michael@0: /*
michael@0:  *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
michael@0:  *
michael@0:  *  Use of this source code is governed by a BSD-style license
michael@0:  *  that can be found in the LICENSE file in the root of the source
michael@0:  *  tree. An additional intellectual property rights grant can be found
michael@0:  *  in the file PATENTS.  All contributing project authors may
michael@0:  *  be found in the AUTHORS file in the root of the source tree.
michael@0:  */
michael@0: 
michael@0: 
michael@0: #include "vpx_config.h"
michael@0: #include "vp8_rtcd.h"
michael@0: #include "vpx_ports/x86.h"
michael@0: #include "vp8/encoder/block.h"
michael@0: 
michael@0: void vp8_short_fdct4x4_mmx(short *input, short *output, int pitch);
michael@0: void vp8_short_fdct8x4_mmx(short *input, short *output, int pitch)
michael@0: {
michael@0:     vp8_short_fdct4x4_mmx(input,   output,    pitch);
michael@0:     vp8_short_fdct4x4_mmx(input + 4, output + 16, pitch);
michael@0: }
michael@0: 
michael@0: int vp8_fast_quantize_b_impl_mmx(short *coeff_ptr, short *zbin_ptr,
michael@0:                                  short *qcoeff_ptr, short *dequant_ptr,
michael@0:                                  const short *scan_mask, short *round_ptr,
michael@0:                                  short *quant_ptr, short *dqcoeff_ptr);
michael@0: void vp8_fast_quantize_b_mmx(BLOCK *b, BLOCKD *d)
michael@0: {
michael@0:     const short *scan_mask   = vp8_default_zig_zag_mask;
michael@0:     short *coeff_ptr   = b->coeff;
michael@0:     short *zbin_ptr    = b->zbin;
michael@0:     short *round_ptr   = b->round;
michael@0:     short *quant_ptr   = b->quant_fast;
michael@0:     short *qcoeff_ptr  = d->qcoeff;
michael@0:     short *dqcoeff_ptr = d->dqcoeff;
michael@0:     short *dequant_ptr = d->dequant;
michael@0: 
michael@0:     *d->eob = (char)vp8_fast_quantize_b_impl_mmx(
michael@0:                                                  coeff_ptr,
michael@0:                                                  zbin_ptr,
michael@0:                                                  qcoeff_ptr,
michael@0:                                                  dequant_ptr,
michael@0:                                                  scan_mask,
michael@0: 
michael@0:                                                  round_ptr,
michael@0:                                                  quant_ptr,
michael@0:                                                  dqcoeff_ptr
michael@0:                                                  );
michael@0: }
michael@0: 
michael@0: int vp8_mbblock_error_mmx_impl(short *coeff_ptr, short *dcoef_ptr, int dc);
michael@0: int vp8_mbblock_error_mmx(MACROBLOCK *mb, int dc)
michael@0: {
michael@0:     short *coeff_ptr =  mb->block[0].coeff;
michael@0:     short *dcoef_ptr =  mb->e_mbd.block[0].dqcoeff;
michael@0:     return vp8_mbblock_error_mmx_impl(coeff_ptr, dcoef_ptr, dc);
michael@0: }
michael@0: 
michael@0: int vp8_mbuverror_mmx_impl(short *s_ptr, short *d_ptr);
michael@0: int vp8_mbuverror_mmx(MACROBLOCK *mb)
michael@0: {
michael@0:     short *s_ptr = &mb->coeff[256];
michael@0:     short *d_ptr = &mb->e_mbd.dqcoeff[256];
michael@0:     return vp8_mbuverror_mmx_impl(s_ptr, d_ptr);
michael@0: }
michael@0: 
michael@0: void vp8_subtract_b_mmx_impl(unsigned char *z,  int src_stride,
michael@0:                              short *diff, unsigned char *predictor,
michael@0:                              int pitch);
michael@0: void vp8_subtract_b_mmx(BLOCK *be, BLOCKD *bd, int pitch)
michael@0: {
michael@0:     unsigned char *z = *(be->base_src) + be->src;
michael@0:     unsigned int  src_stride = be->src_stride;
michael@0:     short *diff = &be->src_diff[0];
michael@0:     unsigned char *predictor = &bd->predictor[0];
michael@0:     vp8_subtract_b_mmx_impl(z, src_stride, diff, predictor, pitch);
michael@0: }