* module/language/tree-il/fix-letrec.scm (compute-sccs): Instead of
depending on all previous complex bindings, we can just depend on the
most recent one. Decreases the graph size.
We were using list sets, which when you end up with thousands of
bindings in an SCC reaches the point where we are off the quadratic end
of the curve. Fix to use intsets and intmaps instead.
* module/language/tree-il/fix-letrec.scm (compute-ids): New function.
(compute-referenced-and-assigned): Rename from analyze-lexicals, and
compute intsets.
(make-compute-free-variables): Rename from free-variables, return a
procedure instead of a hash table, and use intsets. Use a global cache
to avoid quadratic behavior with regard to binding depth.
(compute-complex): Compute a global set of "complex" variables, as an
intset.
(compute-sccs): Use intsets and intmaps to compute the free-variable and
ordering edges.
(fix-scc, fix-term): Refactorings.
(reorder-bindings): Avoid a linear search.
(fix-letrec): Refactor.
* module/language/cps/specialize-numbers.scm (sigbits-ref): New helper.
(invert-graph*): New helper.
(compute-significant-bits): When visiting a term changes computed
needed-bits for one of its definitions, we need to revisit the variables
that contributed to its result (the uses), because they might need more
bits as well. Previously we were doing this by enqueueing predecessors
to the term, which worked if the uses were defined in predecessors, or
if all defining terms were already in the worklist, which is the case
without loops. But with loops, when revisiting a term, you could see
that it causes sigbits to change, enqueue its predecessors, but then the
predecessors don't change anything and the fixpoint stops before
reaching the definitions of the variables we need. So instead we
compute the use-def graph and enqueue defs directly.
* module/language/cps/specialize-numbers.scm (next-power-of-two): Use
integer-length. No change.
(compute-significant-bits): Fix the fixpoint computation, which was
failing to complete in some cases with loops.
* module/language/cps/specialize-numbers.scm (specialize-operations):
Accept any operand to logand/immediate, provided the result is a u64 in
the right range.
* module/language/cps/types.scm
(ulogand, ulogand/immediate, ulogsub, ulogior, ulogxor): Where we have
u64 inputs, there's no need to `restrict!`; the range will come from the
definition.
* module/language/tree-il/peval.scm (peval)
(inlinable-kwargs-bug-fixup): Before 3.0.10, the inlinable exports pass
was incorrectly serializing functions with keyword arguments. This was
fixed in 2c645571b3, but that meant that
3.0.10 compiling against 3.0.9 binaries could raise an exception at
compile-time; whoops. Add a workaround so that 3.0.9 binaries still
work.
Fixes https://issues.guix.gnu.org/72936.
* module/language/cps/intset.scm (make-intset-folder): intset-fold-right
on a transient intset would dispatch to left fold after making the
persistent set. Sadness!
As suggested in
<https://lists.gnu.org/archive/html/guile-devel/2023-06/msg00008.html>.
* module/language/wisp.scm (wisp-uuid): Remove.
(repr-quote, repr-unquote, repr-quasiquote, repr-unquote-splicing)
(repr-syntax, repr-unsyntax, repr-quasisyntax, repr-unsyntax-splicing):
Turn into uninterned symbols.
(line-continues?, chunk-ends-with-period,
line-code-replace-inline-colons): Adjust comparisons accordingly.
(wisp-replace-paren-quotation-repr)[pred]: New procedure.
Use it to compare against the various ‘repr-’ values.
(wisp-make-improper)[dot?]: New procedure.
Use it to compare against ‘repr-dot’.
* module/language/tree-il/optimize.scm (make-optimizer): If the keyword
argument #:dump-optimized-tree-il? is present, print the Tree-IL that
will be handed to the next compiler. Also re-enable #:verify-tree-il?.
* module/language/cps/effects-analysis.scm (compute-clobber-map):
Previously a whole-object read would not be clobbered by a specific
field write. This crops up for the &read introduced at the site of
`cons` for the synthetic car and cdr definitions. This error was there
before but didn't cause bugs before 3.0.10 because cons got eagerly
lowered to separate allocation and initialization instructions.
* module/language/tree-il/inlinable-exports.scm (inlinable-exp): Call
assq-ref in the right order. Embarrassing!
* module/language/tree-il/peval.scm (peval): Print the symbols so that
we can diagnose later failures.
* module/language/tree-il/peval.scm (peval): Handle all lambda inlining
the same, and extend with support for multiple clauses and keyword
arguments.
* test-suite/tests/peval.test ("case-lambda"): Enable kwarg inlining.
Can help reduce case-lambda* / lambda* at Tree-IL optimization-time.
* module/language/tree-il/demux-lambda.scm: New file.
* am/bootstrap.am (SOURCES): Add new file.
* module/language/tree-il/optimize.scm (make-optimizer):
* module/system/base/optimize.scm (available-optimizations): Enable
demux-lambda at level 2.
