808080 is in f8): yading@10: * yading@10: * fxor f0, f2, f10 yading@10: * fand f10, f4, f10 yading@10: * fmul8x16 f8, f10, f10 yading@10: * fand f10, f6, f10 yading@10: * fand f0, f2, f12 yading@10: * fpadd16 f12, f10, f10 yading@10: */ yading@10: yading@10: static void MC_put_no_round_o_16_vis (uint8_t * dest, const uint8_t * ref, yading@10: const ptrdiff_t stride, int height) yading@10: { yading@10: ref = vis_alignaddr(ref); yading@10: do { /* 5 cycles */ yading@10: vis_ld64(ref[0], TMP0); yading@10: yading@10: vis_ld64_2(ref, 8, TMP2); yading@10: yading@10: vis_ld64_2(ref, 16, TMP4); yading@10: ref += stride; yading@10: yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: vis_st64(REF_0, dest[0]); yading@10: yading@10: vis_faligndata(TMP2, TMP4, REF_2); yading@10: vis_st64_2(REF_2, dest, 8); yading@10: dest += stride; yading@10: } while (--height); yading@10: } yading@10: yading@10: static void MC_put_no_round_o_8_vis (uint8_t * dest, const uint8_t * ref, yading@10: const ptrdiff_t stride, int height) yading@10: { yading@10: ref = vis_alignaddr(ref); yading@10: do { /* 4 cycles */ yading@10: vis_ld64(ref[0], TMP0); yading@10: yading@10: vis_ld64(ref[8], TMP2); yading@10: ref += stride; yading@10: yading@10: /* stall */ yading@10: yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: vis_st64(REF_0, dest[0]); yading@10: dest += stride; yading@10: } while (--height); yading@10: } yading@10: yading@10: yading@10: static void MC_avg_no_round_o_16_vis (uint8_t * dest, const uint8_t * ref, yading@10: const ptrdiff_t stride, int height) yading@10: { yading@10: int stride_8 = stride + 8; yading@10: yading@10: ref = vis_alignaddr(ref); yading@10: yading@10: vis_ld64(ref[0], TMP0); yading@10: yading@10: vis_ld64(ref[8], TMP2); yading@10: yading@10: vis_ld64(ref[16], TMP4); yading@10: yading@10: vis_ld64(dest[0], DST_0); yading@10: yading@10: vis_ld64(dest[8], DST_2); yading@10: yading@10: vis_ld64(constants_fe[0], MASK_fe); yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_ld64(constants_7f[0], MASK_7f); yading@10: vis_faligndata(TMP2, TMP4, REF_2); yading@10: yading@10: vis_ld64(constants128[0], CONST_128); yading@10: yading@10: ref += stride; yading@10: height = (height >> 1) - 1; yading@10: yading@10: do { /* 24 cycles */ yading@10: vis_ld64(ref[0], TMP0); yading@10: vis_xor(DST_0, REF_0, TMP6); yading@10: yading@10: vis_ld64_2(ref, 8, TMP2); yading@10: vis_and(TMP6, MASK_fe, TMP6); yading@10: yading@10: vis_ld64_2(ref, 16, TMP4); yading@10: ref += stride; yading@10: vis_mul8x16(CONST_128, TMP6, TMP6); yading@10: vis_xor(DST_2, REF_2, TMP8); yading@10: yading@10: vis_and(TMP8, MASK_fe, TMP8); yading@10: yading@10: vis_and(DST_0, REF_0, TMP10); yading@10: vis_ld64_2(dest, stride, DST_0); yading@10: vis_mul8x16(CONST_128, TMP8, TMP8); yading@10: yading@10: vis_and(DST_2, REF_2, TMP12); yading@10: vis_ld64_2(dest, stride_8, DST_2); yading@10: yading@10: vis_ld64(ref[0], TMP14); yading@10: vis_and(TMP6, MASK_7f, TMP6); yading@10: yading@10: vis_and(TMP8, MASK_7f, TMP8); yading@10: yading@10: vis_padd16(TMP10, TMP6, TMP6); yading@10: vis_st64(TMP6, dest[0]); yading@10: yading@10: vis_padd16(TMP12, TMP8, TMP8); yading@10: vis_st64_2(TMP8, dest, 8); yading@10: yading@10: dest += stride; yading@10: vis_ld64_2(ref, 8, TMP16); yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_ld64_2(ref, 16, TMP18); yading@10: vis_faligndata(TMP2, TMP4, REF_2); yading@10: ref += stride; yading@10: yading@10: vis_xor(DST_0, REF_0, TMP20); yading@10: yading@10: vis_and(TMP20, MASK_fe, TMP20); yading@10: yading@10: vis_xor(DST_2, REF_2, TMP22); yading@10: vis_mul8x16(CONST_128, TMP20, TMP20); yading@10: yading@10: vis_and(TMP22, MASK_fe, TMP22); yading@10: yading@10: vis_and(DST_0, REF_0, TMP24); yading@10: vis_mul8x16(CONST_128, TMP22, TMP22); yading@10: yading@10: vis_and(DST_2, REF_2, TMP26); yading@10: yading@10: vis_ld64_2(dest, stride, DST_0); yading@10: vis_faligndata(TMP14, TMP16, REF_0); yading@10: yading@10: vis_ld64_2(dest, stride_8, DST_2); yading@10: vis_faligndata(TMP16, TMP18, REF_2); yading@10: yading@10: vis_and(TMP20, MASK_7f, TMP20); yading@10: yading@10: vis_and(TMP22, MASK_7f, TMP22); yading@10: yading@10: vis_padd16(TMP24, TMP20, TMP20); yading@10: vis_st64(TMP20, dest[0]); yading@10: yading@10: vis_padd16(TMP26, TMP22, TMP22); yading@10: vis_st64_2(TMP22, dest, 8); yading@10: dest += stride; yading@10: } while (--height); yading@10: yading@10: vis_ld64(ref[0], TMP0); yading@10: vis_xor(DST_0, REF_0, TMP6); yading@10: yading@10: vis_ld64_2(ref, 8, TMP2); yading@10: vis_and(TMP6, MASK_fe, TMP6); yading@10: yading@10: vis_ld64_2(ref, 16, TMP4); yading@10: vis_mul8x16(CONST_128, TMP6, TMP6); yading@10: vis_xor(DST_2, REF_2, TMP8); yading@10: yading@10: vis_and(TMP8, MASK_fe, TMP8); yading@10: yading@10: vis_and(DST_0, REF_0, TMP10); yading@10: vis_ld64_2(dest, stride, DST_0); yading@10: vis_mul8x16(CONST_128, TMP8, TMP8); yading@10: yading@10: vis_and(DST_2, REF_2, TMP12); yading@10: vis_ld64_2(dest, stride_8, DST_2); yading@10: yading@10: vis_ld64(ref[0], TMP14); yading@10: vis_and(TMP6, MASK_7f, TMP6); yading@10: yading@10: vis_and(TMP8, MASK_7f, TMP8); yading@10: yading@10: vis_padd16(TMP10, TMP6, TMP6); yading@10: vis_st64(TMP6, dest[0]); yading@10: yading@10: vis_padd16(TMP12, TMP8, TMP8); yading@10: vis_st64_2(TMP8, dest, 8); yading@10: yading@10: dest += stride; yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_faligndata(TMP2, TMP4, REF_2); yading@10: yading@10: vis_xor(DST_0, REF_0, TMP20); yading@10: yading@10: vis_and(TMP20, MASK_fe, TMP20); yading@10: yading@10: vis_xor(DST_2, REF_2, TMP22); yading@10: vis_mul8x16(CONST_128, TMP20, TMP20); yading@10: yading@10: vis_and(TMP22, MASK_fe, TMP22); yading@10: yading@10: vis_and(DST_0, REF_0, TMP24); yading@10: vis_mul8x16(CONST_128, TMP22, TMP22); yading@10: yading@10: