///////////////////////////////////////////////////////////////////////////////
//
//      ##  ######
//       ######  ###
//  ## ###############        Shark 3D Engine (www.shark3d.com)
//   ########## # # #
//    ########                Copyright (c) 1996-2006 Spinor GmbH.
//   ######### # # #          All rights reserved.
//  ##   ##########
//      ##
//
///////////////////////////////////////////////////////////////////////////////

//@cpp
#ifndef S3D_DRV_D3D9_MMX_H
#define S3D_DRV_D3D9_MMX_H

#include "general/general_env.h"
#include "sys/core/sys_types.h"
#include "sys/core/sys_assert.h"
#include "sys/core/sys_lang.h"

#include <d3dx9.h>
#include <emmintrin.h>
///////////////////////////////////////////////////////////////////////////////

#ifdef S3D_DRV_D3D9_MMX_USEPROLOGUE

#define S3D_DRV_D3D9_MMX_PROLOGUE \
    push esi \
    push edi \
    push ecx 

#define S3D_DRV_D3D9_MMX_EPILOGUE \
    pop ecx \
    pop edi \
    pop esi

#else

#define S3D_DRV_D3D9_MMX_PROLOGUE
#define S3D_DRV_D3D9_MMX_EPILOGUE

#endif //S3D_DRV_D3D9_MMX_USEPROLOGUE

///////////////////////////////////////////////////////////////////////////////

//@
void s3d_DrvD3d9MemcpyMmx8(void *Dest, const void *Src, s3d_CSysIntps Len);

//@
void s3d_DrvD3d9MemcpyMmx16(void *Dest, const void *Src, s3d_CSysIntps Len);

//@
void s3d_DrvD3d9MemcpyMmx32(void *Dest, const void *Src, s3d_CSysIntps Len);

//@
void s3d_DrvD3d9MemcpyMmx64(void *Dest, const void *Src, s3d_CSysIntps Len);

//@
void s3d_DrvD3d9MemfillMmx64(void *Dest, int Val, s3d_CSysIntps Len);

///////////////////////////////////////////////////////////////////////////////

S3D_SYS_INLINE D3DX_ALIGN16
void s3d_DrvD3d9MemcpyMmx8(void *Dest, const void *Src, s3d_CSysIntps Len)
{
    S3D_SYS_ASSERT((Len % 8) == 0);
    __asm 
    {
        S3D_DRV_D3D9_MMX_PROLOGUE

        mov esi, Src 
        mov edi, Dest 
        mov ecx, Len 
        shr ecx, 3          // 8 bytes per iteration 

    loop1: 
        //prefetchnta 8[ESI] // Prefetch next loop, non-temporal 

        movq mm1,  0[ESI]   // Read in source data 

        movntq  0[EDI], mm1 // Non-temporal stores 

        add esi, 8 
        add edi, 8 
        dec ecx 
        jnz loop1 

        emms

        S3D_DRV_D3D9_MMX_EPILOGUE
    }
} 

S3D_SYS_INLINE D3DX_ALIGN16
void s3d_DrvD3d9MemcpyMmx16(void *Dest, const void *Src, s3d_CSysIntps Len)
{
    S3D_SYS_ASSERT((Len % 16) == 0);
    __asm 
    {
        S3D_DRV_D3D9_MMX_PROLOGUE

        mov esi, Src 
        mov edi, Dest 
        mov ecx, Len 
        shr ecx, 4          // 16 bytes per iteration 

    loop1: 
        //prefetchnta 16[ESI] // Prefetch next loop, non-temporal 

        movq mm1,  0[ESI]   // Read in source data 
        movq mm2,  8[ESI] 

