nengel@2: /*
nengel@2:  * Copyright (c) 2002 Brian Foley
nengel@2:  * Copyright (c) 2002 Dieter Shirley
nengel@2:  * Copyright (c) 2003-2004 Romain Dolbeau <romain@dolbeau.org>
nengel@2:  *
nengel@2:  * This file is part of FFmpeg.
nengel@2:  *
nengel@2:  * FFmpeg is free software; you can redistribute it and/or
nengel@2:  * modify it under the terms of the GNU Lesser General Public
nengel@2:  * License as published by the Free Software Foundation; either
nengel@2:  * version 2.1 of the License, or (at your option) any later version.
nengel@2:  *
nengel@2:  * FFmpeg is distributed in the hope that it will be useful,
nengel@2:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
nengel@2:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
nengel@2:  * Lesser General Public License for more details.
nengel@2:  *
nengel@2:  * You should have received a copy of the GNU Lesser General Public
nengel@2:  * License along with FFmpeg; if not, write to the Free Software
nengel@2:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
nengel@2:  */
nengel@2: 
nengel@2: #include "config.h"
nengel@2: #if HAVE_ALTIVEC_H
nengel@2: #include <altivec.h>
nengel@2: #endif
nengel@2: #include "libavcodec/dsputil.h"
nengel@2: #include "dsputil_ppc.h"
nengel@2: #include "util_altivec.h"
nengel@2: #include "types_altivec.h"
nengel@2: #include "dsputil_altivec.h"
nengel@2: 
nengel@2: 
nengel@2: static void get_pixels_altivec(DCTELEM *restrict block, const uint8_t *pixels, int line_size)
nengel@2: {
nengel@2:     int i;
nengel@2:     vector unsigned char perm, bytes, *pixv;
nengel@2:     const vector unsigned char zero = (const vector unsigned char)vec_splat_u8(0);
nengel@2:     vector signed short shorts;
nengel@2: 
nengel@2:     for (i = 0; i < 8; i++) {
nengel@2:         // Read potentially unaligned pixels.
nengel@2:         // We're reading 16 pixels, and actually only want 8,
nengel@2:         // but we simply ignore the extras.
nengel@2:         perm = vec_lvsl(0, pixels);
nengel@2:         pixv = (vector unsigned char *) pixels;
nengel@2:         bytes = vec_perm(pixv[0], pixv[1], perm);
nengel@2: 
nengel@2:         // convert the bytes into shorts
nengel@2:         shorts = (vector signed short)vec_mergeh(zero, bytes);
nengel@2: 
nengel@2:         // save the data to the block, we assume the block is 16-byte aligned
nengel@2:         vec_st(shorts, i*16, (vector signed short*)block);
nengel@2: 
nengel@2:         pixels += line_size;
nengel@2:     }
nengel@2: }
nengel@2: 
nengel@2: static void diff_pixels_altivec(DCTELEM *restrict block, const uint8_t *s1,
nengel@2:         const uint8_t *s2, int stride)
nengel@2: {
nengel@2:     int i;
nengel@2:     vector unsigned char perm, bytes, *pixv;
nengel@2:     const vector unsigned char zero = (const vector unsigned char)vec_splat_u8(0);
nengel@2:     vector signed short shorts1, shorts2;
nengel@2: 
nengel@2:     for (i = 0; i < 4; i++) {
nengel@2:         // Read potentially unaligned pixels
nengel@2:         // We're reading 16 pixels, and actually only want 8,
nengel@2:         // but we simply ignore the extras.
nengel@2:         perm = vec_lvsl(0, s1);
nengel@2:         pixv = (vector unsigned char *) s1;
nengel@2:         bytes = vec_perm(pixv[0], pixv[1], perm);
nengel@2: 
nengel@2:         // convert the bytes into shorts
nengel@2:         shorts1 = (vector signed short)vec_mergeh(zero, bytes);
nengel@2: 
nengel@2:         // Do the same for the second block of pixels
nengel@2:         perm = vec_lvsl(0, s2);
nengel@2:         pixv = (vector unsigned char *) s2;
nengel@2:         bytes = vec_perm(pixv[0], pixv[1], perm);
nengel@2: 
nengel@2:         // convert the bytes into shorts
nengel@2:         shorts2 = (vector signed short)vec_mergeh(zero, bytes);
nengel@2: 
nengel@2:         // Do the subtraction
nengel@2:         shorts1 = vec_sub(shorts1, shorts2);
nengel@2: 
nengel@2:         // save the data to the block, we assume the block is 16-byte aligned
nengel@2:         vec_st(shorts1, 0, (vector signed short*)block);
nengel@2: 
nengel@2:         s1 += stride;
nengel@2:         s2 += stride;
nengel@2:         block += 8;
nengel@2: 
nengel@2: 
nengel@2:         // The code below is a copy of the code above... This is a manual
nengel@2:         // unroll.
