bpt/guile.git
10 years agoprint: Fix printing of weak vectors.
Mark H Weaver [Fri, 25 Apr 2014 05:39:59 +0000 (01:39 -0400)]
print: Fix printing of weak vectors.

* libguile/print.c (print_vector_or_weak_vector): New static function.
  (iprin1): Use 'print_vector_or_weak_vector' in the vector and weak
  vector cases.

10 years agobuild: Clean up rules for meta/{guild,guile-config}.
Ludovic Courtès [Thu, 24 Apr 2014 21:46:20 +0000 (23:46 +0200)]
build: Clean up rules for meta/{guild,guile-config}.

* meta/Makefile.am (guile-config, guild): Avoid useless 'cat'
  invocation'.  Quote all occurrences of '@'.
  Suggested by Eric Blake <eblake@redhat.com>.

10 years agobuild: Honor program name transformation in 'guild'.
Ludovic Courtès [Thu, 24 Apr 2014 09:16:21 +0000 (11:16 +0200)]
build: Honor program name transformation in 'guild'.

* configure.ac: Remove erroneous 'guile_program_name' substitute; see
  <https://lists.gnu.org/archive/html/autoconf/2014-04/msg00010.html>
  for details.  Remove 'meta/guild' config file.
* meta/Makefile.am (guild): New target.
  (CLEANFILES): Add 'guild'.
* meta/guild.in: Remove 'prefix' and 'exec_prefix' definitions; use
  @installed_guile@.

10 years agoSet $GC_MARKERS to 1 when libgc 7.4.0 is used.
Ludovic Courtès [Wed, 23 Apr 2014 16:12:26 +0000 (18:12 +0200)]
Set $GC_MARKERS to 1 when libgc 7.4.0 is used.

* libguile/gc.c (scm_storage_prehistory) [GC_VERSION == 7.4.0]: Add
  'setenv' call.

10 years agoAdd make-vector opcode
Andy Wingo [Mon, 21 Apr 2014 19:48:39 +0000 (21:48 +0200)]
Add make-vector opcode

* libguile/vm-engine.c (make-vector): New opcode.
* module/language/cps/compile-bytecode.scm (compile-fun):
* module/system/vm/assembler.scm (system): Support the new opcode.
  (*bytecode-minor-version*): Bump.

* libguile/_scm.h (SCM_OBJCODE_MINOR_VERSION): Bump.

* test-suite/tests/compiler.test ("limits"): Add vector test.

10 years agoOperations on 8-bit and 12-bit operands shuffle args into range
Andy Wingo [Mon, 21 Apr 2014 17:28:06 +0000 (19:28 +0200)]
Operations on 8-bit and 12-bit operands shuffle args into range

* module/language/cps/slot-allocation.scm (allocate-slots): Avoid
  allocating locals in the range [253,255].

* module/system/vm/assembler.scm: List exports explicitly.  For
  operations with limited-range operands, export wrapper assemblers that
  handle shuffling their operands into and out of their range.
  (define-assembler): Get rid of enclosing begin.
  (shuffling-assembler, define-shuffling-assembler): New helpers to
  define shuffling wrapper assemblers.
  (emit-mov*, emit-receive*): New functions.
  (shuffle-up-args): New helper.
  (standard-prelude, opt-prelude, kw-prelude): Call shuffle-up-args
  after finishing.

* test-suite/tests/compiler.test ("limits"): Add test cases.

10 years agoMore expansion-time-only definitions in assembler.scm
Andy Wingo [Mon, 21 Apr 2014 10:13:54 +0000 (12:13 +0200)]
More expansion-time-only definitions in assembler.scm

* module/system/vm/assembler.scm (define-inline): Change so that the
  defined macro is only defined at expansion-time.
  (u32-ref, u32-set!, s32-ref, s32-set!, pack-arity-flags): Use
  define-inline.
  (pack-flags, assert-match, *block-size*, id-append, assembler)
  (define-assembler, visit-opcodes, define-macro-assembler): Wrap in
  eval-when expand.

10 years agodefine-inline in assembler.scm
Andy Wingo [Mon, 21 Apr 2014 10:01:46 +0000 (12:01 +0200)]
define-inline in assembler.scm

* module/system/vm/assembler.scm (define-inline): New local helper.
  Update local users of define-inlinable to use it.

10 years agoVM opcodes only have <24-bit slot operands in the first word
Andy Wingo [Sun, 20 Apr 2014 09:52:14 +0000 (11:52 +0200)]
VM opcodes only have <24-bit slot operands in the first word

* libguile/vm-engine.c (make-array): Change to only have
  restricted-width operands in the first word.  This instruction is
  currently unused, however.

* module/system/vm/assembler.scm (assembler):
* module/system/vm/disassembler.scm (disassembler): Disallow
  restricted-width operands in tail words.

10 years agoAdd interface to disable automatic finalization
Andy Wingo [Thu, 17 Apr 2014 13:29:13 +0000 (15:29 +0200)]
Add interface to disable automatic finalization

* libguile/finalizers.h:
* libguile/finalizers.c (run_finalizers_async_thunk)
  (finalization_thread_proc): Call the new scm_run_finalizers helper.
  (scm_i_finalizer_pre_fork): Only spawn the thread if automatic
  finalization is enabled.
  (scm_set_automatic_finalization_enabled, scm_run_finalizers): New
  functions.
  (scm_init_finalizers, scm_init_finalizer_thread): Only set a finalizer
  notifier if automatic finalization is enabled.

* doc/ref/libguile-smobs.texi (Garbage Collecting Smobs): Add discussion
  of concurrency.

* doc/ref/api-smobs.texi (Smobs): Document new functions.

