* libguile/vm-engine.c (allocate-struct, struct-ref, struct-set!): New
instructions, to complement their "immediate" variants.
* module/language/cps/compile-bytecode.scm (compile-fun):
* module/system/vm/assembler.scm (system): Wire up the new instructions.
* module/language/cps/types.scm (&all-types): Represent true and false
as separate bits, so that #f can be removed from types on true
branches. Adapt all users.
* module/language/cps/type-fold.scm (&scalar-types):
(fold-and-reduce): Adapt to boolean type representation change.
* module/language/cps/simplify.scm (redominate): Add micropass to
rewrite the scope tree to reflect the dominator tree. Will enable
better eta reduction.
* 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.