* module/language/tree-il/primitives.scm (*primitive-constructors*):
(append): Recognize append and reduce it to only the two-operand form.
* module/language/tree-il/peval.scm (peval): Add optimizations to
append.
* module/language/tree-il/peval.scm (peval): When visiting (values) in
anything other than an effect or values context,
residualize (values (values)), which will cause a run-time error.
* test-suite/tests/peval.test ("values"): Add test.
Allowing variables to hold an unbound value and requiring a check on
each load is suboptimal; the fixing letrec boolean check is better. So
other runtimes (hoot) might preclude unbound variables by construction.
Allow them to do so.
* module/language/cps/guile-vm.scm (target-has-unbound-boxes?): New
definition.
* module/language/tree-il/compile-cps.scm (target-has-unbound-boxes?):
(%box-ref): Only residualize an unbound check if the target has unbound
boxes.
* module/language/tree-il/peval.scm (peval): Introduce raise-type-error
for dynwind unwinder thunk check.
* module/language/tree-il/compile-cps.scm (raise-type-error):
* module/language/tree-il/compile-bytecode.scm (canonicalize): Handle
raise-type-error, as it can be in Tree-IL now.
This keeps things higher level, and is a step towards structured
exceptions in guile.
* module/language/cps/guile-vm/reify-primitives.scm (reify-primitives):
Turn raise-type-error, raise-range-error, and raise-arity-error into
variants of "throw". Lower raise-exception to a non-tail primcall.
* module/language/tree-il/compile-cps.scm: Instead of residualizing
"throw/value+data" throws, exceptions introduced by CPS lowering are
more structured: raise-type-error, raise-range-error, and
raise-arity-error. Also, lower raise-exception to an ordinary `$throw`
instead of eagerly producing the non-tail call to a $prim.
* module/language/cps/specialize-numbers.scm (logand/immediate): Define
a sigbits handler.
(specialize-operations): Require logand/immediate operand to be u64 to
lower to ulogand/immediate. Shouldn't be necessary but even if only u64
bits are used, negative fixnums will have the sign bit set, which trips
up further unboxed uses which error if the operand to `scm->u64` is
negative.
* module/language/cps/type-fold.scm (rem): Emit logand/immediate.
Also rework so that the symbol hash uses the low bits instead of high
bits. We can do this because, for the guile-vm target, now we compute
the full target hash.
* module/language/cps/guile-vm.scm (jenkins-lookup3-hashword2):
(target-symbol-hash, target-symbol-hash-bits): New exported functions..
* module/language/cps/switch.scm (optimize-branch-chain): Change to use
target-symbol-hash and target-symbol-hash-bits from the current
target-runtime.
* libguile/jit.c (compile_ulogand_immediate, compile_ulogand_immediate_slow)
* libguile/vm-engine.c (ulogand_immediate): New JIT and interpreter
support for ulogand/immediate.
* module/language/cps/guile-vm/lower-primcalls.scm (string-ref):
(vtable-vtable?):
(vtable-field-boxed?): Emit ulogand/immediate.
* module/language/cps/guile-vm/reify-primitives.scm (reify-primitives):
Remove logand/immediate. Only emit ulogand/immediate if the immediate
is a u8. Refactor mul/immediate.
* module/language/cps/specialize-numbers.scm (specialize-operations):
Produce ulogand/immediate if the result is a u64.
* module/language/cps/effects-analysis.scm:
* module/language/cps/types.scm (logand/immediate): Add effect and type
inference for logand/immediate, ulogand/immediate,
* module/language/cps/utils.scm (primcall-raw-representations):
ulogand/immediate makes a u64.
* module/language/tree-il/compile-cps.scm (convert): Generate
logand/immediate if possible.
* module/language/cps/compile-bytecode.scm (compile-function):
* module/system/vm/assembler.scm (system): Add ulogand/immediate
emitter.
* libguile/loader.h (SCM_OBJCODE_MINOR_VERSION): Bump.
* module/language/tree-il/peval.scm (peval): If a primcall is
effect-free, don't require that its args are too: just revisit args as a
sequence in effect context.
* module/language/tree-il/effects.scm (effect-free-primcall?): New
exported function.
* module/language/tree-il/peval.scm (peval): fix-letrec can residualize
useless primcalls, when a let or letrec-bound var is unused. Fix to
elide these.
* module/language/tree-il/effects.scm (add-primcall-effect-analyzer!):
New facility.
* module/language/tree-il/effects.scm (make-effects-analyzer): If a
primcall's args cause no effects, call out to a user-provided
effect-free? primitive for a primcall. If true, the primcall will be
marked as depending on all effects but causing none; this will allow it
to be elided by letrectify or peval.
* module/language/tree-il/primitives.scm (*interesting-primitive-names*):
(*primitive-accessors*): Add string->utf8, utf8->string, and
string-utf8-length.