10 years agoSlot allocation can re-use closure and argument slots
Andy Wingo [Wed, 16 Apr 2014 17:21:50 +0000 (19:21 +0200)]
Slot allocation can re-use closure and argument slots

* module/language/cps/slot-allocation.scm (allocate-slots): Allow slot
  allocation to re-use the closure and argument slots.

10 years agoRemove SCM_FRAME_PROGRAM
Andy Wingo [Wed, 16 Apr 2014 17:20:42 +0000 (19:20 +0200)]
Remove SCM_FRAME_PROGRAM

* libguile/frames.h: Remove SCM_FRAME_PROGRAM, now unused.

10 years agoVM robustness for optimized closures
Andy Wingo [Wed, 16 Apr 2014 17:20:23 +0000 (19:20 +0200)]
VM robustness for optimized closures

* libguile/vm.c (vm_error_unbound, vm_error_unbound_fluid): Remove proc
  argument.  The value in slot 0 is not necessarily the procedure being
  applied, after the prelude is done.

* libguile/vm-engine.c (vm_engine): Use LOCAL_REF (0) instead of
  SCM_FRAME_PROGRAM, and adapt to above changes.

10 years agoFix rewinding continuations when outermost frame has zero locals
Andy Wingo [Wed, 16 Apr 2014 17:18:29 +0000 (19:18 +0200)]
Fix rewinding continuations when outermost frame has zero locals

* libguile/vm.c (vm_reinstate_partial_continuation_inner): Fix boundary
  condition when the outermost frame has zero locals.

10 years agoframe-previous, frame-procedure robustness
Andy Wingo [Wed, 16 Apr 2014 17:17:38 +0000 (19:17 +0200)]
frame-previous, frame-procedure robustness

* libguile/frames.c (scm_c_frame_closure): Don't use SCM_FRAME_PROGRAM,
  as we don't know if the frame actually has any locals.
  (scm_c_frame_previous): More robustly detect end-of-stack.  Allows
  scm_c_frame_previous to work on partial continuations.

10 years agomake-stack works on delimited continuations
Andy Wingo [Wed, 16 Apr 2014 17:16:10 +0000 (19:16 +0200)]
make-stack works on delimited continuations

* libguile/stacks.c (scm_make_stack, scm_stack_id):
* libguile/vm.c (scm_i_vm_cont_to_frame): Allow delimited continuations
  as the argument to make-stack.

10 years agoMore robust coverage tests
Andy Wingo [Wed, 16 Apr 2014 17:13:49 +0000 (19:13 +0200)]
More robust coverage tests

* test-suite/tests/coverage.test ("line-execution-counts"): Allow zero
  or one count on the loop head.

10 years agoChange some make-stack tests to use frame-call-representation
Andy Wingo [Wed, 16 Apr 2014 17:13:09 +0000 (19:13 +0200)]
Change some make-stack tests to use frame-call-representation

* test-suite/tests/eval.test ("stacks"): Use frame-call-representation.

10 years agoFix statprof for optimizations
Andy Wingo [Wed, 16 Apr 2014 17:12:43 +0000 (19:12 +0200)]
Fix statprof for optimizations

* module/statprof.scm (profile-signal-handler): Bind in a letrec.
  Otherwise the compiler may see the closure slot as dead, and the inner
  stack cut won't work.

10 years agoFix a frame-call-representation bug
Andy Wingo [Wed, 16 Apr 2014 12:54:46 +0000 (14:54 +0200)]
Fix a frame-call-representation bug

* module/system/vm/frame.scm (frame-call-representation): Fix logic for
  displaying names of non-procedures and procedures out of arities.

10 years agoBetter backtraces from C, especially for optimized closures
Andy Wingo [Wed, 16 Apr 2014 12:33:20 +0000 (14:33 +0200)]
Better backtraces from C, especially for optimized closures

* libguile/frames.h:
* libguile/frames.c (scm_frame_call_representation): New interface;
  dispatches to Scheme.

* libguile/backtrace.c (display_application): Use
  scm_frame_call_representation.  This should be monotonically better,
  given that scm_frame_arguments (which was previously called) also
  dispatched to Scheme and actually ended up calling
  frame-call-representation.

10 years agoframe-call-representation checks available-bindings, as appropriate
Andy Wingo [Wed, 16 Apr 2014 12:19:18 +0000 (14:19 +0200)]
frame-call-representation checks available-bindings, as appropriate

* module/system/vm/frame.scm (available-bindings): Map indexes in such a
  way that the first argument is index 1.
  (frame-call-representation): Update to search the bindings for live
  bindings.

10 years agoUpdate frame-bindings interface
Andy Wingo [Wed, 16 Apr 2014 11:58:17 +0000 (13:58 +0200)]
Update frame-bindings interface

* module/system/repl/debug.scm (print-locals): Update to work with new
  interface.
  (frame->module): Update.  Still doesn't work due to lack of
  `program-module', though.

* module/system/vm/program.scm (make-binding, binding:name)
  (binding:definition-offset, program-arity-bindings-for-ip): Remove
  these.

* module/system/vm/frame.scm (<binding>): New type.
  (available-bindings): Return a list of <binding> instances.
  (frame-lookup-binding, frame-binding-set!, frame-binding-ref):
  (frame-environment, frame-object-name): Adapt.

10 years agoImplement frame-bindings
Andy Wingo [Wed, 16 Apr 2014 10:59:45 +0000 (12:59 +0200)]
Implement frame-bindings

* module/system/vm/frame.scm (parse-code, compute-predecessors):
  (compute-genv, compute-defs-by-slot, compute-killv, available-bindings):
  (frame-bindings): Add a bunch of hairy code to compute the set of
  bindings that are live in a frame.

10 years agoAdd arity-code
Andy Wingo [Wed, 16 Apr 2014 10:58:35 +0000 (12:58 +0200)]
Add arity-code

* module/system/vm/debug.scm (arity-code): New interface.

