Skip to content
Find file
Fetching contributors…
Cannot retrieve contributors at this time
88 lines (86 sloc) 4 KB
- admit static memoization
+ implement static mutable vectors:
x cogen-abssyn.scm:(define (annIsRef? e)
x cogen-abssyn.scm: ((annIsRef? e)
x cogen-abssyn.scm: ((annIsRef? e)
x cogen-abssyn.scm: ((annIsRef? e)
x cogen-effect.scm: ((annIsRef? e)
x cogen-effect.scm: ((annIsRef? e)
x cogen-eq-flow.scm: ((annIsRef? e)
x cogen-eq-flow.scm: ((annIsRef? e)
x cogen-eq-flow.scm: ((annIsRef? e)
x cogen-eq-flow.scm: ((annIsRef? e)
x cogen-eq-flow.scm: ((annIsRef? e)
x cogen-oca.scm: ((annIsRef? e)
cogen-reach.scm: ((annIsRef? e)
cogen-reach.scm: ((annIsRef? e)
cogen-reach.scm: ((annIsRef? e)
cogen-reach.scm: ((annIsRef? e)
cogen-reach.scm: ((annIsRef? e)
cogen-reach.scm: ((annIsRef? e)
x cogen-skeleton.scm: ((annIsRef? e)
cogen-terminate.scm: ((annIsRef? e)
cogen-terminate.scm: ((annIsRef? e)
x pgg-packages.scm: annIsRef?
x cogen-library.scm
x cogen-direct-anf.scm
x cogen-construct-genext.scm
+ insert memoization points more cleverly: only the outermost lam-D or if-D
+ implement coarity raising
- extend the top-level such that arbitrary Scheme programs (including
non-functional definitions) are allowed
+ use the proposed macro mechanism of R4RS to implement non-primitive forms
- automatically insert representation coercions for (partially) static
top-level arguments (from the standard representation to the memoized one)
- revise symbol handling to pairs of (symbol, version number)
- change the way static skeletons are represented: include information
on sharing of dynamic values by replacing them in a consistent way
- change the way static and dynamic parts are projected: process only one
value/btime pair at once
+ analyze which lambdas and ctors are really memoized
+ incorporate the residual program in the goal function
+ make a macro out of the (_ i ...) functions
+ put type annotations to work: enforce types of primitives in bta-eq-flow
- type annotation for arity of operators
+ inspect the lambda lifter (dependency analysis?)
- insert LET only if the variable appears more than once in the body
+ reconsider LET insertion strategy (constructor arguments!)
- how many memoization points are really necessary?
you may want to read Malmkjaer's PEPM'93 paper
+ support for EVAL
+ complete the above (preprocessor, bta, skeleton)
+ check whether it still performs continuation-based reduction :-)
+ really generate LAMBDA_MEMO and friends
+ control memoization
+ handle (begin ...)
+ handle nested (define ...)
- specialize project-static, project-dynamic, and friends
+ use delay/force in project-* functions
+ write the library functions in direct style using shift/reset to
achieve continuation-based reduction
+ printout mechanism for bt analyzed program?
+ use introspection/reflective facilities to obtain the values of the
free variables
- support variadic procedures (by considering argument lists as a
special partially static datatype)
- the code generating functions return two results, the generated code
and a synthesized attribute which indicates the "seriousness" of the
code. The idea is that functions with non-serious body are unfolded
on the fly.
- post unfolding: keep track of # calls in the memo table by storing
an indirection to the call in the residual program. After generation
of the residual program we traverse the memo table once, performing
the unfoldings.
- allow user control over caching.
- implement more a efficient caching mechanism.
- implement parameterized abstract syntax: first define some macros,
then define ast accessors and constructors using them. How about
using some record facility?
+ add record support
- replace the symtab by environments (cogen-env)
- let the spec transform the residual program to CPS in order to
memoize partially static values (alternative: use call-with-values)
- handle load-time evaluations properly. define load+spec function?
+ type of a top-level value might be a poly, does not seem to be
propagated to application site
Something went wrong with that request. Please try again.