nengel@2: 
nengel@2:         // Read potentially unaligned pixels
nengel@2:         // We're reading 16 pixels, and actually only want 8,
nengel@2:         // but we simply ignore the extras.
nengel@2:         perm = vec_lvsl(0, s1);
nengel@2:         pixv = (vector unsigned char *) s1;
nengel@2:         bytes = vec_perm(pixv[0], pixv[1], perm);
nengel@2: 
nengel@2:         // convert the bytes into shorts
nengel@2:         shorts1 = (vector signed short)vec_mergeh(zero, bytes);
nengel@2: 
nengel@2:         // Do the same for the second block of pixels
nengel@2:         perm = vec_lvsl(0, s2);
nengel@2:         pixv = (vector unsigned char *) s2;
nengel@2:         bytes = vec_perm(pixv[0], pixv[1], perm);
nengel@2: 
nengel@2:         // convert the bytes into shorts
nengel@2:         shorts2 = (vector signed short)vec_mergeh(zero, bytes);
nengel@2: 
nengel@2:         // Do the subtraction
nengel@2:         shorts1 = vec_sub(shorts1, shorts2);
nengel@2: 
nengel@2:         // save the data to the block, we assume the block is 16-byte aligned
nengel@2:         vec_st(shorts1, 0, (vector signed short*)block);
nengel@2: 
nengel@2:         s1 += stride;
nengel@2:         s2 += stride;
nengel@2:         block += 8;
nengel@2:     }
nengel@2: }
nengel@2: 
nengel@2: 
nengel@2: static void clear_block_altivec(DCTELEM *block) {
nengel@2:     LOAD_ZERO;
nengel@2:     vec_st(zero_s16v,   0, block);
nengel@2:     vec_st(zero_s16v,  16, block);
nengel@2:     vec_st(zero_s16v,  32, block);
nengel@2:     vec_st(zero_s16v,  48, block);
nengel@2:     vec_st(zero_s16v,  64, block);
nengel@2:     vec_st(zero_s16v,  80, block);
nengel@2:     vec_st(zero_s16v,  96, block);
nengel@2:     vec_st(zero_s16v, 112, block);
nengel@2: }
nengel@2: 
nengel@2: 
nengel@2: 
nengel@2: /* next one assumes that ((line_size % 16) == 0) */
nengel@2: void put_pixels16_altivec(uint8_t *block, const uint8_t *pixels, int line_size, int h)
nengel@2: {
nengel@2: POWERPC_PERF_DECLARE(altivec_put_pixels16_num, 1);
nengel@2:     register vector unsigned char pixelsv1, pixelsv2;
nengel@2:     register vector unsigned char pixelsv1B, pixelsv2B;
nengel@2:     register vector unsigned char pixelsv1C, pixelsv2C;
nengel@2:     register vector unsigned char pixelsv1D, pixelsv2D;
nengel@2: 
nengel@2:     register vector unsigned char perm = vec_lvsl(0, pixels);
nengel@2:     int i;
nengel@2:     register int line_size_2 = line_size << 1;
nengel@2:     register int line_size_3 = line_size + line_size_2;
nengel@2:     register int line_size_4 = line_size << 2;
nengel@2: 
nengel@2: POWERPC_PERF_START_COUNT(altivec_put_pixels16_num, 1);
nengel@2: // hand-unrolling the loop by 4 gains about 15%
nengel@2: // mininum execution time goes from 74 to 60 cycles
nengel@2: // it's faster than -funroll-loops, but using
nengel@2: // -funroll-loops w/ this is bad - 74 cycles again.
