Commit 140e68f5 authored by Martin Reinecke's avatar Martin Reinecke

more tweaks

parent f926ace2
......@@ -917,18 +917,20 @@ template<bool fwd, typename T> void pass8 (size_t ido, size_t l1,
T a0, a1, a2, a3, a4, a5, a6, a7;
PMC(a1,a5,CC(0,1,k),CC(0,5,k));
PMC(a3,a7,CC(0,3,k),CC(0,7,k));
ROTX90<fwd>(a7);
PMINPLACE(a1,a3);
ROTX90<fwd>(a3);
ROTX90<fwd>(a7);
PMINPLACE(a5,a7);
ROTX45<fwd>(a5);
ROTX135<fwd>(a7);
PMC(a0,a4,CC(0,0,k),CC(0,4,k));
PMC(a2,a6,CC(0,2,k),CC(0,6,k));
ROTX90<fwd>(a6);
PMC(CH(0,k,0),CH(0,k,4),a0+a2,a1);
PMC(CH(0,k,1),CH(0,k,5),a4+a6,a5);
PMC(CH(0,k,2),CH(0,k,6),a0-a2,a3);
ROTX90<fwd>(a6);
PMC(CH(0,k,1),CH(0,k,5),a4+a6,a5);
PMC(CH(0,k,3),CH(0,k,7),a4-a6,a7);
}
else
......@@ -938,18 +940,20 @@ template<bool fwd, typename T> void pass8 (size_t ido, size_t l1,
T a0, a1, a2, a3, a4, a5, a6, a7;
PMC(a1,a5,CC(0,1,k),CC(0,5,k));
PMC(a3,a7,CC(0,3,k),CC(0,7,k));
ROTX90<fwd>(a7);
PMINPLACE(a1,a3);
ROTX90<fwd>(a3);
ROTX90<fwd>(a7);
PMINPLACE(a5,a7);
ROTX45<fwd>(a5);
ROTX135<fwd>(a7);
PMC(a0,a4,CC(0,0,k),CC(0,4,k));
PMC(a2,a6,CC(0,2,k),CC(0,6,k));
ROTX90<fwd>(a6);
PMC(CH(0,k,0),CH(0,k,4),a0+a2,a1);
PMC(CH(0,k,1),CH(0,k,5),a4+a6,a5);
PMC(CH(0,k,2),CH(0,k,6),a0-a2,a3);
ROTX90<fwd>(a6);
PMC(CH(0,k,1),CH(0,k,5),a4+a6,a5);
PMC(CH(0,k,3),CH(0,k,7),a4-a6,a7);
}
for (size_t i=1; i<ido; ++i)
......@@ -965,15 +969,15 @@ template<bool fwd, typename T> void pass8 (size_t ido, size_t l1,
ROTX135<fwd>(a7);
PMC(a0,a4,CC(i,0,k),CC(i,4,k));
PMC(a2,a6,CC(i,2,k),CC(i,6,k));
ROTX90<fwd>(a6);
PMINPLACE(a0,a2);
PMINPLACE(a4,a6);
CH(i,k,0) = a0+a1;
CH(i,k,4) = (a0-a1).template special_mul<fwd>(WA(3,i));
CH(i,k,1) = (a4+a5).template special_mul<fwd>(WA(0,i));
CH(i,k,5) = (a4-a5).template special_mul<fwd>(WA(4,i));
CH(i,k,2) = (a2+a3).template special_mul<fwd>(WA(1,i));
CH(i,k,6) = (a2-a3).template special_mul<fwd>(WA(5,i));
ROTX90<fwd>(a6);
PMINPLACE(a4,a6);
CH(i,k,1) = (a4+a5).template special_mul<fwd>(WA(0,i));
CH(i,k,5) = (a4-a5).template special_mul<fwd>(WA(4,i));
CH(i,k,3) = (a6+a7).template special_mul<fwd>(WA(2,i));
CH(i,k,7) = (a6-a7).template special_mul<fwd>(WA(6,i));
}
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment