diff options
Diffstat (limited to 'contrib/ffmpeg/libavcodec/sparc')
| -rw-r--r-- | contrib/ffmpeg/libavcodec/sparc/dsputil_vis.c | 4053 | ||||
| -rw-r--r-- | contrib/ffmpeg/libavcodec/sparc/simple_idct_vis.c | 528 | ||||
| -rw-r--r-- | contrib/ffmpeg/libavcodec/sparc/vis.h | 332 |
3 files changed, 4913 insertions, 0 deletions
diff --git a/contrib/ffmpeg/libavcodec/sparc/dsputil_vis.c b/contrib/ffmpeg/libavcodec/sparc/dsputil_vis.c new file mode 100644 index 000000000..437260311 --- /dev/null +++ b/contrib/ffmpeg/libavcodec/sparc/dsputil_vis.c @@ -0,0 +1,4053 @@ +/* + * dsputil_vis.c + * Copyright (C) 2003 David S. Miller <davem@redhat.com> + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +/* The *no_round* functions have been added by James A. Morrison, 2003,2004. + The vis code from libmpeg2 was adapted for ffmpeg by James A. Morrison. + */ + +#include "config.h" + +#include <inttypes.h> + +#include "dsputil.h" + +#include "vis.h" + +extern void ff_simple_idct_put_vis(uint8_t *dest, int line_size, DCTELEM *data); +extern void ff_simple_idct_add_vis(uint8_t *dest, int line_size, DCTELEM *data); +extern void ff_simple_idct_vis(DCTELEM *data); + +/* The trick used in some of this file is the formula from the MMX + * motion comp code, which is: + * + * (x+y+1)>>1 == (x|y)-((x^y)>>1) + * + * This allows us to average 8 bytes at a time in a 64-bit FPU reg. + * We avoid overflows by masking before we do the shift, and we + * implement the shift by multiplying by 1/2 using mul8x16. So in + * VIS this is (assume 'x' is in f0, 'y' is in f2, a repeating mask + * of '0xfe' is in f4, a repeating mask of '0x7f' is in f6, and + * the value 0x80808080 is in f8): + * + * fxor f0, f2, f10 + * fand f10, f4, f10 + * fmul8x16 f8, f10, f10 + * fand f10, f6, f10 + * for f0, f2, f12 + * fpsub16 f12, f10, f10 + */ + +#define ATTR_ALIGN(alignd) __attribute__ ((aligned(alignd))) + +#define DUP4(x) {x, x, x, x} +#define DUP8(x) {x, x, x, x, x, x, x, x} +static const int16_t constants1[] ATTR_ALIGN(8) = DUP4 (1); +static const int16_t constants2[] ATTR_ALIGN(8) = DUP4 (2); +static const int16_t constants3[] ATTR_ALIGN(8) = DUP4 (3); +static const int16_t constants6[] ATTR_ALIGN(8) = DUP4 (6); +static const int8_t constants_fe[] ATTR_ALIGN(8) = DUP8 (0xfe); +static const int8_t constants_7f[] ATTR_ALIGN(8) = DUP8 (0x7f); +static const int8_t constants128[] ATTR_ALIGN(8) = DUP8 (128); +static const int16_t constants256_512[] ATTR_ALIGN(8) = + {256, 512, 256, 512}; +static const int16_t constants256_1024[] ATTR_ALIGN(8) = + {256, 1024, 256, 1024}; + +#define REF_0 0 +#define REF_0_1 1 +#define REF_2 2 +#define REF_2_1 3 +#define REF_4 4 +#define REF_4_1 5 +#define REF_6 6 +#define REF_6_1 7 +#define REF_S0 8 +#define REF_S0_1 9 +#define REF_S2 10 +#define REF_S2_1 11 +#define REF_S4 12 +#define REF_S4_1 13 +#define REF_S6 14 +#define REF_S6_1 15 +#define DST_0 16 +#define DST_1 17 +#define DST_2 18 +#define DST_3 19 +#define CONST_1 20 +#define CONST_2 20 +#define CONST_3 20 +#define CONST_6 20 +#define MASK_fe 20 +#define CONST_128 22 +#define CONST_256 22 +#define CONST_512 22 +#define CONST_1024 22 +#define TMP0 24 +#define TMP1 25 +#define TMP2 26 +#define TMP3 27 +#define TMP4 28 +#define TMP5 29 +#define ZERO 30 +#define MASK_7f 30 + +#define TMP6 32 +#define TMP8 34 +#define TMP10 36 +#define TMP12 38 +#define TMP14 40 +#define TMP16 42 +#define TMP18 44 +#define TMP20 46 +#define TMP22 48 +#define TMP24 50 +#define TMP26 52 +#define TMP28 54 +#define TMP30 56 +#define TMP32 58 + +static void MC_put_o_16_vis (uint8_t * dest, const uint8_t * _ref, + const int stride, int height) +{ + uint8_t *ref = (uint8_t *) _ref; + + ref = vis_alignaddr(ref); + do { /* 5 cycles */ + vis_ld64(ref[0], TMP0); + + vis_ld64_2(ref, 8, TMP2); + + vis_ld64_2(ref, 16, TMP4); + ref += stride; + + vis_faligndata(TMP0, TMP2, REF_0); + vis_st64(REF_0, dest[0]); + + vis_faligndata(TMP2, TMP4, REF_2); + vis_st64_2(REF_2, dest, 8); + dest += stride; + } while (--height); +} + +static void MC_put_o_8_vis (uint8_t * dest, const uint8_t * _ref, + const int stride, int height) +{ + uint8_t *ref = (uint8_t *) _ref; + + ref = vis_alignaddr(ref); + do { /* 4 cycles */ + vis_ld64(ref[0], TMP0); + + vis_ld64(ref[8], TMP2); + ref += stride; + + /* stall */ + + vis_faligndata(TMP0, TMP2, REF_0); + vis_st64(REF_0, dest[0]); + dest += stride; + } while (--height); +} + + +static void MC_avg_o_16_vis (uint8_t * dest, const uint8_t * _ref, + const int stride, int height) +{ + uint8_t *ref = (uint8_t *) _ref; + int stride_8 = stride + 8; + + ref = vis_alignaddr(ref); + + vis_ld64(ref[0], TMP0); + + vis_ld64(ref[8], TMP2); + + vis_ld64(ref[16], TMP4); + + vis_ld64(dest[0], DST_0); + + vis_ld64(dest[8], DST_2); + + vis_ld64(constants_fe[0], MASK_fe); + vis_faligndata(TMP0, TMP2, REF_0); + + vis_ld64(constants_7f[0], MASK_7f); + vis_faligndata(TMP2, TMP4, REF_2); + + vis_ld64(constants128[0], CONST_128); + + ref += stride; + height = (height >> 1) - 1; + + do { /* 24 cycles */ + vis_ld64(ref[0], TMP0); + vis_xor(DST_0, REF_0, TMP6); + + vis_ld64_2(ref, 8, TMP2); + vis_and(TMP6, MASK_fe, TMP6); + + vis_ld64_2(ref, 16, TMP4); + ref += stride; + vis_mul8x16(CONST_128, TMP6, TMP6); + vis_xor(DST_2, REF_2, TMP8); + + vis_and(TMP8, MASK_fe, TMP8); + + vis_or(DST_0, REF_0, TMP10); + vis_ld64_2(dest, stride, DST_0); + vis_mul8x16(CONST_128, TMP8, TMP8); + + vis_or(DST_2, REF_2, TMP12); + vis_ld64_2(dest, stride_8, DST_2); + + vis_ld64(ref[0], TMP14); + vis_and(TMP6, MASK_7f, TMP6); + + vis_and(TMP8, MASK_7f, TMP8); + + vis_psub16(TMP10, TMP6, TMP6); + vis_st64(TMP6, dest[0]); + + vis_psub16(TMP12, TMP8, TMP8); + vis_st64_2(TMP8, dest, 8); + + dest += stride; + vis_ld64_2(ref, 8, TMP16); + vis_faligndata(TMP0, TMP2, REF_0); + + vis_ld64_2(ref, 16, TMP18); + vis_faligndata(TMP2, TMP4, REF_2); + ref += stride; + + vis_xor(DST_0, REF_0, TMP20); + + vis_and(TMP20, MASK_fe, TMP20); + + vis_xor(DST_2, REF_2, TMP22); + vis_mul8x16(CONST_128, TMP20, TMP20); + + vis_and(TMP22, MASK_fe, TMP22); + + vis_or(DST_0, REF_0, TMP24); + vis_mul8x16(CONST_128, TMP22, TMP22); + + vis_or(DST_2, REF_2, TMP26); + + vis_ld64_2(dest, stride, DST_0); + vis_faligndata(TMP14, TMP16, REF_0); + + vis_ld64_2(dest, stride_8, DST_2); + vis_faligndata(TMP16, TMP18, REF_2); + + vis_and(TMP20, MASK_7f, TMP20); + + vis_and(TMP22, MASK_7f, TMP22); + + vis_psub16(TMP24, TMP20, TMP20); + vis_st64(TMP20, dest[0]); + + vis_psub16(TMP26, TMP22, TMP22); + vis_st64_2(TMP22, dest, 8); + dest += stride; + } while (--height); + + vis_ld64(ref[0], TMP0); + vis_xor(DST_0, REF_0, TMP6); + + vis_ld64_2(ref, 8, TMP2); + vis_and(TMP6, MASK_fe, TMP6); + + vis_ld64_2(ref, 16, TMP4); + vis_mul8x16(CONST_128, TMP6, TMP6); + vis_xor(DST_2, REF_2, TMP8); + + vis_and(TMP8, MASK_fe, TMP8); + + vis_or(DST_0, REF_0, TMP10); + vis_ld64_2(dest, stride, DST_0); + vis_mul8x16(CONST_128, TMP8, TMP8); + + vis_or(DST_2, REF_2, TMP12); + vis_ld64_2(dest, stride_8, DST_2); + + vis_ld64(ref[0], TMP14); + vis_and(TMP6, MASK_7f, TMP6); + + vis_and(TMP8, MASK_7f, TMP8); + + vis_psub16(TMP10, TMP6, TMP6); + vis_st64(TMP6, dest[0]); + + vis_psub16(TMP12, TMP8, TMP8); + vis_st64_2(TMP8, dest, 8); + + dest += stride; + vis_faligndata(TMP0, TMP2, REF_0); + + vis_faligndata(TMP2, TMP4, REF_2); + + vis_xor(DST_0, REF_0, TMP20); + + vis_and(TMP20, MASK_fe, TMP20); + + vis_xor(DST_2, REF_2, TMP22); + vis_mul8x16(CONST_128, TMP20, TMP20); + + vis_and(TMP22, MASK_fe, TMP22); + + vis_or(DST_0, REF_0, TMP24); + vis_mul8x16(CONST_128, TMP22, TMP22); + + vis_or(DST_2, REF_2, TMP26); + + vis_and(TMP20, MASK_7f, TMP20); + + vis_and(TMP22, MASK_7f, TMP22); + + vis_psub16(TMP24, TMP20, TMP20); + vis_st64(TMP20, dest[0]); + + vis_psub16(TMP26, TMP22, TMP22); + vis_st64_2(TMP22, dest, 8); +} + +static void MC_avg_o_8_vis (uint8_t * dest, const uint8_t * _ref, + const int stride, int height) +{ + uint8_t *ref = (uint8_t *) _ref; + + ref = vis_alignaddr(ref); + + vis_ld64(ref[0], TMP0); + + vis_ld64(ref[8], TMP2); + + vis_ld64(dest[0], DST_0); + + vis_ld64(constants_fe[0], MASK_fe); + + vis_ld64(constants_7f[0], MASK_7f); + vis_faligndata(TMP0, TMP2, REF_0); + + vis_ld64(constants128[0], CONST_128); + + ref += stride; + height = (height >> 1) - 1; + + do { /* 12 cycles */ + vis_ld64(ref[0], TMP0); + vis_xor(DST_0, REF_0, TMP4); + + vis_ld64(ref[8], TMP2); + vis_and(TMP4, MASK_fe, TMP4); + + vis_or(DST_0, REF_0, TMP6); + vis_ld64_2(dest, stride, DST_0); + ref += stride; + vis_mul8x16(CONST_128, TMP4, TMP4); + + vis_ld64(ref[0], TMP12); + vis_faligndata(TMP0, TMP2, REF_0); + + vis_ld64(ref[8], TMP2); + vis_xor(DST_0, REF_0, TMP0); + ref += stride; + + vis_and(TMP0, MASK_fe, TMP0); + + vis_and(TMP4, MASK_7f, TMP4); + + vis_psub16(TMP6, TMP4, TMP4); + vis_st64(TMP4, dest[0]); + dest += stride; + vis_mul8x16(CONST_128, TMP0, TMP0); + + vis_or(DST_0, REF_0, TMP6); + vis_ld64_2(dest, stride, DST_0); + + vis_faligndata(TMP12, TMP2, REF_0); + + vis_and(TMP0, MASK_7f, TMP0); + + vis_psub16(TMP6, TMP0, TMP4); + vis_st64(TMP4, dest[0]); + dest += stride; + } while (--height); + + vis_ld64(ref[0], TMP0); + vis_xor(DST_0, REF_0, TMP4); + + vis_ld64(ref[8], TMP2); + vis_and(TMP4, MASK_fe, TMP4); + + vis_or(DST_0, REF_0, TMP6); + vis_ld64_2(dest, stride, DST_0); + vis_mul8x16(CONST_128, TMP4, TMP4); + + vis_faligndata(TMP0, TMP2, REF_0); + + vis_xor(DST_0, REF_0, TMP0); + + vis_and(TMP0, MASK_fe, TMP0); + + vis_and(TMP4, MASK_7f, TMP4); + + vis_psub16(TMP6, TMP4, TMP4); + vis_st64(TMP4, dest[0]); + dest += stride; + vis_mul8x16(CONST_128, TMP0, TMP0); + + vis_or(DST_0, REF_0, TMP6); + + vis_and(TMP0, MASK_7f, TMP0); + + vis_psub16(TMP6, TMP0, TMP4); + vis_st64(TMP4, dest[0]); +} + +static void MC_put_x_16_vis (uint8_t * dest, const uint8_t * _ref, + const int stride, int height) +{ + uint8_t *ref = (uint8_t *) _ref; + unsigned long off = (unsigned long) ref & 0x7; + unsigned long off_plus_1 = off + 1; + + ref = vis_alignaddr(ref); + + vis_ld64(ref[0], TMP0); + + vis_ld64_2(ref, 8, TMP2); + + vis_ld64_2(ref, 16, TMP4); + + vis_ld64(constants_fe[0], MASK_fe); + + vis_ld64(constants_7f[0], MASK_7f); + vis_faligndata(TMP0, TMP2, REF_0); + + vis_ld64(constants128[0], CONST_128); + vis_faligndata(TMP2, TMP4, REF_4); + + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + vis_faligndata(TMP0, TMP2, REF_2); + vis_faligndata(TMP2, TMP4, REF_6); + } else { + vis_src1(TMP2, REF_2); + vis_src1(TMP4, REF_6); + } + + ref += stride; + height = (height >> 1) - 1; + + do { /* 34 cycles */ + vis_ld64(ref[0], TMP0); + vis_xor(REF_0, REF_2, TMP6); + + vis_ld64_2(ref, 8, TMP2); + vis_xor(REF_4, REF_6, TMP8); + + vis_ld64_2(ref, 16, TMP4); + vis_and(TMP6, MASK_fe, TMP6); + ref += stride; + + vis_ld64(ref[0], TMP14); + vis_mul8x16(CONST_128, TMP6, TMP6); + vis_and(TMP8, MASK_fe, TMP8); + + vis_ld64_2(ref, 8, TMP16); + vis_mul8x16(CONST_128, TMP8, TMP8); + vis_or(REF_0, REF_2, TMP10); + + vis_ld64_2(ref, 16, TMP18); + ref += stride; + vis_or(REF_4, REF_6, TMP12); + + vis_alignaddr_g0((void *)off); + + vis_faligndata(TMP0, TMP2, REF_0); + + vis_faligndata(TMP2, TMP4, REF_4); + + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + vis_faligndata(TMP0, TMP2, REF_2); + vis_faligndata(TMP2, TMP4, REF_6); + } else { + vis_src1(TMP2, REF_2); + vis_src1(TMP4, REF_6); + } + + vis_and(TMP6, MASK_7f, TMP6); + + vis_and(TMP8, MASK_7f, TMP8); + + vis_psub16(TMP10, TMP6, TMP6); + vis_st64(TMP6, dest[0]); + + vis_psub16(TMP12, TMP8, TMP8); + vis_st64_2(TMP8, dest, 8); + dest += stride; + + vis_xor(REF_0, REF_2, TMP6); + + vis_xor(REF_4, REF_6, TMP8); + + vis_and(TMP6, MASK_fe, TMP6); + + vis_mul8x16(CONST_128, TMP6, TMP6); + vis_and(TMP8, MASK_fe, TMP8); + + vis_mul8x16(CONST_128, TMP8, TMP8); + vis_or(REF_0, REF_2, TMP10); + + vis_or(REF_4, REF_6, TMP12); + + vis_alignaddr_g0((void *)off); + + vis_faligndata(TMP14, TMP16, REF_0); + + vis_faligndata(TMP16, TMP18, REF_4); + + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + vis_faligndata(TMP14, TMP16, REF_2); + vis_faligndata(TMP16, TMP18, REF_6); + } else { + vis_src1(TMP16, REF_2); + vis_src1(TMP18, REF_6); + } + + vis_and(TMP6, MASK_7f, TMP6); + + vis_and(TMP8, MASK_7f, TMP8); + + vis_psub16(TMP10, TMP6, TMP6); + vis_st64(TMP6, dest[0]); + + vis_psub16(TMP12, TMP8, TMP8); + vis_st64_2(TMP8, dest, 8); + dest += stride; + } while (--height); + + vis_ld64(ref[0], TMP0); + vis_xor(REF_0, REF_2, TMP6); + + vis_ld64_2(ref, 8, TMP2); + vis_xor(REF_4, REF_6, TMP8); + + vis_ld64_2(ref, 16, TMP4); + vis_and(TMP6, MASK_fe, TMP6); + + vis_mul8x16(CONST_128, TMP6, TMP6); + vis_and(TMP8, MASK_fe, TMP8); + + vis_mul8x16(CONST_128, TMP8, TMP8); + vis_or(REF_0, REF_2, TMP10); + + vis_or(REF_4, REF_6, TMP12); + + vis_alignaddr_g0((void *)off); + + vis_faligndata(TMP0, TMP2, REF_0); + + vis_faligndata(TMP2, TMP4, REF_4); + + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + vis_faligndata(TMP0, TMP2, REF_2); + vis_faligndata(TMP2, TMP4, REF_6); + } else { + vis_src1(TMP2, REF_2); + vis_src1(TMP4, REF_6); + } + + vis_and(TMP6, MASK_7f, TMP6); + + vis_and(TMP8, MASK_7f, TMP8); + + vis_psub16(TMP10, TMP6, TMP6); + vis_st64(TMP6, dest[0]); + + vis_psub16(TMP12, TMP8, TMP8); + vis_st64_2(TMP8, dest, 8); + dest += stride; + + vis_xor(REF_0, REF_2, TMP6); + + vis_xor(REF_4, REF_6, TMP8); + + vis_and(TMP6, MASK_fe, TMP6); + + vis_mul8x16(CONST_128, TMP6, TMP6); + vis_and(TMP8, MASK_fe, TMP8); + + vis_mul8x16(CONST_128, TMP8, TMP8); + vis_or(REF_0, REF_2, TMP10); + + vis_or(REF_4, REF_6, TMP12); + + vis_and(TMP6, MASK_7f, TMP6); + + vis_and(TMP8, MASK_7f, TMP8); + + vis_psub16(TMP10, TMP6, TMP6); + vis_st64(TMP6, dest[0]); + + vis_psub16(TMP12, TMP8, TMP8); + vis_st64_2(TMP8, dest, 8); +} + +static void MC_put_x_8_vis (uint8_t * dest, const uint8_t * _ref, + const int stride, int height) +{ + uint8_t *ref = (uint8_t *) _ref; + unsigned long off = (unsigned long) ref & 0x7; + unsigned long off_plus_1 = off + 1; + + ref = vis_alignaddr(ref); + + vis_ld64(ref[0], TMP0); + + vis_ld64(ref[8], TMP2); + + vis_ld64(constants_fe[0], MASK_fe); + + vis_ld64(constants_7f[0], MASK_7f); + + vis_ld64(constants128[0], CONST_128); + vis_faligndata(TMP0, TMP2, REF_0); + + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + vis_faligndata(TMP0, TMP2, REF_2); + } else { + vis_src1(TMP2, REF_2); + } + + ref += stride; + height = (height >> 1) - 1; + + do { /* 20 cycles */ + vis_ld64(ref[0], TMP0); + vis_xor(REF_0, REF_2, TMP4); + + vis_ld64_2(ref, 8, TMP2); + vis_and(TMP4, MASK_fe, TMP4); + ref += stride; + + vis_ld64(ref[0], TMP8); + vis_or(REF_0, REF_2, TMP6); + vis_mul8x16(CONST_128, TMP4, TMP4); + + vis_alignaddr_g0((void *)off); + + vis_ld64_2(ref, 8, TMP10); + ref += stride; + vis_faligndata(TMP0, TMP2, REF_0); + + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + vis_faligndata(TMP0, TMP2, REF_2); + } else { + vis_src1(TMP2, REF_2); + } + + vis_and(TMP4, MASK_7f, TMP4); + + vis_psub16(TMP6, TMP4, DST_0); + vis_st64(DST_0, dest[0]); + dest += stride; + + vis_xor(REF_0, REF_2, TMP12); + + vis_and(TMP12, MASK_fe, TMP12); + + vis_or(REF_0, REF_2, TMP14); + vis_mul8x16(CONST_128, TMP12, TMP12); + + vis_alignaddr_g0((void *)off); + vis_faligndata(TMP8, TMP10, REF_0); + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + vis_faligndata(TMP8, TMP10, REF_2); + } else { + vis_src1(TMP10, REF_2); + } + + vis_and(TMP12, MASK_7f, TMP12); + + vis_psub16(TMP14, TMP12, DST_0); + vis_st64(DST_0, dest[0]); + dest += stride; + } while (--height); + + vis_ld64(ref[0], TMP0); + vis_xor(REF_0, REF_2, TMP4); + + vis_ld64_2(ref, 8, TMP2); + vis_and(TMP4, MASK_fe, TMP4); + + vis_or(REF_0, REF_2, TMP6); + vis_mul8x16(CONST_128, TMP4, TMP4); + + vis_alignaddr_g0((void *)off); + + vis_faligndata(TMP0, TMP2, REF_0); + + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + vis_faligndata(TMP0, TMP2, REF_2); + } else { + vis_src1(TMP2, REF_2); + } + + vis_and(TMP4, MASK_7f, TMP4); + + vis_psub16(TMP6, TMP4, DST_0); + vis_st64(DST_0, dest[0]); + dest += stride; + + vis_xor(REF_0, REF_2, TMP12); + + vis_and(TMP12, MASK_fe, TMP12); + + vis_or(REF_0, REF_2, TMP14); + vis_mul8x16(CONST_128, TMP12, TMP12); + + vis_and(TMP12, MASK_7f, TMP12); + + vis_psub16(TMP14, TMP12, DST_0); + vis_st64(DST_0, dest[0]); + dest += stride; +} + +static void MC_avg_x_16_vis (uint8_t * dest, const uint8_t * _ref, + const int stride, int height) +{ + uint8_t *ref = (uint8_t *) _ref; + unsigned long off = (unsigned long) ref & 0x7; + unsigned long off_plus_1 = off + 1; + + vis_set_gsr(5 << VIS_GSR_SCALEFACT_SHIFT); + + vis_ld64(constants3[0], CONST_3); + vis_fzero(ZERO); + vis_ld64(constants256_512[0], CONST_256); + + ref = vis_alignaddr(ref); + do { /* 26 cycles */ + vis_ld64(ref[0], TMP0); + + vis_ld64(ref[8], TMP2); + + vis_alignaddr_g0((void *)off); + + vis_ld64(ref[16], TMP4); + + vis_ld64(dest[0], DST_0); + vis_faligndata(TMP0, TMP2, REF_0); + + vis_ld64(dest[8], DST_2); + vis_faligndata(TMP2, TMP4, REF_4); + + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + vis_faligndata(TMP0, TMP2, REF_2); + vis_faligndata(TMP2, TMP4, REF_6); + } else { + vis_src1(TMP2, REF_2); + vis_src1(TMP4, REF_6); + } + + vis_mul8x16au(REF_0, CONST_256, TMP0); + + vis_pmerge(ZERO, REF_2, TMP4); + vis_mul8x16au(REF_0_1, CONST_256, TMP2); + + vis_pmerge(ZERO, REF_2_1, TMP6); + + vis_padd16(TMP0, TMP4, TMP0); + + vis_mul8x16al(DST_0, CONST_512, TMP4); + vis_padd16(TMP2, TMP6, TMP2); + + vis_mul8x16al(DST_1, CONST_512, TMP6); + + vis_mul8x16au(REF_6, CONST_256, TMP12); + + vis_padd16(TMP0, TMP4, TMP0); + vis_mul8x16au(REF_6_1, CONST_256, TMP14); + + vis_padd16(TMP2, TMP6, TMP2); + vis_mul8x16au(REF_4, CONST_256, TMP16); + + vis_padd16(TMP0, CONST_3, TMP8); + vis_mul8x16au(REF_4_1, CONST_256, TMP18); + + vis_padd16(TMP2, CONST_3, TMP10); + vis_pack16(TMP8, DST_0); + + vis_pack16(TMP10, DST_1); + vis_padd16(TMP16, TMP12, TMP0); + + vis_st64(DST_0, dest[0]); + vis_mul8x16al(DST_2, CONST_512, TMP4); + vis_padd16(TMP18, TMP14, TMP2); + + vis_mul8x16al(DST_3, CONST_512, TMP6); + vis_padd16(TMP0, CONST_3, TMP0); + + vis_padd16(TMP2, CONST_3, TMP2); + + vis_padd16(TMP0, TMP4, TMP0); + + vis_padd16(TMP2, TMP6, TMP2); + vis_pack16(TMP0, DST_2); + + vis_pack16(TMP2, DST_3); + vis_st64(DST_2, dest[8]); + + ref += stride; + dest += stride; + } while (--height); +} + +static void MC_avg_x_8_vis (uint8_t * dest, const uint8_t * _ref, + const int stride, int height) +{ + uint8_t *ref = (uint8_t *) _ref; + unsigned long off = (unsigned long) ref & 0x7; + unsigned long off_plus_1 = off + 1; + int stride_times_2 = stride << 1; + + vis_set_gsr(5 << VIS_GSR_SCALEFACT_SHIFT); + + vis_ld64(constants3[0], CONST_3); + vis_fzero(ZERO); + vis_ld64(constants256_512[0], CONST_256); + + ref = vis_alignaddr(ref); + height >>= 2; + do { /* 47 cycles */ + vis_ld64(ref[0], TMP0); + + vis_ld64_2(ref, 8, TMP2); + ref += stride; + + vis_alignaddr_g0((void *)off); + + vis_ld64(ref[0], TMP4); + vis_faligndata(TMP0, TMP2, REF_0); + + vis_ld64_2(ref, 8, TMP6); + ref += stride; + + vis_ld64(ref[0], TMP8); + + vis_ld64_2(ref, 8, TMP10); + ref += stride; + vis_faligndata(TMP4, TMP6, REF_4); + + vis_ld64(ref[0], TMP12); + + vis_ld64_2(ref, 8, TMP14); + ref += stride; + vis_faligndata(TMP8, TMP10, REF_S0); + + vis_faligndata(TMP12, TMP14, REF_S4); + + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + + vis_ld64(dest[0], DST_0); + vis_faligndata(TMP0, TMP2, REF_2); + + vis_ld64_2(dest, stride, DST_2); + vis_faligndata(TMP4, TMP6, REF_6); + + vis_faligndata(TMP8, TMP10, REF_S2); + + vis_faligndata(TMP12, TMP14, REF_S6); + } else { + vis_ld64(dest[0], DST_0); + vis_src1(TMP2, REF_2); + + vis_ld64_2(dest, stride, DST_2); + vis_src1(TMP6, REF_6); + + vis_src1(TMP10, REF_S2); + + vis_src1(TMP14, REF_S6); + } + + vis_pmerge(ZERO, REF_0, TMP0); + vis_mul8x16au(REF_0_1, CONST_256, TMP2); + + vis_pmerge(ZERO, REF_2, TMP4); + vis_mul8x16au(REF_2_1, CONST_256, TMP6); + + vis_padd16(TMP0, CONST_3, TMP0); + vis_mul8x16al(DST_0, CONST_512, TMP16); + + vis_padd16(TMP2, CONST_3, TMP2); + vis_mul8x16al(DST_1, CONST_512, TMP18); + + vis_padd16(TMP0, TMP4, TMP0); + vis_mul8x16au(REF_4, CONST_256, TMP8); + + vis_padd16(TMP2, TMP6, TMP2); + vis_mul8x16au(REF_4_1, CONST_256, TMP10); + + vis_padd16(TMP0, TMP16, TMP0); + vis_mul8x16au(REF_6, CONST_256, TMP12); + + vis_padd16(TMP2, TMP18, TMP2); + vis_mul8x16au(REF_6_1, CONST_256, TMP14); + + vis_padd16(TMP8, CONST_3, TMP8); + vis_mul8x16al(DST_2, CONST_512, TMP16); + + vis_padd16(TMP8, TMP12, TMP8); + vis_mul8x16al(DST_3, CONST_512, TMP18); + + vis_padd16(TMP10, TMP14, TMP10); + vis_pack16(TMP0, DST_0); + + vis_pack16(TMP2, DST_1); + vis_st64(DST_0, dest[0]); + dest += stride; + vis_padd16(TMP10, CONST_3, TMP10); + + vis_ld64_2(dest, stride, DST_0); + vis_padd16(TMP8, TMP16, TMP8); + + vis_ld64_2(dest, stride_times_2, TMP4/*DST_2*/); + vis_padd16(TMP10, TMP18, TMP10); + vis_pack16(TMP8, DST_2); + + vis_pack16(TMP10, DST_3); + vis_st64(DST_2, dest[0]); + dest += stride; + + vis_mul8x16au(REF_S0_1, CONST_256, TMP2); + vis_pmerge(ZERO, REF_S0, TMP0); + + vis_pmerge(ZERO, REF_S2, TMP24); + vis_mul8x16au(REF_S2_1, CONST_256, TMP6); + + vis_padd16(TMP0, CONST_3, TMP0); + vis_mul8x16au(REF_S4, CONST_256, TMP8); + + vis_padd16(TMP2, CONST_3, TMP2); + vis_mul8x16au(REF_S4_1, CONST_256, TMP10); + + vis_padd16(TMP0, TMP24, TMP0); + vis_mul8x16au(REF_S6, CONST_256, TMP12); + + vis_padd16(TMP2, TMP6, TMP2); + vis_mul8x16au(REF_S6_1, CONST_256, TMP14); + + vis_padd16(TMP8, CONST_3, TMP8); + vis_mul8x16al(DST_0, CONST_512, TMP16); + + vis_padd16(TMP10, CONST_3, TMP10); + vis_mul8x16al(DST_1, CONST_512, TMP18); + + vis_padd16(TMP8, TMP12, TMP8); + vis_mul8x16al(TMP4/*DST_2*/, CONST_512, TMP20); + + vis_mul8x16al(TMP5/*DST_3*/, CONST_512, TMP22); + vis_padd16(TMP0, TMP16, TMP0); + + vis_padd16(TMP2, TMP18, TMP2); + vis_pack16(TMP0, DST_0); + + vis_padd16(TMP10, TMP14, TMP10); + vis_pack16(TMP2, DST_1); + vis_st64(DST_0, dest[0]); + dest += stride; + + vis_padd16(TMP8, TMP20, TMP8); + + vis_padd16(TMP10, TMP22, TMP10); + vis_pack16(TMP8, DST_2); + + vis_pack16(TMP10, DST_3); + vis_st64(DST_2, dest[0]); + dest += stride; + } while (--height); +} + +static void MC_put_y_16_vis (uint8_t * dest, const uint8_t * _ref, + const int stride, int height) +{ + uint8_t *ref = (uint8_t *) _ref; + + ref = vis_alignaddr(ref); + vis_ld64(ref[0], TMP0); + + vis_ld64_2(ref, 8, TMP2); + + vis_ld64_2(ref, 16, TMP4); + ref += stride; + + vis_ld64(ref[0], TMP6); + vis_faligndata(TMP0, TMP2, REF_0); + + vis_ld64_2(ref, 8, TMP8); + vis_faligndata(TMP2, TMP4, REF_4); + + vis_ld64_2(ref, 16, TMP10); + ref += stride; + + vis_ld64(constants_fe[0], MASK_fe); + vis_faligndata(TMP6, TMP8, REF_2); + + vis_ld64(constants_7f[0], MASK_7f); + vis_faligndata(TMP8, TMP10, REF_6); + + vis_ld64(constants128[0], CONST_128); + height = (height >> 1) - 1; + do { /* 24 cycles */ + vis_ld64(ref[0], TMP0); + vis_xor(REF_0, REF_2, TMP12); + + vis_ld64_2(ref, 8, TMP2); + vis_xor(REF_4, REF_6, TMP16); + + vis_ld64_2(ref, 16, TMP4); + ref += stride; + vis_or(REF_0, REF_2, TMP14); + + vis_ld64(ref[0], TMP6); + vis_or(REF_4, REF_6, TMP18); + + vis_ld64_2(ref, 8, TMP8); + vis_faligndata(TMP0, TMP2, REF_0); + + vis_ld64_2(ref, 16, TMP10); + ref += stride; + vis_faligndata(TMP2, TMP4, REF_4); + + vis_and(TMP12, MASK_fe, TMP12); + + vis_and(TMP16, MASK_fe, TMP16); + vis_mul8x16(CONST_128, TMP12, TMP12); + + vis_mul8x16(CONST_128, TMP16, TMP16); + vis_xor(REF_0, REF_2, TMP0); + + vis_xor(REF_4, REF_6, TMP2); + + vis_or(REF_0, REF_2, TMP20); + + vis_and(TMP12, MASK_7f, TMP12); + + vis_and(TMP16, MASK_7f, TMP16); + + vis_psub16(TMP14, TMP12, TMP12); + vis_st64(TMP12, dest[0]); + + vis_psub16(TMP18, TMP16, TMP16); + vis_st64_2(TMP16, dest, 8); + dest += stride; + + vis_or(REF_4, REF_6, TMP18); + + vis_and(TMP0, MASK_fe, TMP0); + + vis_and(TMP2, MASK_fe, TMP2); + vis_mul8x16(CONST_128, TMP0, TMP0); + + vis_faligndata(TMP6, TMP8, REF_2); + vis_mul8x16(CONST_128, TMP2, TMP2); + + vis_faligndata(TMP8, TMP10, REF_6); + + vis_and(TMP0, MASK_7f, TMP0); + + vis_and(TMP2, MASK_7f, TMP2); + + vis_psub16(TMP20, TMP0, TMP0); + vis_st64(TMP0, dest[0]); + + vis_psub16(TMP18, TMP2, TMP2); + vis_st64_2(TMP2, dest, 8); + dest += stride; + } while (--height); + + vis_ld64(ref[0], TMP0); + vis_xor(REF_0, REF_2, TMP12); + + vis_ld64_2(ref, 8, TMP2); + vis_xor(REF_4, REF_6, TMP16); + + vis_ld64_2(ref, 16, TMP4); + vis_or(REF_0, REF_2, TMP14); + + vis_or(REF_4, REF_6, TMP18); + + vis_faligndata(TMP0, TMP2, REF_0); + + vis_faligndata(TMP2, TMP4, REF_4); + + vis_and(TMP12, MASK_fe, TMP12); + + vis_and(TMP16, MASK_fe, TMP16); + vis_mul8x16(CONST_128, TMP12, TMP12); + + vis_mul8x16(CONST_128, TMP16, TMP16); + vis_xor(REF_0, REF_2, TMP0); + + vis_xor(REF_4, REF_6, TMP2); + + vis_or(REF_0, REF_2, TMP20); + + vis_and(TMP12, MASK_7f, TMP12); + + vis_and(TMP16, MASK_7f, TMP16); + + vis_psub16(TMP14, TMP12, TMP12); + vis_st64(TMP12, dest[0]); + + vis_psub16(TMP18, TMP16, TMP16); + vis_st64_2(TMP16, dest, 8); + dest += stride; + + vis_or(REF_4, REF_6, TMP18); + + vis_and(TMP0, MASK_fe, TMP0); + + vis_and(TMP2, MASK_fe, TMP2); + vis_mul8x16(CONST_128, TMP0, TMP0); + + vis_mul8x16(CONST_128, TMP2, TMP2); + + vis_and(TMP0, MASK_7f, TMP0); + + vis_and(TMP2, MASK_7f, TMP2); + + vis_psub16(TMP20, TMP0, TMP0); + vis_st64(TMP0, dest[0]); + + vis_psub16(TMP18, TMP2, TMP2); + vis_st64_2(TMP2, dest, 8); +} + +static void MC_put_y_8_vis (uint8_t * dest, const uint8_t * _ref, + const int stride, int height) +{ + uint8_t *ref = (uint8_t *) _ref; + + ref = vis_alignaddr(ref); + vis_ld64(ref[0], TMP0); + + vis_ld64_2(ref, 8, TMP2); + ref += stride; + + vis_ld64(ref[0], TMP4); + + vis_ld64_2(ref, 8, TMP6); + ref += stride; + + vis_ld64(constants_fe[0], MASK_fe); + vis_faligndata(TMP0, TMP2, REF_0); + + vis_ld64(constants_7f[0], MASK_7f); + vis_faligndata(TMP4, TMP6, REF_2); + + vis_ld64(constants128[0], CONST_128); + height = (height >> 1) - 1; + do { /* 12 cycles */ + vis_ld64(ref[0], TMP0); + vis_xor(REF_0, REF_2, TMP4); + + vis_ld64_2(ref, 8, TMP2); + ref += stride; + vis_and(TMP4, MASK_fe, TMP4); + + vis_or(REF_0, REF_2, TMP6); + vis_mul8x16(CONST_128, TMP4, TMP4); + + vis_faligndata(TMP0, TMP2, REF_0); + vis_ld64(ref[0], TMP0); + + vis_ld64_2(ref, 8, TMP2); + ref += stride; + vis_xor(REF_0, REF_2, TMP12); + + vis_and(TMP4, MASK_7f, TMP4); + + vis_and(TMP12, MASK_fe, TMP12); + + vis_mul8x16(CONST_128, TMP12, TMP12); + vis_or(REF_0, REF_2, TMP14); + + vis_psub16(TMP6, TMP4, DST_0); + vis_st64(DST_0, dest[0]); + dest += stride; + + vis_faligndata(TMP0, TMP2, REF_2); + + vis_and(TMP12, MASK_7f, TMP12); + + vis_psub16(TMP14, TMP12, DST_0); + vis_st64(DST_0, dest[0]); + dest += stride; + } while (--height); + + vis_ld64(ref[0], TMP0); + vis_xor(REF_0, REF_2, TMP4); + + vis_ld64_2(ref, 8, TMP2); + vis_and(TMP4, MASK_fe, TMP4); + + vis_or(REF_0, REF_2, TMP6); + vis_mul8x16(CONST_128, TMP4, TMP4); + + vis_faligndata(TMP0, TMP2, REF_0); + + vis_xor(REF_0, REF_2, TMP12); + + vis_and(TMP4, MASK_7f, TMP4); + + vis_and(TMP12, MASK_fe, TMP12); + + vis_mul8x16(CONST_128, TMP12, TMP12); + vis_or(REF_0, REF_2, TMP14); + + vis_psub16(TMP6, TMP4, DST_0); + vis_st64(DST_0, dest[0]); + dest += stride; + + vis_and(TMP12, MASK_7f, TMP12); + + vis_psub16(TMP14, TMP12, DST_0); + vis_st64(DST_0, dest[0]); +} + +static void MC_avg_y_16_vis (uint8_t * dest, const uint8_t * _ref, + const int stride, int height) +{ + uint8_t *ref = (uint8_t *) _ref; + int stride_8 = stride + 8; + int stride_16 = stride + 16; + + vis_set_gsr(5 << VIS_GSR_SCALEFACT_SHIFT); + + ref = vis_alignaddr(ref); + + vis_ld64(ref[ 0], TMP0); + vis_fzero(ZERO); + + vis_ld64(ref[ 8], TMP2); + + vis_ld64(ref[16], TMP4); + + vis_ld64(constants3[0], CONST_3); + vis_faligndata(TMP0, TMP2, REF_2); + + vis_ld64(constants256_512[0], CONST_256); + vis_faligndata(TMP2, TMP4, REF_6); + height >>= 1; + + do { /* 31 cycles */ + vis_ld64_2(ref, stride, TMP0); + vis_pmerge(ZERO, REF_2, TMP12); + vis_mul8x16au(REF_2_1, CONST_256, TMP14); + + vis_ld64_2(ref, stride_8, TMP2); + vis_pmerge(ZERO, REF_6, TMP16); + vis_mul8x16au(REF_6_1, CONST_256, TMP18); + + vis_ld64_2(ref, stride_16, TMP4); + ref += stride; + + vis_ld64(dest[0], DST_0); + vis_faligndata(TMP0, TMP2, REF_0); + + vis_ld64_2(dest, 8, DST_2); + vis_faligndata(TMP2, TMP4, REF_4); + + vis_ld64_2(ref, stride, TMP6); + vis_pmerge(ZERO, REF_0, TMP0); + vis_mul8x16au(REF_0_1, CONST_256, TMP2); + + vis_ld64_2(ref, stride_8, TMP8); + vis_pmerge(ZERO, REF_4, TMP4); + + vis_ld64_2(ref, stride_16, TMP10); + ref += stride; + + vis_ld64_2(dest, stride, REF_S0/*DST_4*/); + vis_faligndata(TMP6, TMP8, REF_2); + vis_mul8x16au(REF_4_1, CONST_256, TMP6); + + vis_ld64_2(dest, stride_8, REF_S2/*DST_6*/); + vis_faligndata(TMP8, TMP10, REF_6); + vis_mul8x16al(DST_0, CONST_512, TMP20); + + vis_padd16(TMP0, CONST_3, TMP0); + vis_mul8x16al(DST_1, CONST_512, TMP22); + + vis_padd16(TMP2, CONST_3, TMP2); + vis_mul8x16al(DST_2, CONST_512, TMP24); + + vis_padd16(TMP4, CONST_3, TMP4); + vis_mul8x16al(DST_3, CONST_512, TMP26); + + vis_padd16(TMP6, CONST_3, TMP6); + + vis_padd16(TMP12, TMP20, TMP12); + vis_mul8x16al(REF_S0, CONST_512, TMP20); + + vis_padd16(TMP14, TMP22, TMP14); + vis_mul8x16al(REF_S0_1, CONST_512, TMP22); + + vis_padd16(TMP16, TMP24, TMP16); + vis_mul8x16al(REF_S2, CONST_512, TMP24); + + vis_padd16(TMP18, TMP26, TMP18); + vis_mul8x16al(REF_S2_1, CONST_512, TMP26); + + vis_padd16(TMP12, TMP0, TMP12); + vis_mul8x16au(REF_2, CONST_256, TMP28); + + vis_padd16(TMP14, TMP2, TMP14); + vis_mul8x16au(REF_2_1, CONST_256, TMP30); + + vis_padd16(TMP16, TMP4, TMP16); + vis_mul8x16au(REF_6, CONST_256, REF_S4); + + vis_padd16(TMP18, TMP6, TMP18); + vis_mul8x16au(REF_6_1, CONST_256, REF_S6); + + vis_pack16(TMP12, DST_0); + vis_padd16(TMP28, TMP0, TMP12); + + vis_pack16(TMP14, DST_1); + vis_st64(DST_0, dest[0]); + vis_padd16(TMP30, TMP2, TMP14); + + vis_pack16(TMP16, DST_2); + vis_padd16(REF_S4, TMP4, TMP16); + + vis_pack16(TMP18, DST_3); + vis_st64_2(DST_2, dest, 