* include/lightning/jit_s390x.h, lib/jit_s390x-cpu.c,
lib/jit_s390x-fpu.c, lib/jit_s390x.c: New files
implementing the new s390x port.
* configure.ac, include/lightning.h,
include/lightning/Makefile.am,
include/lightning/jit_private.h,
lib/Makefile.am, lib/jit_disasm.c, lib/lightning.c:
Minor adaptation for the new s390x backend.
* check/float.tst: Update for the s390x result of
truncating +Inf to integer.
* check/qalu_mul.tst: Add extra test cases to better test
high word of signed multiplication as the result is
adjust from unsigned multiplication on s390x.
* check/lightning.c: Do not assume casting a double NaN or
Inf to float will produce the expected float NaN or Inf.
This is not true at least under s390x.
* check/check.arm.sh, check/check.sh, check/check.swf.sh,
check/check.x87.sh: Properly check test programs output,
not just rely on the test program self testing the results
and not crashing.
* module/ice-9/boot-9.scm (iota): Move up.
(make-record-type, define-record-type): Use allocate-struct and
struct-set!.
* module/srfi/srfi-9.scm (%%set-fields, %define-record-type): Use
allocate-struct and struct-set!.
Note that this makes the stack VM slower, but it will make RTL
compilation faster.
* libguile/struct.h:
* libguile/struct.c (scm_allocate_struct): New interface: allocates a
struct.
* libguile/vm-engine.c (allocate_struct): Instead of make-struct with a
rest arg, separate allocation from initialization.
* libguile/vm-engine.c (push_frame, call): Separate out push_frame from
call, and expect the caller to arrange the proc and arguments.
* test-suite/tests/rtl.test ("call"): Update tests.
* libguile/vm-engine.c: Change the RTL VM to number the procedure as
local 0, and other locals from 1. In the future we will want the FP
to point to local 0 instead of local 1. In the future also we can
elide the procedure for well-known closures (closures in which all
references are known call sites).
(make_closure, free_set): Instead of taking rest arguments, we add a
new free-set! op that initializes closures.
(free_ref): Take the closure as an argument.
* libguile/vm.c (rtl_boot_continuation_code): Remove comments, which
were out of date.
(rtl_apply_code, rtl_values_code): Update comments.
* module/system/vm/assembler.scm (intern-constant, emit-init-constants):
Adapt to locals numbering change.
(begin-kw-arity): For assert-nargs-ee purposes, nreq includes the
procedure.
* module/system/vm/disassembler.scm (code-annotation): Adapt annotation
for assert-nargs-ee/locals.
* test-suite/tests/rtl.test: Adapt tests.
* libguile/instructions.c (scm_rtl_instruction_list): Add an element to
the list to indicate that an instruction outputs to its first
argument.
* module/system/vm/assembler.scm:
* module/system/vm/disassembler.scm: Adapt.
* libguile/programs.c (parse_arity): Lookup rtl-program-minimum-arity
from (system vm debug).
* module/system/vm/debug.scm (find-first-arity): Fix the linear search.
Whoops!
* module/rnrs/arithmetic/bitwise.scm (bitwise-if, bitwise-length,
bitwise-first-bit-set, bitwise-bit-field, bitwise-reverse-bit-field):
Replace these with aliases to the identical SRFI-60 operators
'bitwise-if', 'integer-length', 'first-set-bit', 'bit-field', and
'reverse-bit-field'.
(bitwise-copy-bit, bitwise-copy-bit-field, bitwise-rotate-bit-field):
Reimplement these based upon the similar SRFI-60 operators 'copy-bit',
'copy-bit-field', and 'rotate-bit-field'.
* test-suite/tests/r6rs-arithmetic-bitwise.test (bitwise-copy-bit): Fix
test to conform to the specification, which requires the third
argument to be either 0 or 1.
* test-suite/tests/r6rs-arithmetic-fixnums.test (fxcopy-bit): Fix test
to conform to the specification, which requires the third argument to
be either 0 or 1.
Fixes <http://bugs.gnu.org/14905>.
Reported by Göran Weinholt <goran@weinholt.se>.
* libguile/numbers.c (scm_rationalize): Rewrite. Previously an
incorrect algorithm was used which failed in many cases.
* test-suite/tests/numbers.test (rationalize): Add tests.