10 years agoAdd parsing interfaces to the disassembler
Andy Wingo [Wed, 16 Apr 2014 10:58:20 +0000 (12:58 +0200)]
Add parsing interfaces to the disassembler

* module/system/vm/disassembler.scm (instruction-length):
  (instruction-has-fallthrough?, instruction-relative-jump-targets):
  (instruction-slot-clobbers): New interfaces; to be used when
  determining the bindings available at a given point of a procedure.

10 years agoFix up some opcode metadata
Andy Wingo [Wed, 16 Apr 2014 10:51:34 +0000 (12:51 +0200)]
Fix up some opcode metadata

* libguile/vm-engine.c (make-long-immediate, static-ref): Mark as "dst"
  instructions.

10 years agoAdd ability to query local definitions for a procedure
Andy Wingo [Tue, 15 Apr 2014 20:24:48 +0000 (22:24 +0200)]
Add ability to query local definitions for a procedure

* module/system/vm/debug.scm (arity-definitions): New interface.

* module/system/vm/program.scm (make-binding, binding:boxed?)
  (binding:index, binding:start, binding:end): Remove.
  (binding:definition-offset, binding:slot): Add.
  (program-arity-bindings-for-ip): Rename from program-bindings-for-ip,
  as it gives all definitions in an arity.  The user will have to do
  data-flow analysis to recover the set of variables that are actually
  available at any given point.
  (arity->arguments-alist): Remove crufty code.

10 years agoFix rtl tests
Andy Wingo [Tue, 15 Apr 2014 20:00:30 +0000 (22:00 +0200)]
Fix rtl tests

* module/system/vm/assembler.scm (write-arities): Add a diagnostic.

* test-suite/tests/rtl.test: Fix tests to emit "definition"
  instructions.

10 years agoFix frame-call-representation for primitive applications
Andy Wingo [Tue, 15 Apr 2014 19:47:46 +0000 (21:47 +0200)]
Fix frame-call-representation for primitive applications

* module/system/vm/frame.scm (frame-call-representation): Fix to work
  for primitives.

* test-suite/tests/eval.test ("stacks"): Update expected result for
  substring.

10 years agoAssembler residualizes local variable definition locations
Andy Wingo [Tue, 15 Apr 2014 18:25:16 +0000 (20:25 +0200)]
Assembler residualizes local variable definition locations

* module/system/vm/assembler.scm (write-arities): Serialize definition
  locations after names.
  (definition): Store definition as a byte offset.

10 years agoBump minor objcode version for recent changes
Andy Wingo [Tue, 15 Apr 2014 18:20:01 +0000 (20:20 +0200)]
Bump minor objcode version for recent changes

* libguile/_scm.h (SCM_OBJCODE_MINOR_VERSION):
* module/system/vm/assembler.scm (*bytecode-minor-version*): Bump.

10 years agoWrite all local variable names into the arities section
Andy Wingo [Tue, 15 Apr 2014 15:52:41 +0000 (17:52 +0200)]
Write all local variable names into the arities section

* module/system/vm/assembler.scm (put-uleb128, put-sleb128)
  (port-position): Lift out these helpers.
  (arity-header-len, write-arities, link-arities): Add "nlocals" to the
  arity headers.  Write names of all locals into the arities section,
  not just the arguments.  Write them as uleb128's instead of uint32's,
  to save space.

* module/system/vm/debug.scm (arity-header-len, arity-nlocals*)
  (arity-nlocals, arity-locals, arity-arguments-alist): Adapt to new
  encoding for arities.

10 years agoTweak arities debugging representation
Andy Wingo [Tue, 15 Apr 2014 13:27:19 +0000 (15:27 +0200)]
Tweak arities debugging representation

* module/system/vm/assembler.scm (meta-arities-size, write-arity-links):
* module/system/vm/debug.scm (arity-keyword-args)
  (arity-arguments-alist): Rewrite to put they keyword literals link
  first.  Unfortunately requires a recompile :/

10 years agoBeginnings of local variable information
Andy Wingo [Tue, 15 Apr 2014 12:02:25 +0000 (14:02 +0200)]
Beginnings of local variable information

* module/system/vm/assembler.scm (<arity>, begin-kw-arity, end-arity):
  (definition): Add definition macro-instruction.  Arrange to record
  variable definitions.

* module/language/cps/compile-bytecode.scm (compile-fun): Emit
  definition macro-instructions as appropriate.

10 years agoRemove needless label remapping in slot-allocation
Andy Wingo [Tue, 15 Apr 2014 10:25:26 +0000 (12:25 +0200)]
Remove needless label remapping in slot-allocation

* module/language/cps/slot-allocation.scm (dead-after-def?):
  (dead-after-use?, allocate-slots): Remove some needless remapping
  between label indexes in the CFA, the DFA, and their names.

10 years agoDFA datums don't rename their labels
Andy Wingo [Tue, 15 Apr 2014 10:16:41 +0000 (12:16 +0200)]
DFA datums don't rename their labels

* module/language/cps/dfg.scm (analyze-reverse-control-flow): Don't
  compute and return an order vector; it's not needed.
  ($dfa): Remove label renaming.  We can just rename labels before
  returning the DFA.
  (dfa-k-idx, dfa-k-sym, dfa-k-count): Adapt.
  (compute-live-variables): Adapt, and rename labels before returning.

10 years agoBetter backtraces for optimized closures
Andy Wingo [Tue, 15 Apr 2014 09:18:50 +0000 (11:18 +0200)]
Better backtraces for optimized closures

* module/system/vm/debug.scm (arity-keyword-args, find-program-arity):
  New exports.

* module/system/vm/frame.scm (frame-call-representation): Prefer to use
  the frame IP to get the procedure.

