After exploring the JSR-292 API, looks like dataflow-style keeps the code
free from hardcoded signatures that expose datatypes (int, float, etc). It
would be great to hear from projects testing this approach. For example,
adding three operands (either ints or floats) in a datatype-agnostic manner
becomes:
@ForceInlining public final
MethodHandle add3(MethodHandle rand1,
MethodHandle rand2,
MethodHandle rand3) {
return add2(add2(rand1, rand2),
rand3);
}
@ForceInlining public final
MethodHandle add2(MethodHandle a,
MethodHandle b) {
// assert: both operands have the same datatype (int or float)
if(a.type().returnType() == INT) {
int result = ( (int) a.invokeExact() // this triggers all
side-effects of evaluating the first operand
+ (int) b.invokeExact()
)
return constantMH(result);
} else {
... // ditto for float
}
The compilation scheme above does without reifying everything into MHs,
e.g. loops, try-catch-finally, and all of switch, ternary operation,
if-then stmt; can appear in their plain formulation. On the other hand,
accesses to datatype-dependent fields and local vars has to be mediated via
MHs.
I really hope the above can be made to run at native speed (after inlining
and tracing).
P.S.: It would be great to have an API for the above besides the current
one ;)
Miguel
http://lampwww.epfl.ch/~magarcia/ScalaCompilerCornerReloaded/
--
You received this message because you are subscribed to the Google Groups "JVM
Languages" group.
To view this discussion on the web visit
https://groups.google.com/d/msg/jvm-languages/-/KDJVFz6ZFm0J.
To post to this group, send email to jvm-languages@googlegroups.com.
To unsubscribe from this group, send email to
jvm-languages+unsubscr...@googlegroups.com.
For more options, visit this group at
http://groups.google.com/group/jvm-languages?hl=en.