nengel@2: // all this is on a 7450, tuning for the 7450
nengel@2: #if 0
nengel@2:     for (i = 0; i < h; i++) {
nengel@2:         pixelsv1 = vec_ld(0, pixels);
nengel@2:         pixelsv2 = vec_ld(16, pixels);
nengel@2:         vec_st(vec_perm(pixelsv1, pixelsv2, perm),
nengel@2:                0, block);
nengel@2:         pixels+=line_size;
nengel@2:         block +=line_size;
nengel@2:     }
nengel@2: #else
nengel@2:     for (i = 0; i < h; i += 4) {
nengel@2:         pixelsv1  = vec_ld( 0, pixels);
nengel@2:         pixelsv2  = vec_ld(15, pixels);
nengel@2:         pixelsv1B = vec_ld(line_size, pixels);
nengel@2:         pixelsv2B = vec_ld(15 + line_size, pixels);
nengel@2:         pixelsv1C = vec_ld(line_size_2, pixels);
nengel@2:         pixelsv2C = vec_ld(15 + line_size_2, pixels);
nengel@2:         pixelsv1D = vec_ld(line_size_3, pixels);
nengel@2:         pixelsv2D = vec_ld(15 + line_size_3, pixels);
nengel@2:         vec_st(vec_perm(pixelsv1, pixelsv2, perm),
nengel@2:                0, (unsigned char*)block);
nengel@2:         vec_st(vec_perm(pixelsv1B, pixelsv2B, perm),
nengel@2:                line_size, (unsigned char*)block);
nengel@2:         vec_st(vec_perm(pixelsv1C, pixelsv2C, perm),
nengel@2:                line_size_2, (unsigned char*)block);
nengel@2:         vec_st(vec_perm(pixelsv1D, pixelsv2D, perm),
nengel@2:                line_size_3, (unsigned char*)block);
nengel@2:         pixels+=line_size_4;
nengel@2:         block +=line_size_4;
nengel@2:     }
nengel@2: #endif
nengel@2: POWERPC_PERF_STOP_COUNT(altivec_put_pixels16_num, 1);
nengel@2: }
nengel@2: 
nengel@2: /* next one assumes that ((line_size % 16) == 0) */
nengel@2: #define op_avg(a,b)  a = ( ((a)|(b)) - ((((a)^(b))&0xFEFEFEFEUL)>>1) )
nengel@2: void avg_pixels16_altivec(uint8_t *block, const uint8_t *pixels, int line_size, int h)
nengel@2: {
nengel@2: POWERPC_PERF_DECLARE(altivec_avg_pixels16_num, 1);
nengel@2:     register vector unsigned char pixelsv1, pixelsv2, pixelsv, blockv;
nengel@2:     register vector unsigned char perm = vec_lvsl(0, pixels);
nengel@2:     int i;
nengel@2: 
nengel@2: POWERPC_PERF_START_COUNT(altivec_avg_pixels16_num, 1);
nengel@2: 
nengel@2:     for (i = 0; i < h; i++) {
nengel@2:         pixelsv1 = vec_ld( 0, pixels);
nengel@2:         pixelsv2 = vec_ld(16,pixels);
nengel@2:         blockv = vec_ld(0, block);
nengel@2:         pixelsv = vec_perm(pixelsv1, pixelsv2, perm);
nengel@2:         blockv = vec_avg(blockv,pixelsv);
nengel@2:         vec_st(blockv, 0, (unsigned char*)block);
nengel@2:         pixels+=line_size;
nengel@2:         block +=line_size;
nengel@2:     }
nengel@2: 
nengel@2: POWERPC_PERF_STOP_COUNT(altivec_avg_pixels16_num, 1);
nengel@2: }
nengel@2: 
nengel@2: /* next one assumes that ((line_size % 8) == 0) */
nengel@2: static void avg_pixels8_altivec(uint8_t * block, const uint8_t * pixels, int line_size, int h)
nengel@2: {
nengel@2: POWERPC_PERF_DECLARE(altivec_avg_pixels8_num, 1);
nengel@2:     register vector unsigned char pixelsv1, pixelsv2, pixelsv, blockv;
nengel@2:     int i;
nengel@2: 
nengel@2: POWERPC_PERF_START_COUNT(altivec_avg_pixels8_num, 1);
nengel@2: 
nengel@2:    for (i = 0; i < h; i++) {
nengel@2:        /* block is 8 bytes-aligned, so we're either in the
nengel@2:           left block (16 bytes-aligned) or in the right block (not) */
nengel@2:        int rightside = ((unsigned long)block & 0x0000000F);
nengel@2: 
nengel@2:        blockv = vec_ld(0, block);
nengel@2:        pixelsv1 = vec_ld( 0, pixels);
nengel@2:        pixelsv2 = vec_ld(16, pixels);
nengel@2:        pixelsv = vec_perm(pixelsv1, pixelsv2, vec_lvsl(0, pixels));
nengel@2: 
nengel@2:        if (rightside) {
