This patch results in a 20%-40% speedup in the > 1 argument cases of
the following microbenchmarks:
(define A (make-shared-array #0(1) (const '()) #e1e7))
; 1, 2, 3 arguments.
(define a 0) ,time (array-for-each (lambda (b) (set! a (+ a b))) A)
(define a 0) ,time (array-for-each (lambda (b c) (set! a (+ a b c))) A A)
(define a 0) ,time (array-for-each (lambda (b c d) (set! a (+ a b c d))) A A A)
(define A (make-shared-array (make-array 1) (const '()) #e1e7))
(define B (make-shared-array #0(1) (const '()) #e1e7))
; 1, 2, 3 arguments.
,time (array-map! A + B)
,time (array-map! A + B B)
,time (array-map! A + B B B)
* libguile/array-map.c (scm_ramap): Note on cproc arguments.
(rafill): Assume that dst's lbnd is 0.
(racp): Assume that src's lbnd is 0.
(ramap): Assume that ra0's lbnd is 0. When there're more than two
arguments, compute the array handles before the loop. Allocate the arg
list once and reuse it in the loop.
(rafe): Do as in ramap(), when there's more than one argument.
(AREF, ASET): Remove.
* libguile/sort.c (sort!, sort, restricted-vector-sort!, sorted?):
Support arrays of rank 1, whatever the type.
* libguile/quicksort.i.c: Fix accessors to handle typed arrays.
* test-suite/tests/sort.test: Test also with typed arrays.
* libguile/arrays.c (scm_shared_array_root): Adopt uniform check order.
(scm_shared_array_offset, scm_shared_array_increments): Use the array
fields directly just as scm_shared_array_root does.
(scm_c_array_rank): Moved from libguile/generalized-arrays.c. Don't
use array handles, but follow the same type check sequence as the
other array functions (shared-array-root, etc).
(scm_array_rank): Moved from libguile/generalized-arrays.h.
* libguile/arrays.h: Move prototypes here.
* test-suite/tests/arrays.test: Tests for shared-array-offset,
shared-array-increments.
* libguile/root.h:
* libguile/root.c: Remove these files.
* libguile/deprecated.h:
* libguile/deprecated.c (scm_internal_cwdr, scm_call_with_dynamic_root)
(scm_dynamic_root, scm_apply_with_dynamic_root): Deprecate.
Remove all root.h usage, which was vestigial.
* module/ice-9/serialize.scm: Use (current-thread) instead
of (dynamic-root).
Since the move to BDW-GC this option has not been useful.
* libguile/__scm.h (SCM_DEBUG_CELL_ACCESSES): Remove default
definition.
* libguile/gc.h: Add comment about cells.
(SCM_VALIDATE_CELL): Remove. I did a search on the internet and I
found no external users.
(SCM_CELL_WORD, SCM_CELL_OBJECT, SCM_SET_CELL_WORD)
(SCM_SET_CELL_OBJECT, SCM_CELL_OBJECT_LOC): Remove SCM_VALIDATE_CELL
uses.
* libguile/gc.c (scm_debug_cell_accesses_p)
(scm_expensive_debug_cell_accesses_p)
(scm_debug_cells_gc_interval, scm_i_expensive_validation_check)
(scm_assert_cell_valid): Remove bindings only available when
SCM_DEBUG_CELL_ACCESSES was 1.
* libguile/list.c (scm_list_n): Remove SCM_VALIDATE_CELL usage.
* libguile/foreign.c (CODE, get_foreign_stub_code): Add explicit
handle-interrupts and return-values calls, as foreign-call will fall
through.
* libguile/gsubr.c (A, B, C, AB, AC, BC, ABC, SUBR_STUB_CODE)
(scm_i_primitive_call_ip): Same.
* libguile/vm-engine.c (VM_HANDLE_INTERRUPTS): Inline into
handle-interrupts.
(RETURN_ONE_VALUE, RETURN_VALUE_LIST): Inline into callers, and fall
through instead of returning.
(BR_BINARY, BR_UNARY, BR_ARITHMETIC, BR_U64_ARITHMETIC): Remove
conditional VM_HANDLE_INTERRUPTS, as the compiler already inserted the
handle-interrupts calls if needed.