* libguile/tags.h (SCM_UNPACK): Change (*(SCM*)0=(x)) to
(*(volatile SCM *)0=(x)), to avoid null-dereference warnings
from clang. Reported by Shane Celis <shane.celis@gmail.com>.
* module/web/http.scm (http-proxy-port?, set-http-proxy-port?!): New
exported procedures.
(write-request-line): If we're using an http proxy, write an
absolute-URI in the request line.
* module/web/client.scm: Import (web http).
(current-http-proxy): New exported parameter.
(open-socket-for-uri): If 'current-http-proxy' is not false,
connect to the proxy instead of the URI host, and use
'set-http-proxy-port?!' to make note of that fact.
* doc/ref/web.texi (Web Client): Document 'current-http-proxy'.
* include/lightning/jit_aarch64.h, lib/jit_aarch64-cpu.c,
lib/jit_aarch64-fpu.c, lib/jit_aarch64.c: New files
implementing the new aarch64 port, as a new architecture,
not as an expansion of the existing armv[4-7] port.
* check/lightning.c: Add aarch64 support and a small
change to recognize character constants as immediate
values.
* check/float.tst: Add aarch64 preprocessor conditionals
to select proper expected value when converting [+-]Inf
and NaN to integer.
* include/lightning/jit_arm.h, lib/jit_arm.c: Minor changes
to better match the new aarch64 files.
* configure.ac, include/lightning.h,
include/lightning/Makefile.am, include/lightning/jit_private.h,
lib/Makefile.am, lib/lightning.c: Minor adjustments
for the aarch64 port.
Reported by Göran Weinholt <goran@weinholt.se>.
* module/rnrs/arithmetic/fixnums.scm (fxbit-count): If the argument is
negative, return the 'bitwise-not' of the result of 'logcount', as per
R6RS. Previously, 'fxbit-count' was identical to 'logcount'.
* test-suite/tests/r6rs-arithmetic-fixnums.test (fxbit-count): Add test.
Fixes <http://bugs.gnu.org/14868>.
Reported by Göran Weinholt <goran@weinholt.se>.
* module/rnrs/arithmetic/flonums.scm (flfinite?): If the argument is a
NaN, return false.
* test-suite/tests/r6rs-arithmetic-flonums.test (flfinite?): Add test.
* libguile/numbers.c (scm_num_eq_p): Fix bug comparing fractions to
infinities (reported by Göran Weinholt <goran@weinholt.se>). Fix
erroneous comment describing the logic behind inum/flonum comparison.
Use similar logic for inum/complex comparison to avoid rounding
errors. Make minor indentation fixes and simplifications.
* test-suite/tests/numbers.test (=): Add tests.
* libguile/numbers.c (INUM_LOSSLESSLY_CONVERTIBLE_TO_DOUBLE):
New macro.
(scm_i_divide2double): Use INUM_LOSSLESSLY_CONVERTIBLE_TO_DOUBLE to
determine if our fast path is safe. Previously, negative arguments
were not checked properly.
* test-suite/tests/numbers.test (exact->inexact): Add tests.
Fixes <http://bugs.gnu.org/14864>.
Reported by Göran Weinholt <goran@weinholt.se>.
* libguile/vm-i-scheme.c (ash): Fallback to 'scm_ash' for right shifts
with counts >= SCM_I_FIXNUM_BIT, since '>>' is not guaranteed to work
correctly for large counts.
Fixes <http://bugs.gnu.org/14864>.
Reported by Göran Weinholt <goran@weinholt.se>.
* module/rnrs/arithmetic/bitwise.scm (bitwise-bit-count): If the
argument is negative, return the 'bitwise-not' of the result of
'logcount', as per R6RS. Previously, 'bitwise-bit-count' was
identical to 'logcount'.
* module/language/tree-il.scm (<prompt>): Change to have the body and
handler be lambdas, and add an "escape-only?" field. This will make
generic prompts work better in CPS or ANF with the RTL VM, as it
doesn't make sense in that context to capture only part of a frame.
Escape-only prompts can still be fully inlined.
(parse-tree-il, unparse-tree-il): Add escape-only? to the
serialization.
(make-tree-il-folder, pre-post-order): Deal with escape-only?.
* module/language/tree-il/analyze.scm (analyze-lexicals): Handle
escape-only?, and the new expectations for the body and handler.
* module/language/tree-il/canonicalize.scm (canonicalize): Ensure that
the body of an escape-only continuation is a thunk, and that the
handler is always a lambda.