(primitive-module): New public function, moved here from (language
tree-il compile-bytecode).
* module/language/tree-il/compile-bytecode.scm: Use primitive-module
from (language tree-il primitives).
* module/language/tree-il/peval.scm (peval): A bugfix: load primitives
from their proper module. Allows bytevector primitives to fold.
* module/language/cps/guile-vm/reify-primitives.scm:
* module/language/cps/effects-analysis.scm:
* module/language/cps/types.scm
* module/language/tree-il/primitives.scm:
* module/language/tree-il/cps-primitives.scm:
* module/language/tree-il/effects.scm (make-effects-analyzer):
Add string->utf8, utf8->string, and string-utf8-length.
* module/language/tree-il/compile-cps.scm (string->utf8)
(string-utf8-length, utf8->string): New custom lowerers, including type
checks and an unboxed result for string-utf8-length.
* module/system/vm/assembler.scm:
* libguile/intrinsics.h:
* libguile/intrinsics.c: Because string-utf8-length returns an unboxed
value, we need an intrinsic for it; go ahead and add an intrinsic for
string->utf8 and utf8->string too, as we will likely be able to use
these in the future.
* module/language/tree-il/primitives.scm (*interesting-primitive-names*):
(*effect-free-primitives*): Recognize keyword->symbol, symbol->keyword.
* module/language/tree-il/cps-primitives.scm: Plumb through to CPS.
(keyword->symbol):
* module/language/cps/effects-analysis.scm: New prims have no effect.
Fix effects for string->symbol.
(annotation->memory-kind): Add keywords.
* module/language/cps/guile-vm/lower-primcalls.scm (keyword->symbol):
Lower to scm-ref/immediate.
* module/language/cps/types.scm (annotation->type): Add case for
keywords.
* module/language/tree-il/compile-cps.scm: Add converters for new prims,
with type guards.
These numeric predicates now have CPS branching primcalls, which allows
type inference and folding to reduce them to less-strong instructions.
* module/language/cps/effects-analysis.scm (heap-numbers-equal?): Put
all the number predicates together. None have type checks.
* module/language/cps/guile-vm/lower-primcalls.scm
(define-branching-primcall-alias): New helper.
(complex?): Same as number?.
* module/language/cps/guile-vm/lower-primcalls.scm (real?)
(rational?, integer?, exact-integer?, exact?, inexact?): Define
lowerers.
* module/language/cps/type-fold.scm (number?, complex?, real?)
(rational?, integer?, exact-integer?, exact?, inexact?): Add folders and
reducers for all of these.
* module/language/cps/type.scm (number?, complex?, real?)
(rational?, integer?, exact-integer?, exact?, inexact?): Add type
inference for these.
* module/language/tree-il/compile-cps.scm (convert): Add number? checks
before exact? and inexact?. Remove the eager lowering of
exact-integer?; instead rely on folders.
* module/language/tree-il/cps-primitives.scm (number?, complex?)
(real?, rational?, integer?, exact-integer?, exact?, inexact?): Add
primitive decls. Define as "number-type-predicates?", meaning they need
a number? guard.
* module/language/cps/effects-analysis.scm: Mark more predicates as
effect-free. Sort the list.
* module/language/cps/guile-vm/lower-primcalls.scm (procedure?): Reify a
call to a primitive. Sadly we can't elide the $kreceive, as even though
we know that it's single-valued, $call can't continue to $kargs (has to
be $callk). Perhaps is worth relaxing in the future.
* module/language/cps/type-fold.scm: Define a number of additional
folders for disjoint types.
(procedure?): Define a folder for &procedure. Has to include structs,
though.
* module/language/cps/types.scm: Same as for type-fold.scm.
* module/language/tree-il/cps-primitives.scm: Lower procedure? primcalls
to CPS.
* module/language/cps/utils.scm (primcall-raw-representations): Add
sadd, ssub, etc; these are lowered to uadd, usub, etc later for the
guile-vm target, but it is still useful to record their reprs for
pre-lowering analysis.
Recognize `raise-exception` in the same way we recognize `throw`, though
it is a bit less optimized and the boot story is not as complicated.
* doc/ref/vm.texi (Non-Local Control Flow Instructions):
* libguile/jit.c (compile_unreachable):
(compile_unreachable_slow):
* libguile/vm-engine.c (VM_NAME):
* module/language/cps/compile-bytecode.scm (compile-function):
* module/system/vm/assembler.scm (emit-unreachable): Add new
"unreachable" instruction, inserted after a call to non-continuable
`raise-exception`.
* module/language/tree-il/compile-cps.scm (raise-exception):
* module/language/tree-il/primitives.scm
(*interesting-primitive-names*): Recognize raise-exception, and if it is
called with just one argument, prune that branch of the control-flow
graph.