8); + dest += stride; + vis_padd16(REF_S6, TMP6, TMP18); + + vis_padd16(TMP12, TMP20, TMP12); + + vis_padd16(TMP14, TMP22, TMP14); + vis_pack16(TMP12, DST_0); + + vis_padd16(TMP16, TMP24, TMP16); + vis_pack16(TMP14, DST_1); + vis_st64(DST_0, dest[0]); + + vis_padd16(TMP18, TMP26, TMP18); + vis_pack16(TMP16, DST_2); + + vis_pack16(TMP18, DST_3); + vis_st64_2(DST_2, dest, 8); + dest += stride; + } while (--height); +} + +static void MC_avg_y_8_vis (uint8_t * dest, const uint8_t * _ref, + const int stride, int height) +{ + uint8_t *ref = (uint8_t *) _ref; + int stride_8 = stride + 8; + + vis_set_gsr(5 << VIS_GSR_SCALEFACT_SHIFT); + + ref = vis_alignaddr(ref); + + vis_ld64(ref[ 0], TMP0); + vis_fzero(ZERO); + + vis_ld64(ref[ 8], TMP2); + + vis_ld64(constants3[0], CONST_3); + vis_faligndata(TMP0, TMP2, REF_2); + + vis_ld64(constants256_512[0], CONST_256); + + height >>= 1; + do { /* 20 cycles */ + vis_ld64_2(ref, stride, TMP0); + vis_pmerge(ZERO, REF_2, TMP8); + vis_mul8x16au(REF_2_1, CONST_256, TMP10); + + vis_ld64_2(ref, stride_8, TMP2); + ref += stride; + + vis_ld64(dest[0], DST_0); + + vis_ld64_2(dest, stride, DST_2); + vis_faligndata(TMP0, TMP2, REF_0); + + vis_ld64_2(ref, stride, TMP4); + vis_mul8x16al(DST_0, CONST_512, TMP16); + vis_pmerge(ZERO, REF_0, TMP12); + + vis_ld64_2(ref, stride_8, TMP6); + ref += stride; + vis_mul8x16al(DST_1, CONST_512, TMP18); + vis_pmerge(ZERO, REF_0_1, TMP14); + + vis_padd16(TMP12, CONST_3, TMP12); + vis_mul8x16al(DST_2, CONST_512, TMP24); + + vis_padd16(TMP14, CONST_3, TMP14); + vis_mul8x16al(DST_3, CONST_512, TMP26); + + vis_faligndata(TMP4, TMP6, REF_2); + + vis_padd16(TMP8, TMP12, TMP8); + + vis_padd16(TMP10, TMP14, TMP10); + vis_mul8x16au(REF_2, CONST_256, TMP20); + + vis_padd16(TMP8, TMP16, TMP0); + vis_mul8x16au(REF_2_1, CONST_256, TMP22); + + vis_padd16(TMP10, TMP18, TMP2); + vis_pack16(TMP0, DST_0); + + vis_pack16(TMP2, DST_1); + vis_st64(DST_0, dest[0]); + dest += stride; + vis_padd16(TMP12, TMP20, TMP12); + + vis_padd16(TMP14, TMP22, TMP14); + + vis_padd16(TMP12, TMP24, TMP0); + + vis_padd16(TMP14, TMP26, TMP2); + vis_pack16(TMP0, DST_2); + + vis_pack16(TMP2, DST_3); + vis_st64(DST_2, dest[0]); + dest += stride; + } while (--height); +} + +static void MC_put_xy_16_vis (uint8_t * dest, const uint8_t * _ref, + const int stride, int height) +{ + uint8_t *ref = (uint8_t *) _ref; + unsigned long off = (unsigned long) ref & 0x7; + unsigned long off_plus_1 = off + 1; + int stride_8 = stride + 8; + int stride_16 = stride + 16; + + vis_set_gsr(5 << VIS_GSR_SCALEFACT_SHIFT); + + ref = vis_alignaddr(ref); + + vis_ld64(ref[ 0], TMP0); + vis_fzero(ZERO); + + vis_ld64(ref[ 8], TMP2); + + vis_ld64(ref[16], TMP4); + + vis_ld64(constants2[0], CONST_2); + vis_faligndata(TMP0, TMP2, REF_S0); + + vis_ld64(constants256_512[0], CONST_256); + vis_faligndata(TMP2, TMP4, REF_S4); + + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + vis_faligndata(TMP0, TMP2, REF_S2); + vis_faligndata(TMP2, TMP4, REF_S6); + } else { + vis_src1(TMP2, REF_S2); + vis_src1(TMP4, REF_S6); + } + + height >>= 1; + do { + vis_ld64_2(ref, stride, TMP0); + vis_mul8x16au(REF_S0, CONST_256, TMP12); + vis_pmerge(ZERO, REF_S0_1, TMP14); + + vis_alignaddr_g0((void *)off); + + vis_ld64_2(ref, stride_8, TMP2); + vis_mul8x16au(REF_S2, CONST_256, TMP16); + vis_pmerge(ZERO, REF_S2_1, TMP18); + + vis_ld64_2(ref, stride_16, TMP4); + ref += stride; + vis_mul8x16au(REF_S4, CONST_256, TMP20); + vis_pmerge(ZERO, REF_S4_1, TMP22); + + vis_ld64_2(ref, stride, TMP6); + vis_mul8x16au(REF_S6, CONST_256, TMP24); + vis_pmerge(ZERO, REF_S6_1, TMP26); + + vis_ld64_2(ref, stride_8, TMP8); + vis_faligndata(TMP0, TMP2, REF_0); + + vis_ld64_2(ref, stride_16, TMP10); + ref += stride; + vis_faligndata(TMP2, TMP4, REF_4); + + vis_faligndata(TMP6, TMP8, REF_S0); + + vis_faligndata(TMP8, TMP10, REF_S4); + + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + vis_faligndata(TMP0, TMP2, REF_2); + vis_faligndata(TMP2, TMP4, REF_6); + vis_faligndata(TMP6, TMP8, REF_S2); + vis_faligndata(TMP8, TMP10, REF_S6); + } else { + vis_src1(TMP2, REF_2); + vis_src1(TMP4, REF_6); + vis_src1(TMP8, REF_S2); + vis_src1(TMP10, REF_S6); + } + + vis_mul8x16au(REF_0, CONST_256, TMP0); + vis_pmerge(ZERO, REF_0_1, TMP2); + + vis_mul8x16au(REF_2, CONST_256, TMP4); + vis_pmerge(ZERO, REF_2_1, TMP6); + + vis_padd16(TMP0, CONST_2, TMP8); + vis_mul8x16au(REF_4, CONST_256, TMP0); + + vis_padd16(TMP2, CONST_2, TMP10); + vis_mul8x16au(REF_4_1, CONST_256, TMP2); + + vis_padd16(TMP8, TMP4, TMP8); + vis_mul8x16au(REF_6, CONST_256, TMP4); + + vis_padd16(TMP10, TMP6, TMP10); + vis_mul8x16au(REF_6_1, CONST_256, TMP6); + + vis_padd16(TMP12, TMP8, TMP12); + + vis_padd16(TMP14, TMP10, TMP14); + + vis_padd16(TMP12, TMP16, TMP12); + + vis_padd16(TMP14, TMP18, TMP14); + vis_pack16(TMP12, DST_0); + + vis_pack16(TMP14, DST_1); + vis_st64(DST_0, dest[0]); + vis_padd16(TMP0, CONST_2, TMP12); + + vis_mul8x16au(REF_S0, CONST_256, TMP0); + vis_padd16(TMP2, CONST_2, TMP14); + + vis_mul8x16au(REF_S0_1, CONST_256, TMP2); + vis_padd16(TMP12, TMP4, TMP12); + + vis_mul8x16au(REF_S2, CONST_256, TMP4); + vis_padd16(TMP14, TMP6, TMP14); + + vis_mul8x16au(REF_S2_1, CONST_256, TMP6); + vis_padd16(TMP20, TMP12, TMP20); + + vis_padd16(TMP22, TMP14, TMP22); + + vis_padd16(TMP20, TMP24, TMP20); + + vis_padd16(TMP22, TMP26, TMP22); + vis_pack16(TMP20, DST_2); + + vis_pack16(TMP22, DST_3); + vis_st64_2(DST_2, dest, 8); + dest += stride; + vis_padd16(TMP0, TMP4, TMP24); + + vis_mul8x16au(REF_S4, CONST_256, TMP0); + vis_padd16(TMP2, TMP6, TMP26); + + vis_mul8x16au(REF_S4_1, CONST_256, TMP2); + vis_padd16(TMP24, TMP8, TMP24); + + vis_padd16(TMP26, TMP10, TMP26); + vis_pack16(TMP24, DST_0); + + vis_pack16(TMP26, DST_1); + vis_st64(DST_0, dest[0]); + vis_pmerge(ZERO, REF_S6, TMP4); + + vis_pmerge(ZERO, REF_S6_1, TMP6); + + vis_padd16(TMP0, TMP4, TMP0); + + vis_padd16(TMP2, TMP6, TMP2); + + vis_padd16(TMP0, TMP12, TMP0); + + vis_padd16(TMP2, TMP14, TMP2); + vis_pack16(TMP0, DST_2); + + vis_pack16(TMP2, DST_3); + vis_st64_2(DST_2, dest, 8); + dest += stride; + } while (--height); +} + +static void MC_put_xy_8_vis (uint8_t * dest, const uint8_t * _ref, + const int stride, int height) +{ + uint8_t *ref = (uint8_t *) _ref; + unsigned long off = (unsigned long) ref & 0x7; + unsigned long off_plus_1 = off + 1; + int stride_8 = stride + 8; + + vis_set_gsr(5 << VIS_GSR_SCALEFACT_SHIFT); + + ref = vis_alignaddr(ref); + + vis_ld64(ref[ 0], TMP0); + vis_fzero(ZERO); + + vis_ld64(ref[ 8], TMP2); + + vis_ld64(constants2[0], CONST_2); + + vis_ld64(constants256_512[0], CONST_256); + vis_faligndata(TMP0, TMP2, REF_S0); + + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + vis_faligndata(TMP0, TMP2, REF_S2); + } else { + vis_src1(TMP2, REF_S2); + } + + height >>= 1; + do { /* 26 cycles */ + vis_ld64_2(ref, stride, TMP0); + vis_mul8x16au(REF_S0, CONST_256, TMP8); + vis_pmerge(ZERO, REF_S2, TMP12); + + vis_alignaddr_g0((void *)off); + + vis_ld64_2(ref, stride_8, TMP2); + ref += stride; + vis_mul8x16au(REF_S0_1, CONST_256, TMP10); + vis_pmerge(ZERO, REF_S2_1, TMP14); + + vis_ld64_2(ref, stride, TMP4); + + vis_ld64_2(ref, stride_8, TMP6); + ref += stride; + vis_faligndata(TMP0, TMP2, REF_S4); + + vis_pmerge(ZERO, REF_S4, TMP18); + + vis_pmerge(ZERO, REF_S4_1, TMP20); + + vis_faligndata(TMP4, TMP6, REF_S0); + + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + vis_faligndata(TMP0, TMP2, REF_S6); + vis_faligndata(TMP4, TMP6, REF_S2); + } else { + vis_src1(TMP2, REF_S6); + vis_src1(TMP6, REF_S2); + } + + vis_padd16(TMP18, CONST_2, TMP18); + vis_mul8x16au(REF_S6, CONST_256, TMP22); + + vis_padd16(TMP20, CONST_2, TMP20); + vis_mul8x16au(REF_S6_1, CONST_256, TMP24); + + vis_mul8x16au(REF_S0, CONST_256, TMP26); + vis_pmerge(ZERO, REF_S0_1, TMP28); + + vis_mul8x16au(REF_S2, CONST_256, TMP30); + vis_padd16(TMP18, TMP22, TMP18); + + vis_mul8x16au(REF_S2_1, CONST_256, TMP32); + vis_padd16(TMP20, TMP24, TMP20); + + vis_padd16(TMP8, TMP18, TMP8); + + vis_padd16(TMP10, TMP20, TMP10); + + vis_padd16(TMP8, TMP12, TMP8); + + vis_padd16(TMP10, TMP14, TMP10); + vis_pack16(TMP8, DST_0); + + vis_pack16(TMP10, DST_1); + vis_st64(DST_0, dest[0]); + dest += stride; + vis_padd16(TMP18, TMP26, TMP18); + + vis_padd16(TMP20, TMP28, TMP20); + + vis_padd16(TMP18, TMP30, TMP18); + + vis_padd16(TMP20, TMP32, TMP20); + vis_pack16(TMP18, DST_2); + + vis_pack16(TMP20, DST_3); + vis_st64(DST_2, dest[0]); + dest += stride; + } while (--height); +} + +static void MC_avg_xy_16_vis (uint8_t * dest, const uint8_t * _ref, + const int stride, int height) +{ + uint8_t *ref = (uint8_t *) _ref; + unsigned long off = (unsigned long) ref & 0x7; + unsigned long off_plus_1 = off + 1; + int stride_8 = stride + 8; + int stride_16 = stride + 16; + + vis_set_gsr(4 << VIS_GSR_SCALEFACT_SHIFT); + + ref = vis_alignaddr(ref); + + vis_ld64(ref[ 0], TMP0); + vis_fzero(ZERO); + + vis_ld64(ref[ 8], TMP2); + + vis_ld64(ref[16], TMP4); + + vis_ld64(constants6[0], CONST_6); + vis_faligndata(TMP0, TMP2, REF_S0); + + vis_ld64(constants256_1024[0], CONST_256); + vis_faligndata(TMP2, TMP4, REF_S4); + + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + vis_faligndata(TMP0, TMP2, REF_S2); + vis_faligndata(TMP2, TMP4, REF_S6); + } else { + vis_src1(TMP2, REF_S2); + vis_src1(TMP4, REF_S6); + } + + height >>= 1; + do { /* 55 cycles */ + vis_ld64_2(ref, stride, TMP0); + vis_mul8x16au(REF_S0, CONST_256, TMP12); + vis_pmerge(ZERO, REF_S0_1, TMP14); + + vis_alignaddr_g0((void *)off); + + vis_ld64_2(ref, stride_8, TMP2); + vis_mul8x16au(REF_S2, CONST_256, TMP16); + vis_pmerge(ZERO, REF_S2_1, TMP18); + + vis_ld64_2(ref, stride_16, TMP4); + ref += stride; + vis_mul8x16au(REF_S4, CONST_256, TMP20); + vis_pmerge(ZERO, REF_S4_1, TMP22); + + vis_ld64_2(ref, stride, TMP6); + vis_mul8x16au(REF_S6, CONST_256, TMP24); + vis_pmerge(ZERO, REF_S6_1, TMP26); + + vis_ld64_2(ref, stride_8, TMP8); + vis_faligndata(TMP0, TMP2, REF_0); + + vis_ld64_2(ref, stride_16, TMP10); + ref += stride; + vis_faligndata(TMP2, TMP4, REF_4); + + vis_ld64(dest[0], DST_0); + vis_faligndata(TMP6, TMP8, REF_S0); + + vis_ld64_2(dest, 8, DST_2); + vis_faligndata(TMP8, TMP10, REF_S4); + + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + vis_faligndata(TMP0, TMP2, REF_2); + vis_faligndata(TMP2, TMP4, REF_6); + vis_faligndata(TMP6, TMP8, REF_S2); + vis_faligndata(TMP8, TMP10, REF_S6); + } else { + vis_src1(TMP2, REF_2); + vis_src1(TMP4, REF_6); + vis_src1(TMP8, REF_S2); + vis_src1(TMP10, REF_S6); + } + + vis_mul8x16al(DST_0, CONST_1024, TMP30); + vis_pmerge(ZERO, REF_0, TMP0); + + vis_mul8x16al(DST_1, CONST_1024, TMP32); + vis_pmerge(ZERO, REF_0_1, TMP2); + + vis_mul8x16au(REF_2, CONST_256, TMP4); + vis_pmerge(ZERO, REF_2_1, TMP6); + + vis_mul8x16al(DST_2, CONST_1024, REF_0); + vis_padd16(TMP0, CONST_6, TMP0); + + vis_mul8x16al(DST_3, CONST_1024, REF_2); + vis_padd16(TMP2, CONST_6, TMP2); + + vis_padd16(TMP0, TMP4, TMP0); + vis_mul8x16au(REF_4, CONST_256, TMP4); + + vis_padd16(TMP2, TMP6, TMP2); + vis_mul8x16au(REF_4_1, CONST_256, TMP6); + + vis_padd16(TMP12, TMP0, TMP12); + vis_mul8x16au(REF_6, CONST_256, TMP8); + + vis_padd16(TMP14, TMP2, TMP14); + vis_mul8x16au(REF_6_1, CONST_256, TMP10); + + vis_padd16(TMP12, TMP16, TMP12); + vis_mul8x16au(REF_S0, CONST_256, REF_4); + + vis_padd16(TMP14, TMP18, TMP14); + vis_mul8x16au(REF_S0_1, CONST_256, REF_6); + + vis_padd16(TMP12, TMP30, TMP12); + + vis_padd16(TMP14, TMP32, TMP14); + vis_pack16(TMP12, DST_0); + + vis_pack16(TMP14, DST_1); + vis_st64(DST_0, dest[0]); + vis_padd16(TMP4, CONST_6, TMP4); + + vis_ld64_2(dest, stride, DST_0); + vis_padd16(TMP6, CONST_6, TMP6); + vis_mul8x16au(REF_S2, CONST_256, TMP12); + + vis_padd16(TMP4, TMP8, TMP4); + vis_mul8x16au(REF_S2_1, CONST_256, TMP14); + + vis_padd16(TMP6, TMP10, TMP6); + + vis_padd16(TMP20, TMP4, TMP20); + + vis_padd16(TMP22, TMP6, TMP22); + + vis_padd16(TMP20, TMP24, TMP20); + + vis_padd16(TMP22, TMP26, TMP22); + + vis_padd16(TMP20, REF_0, TMP20); + vis_mul8x16au(REF_S4, CONST_256, REF_0); + + vis_padd16(TMP22, REF_2, TMP22); + vis_pack16(TMP20, DST_2); + + vis_pack16(TMP22, DST_3); + vis_st64_2(DST_2, dest, 8); + dest += stride; + + vis_ld64_2(dest, 8, DST_2); + vis_mul8x16al(DST_0, CONST_1024, TMP30); + vis_pmerge(ZERO, REF_S4_1, REF_2); + + vis_mul8x16al(DST_1, CONST_1024, TMP32); + vis_padd16(REF_4, TMP0, TMP8); + + vis_mul8x16au(REF_S6, CONST_256, REF_4); + vis_padd16(REF_6, TMP2, TMP10); + + vis_mul8x16au(REF_S6_1, CONST_256, REF_6); + vis_padd16(TMP8, TMP12, TMP8); + + vis_padd16(TMP10, TMP14, TMP10); + + vis_padd16(TMP8, TMP30, TMP8); + + vis_padd16(TMP10, TMP32, TMP10); + vis_pack16(TMP8, DST_0); + + vis_pack16(TMP10, DST_1); + vis_st64(DST_0, dest[0]); + + vis_padd16(REF_0, TMP4, REF_0); + + vis_mul8x16al(DST_2, CONST_1024, TMP30); + vis_padd16(REF_2, TMP6, REF_2); + + vis_mul8x16al(DST_3, CONST_1024, TMP32); + vis_padd16(REF_0, REF_4, REF_0); + + vis_padd16(REF_2, REF_6, REF_2); + + vis_padd16(REF_0, TMP30, REF_0); + + /* stall */ + + vis_padd16(REF_2, TMP32, REF_2); + vis_pack16(REF_0, DST_2); + + vis_pack16(REF_2, DST_3); + vis_st64_2(DST_2, dest, 8); + dest += stride; + } while (--height); +} + +static void MC_avg_xy_8_vis (uint8_t * dest, const uint8_t * _ref, + const int stride, int height) +{ + uint8_t *ref = (uint8_t *) _ref; + unsigned long off = (unsigned long) ref & 0x7; + unsigned long off_plus_1 = off + 1; + int stride_8 = stride + 8; + + vis_set_gsr(4 << VIS_GSR_SCALEFACT_SHIFT); + + ref = vis_alignaddr(ref); + + vis_ld64(ref[0], TMP0); + vis_fzero(ZERO); + + vis_ld64_2(ref, 8, TMP2); + + vis_ld64(constants6[0], CONST_6); + + vis_ld64(constants256_1024[0], CONST_256); + vis_faligndata(TMP0, TMP2, REF_S0); + + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + vis_faligndata(TMP0, TMP2, REF_S2); + } else { + vis_src1(TMP2, REF_S2); + } + + height >>= 1; + do { /* 31 cycles */ + vis_ld64_2(ref, stride, TMP0); + vis_mul8x16au(REF_S0, CONST_256, TMP8); + vis_pmerge(ZERO, REF_S0_1, TMP10); + + vis_ld64_2(ref, stride_8, TMP2); + ref += stride; + vis_mul8x16au(REF_S2, CONST_256, TMP12); + vis_pmerge(ZERO, REF_S2_1, TMP14); + + vis_alignaddr_g0((void *)off); + + vis_ld64_2(ref, stride, TMP4); + vis_faligndata(TMP0, TMP2, REF_S4); + + vis_ld64_2(ref, stride_8, TMP6); + ref += stride; + + vis_ld64(dest[0], DST_0); + vis_faligndata(TMP4, TMP6, REF_S0); + + vis_ld64_2(dest, stride, DST_2); + + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + vis_faligndata(TMP0, TMP2, REF_S6); + vis_faligndata(TMP4, TMP6, REF_S2); + } else { + vis_src1(TMP2, REF_S6); + vis_src1(TMP6, REF_S2); + } + + vis_mul8x16al(DST_0, CONST_1024, TMP30); + vis_pmerge(ZERO, REF_S4, TMP22); + + vis_mul8x16al(DST_1, CONST_1024, TMP32); + vis_pmerge(ZERO, REF_S4_1, TMP24); + + vis_mul8x16au(REF_S6, CONST_256, TMP26); + vis_pmerge(ZERO, REF_S6_1, TMP28); + + vis_mul8x16au(REF_S0, CONST_256, REF_S4); + vis_padd16(TMP22, CONST_6, TMP22); + + vis_mul8x16au(REF_S0_1, CONST_256, REF_S6); + vis_padd16(TMP24, CONST_6, TMP24); + + vis_mul8x16al(DST_2, CONST_1024, REF_0); + vis_padd16(TMP22, TMP26, TMP22); + + vis_mul8x16al(DST_3, CONST_1024, REF_2); + vis_padd16(TMP24, TMP28, TMP24); + + vis_mul8x16au(REF_S2, CONST_256, TMP26); + vis_padd16(TMP8, TMP22, TMP8); + + vis_mul8x16au(REF_S2_1, CONST_256, TMP28); + vis_padd16(TMP10, TMP24, TMP10); + + vis_padd16(TMP8, TMP12, TMP8); + + vis_padd16(TMP10, TMP14, TMP10); + + vis_padd16(TMP8, TMP30, TMP8); + + vis_padd16(TMP10, TMP32, TMP10); + vis_pack16(TMP8, DST_0); + + vis_pack16(TMP10, DST_1); + vis_st64(DST_0, dest[0]); + dest += stride; + + vis_padd16(REF_S4, TMP22, TMP12); + + vis_padd16(REF_S6, TMP24, TMP14); + + vis_padd16(TMP12, TMP26, TMP12); + + vis_padd16(TMP14, TMP28, TMP14); + + vis_padd16(TMP12, REF_0, TMP12); + + vis_padd16(TMP14, REF_2, TMP14); + vis_pack16(TMP12, DST_2); + + vis_pack16(TMP14, DST_3); + vis_st64(DST_2, dest[0]); + dest += stride; + } while (--height); +} + +/* End of rounding code */ + +/* Start of no rounding code */ +/* The trick used in some of this file is the formula from the MMX + * motion comp code, which is: + * + * (x+y)>>1 == (x&y)+((x^y)>>1) + * + * This allows us to average 8 bytes at a time in a 64-bit FPU reg. + * We avoid overflows by masking before we do the shift, and we + * implement the shift by multiplying by 1/2 using mul8x16. So in + * VIS this is (assume 'x' is in f0, 'y' is in f2, a repeating mask + * of '0xfe' is in f4, a repeating mask of '0x7f' is in f6, and + * the value 0x80808080 is in f8): + * + * fxor f0, f2, f10 + * fand f10, f4, f10 + * fmul8x16 f8, f10, f10 + * fand f10, f6, f10 + * fand f0, f2, f12 + * fpadd16 f12, f10, f10 + */ + +static void MC_put_no_round_o_16_vis (uint8_t * dest, const uint8_t * _ref, + const int stride, int height) +{ + uint8_t *ref = (uint8_t *) _ref; + + ref = vis_alignaddr(ref); + do { /* 5 cycles */ + vis_ld64(ref[0], TMP0); + + vis_ld64_2(ref, 8, TMP2); + + vis_ld64_2(ref, 16, TMP4); + ref += stride; + + vis_faligndata(TMP0, TMP2, REF_0); + vis_st64(REF_0, dest[0]); + + vis_faligndata(TMP2, TMP4, REF_2); + vis_st64_2(REF_2, dest, 8); + dest += stride; + } while (--height); +} + +static void MC_put_no_round_o_8_vis (uint8_t * dest, const uint8_t * _ref, + const int stride, int height) +{ + uint8_t *ref = (uint8_t *) _ref; + + ref = vis_alignaddr(ref); + do { /* 4 cycles */ + vis_ld64(ref[0], TMP0); + + vis_ld64(ref[8], TMP2); + ref += stride; + + /* stall */ + + vis_faligndata(TMP0, TMP2, REF_0); + vis_st64(REF_0, dest[0]); + dest += stride; + } while (--height); +} + + +static void MC_avg_no_round_o_16_vis (uint8_t * dest, const uint8_t * _ref, + const int stride, int height) +{ + uint8_t *ref = (uint8_t *) _ref; + int stride_8 = stride + 8; + + ref = vis_alignaddr(ref); + + vis_ld64(ref[0], TMP0); + + vis_ld64(ref[8], TMP2); + + vis_ld64(ref[16], TMP4); + + vis_ld64(dest[0], DST_0); + + vis_ld64(dest[8], DST_2); + + vis_ld64(constants_fe[0], MASK_fe); + vis_faligndata(TMP0, TMP2, REF_0); + + vis_ld64(constants_7f[0], MASK_7f); + vis_faligndata(TMP2, TMP4, REF_2); + + vis_ld64(constants128[0], CONST_128); + + ref += stride; + height = (height >> 1) - 1; + + do { /* 24 cycles */ + vis_ld64(ref[0], TMP0); + vis_xor(DST_0, REF_0, TMP6); + + vis_ld64_2(ref, 8, TMP2); + vis_and(TMP6, MASK_fe, TMP6); + + vis_ld64_2(ref, 16, TMP4); + ref += stride; + vis_mul8x16(CONST_128, TMP6, TMP6); + vis_xor(DST_2, REF_2, TMP8); + + vis_and(TMP8, MASK_fe, TMP8); + + vis_and(DST_0, REF_0, TMP10); + vis_ld64_2(dest, stride, DST_0); + vis_mul8x16(CONST_128, TMP8, TMP8); + + vis_and(DST_2, REF_2, TMP12); + vis_ld64_2(dest, stride_8, DST_2); + + vis_ld64(ref[0], TMP14); + vis_and(TMP6, MASK_7f, TMP6); + + vis_and(TMP8, MASK_7f, TMP8); + + vis_padd16(TMP10, TMP6, TMP6); + vis_st64(TMP6, dest[0]); + + vis_padd16(TMP12, TMP8, TMP8); + vis_st64_2(TMP8, dest, 8); + + dest += stride; + vis_ld64_2(ref, 8, TMP16); + vis_faligndata(TMP0, TMP2, REF_0); + + vis_ld64_2(ref, 16, TMP18); + vis_faligndata(TMP2, TMP4, REF_2); + ref += stride; + + vis_xor(DST_0, REF_0, TMP20); + + vis_and(TMP20, MASK_fe, TMP20); + + vis_xor(DST_2, REF_2, TMP22); + vis_mul8x16(CONST_128, TMP20, TMP20); + + vis_and(TMP22, MASK_fe, TMP22); + + vis_and(DST_0, REF_0, TMP24); + vis_mul8x16(CONST_128, TMP22, TMP22); + + vis_and(DST_2, REF_2, TMP26); + + vis_ld64_2(dest, stride, DST_0); + vis_faligndata(TMP14, TMP16, REF_0); + + vis_ld64_2(dest, stride_8, DST_2); + vis_faligndata(TMP16, TMP18, REF_2); + + vis_and(TMP20, MASK_7f, TMP20); + + vis_and(TMP22, MASK_7f, TMP22); + + vis_padd16(TMP24, TMP20, TMP20); + vis_st64(TMP20, dest[0]); + + vis_padd16(TMP26, TMP22, TMP22); + vis_st64_2(TMP22, dest, 8); + dest += stride; + } while (--height); + + vis_ld64(ref[0], TMP0); + vis_xor(DST_0, REF_0, TMP6); + + vis_ld64_2(ref, 8, TMP2); + vis_and(TMP6, MASK_fe, TMP6); + + vis_ld64_2(ref, 16, TMP4); + vis_mul8x16(CONST_128, TMP6, TMP6); + vis_xor(DST_2, REF_2, TMP8); + + vis_and(TMP8, MASK_fe, TMP8); + + vis_and(DST_0, REF_0, TMP10); + vis_ld64_2(dest, stride, DST_0); + vis_mul8x16(CONST_128, TMP8, TMP8); + + vis_and(DST_2, REF_2, TMP12); + vis_ld64_2(dest, stride_8, DST_2); + + vis_ld64(ref[0], TMP14); + vis_and(TMP6, MASK_7f, TMP6); + + vis_and(TMP8, MASK_7f, TMP8); + + vis_padd16(TMP10, TMP6, TMP6); + vis_st64(TMP6, dest[0]); + + vis_padd16(TMP12, TMP8, TMP8); + vis_st64_2(TMP8, dest, 8); + + dest += stride; + vis_faligndata(TMP0, TMP2, REF_0); + + vis_faligndata(TMP2, TMP4, REF_2); + + vis_xor(DST_0, REF_0, TMP20); + + vis_and(TMP20, MASK_fe, TMP20); + + vis_xor(DST_2, REF_2, TMP22); + vis_mul8x16(CONST_128, TMP20, TMP20); + + vis_and(TMP22, MASK_fe, TMP22); + + vis_and(DST_0, REF_0, TMP24); + vis_mul8x16(CONST_128, TMP22, TMP22); + + vis_and(DST_2, REF_2, TMP26); + + vis_and(TMP20, MASK_7f, TMP20); + + vis_and(TMP22, MASK_7f, TMP22); + + vis_padd16(TMP24, TMP20, TMP20); + vis_st64(TMP20, dest[0]); + + vis_padd16(TMP26, TMP22, TMP22); + vis_st64_2(TMP22, dest, 8); +} + +static void MC_avg_no_round_o_8_vis (uint8_t * dest, const uint8_t * _ref, + const int stride, int height) +{ + uint8_t *ref = (uint8_t *) _ref; + + ref = vis_alignaddr(ref); + + vis_ld64(ref[0], TMP0); + + vis_ld64(ref[8], TMP2); + + vis_ld64(dest[0], DST_0); + + vis_ld64(constants_fe[0], MASK_fe); + + vis_ld64(constants_7f[0], MASK_7f); + vis_faligndata(TMP0, TMP2, REF_0); + + vis_ld64(constants128[0], CONST_128); + + ref += stride; + height = (height >> 1) - 1; + + do { /* 12 cycles */ + vis_ld64(ref[0], TMP0); + vis_xor(DST_0, REF_0, TMP4); + + vis_ld64(ref[8], TMP2); + vis_and(TMP4, MASK_fe, TMP4); + + vis_and(DST_0, REF_0, TMP6); + vis_ld64_2(dest, stride, DST_0); + ref += stride; + vis_mul8x16(CONST_128, TMP4, TMP4); + + vis_ld64(ref[0], TMP12); + vis_faligndata(TMP0, TMP2, REF_0); + + vis_ld64(ref[8], TMP2); + vis_xor(DST_0, REF_0, TMP0); + ref += stride; + + vis_and(TMP0, MASK_fe, TMP0); + + vis_and(TMP4, MASK_7f, TMP4); + + vis_padd16(TMP6, TMP4, TMP4); + vis_st64(TMP4, dest[0]); + dest += stride; + vis_mul8x16(CONST_128, TMP0, TMP0); + + vis_and(DST_0, REF_0, TMP6); + vis_ld64_2(dest, stride, DST_0); + + vis_faligndata(TMP12, TMP2, REF_0); + + vis_and(TMP0, MASK_7f, TMP0); + + vis_padd16(TMP6, TMP0, TMP4); + vis_st64(TMP4, dest[0]); + dest += stride; + } while (--height); + + vis_ld64(ref[0], TMP0); + vis_xor(DST_0, REF_0, TMP4); + + vis_ld64(ref[8], TMP2); + vis_and(TMP4, MASK_fe, TMP4); + + vis_and(DST_0, REF_0, TMP6); + vis_ld64_2(dest, stride, DST_0); + vis_mul8x16(CONST_128, TMP4, TMP4); + + vis_faligndata(TMP0, TMP2, REF_0); + + vis_xor(DST_0, REF_0, TMP0); + + vis_and(TMP0, MASK_fe, TMP0); + + vis_and(TMP4, MASK_7f, TMP4); + + vis_padd16(TMP6, TMP4, TMP4); + vis_st64(TMP4, dest[0]); + dest += stride; + vis_mul8x16(CONST_128, TMP0, TMP0); + + vis_and(DST_0, REF_0, TMP6); + + vis_and(TMP0, MASK_7f, TMP0); + + vis_padd16(TMP6, TMP0, TMP4); + vis_st64(TMP4, dest[0]); +} + +static void MC_put_no_round_x_16_vis (uint8_t * dest, const uint8_t * _ref, + const int stride, int height) +{ + uint8_t *ref = (uint8_t *) _ref; + unsigned long off = (unsigned long) ref & 0x7; + unsigned long off_plus_1 = off + 1; + + ref = vis_alignaddr(ref); + + vis_ld64(ref[0], TMP0); + + vis_ld64_2(ref, 8, TMP2); + + vis_ld64_2(ref, 16, TMP4); + + vis_ld64(constants_fe[0], MASK_fe); + + vis_ld64(constants_7f[0], MASK_7f); + vis_faligndata(TMP0, TMP2, REF_0); + + vis_ld64(constants128[0], CONST_128); + vis_faligndata(TMP2, TMP4, REF_4); + + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + vis_faligndata(TMP0, TMP2, REF_2); + vis_faligndata(TMP2, TMP4, REF_6); + } else { + vis_src1(TMP2, REF_2); + vis_src1(TMP4, REF_6); + } + + ref += stride; + height = (height >> 1) - 1; + + do { /* 34 cycles */ + vis_ld64(ref[0], TMP0); + vis_xor(REF_0, REF_2, TMP6); + + vis_ld64_2(ref, 8, TMP2); + vis_xor(REF_4, REF_6, TMP8); + + vis_ld64_2(ref, 16, TMP4); + vis_and(TMP6, MASK_fe, TMP6); + ref += stride; + + vis_ld64(ref[0], TMP14); + vis_mul8x16(CONST_128, TMP6, TMP6); + vis_and(TMP8, MASK_fe, TMP8); + + vis_ld64_2(ref, 8, TMP16); + vis_mul8x16(CONST_128, TMP8, TMP8); + vis_and(REF_0, REF_2, TMP10); + + vis_ld64_2(ref, 16, TMP18); + ref += stride; + vis_and(REF_4, REF_6, TMP12); + + vis_alignaddr_g0((void *)off); + + vis_faligndata(TMP0, TMP2, REF_0); + + vis_faligndata(TMP2, TMP4, REF_4); + + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + vis_faligndata(TMP0, TMP2, REF_2); + vis_faligndata(TMP2, TMP4, REF_6); + } else { + vis_src1(TMP2, REF_2); + vis_src1(TMP4, REF_6); + } + + vis_and(TMP6, MASK_7f, TMP6); + + vis_and(TMP8, MASK_7f, TMP8); + + vis_padd16(TMP10, TMP6, TMP6); + vis_st64(TMP6, dest[0]); + + vis_padd16(TMP12, TMP8, TMP8); + vis_st64_2(TMP8, dest, 8); + dest += stride; + + vis_xor(REF_0, REF_2, TMP6); + + vis_xor(REF_4, REF_6, TMP8); + + vis_and(TMP6, MASK_fe, TMP6); + + vis_mul8x16(CONST_128, TMP6, TMP6); + vis_and(TMP8, MASK_fe, TMP8); + + vis_mul8x16(CONST_128, TMP8, TMP8); + vis_and(REF_0, REF_2, TMP10); + + vis_and(REF_4, REF_6, TMP12); + + vis_alignaddr_g0((void *)off); + + vis_faligndata(TMP14, TMP16, REF_0); + + vis_faligndata(TMP16, TMP18, REF_4); + + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + vis_faligndata(TMP14, TMP16, REF_2); + vis_faligndata(TMP16, TMP18, REF_6); + } else { + vis_src1(TMP16, REF_2); + vis_src1(TMP18, REF_6); + } + + vis_and(TMP6, MASK_7f, TMP6); + + vis_and(TMP8, MASK_7f, TMP8); + + vis_padd16(TMP10, TMP6, TMP6); + vis_st64(TMP6, dest[0]); + + vis_padd16(TMP12, TMP8, TMP8); + vis_st64_2(TMP8, dest, 8); + dest += stride; + } while (--height); + + vis_ld64(ref[0], TMP0); + vis_xor(REF_0, REF_2, TMP6); + + vis_ld64_2(ref, 8, TMP2); + vis_xor(REF_4, REF_6, TMP8); + + vis_ld64_2(ref, 16, TMP4); + vis_and(TMP6, MASK_fe, TMP6); + + vis_mul8x16(CONST_128, TMP6, TMP6); + vis_and(TMP8, MASK_fe, TMP8); + + vis_mul8x16(CONST_128, TMP8, TMP8); + vis_and(REF_0, REF_2, TMP10); + + vis_and(REF_4, REF_6, TMP12); + + vis_alignaddr_g0((void *)off); + + vis_faligndata(TMP0, TMP2, REF_0); + + vis_faligndata(TMP2, TMP4, REF_4); + + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + vis_faligndata(TMP0, TMP2, REF_2); + vis_faligndata(TMP2, TMP4, REF_6); + } else { + vis_src1(TMP2, REF_2); + vis_src1(TMP4, REF_6); + } + + vis_and(TMP6, MASK_7f, TMP6); + + vis_and(TMP8, MASK_7f, TMP8); + + vis_padd16(TMP10, TMP6, TMP6); + vis_st64(TMP6, dest[0]); + + vis_padd16(TMP12, TMP8, TMP8); + vis_st64_2(TMP8, dest, 8); + dest += stride; + + vis_xor(REF_0, REF_2, TMP6); + + vis_xor(REF_4, REF_6, TMP8); + + vis_and(TMP6, MASK_fe, TMP6); + + vis_mul8x16(CONST_128, TMP6, TMP6); + vis_and(TMP8, MASK_fe, TMP8); + + vis_mul8x16(CONST_128, TMP8, TMP8); + vis_and(REF_0, REF_2, TMP10); + + vis_and(REF_4, REF_6, TMP12); + + vis_and(TMP6, MASK_7f, TMP6); + + vis_and(TMP8, MASK_7f, TMP8); + + vis_padd16(TMP10, TMP6, TMP6); + vis_st64(TMP6, dest[0]); + + vis_padd16(TMP12, TMP8, TMP8); + vis_st64_2(TMP8, dest, 8); +} + +static void MC_put_no_round_x_8_vis (uint8_t * dest, const uint8_t * _ref, + const int stride, int height) +{ + uint8_t *ref = (uint8_t *) _ref; + unsigned long off = (unsigned long) ref & 0x7; + unsigned long off_plus_1 = off + 1; + + ref = vis_alignaddr(ref); + + vis_ld64(ref[0], TMP0); + + vis_ld64(ref[8], TMP2); + + vis_ld64(constants_fe[0], MASK_fe); + + vis_ld64(constants_7f[0], MASK_7f); + + vis_ld64(constants128[0], CONST_128); + vis_faligndata(TMP0, TMP2, REF_0); + + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + vis_faligndata(TMP0, TMP2, REF_2); + } else { + vis_src1(TMP2, REF_2); + } + + ref += stride; + height = (height >> 1) - 1; + + do { /* 20 cycles */ + vis_ld64(ref[0], TMP0); + vis_xor(REF_0, REF_2, TMP4); + + vis_ld64_2(ref, 8, TMP2); + vis_and(TMP4, MASK_fe, TMP4); + ref += stride; + + vis_ld64(ref[0], TMP8); + vis_and(REF_0, REF_2, TMP6); + vis_mul8x16(CONST_128, TMP4, TMP4); + + vis_alignaddr_g0((void *)off); + + vis_ld64_2(ref, 8, TMP10); + ref += stride; + vis_faligndata(TMP0, TMP2, REF_0); + + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + vis_faligndata(TMP0, TMP2, REF_2); + } else { + vis_src1(TMP2, REF_2); + } + + vis_and(TMP4, MASK_7f, TMP4); + + vis_padd16(TMP6, TMP4, DST_0); + vis_st64(DST_0, dest[0]); + dest += stride; + + vis_xor(REF_0, REF_2, TMP12); + + vis_and(TMP12, MASK_fe, TMP12); + + vis_and(REF_0, REF_2, TMP14); + vis_mul8x16(CONST_128, TMP12, TMP12); + + vis_alignaddr_g0((void *)off); + vis_faligndata(TMP8, TMP10, REF_0); + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + vis_faligndata(TMP8, TMP10, REF_2); + } else { + vis_src1(TMP10, REF_2); + } + + vis_and(TMP12, MASK_7f, TMP12); + + vis_padd16(TMP14, TMP12, DST_0); + vis_st64(DST_0, dest[0]); + dest += stride; + } while (--height); + + vis_ld64(ref[0], TMP0); + vis_xor(REF_0, REF_2, TMP4); + + vis_ld64_2(ref, 8, TMP2); + vis_and(TMP4, MASK_fe, TMP4); + + vis_and(REF_0, REF_2, TMP6); + vis_mul8x16(CONST_128, TMP4, TMP4); + + vis_alignaddr_g0((void *)off); + + vis_faligndata(TMP0, TMP2, REF_0); + + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + vis_faligndata(TMP0, TMP2, REF_2); + } else { + vis_src1(TMP2, REF_2); + } + + vis_and(TMP4, MASK_7f, TMP4); + + vis_padd16(TMP6, TMP4, DST_0); + vis_st64(DST_0, dest[0]); + dest += stride; + + vis_xor(REF_0, REF_2, TMP12); + + vis_and(TMP12, MASK_fe, TMP12); + + vis_and(REF_0, REF_2, TMP14); + vis_mul8x16(CONST_128, TMP12, TMP12); + + vis_and(TMP12, MASK_7f, TMP12); + + vis_padd16(TMP14, TMP12, DST_0); + vis_st64(DST_0, dest[0]); + dest += stride; +} + +static void MC_avg_no_round_x_16_vis (uint8_t * dest, const uint8_t * _ref, + const int stride, int height) +{ + uint8_t *ref = (uint8_t *) _ref; + unsigned long off = (unsigned long) ref & 0x7; + unsigned long off_plus_1 = off + 1; + + vis_set_gsr(5 << VIS_GSR_SCALEFACT_SHIFT); + + vis_ld64(constants3[0], CONST_3); + vis_fzero(ZERO); + vis_ld64(constants256_512[0], CONST_256); + + ref = vis_alignaddr(ref); + do { /* 26 cycles */ + vis_ld64(ref[0], TMP0); + + vis_ld64(ref[8], TMP2); + + vis_alignaddr_g0((void *)off); + + vis_ld64(ref[16], TMP4); + + vis_ld64(dest[0], DST_0); + vis_faligndata(TMP0, TMP2, REF_0); + + vis_ld64(dest[8], DST_2); + vis_faligndata(TMP2, TMP4, REF_4); + + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + vis_faligndata(TMP0, TMP2, REF_2); + vis_faligndata(TMP2, TMP4, REF_6); + } else { + vis_src1(TMP2, REF_2); + vis_src1(TMP4, REF_6); + } + + vis_mul8x16au(REF_0, CONST_256, TMP0); + + vis_pmerge(ZERO, REF_2, TMP4); + vis_mul8x16au(REF_0_1, CONST_256, TMP2); + + vis_pmerge(ZERO, REF_2_1, TMP6); + + vis_padd16(TMP0, TMP4, TMP0); + + vis_mul8x16al(DST_0, CONST_512, TMP4); + vis_padd16(TMP2, TMP6, TMP2); + + vis_mul8x16al(DST_1, CONST_512, TMP6); + + vis_mul8x16au(REF_6, CONST_256, TMP12); + + vis_padd16(TMP0, TMP4, TMP0); + vis_mul8x16au(REF_6_1, CONST_256, TMP14); + + vis_padd16(TMP2, TMP6, TMP2); + vis_mul8x16au(REF_4, CONST_256, TMP16); + + vis_padd16(TMP0, CONST_3, TMP8); + vis_mul8x16au(REF_4_1, CONST_256, TMP18); + + vis_padd16(TMP2, CONST_3, TMP10); + vis_pack16(TMP8, DST_0); + + vis_pack16(TMP10, DST_1); + vis_padd16(TMP16, TMP12, TMP0); + + vis_st64(DST_0, dest[0]); + vis_mul8x16al(DST_2, CONST_512, TMP4); + vis_padd16(TMP18, TMP14, TMP2); + + vis_mul8x16al(DST_3, CONST_512, TMP6); + vis_padd16(TMP0, CONST_3, TMP0); + + vis_padd16(TMP2, CONST_3, TMP2); + + vis_padd16(TMP0, TMP4, TMP0); + + vis_padd16(TMP2, TMP6, TMP2); + vis_pack16(TMP0, DST_2); + + vis_pack16(TMP2, DST_3); + vis_st64(DST_2, dest[8]); + + ref += stride; + dest += stride; + } while (--height); +} + +static void MC_avg_no_round_x_8_vis (uint8_t * dest, const uint8_t * _ref, + const int stride, int height) +{ + uint8_t *ref = (uint8_t *) _ref; + unsigned long off = (unsigned long) ref & 0x7; + unsigned long off_plus_1 = off + 1; + int stride_times_2 = stride << 1; + + vis_set_gsr(5 << VIS_GSR_SCALEFACT_SHIFT); + + vis_ld64(constants3[0], CONST_3); + vis_fzero(ZERO); + vis_ld64(constants256_512[0], CONST_256); + + ref = vis_alignaddr(ref); + height >>= 2; + do { /* 47 cycles */ + vis_ld64(ref[0], TMP0); + + vis_ld64_2(ref, 8, TMP2); + ref += stride; + + vis_alignaddr_g0((void *)off); + + vis_ld64(ref[0], TMP4); + vis_faligndata(TMP0, TMP2, REF_0); + + vis_ld64_2(ref, 8, TMP6); + ref += stride; + + vis_ld64(ref[0], TMP8); + + vis_ld64_2(ref, 8, TMP10); + ref += stride; + vis_faligndata(TMP4, TMP6, REF_4); + + vis_ld64(ref[0], TMP12); + + vis_ld64_2(ref, 8, TMP14); + ref += stride; + vis_faligndata(TMP8, TMP10, REF_S0); + + vis_faligndata(TMP12, TMP14, REF_S4); + + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + + vis_ld64(dest[0], DST_0); + vis_faligndata(TMP0, TMP2, REF_2); + + vis_ld64_2(dest, stride, DST_2); + vis_faligndata(TMP4, TMP6, REF_6); + + vis_faligndata(TMP8, TMP10, REF_S2); + + vis_faligndata(TMP12, TMP14, REF_S6); + } else { + vis_ld64(dest[0], DST_0); + vis_src1(TMP2, REF_2); + + vis_ld64_2(dest, stride, DST_2); + vis_src1(TMP6, REF_6); + + vis_src1(TMP10, REF_S2); + + vis_src1(TMP14, REF_S6); + } + + vis_pmerge(ZERO, REF_0, TMP0); + vis_mul8x16au(REF_0_1, CONST_256, TMP2); + + vis_pmerge(ZERO, REF_2, TMP4); + vis_mul8x16au(REF_2_1, CONST_256, TMP6); + + vis_padd16(TMP0, CONST_3, TMP0); + vis_mul8x16al(DST_0, CONST_512, TMP16); + + vis_padd16(TMP2, CONST_3, TMP2); + vis_mul8x16al(DST_1, CONST_512, TMP18); + + vis_padd16(TMP0, TMP4, TMP0); + vis_mul8x16au(REF_4, CONST_256, TMP8); + + vis_padd16(TMP2, TMP6, TMP2); + vis_mul8x16au(REF_4_1, CONST_256, TMP10); + + vis_padd16(TMP0, TMP16, TMP0); + vis_mul8x16au(REF_6, CONST_256, TMP12); + + vis_padd16(TMP2, TMP18, TMP2); + vis_mul8x16au(REF_6_1, CONST_256, TMP14); + + vis_padd16(TMP8, CONST_3, TMP8); + vis_mul8x16al(DST_2, CONST_512, TMP16); + + vis_padd16(TMP8, TMP12, TMP8); + vis_mul8x16al(DST_3, CONST_512, TMP18); + + vis_padd16(TMP10, TMP14, TMP10); + vis_pack16(TMP0, DST_0); + + vis_pack16(TMP2, DST_1); + vis_st64(DST_0, dest[0]); + dest += stride; + vis_padd16(TMP10, CONST_3, TMP10); + + vis_ld64_2(dest, stride, DST_0); + vis_padd16(TMP8, TMP16, TMP8); + + vis_ld64_2(dest, stride_times_2, TMP4/*DST_2*/); + vis_padd16(TMP10, TMP18, TMP10); + vis_pack16(TMP8, DST_2); + + vis_pack16(TMP10, DST_3); + vis_st64(DST_2, dest[0]); + dest += stride; + + vis_mul8x16au(REF_S0_1, CONST_256, TMP2); + vis_pmerge(ZERO, REF_S0, TMP0); + + vis_pmerge(ZERO, REF_S2, TMP24); + vis_mul8x16au(REF_S2_1, CONST_256, TMP6); + + vis_padd16(TMP0, CONST_3, TMP0); + vis_mul8x16au(REF_S4, CONST_256, TMP8); + + vis_padd16(TMP2, CONST_3, TMP2); + vis_mul8x16au(REF_S4_1, CONST_256, TMP10); + + vis_padd16(TMP0, TMP24, TMP0); + vis_mul8x16au(REF_S6, CONST_256, TMP12); + + vis_padd16(TMP2, TMP6, TMP2); + vis_mul8x16au(REF_S6_1, CONST_256, TMP14); + + vis_padd16(TMP8, CONST_3, TMP8); + vis_mul8x16al(DST_0, CONST_512, TMP16); + + vis_padd16(TMP10, CONST_3, TMP10); + vis_mul8x16al(DST_1, CONST_512, TMP18); + + vis_padd16(TMP8, TMP12, TMP8); + vis_mul8x16al(TMP4/*DST_2*/, CONST_512, TMP20); + + vis_mul8x16al(TMP5/*DST_3*/, CONST_512, TMP22); + vis_padd16(TMP0, TMP16, TMP0); + + vis_padd16(TMP2, TMP18, TMP2); + vis_pack16(TMP0, DST_0); + + vis_padd16(TMP10, TMP14, TMP10); + vis_pack16(TMP2, DST_1); + vis_st64(DST_0, dest[0]); + dest += stride; + + vis_padd16(TMP8, TMP20, TMP8); + + vis_padd16(TMP10, TMP22, TMP10); + vis_pack16(TMP8, DST_2); + + vis_pack16(TMP10, DST_3); + vis_st64(DST_2, dest[0]); + dest += stride; + } while (--height); +} + +static void MC_put_no_round_y_16_vis (uint8_t * dest, const uint8_t * _ref, + const int stride, int height) +{ + uint8_t *ref = (uint8_t *) _ref; + + ref = vis_alignaddr(ref); + vis_ld64(ref[0], TMP0); + + vis_ld64_2(ref, 8, TMP2); + + vis_ld64_2(ref, 16, TMP4); + ref += stride; + + vis_ld64(ref[0], TMP6); + vis_faligndata(TMP0, TMP2, REF_0); + + vis_ld64_2(ref, 8, TMP8); + vis_faligndata(TMP2, TMP4, REF_4); + + vis_ld64_2(ref, 16, TMP10); + ref += stride; + + vis_ld64(constants_fe[0], MASK_fe); + vis_faligndata(TMP6, TMP8, REF_2); + + vis_ld64(constants_7f[0], MASK_7f); + vis_faligndata(TMP8, TMP10, REF_6); + + vis_ld64(constants128[0], CONST_128); + height = (height >> 1) - 1; + do { /* 24 cycles */ + vis_ld64(ref[0], TMP0); + vis_xor(REF_0, REF_2, TMP12); + + vis_ld64_2(ref, 8, TMP2); + vis_xor(REF_4, REF_6, TMP16); + + vis_ld64_2(ref, 16, TMP4); + ref += stride; + vis_and(REF_0, REF_2, TMP14); + + vis_ld64(ref[0], TMP6); + vis_and(REF_4, REF_6, TMP18); + + vis_ld64_2(ref, 8, TMP8); + vis_faligndata(TMP0, TMP2, REF_0); + + vis_ld64_2(ref, 16, TMP10); + ref += stride; + vis_faligndata(TMP2, TMP4, REF_4); + + vis_and(TMP12, MASK_fe, TMP12); + + vis_and(TMP16, MASK_fe, TMP16); + vis_mul8x16(CONST_128, TMP12, TMP12); + + vis_mul8x16(CONST_128, TMP16, TMP16); + vis_xor(REF_0, REF_2, TMP0); + + vis_xor(REF_4, REF_6, TMP2); + + vis_and(REF_0, REF_2, TMP20); + + vis_and(TMP12, MASK_7f, TMP12); + + vis_and(TMP16, MASK_7f, TMP16); + + vis_padd16(TMP14, TMP12, TMP12); + vis_st64(TMP12, dest[0]); + + vis_padd16(TMP18, TMP16, TMP16); + vis_st64_2(TMP16, dest, 8); + dest += stride; + + vis_and(REF_4, REF_6, TMP18); + + vis_and(TMP0, MASK_fe, TMP0); + + vis_and(TMP2, MASK_fe, TMP2); + vis_mul8x16(CONST_128, TMP0, TMP0); + + vis_faligndata(TMP6, TMP8, REF_2); + vis_mul8x16(CONST_128, TMP2, TMP2); + + vis_faligndata(TMP8, TMP10, REF_6); + + vis_and(TMP0, MASK_7f, TMP0); + + vis_and(TMP2, MASK_7f, TMP2); + + vis_padd16(TMP20, TMP0, TMP0); + vis_st64(TMP0, dest[0]); + + vis_padd16(TMP18, TMP2, TMP2); + vis_st64_2(TMP2, dest, 8); + dest += stride; + } while (--height); + + vis_ld64(ref[0], TMP0); + vis_xor(REF_0, REF_2, TMP12); + + vis_ld64_2(ref, 8, TMP2); + vis_xor(REF_4, REF_6, TMP16); + + vis_ld64_2(ref, 16, TMP4); + vis_and(REF_0, REF_2, TMP14); + + vis_and(REF_4, REF_6, TMP18); + + vis_faligndata(TMP0, TMP2, REF_0); + + vis_faligndata(TMP2, TMP4, REF_4); + + vis_and(TMP12, MASK_fe, TMP12); + + vis_and(TMP16, MASK_fe, TMP16); + vis_mul8x16(CONST_128, TMP12, TMP12); + + vis_mul8x16(CONST_128, TMP16, TMP16); + vis_xor(REF_0, REF_2, TMP0); + + vis_xor(REF_4, REF_6, TMP2); + + vis_and(REF_0, REF_2, TMP20); + + vis_and(TMP12, MASK_7f, TMP12); + + vis_and(TMP16, MASK_7f, TMP16); + + vis_padd16(TMP14, TMP12, TMP12); + vis_st64(TMP12, dest[0]); + + vis_padd16(TMP18, TMP16, TMP16); + vis_st64_2(TMP16, dest, 8); + dest += stride; + + vis_and(REF_4, REF_6, TMP18); + + vis_and(TMP0, MASK_fe, TMP0); + + vis_and(TMP2, MASK_fe, TMP2); + vis_mul8x16(CONST_128, TMP0, TMP0); + + vis_mul8x16(CONST_128, TMP2, TMP2); + + vis_and(TMP0, MASK_7f, TMP0); + + vis_and(TMP2, MASK_7f, TMP2); + + vis_padd16(TMP20, TMP0, TMP0); + vis_st64(TMP0, dest[0]); + + vis_padd16(TMP18, TMP2, TMP2); + vis_st64_2(TMP2, dest, 8); +} + +static void MC_put_no_round_y_8_vis (uint8_t * dest, const uint8_t * _ref, + const int stride, int height) +{ + uint8_t *ref = (uint8_t *) _ref; + + ref = vis_alignaddr(ref); + vis_ld64(ref[0], TMP0); + + vis_ld64_2(ref, 8, TMP2); + ref += stride; + + vis_ld64(ref[0], TMP4); + + vis_ld64_2(ref, 8, TMP6); + ref += stride; + + vis_ld64(constants_fe[0], MASK_fe); + vis_faligndata(TMP0, TMP2, REF_0); + + vis_ld64(constants_7f[0], MASK_7f); + vis_faligndata(TMP4, TMP6, REF_2); + + vis_ld64(constants128[0], CONST_128); + height = (height >> 1) - 1; + do { /* 12 cycles */ + vis_ld64(ref[0], TMP0); + vis_xor(REF_0, REF_2, TMP4); + + vis_ld64_2(ref, 8, TMP2); + ref += stride; + vis_and(TMP4, MASK_fe, TMP4); + + vis_and(REF_0, REF_2, TMP6); + vis_mul8x16(CONST_128, TMP4, TMP4); + + vis_faligndata(TMP0, TMP2, REF_0); + vis_ld64(ref[0], TMP0); + + vis_ld64_2(ref, 8, TMP2); + ref += stride; + vis_xor(REF_0, REF_2, TMP12); + + vis_and(TMP4, MASK_7f, TMP4); + + vis_and(TMP12, MASK_fe, TMP12); + + vis_mul8x16(CONST_128, TMP12, TMP12); + vis_and(REF_0, REF_2, TMP14); + + vis_padd16(TMP6, TMP4, DST_0); + vis_st64(DST_0, dest[0]); + dest += stride; + + vis_faligndata(TMP0, TMP2, REF_2); + + vis_and(TMP12, MASK_7f, TMP12); + + vis_padd16(TMP14, TMP12, DST_0); + vis_st64(DST_0, dest[0]); + dest += stride; + } while (--height); + + vis_ld64(ref[0], TMP0); + vis_xor(REF_0, REF_2, TMP4); + + vis_ld64_2(ref, 8, TMP2); + vis_and(TMP4, MASK_fe, TMP4); + + vis_and(REF_0, REF_2, TMP6); + vis_mul8x16(CONST_128, TMP4, TMP4); + + vis_faligndata(TMP0, TMP2, REF_0); + + vis_xor(REF_0, REF_2, TMP12); + + vis_and(TMP4, MASK_7f, TMP4); + + vis_and(TMP12, MASK_fe, TMP12); + + vis_mul8x16(CONST_128, TMP12, TMP12); + vis_and(REF_0, REF_2, TMP14); + + vis_padd16(TMP6, TMP4, DST_0); + vis_st64(DST_0, dest[0]); + dest += stride; + + vis_and(TMP12, MASK_7f, TMP12); + + vis_padd16(TMP14, TMP12, DST_0); + vis_st64(DST_0, dest[0]); +} + +static void MC_avg_no_round_y_16_vis (uint8_t * dest, const uint8_t * _ref, + const int stride, int height) +{ + uint8_t *ref = (uint8_t *) _ref; + int stride_8 = stride + 8; + int stride_16 = stride + 16; + + vis_set_gsr(5 << VIS_GSR_SCALEFACT_SHIFT); + + ref = vis_alignaddr(ref); + + vis_ld64(ref[ 0], TMP0); + vis_fzero(ZERO); + + vis_ld64(ref[ 8], TMP2); + + vis_ld64(ref[16], TMP4); + + vis_ld64(constants3[0], CONST_3); + vis_faligndata(TMP0, TMP2, REF_2); + + vis_ld64(constants256_512[0], CONST_256); + vis_faligndata(TMP2, TMP4, REF_6); + height >>= 1; + + do { /* 31 cycles */ + vis_ld64_2(ref, stride, TMP0); + vis_pmerge(ZERO, REF_2, TMP12); + vis_mul8x16au(REF_2_1, CONST_256, TMP14); + + vis_ld64_2(ref, stride_8, TMP2); + vis_pmerge(ZERO, REF_6, TMP16); + vis_mul8x16au(REF_6_1, CONST_256, TMP18); + + vis_ld64_2(ref, stride_16, TMP4); + ref += stride; + + vis_ld64(dest[0], DST_0); + vis_faligndata(TMP0, TMP2, REF_0); + + vis_ld64_2(dest, 8, DST_2); + vis_faligndata(TMP2, TMP4, REF_4); + + vis_ld64_2(ref, stride, TMP6); + vis_pmerge(ZERO, REF_0, TMP0); + vis_mul8x16au(REF_0_1, CONST_256, TMP2); + + vis_ld64_2(ref, stride_8, TMP8); + vis_pmerge(ZERO, REF_4, TMP4); + + vis_ld64_2(ref, stride_16, TMP10); + ref += stride; + + vis_ld64_2(dest, stride, REF_S0/*DST_4*/); + vis_faligndata(TMP6, TMP8, REF_2); + vis_mul8x16au(REF_4_1, CONST_256, TMP6); + + vis_ld64_2(dest, stride_8, REF_S2/*DST_6*/); + vis_faligndata(TMP8, TMP10, REF_6); + vis_mul8x16al(DST_0, CONST_512, TMP20); + + vis_padd16(TMP0, CONST_3, TMP0); + vis_mul8x16al(DST_1, CONST_512, TMP22); + + vis_padd16(TMP2, CONST_3, TMP2); + vis_mul8x16al(DST_2, CONST_512, TMP24); + + vis_padd16(TMP4, CONST_3, TMP4); + vis_mul8x16al(DST_3, CONST_512, TMP26); + + vis_padd16(TMP6, CONST_3, TMP6); + + vis_padd16(TMP12, TMP20, TMP12); + vis_mul8x16al(REF_S0, CONST_512, TMP20); + + vis_padd16(TMP14, TMP22, TMP14); + vis_mul8x16al(REF_S0_1, CONST_512, TMP22); + + vis_padd16(TMP16, TMP24, TMP16); + vis_mul8x16al(REF_S2, CONST_512, TMP24); + + vis_padd16(TMP18, TMP26, TMP18); + vis_mul8x16al(REF_S2_1, CONST_512, TMP26); + + vis_padd16(TMP12, TMP0, TMP12); + vis_mul8x16au(REF_2, CONST_256, TMP28); + + vis_padd16(TMP14, TMP2, TMP14); + vis_mul8x16au(REF_2_1, CONST_256, TMP30); + + vis_padd16(TMP16, TMP4, TMP16); + vis_mul8x16au(REF_6, CONST_256, REF_S4); + + vis_padd16(TMP18, TMP6, TMP18); + vis_mul8x16au(REF_6_1, CONST_256, REF_S6); + + vis_pack16(TMP12, DST_0); + vis_padd16(TMP28, TMP0, TMP12); + + vis_pack16(TMP14, DST_1); + vis_st64(DST_0, dest[0]); + vis_padd16(TMP30, TMP2, TMP14); + + vis_pack16(TMP16, DST_2); + vis_padd16(REF_S4, TMP4, TMP16); + + vis_pack16(TMP18, DST_3); + vis_st64_2(DST_2, dest, 