* module/language/tree-il/debug.scm (verify-tree-il): Assert that
escape-only? is a boolean.
* module/language/tree-il/cse.scm (cse):
* module/language/tree-il/effects.scm (make-effects-analyzer):
* module/language/tree-il/peval.scm (peval):
* module/language/tree-il/primitives.scm (*primitive-expand-table*):
* test-suite/tests/peval.test ("partial evaluation"):
* module/language/tree-il/compile-glil.scm (flatten-lambda-case): Adapt
to <prompt> change.
* libguile/dynstack.h (SCM_DYNSTACK_TYPE_WITH_FLUID): Rename from
with-fluids.
* libguile/dynstack.c (scm_dynstack_push_fluid):
(scm_dynstack_unwind_fluid): Change API to only wind/unwind one
fluid binding.
(WITH_FLUID_WORDS): New define, always 2 words (fluid and value box).
(WITH_FLUID_FLUID, WITH_FLUID_VALUE_BOX): New macros to get offsets of
fluid and value box.
(scm_dynstack_push_rewinder, scm_dynstack_push_unwinder): Use
WINDER_WORDS.
(scm_dynstack_push_dynwind): Use DYNWIND_WORDS.
(scm_dynstack_wind_1): Update for scm_swap_fluid API change.
* libguile/fluids.h:
* libguile/fluids.c (scm_prepare_fluids): Remove; no longer needed.
(scm_swap_fluid): Update to just swap one fluid binding.
(scm_c_with_fluids, scm_c_with_fluid): Update to use
scm_dynstack_push_fluid.
* libguile/memoize.c (do_push_fluid, do_pop_fluid): Adapt to API
change.
* libguile/vm-engine.c (rtl_vm_engine): Change wind-fluids / unwind-fluids
to push-fluid / pop-fluid, and actually enable. Woo!
* libguile/vm-i-system.c (push-fluid, pop-fluid): Update to new API.
* libguile/vm-i-system.c (push-fluid, pop-fluid):
* doc/ref/vm.texi (Dynamic Environment Instructions): Rename wind-fluids
to push-fluid, and unwind-fluids to pop-fluid. They now only work on
one fluid binding at a time.
* module/ice-9/boot-9.scm (with-fluid*): Implement in Scheme in terms of
primcalls to push-fluid and pop-fluid.
(custom-throw-handler, catch, with-throw-handler): Use with-fluid*
instead of with-fluids, as with-fluids is no longer available before
psyntax is loaded.
(with-fluids): Define in Scheme in terms of with-fluid*.
* libguile/fluids.c (scm_with_fluid): Rename from scm_with_fluids, and
don't expose to Scheme.
* libguile/eval.c (eval): Remove SCM_M_WITH_FLUIDS case.
* libguile/expand.c (expand_with_fluids): Remove with-fluids syntax.
(DYNLET): Remove, no longer defining dynlet in the %expanded-vtables.
* libguile/expand.h: Remove dynlet definitions.
* module/ice-9/eval.scm (primitive-eval): Remove with-fluids case.
* libguile/memoize.c (do_push_fluid, do_pop_fluid): New primitive
helpers, like wind and unwind.
(memoize): Memoize wind and unwind primcalls. Don't memoize dynlet to
with-fluids.
(scm_init_memoize): Initialize push_fluid and pop_fluid here.
* libguile/memoize.h (SCM_M_WITH_FLUIDS): Remove definition.
* module/ice-9/psyntax.scm (build-dynlet): Remove; this just supported
with-fluids, which is now defined in boot-9.
* module/ice-9/psyntax-pp.scm: Regenerate.
* doc/ref/compiler.texi (Tree-IL):
* module/language/tree-il.scm:
* module/language/tree-il/analyze.scm:
* module/language/tree-il/canonicalize.scm:
* module/language/tree-il/compile-glil.scm:
* module/language/tree-il/cse.scm:
* module/language/tree-il/debug.scm:
* module/language/tree-il/effects.scm: Remove <dynlet>. Add cases for
primcalls to push-fluid and pop-fluid in compile-glil.scm and
effects.scm.
* module/language/tree-il/peval.scm (peval): Factor out
with-temporaries; probably a bad idea, but works for now. Factor out
make-begin0 (a better idea). Inline primcalls to with-fluid*, and
remove dynlet cases.
* module/language/tree-il/primitives.scm (*interesting-primitive-names*):
Add with-fluid*.
