sl@0: /*
sl@0: * LIBOIL - Library of Optimized Inner Loops
sl@0: * Copyright (c) 2003,2004 David A. Schleef <ds@schleef.org>
sl@0: * All rights reserved.
sl@0: *
sl@0: * Redistribution and use in source and binary forms, with or without
sl@0: * modification, are permitted provided that the following conditions
sl@0: * are met:
sl@0: * 1. Redistributions of source code must retain the above copyright
sl@0: * notice, this list of conditions and the following disclaimer.
sl@0: * 2. Redistributions in binary form must reproduce the above copyright
sl@0: * notice, this list of conditions and the following disclaimer in the
sl@0: * documentation and/or other materials provided with the distribution.
sl@0: *
sl@0: * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
sl@0: * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
sl@0: * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
sl@0: * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT,
sl@0: * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
sl@0: * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
sl@0: * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
sl@0: * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
sl@0: * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
sl@0: * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
sl@0: * POSSIBILITY OF SUCH DAMAGE.
sl@0: */
sl@0: //Portions Copyright (c) 2008-2009 Nokia Corporation and/or its subsidiary(-ies). All rights reserved.
sl@0:
sl@0: #ifdef HAVE_CONFIG_H
sl@0: #include "config.h"
sl@0: #endif
sl@0:
sl@0: #include <liboil/liboilfunction.h>
sl@0: #include "liboil/simdpack/simdpack.h"
sl@0:
sl@0: OIL_DECLARE_CLASS (recon8x8_intra);
sl@0: OIL_DECLARE_CLASS (recon8x8_inter);
sl@0: OIL_DECLARE_CLASS (recon8x8_inter2);
sl@0:
sl@0: const uint8_t c0x80[8] = {
sl@0: 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80
sl@0: };
sl@0:
sl@0: static void
sl@0: recon8x8_intra_i386_mmx (uint8_t *dest, int ds, int16_t *change)
sl@0: {
sl@0: #if !defined(__WINSCW__) && !defined(__WINS__)
sl@0: __asm__ __volatile__ (
sl@0: " movq (%3), %%mm0 \n\t" /* Set mm0 to 0x8080808080808080 */
sl@0:
sl@0: " lea 128(%1), %%edi \n\t" /* Endpoint in input buffer */
sl@0: "1: \n\t"
sl@0: " movq (%1), %%mm2 \n\t" /* First four input values */
sl@0:
sl@0: " packsswb 8(%1), %%mm2 \n\t" /* pack with next(high) four values */
sl@0: " por %%mm0, %%mm0 \n\t"
sl@0: " pxor %%mm0, %%mm2 \n\t" /* Convert result to unsigned (same as add 128) */
sl@0: " lea 16(%1), %1 \n\t" /* Step source buffer */
sl@0: " cmp %%edi, %1 \n\t" /* are we done */
sl@0:
sl@0: " movq %%mm2, (%0) \n\t" /* store results */
sl@0:
sl@0: " lea (%0, %2), %0 \n\t" /* Step output buffer */
sl@0: " jc 1b \n\t" /* Loop back if we are not done */
sl@0: " emms \n\t"
sl@0: : "+r" (dest)
sl@0: : "r" (change),
sl@0: "r" (ds),
sl@0: "r" (c0x80)
sl@0: : "memory", "edi"
sl@0: );
sl@0: #endif
sl@0: }
sl@0:
sl@0: OIL_DEFINE_IMPL_FULL (recon8x8_intra_i386_mmx, recon8x8_intra, OIL_IMPL_FLAG_MMX);
sl@0:
sl@0: #if 0
sl@0: static void
sl@0: recon8x8_inter_i386_mmx (uint8_t *dest, int ds, uint8_t *src, int ss, int16_t *change)
sl@0: {
sl@0: /* FIXME doesn't handle ss */
sl@0: __asm__ __volatile__ (
sl@0: " pxor %%mm0, %%mm0 \n\t"
sl@0: " lea 128(%1), %%edi \n\t"
sl@0:
sl@0: "1: \n\t"
sl@0: " movq (%2), %%mm2 \n\t" /* (+3 misaligned) 8 reference pixels */
sl@0:
sl@0: " movq (%1), %%mm4 \n\t" /* first 4 changes */
sl@0: " movq %%mm2, %%mm3 \n\t"
sl@0: " movq 8(%1), %%mm5 \n\t" /* last 4 changes */