(vm_engine): Remove VM_HANDLE_INTERRUPTS invocations except in the
handle-interrupts instruction.
* module/ice-9/threads.scm (join-thread): Implement in Scheme.
(call-with-new-thread): Arrange to record values in a weak table and
signal the join cond.
(with-mutex): Move up definition; call-with-new-thread needs it. (How
was this working before?)
* libguile/threads.c (guilify_self_1, guilify_self_2, do_thread_exit):
Remove join queue management.
* libguile/threads.c (scm_join_thread, scm_join_thread_timed): Call out
to Scheme.
(scm_init_ice_9_threads): Capture join-thread var.
* libguile/throw.h (scm_i_make_catch_body_closure)
(scm_i_make_catch_handler_closure): Add scm_i_ prefix and make
available for internal use.
* libguile/throw.c: Adapt.
* libguile/threads.c (scm_spawn_thread): Rewrite in terms of
scm_call_with_new_thread.
* libguile/threads.c (lock_mutex, scm_timed_lock_mutex): As with
unlock_mutex before, optimize by specializing to the mutex kind.
Also, avoid lock thrashing after a return from block_self.
* libguile/threads.c (scm_timed_lock_mutex): Use "level" field only for
recursive mutexes.
(unlock_mutex, scm_unlock_mutex): Factor implementation back out of
scm_unlock_mutex (doh?), but in such a way that can specialize against
mutex type.
(scm_mutex_level): Only look at level for recursive mutexes.
(scm_mutex_locked_p): Look at owner field, not level field.
(timed_wait, scm_timed_wait_condition_variable): Factor implementation
out, as above with unlock-mutex. Specialize relocking of the Guile
mutex.
* libguile/threads.c (scm_unlock_mutex): Bind to unlock-mutex.
* libguile/threads.h: Remove scm_unlock_mutex_timed.
* libguile/deprecated.h: Add scm_unlock_mutex_timed.
* libguile/deprecated.c (scm_unlock_mutex_timed): Deprecate.
* test-suite/tests/threads.test: Update unlock-mutex tests to use
wait-condition-variable if they would wait on a cond.
* libguile/threads.c (enum fat_mutex_kind): New data type, replacing
separate flags.
(struct fat_mutex): Adapt.
(make_fat_mutex): Fat mutexes can only be one of three kinds, not one
of 4 kinds. (Recursive unowned mutexes are not a thing.)
(scm_make_mutex): Adapt.
(scm_make_mutex_with_kind): New function, replacing
scm_make_mutex_with_flags. Still bound to make-mutex.
(scm_make_recursive_mutex): Adapt.
(fat_mutex_lock, fat_mutex_unlock): Adapt.
* libguile/threads.h (scm_make_mutex_with_kind): New decl.
* libguile/deprecated.h:
* libguile/deprecated.c (scm_make_mutex_with_flags): Deprecate.
* libguile/threads.c (fat_mutex_lock): allow-external-unlock mutexes
can't be recursive, but a recursive lock attempt can be unblocked by
an external thread, so these mutexes shouldn't throw an error on
recursive lock attempts.
* test-suite/tests/srfi-18.test: Add tests.
* libguile/threads.h (scm_i_thread):
* libguile/threads.c (guilify_self_1, on_thread_exit)
(scm_pthread_cond_wait, scm_pthread_cond_timedwait): The thread-local
held_mutex field is no longer needed, now that we cancel threads via
interrupts instead of pthread_cancel.
* libguile/scmsigs.c (signal_handler_lock): New variable.
(scm_sigaction_for_thread): Block asyncs and use a mutex instead of
using scm_dynwind_critical_section.
* libguile/debug.c (scm_debug_options): Remove critical section. There
is no memory corruption hazard and racing callers to debug options
won't produce sensible results anyway; or anyway they will be the same
as racing vector-set!.
* libguile/stime.c (scm_current_time): No need for a critical section
around time().
(tz_lock): New mutex.
(scm_localtime): Use tz_lock in a dynwind. Fixes bug in which
critical section was never left on error.
(scm_gmtime, scm_mktime, scm_strftime, scm_strptime): Use a normal
mutex instead of a critical section.