8); + dest += stride; + vis_padd16(REF_S6, TMP6, TMP18); + + vis_padd16(TMP12, TMP20, TMP12); + + vis_padd16(TMP14, TMP22, TMP14); + vis_pack16(TMP12, DST_0); + + vis_padd16(TMP16, TMP24, TMP16); + vis_pack16(TMP14, DST_1); + vis_st64(DST_0, dest[0]); + + vis_padd16(TMP18, TMP26, TMP18); + vis_pack16(TMP16, DST_2); + + vis_pack16(TMP18, DST_3); + vis_st64_2(DST_2, dest, 8); + dest += stride; + } while (--height); +} + +static void MC_avg_no_round_y_8_vis (uint8_t * dest, const uint8_t * _ref, + const int stride, int height) +{ + uint8_t *ref = (uint8_t *) _ref; + int stride_8 = stride + 8; + + vis_set_gsr(5 << VIS_GSR_SCALEFACT_SHIFT); + + ref = vis_alignaddr(ref); + + vis_ld64(ref[ 0], TMP0); + vis_fzero(ZERO); + + vis_ld64(ref[ 8], TMP2); + + vis_ld64(constants3[0], CONST_3); + vis_faligndata(TMP0, TMP2, REF_2); + + vis_ld64(constants256_512[0], CONST_256); + + height >>= 1; + do { /* 20 cycles */ + vis_ld64_2(ref, stride, TMP0); + vis_pmerge(ZERO, REF_2, TMP8); + vis_mul8x16au(REF_2_1, CONST_256, TMP10); + + vis_ld64_2(ref, stride_8, TMP2); + ref += stride; + + vis_ld64(dest[0], DST_0); + + vis_ld64_2(dest, stride, DST_2); + vis_faligndata(TMP0, TMP2, REF_0); + + vis_ld64_2(ref, stride, TMP4); + vis_mul8x16al(DST_0, CONST_512, TMP16); + vis_pmerge(ZERO, REF_0, TMP12); + + vis_ld64_2(ref, stride_8, TMP6); + ref += stride; + vis_mul8x16al(DST_1, CONST_512, TMP18); + vis_pmerge(ZERO, REF_0_1, TMP14); + + vis_padd16(TMP12, CONST_3, TMP12); + vis_mul8x16al(DST_2, CONST_512, TMP24); + + vis_padd16(TMP14, CONST_3, TMP14); + vis_mul8x16al(DST_3, CONST_512, TMP26); + + vis_faligndata(TMP4, TMP6, REF_2); + + vis_padd16(TMP8, TMP12, TMP8); + + vis_padd16(TMP10, TMP14, TMP10); + vis_mul8x16au(REF_2, CONST_256, TMP20); + + vis_padd16(TMP8, TMP16, TMP0); + vis_mul8x16au(REF_2_1, CONST_256, TMP22); + + vis_padd16(TMP10, TMP18, TMP2); + vis_pack16(TMP0, DST_0); + + vis_pack16(TMP2, DST_1); + vis_st64(DST_0, dest[0]); + dest += stride; + vis_padd16(TMP12, TMP20, TMP12); + + vis_padd16(TMP14, TMP22, TMP14); + + vis_padd16(TMP12, TMP24, TMP0); + + vis_padd16(TMP14, TMP26, TMP2); + vis_pack16(TMP0, DST_2); + + vis_pack16(TMP2, DST_3); + vis_st64(DST_2, dest[0]); + dest += stride; + } while (--height); +} + +static void MC_put_no_round_xy_16_vis (uint8_t * dest, const uint8_t * _ref, + const int stride, int height) +{ + uint8_t *ref = (uint8_t *) _ref; + unsigned long off = (unsigned long) ref & 0x7; + unsigned long off_plus_1 = off + 1; + int stride_8 = stride + 8; + int stride_16 = stride + 16; + + vis_set_gsr(5 << VIS_GSR_SCALEFACT_SHIFT); + + ref = vis_alignaddr(ref); + + vis_ld64(ref[ 0], TMP0); + vis_fzero(ZERO); + + vis_ld64(ref[ 8], TMP2); + + vis_ld64(ref[16], TMP4); + + vis_ld64(constants1[0], CONST_1); + vis_faligndata(TMP0, TMP2, REF_S0); + + vis_ld64(constants256_512[0], CONST_256); + vis_faligndata(TMP2, TMP4, REF_S4); + + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + vis_faligndata(TMP0, TMP2, REF_S2); + vis_faligndata(TMP2, TMP4, REF_S6); + } else { + vis_src1(TMP2, REF_S2); + vis_src1(TMP4, REF_S6); + } + + height >>= 1; + do { + vis_ld64_2(ref, stride, TMP0); + vis_mul8x16au(REF_S0, CONST_256, TMP12); + vis_pmerge(ZERO, REF_S0_1, TMP14); + + vis_alignaddr_g0((void *)off); + + vis_ld64_2(ref, stride_8, TMP2); + vis_mul8x16au(REF_S2, CONST_256, TMP16); + vis_pmerge(ZERO, REF_S2_1, TMP18); + + vis_ld64_2(ref, stride_16, TMP4); + ref += stride; + vis_mul8x16au(REF_S4, CONST_256, TMP20); + vis_pmerge(ZERO, REF_S4_1, TMP22); + + vis_ld64_2(ref, stride, TMP6); + vis_mul8x16au(REF_S6, CONST_256, TMP24); + vis_pmerge(ZERO, REF_S6_1, TMP26); + + vis_ld64_2(ref, stride_8, TMP8); + vis_faligndata(TMP0, TMP2, REF_0); + + vis_ld64_2(ref, stride_16, TMP10); + ref += stride; + vis_faligndata(TMP2, TMP4, REF_4); + + vis_faligndata(TMP6, TMP8, REF_S0); + + vis_faligndata(TMP8, TMP10, REF_S4); + + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + vis_faligndata(TMP0, TMP2, REF_2); + vis_faligndata(TMP2, TMP4, REF_6); + vis_faligndata(TMP6, TMP8, REF_S2); + vis_faligndata(TMP8, TMP10, REF_S6); + } else { + vis_src1(TMP2, REF_2); + vis_src1(TMP4, REF_6); + vis_src1(TMP8, REF_S2); + vis_src1(TMP10, REF_S6); + } + + vis_mul8x16au(REF_0, CONST_256, TMP0); + vis_pmerge(ZERO, REF_0_1, TMP2); + + vis_mul8x16au(REF_2, CONST_256, TMP4); + vis_pmerge(ZERO, REF_2_1, TMP6); + + vis_padd16(TMP0, CONST_2, TMP8); + vis_mul8x16au(REF_4, CONST_256, TMP0); + + vis_padd16(TMP2, CONST_1, TMP10); + vis_mul8x16au(REF_4_1, CONST_256, TMP2); + + vis_padd16(TMP8, TMP4, TMP8); + vis_mul8x16au(REF_6, CONST_256, TMP4); + + vis_padd16(TMP10, TMP6, TMP10); + vis_mul8x16au(REF_6_1, CONST_256, TMP6); + + vis_padd16(TMP12, TMP8, TMP12); + + vis_padd16(TMP14, TMP10, TMP14); + + vis_padd16(TMP12, TMP16, TMP12); + + vis_padd16(TMP14, TMP18, TMP14); + vis_pack16(TMP12, DST_0); + + vis_pack16(TMP14, DST_1); + vis_st64(DST_0, dest[0]); + vis_padd16(TMP0, CONST_1, TMP12); + + vis_mul8x16au(REF_S0, CONST_256, TMP0); + vis_padd16(TMP2, CONST_1, TMP14); + + vis_mul8x16au(REF_S0_1, CONST_256, TMP2); + vis_padd16(TMP12, TMP4, TMP12); + + vis_mul8x16au(REF_S2, CONST_256, TMP4); + vis_padd16(TMP14, TMP6, TMP14); + + vis_mul8x16au(REF_S2_1, CONST_256, TMP6); + vis_padd16(TMP20, TMP12, TMP20); + + vis_padd16(TMP22, TMP14, TMP22); + + vis_padd16(TMP20, TMP24, TMP20); + + vis_padd16(TMP22, TMP26, TMP22); + vis_pack16(TMP20, DST_2); + + vis_pack16(TMP22, DST_3); + vis_st64_2(DST_2, dest, 8); + dest += stride; + vis_padd16(TMP0, TMP4, TMP24); + + vis_mul8x16au(REF_S4, CONST_256, TMP0); + vis_padd16(TMP2, TMP6, TMP26); + + vis_mul8x16au(REF_S4_1, CONST_256, TMP2); + vis_padd16(TMP24, TMP8, TMP24); + + vis_padd16(TMP26, TMP10, TMP26); + vis_pack16(TMP24, DST_0); + + vis_pack16(TMP26, DST_1); + vis_st64(DST_0, dest[0]); + vis_pmerge(ZERO, REF_S6, TMP4); + + vis_pmerge(ZERO, REF_S6_1, TMP6); + + vis_padd16(TMP0, TMP4, TMP0); + + vis_padd16(TMP2, TMP6, TMP2); + + vis_padd16(TMP0, TMP12, TMP0); + + vis_padd16(TMP2, TMP14, TMP2); + vis_pack16(TMP0, DST_2); + + vis_pack16(TMP2, DST_3); + vis_st64_2(DST_2, dest, 8); + dest += stride; + } while (--height); +} + +static void MC_put_no_round_xy_8_vis (uint8_t * dest, const uint8_t * _ref, + const int stride, int height) +{ + uint8_t *ref = (uint8_t *) _ref; + unsigned long off = (unsigned long) ref & 0x7; + unsigned long off_plus_1 = off + 1; + int stride_8 = stride + 8; + + vis_set_gsr(5 << VIS_GSR_SCALEFACT_SHIFT); + + ref = vis_alignaddr(ref); + + vis_ld64(ref[ 0], TMP0); + vis_fzero(ZERO); + + vis_ld64(ref[ 8], TMP2); + + vis_ld64(constants1[0], CONST_1); + + vis_ld64(constants256_512[0], CONST_256); + vis_faligndata(TMP0, TMP2, REF_S0); + + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + vis_faligndata(TMP0, TMP2, REF_S2); + } else { + vis_src1(TMP2, REF_S2); + } + + height >>= 1; + do { /* 26 cycles */ + vis_ld64_2(ref, stride, TMP0); + vis_mul8x16au(REF_S0, CONST_256, TMP8); + vis_pmerge(ZERO, REF_S2, TMP12); + + vis_alignaddr_g0((void *)off); + + vis_ld64_2(ref, stride_8, TMP2); + ref += stride; + vis_mul8x16au(REF_S0_1, CONST_256, TMP10); + vis_pmerge(ZERO, REF_S2_1, TMP14); + + vis_ld64_2(ref, stride, TMP4); + + vis_ld64_2(ref, stride_8, TMP6); + ref += stride; + vis_faligndata(TMP0, TMP2, REF_S4); + + vis_pmerge(ZERO, REF_S4, TMP18); + + vis_pmerge(ZERO, REF_S4_1, TMP20); + + vis_faligndata(TMP4, TMP6, REF_S0); + + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + vis_faligndata(TMP0, TMP2, REF_S6); + vis_faligndata(TMP4, TMP6, REF_S2); + } else { + vis_src1(TMP2, REF_S6); + vis_src1(TMP6, REF_S2); + } + + vis_padd16(TMP18, CONST_1, TMP18); + vis_mul8x16au(REF_S6, CONST_256, TMP22); + + vis_padd16(TMP20, CONST_1, TMP20); + vis_mul8x16au(REF_S6_1, CONST_256, TMP24); + + vis_mul8x16au(REF_S0, CONST_256, TMP26); + vis_pmerge(ZERO, REF_S0_1, TMP28); + + vis_mul8x16au(REF_S2, CONST_256, TMP30); + vis_padd16(TMP18, TMP22, TMP18); + + vis_mul8x16au(REF_S2_1, CONST_256, TMP32); + vis_padd16(TMP20, TMP24, TMP20); + + vis_padd16(TMP8, TMP18, TMP8); + + vis_padd16(TMP10, TMP20, TMP10); + + vis_padd16(TMP8, TMP12, TMP8); + + vis_padd16(TMP10, TMP14, TMP10); + vis_pack16(TMP8, DST_0); + + vis_pack16(TMP10, DST_1); + vis_st64(DST_0, dest[0]); + dest += stride; + vis_padd16(TMP18, TMP26, TMP18); + + vis_padd16(TMP20, TMP28, TMP20); + + vis_padd16(TMP18, TMP30, TMP18); + + vis_padd16(TMP20, TMP32, TMP20); + vis_pack16(TMP18, DST_2); + + vis_pack16(TMP20, DST_3); + vis_st64(DST_2, dest[0]); + dest += stride; + } while (--height); +} + +static void MC_avg_no_round_xy_16_vis (uint8_t * dest, const uint8_t * _ref, + const int stride, int height) +{ + uint8_t *ref = (uint8_t *) _ref; + unsigned long off = (unsigned long) ref & 0x7; + unsigned long off_plus_1 = off + 1; + int stride_8 = stride + 8; + int stride_16 = stride + 16; + + vis_set_gsr(4 << VIS_GSR_SCALEFACT_SHIFT); + + ref = vis_alignaddr(ref); + + vis_ld64(ref[ 0], TMP0); + vis_fzero(ZERO); + + vis_ld64(ref[ 8], TMP2); + + vis_ld64(ref[16], TMP4); + + vis_ld64(constants6[0], CONST_6); + vis_faligndata(TMP0, TMP2, REF_S0); + + vis_ld64(constants256_1024[0], CONST_256); + vis_faligndata(TMP2, TMP4, REF_S4); + + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + vis_faligndata(TMP0, TMP2, REF_S2); + vis_faligndata(TMP2, TMP4, REF_S6); + } else { + vis_src1(TMP2, REF_S2); + vis_src1(TMP4, REF_S6); + } + + height >>= 1; + do { /* 55 cycles */ + vis_ld64_2(ref, stride, TMP0); + vis_mul8x16au(REF_S0, CONST_256, TMP12); + vis_pmerge(ZERO, REF_S0_1, TMP14); + + vis_alignaddr_g0((void *)off); + + vis_ld64_2(ref, stride_8, TMP2); + vis_mul8x16au(REF_S2, CONST_256, TMP16); + vis_pmerge(ZERO, REF_S2_1, TMP18); + + vis_ld64_2(ref, stride_16, TMP4); + ref += stride; + vis_mul8x16au(REF_S4, CONST_256, TMP20); + vis_pmerge(ZERO, REF_S4_1, TMP22); + + vis_ld64_2(ref, stride, TMP6); + vis_mul8x16au(REF_S6, CONST_256, TMP24); + vis_pmerge(ZERO, REF_S6_1, TMP26); + + vis_ld64_2(ref, stride_8, TMP8); + vis_faligndata(TMP0, TMP2, REF_0); + + vis_ld64_2(ref, stride_16, TMP10); + ref += stride; + vis_faligndata(TMP2, TMP4, REF_4); + + vis_ld64(dest[0], DST_0); + vis_faligndata(TMP6, TMP8, REF_S0); + + vis_ld64_2(dest, 8, DST_2); + vis_faligndata(TMP8, TMP10, REF_S4); + + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + vis_faligndata(TMP0, TMP2, REF_2); + vis_faligndata(TMP2, TMP4, REF_6); + vis_faligndata(TMP6, TMP8, REF_S2); + vis_faligndata(TMP8, TMP10, REF_S6); + } else { + vis_src1(TMP2, REF_2); + vis_src1(TMP4, REF_6); + vis_src1(TMP8, REF_S2); + vis_src1(TMP10, REF_S6); + } + + vis_mul8x16al(DST_0, CONST_1024, TMP30); + vis_pmerge(ZERO, REF_0, TMP0); + + vis_mul8x16al(DST_1, CONST_1024, TMP32); + vis_pmerge(ZERO, REF_0_1, TMP2); + + vis_mul8x16au(REF_2, CONST_256, TMP4); + vis_pmerge(ZERO, REF_2_1, TMP6); + + vis_mul8x16al(DST_2, CONST_1024, REF_0); + vis_padd16(TMP0, CONST_6, TMP0); + + vis_mul8x16al(DST_3, CONST_1024, REF_2); + vis_padd16(TMP2, CONST_6, TMP2); + + vis_padd16(TMP0, TMP4, TMP0); + vis_mul8x16au(REF_4, CONST_256, TMP4); + + vis_padd16(TMP2, TMP6, TMP2); + vis_mul8x16au(REF_4_1, CONST_256, TMP6); + + vis_padd16(TMP12, TMP0, TMP12); + vis_mul8x16au(REF_6, CONST_256, TMP8); + + vis_padd16(TMP14, TMP2, TMP14); + vis_mul8x16au(REF_6_1, CONST_256, TMP10); + + vis_padd16(TMP12, TMP16, TMP12); + vis_mul8x16au(REF_S0, CONST_256, REF_4); + + vis_padd16(TMP14, TMP18, TMP14); + vis_mul8x16au(REF_S0_1, CONST_256, REF_6); + + vis_padd16(TMP12, TMP30, TMP12); + + vis_padd16(TMP14, TMP32, TMP14); + vis_pack16(TMP12, DST_0); + + vis_pack16(TMP14, DST_1); + vis_st64(DST_0, dest[0]); + vis_padd16(TMP4, CONST_6, TMP4); + + vis_ld64_2(dest, stride, DST_0); + vis_padd16(TMP6, CONST_6, TMP6); + vis_mul8x16au(REF_S2, CONST_256, TMP12); + + vis_padd16(TMP4, TMP8, TMP4); + vis_mul8x16au(REF_S2_1, CONST_256, TMP14); + + vis_padd16(TMP6, TMP10, TMP6); + + vis_padd16(TMP20, TMP4, TMP20); + + vis_padd16(TMP22, TMP6, TMP22); + + vis_padd16(TMP20, TMP24, TMP20); + + vis_padd16(TMP22, TMP26, TMP22); + + vis_padd16(TMP20, REF_0, TMP20); + vis_mul8x16au(REF_S4, CONST_256, REF_0); + + vis_padd16(TMP22, REF_2, TMP22); + vis_pack16(TMP20, DST_2); + + vis_pack16(TMP22, DST_3); + vis_st64_2(DST_2, dest, 8); + dest += stride; + + vis_ld64_2(dest, 8, DST_2); + vis_mul8x16al(DST_0, CONST_1024, TMP30); + vis_pmerge(ZERO, REF_S4_1, REF_2); + + vis_mul8x16al(DST_1, CONST_1024, TMP32); + vis_padd16(REF_4, TMP0, TMP8); + + vis_mul8x16au(REF_S6, CONST_256, REF_4); + vis_padd16(REF_6, TMP2, TMP10); + + vis_mul8x16au(REF_S6_1, CONST_256, REF_6); + vis_padd16(TMP8, TMP12, TMP8); + + vis_padd16(TMP10, TMP14, TMP10); + + vis_padd16(TMP8, TMP30, TMP8); + + vis_padd16(TMP10, TMP32, TMP10); + vis_pack16(TMP8, DST_0); + + vis_pack16(TMP10, DST_1); + vis_st64(DST_0, dest[0]); + + vis_padd16(REF_0, TMP4, REF_0); + + vis_mul8x16al(DST_2, CONST_1024, TMP30); + vis_padd16(REF_2, TMP6, REF_2); + + vis_mul8x16al(DST_3, CONST_1024, TMP32); + vis_padd16(REF_0, REF_4, REF_0); + + vis_padd16(REF_2, REF_6, REF_2); + + vis_padd16(REF_0, TMP30, REF_0); + + /* stall */ + + vis_padd16(REF_2, TMP32, REF_2); + vis_pack16(REF_0, DST_2); + + vis_pack16(REF_2, DST_3); + vis_st64_2(DST_2, dest, 8); + dest += stride; + } while (--height); +} + +static void MC_avg_no_round_xy_8_vis (uint8_t * dest, const uint8_t * _ref, + const int stride, int height) +{ + uint8_t *ref = (uint8_t *) _ref; + unsigned long off = (unsigned long) ref & 0x7; + unsigned long off_plus_1 = off + 1; + int stride_8 = stride + 8; + + vis_set_gsr(4 << VIS_GSR_SCALEFACT_SHIFT); + + ref = vis_alignaddr(ref); + + vis_ld64(ref[0], TMP0); + vis_fzero(ZERO); + + vis_ld64_2(ref, 8, TMP2); + + vis_ld64(constants6[0], CONST_6); + + vis_ld64(constants256_1024[0], CONST_256); + vis_faligndata(TMP0, TMP2, REF_S0); + + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + vis_faligndata(TMP0, TMP2, REF_S2); + } else { + vis_src1(TMP2, REF_S2); + } + + height >>= 1; + do { /* 31 cycles */ + vis_ld64_2(ref, stride, TMP0); + vis_mul8x16au(REF_S0, CONST_256, TMP8); + vis_pmerge(ZERO, REF_S0_1, TMP10); + + vis_ld64_2(ref, stride_8, TMP2); + ref += stride; + vis_mul8x16au(REF_S2, CONST_256, TMP12); + vis_pmerge(ZERO, REF_S2_1, TMP14); + + vis_alignaddr_g0((void *)off); + + vis_ld64_2(ref, stride, TMP4); + vis_faligndata(TMP0, TMP2, REF_S4); + + vis_ld64_2(ref, stride_8, TMP6); + ref += stride; + + vis_ld64(dest[0], DST_0); + vis_faligndata(TMP4, TMP6, REF_S0); + + vis_ld64_2(dest, stride, DST_2); + + if (off != 0x7) { + vis_alignaddr_g0((void *)off_plus_1); + vis_faligndata(TMP0, TMP2, REF_S6); + vis_faligndata(TMP4, TMP6, REF_S2); + } else { + vis_src1(TMP2, REF_S6); + vis_src1(TMP6, REF_S2); + } + + vis_mul8x16al(DST_0, CONST_1024, TMP30); + vis_pmerge(ZERO, REF_S4, TMP22); + + vis_mul8x16al(DST_1, CONST_1024, TMP32); + vis_pmerge(ZERO, REF_S4_1, TMP24); + + vis_mul8x16au(REF_S6, CONST_256, TMP26); + vis_pmerge(ZERO, REF_S6_1, TMP28); + + vis_mul8x16au(REF_S0, CONST_256, REF_S4); + vis_padd16(TMP22, CONST_6, TMP22); + + vis_mul8x16au(REF_S0_1, CONST_256, REF_S6); + vis_padd16(TMP24, CONST_6, TMP24); + + vis_mul8x16al(DST_2, CONST_1024, REF_0); + vis_padd16(TMP22, TMP26, TMP22); + + vis_mul8x16al(DST_3, CONST_1024, REF_2); + vis_padd16(TMP24, TMP28, TMP24); + + vis_mul8x16au(REF_S2, CONST_256, TMP26); + vis_padd16(TMP8, TMP22, TMP8); + + vis_mul8x16au(REF_S2_1, CONST_256, TMP28); + vis_padd16(TMP10, TMP24, TMP10); + + vis_padd16(TMP8, TMP12, TMP8); + + vis_padd16(TMP10, TMP14, TMP10); + + vis_padd16(TMP8, TMP30, TMP8); + + vis_padd16(TMP10, TMP32, TMP10); + vis_pack16(TMP8, DST_0); + + vis_pack16(TMP10, DST_1); + vis_st64(DST_0, dest[0]); + dest += stride; + + vis_padd16(REF_S4, TMP22, TMP12); + + vis_padd16(REF_S6, TMP24, TMP14); + + vis_padd16(TMP12, TMP26, TMP12); + + vis_padd16(TMP14, TMP28, TMP14); + + vis_padd16(TMP12, REF_0, TMP12); + + vis_padd16(TMP14, REF_2, TMP14); + vis_pack16(TMP12, DST_2); + + vis_pack16(TMP14, DST_3); + vis_st64(DST_2, dest[0]); + dest += stride; + } while (--height); +} + +/* End of no rounding code */ + +#define ACCEL_SPARC_VIS 1 +#define ACCEL_SPARC_VIS2 2 + +static int vis_level () +{ + int accel = 0; + accel |= ACCEL_SPARC_VIS; + accel |= ACCEL_SPARC_VIS2; + return accel; +} + +/* libavcodec initialization code */ +void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx) +{ + /* VIS-specific optimizations */ + int accel = vis_level (); + + if (accel & ACCEL_SPARC_VIS) { + if(avctx->idct_algo==FF_IDCT_SIMPLEVIS){ + c->idct_put = ff_simple_idct_put_vis; + c->idct_add = ff_simple_idct_add_vis; + c->idct = ff_simple_idct_vis; + c->idct_permutation_type = FF_TRANSPOSE_IDCT_PERM; + } + + c->put_pixels_tab[0][0] = MC_put_o_16_vis; + c->put_pixels_tab[0][1] = MC_put_x_16_vis; + c->put_pixels_tab[0][2] = MC_put_y_16_vis; + c->put_pixels_tab[0][3] = MC_put_xy_16_vis; + + c->put_pixels_tab[1][0] = MC_put_o_8_vis; + c->put_pixels_tab[1][1] = MC_put_x_8_vis; + c->put_pixels_tab[1][2] = MC_put_y_8_vis; + c->put_pixels_tab[1][3] = MC_put_xy_8_vis; + + c->avg_pixels_tab[0][0] = MC_avg_o_16_vis; + c->avg_pixels_tab[0][1] = MC_avg_x_16_vis; + c->avg_pixels_tab[0][2] = MC_avg_y_16_vis; + c->avg_pixels_tab[0][3] = MC_avg_xy_16_vis; + + c->avg_pixels_tab[1][0] = MC_avg_o_8_vis; + c->avg_pixels_tab[1][1] = MC_avg_x_8_vis; + c->avg_pixels_tab[1][2] = MC_avg_y_8_vis; + c->avg_pixels_tab[1][3] = MC_avg_xy_8_vis; + + c->put_no_rnd_pixels_tab[0][0] = MC_put_no_round_o_16_vis; + c->put_no_rnd_pixels_tab[0][1] = MC_put_no_round_x_16_vis; + c->put_no_rnd_pixels_tab[0][2] = MC_put_no_round_y_16_vis; + c->put_no_rnd_pixels_tab[0][3] = MC_put_no_round_xy_16_vis; + + c->put_no_rnd_pixels_tab[1][0] = MC_put_no_round_o_8_vis; + c->put_no_rnd_pixels_tab[1][1] = MC_put_no_round_x_8_vis; + c->put_no_rnd_pixels_tab[1][2] = MC_put_no_round_y_8_vis; + c->put_no_rnd_pixels_tab[1][3] = MC_put_no_round_xy_8_vis; + + c->avg_no_rnd_pixels_tab[0][0] = MC_avg_no_round_o_16_vis; + c->avg_no_rnd_pixels_tab[0][1] = MC_avg_no_round_x_16_vis; + c->avg_no_rnd_pixels_tab[0][2] = MC_avg_no_round_y_16_vis; + c->avg_no_rnd_pixels_tab[0][3] = MC_avg_no_round_xy_16_vis; + + c->avg_no_rnd_pixels_tab[1][0] = MC_avg_no_round_o_8_vis; + c->avg_no_rnd_pixels_tab[1][1] = MC_avg_no_round_x_8_vis; + c->avg_no_rnd_pixels_tab[1][2] = MC_avg_no_round_y_8_vis; + c->avg_no_rnd_pixels_tab[1][3] = MC_avg_no_round_xy_8_vis; + } +} diff --git a/contrib/ffmpeg/libavcodec/sparc/simple_idct_vis.c b/contrib/ffmpeg/libavcodec/sparc/simple_idct_vis.c new file mode 100644 index 000000000..716d3de48 --- /dev/null +++ b/contrib/ffmpeg/libavcodec/sparc/simple_idct_vis.c @@ -0,0 +1,528 @@ +/* + * SPARC VIS optimized inverse DCT + * Copyright (c) 2007 Denes Balatoni < dbalatoni XatX interware XdotX hu > + * + * I did consult the following fine web page about dct + * http://www.geocities.com/ssavekar/dct.htm + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include "dsputil.h" + +static const DECLARE_ALIGNED_8(int16_t, coeffs[28]) = { + - 1259,- 1259,- 1259,- 1259, + - 4989,- 4989,- 4989,- 4989, + -11045,-11045,-11045,-11045, + -19195,-19195,-19195,-19195, + -29126,-29126,-29126,-29126, + 25080, 25080, 25080, 25080, + 12785, 12785, 12785, 12785 +}; +static const DECLARE_ALIGNED_8(uint16_t, scale[4]) = { + 65536>>6, 65536>>6, 65536>>6, 65536>>6 +}; +static const DECLARE_ALIGNED_8(uint16_t, rounder[4]) = { + 1<<5, 1<<5, 1<<5, 1<<5 +}; +static const DECLARE_ALIGNED_8(uint16_t, expand[4]) = { + 1<<14, 1<<14, 1<<14, 1<<14 +}; + +#define INIT_IDCT \ + "ldd [%1], %%f32 \n\t"\ + "ldd [%1+8], %%f34 \n\t"\ + "ldd [%1+16], %%f36 \n\t"\ + "ldd [%1+24], %%f38 \n\t"\ + "ldd [%1+32], %%f40 \n\t"\ + "ldd [%1+40], %%f42 \n\t"\ + "ldd [%1+48], %%f44 \n\t"\ + "ldd [%0], %%f46 \n\t"\ + "fzero %%f62 \n\t"\ + +#define LOADSCALE(in) \ + "ldd [" in "], %%f0 \n\t"\ + "ldd [" in "+16], %%f2 \n\t"\ + "ldd [" in "+32], %%f4 \n\t"\ + "ldd [" in "+48], %%f6 \n\t"\ + "ldd [" in "+64], %%f8 \n\t"\ + "ldd [" in "+80], %%f10 \n\t"\ + "ldd [" in "+96], %%f12 \n\t"\ + "ldd [" in "+112], %%f14 \n\t"\ + "fpadd16 %%f0, %%f0, %%f0 \n\t"\ + "fpadd16 %%f2, %%f2, %%f2 \n\t"\ + "fpadd16 %%f4, %%f4, %%f4 \n\t"\ + "fpadd16 %%f6, %%f6, %%f6 \n\t"\ + "fpadd16 %%f8, %%f8, %%f8 \n\t"\ + "fpadd16 %%f10, %%f10, %%f10 \n\t"\ + "fpadd16 %%f12, %%f12, %%f12 \n\t"\ + "fpadd16 %%f14, %%f14, %%f14 \n\t"\ +\ + "fpadd16 %%f0, %%f0, %%f0 \n\t"\ + "fpadd16 %%f2, %%f2, %%f2 \n\t"\ + "fpadd16 %%f4, %%f4, %%f4 \n\t"\ + "fpadd16 %%f6, %%f6, %%f6 \n\t"\ + "fpadd16 %%f8, %%f8, %%f8 \n\t"\ + "fpadd16 %%f10, %%f10, %%f10 \n\t"\ + "fpadd16 %%f12, %%f12, %%f12 \n\t"\ + "fpadd16 %%f14, %%f14, %%f14 \n\t"\ +\ + "fpadd16 %%f0, %%f0, %%f0 \n\t"\ + "fpadd16 %%f2, %%f2, %%f2 \n\t"\ + "fpadd16 %%f4, %%f4, %%f4 \n\t"\ + "fpadd16 %%f6, %%f6, %%f6 \n\t"\ + "fpadd16 %%f8, %%f8, %%f8 \n\t"\ + "fpadd16 %%f10, %%f10, %%f10 \n\t"\ + "fpadd16 %%f12, %%f12, %%f12 \n\t"\ + "fpadd16 %%f14, %%f14, %%f14 \n\t"\ +\ + "fpadd16 %%f0, %%f0, %%f0 \n\t"\ + "fpadd16 %%f2, %%f2, %%f2 \n\t"\ + "fpadd16 %%f4, %%f4, %%f4 \n\t"\ + "fpadd16 %%f6, %%f6, %%f6 \n\t"\ + "fpadd16 %%f8, %%f8, %%f8 \n\t"\ + "fpadd16 %%f10, %%f10, %%f10 \n\t"\ + "fpadd16 %%f12, %%f12, %%f12 \n\t"\ + "fpadd16 %%f14, %%f14, %%f14 \n\t"\ + +#define LOAD(in) \ + "ldd [" in "], %%f16 \n\t"\ + "ldd [" in "+8], %%f18 \n\t"\ + "ldd [" in "+16], %%f20 \n\t"\ + "ldd [" in "+24], %%f22 \n\t"\ + "ldd [" in "+32], %%f24 \n\t"\ + "ldd [" in "+40], %%f26 \n\t"\ + "ldd [" in "+48], %%f28 \n\t"\ + "ldd [" in "+56], %%f30 \n\t"\ + +#define TRANSPOSE \ + "fpmerge %%f16, %%f24, %%f0 \n\t"\ + "fpmerge %%f20, %%f28, %%f2 \n\t"\ + "fpmerge %%f17, %%f25, %%f4 \n\t"\ + "fpmerge %%f21, %%f29, %%f6 \n\t"\ + "fpmerge %%f18, %%f26, %%f8 \n\t"\ + "fpmerge %%f22, %%f30, %%f10 \n\t"\ + "fpmerge %%f19, %%f27, %%f12 \n\t"\ + "fpmerge %%f23, %%f31, %%f14 \n\t"\ +\ + "fpmerge %%f0, %%f2, %%f16 \n\t"\ + "fpmerge %%f1, %%f3, %%f18 \n\t"\ + "fpmerge %%f4, %%f6, %%f20 \n\t"\ + "fpmerge %%f5, %%f7, %%f22 \n\t"\ + "fpmerge %%f8, %%f10, %%f24 \n\t"\ + "fpmerge %%f9, %%f11, %%f26 \n\t"\ + "fpmerge %%f12, %%f14, %%f28 \n\t"\ + "fpmerge %%f13, %%f15, %%f30 \n\t"\ +\ + "fpmerge %%f16, %%f17, %%f0 \n\t"\ + "fpmerge %%f18, %%f19, %%f2 \n\t"\ + "fpmerge %%f20, %%f21, %%f4 \n\t"\ + "fpmerge %%f22, %%f23, %%f6 \n\t"\ + "fpmerge %%f24, %%f25, %%f8 \n\t"\ + "fpmerge %%f26, %%f27, %%f10 \n\t"\ + "fpmerge %%f28, %%f29, %%f12 \n\t"\ + "fpmerge %%f30, %%f31, %%f14 \n\t"\ + +#define IDCT4ROWS \ + /* 1. column */\ + "fmul8ulx16 %%f0, %%f38, %%f28 \n\t"\ + "for %%f4, %%f6, %%f60 \n\t"\ + "fmul8ulx16 %%f2, %%f32, %%f18 \n\t"\ + "fmul8ulx16 %%f2, %%f36, %%f22 \n\t"\ + "fmul8ulx16 %%f2, %%f40, %%f26 \n\t"\ + "fmul8ulx16 %%f2, %%f44, %%f30 \n\t"\ +\ + ADDROUNDER\ +\ + "fmul8sux16 %%f0, %%f38, %%f48 \n\t"\ + "fcmpd %%fcc0, %%f62, %%f60 \n\t"\ + "for %%f8, %%f10, %%f60 \n\t"\ + "fmul8sux16 %%f2, %%f32, %%f50 \n\t"\ + "fmul8sux16 %%f2, %%f36, %%f52 \n\t"\ + "fmul8sux16 %%f2, %%f40, %%f54 \n\t"\ + "fmul8sux16 %%f2, %%f44, %%f56 \n\t"\ +\ + "fpadd16 %%f48, %%f28, %%f28 \n\t"\ + "fcmpd %%fcc1, %%f62, %%f60 \n\t"\ + "for %%f12, %%f14, %%f60 \n\t"\ + "fpadd16 %%f50, %%f18, %%f18 \n\t"\ + "fpadd16 %%f52, %%f22, %%f22 \n\t"\ + "fpadd16 %%f54, %%f26, %%f26 \n\t"\ + "fpadd16 %%f56, %%f30, %%f30 \n\t"\ +\ + "fpadd16 %%f28, %%f0, %%f16 \n\t"\ + "fcmpd %%fcc2, %%f62, %%f60 \n\t"\ + "fpadd16 %%f28, %%f0, %%f20 \n\t"\ + "fpadd16 %%f28, %%f0, %%f24 \n\t"\ + "fpadd16 %%f28, %%f0, %%f28 \n\t"\ + "fpadd16 %%f18, %%f2, %%f18 \n\t"\ + "fpadd16 %%f22, %%f2, %%f22 \n\t"\ + /* 2. column */\ + "fbe %%fcc0, 3f \n\t"\ + "fpadd16 %%f26, %%f2, %%f26 \n\t"\ + "fmul8ulx16 %%f4, %%f34, %%f48 \n\t"\ + "fmul8ulx16 %%f4, %%f42, %%f50 \n\t"\ + "fmul8ulx16 %%f6, %%f36, %%f52 \n\t"\ + "fmul8ulx16 %%f6, %%f44, %%f54 \n\t"\ + "fmul8ulx16 %%f6, %%f32, %%f56 \n\t"\ + "fmul8ulx16 %%f6, %%f40, %%f58 \n\t"\ +\ + "fpadd16 %%f16, %%f48, %%f16 \n\t"\ + "fpadd16 %%f20, %%f50, %%f20 \n\t"\ + "fpsub16 %%f24, %%f50, %%f24 \n\t"\ + "fpsub16 %%f28, %%f48, %%f28 \n\t"\ + "fpadd16 %%f18, %%f52, %%f18 \n\t"\ + "fpsub16 %%f22, %%f54, %%f22 \n\t"\ + "fpsub16 %%f26, %%f56, %%f26 \n\t"\ + "fpsub16 %%f30, %%f58, %%f30 \n\t"\ +\ + "fmul8sux16 %%f4, %%f34, %%f48 \n\t"\ + "fmul8sux16 %%f4, %%f42, %%f50 \n\t"\ + "fmul8sux16 %%f6, %%f36, %%f52 \n\t"\ + "fmul8sux16 %%f6, %%f44, %%f54 \n\t"\ + "fmul8sux16 %%f6, %%f32, %%f56 \n\t"\ + "fmul8sux16 %%f6, %%f40, %%f58 \n\t"\ +\ + "fpadd16 %%f16, %%f48, %%f16 \n\t"\ + "fpadd16 %%f20, %%f50, %%f20 \n\t"\ + "fpsub16 %%f24, %%f50, %%f24 \n\t"\ + "fpsub16 %%f28, %%f48, %%f28 \n\t"\ + "fpadd16 %%f18, %%f52, %%f18 \n\t"\ + "fpsub16 %%f22, %%f54, %%f22 \n\t"\ + "fpsub16 %%f26, %%f56, %%f26 \n\t"\ + "fpsub16 %%f30, %%f58, %%f30 \n\t"\ +\ + "fpadd16 %%f16, %%f4, %%f16 \n\t"\ + "fpsub16 %%f28, %%f4, %%f28 \n\t"\ + "fpadd16 %%f18, %%f6, %%f18 \n\t"\ + "fpsub16 %%f26, %%f6, %%f26 \n\t"\ + /* 3. column */\ + "3: \n\t"\ + "fbe %%fcc1, 4f \n\t"\ + "fpsub16 %%f30, %%f6, %%f30 \n\t"\ + "fmul8ulx16 %%f8, %%f38, %%f48 \n\t"\ + "fmul8ulx16 %%f10, %%f40, %%f50 \n\t"\ + "fmul8ulx16 %%f10, %%f32, %%f52 \n\t"\ + "fmul8ulx16 %%f10, %%f44, %%f54 \n\t"\ + "fmul8ulx16 %%f10, %%f36, %%f56 \n\t"\ +\ + "fpadd16 %%f16, %%f48, %%f16 \n\t"\ + "fpsub16 %%f20, %%f48, %%f20 \n\t"\ + "fpsub16 %%f24, %%f48, %%f24 \n\t"\ + "fpadd16 %%f28, %%f48, %%f28 \n\t"\ + "fpadd16 %%f18, %%f50, %%f18 \n\t"\ + "fpsub16 %%f22, %%f52, %%f22 \n\t"\ + "fpadd16 %%f26, %%f54, %%f26 \n\t"\ + "fpadd16 %%f30, %%f56, %%f30 \n\t"\ +\ + "fmul8sux16 %%f8, %%f38, %%f48 \n\t"\ + "fmul8sux16 %%f10, %%f40, %%f50 \n\t"\ + "fmul8sux16 %%f10, %%f32, %%f52 \n\t"\ + "fmul8sux16 %%f10, %%f44, %%f54 \n\t"\ + "fmul8sux16 %%f10, %%f36, %%f56 \n\t"\ +\ + "fpadd16 %%f16, %%f48, %%f16 \n\t"\ + "fpsub16 %%f20, %%f48, %%f20 \n\t"\ + "fpsub16 %%f24, %%f48, %%f24 \n\t"\ + "fpadd16 %%f28, %%f48, %%f28 \n\t"\ + "fpadd16 %%f18, %%f50, %%f18 \n\t"\ + "fpsub16 %%f22, %%f52, %%f22 \n\t"\ + "fpadd16 %%f26, %%f54, %%f26 \n\t"\ + "fpadd16 %%f30, %%f56, %%f30 \n\t"\ +\ + "fpadd16 %%f16, %%f8, %%f16 \n\t"\ + "fpsub16 %%f20, %%f8, %%f20 \n\t"\ + "fpsub16 %%f24, %%f8, %%f24 \n\t"\ + "fpadd16 %%f28, %%f8, %%f28 \n\t"\ + "fpadd16 %%f18, %%f10, %%f18 \n\t"\ + "fpsub16 %%f22, %%f10, %%f22 \n\t"\ + /* 4. column */\ + "4: \n\t"\ + "fbe %%fcc2, 5f \n\t"\ + "fpadd16 %%f30, %%f10, %%f30 \n\t"\ + "fmul8ulx16 %%f12, %%f42, %%f48 \n\t"\ + "fmul8ulx16 %%f12, %%f34, %%f50 \n\t"\ + "fmul8ulx16 %%f14, %%f44, %%f52 \n\t"\ + "fmul8ulx16 %%f14, %%f40, %%f54 \n\t"\ + "fmul8ulx16 %%f14, %%f36, %%f56 \n\t"\ + "fmul8ulx16 %%f14, %%f32, %%f58 \n\t"\ +\ + "fpadd16 %%f16, %%f48, %%f16 \n\t"\ + "fpsub16 %%f20, %%f50, %%f20 \n\t"\ + "fpadd16 %%f24, %%f50, %%f24 \n\t"\ + "fpsub16 %%f28, %%f48, %%f28 \n\t"\ + "fpadd16 %%f18, %%f52, %%f18 \n\t"\ + "fpsub16 %%f22, %%f54, %%f22 \n\t"\ + "fpadd16 %%f26, %%f56, %%f26 \n\t"\ + "fpsub16 %%f30, %%f58, %%f30 \n\t"\ +\ + "fmul8sux16 %%f12, %%f42, %%f48 \n\t"\ + "fmul8sux16 %%f12, %%f34, %%f50 \n\t"\ + "fmul8sux16 %%f14, %%f44, %%f52 \n\t"\ + "fmul8sux16 %%f14, %%f40, %%f54 \n\t"\ + "fmul8sux16 %%f14, %%f36, %%f56 \n\t"\ + "fmul8sux16 %%f14, %%f32, %%f58 \n\t"\ +\ + "fpadd16 %%f16, %%f48, %%f16 \n\t"\ + "fpsub16 %%f20, %%f50, %%f20 \n\t"\ + "fpadd16 %%f24, %%f50, %%f24 \n\t"\ + "fpsub16 %%f28, %%f48, %%f28 \n\t"\ + "fpadd16 %%f18, %%f52, %%f18 \n\t"\ + "fpsub16 %%f22, %%f54, %%f22 \n\t"\ + "fpadd16 %%f26, %%f56, %%f26 \n\t"\ + "fpsub16 %%f30, %%f58, %%f30 \n\t"\ +\ + "fpsub16 %%f20, %%f12, %%f20 \n\t"\ + "fpadd16 %%f24, %%f12, %%f24 \n\t"\ + "fpsub16 %%f22, %%f14, %%f22 \n\t"\ + "fpadd16 %%f26, %%f14, %%f26 \n\t"\ + "fpsub16 %%f30, %%f14, %%f30 \n\t"\ + /* final butterfly */\ + "5: \n\t"\ + "fpsub16 %%f16, %%f18, %%f48 \n\t"\ + "fpsub16 %%f20, %%f22, %%f50 \n\t"\ + "fpsub16 %%f24, %%f26, %%f52 \n\t"\ + "fpsub16 %%f28, %%f30, %%f54 \n\t"\ + "fpadd16 %%f16, %%f18, %%f16 \n\t"\ + "fpadd16 %%f20, %%f22, %%f20 \n\t"\ + "fpadd16 %%f24, %%f26, %%f24 \n\t"\ + "fpadd16 %%f28, %%f30, %%f28 \n\t"\ + +#define STOREROWS(out) \ + "std %%f48, [" out "+112] \n\t"\ + "std %%f50, [" out "+96] \n\t"\ + "std %%f52, [" out "+80] \n\t"\ + "std %%f54, [" out "+64] \n\t"\ + "std %%f16, [" out "] \n\t"\ + "std %%f20, [" out "+16] \n\t"\ + "std %%f24, [" out "+32] \n\t"\ + "std %%f28, [" out "+48] \n\t"\ + +#define SCALEROWS \ + "fmul8sux16 %%f46, %%f48, %%f48 \n\t"\ + "fmul8sux16 %%f46, %%f50, %%f50 \n\t"\ + "fmul8sux16 %%f46, %%f52, %%f52 \n\t"\ + "fmul8sux16 %%f46, %%f54, %%f54 \n\t"\ + "fmul8sux16 %%f46, %%f16, %%f16 \n\t"\ + "fmul8sux16 %%f46, %%f20, %%f20 \n\t"\ + "fmul8sux16 %%f46, %%f24, %%f24 \n\t"\ + "fmul8sux16 %%f46, %%f28, %%f28 \n\t"\ + +#define PUTPIXELSCLAMPED(dest) \ + "fpack16 %%f48, %%f14 \n\t"\ + "fpack16 %%f50, %%f12 \n\t"\ + "fpack16 %%f16, %%f0 \n\t"\ + "fpack16 %%f20, %%f2 \n\t"\ + "fpack16 %%f24, %%f4 \n\t"\ + "fpack16 %%f28, %%f6 \n\t"\ + "fpack16 %%f54, %%f8 \n\t"\ + "fpack16 %%f52, %%f10 \n\t"\ + "st %%f0, [%3+" dest "] \n\t"\ + "st %%f2, [%5+" dest "] \n\t"\ + "st %%f4, [%6+" dest "] \n\t"\ + "st %%f6, [%7+" dest "] \n\t"\ + "st %%f8, [%8+" dest "] \n\t"\ + "st %%f10, [%9+" dest "] \n\t"\ + "st %%f12, [%10+" dest "] \n\t"\ + "st %%f14, [%11+" dest "] \n\t"\ + +#define ADDPIXELSCLAMPED(dest) \ + "ldd [%5], %%f18 \n\t"\ + "ld [%3+" dest"], %%f0 \n\t"\ + "ld [%6+" dest"], %%f2 \n\t"\ + "ld [%7+" dest"], %%f4 \n\t"\ + "ld [%8+" dest"], %%f6 \n\t"\ + "ld [%9+" dest"], %%f8 \n\t"\ + "ld [%10+" dest"], %%f10 \n\t"\ + "ld [%11+" dest"], %%f12 \n\t"\ + "ld [%12+" dest"], %%f14 \n\t"\ + "fmul8x16 %%f0, %%f18, %%f0 \n\t"\ + "fmul8x16 %%f2, %%f18, %%f2 \n\t"\ + "fmul8x16 %%f4, %%f18, %%f4 \n\t"\ + "fmul8x16 %%f6, %%f18, %%f6 \n\t"\ + "fmul8x16 %%f8, %%f18, %%f8 \n\t"\ + "fmul8x16 %%f10, %%f18, %%f10 \n\t"\ + "fmul8x16 %%f12, %%f18, %%f12 \n\t"\ + "fmul8x16 %%f14, %%f18, %%f14 \n\t"\ + "fpadd16 %%f0, %%f16, %%f0 \n\t"\ + "fpadd16 %%f2, %%f20, %%f2 \n\t"\ + "fpadd16 %%f4, %%f24, %%f4 \n\t"\ + "fpadd16 %%f6, %%f28, %%f6 \n\t"\ + "fpadd16 %%f8, %%f54, %%f8 \n\t"\ + "fpadd16 %%f10, %%f52, %%f10 \n\t"\ + "fpadd16 %%f12, %%f50, %%f12 \n\t"\ + "fpadd16 %%f14, %%f48, %%f14 \n\t"\ + "fpack16 %%f0, %%f0 \n\t"\ + "fpack16 %%f2, %%f2 \n\t"\ + "fpack16 %%f4, %%f4 \n\t"\ + "fpack16 %%f6, %%f6 \n\t"\ + "fpack16 %%f8, %%f8 \n\t"\ + "fpack16 %%f10, %%f10 \n\t"\ + "fpack16 %%f12, %%f12 \n\t"\ + "fpack16 %%f14, %%f14 \n\t"\ + "st %%f0, [%3+" dest "] \n\t"\ + "st %%f2, [%6+" dest "] \n\t"\ + "st %%f4, [%7+" dest "] \n\t"\ + "st %%f6, [%8+" dest "] \n\t"\ + "st %%f8, [%9+" dest "] \n\t"\ + "st %%f10, [%10+" dest "] \n\t"\ + "st %%f12, [%11+" dest "] \n\t"\ + "st %%f14, [%12+" dest "] \n\t"\ + + +inline void ff_simple_idct_vis(DCTELEM *data) { + int out1, out2, out3, out4; + DECLARE_ALIGNED_8(int16_t, temp[8*8]); + + asm volatile( + INIT_IDCT + +#define ADDROUNDER + + // shift right 16-4=12 + LOADSCALE("%2+8") + IDCT4ROWS + STOREROWS("%3+8") + LOADSCALE("%2+0") + IDCT4ROWS + "std %%f48, [%3+112] \n\t" + "std %%f50, [%3+96] \n\t" + "std %%f52, [%3+80] \n\t" + "std %%f54, [%3+64] \n\t" + + // shift right 16+4 + "ldd [%3+8], %%f18 \n\t" + "ldd [%3+24], %%f22 \n\t" + "ldd [%3+40], %%f26 \n\t" + "ldd [%3+56], %%f30 \n\t" + TRANSPOSE + IDCT4ROWS + SCALEROWS + STOREROWS("%2+0") + LOAD("%3+64") + TRANSPOSE + IDCT4ROWS + SCALEROWS + STOREROWS("%2+8") + + : "=r" (out1), "=r" (out2), "=r" (out3), "=r" (out4) + : "0" (scale), "1" (coeffs), "2" (data), "3" (temp) + ); +} + +void ff_simple_idct_put_vis(uint8_t *dest, int line_size, DCTELEM *data) { + int out1, out2, out3, out4, out5; + int r1, r2, r3, r4, r5, r6, r7; + + asm volatile( + "wr %%g0, 0x8, %%gsr \n\t" + + INIT_IDCT + + "add %3, %4, %5 \n\t" + "add %5, %4, %6 \n\t" + "add %6, %4, %7 \n\t" + "add %7, %4, %8 \n\t" + "add %8, %4, %9 \n\t" + "add %9, %4, %10 \n\t" + "add %10, %4, %11 \n\t" + + // shift right 16-4=12 + LOADSCALE("%2+8") + IDCT4ROWS + STOREROWS("%2+8") + LOADSCALE("%2+0") + IDCT4ROWS + "std %%f48, [%2+112] \n\t" + "std %%f50, [%2+96] \n\t" + "std %%f52, [%2+80] \n\t" + "std %%f54, [%2+64] \n\t" + +#undef ADDROUNDER +#define ADDROUNDER "fpadd16 %%f28, %%f46, %%f28 \n\t" + + // shift right 16+4 + "ldd [%2+8], %%f18 \n\t" + "ldd [%2+24], %%f22 \n\t" + "ldd [%2+40], %%f26 \n\t" + "ldd [%2+56], %%f30 \n\t" + TRANSPOSE + IDCT4ROWS + PUTPIXELSCLAMPED("0") + LOAD("%2+64") + TRANSPOSE + IDCT4ROWS + PUTPIXELSCLAMPED("4") + + : "=r" (out1), "=r" (out2), "=r" (out3), "=r" (out4), "=r" (out5), + "=r" (r1), "=r" (r2), "=r" (r3), "=r" (r4), "=r" (r5), "=r" (r6), "=r" (r7) + : "0" (rounder), "1" (coeffs), "2" (data), "3" (dest), "4" (line_size) + ); +} + +void ff_simple_idct_add_vis(uint8_t *dest, int line_size, DCTELEM *data) { + int out1, out2, out3, out4, out5, out6; + int r1, r2, r3, r4, r5, r6, r7; + + asm volatile( + "wr %%g0, 0x8, %%gsr \n\t" + + INIT_IDCT + + "add %3, %4, %6 \n\t" + "add %6, %4, %7 \n\t" + "add %7, %4, %8 \n\t" + "add %8, %4, %9 \n\t" + "add %9, %4, %10 \n\t" + "add %10, %4, %11 \n\t" + "add %11, %4, %12 \n\t" + +#undef ADDROUNDER +#define ADDROUNDER + + // shift right 16-4=12 + LOADSCALE("%2+8") + IDCT4ROWS + STOREROWS("%2+8") + LOADSCALE("%2+0") + IDCT4ROWS + "std %%f48, [%2+112] \n\t" + "std %%f50, [%2+96] \n\t" + "std %%f52, [%2+80] \n\t" + "std %%f54, [%2+64] \n\t" + +#undef ADDROUNDER +#define ADDROUNDER "fpadd16 %%f28, %%f46, %%f28 \n\t" + + // shift right 16+4 + "ldd [%2+8], %%f18 \n\t" + "ldd [%2+24], %%f22 \n\t" + "ldd [%2+40], %%f26 \n\t" + "ldd [%2+56], %%f30 \n\t" + TRANSPOSE + IDCT4ROWS + ADDPIXELSCLAMPED("0") + LOAD("%2+64") + TRANSPOSE + IDCT4ROWS + ADDPIXELSCLAMPED("4") + + : "=r" (out1), "=r" (out2), "=r" (out3), "=r" (out4), "=r" (out5), "=r" (out6), + "=r" (r1), "=r" (r2), "=r" (r3), "=r" (r4), "=r" (r5), "=r" (r6), "=r" (r7) + : "0" (rounder), "1" (coeffs), "2" (data), "3" (dest), "4" (line_size), "5" (expand) + ); +} diff --git a/contrib/ffmpeg/libavcodec/sparc/vis.h b/contrib/ffmpeg/libavcodec/sparc/vis.h new file mode 100644 index 000000000..a0b09f5bc --- /dev/null +++ b/contrib/ffmpeg/libavcodec/sparc/vis.h @@ -0,0 +1,332 @@ +/* + * vis.h + * Copyright (C) 2003 David S. Miller <davem@redhat.com> + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +/* You may be asking why I hard-code the instruction opcodes and don't + * use the normal VIS assembler mnenomics for the VIS instructions. + * + * The reason is that Sun, in their infinite wisdom, decided that a binary + * using a VIS instruction will cause it to be marked (in the ELF headers) + * as doing so, and this prevents the OS from loading such binaries if the + * current cpu doesn't have VIS. There is no way to easily override this + * behavior of the assembler that I am aware of. + * + * This totally defeats what libmpeg2 is trying to do which is allow a + * single binary to be created, and then detect the availability of VIS + * at runtime. + * + * I'm not saying that tainting the binary by default is bad, rather I'm + * saying that not providing a way to override this easily unnecessarily + * ties people's hands. + * + * Thus, we do the opcode encoding by hand and output 32-bit words in + * the assembler to keep the binary from becoming tainted. + */ + +#ifndef FFMPEG_VIS_H +#define FFMPEG_VIS_H + +#define vis_opc_base ((0x1 << 31) | (0x36 << 19)) +#define vis_opf(X) ((X) << 5) +#define vis_sreg(X) (X) +#define vis_dreg(X) (((X)&0x1f)|((X)>>5)) +#define vis_rs1_s(X) (vis_sreg(X) << 14) +#define vis_rs1_d(X) (vis_dreg(X) << 14) +#define vis_rs2_s(X) (vis_sreg(X) << 0) +#define vis_rs2_d(X) (vis_dreg(X) << 0) +#define vis_rd_s(X) (vis_sreg(X) << 25) +#define vis_rd_d(X) (vis_dreg(X) << 25) + +#define vis_ss2s(opf,rs1,rs2,rd) \ + __asm__ __volatile__ (".word %0" \ + : : "i" (vis_opc_base | vis_opf(opf) | \ + vis_rs1_s(rs1) | \ + vis_rs2_s(rs2) | \ + vis_rd_s(rd))) + +#define vis_dd2d(opf,rs1,rs2,rd) \ + __asm__ __volatile__ (".word %0" \ + : : "i" (vis_opc_base | vis_opf(opf) | \ + vis_rs1_d(rs1) | \ + vis_rs2_d(rs2) | \ + vis_rd_d(rd))) + +#define vis_ss2d(opf,rs1,rs2,rd) \ + __asm__ __volatile__ (".word %0" \ + : : "i" (vis_opc_base | vis_opf(opf) | \ + vis_rs1_s(rs1) | \ + vis_rs2_s(rs2) | \ + vis_rd_d(rd))) + +#define vis_sd2d(opf,rs1,rs2,rd) \ + __asm__ __volatile__ (".word %0" \ + : : "i" (vis_opc_base | vis_opf(opf) | \ + vis_rs1_s(rs1) | \ + vis_rs2_d(rs2) | \ + vis_rd_d(rd))) + +#define vis_d2s(opf,rs2,rd) \ + __asm__ __volatile__ (".word %0" \ + : : "i" (vis_opc_base | vis_opf(opf) | \ + vis_rs2_d(rs2) | \ + vis_rd_s(rd))) + +#define vis_s2d(opf,rs2,rd) \ + __asm__ __volatile__ (".word %0" \ + : : "i" (vis_opc_base | vis_opf(opf) | \ + vis_rs2_s(rs2) | \ + vis_rd_d(rd))) + +#define vis_d12d(opf,rs1,rd) \ + __asm__ __volatile__ (".word %0" \ + : : "i" (vis_opc_base | vis_opf(opf) | \ + vis_rs1_d(rs1) | \ + vis_rd_d(rd))) + +#define vis_d22d(opf,rs2,rd) \ + __asm__ __volatile__ (".word %0" \ + : : "i" (vis_opc_base | vis_opf(opf) | \ + vis_rs2_d(rs2) | \ + vis_rd_d(rd))) + +#define vis_s12s(opf,rs1,rd) \ + __asm__ __volatile__ (".word %0" \ + : : "i" (vis_opc_base | vis_opf(opf) | \ + vis_rs1_s(rs1) | \ + vis_rd_s(rd))) + +#define vis_s22s(opf,rs2,rd) \ + __asm__ __volatile__ (".word %0" \ + : : "i" (vis_opc_base | vis_opf(opf) | \ + vis_rs2_s(rs2) | \ + vis_rd_s(rd))) + +#define vis_s(opf,rd) \ + __asm__ __volatile__ (".word %0" \ + : : "i" (vis_opc_base | vis_opf(opf) | \ + vis_rd_s(rd))) + +#define vis_d(opf,rd) \ + __asm__ __volatile__ (".word %0" \ + : : "i" (vis_opc_base | vis_opf(opf) | \ + vis_rd_d(rd))) + +#define vis_r2m(op,rd,mem) \ + __asm__ __volatile__ (#op "\t%%f" #rd ", [%0]" : : "r" (&(mem)) ) + +#define vis_r2m_2(op,rd,mem1,mem2) \ + __asm__ __volatile__ (#op "\t%%f" #rd ", [%0 + %1]" : : "r" (mem1), "r" (mem2) ) + +#define vis_m2r(op,mem,rd) \ + __asm__ __volatile__ (#op "\t[%0], %%f" #rd : : "r" (&(mem)) ) + +#define vis_m2r_2(op,mem1,mem2,rd) \ + __asm__ __volatile__ (#op "\t[%0 + %1], %%f" #rd : : "r" (mem1), "r" (mem2) ) + +static inline void vis_set_gsr(unsigned int _val) +{ + register unsigned int val asm("g1"); + + val = _val; + __asm__ __volatile__(".word 0xa7804000" + : : "r" (val)); +} + +#define VIS_GSR_ALIGNADDR_MASK 0x0000007 +#define VIS_GSR_ALIGNADDR_SHIFT 0 +#define VIS_GSR_SCALEFACT_MASK 0x0000078 +#define VIS_GSR_SCALEFACT_SHIFT 3 + +#define vis_ld32(mem,rs1) vis_m2r(ld, mem, rs1) +#define vis_ld32_2(mem1,mem2,rs1) vis_m2r_2(ld, mem1, mem2, rs1) +#define vis_st32(rs1,mem) vis_r2m(st, rs1, mem) +#define vis_st32_2(rs1,mem1,mem2) vis_r2m_2(st, rs1, mem1, mem2) +#define vis_ld64(mem,rs1) vis_m2r(ldd, mem, rs1) +#define vis_ld64_2(mem1,mem2,rs1) vis_m2r_2(ldd, mem1, mem2, rs1) +#define vis_st64(rs1,mem) vis_r2m(std, rs1, mem) +#define vis_st64_2(rs1,mem1,mem2) vis_r2m_2(std, rs1, mem1, mem2) + +#define vis_ldblk(mem, rd) \ +do { register void *__mem asm("g1"); \ + __mem = &(mem); \ + __asm__ __volatile__(".word 0xc1985e00 | %1" \ + : \ + : "r" (__mem), \ + "i" (vis_rd_d(rd)) \ + : "memory"); \ +} while (0) + +#define vis_stblk(rd, mem) \ +do { register void *__mem asm("g1"); \ + __mem = &(mem); \ + __asm__ __volatile__(".word 0xc1b85e00 | %1" \ + : \ + : "r" (__mem), \ + "i" (vis_rd_d(rd)) \ + : "memory"); \ +} while (0) + +#define vis_membar_storestore() \ + __asm__ __volatile__(".word 0x8143e008" : : : "memory") + +#define vis_membar_sync() \ + __asm__ __volatile__(".word 0x8143e040" : : : "memory") + +/* 16 and 32 bit partitioned addition and subtraction. The normal + * versions perform 4 16-bit or 2 32-bit additions or subtractions. + * The 's' versions perform 2 16-bit or 1 32-bit additions or + * subtractions. + */ + +#define vis_padd16(rs1,rs2,rd) vis_dd2d(0x50, rs1, rs2, rd) +#define vis_padd16s(rs1,rs2,rd) vis_ss2s(0x51, rs1, rs2, rd) +#define vis_padd32(rs1,rs2,rd) vis_dd2d(0x52, rs1, rs2, rd) +#define vis_padd32s(rs1,rs2,rd) vis_ss2s(0x53, rs1, rs2, rd) +#define vis_psub16(rs1,rs2,rd) vis_dd2d(0x54, rs1, rs2, rd) +#define vis_psub16s(rs1,rs2,rd) vis_ss2s(0x55, rs1, rs2, rd) +#define vis_psub32(rs1,rs2,rd) vis_dd2d(0x56, rs1, rs2, rd) +#define vis_psub32s(rs1,rs2,rd) vis_ss2s(0x57, rs1, rs2, rd) + +/* Pixel formatting instructions. */ + +#define vis_pack16(rs2,rd) vis_d2s( 0x3b, rs2, rd) +#define vis_pack32(rs1,rs2,rd) vis_dd2d(0x3a, rs1, rs2, rd) +#define vis_packfix(rs2,rd) vis_d2s( 0x3d, rs2, rd) +#define vis_expand(rs2,rd) vis_s2d( 0x4d, rs2, rd) +#define vis_pmerge(rs1,rs2,rd) vis_ss2d(0x4b, rs1, rs2, rd) + +/* Partitioned multiply instructions. */ + +#define vis_mul8x16(rs1,rs2,rd) vis_sd2d(0x31, rs1, rs2, rd) +#define vis_mul8x16au(rs1,rs2,rd) vis_ss2d(0x33, rs1, rs2, rd) +#define vis_mul8x16al(rs1,rs2,rd) vis_ss2d(0x35, rs1, rs2, rd) +#define vis_mul8sux16(rs1,rs2,rd) vis_dd2d(0x36, rs1, rs2, rd) +#define vis_mul8ulx16(rs1,rs2,rd) vis_dd2d(0x37, rs1, rs2, rd) +#define vis_muld8sux16(rs1,rs2,rd) vis_ss2d(0x38, rs1, rs2, rd) +#define vis_muld8ulx16(rs1,rs2,rd) vis_ss2d(0x39, rs1, rs2, rd) + +/* Alignment instructions. */ + +static inline void *vis_alignaddr(void *_ptr) +{ + register void *ptr asm("g1"); + + ptr = _ptr; + + __asm__ __volatile__(".word %2" + : "=&r" (ptr) + : "0" (ptr), + "i" (vis_opc_base | vis_opf(0x18) | + vis_rs1_s(1) | + vis_rs2_s(0) | + vis_rd_s(1))); + + return ptr; +} + +static inline void vis_alignaddr_g0(void *_ptr) +{ + register void *ptr asm("g1"); + + ptr = _ptr; + + __asm__ __volatile__(".word %2" + : "=&r" (ptr) + : "0" (ptr), + "i" (vis_opc_base | vis_opf(0x18) | + vis_rs1_s(1) | + vis_rs2_s(0) | + vis_rd_s(0))); +} + +static inline void *vis_alignaddrl(void *_ptr) +{ + register void *ptr asm("g1"); + + ptr = _ptr; + + __asm__ __volatile__(".word %2" + : "=&r" (ptr) + : "0" (ptr), + "i" (vis_opc_base | vis_opf(0x19) | + vis_rs1_s(1) | + vis_rs2_s(0) | + vis_rd_s(1))); + + return ptr; +} + +static inline void vis_alignaddrl_g0(void *_ptr) +{ + register void *ptr asm("g1"); + + ptr = _ptr; + + __asm__ __volatile__(".word %2" + : "=&r" (ptr) + : "0" (ptr), + "i" (vis_opc_base | vis_opf(0x19) | + vis_rs1_s(1) | + vis_rs2_s(0) | + vis_rd_s(0))); +} + +#define vis_faligndata(rs1,rs2,rd) vis_dd2d(0x48, rs1, rs2, rd) + +/* Logical operate instructions. */ + +#define vis_fzero(rd) vis_d( 0x60, rd) +#define vis_fzeros(rd) vis_s( 0x61, rd) +#define vis_fone(rd) vis_d( 0x7e, rd) +#define vis_fones(rd) vis_s( 0x7f, rd) +#define vis_src1(rs1,rd) vis_d12d(0x74, rs1, rd) +#define vis_src1s(rs1,rd) vis_s12s(0x75, rs1, rd) +#define vis_src2(rs2,rd) vis_d22d(0x78, rs2, rd) +#define vis_src2s(rs2,rd) vis_s22s(0x79, rs2, rd) +#define vis_not1(rs1,rd) vis_d12d(0x6a, rs1, rd) +#define vis_not1s(rs1,rd) vis_s12s(0x6b, rs1, rd) +#define vis_not2(rs2,rd) vis_d22d(0x66, rs2, rd) +#define vis_not2s(rs2,rd) vis_s22s(0x67, rs2, rd) +#define vis_or(rs1,rs2,rd) vis_dd2d(0x7c, rs1, rs2, rd) +#define vis_ors(rs1,rs2,rd) vis_ss2s(0x7d, rs1, rs2, rd) +#define vis_nor(rs1,rs2,rd) vis_dd2d(0x62, rs1, rs2, rd) +#define vis_nors(rs1,rs2,rd) vis_ss2s(0x63, rs1, rs2, rd) +#define vis_and(rs1,rs2,rd) vis_dd2d(0x70, rs1, rs2, rd) +#define vis_ands(rs1,rs2,rd) vis_ss2s(0x71, rs1, rs2, rd) +#define vis_nand(rs1,rs2,rd) vis_dd2d(0x6e, rs1, rs2, rd) +#define vis_nands(rs1,rs2,rd) vis_ss2s(0x6f, rs1, rs2, rd) +#define vis_xor(rs1,rs2,rd) vis_dd2d(0x6c, rs1, rs2, rd) +#define vis_xors(rs1,rs2,rd) vis_ss2s(0x6d, rs1, rs2, rd) +#define vis_xnor(rs1,rs2,rd) vis_dd2d(0x72, rs1, rs2, rd) +#define vis_xnors(rs1,rs2,rd) vis_ss2s(0x73, rs1, rs2, rd) +#define vis_ornot1(rs1,rs2,rd) vis_dd2d(0x7a, rs1, rs2, rd) +#define vis_ornot1s(rs1,rs2,rd) vis_ss2s(0x7b, rs1, rs2, rd) +#define vis_ornot2(rs1,rs2,rd) vis_dd2d(0x76, rs1, rs2, rd) +#define vis_ornot2s(rs1,rs2,rd) vis_ss2s(0x77, rs1, rs2, rd) +#define vis_andnot1(rs1,rs2,rd) vis_dd2d(0x68, rs1, rs2, rd) +#define vis_andnot1s(rs1,rs2,rd) vis_ss2s(0x69, rs1, rs2, rd) +#define vis_andnot2(rs1,rs2,rd) vis_dd2d(0x64, rs1, rs2, rd) +#define vis_andnot2s(rs1,rs2,rd) vis_ss2s(0x65, rs1, rs2, rd) + +/* Pixel component distance. */ + +#define vis_pdist(rs1,rs2,rd) vis_dd2d(0x3e, rs1, rs2, rd) + +#endif /* FFMPEG_VIS_H */ |