10 years agostatprof avoids mucking with VM trace levels when not counting calls
Andy Wingo [Mon, 14 Apr 2014 15:06:05 +0000 (17:06 +0200)]
statprof avoids mucking with VM trace levels when not counting calls

* module/statprof.scm (statprof-start, statprof-stop): Don't futz the vm
  trace level when we aren't counting calls.  With this change, statprof
  now imposes no overhead on the measured program.

10 years agoBetter state handling in statprof
Andy Wingo [Mon, 14 Apr 2014 14:54:51 +0000 (16:54 +0200)]
Better state handling in statprof

* module/statprof.scm (statprof-fold-call-data)
  (statprof-proc-call-data): Add optional state arg.
  (gcprof): Add optional port arg, and pass state arg explicitly.
  (statprof-display-anomalies, statprof-display)
  (statprof-call-data->stats): Pass state explicitly.

10 years agoOptimize make-stack
Andy Wingo [Mon, 14 Apr 2014 14:31:02 +0000 (16:31 +0200)]
Optimize make-stack

* libguile/continuations.h:
* libguile/continuations.c (scm_i_continuation_to_frame): Operate on
  low-level C structures instead of heap objects.

* libguile/frames.h:
* libguile/frames.c (frame_offset, frame_stack_base): Const args.
  (scm_c_frame_closure): New helper.
  (scm_frame_procedure): Use the new helper.

* libguile/stacks.c (stack_depth, narrow_stack, scm_make_stack): Rework
  to avoid allocating frames as we traverse the stack, and to avoid an
  n**2 case where there are outer cuts.

10 years agoscm_c_make_frame takes struct scm_frame as arg
Andy Wingo [Mon, 14 Apr 2014 13:14:26 +0000 (15:14 +0200)]
scm_c_make_frame takes struct scm_frame as arg

* libguile/frames.h:
* libguile/frames.c (scm_c_make_frame): Adapt to take a const struct
  scm_frame as the argument.  Adapt callers.

* libguile/continuations.c:
* libguile/stacks.c: Adapt callers.

10 years agoRefactor to frames code
Andy Wingo [Mon, 14 Apr 2014 12:54:14 +0000 (14:54 +0200)]
Refactor to frames code

* libguile/frames.h:
* libguile/frames.c (scm_c_frame_previous): New internal helper.
  (scm_frame_previous): Use the helper.
  (RELOC): Take kind and low-level frame args separately.  Adapt
  callers.
  (frame_stack_base, frame_offset): New helpers.
  (scm_i_frame_offset, scm_i_frame_stack_base): Use low-level helpers.

10 years agoOptimize make-global-cont-folder
Andy Wingo [Mon, 14 Apr 2014 11:53:35 +0000 (13:53 +0200)]
Optimize make-global-cont-folder

* module/language/cps.scm (make-global-cont-folder): Inline the
  fold-values, as peval doesn't do so.  Allows closure conversion to
  avoid any closure creation.

10 years agoImprove disassembly for optimized closures
Andy Wingo [Sun, 13 Apr 2014 12:40:22 +0000 (14:40 +0200)]
Improve disassembly for optimized closures

* module/system/vm/disassembler.scm (code-annotation): Add call-label
  and tail-call-label cases.
  (disassemble-addr): With call-label we can see sets of mutually
  recursive functions, so keep a global "visited?" set.

10 years agoRemove debugging code in closure-conversion
Andy Wingo [Sun, 13 Apr 2014 12:26:03 +0000 (14:26 +0200)]
Remove debugging code in closure-conversion

* module/language/cps/closure-conversion.scm (prune-free-vars): Remove
  pk.

10 years agoEval has no more free variables
Andy Wingo [Sun, 13 Apr 2014 12:22:22 +0000 (14:22 +0200)]
Eval has no more free variables

* module/ice-9/eval.scm (primitive-eval): Expand out the call to
  make-general-closure, so that make-general-closure becomes
  well-known.  Now eval has no more free variables!

10 years agoClosure conversion eliminates self-references introduced by fixpoint
Andy Wingo [Sun, 13 Apr 2014 12:21:25 +0000 (14:21 +0200)]
Closure conversion eliminates self-references introduced by fixpoint

* module/language/cps/closure-conversion.scm (analyze-closures): Build a
  bound-vars set as well, to resolve introduced self-references.
  (prune-free-vars, convert-one): Arrange to eliminate self-references.

10 years agoRefactor to closure-conversion
Andy Wingo [Sun, 13 Apr 2014 11:52:56 +0000 (13:52 +0200)]
Refactor to closure-conversion

* module/language/cps/closure-conversion.scm (convert-one): Refactor to
  pull in helpers locally, as they will need more state.

10 years agoAvoid consing an unbound-arg marker in the evaluator
Andy Wingo [Sun, 13 Apr 2014 10:21:36 +0000 (12:21 +0200)]
Avoid consing an unbound-arg marker in the evaluator

* module/ice-9/eval.scm (primitive-eval): Turns out we don't need to
  cons to make the unbound-arg marker.

10 years agoOptimize closures with one free variable
Andy Wingo [Sun, 13 Apr 2014 09:47:17 +0000 (11:47 +0200)]
Optimize closures with one free variable

* module/language/cps/closure-conversion.scm (convert-free-var)
  (allocate-closure, init-closure, prune-free-vars, convert-one)
  (convert-closures): Optimize closures with one free variable.

10 years agoWell-known closures represented using pairs or vectors
Andy Wingo [Sat, 12 Apr 2014 21:31:08 +0000 (23:31 +0200)]
Well-known closures represented using pairs or vectors

* module/language/cps/closure-conversion.scm (convert-free-var):
  (convert-free-vars): Take self-known? param, to do the right thing for
  well-known closures.
  (allocate-closure): New helper.  Well-known closures are represented
  using pairs or vectors.
  (init-closure): Adapt tpo DTRT for well-known closures.
  (prune-free-vars): Move up.
  (convert-one): Adapt to new well-known closure representation.

