/******************************************************** * Some code. Copyright (C) 2003 by Pascal Massimino.   * * All Rights Reserved.      (http://skal.planet-d.net) * * For Educational/Academic use ONLY. See 'LICENSE.TXT'.* ********************************************************//* * skl_gmc_dsp.cpp * *   level Sprite (MPEG4's GMC) processing ********************************************************/#include "skl.h"#include "skl_syst/skl_dsp.h"#include "skl_syst/skl_bits.h"extern "C" {//////////////////////////////////////////////////////////// Raw-C version#define RDIV(a,b) (((a)>0 ? (a) + ((b)>>1) : (a) - ((b)>>1))/(b))#define RSHIFT(a,b) ( (a)>0 ? ((a) + (1<<((b)-1)))>>(b) : ((a) + (1<<((b)-1))-1)>>(b))#define MLT(i)  (((16-(i))<<16) + (i))static const SKL_UINT32 MTab[16] = {    MLT( 0), MLT( 1), MLT( 2), MLT( 3), MLT( 4), MLT( 5), MLT( 6), MLT( 7),  MLT( 8), MLT( 9), MLT(10), MLT(11), MLT(12), MLT(13), MLT(14), MLT(15)};#undef MLT//////////////////////////////////////////////////////////// Pts = 2 or 3staticvoid Predict_16x16_C(const SKL_GMC_DSP * const This,                     SKL_BYTE *Dst, const SKL_BYTE *Src,                      int BpS, int x, int y, int Rounding){  const int W   = This->Width;  const int H   = This->Height;  const int Rho = 3-This->Accuracy;  const SKL_INT32 Rounder = ( 128 - (Rounding<<(2*Rho)) ) << 16;  const SKL_UINT32 W2 = W<<(16-Rho);  const SKL_UINT32 H2 = H<<(16-Rho);  Dst += 16;  const SKL_INT32 dUx = This->dU[0];  const SKL_INT32 dVx = This->dV[0];  const SKL_INT32 dUy = This->dU[1];  const SKL_INT32 dVy = This->dV[1];  SKL_INT32 Uo = This->Uo + 16*(dUy*y + dUx*x);  SKL_INT32 Vo = This->Vo + 16*(dVy*y + dVx*x);  for(int j=16; j>0; --j)  {#ifdef USE_ASM//    __asm__ __volatile__ ("prefetcht0 (%0)\n\t" : : "r" (Dst-16));#endif    SKL_INT32 U = Uo, V = Vo;    Uo += dUy; Vo += dVy;    if ( (SKL_UINT32)U<W2 && (SKL_UINT32)(U+15*dUx)<W2 &&         (SKL_UINT32)V<H2 && (SKL_UINT32)(V+15*dVx)<H2 )    {          for(int i=-16; i<0; ++i)      {        SKL_INT32 u = ( U >> 16 ) << Rho;        SKL_INT32 v = ( V >> 16 ) << Rho;        U += dUx;  V += dVx;        int Offset = (u>>4) + (v>>4)*BpS;        const SKL_UINT32 ri = MTab[u&15];        const SKL_UINT32 rj = MTab[v&15];        SKL_UINT32 f0, f1;        f0  = Src[ Offset     +0 ];        f0 |= Src[ Offset     +1 ] << 16;        f1  = Src[ Offset+BpS +0 ];        f1 |= Src[ Offset+BpS +1 ] << 16;        f0 = (ri*f0)>>16;        f1 = (ri*f1) & 0x0fff0000;        f0 |= f1;         f0 = ( rj*f0 + Rounder ) >> 24;        Dst[i] = (SKL_BYTE)f0;      }    }    else {      for(int i=-16; i<0; ++i)      {        int Offset;        SKL_INT32 u = ( U >> 16 ) << Rho;        SKL_INT32 v = ( V >> 16 ) << Rho;        U += dUx; V += dVx;        if ((SKL_UINT32)u<(SKL_UINT32)W) Offset = u>>4;        