Leo --
Here's the simple bench.jako example (again):
const int N = 100;
var int q = 1;
while (q < N) {
var int i, j, w = 1;
while (w < N) {
i++;
j += i;
w++;
# print("$q, $w\n");
}
q++;
}
I used to work harder in the Jako compiler to track constants
and substitute their values, but I figured thats the sort of
shared functionality IMCC would either have now or in future.
So, I've removed those smarts from jakoc (which has the added
benefit, IMO, of leaving symbolic constants symbolic in the
..imc file). Now, I generate the following IMC code for the
above Jako code:
.sub __MODULE__
.local int N # const int N = 100;
N = 100
.local int q # var int q;
q = 1 # q = 1;
.namespace WHILE_BLOCK
_W1_WHILE:
_W1_NEXT: ge q, N, _W1_LAST # _W1: while (q < N)
{
_W1_REDO: .local int i # var int i;
i = 1 # i = 1;
.local int j # var int j;
j = 1 # j = 1;
.local int w # var int w;
w = 1 # w = 1;
.namespace WHILE_BLOCK
_W2_WHILE:
_W2_NEXT: ge w, N, _W2_LAST # _W2: while (w <
N) {
_W2_REDO: add i, 1 # i += 1;
add j, i # j += i;
add w, 1 # w += 1;
_W2_CONT: branch _W2_NEXT # }
_W2_LAST: .endnamespace WHILE_BLOCK
add q, 1 # q += 1;
_W1_CONT: branch _W1_NEXT # }
_W1_LAST: .endnamespace WHILE_BLOCK
end
.end
(I'm still not making use of all the imcc language features)
You can see that as far as the generated IMC is concerned,
variables and constants are treated equivalently. I do track
the const-ness of the identifier to prevent Jako code from
trying to do "N = 5" when N was declared earlier as a constant.
The current IMCC converts the above to this:
__MODULE__:
set I4, 100
set I3, 1
_W1_WHILE:
_W1_NEXT:
ge I3, I4, _W1_LAST
_W1_REDO:
set I1, 1
set I2, 1
set I0, 1
_W2_WHILE:
_W2_NEXT:
ge I0, I4, _W2_LAST
_W2_REDO:
add I1, 1
add I2, I1
add I0, 1
_W2_CONT:
branch _W2_NEXT
_W2_LAST:
add I3, 1
_W1_CONT:
branch _W1_NEXT
_W1_LAST:
end
Which looks pretty good. Here we can see that IMCC does
not discover that I4 can be optimized away. Is that something
I should be expecting in forthcoming IMCC releases? Or, should
I go back to handling it in the Jako compiler?
I was thinking it might be nice to be able to do something like
this in the .imc file:
.local int N 100 # Constant implied by value after name
BTW, thanks for producing imcc. I'm very happy I don't have to
do register allocation anymore. The Jako compiler is still not
a work of art, but imcc is allowing me to incrementally remove
complexity (which I'm turning right around and re-adding in
other areas... :). And, thanks also for fielding my numerous
questions as promptly as you have!
Regards,
-- Gregor
I am that which is not everything else.