nengel@2:            pixelsv = vec_perm(blockv, pixelsv, vcprm(0,1,s0,s1));
nengel@2:        } else {
nengel@2:            pixelsv = vec_perm(blockv, pixelsv, vcprm(s0,s1,2,3));
nengel@2:        }
nengel@2: 
nengel@2:        blockv = vec_avg(blockv, pixelsv);
nengel@2: 
nengel@2:        vec_st(blockv, 0, block);
nengel@2: 
nengel@2:        pixels += line_size;
nengel@2:        block += line_size;
nengel@2:    }
nengel@2: 
nengel@2: POWERPC_PERF_STOP_COUNT(altivec_avg_pixels8_num, 1);
nengel@2: }
nengel@2: 
nengel@2: /* next one assumes that ((line_size % 8) == 0) */
nengel@2: static void put_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, int line_size, int h)
nengel@2: {
nengel@2: POWERPC_PERF_DECLARE(altivec_put_pixels8_xy2_num, 1);
nengel@2:     register int i;
nengel@2:     register vector unsigned char pixelsv1, pixelsv2, pixelsavg;
nengel@2:     register vector unsigned char blockv, temp1, temp2;
nengel@2:     register vector unsigned short pixelssum1, pixelssum2, temp3;
nengel@2:     register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
nengel@2:     register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2);
nengel@2: 
nengel@2:     temp1 = vec_ld(0, pixels);
nengel@2:     temp2 = vec_ld(16, pixels);
nengel@2:     pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels));
nengel@2:     if ((((unsigned long)pixels) & 0x0000000F) ==  0x0000000F) {
nengel@2:         pixelsv2 = temp2;
nengel@2:     } else {
nengel@2:         pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(1, pixels));
nengel@2:     }
nengel@2:     pixelsv1 = vec_mergeh(vczero, pixelsv1);
nengel@2:     pixelsv2 = vec_mergeh(vczero, pixelsv2);
nengel@2:     pixelssum1 = vec_add((vector unsigned short)pixelsv1,
nengel@2:                          (vector unsigned short)pixelsv2);
nengel@2:     pixelssum1 = vec_add(pixelssum1, vctwo);
nengel@2: 
nengel@2: POWERPC_PERF_START_COUNT(altivec_put_pixels8_xy2_num, 1);
nengel@2:     for (i = 0; i < h ; i++) {
nengel@2:         int rightside = ((unsigned long)block & 0x0000000F);
nengel@2:         blockv = vec_ld(0, block);
nengel@2: 
nengel@2:         temp1 = vec_ld(line_size, pixels);
nengel@2:         temp2 = vec_ld(line_size + 16, pixels);
nengel@2:         pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels));
nengel@2:         if (((((unsigned long)pixels) + line_size) & 0x0000000F) ==  0x0000000F) {
nengel@2:             pixelsv2 = temp2;
nengel@2:         } else {
nengel@2:             pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(line_size + 1, pixels));
nengel@2:         }
nengel@2: 
nengel@2:         pixelsv1 = vec_mergeh(vczero, pixelsv1);
nengel@2:         pixelsv2 = vec_mergeh(vczero, pixelsv2);
nengel@2:         pixelssum2 = vec_add((vector unsigned short)pixelsv1,
nengel@2:                              (vector unsigned short)pixelsv2);
nengel@2:         temp3 = vec_add(pixelssum1, pixelssum2);
nengel@2:         temp3 = vec_sra(temp3, vctwo);
nengel@2:         pixelssum1 = vec_add(pixelssum2, vctwo);
nengel@2:         pixelsavg = vec_packsu(temp3, (vector unsigned short) vczero);
nengel@2: 
nengel@2:         if (rightside) {
nengel@2:             blockv = vec_perm(blockv, pixelsavg, vcprm(0, 1, s0, s1));
nengel@2:         } else {
nengel@2:             blockv = vec_perm(blockv, pixelsavg, vcprm(s0, s1, 2, 3));
nengel@2:         }
nengel@2: 
nengel@2:         vec_st(blockv, 0, block);
nengel@2: 
nengel@2:         block += line_size;
nengel@2:         pixels += line_size;
nengel@2:     }
nengel@2: 
nengel@2: POWERPC_PERF_STOP_COUNT(altivec_put_pixels8_xy2_num, 1);
nengel@2: }
nengel@2: 
nengel@2: /* next one assumes that ((line_size % 8) == 0) */
