/* Copyright (c) 2008 by FlightSafety International, Inc. - Unpublished.
** All rights reserved under the copyright laws of the United States.
** @(#) $Id: DXTcompress.c,v 1.2 2008/03/25 18:25:24 carter Exp $
*/

// gcc -O3 -c -masm=intel DXTcompress.c

#include <stdlib.h>
#include <string.h>

typedef unsigned char byte;

#define ALIGN16(a)	a __attribute__ ((aligned (16)))

/*
 SIMD Optimized Calculation of Line Through Color Space
 Copyright (C) 2006 Id Software, Inc.
 Written by J.M.P. van Waveren

 This code is free software; you can redistribute it and/or
 modify it under the terms of the GNU Lesser General Public
 License as published by the Free Software Foundation; either
 version 2.1 of the License, or (at your option) any later version.

 This code is distributed in the hope that it will be useful,
 but WITHOUT ANY WARRANTY; without even the implied warranty of
 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
 Lesser General Public License for more details.
*/

ALIGN16(const byte SIMD_SSE2_byte_0[16]) =
    { 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
      0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00 };

static void
GetMinMaxColors_SSE2( const byte *colorBlock, byte *minColor, byte *maxColor)
{
        asm (
	"mov eax, %2 \n\t"
	"mov esi, %0 \n\t"
	"mov edi, %1 \n\t"

	// get bounding box
	"movdqa xmm0, qword ptr [eax+ 0] \n\t"
	"movdqa xmm1, qword ptr [eax+ 0] \n\t"
	"pminub xmm0, qword ptr [eax+16] \n\t"
	"pmaxub xmm1, qword ptr [eax+16] \n\t"
	"pminub xmm0, qword ptr [eax+32] \n\t"
	"pmaxub xmm1, qword ptr [eax+32] \n\t"
	"pminub xmm0, qword ptr [eax+48] \n\t"
	"pmaxub xmm1, qword ptr [eax+48] \n\t"
	"pshufd xmm3, xmm0, 2 | (3<<2) | (2<<4) | (3<<6) \n\t"
	"pshufd xmm4, xmm1, 2 | (3<<2) | (2<<4) | (3<<6) \n\t"
	"pminub xmm0, xmm3 \n\t"
	"pmaxub xmm1, xmm4 \n\t"
	"pshuflw xmm6, xmm0, 2 | (3<<2) | (2<<4) | (3<<6) \n\t"
	"pshuflw xmm7, xmm1, 2 | (3<<2) | (2<<4) | (3<<6) \n\t"
	"pminub xmm0, xmm6 \n\t"
	"pmaxub xmm1, xmm7 \n\t"

	// inset the bounding box
	"punpcklbw xmm0, _SIMD_SSE2_byte_0 \n\t"
	"punpcklbw xmm1, _SIMD_SSE2_byte_0 \n\t"
	"movdqa xmm2, xmm1 \n\t"
	"psubw xmm2, xmm0 \n\t"
	"psrlw xmm2, 4 \n\t"
	"paddw xmm0, xmm2 \n\t"
	"psubw xmm1, xmm2 \n\t"
	"packuswb xmm0, xmm0 \n\t"
	"packuswb xmm1, xmm1 \n\t"

	// store bounding box extents
	"movd dword ptr [esi], xmm0 \n\t"
	"movd dword ptr [edi], xmm1 \n\t"
	: "=m" (minColor), "=m" (maxColor)
	: "m" (colorBlock)
	: "eax", "esi", "edi",
	  "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm6", "xmm7" 
    );
}


// default to SSE2 (the fastest)

static typeof(GetMinMaxColors_SSE2)  *GetMinMaxColors_  = GetMinMaxColors_SSE2;

int main(int argc, char *argv[])
{
    byte block[64];
    byte minColor[4];
    byte maxColor[4];

    GetMinMaxColors_(block, minColor, maxColor);
    return 0;
}