vis_and(DST_2, REF_2, TMP26); yading@10: yading@10: vis_and(TMP20, MASK_7f, TMP20); yading@10: yading@10: vis_and(TMP22, MASK_7f, TMP22); yading@10: yading@10: vis_padd16(TMP24, TMP20, TMP20); yading@10: vis_st64(TMP20, dest[0]); yading@10: yading@10: vis_padd16(TMP26, TMP22, TMP22); yading@10: vis_st64_2(TMP22, dest, 8); yading@10: } yading@10: yading@10: static void MC_put_no_round_x_16_vis (uint8_t * dest, const uint8_t * ref, yading@10: const ptrdiff_t stride, int height) yading@10: { yading@10: unsigned long off = (unsigned long) ref & 0x7; yading@10: unsigned long off_plus_1 = off + 1; yading@10: yading@10: ref = vis_alignaddr(ref); yading@10: yading@10: vis_ld64(ref[0], TMP0); yading@10: yading@10: vis_ld64_2(ref, 8, TMP2); yading@10: yading@10: vis_ld64_2(ref, 16, TMP4); yading@10: yading@10: vis_ld64(constants_fe[0], MASK_fe); yading@10: yading@10: vis_ld64(constants_7f[0], MASK_7f); yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_ld64(constants128[0], CONST_128); yading@10: vis_faligndata(TMP2, TMP4, REF_4); yading@10: yading@10: if (off != 0x7) { yading@10: vis_alignaddr_g0((void *)off_plus_1); yading@10: vis_faligndata(TMP0, TMP2, REF_2); yading@10: vis_faligndata(TMP2, TMP4, REF_6); yading@10: } else { yading@10: vis_src1(TMP2, REF_2); yading@10: vis_src1(TMP4, REF_6); yading@10: } yading@10: yading@10: ref += stride; yading@10: height = (height >> 1) - 1; yading@10: yading@10: do { /* 34 cycles */ yading@10: vis_ld64(ref[0], TMP0); yading@10: vis_xor(REF_0, REF_2, TMP6); yading@10: yading@10: vis_ld64_2(ref, 8, TMP2); yading@10: vis_xor(REF_4, REF_6, TMP8); yading@10: yading@10: vis_ld64_2(ref, 16, TMP4); yading@10: vis_and(TMP6, MASK_fe, TMP6); yading@10: ref += stride; yading@10: yading@10: vis_ld64(ref[0], TMP14); yading@10: vis_mul8x16(CONST_128, TMP6, TMP6); yading@10: vis_and(TMP8, MASK_fe, TMP8); yading@10: yading@10: vis_ld64_2(ref, 8, TMP16); yading@10: vis_mul8x16(CONST_128, TMP8, TMP8); yading@10: vis_and(REF_0, REF_2, TMP10); yading@10: yading@10: vis_ld64_2(ref, 16, TMP18); yading@10: ref += stride; yading@10: vis_and(REF_4, REF_6, TMP12); yading@10: yading@10: vis_alignaddr_g0((void *)off); yading@10: yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_faligndata(TMP2, TMP4, REF_4); yading@10: yading@10: if (off != 0x7) { yading@10: vis_alignaddr_g0((void *)off_plus_1); yading@10: vis_faligndata(TMP0, TMP2, REF_2); yading@10: vis_faligndata(TMP2, TMP4, REF_6); yading@10: } else { yading@10: vis_src1(TMP2, REF_2); yading@10: vis_src1(TMP4, REF_6); yading@10: } yading@10: yading@10: vis_and(TMP6, MASK_7f, TMP6); yading@10: yading@10: vis_and(TMP8, MASK_7f, TMP8); yading@10: yading@10: vis_padd16(TMP10, TMP6, TMP6); yading@10: vis_st64(TMP6, dest[0]); yading@10: yading@10: vis_padd16(TMP12, TMP8, TMP8); yading@10: vis_st64_2(TMP8, dest, 8); yading@10: dest += stride; yading@10: yading@10: vis_xor(REF_0, REF_2, TMP6); yading@10: yading@10: vis_xor(REF_4, REF_6, TMP8); yading@10: yading@10: vis_and(TMP6, MASK_fe, TMP6); yading@10: yading@10: vis_mul8x16(CONST_128, TMP6, TMP6); yading@10: vis_and(TMP8, MASK_fe, TMP8); yading@10: yading@10: vis_mul8x16(CONST_128, TMP8, TMP8); yading@10: vis_and(REF_0, REF_2, TMP10); yading@10: yading@10: vis_and(REF_4, REF_6, TMP12); yading@10: yading@10: vis_alignaddr_g0((void *)off); yading@10: yading@10: vis_faligndata(TMP14, TMP16, REF_0); yading@10: yading@10: vis_faligndata(TMP16, TMP18, REF_4); yading@10: yading@10: if (off != 0x7) { yading@10: vis_alignaddr_g0((void *)off_plus_1); yading@10: vis_faligndata(TMP14, TMP16, REF_2); yading@10: vis_faligndata(TMP16, TMP18, REF_6); yading@10: } else { yading@10: vis_src1(TMP16, REF_2); yading@10: vis_src1(TMP18, REF_6); yading@10: } yading@10: yading@10: vis_and(TMP6, MASK_7f, TMP6); yading@10: yading@10: vis_and(TMP8, MASK_7f, TMP8); yading@10: yading@10: vis_padd16(TMP10, TMP6, TMP6); yading@10: vis_st64(TMP6, dest[0]); yading@10: yading@10: vis_padd16(TMP12, TMP8, TMP8); yading@10: vis_st64_2(TMP8, dest, 8); yading@10: dest += stride; yading@10: } while (--height); yading@10: yading@10: vis_ld64(ref[0], TMP0); yading@10: vis_xor(REF_0, REF_2, TMP6); yading@10: yading@10: vis_ld64_2(ref, 8, TMP2); yading@10: vis_xor(REF_4, REF_6, TMP8); yading@10: yading@10: vis_ld64_2(ref, 16, TMP4); yading@10: vis_and(TMP6, MASK_fe, TMP6); yading@10: yading@10: vis_mul8x16(CONST_128, TMP6, TMP6); yading@10: vis_and(TMP8, MASK_fe, TMP8); yading@10: yading@10: vis_mul8x16(CONST_128, TMP8, TMP8); yading@10: vis_and(REF_0, REF_2, TMP10); yading@10: yading@10: vis_and(REF_4, REF_6, TMP12); yading@10: yading@10: vis_alignaddr_g0((void *)off); yading@10: yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_faligndata(TMP2, TMP4, REF_4); yading@10: yading@10: if (off != 0x7) { yading@10: vis_alignaddr_g0((void *)off_plus_1); yading@10: vis_faligndata(TMP0, TMP2, REF_2); yading@10: vis_faligndata(TMP2, TMP4, REF_6); yading@10: } else { yading@10: vis_src1(TMP2, REF_2); yading@10: vis_src1(TMP4, REF_6); yading@10: } yading@10: yading@10: vis_and(TMP6, MASK_7f, TMP6); yading@10: yading@10: vis_and(TMP8, MASK_7f, TMP8); yading@10: yading@10: vis_padd16(TMP10, TMP6, TMP6); yading@10: vis_st64(TMP6, dest[0]); yading@10: yading@10: vis_padd16(TMP12, TMP8, TMP8); yading@10: vis_st64_2(TMP8, dest, 8); yading@10: dest += stride; yading@10: yading@10: vis_xor(REF_0, REF_2, TMP6); yading@10: yading@10: vis_xor(REF_4, REF_6, TMP8); yading@10: yading@10: vis_and(TMP6, MASK_fe, TMP6); yading@10: yading@10: vis_mul8x16(CONST_128, TMP6, TMP6); yading@10: vis_and(TMP8, MASK_fe, TMP8); yading@10: yading@10: vis_mul8x16(CONST_128, TMP8, TMP8); yading@10: vis_and(REF_0, REF_2, TMP10); yading@10: yading@10: vis_and(REF_4, REF_6, TMP12); yading@10: yading@10: vis_and(TMP6, MASK_7f, TMP6); yading@10: yading@10: vis_and(TMP8, MASK_7f, TMP8); yading@10: yading@10: vis_padd16(TMP10, TMP6, TMP6); yading@10: vis_st64(TMP6, dest[0]); yading@10: yading@10: vis_padd16(TMP12, TMP8, TMP8); yading@10: vis_st64_2(TMP8, dest, 8); yading@10: } yading@10: yading@10: static void MC_put_no_round_x_8_vis (uint8_t * dest, const uint8_t * ref, yading@10: const ptrdiff_t stride, int height) yading@10: { yading@10: unsigned long off = (unsigned long) ref & 0x7; yading@10: unsigned long off_plus_1 = off + 1; yading@10: yading@10: ref = vis_alignaddr(ref); yading@10: yading@10: vis_ld64(ref[0], TMP0); yading@10: yading@10: vis_ld64(ref[8], TMP2); yading@10: yading@10: vis_ld64(constants_fe[0], MASK_fe); yading@10: yading@10: vis_ld64(constants_7f[0], MASK_7f); yading@10: yading@10: vis_ld64(constants128[0], CONST_128); yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: if (off != 0x7) { yading@10: vis_alignaddr_g0((void *)off_plus_1); yading@10: vis_faligndata(TMP0, TMP2, REF_2); yading@10: } else { yading@10: vis_src1(TMP2, REF_2); yading@10: } yading@10: yading@10: ref += stride; yading@10: height = (height >> 1) - 1; yading@10: yading@10: do { /* 20 cycles */ yading@10: vis_ld64(ref[0], TMP0); yading@10: vis_xor(REF_0, REF_2, TMP4); yading@10: yading@10: vis_ld64_2(ref, 8, TMP2); yading@10: vis_and(TMP4, MASK_fe, TMP4); yading@10: ref += stride; yading@10: yading@10: vis_ld64(ref[0], TMP8); yading@10: vis_and(REF_0, REF_2, TMP6); yading@10: vis_mul8x16(CONST_128, TMP4, TMP4); yading@10: yading@10: vis_alignaddr_g0((void *)off); yading@10: yading@10: vis_ld64_2(ref, 8, TMP10); yading@10: ref += stride; yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: if (off != 0x7) { yading@10: vis_alignaddr_g0((void *)off_plus_1); yading@10: vis_faligndata(TMP0, TMP2, REF_2); yading@10: } else { yading@10: vis_src1(TMP2, REF_2); yading@10: } yading@10: yading@10: vis_and(TMP4, MASK_7f, TMP4); yading@10: yading@10: vis_padd16(TMP6, TMP4, DST_0); yading@10: vis_st64(DST_0, dest[0]); yading@10: dest += stride; yading@10: yading@10: vis_xor(REF_0, REF_2, TMP12); yading@10: yading@10: vis_and(TMP12, MASK_fe, TMP12); yading@10: yading@10: vis_and(REF_0, REF_2, TMP14); yading@10: vis_mul8x16(CONST_128, TMP12, TMP12); yading@10: yading@10: vis_alignaddr_g0((void *)off); yading@10: vis_faligndata(TMP8, TMP10, REF_0); yading@10: if (off != 0x7) { yading@10: vis_alignaddr_g0((void *)off_plus_1); yading@10: vis_faligndata(TMP8, TMP10, REF_2); yading@10: } else { yading@10: vis_src1(TMP10, REF_2); yading@10: } yading@10: yading@10: vis_and(TMP12, MASK_7f, TMP12); yading@10: yading@10: vis_padd16(TMP14, TMP12, DST_0); yading@10: vis_st64(DST_0, dest[0]); yading@10: dest += stride; yading@10: } while (--height); yading@10: yading@10: vis_ld64(ref[0], TMP0); yading@10: vis_xor(REF_0, REF_2, TMP4); yading@10: yading@10: vis_ld64_2(ref, 8, TMP2); yading@10: vis_and(TMP4, MASK_fe, TMP4); yading@10: yading@10: vis_and(REF_0, REF_2, TMP6); yading@10: vis_mul8x16(CONST_128, TMP4, TMP4); yading@10: yading@10: vis_alignaddr_g0((void *)off); yading@10: yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: if (off != 0x7) { yading@10: vis_alignaddr_g0((void *)off_plus_1); yading@10: vis_faligndata(TMP0, TMP2, REF_2); yading@10: } else { yading@10: vis_src1(TMP2, REF_2); yading@10: } yading@10: yading@10: vis_and(TMP4, MASK_7f, TMP4); yading@10: yading@10: vis_padd16(TMP6, TMP4, DST_0); yading@10: vis_st64(DST_0, dest[0]); yading@10: dest += stride; yading@10: yading@10: vis_xor(REF_0, REF_2, TMP12); yading@10: yading@10: vis_and(TMP12, MASK_fe, TMP12); yading@10: yading@10: vis_and(REF_0, REF_2, TMP14); yading@10: vis_mul8x16(CONST_128, TMP12, TMP12); yading@10: yading@10: vis_and(TMP12, MASK_7f, TMP12); yading@10: yading@10: vis_padd16(TMP14, TMP12, DST_0); yading@10: vis_st64(DST_0, dest[0]); yading@10: dest += stride; yading@10: } yading@10: yading@10: static void MC_avg_no_round_x_16_vis (uint8_t * dest, const uint8_t * ref, yading@10: const ptrdiff_t stride, int height) yading@10: { yading@10: unsigned long off = (unsigned long) ref & 0x7; yading@10: unsigned long off_plus_1 = off + 1; yading@10: yading@10: vis_set_gsr(5 << VIS_GSR_SCALEFACT_SHIFT); yading@10: yading@10: vis_ld64(constants3[0], CONST_3); yading@10: vis_fzero(ZERO); yading@10: vis_ld64(constants256_512[0], CONST_256); yading@10: yading@10: ref = vis_alignaddr(ref); yading@10: do { /* 26 cycles */ yading@10: vis_ld64(ref[0], TMP0); yading@10: yading@10: vis_ld64(ref[8], TMP2); yading@10: yading@10: vis_alignaddr_g0((void *)off); yading@10: yading@10: vis_ld64(ref[16], TMP4); yading@10: yading@10: vis_ld64(dest[0], DST_0); yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_ld64(dest[8], DST_2); yading@10: vis_faligndata(TMP2, TMP4, REF_4); yading@10: yading@10: if (off != 0x7) { yading@10: vis_alignaddr_g0((void *)off_plus_1); yading@10: vis_faligndata(TMP0, TMP2, REF_2); yading@10: vis_faligndata(TMP2, TMP4, REF_6); yading@10: } else { yading@10: vis_src1(TMP2, REF_2); yading@10: vis_src1(TMP4, REF_6); yading@10: } yading@10: yading@10: vis_mul8x16au(REF_0, CONST_256, TMP0); yading@10: yading@10: vis_pmerge(ZERO, REF_2, TMP4); yading@10: vis_mul8x16au(REF_0_1, CONST_256, TMP2); yading@10: yading@10: vis_pmerge(ZERO, REF_2_1, TMP6); yading@10: yading@10: vis_padd16(TMP0, TMP4, TMP0); yading@10: yading@10: vis_mul8x16al(DST_0, CONST_512, TMP4); yading@10: vis_padd16(TMP2, TMP6, TMP2); yading@10: yading@10: vis_mul8x16al(DST_1, CONST_512, TMP6); yading@10: yading@10: vis_mul8x16au(REF_6, CONST_256, TMP12); yading@10: yading@10: vis_padd16(TMP0, TMP4, TMP0); yading@10: vis_mul8x16au(REF_6_1, CONST_256, TMP14); yading@10: yading@10: vis_padd16(TMP2, TMP6, TMP2); yading@10: vis_mul8x16au(REF_4, CONST_256, TMP16); yading@10: yading@10: vis_padd16(TMP0, CONST_3, TMP8); yading@10: vis_mul8x16au(REF_4_1, CONST_256, TMP18); yading@10: yading@10: vis_padd16(TMP2, CONST_3, TMP10); yading@10: vis_pack16(TMP8, DST_0); yading@10: yading@10: vis_pack16(TMP10, DST_1); yading@10: vis_padd16(TMP16, TMP12, TMP0); yading@10: yading@10: vis_st64(DST_0, dest[0]); yading@10: vis_mul8x16al(DST_2, CONST_512, TMP4); yading@10: vis_padd16(TMP18, TMP14, TMP2); yading@10: yading@10: vis_mul8x16al(DST_3, CONST_512, TMP6); yading@10: vis_padd16(TMP0, CONST_3, TMP0); yading@10: yading@10: vis_padd16(TMP2, CONST_3, TMP2); yading@10: yading@10: vis_padd16(TMP0, TMP4, TMP0); yading@10: yading@10: vis_padd16(TMP2, TMP6, TMP2); yading@10: vis_pack16(TMP0, DST_2); yading@10: yading@10: vis_pack16(TMP2, DST_3); yading@10: vis_st64(DST_2, dest[8]); yading@10: yading@10: ref += stride; yading@10: dest += stride; yading@10: } while (--height); yading@10: } yading@10: yading@10: static void MC_put_no_round_y_16_vis (uint8_t * dest, const uint8_t * ref, yading@10: const ptrdiff_t stride, int height) yading@10: { yading@10: ref = vis_alignaddr(ref); yading@10: vis_ld64(ref[0], TMP0); yading@10: yading@10: vis_ld64_2(ref, 8, TMP2); yading@10: yading@10: vis_ld64_2(ref, 16, TMP4); yading@10: ref += stride; yading@10: yading@10: vis_ld64(ref[0], TMP6); yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_ld64_2(ref, 8, TMP8); yading@10: vis_faligndata(TMP2, TMP4, REF_4); yading@10: yading@10: vis_ld64_2(ref, 16, TMP10); yading@10: ref += stride; yading@10: yading@10: vis_ld64(constants_fe[0], MASK_fe); yading@10: vis_faligndata(TMP6, TMP8, REF_2); yading@10: yading@10: vis_ld64(constants_7f[0], MASK_7f); yading@10: vis_faligndata(TMP8, TMP10, REF_6); yading@10: yading@10: vis_ld64(constants128[0], CONST_128); yading@10: height = (height >> 1) - 1; yading@10: do { /* 24 cycles */ yading@10: vis_ld64(ref[0], TMP0); yading@10: vis_xor(REF_0, REF_2, TMP12); yading@10: yading@10: vis_ld64_2(ref, 8, TMP2); yading@10: vis_xor(REF_4, REF_6, TMP16); yading@10: yading@10: vis_ld64_2(ref, 16, TMP4); yading@10: ref += stride; yading@10: vis_and(REF_0, REF_2, TMP14); yading@10: yading@10: vis_ld64(ref[0], TMP6); yading@10: vis_and(REF_4, REF_6, TMP18); yading@10: yading@10: vis_ld64_2(ref, 8, TMP8); yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_ld64_2(ref, 16, TMP10); yading@10: ref += stride; yading@10: vis_faligndata(TMP2, TMP4, REF_4); yading@10: yading@10: vis_and(TMP12, MASK_fe, TMP12); yading@10: yading@10: vis_and(TMP16, MASK_fe, TMP16); yading@10: vis_mul8x16(CONST_128, TMP12, TMP12); yading@10: yading@10: vis_mul8x16(CONST_128, TMP16, TMP16); yading@10: vis_xor(REF_0, REF_2, TMP0); yading@10: yading@10: vis_xor(REF_4, REF_6, TMP2); yading@10: yading@10: vis_and(REF_0, REF_2, TMP20); yading@10: yading@10: vis_and(TMP12, MASK_7f, TMP12); yading@10: yading@10: vis_and(TMP16, MASK_7f, TMP16); yading@10: yading@10: vis_padd16(TMP14, TMP12, TMP12); yading@10: vis_st64(TMP12, dest[0]); yading@10: yading@10: vis_padd16(TMP18, TMP16, TMP16); yading@10: vis_st64_2(TMP16, dest, 8); yading@10: dest += stride; yading@10: yading@10: vis_and(REF_4, REF_6, TMP18); yading@10: yading@10: vis_and(TMP0, MASK_fe, TMP0); yading@10: yading@10: vis_and(TMP2, MASK_fe, TMP2); yading@10: vis_mul8x16(CONST_128, TMP0, TMP0); yading@10: yading@10: vis_faligndata(TMP6, TMP8, REF_2); yading@10: vis_mul8x16(CONST_128, TMP2, TMP2); yading@10: yading@10: vis_faligndata(TMP8, TMP10, REF_6); yading@10: yading@10: vis_and(TMP0, MASK_7f, TMP0); yading@10: yading@10: vis_and(TMP2, MASK_7f, TMP2); yading@10: yading@10: vis_padd16(TMP20, TMP0, TMP0); yading@10: vis_st64(TMP0, dest[0]); yading@10: yading@10: vis_padd16(TMP18, TMP2, TMP2); yading@10: vis_st64_2(TMP2, dest, 8); yading@10: dest += stride; yading@10: } while (--height); yading@10: yading@10: vis_ld64(ref[0], TMP0); yading@10: vis_xor(REF_0, REF_2, TMP12); yading@10: yading@10: vis_ld64_2(ref, 8, TMP2); yading@10: vis_xor(REF_4, REF_6, TMP16); yading@10: yading@10: vis_ld64_2(ref, 16, TMP4); yading@10: vis_and(REF_0, REF_2, TMP14); yading@10: yading@10: vis_and(REF_4, REF_6, TMP18); yading@10: yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_faligndata(TMP2, TMP4, REF_4); yading@10: yading@10: vis_and(TMP12, MASK_fe, TMP12); yading@10: yading@10: vis_and(TMP16, MASK_fe, TMP16); yading@10: vis_mul8x16(CONST_128, TMP12, TMP12); yading@10: yading@10: vis_mul8x16(CONST_128, TMP16, TMP16); yading@10: vis_xor(REF_0, REF_2, TMP0); yading@10: yading@10: vis_xor(REF_4, REF_6, TMP2); yading@10: yading@10: vis_and(REF_0, REF_2, TMP20); yading@10: yading@10: vis_and(TMP12, MASK_7f, TMP12); yading@10: yading@10: vis_and(TMP16, MASK_7f, TMP16); yading@10: yading@10: vis_padd16(TMP14, TMP12, TMP12); yading@10: vis_st64(TMP12, dest[0]); yading@10: yading@10: vis_padd16(TMP18, TMP16, TMP16); yading@10: vis_st64_2(TMP16, dest, 8); yading@10: dest += stride; yading@10: yading@10: vis_and(REF_4, REF_6, TMP18); yading@10: yading@10: vis_and(TMP0, MASK_fe, TMP0); yading@10: yading@10: vis_and(TMP2, MASK_fe, TMP2); yading@10: vis_mul8x16(CONST_128, TMP0, TMP0); yading@10: yading@10: vis_mul8x16(CONST_128, TMP2, TMP2); yading@10: yading@10: vis_and(TMP0, MASK_7f, TMP0); yading@10: yading@10: vis_and(TMP2, MASK_7f, TMP2); yading@10: yading@10: vis_padd16(TMP20, TMP0, TMP0); yading@10: vis_st64(TMP0, dest[0]); yading@10: yading@10: vis_padd16(TMP18, TMP2, TMP2); yading@10: vis_st64_2(TMP2, dest, 8); yading@10: } yading@10: yading@10: static void MC_put_no_round_y_8_vis (uint8_t * dest, const uint8_t * ref, yading@10: const ptrdiff_t stride, int height) yading@10: { yading@10: ref = vis_alignaddr(ref); yading@10: vis_ld64(ref[0], TMP0); yading@10: yading@10: vis_ld64_2(ref, 8, TMP2); yading@10: ref += stride; yading@10: yading@10: vis_ld64(ref[0], TMP4); yading@10: yading@10: vis_ld64_2(ref, 8, TMP6); yading@10: ref += stride; yading@10: yading@10: vis_ld64(constants_fe[0], MASK_fe); yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_ld64(constants_7f[0], MASK_7f); yading@10: vis_faligndata(TMP4, TMP6, REF_2); yading@10: yading@10: vis_ld64(constants128[0], CONST_128); yading@10: height = (height >> 1) - 1; yading@10: do { /* 12 cycles */ yading@10: vis_ld64(ref[0], TMP0); yading@10: vis_xor(REF_0, REF_2, TMP4); yading@10: yading@10: vis_ld64_2(ref, 8, TMP2); yading@10: ref += stride; yading@10: vis_and(TMP4, MASK_fe, TMP4); yading@10: yading@10: vis_and(REF_0, REF_2, TMP6); yading@10: vis_mul8x16(CONST_128, TMP4, TMP4); yading@10: yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: vis_ld64(ref[0], TMP0); yading@10: yading@10: vis_ld64_2(ref, 8, TMP2); yading@10: ref += stride; yading@10: vis_xor(REF_0, REF_2, TMP12); yading@10: yading@10: vis_and(TMP4, MASK_7f, TMP4); yading@10: yading@10: vis_and(TMP12, MASK_fe, TMP12); yading@10: yading@10: vis_mul8x16(CONST_128, TMP12, TMP12); yading@10: vis_and(REF_0, REF_2, TMP14); yading@10: yading@10: vis_padd16(TMP6, TMP4, DST_0); yading@10: vis_st64(DST_0, dest[0]); yading@10: dest += stride; yading@10: yading@10: vis_faligndata(TMP0, TMP2, REF_2); yading@10: yading@10: vis_and(TMP12, MASK_7f, TMP12); yading@10: yading@10: vis_padd16(TMP14, TMP12, DST_0); yading@10: vis_st64(DST_0, dest[0]); yading@10: dest += stride; yading@10: } while (--height); yading@10: yading@10: vis_ld64(ref[0], TMP0); yading@10: vis_xor(REF_0, REF_2, TMP4); yading@10: yading@10: vis_ld64_2(ref, 8, TMP2); yading@10: vis_and(TMP4, MASK_fe, TMP4); yading@10: yading@10: vis_and(REF_0, REF_2, TMP6); yading@10: vis_mul8x16(CONST_128, TMP4, TMP4); yading@10: yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_xor(REF_0, REF_2, TMP12); yading@10: yading@10: vis_and(TMP4, MASK_7f, TMP4); yading@10: yading@10: vis_and(TMP12, MASK_fe, TMP12); yading@10: yading@10: vis_mul8x16(CONST_128, TMP12, TMP12); yading@10: vis_and(REF_0, REF_2, TMP14); yading@10: yading@10: vis_padd16(TMP6, TMP4, DST_0); yading@10: vis_st64(DST_0, dest[0]); yading@10: dest += stride; yading@10: yading@10: vis_and(TMP12, MASK_7f, TMP12); yading@10: yading@10: vis_padd16(TMP14, TMP12, DST_0); yading@10: vis_st64(DST_0, dest[0]); yading@10: } yading@10: yading@10: static void MC_avg_no_round_y_16_vis (uint8_t * dest, const uint8_t * ref, yading@10: const ptrdiff_t stride, int height) yading@10: { yading@10: int stride_8 = stride + 8; yading@10: int stride_16 = stride + 16; yading@10: yading@10: vis_set_gsr(5 << VIS_GSR_SCALEFACT_SHIFT); yading@10: yading@10: ref = vis_alignaddr(ref); yading@10: yading@10: vis_ld64(ref[ 0], TMP0); yading@10: vis_fzero(ZERO); yading@10: yading@10: vis_ld64(ref[ 8], TMP2); yading@10: yading@10: vis_ld64(ref[16], TMP4); yading@10: yading@10: vis_ld64(constants3[0], CONST_3); yading@10: vis_faligndata(TMP0, TMP2, REF_2); yading@10: yading@10: vis_ld64(constants256_512[0], CONST_256); yading@10: vis_faligndata(TMP2, TMP4, REF_6); yading@10: height >>= 1; yading@10: yading@10: do { /* 31 cycles */ yading@10: vis_ld64_2(ref, stride, TMP0); yading@10: vis_pmerge(ZERO, REF_2, TMP12); yading@10: vis_mul8x16au(REF_2_1, CONST_256, TMP14); yading@10: yading@10: vis_ld64_2(ref, stride_8, TMP2); yading@10: vis_pmerge(ZERO, REF_6, TMP16); yading@10: vis_mul8x16au(REF_6_1, CONST_256, TMP18); yading@10: yading@10: vis_ld64_2(ref, stride_16, TMP4); yading@10: ref += stride; yading@10: yading@10: vis_ld64(dest[0], DST_0); yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_ld64_2(dest, 8, DST_2); yading@10: vis_faligndata(TMP2, TMP4, REF_4); yading@10: yading@10: vis_ld64_2(ref, stride, TMP6); yading@10: vis_pmerge(ZERO, REF_0, TMP0); yading@10: vis_mul8x16au(REF_0_1, CONST_256, TMP2); yading@10: yading@10: vis_ld64_2(ref, stride_8, TMP8); yading@10: vis_pmerge(ZERO, REF_4, TMP4); yading@10: yading@10: vis_ld64_2(ref, stride_16, TMP10); yading@10: ref += stride; yading@10: yading@10: vis_ld64_2(dest, stride, REF_S0/*DST_4*/); yading@10: vis_faligndata(TMP6, TMP8, REF_2); yading@10: vis_mul8x16au(REF_4_1, CONST_256, TMP6); yading@10: yading@10: vis_ld64_2(dest, stride_8, REF_S2/*DST_6*/); yading@10: vis_faligndata(TMP8, TMP10, REF_6); yading@10: vis_mul8x16al(DST_0, CONST_512, TMP20); yading@10: yading@10: vis_padd16(TMP0, CONST_3, TMP0); yading@10: vis_mul8x16al(DST_1, CONST_512, TMP22); yading@10: yading@10: vis_padd16(TMP2, CONST_3, TMP2); yading@10: vis_mul8x16al(DST_2, CONST_512, TMP24); yading@10: yading@10: vis_padd16(TMP4, CONST_3, TMP4); yading@10: vis_mul8x16al(DST_3, CONST_512, TMP26); yading@10: yading@10: vis_padd16(TMP6, CONST_3, TMP6); yading@10: yading@10: vis_padd16(TMP12, TMP20, TMP12); yading@10: vis_mul8x16al(REF_S0, CONST_512, TMP20); yading@10: yading@10: vis_padd16(TMP14, TMP22, TMP14); yading@10: vis_mul8x16al(REF_S0_1, CONST_512, TMP22); yading@10: yading@10: vis_padd16(TMP16, TMP24, TMP16); yading@10: vis_mul8x16al(REF_S2, CONST_512, TMP24); yading@10: yading@10: vis_padd16(TMP18, TMP26, TMP18); yading@10: vis_mul8x16al(REF_S2_1, CONST_512, TMP26); yading@10: yading@10: vis_padd16(TMP12, TMP0, TMP12); yading@10: vis_mul8x16au(REF_2, CONST_256, TMP28); yading@10: yading@10: vis_padd16(TMP14, TMP2, TMP14); yading@10: vis_mul8x16au(REF_2_1, CONST_256, TMP30); yading@10: yading@10: vis_padd16(TMP16, TMP4, TMP16); yading@10: vis_mul8x16au(REF_6, CONST_256, REF_S4); yading@10: yading@10: vis_padd16(TMP18, TMP6, TMP18); yading@10: vis_mul8x16au(REF_6_1, CONST_256, REF_S6); yading@10: yading@10: vis_pack16(TMP12, DST_0); yading@10: vis_padd16(TMP28, TMP0, TMP12); yading@10: yading@10: vis_pack16(TMP14, DST_1); yading@10: vis_st64(DST_0, dest[0]); yading@10: vis_padd16(TMP30, TMP2, TMP14); yading@10: yading@10: vis_pack16(TMP16, DST_2); yading@10: vis_padd16(REF_S4, TMP4, TMP16); yading@10: yading@10: vis_pack16(TMP18, DST_3); yading@10: vis_st64_2(DST_2, dest, 8); yading@10: dest += stride; yading@10: vis_padd16(REF_S6, TMP6, TMP18); yading@10: yading@10: vis_padd16(TMP12, TMP20, TMP12); yading@10: