sse2mmx.c

This a framework to test new ideas in transmission technology. Actual development is a LDPC-coder in

#include "sse2mmx.h"
#include "spc.h"

/* Initializes xmm6 and xmm7 with TA loaded in xmm2
 */
void init67TA() {
  asm("movdqa %xmm2, %xmm7");
  asm("movdqa %xmm2, %xmm6");
}

/* loads TA in xmm2
 */
float a[8] = { TA, TA, TA, TA, TA, TA, TA, TA };
void load2TA() {
  asm("movdqa %0, %%xmm2"::"m"(a[0]));
}
/* adds a to xmm7 and stores result in xmm7
 */
void add
  (short int *a) {
  asm("movdqa %0, %%xmm0"::"m"(a[0]));
  asm("paddw %xmm0, %xmm7");
}

/* loads the received values a in xmm7
 */
void load7(short int *a) {
  asm("movdqu %0, %%xmm7"::"m"(a[0]));
}

/* loads 8 short integers from a as 4 floats each into xmm4 and xmm5
 */
void load45data(short int *a) {
  asm("movdqa %0, %%xmm3"::"m"(a[0]));
  asm("movdqa %xmm3,%xmm4");
  asm("movdqa %xmm4,%xmm5");
  asm("psraw $15,%xmm3");
  asm("punpcklwd %xmm3,%xmm4");
  asm("punpckhwd %xmm3,%xmm5");
  asm("cvtdq2ps %xmm4,%xmm4");
  asm("cvtdq2ps %xmm5,%xmm5");
}

/* loads the 8 received short integers as 4 floats each in xmm6 and xmm7
 */
void load67data(short int *a) {
  asm("movdqu %0, %%xmm3"::"m"(a[0]));
  asm("movdqu %xmm3,%xmm6");
  asm("movdqu %xmm4,%xmm7");
  asm("psraw $15,%xmm3");
  asm("punpcklwd %xmm3,%xmm6");
  asm("punpckhwd %xmm3,%xmm7");
  asm("cvtdq2ps %xmm6,%xmm6");
  asm("cvtdq2ps %xmm7,%xmm7");
}

/* does prod * msg /TA where
 * prod is in xmm6 and xmm7
 * TA is in xmm2
 * msg is in xmm4 and xmm5
 */
void mulfloat() {
  asm("mulps %xmm4,%xmm6");
  asm("mulps %xmm5,%xmm7");
  asm("divps %xmm2,%xmm7");
  asm("divps %xmm2,%xmm6");
}

/* saves xmm4 in a and xmm5 in b
 */
void save45temp(float * a,float * b) {
  asm("movdqa %%xmm4,%0"::"m"(a[0]));
  asm("movdqa %%xmm5,%0"::"m"(b[0]));

}
/* saves xmm6 in a and xmm7 in b
 */
void save67temp(float * a,float * b) {
  asm("movdqu %%xmm6,%0"::"m"(a[0]));
  asm("movdqu %%xmm7,%0"::"m"(b[0]));

}

/* puts prod * TA / message in xmm4 and xmm5
 */
void divide(float * a,float *b) {
  asm("movdqa %xmm6,%xmm4");
  asm("movdqa %xmm7,%xmm5");
  asm("mulps %xmm2,%xmm4");
  asm("mulps %xmm2,%xmm5");
  asm("movdqa %0,%%xmm1"::"m"(a[0]));
  asm("movdqa %0,%%xmm3"::"m"(b[0]));
  asm("divps %xmm1,%xmm4");
  asm("divps %xmm3,%xmm5");
}

/* converts 4 floats each from xmm4 and xmm5
 * to short integers and stores in a
 */
void loadmessage45(short int * a) {
  asm("cvtps2dq %xmm4,%xmm4");
  asm("cvtps2dq %xmm5,%xmm5");
  asm("packssdw %xmm5,%xmm4");
asm("movdqa %%xmm4,%0":"=m"(a[0]));
}

/* converts 4 floats each from xmm6 and xmm7
 * to short integers and stores in a
 */
void loadmessage67(short int * a) {
  asm("cvtps2dq %xmm6,%xmm6");
  asm("cvtps2dq %xmm7,%xmm7");
  asm("packssdw %xmm7,%xmm6");
asm("movdqa %%xmm6,%0":"=m"(a[0]));
}

/* multiplies xmm7 with a and stores result in xmm7
 */
void mul(short int *a) {
  asm("pmullw %0, %%xmm7"::"m"(a[0]));
}

/* subtracts message a from xmm7 and stores result in xmm0
 */
void sub(short int *a) {
  asm("movdqu %0, %%xmm1"::"m"(a[0]));
  asm("movdqu %xmm7,%xmm0");
  asm("psubw %xmm1,%xmm0");
}

/*loads xmm0 in a
 */
void retrieve0(short int *a) {
asm("movdqu %%xmm0, %0":"=m"(a[0]));
}

/* loads xmm7 in a
 */
void retrieve7(short int *a) {
asm("movdqu %%xmm7, %0":"=m"(a[0]));
}

/* loads x in xmm6 and -x in xmm5
 */
short int init_a[8];
short int init_b[8];
void init(short int x) {
  init_a[0]=-x;
  init_a[1]=-x;
  init_a[2]=-x;
  init_a[3]=-x;
  init_a[4]=-x;
  init_a[5]=-x;
  init_a[6]=-x;
  init_a[7]=-x;
  init_b[0]=x;
  init_b[1]=x;
  init_b[2]=x;
  init_b[3]=x;
  init_b[4]=x;
  init_b[5]=x;
  init_b[6]=x;
  init_b[7]=x;

  asm("movdqu %0, %%xmm5"::"m"(init_a[0]));
  asm("movdqu %0, %%xmm6"::"m"(init_b[0]));
}

/* limits xmm0 to lie between values in xmm5 and xmm6
 */
void limit() {
  asm("pminsw %xmm6 , %xmm0");
  asm("pmaxsw %xmm5 , %xmm0");

}