nengel@2: static void put_no_rnd_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, int line_size, int h)
nengel@2: {
nengel@2: POWERPC_PERF_DECLARE(altivec_put_no_rnd_pixels8_xy2_num, 1);
nengel@2:     register int i;
nengel@2:     register vector unsigned char pixelsv1, pixelsv2, pixelsavg;
nengel@2:     register vector unsigned char blockv, temp1, temp2;
nengel@2:     register vector unsigned short pixelssum1, pixelssum2, temp3;
nengel@2:     register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
nengel@2:     register const vector unsigned short vcone = (const vector unsigned short)vec_splat_u16(1);
nengel@2:     register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2);
nengel@2: 
nengel@2:     temp1 = vec_ld(0, pixels);
nengel@2:     temp2 = vec_ld(16, pixels);
nengel@2:     pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels));
nengel@2:     if ((((unsigned long)pixels) & 0x0000000F) ==  0x0000000F) {
nengel@2:         pixelsv2 = temp2;
nengel@2:     } else {
nengel@2:         pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(1, pixels));
nengel@2:     }
nengel@2:     pixelsv1 = vec_mergeh(vczero, pixelsv1);
nengel@2:     pixelsv2 = vec_mergeh(vczero, pixelsv2);
nengel@2:     pixelssum1 = vec_add((vector unsigned short)pixelsv1,
nengel@2:                          (vector unsigned short)pixelsv2);
nengel@2:     pixelssum1 = vec_add(pixelssum1, vcone);
nengel@2: 
nengel@2: POWERPC_PERF_START_COUNT(altivec_put_no_rnd_pixels8_xy2_num, 1);
nengel@2:     for (i = 0; i < h ; i++) {
nengel@2:         int rightside = ((unsigned long)block & 0x0000000F);
nengel@2:         blockv = vec_ld(0, block);
nengel@2: 
nengel@2:         temp1 = vec_ld(line_size, pixels);
nengel@2:         temp2 = vec_ld(line_size + 16, pixels);
nengel@2:         pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels));
nengel@2:         if (((((unsigned long)pixels) + line_size) & 0x0000000F) ==  0x0000000F) {
nengel@2:             pixelsv2 = temp2;
nengel@2:         } else {
nengel@2:             pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(line_size + 1, pixels));
nengel@2:         }
nengel@2: 
nengel@2:         pixelsv1 = vec_mergeh(vczero, pixelsv1);
nengel@2:         pixelsv2 = vec_mergeh(vczero, pixelsv2);
nengel@2:         pixelssum2 = vec_add((vector unsigned short)pixelsv1,
nengel@2:                              (vector unsigned short)pixelsv2);
nengel@2:         temp3 = vec_add(pixelssum1, pixelssum2);
nengel@2:         temp3 = vec_sra(temp3, vctwo);
nengel@2:         pixelssum1 = vec_add(pixelssum2, vcone);
nengel@2:         pixelsavg = vec_packsu(temp3, (vector unsigned short) vczero);
nengel@2: 
nengel@2:         if (rightside) {
nengel@2:             blockv = vec_perm(blockv, pixelsavg, vcprm(0, 1, s0, s1));
nengel@2:         } else {
nengel@2:             blockv = vec_perm(blockv, pixelsavg, vcprm(s0, s1, 2, 3));
nengel@2:         }
nengel@2: 
nengel@2:         vec_st(blockv, 0, block);
nengel@2: 
nengel@2:         block += line_size;
nengel@2:         pixels += line_size;
nengel@2:     }
nengel@2: 
nengel@2: POWERPC_PERF_STOP_COUNT(altivec_put_no_rnd_pixels8_xy2_num, 1);
nengel@2: }
nengel@2: 
nengel@2: /* next one assumes that ((line_size % 16) == 0) */
nengel@2: static void put_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pixels, int line_size, int h)
nengel@2: {
nengel@2: POWERPC_PERF_DECLARE(altivec_put_pixels16_xy2_num, 1);
nengel@2:     register int i;
nengel@2:     register vector unsigned char pixelsv1, pixelsv2, pixelsv3, pixelsv4;
nengel@2:     register vector unsigned char blockv, temp1, temp2;
nengel@2:     register vector unsigned short temp3, temp4,
nengel@2:         pixelssum1, pixelssum2, pixelssum3, pixelssum4;