10 years agoUpdate verify-cps
Andy Wingo [Sat, 12 Apr 2014 20:42:23 +0000 (22:42 +0200)]
Update verify-cps

* module/language/cps/verify.scm (verify-cps): Update for recent CPS
  changes.

10 years agoAvoid creating closures with no free variables
Andy Wingo [Sat, 12 Apr 2014 17:46:23 +0000 (19:46 +0200)]
Avoid creating closures with no free variables

* module/language/cps/closure-conversion.scm (init-closure): Return just
  one value.
  (analyze-closures): Rewrite the well-known set to key off the label
  instead of the closure identifiers before returning.
  (convert-one): Avoid creating closure objects at runtime or load-time
  when "instantiating" or calling well-known closures with no free
  variables.
  (prune-free-vars): New pass.
  (convert-closures): Adapt.

10 years agoHard-wire calls to known procedures
Andy Wingo [Sat, 12 Apr 2014 14:12:33 +0000 (16:12 +0200)]
Hard-wire calls to known procedures

* module/language/cps/closure-conversion.scm (analyze-closures):
  (convert-one, convert-closures): Hard-wire calls to known procedures
  by transforming $call to $callk.

10 years agoclosure conversion computes well-known functions
Andy Wingo [Sat, 12 Apr 2014 13:53:58 +0000 (15:53 +0200)]
closure conversion computes well-known functions

* module/language/cps/closure-conversion.scm (analyze-closures)
  (convert-closures, convert-one): Adapt to compute well-known
  functions.  We don't yet produce $callk though.

10 years agoFirst-order CPS has $program and $closure forms
Andy Wingo [Sat, 12 Apr 2014 09:52:38 +0000 (11:52 +0200)]
First-order CPS has $program and $closure forms

* module/language/cps.scm ($closure, $program): New CPS types, part of
  low-level (first-order) CPS.
  (build-cps-exp, build-cps-term, parse-cps, unparse-cps)
  (compute-max-label-and-var): Update for new CPS types.

* module/language/cps/closure-conversion.scm: Rewrite to produce a
  $program with $closures, and no $funs.

* module/language/cps/reify-primitives.scm:
* module/language/cps/compile-bytecode.scm (compile-fun):
  (compile-bytecode): Adapt to new first-order format.

* module/language/cps/dfg.scm (compute-dfg): Add $closure case.

* module/language/cps/renumber.scm (renumber): Allow this pass to work
  on either format.

* module/language/cps/slot-allocation.scm (allocate-slots): Add $closure
  case.

10 years agoSeparate make-cont-folder into global and local variants
Andy Wingo [Fri, 11 Apr 2014 16:01:23 +0000 (18:01 +0200)]
Separate make-cont-folder into global and local variants

* module/language/cps.scm (make-global-cont-folder)
  (make-local-cont-folder): Separate this macro in two.  It's hot and
  the difference can be important for perf.

* module/language/cps/dfg.scm (compute-label-and-var-ranges):
* module/language/cps/cse.scm (compute-label-and-var-ranges):
* module/language/cps/dce.scm (compute-live-code): Adapt.

10 years agoRoot higher-order CPS term is always $kfun $cont
Andy Wingo [Fri, 11 Apr 2014 12:01:27 +0000 (14:01 +0200)]
Root higher-order CPS term is always $kfun $cont

* module/language/cps/arities.scm:
* module/language/cps/closure-conversion.scm:
* module/language/cps/compile-bytecode.scm:
* module/language/cps/constructors.scm:
* module/language/cps/contification.scm:
* module/language/cps/cse.scm:
* module/language/cps/dce.scm:
* module/language/cps/elide-values.scm:
* module/language/cps/prune-bailouts.scm:
* module/language/cps/prune-top-level-scopes.scm:
* module/language/cps/renumber.scm:
* module/language/cps/self-references.scm:
* module/language/cps/simplify.scm:
* module/language/cps/specialize-primcalls.scm:
* module/language/tree-il/compile-cps.scm: Adapt to produce and consume
  raw $kfun $cont instances.

* .dir-locals.el: Update $letrec indentation.

10 years agoClosure conversion, reify-primitives use $kfun $cont
Andy Wingo [Fri, 11 Apr 2014 09:51:34 +0000 (11:51 +0200)]
Closure conversion, reify-primitives use $kfun $cont

* module/language/cps/closure-conversion.scm: Produce a $kfun $cont.
* module/language/cps/reify-primitives.scm: Produce and consume $kfun
  $cont.
* module/language/cps/compile-bytecode.scm: Adapt.

10 years agoPreparation for compile-bytecode to work on $kfun $conts
Andy Wingo [Fri, 11 Apr 2014 09:34:50 +0000 (11:34 +0200)]
Preparation for compile-bytecode to work on $kfun $conts

* module/language/cps/compile-bytecode.scm (compile-fun): Change to take
  a $kfun $cont instead of a $fun.
  (visit-funs): Change likewise, and call the proc on $kfun $cont's, not
  $fun's.
  (compile-bytecode): Adapt.

* module/language/cps/dfg.scm (analyze-reverse-control-flow): Adapt to
  expect a $kfun $cont.

10 years agocompute-dfg takes a $kfun $cont, not a $fun
Andy Wingo [Fri, 11 Apr 2014 09:22:06 +0000 (11:22 +0200)]
compute-dfg takes a $kfun $cont, not a $fun

* module/language/cps/dfg.scm (compute-dfg): Take a $kfun $cont instead
  of a $fun.

* module/language/cps/arities.scm:
* module/language/cps/compile-bytecode.scm:
* module/language/cps/contification.scm:
* module/language/cps/cse.scm:
* module/language/cps/dce.scm:
* module/language/cps/simplify.scm:
* module/language/cps/specialize-primcalls.scm: Adapt callers.