        movntq  0[EDI], mm1 // Non-temporal stores 
        movntq  8[EDI], mm2 

        add esi, 16 
        add edi, 16 
        dec ecx 
        jnz loop1 

        emms

        S3D_DRV_D3D9_MMX_EPILOGUE
    }
} 

S3D_SYS_INLINE D3DX_ALIGN16
void s3d_DrvD3d9MemcpyMmx32(void *Dest, const void *Src, s3d_CSysIntps Len)
{
    S3D_SYS_ASSERT((Len % 32) == 0);
    __asm 
    {
        S3D_DRV_D3D9_MMX_PROLOGUE

        mov esi, Src 
        mov edi, Dest 
        mov ecx, Len 
        shr ecx, 5          // 32 bytes per iteration 

    loop1: 
        //prefetchnta 32[ESI] // Prefetch next loop, non-temporal 

        movq mm1,  0[ESI]   // Read in source data 
        movq mm2,  8[ESI] 
        movq mm3, 16[ESI] 
        movq mm4, 24[ESI] 

        movntq  0[EDI], mm1 // Non-temporal stores 
        movntq  8[EDI], mm2 
        movntq 16[EDI], mm3
        movntq 24[EDI], mm4 

        add esi, 32 
        add edi, 32 
        dec ecx 
        jnz loop1 

        emms

        S3D_DRV_D3D9_MMX_EPILOGUE
    }
} 

S3D_SYS_INLINE D3DX_ALIGN16
void s3d_DrvD3d9MemcpyMmx64(void *Dest, const void *Src, s3d_CSysIntps Len)
{
    S3D_SYS_ASSERT((Len % 64) == 0);
    __asm 
    {
        S3D_DRV_D3D9_MMX_PROLOGUE

        mov esi, Src 
        mov edi, Dest 
        mov ecx, Len 
        shr ecx, 6          // 64 bytes per iteration 

    loop1: 
        //prefetchnta 64[ESI] // Prefetch next loop, non-temporal 

        movq mm1,  0[ESI]   // Read in source data 
        movq mm2,  8[ESI] 
        movq mm3, 16[ESI] 
        movq mm4, 24[ESI] 
        movq mm5, 32[ESI] 
        movq mm6, 40[ESI] 
        movq mm7, 48[ESI] 
        movq mm0, 56[ESI] 

        movntq  0[EDI], mm1 // Non-temporal stores 
        movntq  8[EDI], mm2 
        movntq 16[EDI], mm3
        movntq 24[EDI], mm4 
        movntq 32[EDI], mm5
        movntq 40[EDI], mm6 
        movntq 48[EDI], mm7
        movntq 56[EDI], mm0 

        add esi, 64 
        add edi, 64 
        dec ecx 
        jnz loop1 

        emms

        S3D_DRV_D3D9_MMX_EPILOGUE
    }
} 

///////////////////////////////////////////////////////////////////////////////

S3D_SYS_INLINE D3DX_ALIGN16
void s3d_DrvD3d9MemfillMmx64(void *Dest, int Val, s3d_CSysIntps Len)
{
    S3D_SYS_ASSERT(((UINT_PTR)Dest & 0xF) == 0);
    S3D_SYS_ASSERT((Len % 64) == 0);

    __m64 Val64;
    Val64.m64_i32[0] = Val;
    Val64.m64_i32[1] = Val;
    __asm 
    {
        S3D_DRV_D3D9_MMX_PROLOGUE

        mov edi, Dest 
        mov ecx, Len 
        shr ecx, 6          // 64 bytes per iteration 

        movq mm1, Val64   // Read in source data 
    loop1: 
        //prefetchnta 64[ESI] // Prefetch next loop, non-temporal 


        movntq  0[EDI], mm1 // Non-temporal stores 
        movntq  8[EDI], mm1 
        movntq 16[EDI], mm1
        movntq 24[EDI], mm1 
        movntq 32[EDI], mm1
        movntq 40[EDI], mm1 
        movntq 48[EDI], mm1
        movntq 56[EDI], mm1 

        add edi, 64 
        dec ecx 
        jnz loop1 

        emms

        S3D_DRV_D3D9_MMX_EPILOGUE
    }
}

///////////////////////////////////////////////////////////////////////////////

#endif