nengel@2:     register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
nengel@2:     register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2);
nengel@2: 
nengel@2: POWERPC_PERF_START_COUNT(altivec_put_pixels16_xy2_num, 1);
nengel@2: 
nengel@2:     temp1 = vec_ld(0, pixels);
nengel@2:     temp2 = vec_ld(16, pixels);
nengel@2:     pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels));
nengel@2:     if ((((unsigned long)pixels) & 0x0000000F) ==  0x0000000F) {
nengel@2:         pixelsv2 = temp2;
nengel@2:     } else {
nengel@2:         pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(1, pixels));
nengel@2:     }
nengel@2:     pixelsv3 = vec_mergel(vczero, pixelsv1);
nengel@2:     pixelsv4 = vec_mergel(vczero, pixelsv2);
nengel@2:     pixelsv1 = vec_mergeh(vczero, pixelsv1);
nengel@2:     pixelsv2 = vec_mergeh(vczero, pixelsv2);
nengel@2:     pixelssum3 = vec_add((vector unsigned short)pixelsv3,
nengel@2:                          (vector unsigned short)pixelsv4);
nengel@2:     pixelssum3 = vec_add(pixelssum3, vctwo);
nengel@2:     pixelssum1 = vec_add((vector unsigned short)pixelsv1,
nengel@2:                          (vector unsigned short)pixelsv2);
nengel@2:     pixelssum1 = vec_add(pixelssum1, vctwo);
nengel@2: 
nengel@2:     for (i = 0; i < h ; i++) {
nengel@2:         blockv = vec_ld(0, block);
nengel@2: 
nengel@2:         temp1 = vec_ld(line_size, pixels);
nengel@2:         temp2 = vec_ld(line_size + 16, pixels);
nengel@2:         pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels));
nengel@2:         if (((((unsigned long)pixels) + line_size) & 0x0000000F) ==  0x0000000F) {
nengel@2:             pixelsv2 = temp2;
nengel@2:         } else {
nengel@2:             pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(line_size + 1, pixels));
nengel@2:         }
nengel@2: 
nengel@2:         pixelsv3 = vec_mergel(vczero, pixelsv1);
nengel@2:         pixelsv4 = vec_mergel(vczero, pixelsv2);
nengel@2:         pixelsv1 = vec_mergeh(vczero, pixelsv1);
nengel@2:         pixelsv2 = vec_mergeh(vczero, pixelsv2);
nengel@2: 
nengel@2:         pixelssum4 = vec_add((vector unsigned short)pixelsv3,
nengel@2:                              (vector unsigned short)pixelsv4);
nengel@2:         pixelssum2 = vec_add((vector unsigned short)pixelsv1,
nengel@2:                              (vector unsigned short)pixelsv2);
nengel@2:         temp4 = vec_add(pixelssum3, pixelssum4);
nengel@2:         temp4 = vec_sra(temp4, vctwo);
nengel@2:         temp3 = vec_add(pixelssum1, pixelssum2);
nengel@2:         temp3 = vec_sra(temp3, vctwo);
nengel@2: 
nengel@2:         pixelssum3 = vec_add(pixelssum4, vctwo);
nengel@2:         pixelssum1 = vec_add(pixelssum2, vctwo);
nengel@2: 
nengel@2:         blockv = vec_packsu(temp3, temp4);
nengel@2: 
nengel@2:         vec_st(blockv, 0, block);
nengel@2: 
nengel@2:         block += line_size;
nengel@2:         pixels += line_size;
nengel@2:     }
nengel@2: 
nengel@2: POWERPC_PERF_STOP_COUNT(altivec_put_pixels16_xy2_num, 1);
nengel@2: }
nengel@2: 
nengel@2: /* next one assumes that ((line_size % 16) == 0) */
nengel@2: static void put_no_rnd_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pixels, int line_size, int h)
nengel@2: {
nengel@2: POWERPC_PERF_DECLARE(altivec_put_no_rnd_pixels16_xy2_num, 1);
nengel@2:     register int i;
nengel@2:     register vector unsigned char pixelsv1, pixelsv2, pixelsv3, pixelsv4;
nengel@2:     register vector unsigned char blockv, temp1, temp2;
nengel@2:     register vector unsigned short temp3, temp4,
nengel@2:         pixelssum1, pixelssum2, pixelssum3, pixelssum4;
nengel@2:     register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