else {          if (u>=W) Offset = W>>4;          else Offset = 0;          u = 0;        }        if ((SKL_UINT32)v<(SKL_UINT32)H) Offset += (v>>4)*BpS;        else {          if (v>=H) Offset += (H>>4)*BpS;          v = 0;        }        if (u&15) {          const SKL_UINT32 ri = MTab[u&15];          if (v&15) {            const SKL_UINT32 rj = MTab[v&15];            SKL_UINT32 f0, f1;            f0  = Src[ Offset     +0 ];            f0 |= Src[ Offset     +1 ] << 16;            f1  = Src[ Offset+BpS +0 ];            f1 |= Src[ Offset+BpS +1 ] << 16;            f0 = (ri*f0)>>16;            f1 = (ri*f1) & 0x0fff0000;            f0 |= f1;             f0 = ( rj*f0 + Rounder ) >> 24;            Dst[i] = (SKL_BYTE)f0;          }          else {            SKL_UINT32 f0;            f0  = Src[ Offset   ];            f0 |= Src[ Offset+1 ] << 16;            f0 = (ri*f0 + (Rounder>>4))>>20;            Dst[i] = (SKL_BYTE)f0;          }        }        else {          if (v&15) {            const SKL_UINT32 rj = MTab[v&15];            SKL_UINT32 f0;            f0  = Src[ Offset      ];            f0 |= Src[ Offset +BpS ] << 16;            f0 = (rj*f0 + (Rounder>>4))>>20;            Dst[i] = (SKL_BYTE)f0;          }          else {            Dst[i] = Src[Offset];          }        }      }    }    Dst += BpS;  }}staticvoid Predict_8x8_C(const SKL_GMC_DSP * const This,                   SKL_BYTE *uDst, const SKL_BYTE *uSrc,                    SKL_SAFE_INT uv_Coloc,                   int BpS, int x, int y, int Rounding){  const int W   = This->Width >> 1;  const int H   = This->Height>> 1;  const int Rho = 3-This->Accuracy;  const SKL_INT32 Rounder = ( 128 - (Rounding<<(2*Rho)) ) << 16;  const SKL_UINT32 W2 = W<<(16-Rho);     const SKL_UINT32 H2 = H<<(16-Rho);  uDst += 8;  SKL_BYTE       *vDst = uDst + uv_Coloc;  const SKL_BYTE *vSrc = uSrc + uv_Coloc;  const SKL_INT32 dUx = This->dU[0];  const SKL_INT32 dVx = This->dV[0];  const SKL_INT32 dUy = This->dU[1];  const SKL_INT32 dVy = This->dV[1];  SKL_INT32 Uo = This->Uco + 8*(dUy*y + dUx*x);  SKL_INT32 Vo = This->Vco + 8*(dVy*y + dVx*x);  for(int j=8; j>0; --j)  {#ifdef USE_ASM//    __asm__ __volatile__ ("prefetcht0 (%0)\n\t" : : "r" (uDst-8));//    __asm__ __volatile__ ("prefetcht0 (%0)\n\t" : : "r" (vDst-8));#endif    SKL_INT32 U = Uo, V = Vo;    Uo += dUy; Vo += dVy;    if ( (SKL_UINT32)U<W2 && (SKL_UINT32)(U+15*dUx)<W2 &&         (SKL_UINT32)V<H2 && (SKL_UINT32)(V+15*dVx)<H2 )    {      for(int i=-8; i<0; ++i)       {        SKL_INT32 u = ( U >> 16 ) << Rho;        SKL_INT32 v = ( V >> 16 ) << Rho;        U += dUx; V += dVx;        const int Offset = (u>>4) + (v>>4)*BpS;        const SKL_UINT32 ri = MTab[u&15];        const SKL_UINT32 rj = MTab[v&15];        SKL_UINT32 f0, f1;        f0  = uSrc[ Offset     +0 ];        f0 |= uSrc[ Offset     +1 ] << 16;        f1  = uSrc[ Offset+BpS +0 ];        f1 |= uSrc[ Offset+BpS +1 ] << 16;        f0 = (ri*f0)>>16;        f1 = (ri*f1) & 0x0fff0000;        f0 |= f1;         f0 = ( rj*f0 + Rounder ) >> 24;        uDst[i] = (SKL_BYTE)f0;        f0  = vSrc[ Offset     +0 ];        f0 |= vSrc[ Offset     +1 ] << 16;        f1  = vSrc[ Offset+BpS +0 ];        f1 |= vSrc[ Offset+BpS +1 ] << 16;        f0 = (ri*f0)>>16;          f1 = (ri*f1) & 0x0fff0000;        f0 |= f1;         f0 = ( rj*f0 + Rounder ) >> 24;        vDst[i] = (SKL_BYTE)f0;      }    }    else {      for(int i=-8; i<0; ++i)       {        SKL_INT32 u = ( U >> 16 ) << Rho;        SKL_INT32 v = ( V >> 16 ) << Rho;        U += dUx; V += dVx;        int Offset;        if ((SKL_UINT32)u<(SKL_UINT32)W) Offset = u>>4;        else {          if (u>=W) Offset = W>>4;          else      Offset = 0;          u = 0;        }        if ((SKL_UINT32)v<(SKL_UINT32)H) Offset += (v>>4)*BpS;        else {          if (v>=H) Offset += (H>>4)*BpS;          v = 0;        }        if (u&15) {          const SKL_UINT32 ri = MTab[u&15];          if (v&15) {            SKL_UINT32 f0, f1;            const SKL_UINT32 rj = MTab[v&15];            f0  = uSrc[ Offset     +0 ];            f0 |= uSrc[ Offset     +1 ] << 16;            f1  = uSrc[ Offset+BpS +0 ];            f1 |= uSrc[ Offset+BpS +1 ] << 16;            f0 = (ri*f0)>>16;            f1 = (ri*f1) & 0x0fff0000;            f0 |= f1;             f0 = ( rj*f0 + Rounder ) >> 24;            uDst[i] = (SKL_BYTE)f0;            f0  = vSrc[ Offset     +0 ];            f0 |= vSrc[ Offset     +1 ] << 16;            f1  = vSrc[ Offset+BpS +0 ];            f1 |= vSrc[ Offset+BpS +1 ] << 16;            f0 = (ri*f0)>>16;              f1 = (ri*f1) & 0x0fff0000;            f0 |= f1;             f0 = ( rj*f0 + Rounder ) >> 24;            vDst[i] = (SKL_BYTE)f0;          }          else {            SKL_UINT32 f0;            f0  = uSrc[ Offset   ];            f0 |= uSrc[ Offset+1 ] << 16;            f0 = (ri*f0 + (Rounder>>4))>>20;            uDst[i] = (SKL_BYTE)f0;            f0  = vSrc[ Offset   ];            f0 |= vSrc[ Offset+1 ] << 16;            f0 = (ri*f0 + (Rounder>>4))>>20;            vDst[i] = (SKL_BYTE)f0;          }        }        else {          if (v&15) {            const SKL_UINT32 rj = MTab[v&15];            SKL_UINT32 f0;            f0  = uSrc[ Offset      ];            f0 |= uSrc[ Offset +BpS ] << 16;            f0 = (rj*f0 + (Rounder>>4))>>20;            uDst[i] = (SKL_BYTE)f0;            f0  = vSrc[ Offset      ];            f0 |= vSrc[ Offset +BpS ] << 16;            f0 = (rj*f0 + (Rounder>>4))>>20;            vDst[i] = (SKL_BYTE)f0;          }          else {            uDst[i] = uSrc[Offset];            vDst[i] = vSrc[Offset];          }        }