10 years agowith-fresh-name-state takes a cont, not a $fun
Andy Wingo [Fri, 11 Apr 2014 08:21:04 +0000 (10:21 +0200)]
with-fresh-name-state takes a cont, not a $fun

* module/language/cps.scm (with-fresh-name-state): Take a cont instead
  of a fun.

* module/language/cps/closure-conversion.scm:
* module/language/cps/constructors.scm:
* module/language/cps/elide-values.scm:
* module/language/cps/prune-bailouts.scm:
* module/language/cps/reify-primitives.scm: Adapt.

10 years agoFunction defined by make-cont-folder takes a cont, not a $fun
Andy Wingo [Fri, 11 Apr 2014 08:12:37 +0000 (10:12 +0200)]
Function defined by make-cont-folder takes a cont, not a $fun

* module/language/cps.scm (make-cont-folder): Take a cont instead of a
  $fun.
  (with-fresh-name-state): Adapt.

* module/language/cps/cse.scm (compute-label-and-var-ranges):
* module/language/cps/dce.scm (compute-live-code):
* module/language/cps/dfg.scm (compute-dfg):
* module/language/cps/elide-values.scm (elide-values):
* module/language/cps/reify-primitives.scm (reify-primitives):
* module/language/cps/renumber.scm (compute-new-labels-and-vars):
  (renumber): Adapt.

10 years agoRename $kentry to $kfun
Andy Wingo [Thu, 10 Apr 2014 10:11:35 +0000 (12:11 +0200)]
Rename $kentry to $kfun

* module/language/cps.scm ($kfun): Rename from $kentry.

* module/language/cps/arities.scm:
* module/language/cps/closure-conversion.scm:
* module/language/cps/compile-bytecode.scm:
* module/language/cps/constructors.scm:
* module/language/cps/contification.scm:
* module/language/cps/cse.scm:
* module/language/cps/dce.scm:
* module/language/cps/dfg.scm:
* module/language/cps/effects-analysis.scm:
* module/language/cps/elide-values.scm:
* module/language/cps/prune-bailouts.scm:
* module/language/cps/prune-top-level-scopes.scm:
* module/language/cps/reify-primitives.scm:
* module/language/cps/renumber.scm:
* module/language/cps/self-references.scm:
* module/language/cps/simplify.scm:
* module/language/cps/slot-allocation.scm:
* module/language/cps/specialize-primcalls.scm:
* module/language/cps/verify.scm:
* module/language/tree-il/compile-cps.scm: Adapt users.

10 years agosrc and meta are fields of $kentry, not $fun
Andy Wingo [Thu, 10 Apr 2014 08:50:17 +0000 (10:50 +0200)]
src and meta are fields of $kentry, not $fun

* module/language/cps.scm ($kentry, $fun): Attach "src" and "meta" on
  the $kentry, not the $fun.  This prepares us for $callk to $kentry
  continuations that have no corresponding $fun.

* module/language/cps/arities.scm:
* module/language/cps/closure-conversion.scm:
* module/language/cps/compile-bytecode.scm:
* module/language/cps/constructors.scm:
* module/language/cps/contification.scm:
* module/language/cps/cse.scm:
* module/language/cps/dce.scm:
* module/language/cps/dfg.scm:
* module/language/cps/elide-values.scm:
* module/language/cps/prune-bailouts.scm:
* module/language/cps/prune-top-level-scopes.scm:
* module/language/cps/reify-primitives.scm:
* module/language/cps/renumber.scm:
* module/language/cps/self-references.scm:
* module/language/cps/simplify.scm:
* module/language/cps/slot-allocation.scm:
* module/language/cps/specialize-primcalls.scm:
* module/language/cps/verify.scm:
* module/language/tree-il/compile-cps.scm: Adapt.

10 years agoRemove tests for old Tree-IL CSE module
Andy Wingo [Thu, 10 Apr 2014 07:25:38 +0000 (09:25 +0200)]
Remove tests for old Tree-IL CSE module

* test-suite/tests/cse.test: Remove.
* test-suite/Makefile.am:

10 years agoRemove obsolete comment in compile-bytecode.scm
Andy Wingo [Tue, 8 Apr 2014 12:10:36 +0000 (14:10 +0200)]
Remove obsolete comment in compile-bytecode.scm

* module/language/cps/compile-bytecode.scm (optimize): Remove an
  obsolete comment.

10 years agoNew pass to avoid free variable creation for self-recursion
Andy Wingo [Tue, 8 Apr 2014 19:41:42 +0000 (21:41 +0200)]
New pass to avoid free variable creation for self-recursion

* module/language/cps/self-references.scm: New pass, avoids the need for
  self-recursion to allocate free variables.

* module/Makefile.am:
* module/language/cps/compile-bytecode.scm: Wire up the new pass.

10 years agoCompile some standalone tests to bytecode
Andy Wingo [Tue, 8 Apr 2014 08:06:40 +0000 (10:06 +0200)]
Compile some standalone tests to bytecode

* test-suite/standalone/test-out-of-memory:
* test-suite/standalone/test-stack-overflow: Compile these files before
  running them.  That way, recursion can check the stack-overflow
  mechanism instead of the memory allocation mechanism.  We compile
  beforehand as a prepass so as not to impose an rlimit on a Guile that
  previously ran auto-compilation.

10 years agoRemove old Tree-IL CSE pass
Andy Wingo [Sun, 6 Apr 2014 09:08:05 +0000 (11:08 +0200)]
Remove old Tree-IL CSE pass

* module/language/tree-il/cse.scm: Delete.

* module/language/tree-il/optimize.scm: Remove use of Tree-IL CSE.

* module/Makefile.am: Remove language/tree-il/cse.scm.