nengel@2:     register const vector unsigned short vcone = (const vector unsigned short)vec_splat_u16(1);
nengel@2:     register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2);
nengel@2: 
nengel@2: POWERPC_PERF_START_COUNT(altivec_put_no_rnd_pixels16_xy2_num, 1);
nengel@2: 
nengel@2:     temp1 = vec_ld(0, pixels);
nengel@2:     temp2 = vec_ld(16, pixels);
nengel@2:     pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels));
nengel@2:     if ((((unsigned long)pixels) & 0x0000000F) ==  0x0000000F) {
nengel@2:         pixelsv2 = temp2;
nengel@2:     } else {
nengel@2:         pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(1, pixels));
nengel@2:     }
nengel@2:     pixelsv3 = vec_mergel(vczero, pixelsv1);
nengel@2:     pixelsv4 = vec_mergel(vczero, pixelsv2);
nengel@2:     pixelsv1 = vec_mergeh(vczero, pixelsv1);
nengel@2:     pixelsv2 = vec_mergeh(vczero, pixelsv2);
nengel@2:     pixelssum3 = vec_add((vector unsigned short)pixelsv3,
nengel@2:                          (vector unsigned short)pixelsv4);
nengel@2:     pixelssum3 = vec_add(pixelssum3, vcone);
nengel@2:     pixelssum1 = vec_add((vector unsigned short)pixelsv1,
nengel@2:                          (vector unsigned short)pixelsv2);
nengel@2:     pixelssum1 = vec_add(pixelssum1, vcone);
nengel@2: 
nengel@2:     for (i = 0; i < h ; i++) {
nengel@2:         blockv = vec_ld(0, block);
nengel@2: 
nengel@2:         temp1 = vec_ld(line_size, pixels);
nengel@2:         temp2 = vec_ld(line_size + 16, pixels);
nengel@2:         pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels));
nengel@2:         if (((((unsigned long)pixels) + line_size) & 0x0000000F) ==  0x0000000F) {
nengel@2:             pixelsv2 = temp2;
nengel@2:         } else {
nengel@2:             pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(line_size + 1, pixels));
nengel@2:         }
nengel@2: 
nengel@2:         pixelsv3 = vec_mergel(vczero, pixelsv1);
nengel@2:         pixelsv4 = vec_mergel(vczero, pixelsv2);
nengel@2:         pixelsv1 = vec_mergeh(vczero, pixelsv1);
nengel@2:         pixelsv2 = vec_mergeh(vczero, pixelsv2);
nengel@2: 
nengel@2:         pixelssum4 = vec_add((vector unsigned short)pixelsv3,
nengel@2:                              (vector unsigned short)pixelsv4);
nengel@2:         pixelssum2 = vec_add((vector unsigned short)pixelsv1,
nengel@2:                              (vector unsigned short)pixelsv2);
nengel@2:         temp4 = vec_add(pixelssum3, pixelssum4);
nengel@2:         temp4 = vec_sra(temp4, vctwo);
nengel@2:         temp3 = vec_add(pixelssum1, pixelssum2);
nengel@2:         temp3 = vec_sra(temp3, vctwo);
nengel@2: 
nengel@2:         pixelssum3 = vec_add(pixelssum4, vcone);
nengel@2:         pixelssum1 = vec_add(pixelssum2, vcone);
nengel@2: 
nengel@2:         blockv = vec_packsu(temp3, temp4);
nengel@2: 
nengel@2:         vec_st(blockv, 0, block);
nengel@2: 
nengel@2:         block += line_size;
nengel@2:         pixels += line_size;
nengel@2:     }
nengel@2: 
nengel@2: POWERPC_PERF_STOP_COUNT(altivec_put_no_rnd_pixels16_xy2_num, 1);
nengel@2: }
nengel@2: 
nengel@2: /* next one assumes that ((line_size % 8) == 0) */
nengel@2: static void avg_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, int line_size, int h)
nengel@2: {
nengel@2: POWERPC_PERF_DECLARE(altivec_avg_pixels8_xy2_num, 1);
nengel@2:     register int i;
nengel@2:     register vector unsigned char pixelsv1, pixelsv2, pixelsavg;
nengel@2:     register vector unsigned char blockv, temp1, temp2, blocktemp;
nengel@2:     register vector unsigned short pixelssum1, pixelssum2, temp3;
nengel@2: 
nengel@2:     register const vector unsigned char vczero = (const vector unsigned char)
