bn/asm/x86_64-gcc.c

298 /* mul_add_c(a,b,c0,c1,c2)  -- c+=a*b for three word number c=(c2,c1,c0) */
312 #  define mul_add_c(a,b,c0,c1,c2) {       \
330 #  define mul_add_c(a,b,c0,c1,c2) do {    \
387     mul_add_c(a[0], b[0], c1, c2, c3);
390     mul_add_c(a[0], b[1], c2, c3, c1);
391     mul_add_c(a[1], b[0], c2, c3, c1);
394     mul_add_c(a[2], b[0], c3, c1, c2);
395     mul_add_c(a[1], b[1], c3, c1, c2);
396     mul_add_c(a[0], b[2], c3, c1, c2);
399     mul_add_c(a[0], b[3], c1, c2, c3);
400     mul_add_c(a[1], b[2], c1, c2, c3);
401     mul_add_c(a[2], b[1], c1, c2, c3);
402     mul_add_c(a[3], b[0], c1, c2, c3);
405     mul_add_c(a[4], b[0], c2, c3, c1);
406     mul_add_c(a[3], b[1], c2, c3, c1);
407     mul_add_c(a[2], b[2], c2, c3, c1);
408     mul_add_c(a[1], b[3], c2, c3, c1);
409     mul_add_c(a[0], b[4], c2, c3, c1);
412     mul_add_c(a[0], b[5], c3, c1, c2);
413     mul_add_c(a[1], b[4], c3, c1, c2);
414     mul_add_c(a[2], b[3], c3, c1, c2);
415     mul_add_c(a[3], b[2], c3, c1, c2);
416     mul_add_c(a[4], b[1], c3, c1, c2);
417     mul_add_c(a[5], b[0], c3, c1, c2);
420     mul_add_c(a[6], b[0], c1, c2, c3);
421     mul_add_c(a[5], b[1], c1, c2, c3);
422     mul_add_c(a[4], b[2], c1, c2, c3);
423     mul_add_c(a[3], b[3], c1, c2, c3);
424     mul_add_c(a[2], b[4], c1, c2, c3);
425     mul_add_c(a[1], b[5], c1, c2, c3);
426     mul_add_c(a[0], b[6], c1, c2, c3);
429     mul_add_c(a[0], b[7], c2, c3, c1);
430     mul_add_c(a[1], b[6], c2, c3, c1);
431     mul_add_c(a[2], b[5], c2, c3, c1);
432     mul_add_c(a[3], b[4], c2, c3, c1);
433     mul_add_c(a[4], b[3], c2, c3, c1);
434     mul_add_c(a[5], b[2], c2, c3, c1);
435     mul_add_c(a[6], b[1], c2, c3, c1);
436     mul_add_c(a[7], b[0], c2, c3, c1);
439     mul_add_c(a[7], b[1], c3, c1, c2);
440     mul_add_c(a[6], b[2], c3, c1, c2);
441     mul_add_c(a[5], b[3], c3, c1, c2);
442     mul_add_c(a[4], b[4], c3, c1, c2);
443     mul_add_c(a[3], b[5], c3, c1, c2);
444     mul_add_c(a[2], b[6], c3, c1, c2);
445     mul_add_c(a[1], b[7], c3, c1, c2);
448     mul_add_c(a[2], b[7], c1, c2, c3);
449     mul_add_c(a[3], b[6], c1, c2, c3);
450     mul_add_c(a[4], b[5], c1, c2, c3);
451     mul_add_c(a[5], b[4], c1, c2, c3);
452     mul_add_c(a[6], b[3], c1, c2, c3);
453     mul_add_c(a[7], b[2], c1, c2, c3);
456     mul_add_c(a[7], b[3], c2, c3, c1);
457     mul_add_c(a[6], b[4], c2, c3, c1);
458     mul_add_c(a[5], b[5], c2, c3, c1);
459     mul_add_c(a[4], b[6], c2, c3, c1);
460     mul_add_c(a[3], b[7], c2, c3, c1);
463     mul_add_c(a[4], b[7], c3, c1, c2);
464     mul_add_c(a[5], b[6], c3, c1, c2);
465     mul_add_c(a[6], b[5], c3, c1, c2);
466     mul_add_c(a[7], b[4], c3, c1, c2);
469     mul_add_c(a[7], b[5], c1, c2, c3);
470     mul_add_c(a[6], b[6], c1, c2, c3);
471     mul_add_c(a[5], b[7], c1, c2, c3);
474     mul_add_c(a[6], b[7], c2, c3, c1);
475     mul_add_c(a[7], b[6], c2, c3, c1);
478     mul_add_c(a[7], b[7], c3, c1, c2);
491     mul_add_c(a[0], b[0], c1, c2, c3);
494     mul_add_c(a[0], b[1], c2, c3, c1);
495     mul_add_c(a[1], b[0], c2, c3, c1);
498     mul_add_c(a[2], b[0], c3, c1, c2);
499     mul_add_c(a[1], b[1], c3, c1, c2);
500     mul_add_c(a[0], b[2], c3, c1, c2);
503     mul_add_c(a[0], b[3], c1, c2, c3);
504     mul_add_c(a[1], b[2], c1, c2, c3);
505     mul_add_c(a[2], b[1], c1, c2, c3);
506     mul_add_c(a[3], b[0], c1, c2, c3);
509     mul_add_c(a[3], b[1], c2, c3, c1);
510     mul_add_c(a[2], b[2], c2, c3, c1);
511     mul_add_c(a[1], b[3], c2, c3, c1);
514     mul_add_c(a[2], b[3], c3, c1, c2);
515     mul_add_c(a[3], b[2], c3, c1, c2);
518     mul_add_c(a[3], b[3], c1, c2, c3);