ELECH473 - Arthur Valingot - Lab 2 AT&T version

main.c

// ConsoleApplication2.cpp : Defines the entry point for the console application.
//

#include <stdlib.h>  
#include <stdio.h>
// #include <windows.h>
// #pragma warning(disable : 4996)


/*
	Result
	Code C debug: 0.003631, release: 0.002187 gain significatif
	Code assembleur:  debug: 0.001466, release: 0.001410 gain non significatif // pas beaucoup de changement entre debug et release
	Conclusion: La partie en assembleur est plus rapide, deplus les changements de mode ( debug / release) n'ont pas d'effet, alors que les effets sont spectaculaire sur le code en C
*/
int main()
{
	// LARGE_INTEGER frequency;
	// LARGE_INTEGER start;
	// LARGE_INTEGER end;
	// double interval;
	int W = 1024, H = 1024;
	unsigned char *threshold;
	unsigned char *mask;
	int threshold_1 = 125;
	unsigned char *src;
	unsigned char *dst;
	char c;
	
	src = (unsigned char *) malloc (W*H*sizeof(unsigned char));
	dst = (unsigned char *) malloc (W*H*sizeof(unsigned char));




	threshold = (unsigned char *)malloc(16 * sizeof(unsigned char));
	mask = (unsigned char *)malloc(16 * sizeof(unsigned char));
	for (int i = 0; i < 16; i++) {
		threshold[i] = (unsigned char)threshold_1;
		mask[i] = (unsigned char)128;
	}

	if (src == NULL || dst == NULL) {
		printf("Out of memory!");
		exit(1);
	}
	
	// FILE *fp1 = fopen("C:\\Users\\Arthur\\Desktop\\ULB Cours\\Microprocesseur Archi\\Labs\\test.raw", "r");
	// FILE *fp2 = fopen("C:\\Users\\Arthur\\Desktop\\ULB Cours\\Microprocesseur Archi\\Labs\\test_contourDetector.raw", "w");
	FILE *fp1 = fopen("/home/para/Documents/ULB/ELECH473/2017-2018/SIMD/students/LAB7ArthurValingot/test.raw", "r");
	FILE *fp2 = fopen("/home/para/Documents/ULB/ELECH473/2017-2018/SIMD/students/LAB7ArthurValingot/test_contourDetector.raw", "w");

	// QueryPerformanceFrequency(&frequency);
	// QueryPerformanceCounter(&start);
	if (fp1 != NULL) {
		fread(src, sizeof(unsigned char), W*H, fp1);
		for (int i = 0; i < W*H; i++)
			dst[i] = src[i];
		// for (int i = 0; i < 16; i++)
		// 	printf("%d \n", (int)src[888670 + i]);
		// src += 888670;
	// put our code here
		 //C code, uncomment to get it
		//for (int i = 0; i < W*H; i++) {
		//	if ((int)src[i] < threshold_1) {
		//		dst[i] = 0;
		//	}
		//	else {
		//		dst[i] = 255;
		//	}
		//}
	// assembly code
		int ii = W*H / 14;
		ii = ii - 2;
		__asm__(
			"mov	%[src], %%esi\n"
			"mov	%[ii], %%ecx\n"
			"mov 	%[dst], %%edi\n"
		"l1:\n"
			"movdqu (%%esi), %%xmm0\n"
			"movdqu 1024(%%esi), %%xmm1\n"
			"movdqu 2048(%%esi), %%xmm2\n"
			"movdqu %%xmm0, %%xmm3\n"
			"pmaxub %%xmm1, %%xmm0\n"
			"pmaxub %%xmm2, %%xmm0\n"
			"pminub %%xmm1, %%xmm3\n"
			"pminub %%xmm2, %%xmm3\n"
			"movdqu %%xmm0, %%xmm6\n"
			"movdqu %%xmm0, %%xmm7\n"
			"movdqu %%xmm3, %%xmm4\n"
			"movdqu %%xmm3, %%xmm5\n"
			"psrldq $1, %%xmm6\n"
			"psrldq $2, %%xmm7\n"
			"psrldq $1, %%xmm4\n"
			"psrldq $2, %%xmm5\n"
			"pmaxub %%xmm7, %%xmm6\n"
			"pmaxub %%xmm0, %%xmm6\n"
			"pminub %%xmm5, %%xmm4\n"
			"pminub %%xmm3, %%xmm4\n"
			"psubusb %%xmm4, %%xmm6\n"
			"movdqu %%xmm6, (%%edi)\n"
			"add $14, %%esi\n"
			"add $14, %%edi\n"
			"sub $1, %%ecx\n"
		"jnz l1\n"
        :"=m"(src), "=m"(dst)//outputs
        :[src]"m" (src), [dst]"m" (dst), [ii]"m" (ii)//inputs
        :"%esi", "%ecx", "%edx", "%xmm1"//clobbers
		);
		printf("ASM computation over.\n");
		// QueryPerformanceCounter(&end);
		fclose(fp1);}
	else {
		printf("Can't open specified file!");
		exit(1);
	}

	if (fp2 != NULL) {
		fwrite(dst, sizeof(unsigned char), W*H, fp2);
		fclose(fp2);
	}
	else {
		printf("Can't open specified file!");
		exit(1);
	}
	free(dst);
	free(src);
	free(threshold);
	// interval = (double)(end.QuadPart - start.QuadPart) / frequency.QuadPart;
	// printf("%f", interval);
	// scanf("%c", &c);
    return 0;
}