yading@10: vis_padd16(TMP14, TMP22, TMP14); yading@10: vis_pack16(TMP12, DST_0); yading@10: yading@10: vis_padd16(TMP16, TMP24, TMP16); yading@10: vis_pack16(TMP14, DST_1); yading@10: vis_st64(DST_0, dest[0]); yading@10: yading@10: vis_padd16(TMP18, TMP26, TMP18); yading@10: vis_pack16(TMP16, DST_2); yading@10: yading@10: vis_pack16(TMP18, DST_3); yading@10: vis_st64_2(DST_2, dest, 8); yading@10: dest += stride; yading@10: } while (--height); yading@10: } yading@10: yading@10: static void MC_put_no_round_xy_16_vis (uint8_t * dest, const uint8_t * ref, yading@10: const ptrdiff_t stride, int height) yading@10: { yading@10: unsigned long off = (unsigned long) ref & 0x7; yading@10: unsigned long off_plus_1 = off + 1; yading@10: int stride_8 = stride + 8; yading@10: int stride_16 = stride + 16; yading@10: yading@10: vis_set_gsr(5 << VIS_GSR_SCALEFACT_SHIFT); yading@10: yading@10: ref = vis_alignaddr(ref); yading@10: yading@10: vis_ld64(ref[ 0], TMP0); yading@10: vis_fzero(ZERO); yading@10: yading@10: vis_ld64(ref[ 8], TMP2); yading@10: yading@10: vis_ld64(ref[16], TMP4); yading@10: yading@10: vis_ld64(constants1[0], CONST_1); yading@10: vis_faligndata(TMP0, TMP2, REF_S0); yading@10: yading@10: vis_ld64(constants256_512[0], CONST_256); yading@10: vis_faligndata(TMP2, TMP4, REF_S4); yading@10: yading@10: if (off != 0x7) { yading@10: vis_alignaddr_g0((void *)off_plus_1); yading@10: vis_faligndata(TMP0, TMP2, REF_S2); yading@10: vis_faligndata(TMP2, TMP4, REF_S6); yading@10: } else { yading@10: vis_src1(TMP2, REF_S2); yading@10: vis_src1(TMP4, REF_S6); yading@10: } yading@10: yading@10: height >>= 1; yading@10: do { yading@10: vis_ld64_2(ref, stride, TMP0); yading@10: vis_mul8x16au(REF_S0, CONST_256, TMP12); yading@10: vis_pmerge(ZERO, REF_S0_1, TMP14); yading@10: yading@10: vis_alignaddr_g0((void *)off); yading@10: yading@10: vis_ld64_2(ref, stride_8, TMP2); yading@10: vis_mul8x16au(REF_S2, CONST_256, TMP16); yading@10: vis_pmerge(ZERO, REF_S2_1, TMP18); yading@10: yading@10: vis_ld64_2(ref, stride_16, TMP4); yading@10: ref += stride; yading@10: vis_mul8x16au(REF_S4, CONST_256, TMP20); yading@10: vis_pmerge(ZERO, REF_S4_1, TMP22); yading@10: yading@10: vis_ld64_2(ref, stride, TMP6); yading@10: vis_mul8x16au(REF_S6, CONST_256, TMP24); yading@10: vis_pmerge(ZERO, REF_S6_1, TMP26); yading@10: yading@10: vis_ld64_2(ref, stride_8, TMP8); yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_ld64_2(ref, stride_16, TMP10); yading@10: ref += stride; yading@10: vis_faligndata(TMP2, TMP4, REF_4); yading@10: yading@10: vis_faligndata(TMP6, TMP8, REF_S0); yading@10: yading@10: vis_faligndata(TMP8, TMP10, REF_S4); yading@10: yading@10: if (off != 0x7) { yading@10: vis_alignaddr_g0((void *)off_plus_1); yading@10: vis_faligndata(TMP0, TMP2, REF_2); yading@10: vis_faligndata(TMP2, TMP4, REF_6); yading@10: vis_faligndata(TMP6, TMP8, REF_S2); yading@10: vis_faligndata(TMP8, TMP10, REF_S6); yading@10: } else { yading@10: vis_src1(TMP2, REF_2); yading@10: vis_src1(TMP4, REF_6); yading@10: vis_src1(TMP8, REF_S2); yading@10: vis_src1(TMP10, REF_S6); yading@10: } yading@10: yading@10: vis_mul8x16au(REF_0, CONST_256, TMP0); yading@10: vis_pmerge(ZERO, REF_0_1, TMP2); yading@10: yading@10: vis_mul8x16au(REF_2, CONST_256, TMP4); yading@10: vis_pmerge(ZERO, REF_2_1, TMP6); yading@10: yading@10: vis_padd16(TMP0, CONST_2, TMP8); yading@10: vis_mul8x16au(REF_4, CONST_256, TMP0); yading@10: yading@10: vis_padd16(TMP2, CONST_1, TMP10); yading@10: vis_mul8x16au(REF_4_1, CONST_256, TMP2); yading@10: yading@10: vis_padd16(TMP8, TMP4, TMP8); yading@10: vis_mul8x16au(REF_6, CONST_256, TMP4); yading@10: yading@10: vis_padd16(TMP10, TMP6, TMP10); yading@10: vis_mul8x16au(REF_6_1, CONST_256, TMP6); yading@10: yading@10: vis_padd16(TMP12, TMP8, TMP12); yading@10: yading@10: vis_padd16(TMP14, TMP10, TMP14); yading@10: yading@10: vis_padd16(TMP12, TMP16, TMP12); yading@10: yading@10: vis_padd16(TMP14, TMP18, TMP14); yading@10: vis_pack16(TMP12, DST_0); yading@10: yading@10: vis_pack16(TMP14, DST_1); yading@10: vis_st64(DST_0, dest[0]); yading@10: vis_padd16(TMP0, CONST_1, TMP12); yading@10: yading@10: vis_mul8x16au(REF_S0, CONST_256, TMP0); yading@10: vis_padd16(TMP2, CONST_1, TMP14); yading@10: yading@10: vis_mul8x16au(REF_S0_1, CONST_256, TMP2); yading@10: vis_padd16(TMP12, TMP4, TMP12); yading@10: yading@10: vis_mul8x16au(REF_S2, CONST_256, TMP4); yading@10: vis_padd16(TMP14, TMP6, TMP14); yading@10: yading@10: vis_mul8x16au(REF_S2_1, CONST_256, TMP6); yading@10: vis_padd16(TMP20, TMP12, TMP20); yading@10: yading@10: vis_padd16(TMP22, TMP14, TMP22); yading@10: yading@10: vis_padd16(TMP20, TMP24, TMP20); yading@10: yading@10: vis_padd16(TMP22, TMP26, TMP22); yading@10: vis_pack16(TMP20, DST_2); yading@10: yading@10: vis_pack16(TMP22, DST_3); yading@10: vis_st64_2(DST_2, dest, 8); yading@10: dest += stride; yading@10: vis_padd16(TMP0, TMP4, TMP24); yading@10: yading@10: vis_mul8x16au(REF_S4, CONST_256, TMP0); yading@10: vis_padd16(TMP2, TMP6, TMP26); yading@10: yading@10: vis_mul8x16au(REF_S4_1, CONST_256, TMP2); yading@10: vis_padd16(TMP24, TMP8, TMP24); yading@10: yading@10: vis_padd16(TMP26, TMP10, TMP26); yading@10: vis_pack16(TMP24, DST_0); yading@10: yading@10: vis_pack16(TMP26, DST_1); yading@10: vis_st64(DST_0, dest[0]); yading@10: vis_pmerge(ZERO, REF_S6, TMP4); yading@10: yading@10: vis_pmerge(ZERO, REF_S6_1, TMP6); yading@10: yading@10: vis_padd16(TMP0, TMP4, TMP0); yading@10: yading@10: vis_padd16(TMP2, TMP6, TMP2); yading@10: yading@10: vis_padd16(TMP0, TMP12, TMP0); yading@10: yading@10: vis_padd16(TMP2, TMP14, TMP2); yading@10: vis_pack16(TMP0, DST_2); yading@10: yading@10: vis_pack16(TMP2, DST_3); yading@10: vis_st64_2(DST_2, dest, 8); yading@10: dest += stride; yading@10: } while (--height); yading@10: } yading@10: yading@10: static void MC_put_no_round_xy_8_vis (uint8_t * dest, const