* module/language/cps/compile-bytecode.scm: Rename CSE keyword to
  #:cse?.

10 years agoFlow-sensitive analysis of truth values
Andy Wingo [Sat, 5 Apr 2014 19:08:09 +0000 (21:08 +0200)]
Flow-sensitive analysis of truth values

* module/language/cps/cse.scm (compute-truthy-expressions):
  (compute-equivalent-subexpressions, apply-cse): Arrange to infer
  truthiness of expressions, and use that information to elide redundant
  tests.

10 years agoAdd effects for specialized primitives
Andy Wingo [Sat, 5 Apr 2014 19:06:35 +0000 (21:06 +0200)]
Add effects for specialized primitives

* module/language/cps/effects-analysis.scm (make-vector)
  (make-vector/immediate, vector-ref/immediate, vector-set!/immediate)
  (struct-ref/immediate, struct-set!/immediate): Add effects.

10 years agoMinor cleanup/optimization in CSE
Andy Wingo [Sat, 5 Apr 2014 12:38:37 +0000 (14:38 +0200)]
Minor cleanup/optimization in CSE

* module/language/cps/cse.scm (compute-available-expressions): Remove
  needless for-each definition.
  (compute-equivalent-subexpressions): Optimize for-each/2.

10 years agoPrune bailouts after contification
Andy Wingo [Sat, 5 Apr 2014 10:16:34 +0000 (12:16 +0200)]
Prune bailouts after contification

* module/language/cps/compile-bytecode.scm (optimize): Prune bailouts
  after contifying, so that we return to the tail of the contified
  function.

10 years agoMatch and srfi-9 expose their bailouts to the CSE pass
Andy Wingo [Sat, 5 Apr 2014 09:56:44 +0000 (11:56 +0200)]
Match and srfi-9 expose their bailouts to the CSE pass

* module/ice-9/match.upstream.scm (match-next): Inline a call to
  "error", so the new CSE pass will see this case as a bailout.

* module/srfi/srfi-9.scm (throw-bad-struct): Reimplement as a syntax
  rule, so that the CSE pass sees the "throw" call.

10 years agoRemove &bailout; replace uses of &unknown-effects with &all-effects
Andy Wingo [Sat, 5 Apr 2014 09:21:33 +0000 (11:21 +0200)]
Remove &bailout; replace uses of &unknown-effects with &all-effects

* module/language/cps/effects-analysis.scm (&bailout): Remove effect.
  (&unknown-effects): Remove.  Replace uses with &all-effects.
* module/language/cps/cse.scm:

10 years agoRemove parts of CSE that deal with bailout
Andy Wingo [Sat, 5 Apr 2014 09:18:20 +0000 (11:18 +0200)]
Remove parts of CSE that deal with bailout

* module/language/cps/cse.scm (compute-available-expressions, cse):
  (compute-idoms, compute-equivalent-subexpressions, apply-cse): Remove
  attempts to deal with bailout, as the bailout pass handles that
  already.

10 years agoAdd prune-bailouts pass
Andy Wingo [Sat, 5 Apr 2014 09:08:47 +0000 (11:08 +0200)]
Add prune-bailouts pass

* module/language/cps/prune-bailouts.scm: New pass.
* module/language/cps/compile-bytecode.scm: Wire it up.
* module/Makefile.am: Add new file.

10 years agoDisable Tree-IL CSE
Andy Wingo [Sat, 5 Apr 2014 08:27:26 +0000 (10:27 +0200)]
Disable Tree-IL CSE

* module/language/tree-il/optimize.scm (optimize): Disable Tree-IL CSE
  by default.

10 years agoFix effects analysis for cached-module-box
Andy Wingo [Sat, 5 Apr 2014 09:32:06 +0000 (11:32 +0200)]
Fix effects analysis for cached-module-box

* module/language/cps/effects-analysis.scm (cached-module-box): Fix
  expected arity.

10 years agoFix coverage expectations
Andy Wingo [Fri, 4 Apr 2014 14:49:59 +0000 (16:49 +0200)]
Fix coverage expectations

* test-suite/tests/coverage.test ("line-execution-counts"): Update
  expectations.  Since there's nothing to do at the loop header and the
  renaming of X happens at the end of the loops, the compiled code only
  sees the loop header once.

10 years agoMore bailout preparation work
Andy Wingo [Fri, 4 Apr 2014 12:29:11 +0000 (14:29 +0200)]
More bailout preparation work

* module/language/cps/cse.scm (compute-available-expressions): Compute a
  bailout set -- or at least, set things up so that we can do so.
  (compute-idoms): Don't add predecessors that bail out.
  (apply-cse, cse, compute-equivalent-subexpressions): Thread the
  bailout set through the computations.

10 years agoPrepare for CSE bailout propagation
Andy Wingo [Fri, 4 Apr 2014 11:42:54 +0000 (13:42 +0200)]
Prepare for CSE bailout propagation

* module/language/cps/cse.scm (compute-available-expressions): Prepare
  for being able to prune joins from bailouts.  Always loop after the
  first iteration.

* module/language/cps/effects-analysis.scm: Remove &possible-bailout.
  Rename &definite-bailout to &bailout, and rename
  &all-effects-but-bailout to &unknown-effects.

10 years agoAdd common subexpression elimination pass on CPS
Andy Wingo [Fri, 4 Apr 2014 10:15:10 +0000 (12:15 +0200)]
Add common subexpression elimination pass on CPS

* module/language/cps/cse.scm: New file.
* module/language/cps/compile-bytecode.scm: Wire up CSE, on by default.
  Currently using the #:cps-cse? keyword.

* module/Makefile.am: Add new file.