nengel@2:                                         vec_splat_u8(0);
nengel@2:     register const vector unsigned short vctwo = (const vector unsigned short)
nengel@2:                                         vec_splat_u16(2);
nengel@2: 
nengel@2:     temp1 = vec_ld(0, pixels);
nengel@2:     temp2 = vec_ld(16, pixels);
nengel@2:     pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels));
nengel@2:     if ((((unsigned long)pixels) & 0x0000000F) ==  0x0000000F) {
nengel@2:         pixelsv2 = temp2;
nengel@2:     } else {
nengel@2:         pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(1, pixels));
nengel@2:     }
nengel@2:     pixelsv1 = vec_mergeh(vczero, pixelsv1);
nengel@2:     pixelsv2 = vec_mergeh(vczero, pixelsv2);
nengel@2:     pixelssum1 = vec_add((vector unsigned short)pixelsv1,
nengel@2:                          (vector unsigned short)pixelsv2);
nengel@2:     pixelssum1 = vec_add(pixelssum1, vctwo);
nengel@2: 
nengel@2: POWERPC_PERF_START_COUNT(altivec_avg_pixels8_xy2_num, 1);
nengel@2:     for (i = 0; i < h ; i++) {
nengel@2:         int rightside = ((unsigned long)block & 0x0000000F);
nengel@2:         blockv = vec_ld(0, block);
nengel@2: 
nengel@2:         temp1 = vec_ld(line_size, pixels);
nengel@2:         temp2 = vec_ld(line_size + 16, pixels);
nengel@2:         pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels));
nengel@2:         if (((((unsigned long)pixels) + line_size) & 0x0000000F) ==  0x0000000F) {
nengel@2:             pixelsv2 = temp2;
nengel@2:         } else {
nengel@2:             pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(line_size + 1, pixels));
nengel@2:         }
nengel@2: 
nengel@2:         pixelsv1 = vec_mergeh(vczero, pixelsv1);
nengel@2:         pixelsv2 = vec_mergeh(vczero, pixelsv2);
nengel@2:         pixelssum2 = vec_add((vector unsigned short)pixelsv1,
nengel@2:                              (vector unsigned short)pixelsv2);
nengel@2:         temp3 = vec_add(pixelssum1, pixelssum2);
nengel@2:         temp3 = vec_sra(temp3, vctwo);
nengel@2:         pixelssum1 = vec_add(pixelssum2, vctwo);
nengel@2:         pixelsavg = vec_packsu(temp3, (vector unsigned short) vczero);
nengel@2: 
nengel@2:         if (rightside) {
nengel@2:             blocktemp = vec_perm(blockv, pixelsavg, vcprm(0, 1, s0, s1));
nengel@2:         } else {
nengel@2:             blocktemp = vec_perm(blockv, pixelsavg, vcprm(s0, s1, 2, 3));
nengel@2:         }
nengel@2: 
nengel@2:         blockv = vec_avg(blocktemp, blockv);
nengel@2:         vec_st(blockv, 0, block);
nengel@2: 
nengel@2:         block += line_size;
nengel@2:         pixels += line_size;
nengel@2:     }
nengel@2: 
nengel@2: POWERPC_PERF_STOP_COUNT(altivec_avg_pixels8_xy2_num, 1);
nengel@2: }
nengel@2: 
nengel@2: void dsputil_init_altivec(DSPContext* c)
nengel@2: {
nengel@2:     c->diff_pixels = diff_pixels_altivec;
nengel@2:     c->get_pixels = get_pixels_altivec;
nengel@2:     c->clear_block = clear_block_altivec;
nengel@2: 
nengel@2:     c->put_pixels_tab[0][0] = put_pixels16_altivec;
nengel@2:     /* the two functions do the same thing, so use the same code */
nengel@2:     c->put_no_rnd_pixels_tab[0][0] = put_pixels16_altivec;
nengel@2:     c->avg_pixels_tab[0][0] = avg_pixels16_altivec;
nengel@2:     c->avg_pixels_tab[1][0] = avg_pixels8_altivec;
nengel@2:     c->avg_pixels_tab[1][3] = avg_pixels8_xy2_altivec;
nengel@2:     c->put_pixels_tab[1][3] = put_pixels8_xy2_altivec;
nengel@2:     c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels8_xy2_altivec;
nengel@2:     c->put_pixels_tab[0][3] = put_pixels16_xy2_altivec;
nengel@2:     c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy2_altivec;
nengel@2: 
nengel@2: }