uint8_t * ref, yading@10: const ptrdiff_t stride, int height) yading@10: { yading@10: unsigned long off = (unsigned long) ref & 0x7; yading@10: unsigned long off_plus_1 = off + 1; yading@10: int stride_8 = stride + 8; yading@10: yading@10: vis_set_gsr(5 << VIS_GSR_SCALEFACT_SHIFT); yading@10: yading@10: ref = vis_alignaddr(ref); yading@10: yading@10: vis_ld64(ref[ 0], TMP0); yading@10: vis_fzero(ZERO); yading@10: yading@10: vis_ld64(ref[ 8], TMP2); yading@10: yading@10: vis_ld64(constants1[0], CONST_1); yading@10: yading@10: vis_ld64(constants256_512[0], CONST_256); yading@10: vis_faligndata(TMP0, TMP2, REF_S0); yading@10: yading@10: if (off != 0x7) { yading@10: vis_alignaddr_g0((void *)off_plus_1); yading@10: vis_faligndata(TMP0, TMP2, REF_S2); yading@10: } else { yading@10: vis_src1(TMP2, REF_S2); yading@10: } yading@10: yading@10: height >>= 1; yading@10: do { /* 26 cycles */ yading@10: vis_ld64_2(ref, stride, TMP0); yading@10: vis_mul8x16au(REF_S0, CONST_256, TMP8); yading@10: vis_pmerge(ZERO, REF_S2, TMP12); yading@10: yading@10: vis_alignaddr_g0((void *)off); yading@10: yading@10: vis_ld64_2(ref, stride_8, TMP2); yading@10: ref += stride; yading@10: vis_mul8x16au(REF_S0_1, CONST_256, TMP10); yading@10: vis_pmerge(ZERO, REF_S2_1, TMP14); yading@10: yading@10: vis_ld64_2(ref, stride, TMP4); yading@10: yading@10: vis_ld64_2(ref, stride_8, TMP6); yading@10: ref += stride; yading@10: vis_faligndata(TMP0, TMP2, REF_S4); yading@10: yading@10: vis_pmerge(ZERO, REF_S4, TMP18); yading@10: yading@10: vis_pmerge(ZERO, REF_S4_1, TMP20); yading@10: yading@10: vis_faligndata(TMP4, TMP6, REF_S0); yading@10: yading@10: if (off != 0x7) { yading@10: vis_alignaddr_g0((void *)off_plus_1); yading@10: vis_faligndata(TMP0, TMP2, REF_S6); yading@10: vis_faligndata(TMP4, TMP6, REF_S2); yading@10: } else { yading@10: vis_src1(TMP2, REF_S6); yading@10: vis_src1(TMP6, REF_S2); yading@10: } yading@10: yading@10: vis_padd16(TMP18, CONST_1, TMP18); yading@10: vis_mul8x16au(REF_S6, CONST_256, TMP22); yading@10: yading@10: vis_padd16(TMP20, CONST_1, TMP20); yading@10: vis_mul8x16au(REF_S6_1, CONST_256, TMP24); yading@10: yading@10: vis_mul8x16au(REF_S0, CONST_256, TMP26); yading@10: vis_pmerge(ZERO, REF_S0_1, TMP28); yading@10: yading@10: vis_mul8x16au(REF_S2, CONST_256, TMP30); yading@10: vis_padd16(TMP18, TMP22, TMP18); yading@10: yading@10: vis_mul8x16au(REF_S2_1, CONST_256, TMP32); yading@10: vis_padd16(TMP20, TMP24, TMP20); yading@10: yading@10: vis_padd16(TMP8, TMP18, TMP8); yading@10: yading@10: vis_padd16(TMP10, TMP20, TMP10); yading@10: yading@10: vis_padd16(TMP8, TMP12, TMP8); yading@10: yading@10: vis_padd16(TMP10, TMP14, TMP10); yading@10: vis_pack16(TMP8, DST_0); yading@10: yading@10: vis_pack16(TMP10, DST_1); yading@10: vis_st64(DST_0, dest[0]); yading@10: dest += stride; yading@10: vis_padd16(TMP18, TMP26, TMP18); yading@10: yading@10: vis_padd16(TMP20, TMP28, TMP20); yading@10: yading@10: vis_padd16(TMP18, TMP30, TMP18); yading@10: yading@10: vis_padd16(TMP20, TMP32, TMP20); yading@10: vis_pack16(TMP18, DST_2); yading@10: yading@10: vis_pack16(TMP20, DST_3); yading@10: vis_st64(DST_2, dest[0]); yading@10: dest += stride; yading@10: } while (--height); yading@10: } yading@10: yading@10: static void MC_avg_no_round_xy_16_vis (uint8_t * dest, const uint8_t * ref, yading@10: const ptrdiff_t stride, int height) yading@10: { yading@10: unsigned long off = (unsigned long) ref & 0x7; yading@10: unsigned long off_plus_1 = off + 1; yading@10: int stride_8 = stride + 8; yading@10: int stride_16 = stride + 16; yading@10: yading@10: vis_set_gsr(4 << VIS_GSR_SCALEFACT_SHIFT); yading@10: yading@10: ref = vis_alignaddr(ref); yading@10: yading@10: vis_ld64(ref[ 0], TMP0); yading@10: vis_fzero(ZERO); yading@10: yading@10: vis_ld64(ref[ 8], TMP2); yading@10: yading@10: vis_ld64(ref[16], TMP4); yading@10: yading@10: vis_ld64(constants6[0], CONST_6); yading@10: vis_faligndata(TMP0, TMP2, REF_S0); yading@10: yading@10: vis_ld64(constants256_1024[0], CONST_256); yading@10: vis_faligndata(TMP2, TMP4, REF_S4); yading@10: yading@10: if (off != 0x7) { yading@10: vis_alignaddr_g0((void *)off_plus_1); yading@10: vis_faligndata(TMP0, TMP2, REF_S2); yading@10: vis_faligndata(TMP2, TMP4, REF_S6); yading@10: } else { yading@10: vis_src1(TMP2, REF_S2); yading@10: vis_src1(TMP4, REF_S6); yading@10: } yading@10: yading@10: height >>= 1; yading@10: do { /* 55 cycles */ yading@10: vis_ld64_2(ref, stride, TMP0); yading@10: vis_mul8x16au(REF_S0, CONST_256, TMP12); yading@10: vis_pmerge(ZERO, REF_S0_1, TMP14); yading@10: yading@10: vis_alignaddr_g0((void *)off); yading@10: yading@10: vis_ld64_2(ref, stride_8, TMP2); yading@10: vis_mul8x16au(REF_S2, CONST_256, TMP16); yading@10: vis_pmerge(ZERO, REF_S2_1, TMP18); yading@10: yading@10: vis_ld64_2(ref, stride_16, TMP4); yading@10: ref += stride; yading@10: vis_mul8x16au(REF_S4, CONST_256, TMP20); yading@10: vis_pmerge(ZERO, REF_S4_1, TMP22); yading@10: yading@10: vis_ld64_2(ref, stride, TMP6); yading@10: vis_mul8x16au(REF_S6, CONST_256, TMP24); yading@10: vis_pmerge(ZERO, REF_S6_1, TMP26); yading@10: yading@10: vis_ld64_2(ref, stride_8, TMP8); yading@10: vis_faligndata(TMP0, TMP2, REF_0); yading@10: yading@10: vis_ld64_2(ref, stride_16, TMP10); yading@10: ref += stride; yading@10: vis_faligndata(TMP2, TMP4, REF_4); yading@10: yading@10: vis_ld64(dest[0], DST_0); yading@10: vis_faligndata(TMP6, TMP8, REF_S0); yading@10: yading@10: vis_ld64_2(dest, 8, DST_2); yading@10: vis_faligndata(TMP8, TMP10, REF_S4); yading@10: yading@10: if (off != 0x7) { yading@10: vis_alignaddr_g0((void *)off_plus_1); yading@10: vis_faligndata(TMP0, TMP2, REF_2); yading@10: vis_faligndata(TMP2, TMP4, REF_6); yading@10: vis_faligndata(TMP6, TMP8, REF_S2); yading@10: vis_faligndata(TMP8, TMP10, REF_S6); yading@10: } else { yading@10: vis_src1(TMP2, REF_2); yading@10: vis_src1(TMP4, REF_6); yading@10: vis_src1(TMP8, REF_S2); yading@10: vis_src1(TMP10, REF_S6); yading@10: } yading@10: yading@10: vis_mul8x16al(DST_0, CONST_1024, TMP30); yading@10: vis_pmerge(ZERO, REF_0, TMP0); yading@10: yading@10: vis_mul8x16al(DST_1, CONST_1024, TMP32); yading@10: vis_pmerge(ZERO, REF_0_1, TMP2); yading@10: yading@10: vis_mul8x16au(REF_2, CONST_256, TMP4); yading@10: vis_pmerge(ZERO, REF_2_1, TMP6); yading@10: yading@10: vis_mul8x16al(DST_2, CONST_1024, REF_0); yading@10: vis_padd16(TMP0, CONST_6, TMP0); yading@10: yading@10: vis_mul8x16al(DST_3, CONST_1024, REF_2); yading@10: vis_padd16(TMP2, CONST_6, TMP2); yading@10: yading@10: vis_padd16(TMP0, TMP4, TMP0); yading@10: vis_mul8x16au(REF_4, CONST_256, TMP4); yading@10: yading@10: vis_padd16(TMP2, TMP6, TMP2); yading@10: vis_mul8x16au(REF_4_1, CONST_256, TMP6); yading@10: yading@10: vis_padd16(TMP12, TMP0, TMP12); yading@10: vis_mul8x16au(REF_6, CONST_256, TMP8); yading@10: yading@10: vis_padd16(TMP14, TMP2, TMP14); yading@10: vis_mul8x16au(REF_6_1, CONST_256, TMP10); yading@10: yading@10: vis_padd16(TMP12, TMP16, TMP12); yading@10: vis_mul8x16au(REF_S0, CONST_256, REF_4); yading@10: yading@10: vis_padd16(TMP14, TMP18, TMP14); yading@10: vis_mul8x16au(REF_S0_1, CONST_256, REF_6); yading@10: yading@10: vis_padd16(TMP12, TMP30, TMP12); yading@10: yading@10: vis_padd16(TMP14, TMP32, TMP14); yading@10: vis_pack16(TMP12, DST_0); yading@10: yading@10: vis_pack16(TMP14, DST_1); yading@10: vis_st64(DST_0, dest[0]); yading@10: vis_padd16(TMP4, CONST_6, TMP4); yading@10: yading@10: vis_ld64_2(dest, stride, DST_0); yading@10: vis_padd16(TMP6, CONST_6, TMP6); yading@10: vis_mul8x16au(REF_S2, CONST_256, TMP12); yading@10: yading@10: vis_padd16(TMP4, TMP8, TMP4); yading@10: vis_mul8x16au(REF_S2_1, CONST_256, TMP14); yading@10: yading@10: vis_padd16(TMP6, TMP10, TMP6); yading@10: yading@10: vis_padd16(TMP20, TMP4, TMP20); yading@10: yading@10: vis_padd16(TMP22, TMP6, TMP22); yading@10: yading@10: vis_padd16(TMP20, TMP24, TMP20); yading@10: yading@10: vis_padd16(TMP22, TMP26, TMP22); yading@10: yading@10: vis_padd16(TMP20, REF_0, TMP20); yading@10: vis_mul8x16au(REF_S4, CONST_256, REF_0); yading@10: yading@10: vis_padd16(TMP22, REF_2, TMP22); yading@10: vis_pack16(TMP20, DST_2); yading@10: yading@10: vis_pack16(TMP22, DST_3); yading@10: vis_st64_2(DST_2, dest, 8); yading@10: dest += stride; yading@10: yading@10: vis_ld64_2(dest, 8, DST_2); yading@10: vis_mul8x16al(DST_0, CONST_1024, TMP30); yading@10: vis_pmerge(ZERO, REF_S4_1, REF_2); yading@10: yading@10: vis_mul8x16al(DST_1, CONST_1024, TMP32); yading@10: vis_padd16(REF_4, TMP0, TMP8); yading@10: yading@10: vis_mul8x16au(REF_S6, CONST_256, REF_4); yading@10: vis_padd16(REF_6, TMP2, TMP10); yading@10: yading@10: vis_mul8x16au(REF_S6_1, CONST_256, REF_6); yading@10: vis_padd16(TMP8, TMP12, TMP8); yading@10: yading@10: vis_padd16(TMP10, TMP14, TMP10); yading@10: yading@10: vis_padd16(TMP8, TMP30, TMP8); yading@10: yading@10: vis_padd16(TMP10, TMP32, TMP10); yading@10: vis_pack16(TMP8, DST_0); yading@10: yading@10: vis_pack16(TMP10, DST_1); yading@10: vis_st64(DST_0, dest[0]); yading@10: yading@10: vis_padd16(REF_0, TMP4, REF_0); yading@10: yading@10: vis_mul8x16al(DST_2, CONST_1024, TMP30); yading@10: vis_padd16(REF_2, TMP6, REF_2); yading@10: yading@10: vis_mul8x16al(DST_3, CONST_1024, TMP32); yading@10: vis_padd16(REF_0, REF_4, REF_0); yading@10: yading@10: vis_padd16(REF_2, REF_6, REF_2); yading@10: yading@10: vis_padd16(REF_0, TMP30, REF_0); yading@10: yading@10: /* stall */ yading@10: yading@10: vis_padd16(REF_2, TMP32, REF_2); yading@10: vis_pack16(REF_0, DST_2); yading@10: yading@10: vis_pack16(REF_2, DST_3); yading@10: vis_st64_2(DST_2, dest, 8); yading@10: dest += stride; yading@10: } while (--height); yading@10: } yading@10: yading@10: /* End of no rounding code */ yading@10: yading@10: av_cold void ff_hpeldsp_init_vis(HpelDSPContext *c, int flags) yading@10: { yading@10: /* VIS-specific optimizations */ yading@10: int accel = vis_level (); yading@10: yading@10: if (accel & ACCEL_SPARC_VIS) { yading@10: c->put_pixels_tab[0][0] = MC_put_o_16_vis; yading@10: c->put_pixels_tab[0][1] = MC_put_x_16_vis; yading@10: c->put_pixels_tab[0][2] = MC_put_y_16_vis; yading@10: c->put_pixels_tab[0][3] = MC_put_xy_16_vis; yading@10: yading@10: c->put_pixels_tab[1][0] = MC_put_o_8_vis; yading@10: c->put_pixels_tab[1][1] = MC_put_x_8_vis; yading@10: c->put_pixels_tab[1][2] = MC_put_y_8_vis; yading@10: c->put_pixels_tab[1][3] = MC_put_xy_8_vis; yading@10: yading@10: c->avg_pixels_tab[0][0] = MC_avg_o_16_vis; yading@10: c->avg_pixels_tab[0][1] = MC_avg_x_16_vis; yading@10: c->avg_pixels_tab[0][2] = MC_avg_y_16_vis; yading@10: c->avg_pixels_tab[0][3] = MC_avg_xy_16_vis; yading@10: yading@10: c->avg_pixels_tab[1][0] = MC_avg_o_8_vis; yading@10: c->avg_pixels_tab[1][1] = MC_avg_x_8_vis; yading@10: c->avg_pixels_tab[1][2] = MC_avg_y_8_vis; yading@10: c->avg_pixels_tab[1][3] = MC_avg_xy_8_vis; yading@10: yading@10: c->put_no_rnd_pixels_tab[0][0] = MC_put_no_round_o_16_vis; yading@10: c->put_no_rnd_pixels_tab[0][1] = MC_put_no_round_x_16_vis; yading@10: c->put_no_rnd_pixels_tab[0][2] = MC_put_no_round_y_16_vis; yading@10: c->put_no_rnd_pixels_tab[0][3] = MC_put_no_round_xy_16_vis; yading@10: yading@10: c->put_no_rnd_pixels_tab[1][0] = MC_put_no_round_o_8_vis; yading@10: c->put_no_rnd_pixels_tab[1][1] = MC_put_no_round_x_8_vis; yading@10: c->put_no_rnd_pixels_tab[1][2] = MC_put_no_round_y_8_vis; yading@10: c->put_no_rnd_pixels_tab[1][3] = MC_put_no_round_xy_8_vis; yading@10: yading@10: c->avg_no_rnd_pixels_tab[0] = MC_avg_no_round_o_16_vis; yading@10: c->avg_no_rnd_pixels_tab[1] = MC_avg_no_round_x_16_vis; yading@10: c->avg_no_rnd_pixels_tab[2] = MC_avg_no_round_y_16_vis; yading@10: c->avg_no_rnd_pixels_tab[3] = MC_avg_no_round_xy_16_vis; yading@10: } yading@10: }