10 years agoEffects analysis tweaks
Andy Wingo [Fri, 4 Apr 2014 10:08:52 +0000 (12:08 +0200)]
Effects analysis tweaks

* module/language/cps/effects-analysis.scm: Add &fluid-environment
  effect, a dependency of fluid-ref and fluid-set!, and an effect of
  push-fluid/pop-fluid.
  (list): Depend on &cdr.
  (resolve, cached-toplevel-box, cached-module-box): Don't depend on
  &box.

10 years agoFix verify-cps to work
Andy Wingo [Fri, 4 Apr 2014 10:07:24 +0000 (12:07 +0200)]
Fix verify-cps to work

* module/language/cps/verify.scm (verify-cps): Relax requirements for
  variable names to be symbols.

10 years agoconstant-needs-allocation? fix
Andy Wingo [Fri, 4 Apr 2014 10:06:59 +0000 (12:06 +0200)]
constant-needs-allocation? fix

* module/language/cps/dfg.scm (constant-needs-allocation?): Constants
  need allocation when they are used as a slot-needing operand, not when
  they are not used as an immediate operand.  Fixes the case where one
  var is used in both ways after CSE, in struct-set!/immediate.

10 years agoRemove variable-set! clause from compile-fun
Andy Wingo [Fri, 4 Apr 2014 08:39:42 +0000 (10:39 +0200)]
Remove variable-set! clause from compile-fun

* module/language/cps/compile-bytecode.scm (compile-fun): Remove
  vestigial `variable-set!' clause.

10 years agoEffects analysis: define causes-all-effects?
Andy Wingo [Thu, 3 Apr 2014 14:37:07 +0000 (16:37 +0200)]
Effects analysis: define causes-all-effects?

* module/language/cps/effects-analysis.scm (causes-all-effects?): New
  export.

10 years agobuild-cps niceties
Andy Wingo [Thu, 3 Apr 2014 14:36:23 +0000 (16:36 +0200)]
build-cps niceties

* module/language/cps.scm (build-cps-exp, build-cont-body): Respect
  unquote in list builders (kargs, call, callk, primcall, and values).

10 years agoMinor CSE optimization
Andy Wingo [Thu, 3 Apr 2014 07:40:18 +0000 (09:40 +0200)]
Minor CSE optimization

* module/language/tree-il/cse.scm (cse): Use hashq instead of modulo to
  convert a full-width hash value to a vector index.

10 years agoAdd with-fresh-name-state-from-dfg
Andy Wingo [Wed, 2 Apr 2014 14:25:07 +0000 (16:25 +0200)]
Add with-fresh-name-state-from-dfg

* module/language/cps/dfg.scm (with-fresh-name-state-from-dfg): New
  helper.
  ($dfg, compute-dfg): Store max-var and max-label in the dfg.

* module/language/cps.scm (with-fresh-name-state): Don't raise an error
  on recursive invocation; that was mostly useful when finding a bug.

* module/language/cps/arities.scm (fix-arities):
* module/language/cps/specialize-primcalls.scm (specialize-primcalls):
  Use the new helper.

* .dir-locals.el: Update.

10 years ago(test-suite lib) uses plain old catch, not stack-catch
Andy Wingo [Wed, 2 Apr 2014 20:00:14 +0000 (22:00 +0200)]
(test-suite lib) uses plain old catch, not stack-catch

* test-suite/test-suite/lib.scm (run-test-exception): Refactor to just
  use "catch" instead of stack-catch.

10 years agoRefactor toplevel scope name generation in compile-cps
Andy Wingo [Wed, 2 Apr 2014 13:58:06 +0000 (15:58 +0200)]
Refactor toplevel scope name generation in compile-cps

* module/language/tree-il/compile-cps.scm (scope-counter, fresh-scope-id):
  (toplevel-box, capture-toplevel-scope, convert, cps-convert/thunk):
  Refactor to avoid abusing the var counter to generate scope
  identifiers.

10 years agocompute-max-label-and-var takes letrec vars into account.
Andy Wingo [Wed, 2 Apr 2014 13:48:13 +0000 (15:48 +0200)]
compute-max-label-and-var takes letrec vars into account.

* module/language/cps.scm (compute-max-label-and-var): Fix to take
  letrec vars into account.

10 years agoFix DCE for refactor-introduced borkage
Andy Wingo [Wed, 2 Apr 2014 13:41:14 +0000 (15:41 +0200)]
Fix DCE for refactor-introduced borkage

* module/language/cps/dce.scm ($fun-data, compute-live-code)
  (process-eliminations): Fix clownshoes regarding fun-data field names
  and order.

10 years agoFix prune-top-level-scopes to allow collisions between var, scope, cont names
Andy Wingo [Wed, 2 Apr 2014 13:40:03 +0000 (15:40 +0200)]
Fix prune-top-level-scopes to allow collisions between var, scope, cont names

* module/language/cps/prune-top-level-scopes.scm (compute-referenced-scopes):
  Fix to not assume that scope names, continuation names, and var names
  are mutually unique.
  (prune-top-level-scopes): Better variable names.

10 years agoUpdate old-style REPL code for deprecation
Andy Wingo [Wed, 2 Apr 2014 10:08:48 +0000 (12:08 +0200)]
Update old-style REPL code for deprecation

* module/ice-9/scm-style-repl.scm:
* module/ice-9/save-stack.scm: As the deprecated bindings have been
  removed from the default environment, use #:export instead of
  #:replace.

10 years agoRemove CFA data type
Andy Wingo [Wed, 2 Apr 2014 10:00:09 +0000 (12:00 +0200)]
Remove CFA data type

* module/language/cps/dfg.scm: Remove CFA data type.
  (analyze-reverse-control-flow): Take min-label and label-count as
  args, and return multiple values instead of returning a CFA object.
  (compute-live-variables): Rework to accept multiple values from
  analyze-reverse-control-flow.
  ($dfa): Update comments.