sl@0: " punpcklbw %%mm0, %%mm2 \n\t" /* turn first 4 refs into positive 16-bit #s */
sl@0: " paddsw %%mm4, %%mm2 \n\t" /* add in first 4 changes */
sl@0: " punpckhbw %%mm0, %%mm3 \n\t" /* turn last 4 refs into positive 16-bit #s */
sl@0: " paddsw %%mm5, %%mm3 \n\t" /* add in last 4 changes */
sl@0: " add %3, %2 \n\t" /* next row of reference pixels */
sl@0: " packuswb %%mm3, %%mm2 \n\t" /* pack result to unsigned 8-bit values */
sl@0: " lea 16(%1), %1 \n\t" /* next row of changes */
sl@0: " cmp %%edi, %1 \n\t" /* are we done? */
sl@0:
sl@0: " movq %%mm2, (%0) \n\t" /* store result */
sl@0:
sl@0: " lea (%0, %3), %0 \n\t" /* next row of output */
sl@0: " jc 1b \n\t"
sl@0: " emms \n\t"
sl@0: : "+r" (dest)
sl@0: : "r" (change),
sl@0: "r" (src),
sl@0: "r" (ds)
sl@0: : "memory", "edi"
sl@0: );
sl@0: }
sl@0:
sl@0: OIL_DEFINE_IMPL_FULL (recon8x8_inter_i386_mmx, recon8x8_inter, OIL_IMPL_FLAG_MMX);
sl@0:
sl@0: static void
sl@0: recon8x8_inter2_i386_mmx (uint8_t *dest, int ds, uint8_t *s1, int ss1, uint8_t *s2, int ss2, int16_t *change)
sl@0: {
sl@0: /* FIXME doesn't handle ss1, ss2 */
sl@0: __asm__ __volatile__ (
sl@0: " pxor %%mm0, %%mm0 \n\t"
sl@0: " lea 128(%1), %%edi \n\t"
sl@0:
sl@0: "1: \n\t"
sl@0: " movq (%2), %%mm2 \n\t" /* (+3 misaligned) 8 reference pixels */
sl@0: " movq (%3), %%mm4 \n\t" /* (+3 misaligned) 8 reference pixels */
sl@0:
sl@0: " movq %%mm2, %%mm3 \n\t"
sl@0: " punpcklbw %%mm0, %%mm2 \n\t" /* mm2 = start ref1 as positive 16-bit #s */
sl@0: " movq %%mm4, %%mm5 \n\t"
sl@0: " movq (%1), %%mm6 \n\t" /* first 4 changes */
sl@0: " punpckhbw %%mm0, %%mm3 \n\t" /* mm3 = end ref1 as positive 16-bit #s */
sl@0: " movq 8(%1), %%mm7 \n\t" /* last 4 changes */
sl@0: " punpcklbw %%mm0, %%mm4 \n\t" /* mm4 = start ref2 as positive 16-bit #s */
sl@0: " punpckhbw %%mm0, %%mm5 \n\t" /* mm5 = end ref2 as positive 16-bit #s */
sl@0: " paddw %%mm4, %%mm2 \n\t" /* mm2 = start (ref1 + ref2) */
sl@0: " paddw %%mm5, %%mm3 \n\t" /* mm3 = end (ref1 + ref2) */
sl@0: " psrlw $1, %%mm2 \n\t" /* mm2 = start (ref1 + ref2)/2 */
sl@0: " psrlw $1, %%mm3 \n\t" /* mm3 = end (ref1 + ref2)/2 */
sl@0: " paddw %%mm6, %%mm2 \n\t" /* add changes to start */
sl@0: " paddw %%mm7, %%mm3 \n\t" /* add changes to end */
sl@0: " lea 16(%1), %1 \n\t" /* next row of changes */
sl@0: " packuswb %%mm3, %%mm2 \n\t" /* pack start|end to unsigned 8-bit */
sl@0: " add %4, %2 \n\t" /* next row of reference pixels */
sl@0: " add %4, %3 \n\t" /* next row of reference pixels */
sl@0: " movq %%mm2, (%0) \n\t" /* store result */
sl@0: " add %4, %0 \n\t" /* next row of output */
sl@0: " cmp %%edi, %1 \n\t" /* are we done? */
sl@0: " jc 1b \n\t"
sl@0: " emms \n\t"
sl@0: : "+r" (dest)
sl@0: : "r" (change),
sl@0: "r" (s1),
sl@0: "r" (s2),
sl@0: "m" (ds)
sl@0: : "memory", "edi"
sl@0: );
sl@0: }
sl@0:
sl@0: OIL_DEFINE_IMPL_FULL (recon8x8_inter2_i386_mmx, recon8x8_inter2, OIL_IMPL_FLAG_MMX);
sl@0: #endif
sl@0:
sl@0:
sl@0: #ifdef __SYMBIAN32__
sl@0:
sl@0: OilFunctionImpl* __oil_function_impl_recon8x8_intra_i386_mmx, recon8x8_intra() {
sl@0: return &_oil_function_impl_recon8x8_intra_i386_mmx, recon8x8_intra;
sl@0: }
sl@0: #endif
sl@0:
sl@0: #ifdef __SYMBIAN32__
sl@0:
sl@0: OilFunctionImpl* __oil_function_impl_recon8x8_inter_i386_mmx, recon8x8_inter() {
sl@0: return &_oil_function_impl_recon8x8_inter_i386_mmx, recon8x8_inter;
sl@0: }
sl@0: #endif
sl@0:
sl@0: #ifdef __SYMBIAN32__
sl@0:
sl@0: OilFunctionImpl* __oil_function_impl_recon8x8_inter2_i386_mmx, recon8x8_inter2() {
sl@0: return &_oil_function_impl_recon8x8_inter2_i386_mmx, recon8x8_inter2;
sl@0: }
sl@0: #endif
sl@0: