/*
	Transpose 4x4 matrix (in-place) using SSE instructions, $Revision: 1.4 $
	
	Author: Wojciech Muła
	e-mail: wojciech_mula@poczta.onet.pl
	www:    http://0x80.pl/
	
	License: public domain
	
	initial release 10.03.2002, last update $Date: 2008-06-08 23:00:44 $
*/

#include "sse-aux.c"

// start-snippet
void sse_transpose(float mat[4*4]) {
__asm__ volatile (
                                //         0  1  2  3 <- index
   "movups 0x00(%0), %%xmm0 \n" // xmm0 := a0 a1 a2 a3
   "movups 0x10(%0), %%xmm1 \n" // xmm1 := b0 b1 b2 b3
   "movups 0x20(%0), %%xmm2 \n" // xmm2 := c0 c1 c2 c3
   "movups 0x30(%0), %%xmm3 \n" // xmm3 := d0 d1 d2 d3
   
   "movaps   %%xmm0, %%xmm4 \n"
   "unpcklps %%xmm2, %%xmm0 \n" // xmm0 := a0 c0 a1 c1
   "unpckhps %%xmm2, %%xmm4 \n" // xmm4 := a2 c2 a3 c3
   
   "movaps   %%xmm1, %%xmm2 \n"
   "unpcklps %%xmm3, %%xmm1 \n" // xmm1 := b0 d0 b1 d1
   "unpckhps %%xmm3, %%xmm2 \n" // xmm2 := b2 d2 b3 d3
   
   "movaps   %%xmm0, %%xmm3 \n"
   "unpcklps %%xmm1, %%xmm0 \n" // xmm0 := a0 b0 c0 d0
   "unpckhps %%xmm1, %%xmm3 \n" // xmm3 := a1 b1 c1 d1
   
   "movaps   %%xmm4, %%xmm1 \n"
   "unpcklps %%xmm2, %%xmm1 \n" // xmm1 := a2 b2 c2 d2
   "unpckhps %%xmm2, %%xmm4 \n" // xmm4 := a3 b3 c3 d3
   
   "movups %%xmm0, 0x00(%0) \n"
   "movups %%xmm3, 0x10(%0) \n"
   "movups %%xmm1, 0x20(%0) \n"
   "movups %%xmm4, 0x30(%0) \n"
   : "=r" (mat)
   : "0" (mat)
   : "memory"
);
}
// end-snippet


int main() {
	initfrand();

	float mat[4*4];
	int i=0;
	for (i=0; i < 4*4; i++)
		mat[i] = i;

	printf("M = \n");
	for (i=0; i < 4; i++) print_vec_float(&mat[i*4]);

	sse_transpose(mat);

	printf("M^T = \n");
	for (i=0; i < 4; i++) print_vec_float(&mat[i*4]);

	return 0;
}

// vim: ts=4 sw=4 nowrap