freenode/#clasp - IRC Chatlog
Search
7:01:43
drmeister
With *optimization-level* of 3 Compile-file time seconds real(24.4) run(24.4) llvm(13.1) link(5.3)
7:03:00
drmeister
With *optimization-level* of 0 Compile-file time seconds real(1.9) run(1.9) llvm(0.1) link(0.8)
7:07:44
drmeister
(cmp:*optimization-level* 3)Compile-file time seconds real(14.1) run(14.1) llvm(3.3) link(4.9)
7:08:01
drmeister
(cmp:*optimization-level* 0)Compile-file time seconds real(10.9) run(10.9) llvm(0.4) link(4.3)
12:59:23
beach
karlosz: Yes, and what they mean is that it won't work if you use ONLY those, without taking executability into account.
13:00:33
karlosz
yeah, but i feel like there still needs to be a clause handling merge points, even in SFA
13:01:50
beach
Yes, some variables will have several definitions. Hence the UD chains rather than just a single SSA edge.
13:03:25
karlosz
if we let v represent the merge point variable, then the UD chain would look like (v, {3, 4})
13:03:56
karlosz
i think you might be missing a clause saying that the lattice cell of the variable with multiple defs is derived from the executable defs in the UD chain
13:05:01
beach
Evaluate the expression obtaining the values of the operands from the LatticCells at the DEFINE side of the UD chains of this node.