trydct.cpp

这是一组DCT和iDCT的代码

// TryDCT.cpp : Defines the entry point for the console application.
//

#include "stdafx.h"
#include <stdio.h>
#include <stdlib.h>
#include <time.h>
#include <string.h>
#include "Common.h"

#define	LOOP	(1048576)

inline void query_performance_count(__int64& c)
{
	__asm
	{
		mov ecx, c
		rdtsc
		mov [ecx], eax
		mov [ecx + 4], edx
	}
}

int main(int argc, char* argv[])
{
#if 1
	__declspec(align(16)) unsigned char buf[128], buf2[128];
	__declspec(align(16)) short tmpbuf[64], preserve_dct[64];
//	__declspec(align(16)) double co[64];
#else
	__declspec(align(16)) unsigned char dummy;
	unsigned char buf[128], buf2[128];
	short tmpbuf[64], preserve_dct[64];
//	double co[64];
#endif
//	double tmpl[64];
	int i, j;
	__int64 n64Start, n64End;
	__int64 n64BBmpegAAN = 0, n64BBmpegLLM = 0, n64FdctAM32 = 0, n64FdctMM32 = 0, n64FdctXvidMMX = 0, n64FdctXvidSSE2 = 0;
	__int64 n64IdctAAN = 0, n64IdctChen = 0, n64IdctInt32 = 0, n64Idct8x8SSE = 0, n64IdctAP922FloatSSE = 0, n64IdctAP922FloatX87 = 0, n64IdctAP922HybridX87 = 0, n64IdctXvidMMX = 0, n64IdctXvidXMM = 0, n64IdctXvidSSE2 = 0;
	__int64 n64SubPredMMX = 0, n64AddPredMMX = 0, n64SubPredSSE2 = 0;

	Init();
	srand((unsigned)time(NULL));

	for ( j = 0; j < LOOP; j++ )
	{
		for ( i = 0; i < 64; i++ )
		{
			buf[i] = rand() * 255 / RAND_MAX;
			buf2[i] = rand() * 255 / RAND_MAX;
		}
//		Standard_fDCT(buf, tmpl);

/*
		query_performance_count(n64Start);
		BBmpeg_AAN_fDct(buf, co);
		query_performance_count(n64End);
		n64BBmpegAAN += n64End - n64Start;
//		printf("The max absolute error (BBmpeg_AAN_fDct): %E\n", CompareDCT(co, tmpl));

		query_performance_count(n64Start);
		BBmpeg_LLM_fDct(buf, co);
		query_performance_count(n64End);
		n64BBmpegLLM += n64End - n64Start;
//		printf("The max absolute error (BBmpeg_LLM_fDct): %E\n", CompareDCT(co, tmpl));

		for ( i = 0; i < 64; i++ )
			tmpbuf[i] = (short)buf[i];
		query_performance_count(n64Start);
		fdct_am32(tmpbuf);
		query_performance_count(n64End);
		n64FdctAM32 += n64End - n64Start;
//		for ( i = 0; i < 64; i++ )
//			co[i] = (double)tmpbuf[i];
//		printf("The max absolute error (fdct_am32): %E\n", CompareDCT(co, tmpl));
*/
		for ( i = 0; i < 64; i++ )
			tmpbuf[i] = (short)buf[i];
		query_performance_count(n64Start);
		fdct_mm32(tmpbuf);
		query_performance_count(n64End);
		n64FdctMM32 += n64End - n64Start;
//		for ( i = 0; i < 64; i++ )
//			co[i] = (double)tmpbuf[i];
//		printf("The max absolute error (fdct_mm32): %E\n", CompareDCT(co, tmpl));

		for ( i = 0; i < 64; i++ )
			tmpbuf[i] = (short)buf[i];
		query_performance_count(n64Start);
		fdct_mmx(tmpbuf);
		query_performance_count(n64End);
		n64FdctXvidMMX += n64End - n64Start;

		for ( i = 0; i < 64; i++ )
			tmpbuf[i] = (short)buf[i];
		query_performance_count(n64Start);
//		fdct_sse2(tmpbuf);
		query_performance_count(n64End);
		n64FdctXvidSSE2 += n64End - n64Start;

		memcpy(preserve_dct, tmpbuf, sizeof(tmpbuf));
		query_performance_count(n64Start);
		idct_aan(tmpbuf);
		query_performance_count(n64End);
		n64IdctAAN += n64End - n64Start;

		memcpy(tmpbuf, preserve_dct, sizeof(tmpbuf));
		query_performance_count(n64Start);
		idct_chen(tmpbuf);
		query_performance_count(n64End);
		n64IdctChen += n64End - n64Start;

		memcpy(tmpbuf, preserve_dct, sizeof(tmpbuf));
		query_performance_count(n64Start);
		idct_int32(tmpbuf);
		query_performance_count(n64End);
		n64IdctInt32 += n64End - n64Start;

		memcpy(tmpbuf, preserve_dct, sizeof(tmpbuf));
		query_performance_count(n64Start);
		idct8x8_sse(tmpbuf);
		query_performance_count(n64End);
		n64Idct8x8SSE += n64End - n64Start;
/*
		memcpy(tmpbuf, preserve_dct, sizeof(tmpbuf));
		query_performance_count(n64Start);
		idct_ap922float_x87(tmpbuf);
		query_performance_count(n64End);
		n64IdctAP922FloatX87 += n64End - n64Start;

		memcpy(tmpbuf, preserve_dct, sizeof(tmpbuf));
		query_performance_count(n64Start);
		idct_ap922hybr_x87(tmpbuf);
		query_performance_count(n64End);
		n64IdctAP922HybridX87 += n64End - n64Start;

		memcpy(tmpbuf, preserve_dct, sizeof(tmpbuf));
		query_performance_count(n64Start);
		idct_ap922float_sse(tmpbuf);
		query_performance_count(n64End);
		n64IdctAP922FloatSSE += n64End - n64Start;
*/
		memcpy(tmpbuf, preserve_dct, sizeof(tmpbuf));
		query_performance_count(n64Start);
		idct_mmx(tmpbuf);
		query_performance_count(n64End);
		n64IdctXvidMMX += n64End - n64Start;

		memcpy(tmpbuf, preserve_dct, sizeof(tmpbuf));
		query_performance_count(n64Start);
		idct_xmm(tmpbuf);
		query_performance_count(n64End);
		n64IdctXvidXMM += n64End - n64Start;

		memcpy(tmpbuf, preserve_dct, sizeof(tmpbuf));
		query_performance_count(n64Start);
//		idct_sse2(tmpbuf);
		query_performance_count(n64End);
		n64IdctXvidSSE2 += n64End - n64Start;

		query_performance_count(n64Start);
		sub_pred_mmx(buf, buf2, 16, tmpbuf);
		query_performance_count(n64End);
		n64SubPredMMX += n64End - n64Start;

		query_performance_count(n64Start);
		add_pred_mmx(buf, buf2, 16, tmpbuf);
		query_performance_count(n64End);
		n64AddPredMMX += n64End - n64Start;

		query_performance_count(n64Start);
		sub_pred_sse2(buf, buf2, 16, tmpbuf);
		query_performance_count(n64End);
		n64SubPredSSE2 += n64End - n64Start;
	}

//	printf("n64BBmpegAAN   = %I64d,\tn64BBmpegLLM   = %I64d\n", n64BBmpegAAN / LOOP, n64BBmpegLLM / LOOP);
//	printf("n64FdctAM32    = %I64d,\tn64FdctMM32    = %I64d\n", n64FdctAM32 / LOOP, n64FdctMM32 / LOOP);
	printf("n64FdctMM32    = %I64d,\tn64FdctXvidMMX = %I64d\n", n64FdctMM32 / LOOP, n64FdctXvidMMX / LOOP);
	printf("n64FdctXvidSSE2 = %I64d\n", n64FdctXvidSSE2 / LOOP);

	printf("\n");

	printf("n64IdctAAN   = %I64d,\tn64IdctChen   = %I64d\n", n64IdctAAN / LOOP, n64IdctChen / LOOP);
	printf("n64IdctInt32 = %I64d,\tn64Idct8x8SSE = %I64d\n", n64IdctInt32 / LOOP, n64Idct8x8SSE / LOOP);
//	printf("n64IdctAP922HybridX87 = %I64d,\tn64IdctAP922FloatX87 = %I64d\n", n64IdctAP922HybridX87 / LOOP, n64IdctAP922FloatX87 / LOOP);
//	printf("n64IdctAP922FloatSSE  = %I64d\n", n64IdctAP922FloatSSE / LOOP);
	printf("n64IdctXvidMMX        = %I64d,\tn64IdctXvidXMM       = %I64d\n", n64IdctXvidMMX / LOOP, n64IdctXvidXMM / LOOP);
	printf("n64IdctXvidSSE2       = %I64d\n", n64IdctXvidSSE2 / LOOP);

	printf("\n");

	printf("n64SubPredMMX  = %I64d,\tn64AddPredMMX  = %I64d\n", n64SubPredMMX / LOOP, n64AddPredMMX / LOOP);
	printf("n64SubPredSSE2 = %I64d\n", n64SubPredSSE2 / LOOP);

	return 0;
}