* module/language/cps/slot-allocation.scm (allocate-slots): For
continuations of $call, $callk, and $values with multiple
predecessors, recalculate the set of live slots. Fixes miscompilation
of ice-9/futures.scm:process-future!, broken since the previous patch,
now that $kreceive continuations can have multiple predecessors.
* module/language/cps/renumber.scm (compute-new-labels-and-vars):
(compute-tail-path-lengths, sort-conts): Arrange to visit successors
in such a way that if branches are unsorted, the longest path length
will appear first. This keeps loop bodies together.
* module/language/cps/type-fold.scm (fold-and-reduce): Don't require
types to check out; it could be that the reduced expression can
exhibit the same type-check effects. Reduce for all continuations,
even $kreceive. Pass dfg to reducer.
(mul): Check types.
(logbit?): New reducer.
* module/language/cps/effects-analysis.scm: Add entries for logtest and
logbit?.
* module/language/cps/types.scm (logtest, logbit?): New checkers and
inferrers.
* module/language/tree-il/peval.scm (peval): Convert (zero? (logand a
b)) to (logtest a b), in anticipation of opcode support for logtest.
*
module/language/tree-il/primitives.scm (*interesting-primitive-names*):
(*effect-free-primitives*): Add logtest and logbit?.
* module/language/cps/dfg.scm (compute-live-variables): Convert to use
intsets, and fold in compute-maximum-fixed-point.
(print-dfa): Update.
* module/language/cps/slot-allocation.scm (dead-after-def?)
(dead-after-use?, allocate-slots): Convert to use intsets.
This was a failed experiment. It had good space complexity but terrible
time complexity.
* module/Makefile.am: Update.
* module/language/cps/nameset.scm: Remove.
* module/Makefile.am: Build types.scm early, but don't block the rest of
the build on it.
* module/language/cps/types.scm: Rewrite to use namesets.
* module/language/cps/dce.scm:
* module/language/cps/type-fold.scm: Adapt to interface changes.
* module/language/cps/cse.scm:
* module/language/cps/dfg.scm (compute-idoms, compute-dom-edges): Move
these procedures from cse.scm to dfg.scm.
Remove loop-detection code; that can come back later but it is
bitrotten for now.
* module/language/cps/effects-analysis.scm (define-enumeration): New
helper.
(&memory-kind-mask): Define as an enumeration, not a bitfield. Add
&unknown-memory-kinds.
(&all-effects, effect-clobbers?, make-prompt-tag, expression-effects):
Adapt.
Note that this change requires dce.go and cse.go to be recompiled.
* module/language/cps/dce.scm (elide-type-checks!, compute-live-code):
Replace old ad-hoc type check elision with one driven from type
analysis. Type check elision only operates on smallish functions, to
avoid n**2 explosion in type inference.
* module/language/cps/types.scm (infer-types): Add #:max-label-count
argument.
* module/language/cps/type-fold.scm (compute-folded, fold-constants*):
Disable for big functions. Perhaps we can relax this if we find an
O(n log n) way to represent types.
* module/language/cps/types.scm: New file, implementing type and range
inference over CPS.
* module/language/cps/type-fold.scm: New file, implementing abstract
constant folding for CPS.
* module/Makefile.am: Add the new files.
* module/language/cps/compile-bytecode.scm: Wire up type-fold, but
currently disabled.
* module/language/cps/effects-analysis.scm: Rewrite so that instead of
the depends/causes effects, there is just &type-check, &allocation,
&read, and &write. The object kind is a separate part of the
bitfield, and the field in the object (if appropriate) is another
field. Effects are still a fixnum. This enables precise effects for
vectors and structs on all architectures.
This kind of effects analysis was not possible in Tree-IL because
Tree-IL relied on logior-ing effects of subexpressions, whereas with
CPS we have no sub-expressions and we do flow analysis instead.
(effect-clobbers?): Replace effects-commute? with this inherently
directional and precise predicate.
* module/language/cps/cse.scm (compute-always-available-expressions):
(compute-equivalent-subexpressions): Adapt to effects analysis
change.
* module/language/cps/dce.scm (compute-live-code): Likewise.
* module/language/cps/cse.scm (compute-always-available-expressions):
Use constant? instead of zero?, to avoid punching through the effects
abstraction.
* module/language/cps/cse.scm (compute-available-expressions):
Simplify initialization.
(compute-equivalent-subexpressions): When synthesizing definitions,
use substed vars. Add synthetic definitions after processing an
expression, to take advantage of the substed vars.
* module/language/cps/effects-analysis.scm (synthesize-definition-effects!):
Fix a boneheaded thinko that caused all primcalls to be marked as
causing car, cdr, vector, struct, and box effects.
* module/language/cps/dce.scm (constant-type, lookup-type)
(default-type-checker, *primcall-type-checkers*)
(define-primcall-type-checker, define-simple-primcall-types)
(check-primcall-arg-types): Define a really lame type analysis that
can elide some expressions causing &type-check.
(compute-live-code): Wire up the type checker.
* module/language/cps/effects-analysis.scm (effects-clobber): New
helper.
(length): Only depend on &cdr.
(synthesize-definition-effects!): New interface.
* module/language/cps/cse.scm (compute-available-expressions): Don't
count out constructors here -- we'll do that below.
(compute-defs): Add a comment.
(compute-equivalent-subexpressions): Synthesize getter calls at
constructor/setter sites, so that (set-car! x y) can cause a
future (car x) to just reference y. The equiv-labels set now stores
the defined vars, so there is no need for the defs vector.
(cse, apply-cse): Adapt to compute-equivalent-subexpressions change.
