* include/lightning/jit_private.h, lib/jit_note.c, lib/lightning.c:
Store all annotation information in the mmap'ed area reserved for
read only data. This adds code to not allocate memory for jit_note_t
objects, and to relocate jit_line_t objects and its contents after
calculating annotation information. The jit_line_t objects are
relocated because it is not possible to always calculate before
hand data layout because note information may be extended or
redundant entries removed, as well as allowed to be added in
non sequential order.
A bug was also corrected in _jit_set_note, that was causing it
to allocate new jit_line_t objects when not needed. It was still
working correctly, but allocating way more memory than required.
*include/lightning.h, lib/lightning.c: Add the new jit_live code
to explicitly mark a register as live. It is required to avoid
assuming functions always return a value in the gpr and fpr return
register, and to avoid the need of some very specialized codes
that vary too much from backend to backend, to instruct the
optimization code the return register is live.
* lib/jit_arm.c, lib/jit_mips.c, lib/jit_ppc.c, lib/jit_print.c,
lib/jit_x86.c: Update for the new jit_live code.
* check/ret.ok, check/ret.tst: New files implementing a simple
test case that would previously fail at least in ix86/x86_64.
* check/Makefile.am: Update for new "ret" test case.
2013-02-04 Paulo Andrade <pcpa@gnu.org>
* include/lightning.h, include/lightning/jit_private.h,
lib/jit_arm-cpu.c, lib/jit_arm.c, lib/jit_mips-cpu.c,
lib/jit_mips.c, lib/jit_ppc-cpu.c, lib/jit_ppc.c,
lib/jit_x86-cpu.c, lib/jit_x86.c, lib/lightning.c:
Implement the new qmul and qdiv instructions that return signed
and unsigned lo/hi multiplication result and div/rem division result.
These should be useful for jit translation of code that needs to
know if a multiplication overflows (no branch opcode added) or if
a division is exact (easy check if remainder is zero).
* check/lightning.c, lib/jit_print.c, check/Makefile.am,
check/all.tst: Update for the new qmul and qdiv instructions.
* check/qalu.inc, check/qalu_div.ok, check/qalu_div.tst,
check/qalu_mul.ok, check/qalu_mul.tst: New files implementing
simple test cases for qmul and qdiv.
* check/3to2.tst, check/add.tst, check/allocai.tst, check/bp.tst,
check/call.tst, check/ccall.c, check/clobber.tst, check/divi.tst,
check/fib.tst, check/ldsti.tst, check/ldstr-c.tst, check/ldstr.tst,
check/ldstxi-c.tst, check/ldstxi.tst, check/ldstxr-c.tst,
check/ldstxr.tst, check/lightning.c, check/rpn.tst, check/stack.tst,
check/varargs.tst, include/lightning.h,
include/lightning/jit_private.h, lib/jit_arm.c, lib/jit_disasm.c,
lib/jit_mips.c, lib/jit_note.c, lib/jit_ppc.c, lib/jit_print.c,
lib/jit_x86.c, lib/lightning.c: Extend the "jit_note" abstraction
with the new "jit_name" call, that receives a string argument, and
should usually be called to mark boundaries of functions of code
generating jit (that is, it is not expected that the language
generating jit map its functions to jit functions).
2013-01-14 Paulo Andrade <pcpa@gnu.org>
* include/lightning.h, lib/lightning.c: Add an extra align
argument to the jit_data call (that should be made private),
so that it should not align strings at 8 bytes.
Correct the jit_note call to include the null ending byte
when adding label/note names to the "jit data section".
* lib/jit_note.c: New file implementing a simple string+integer
annotation, that should be used to map filename and line number
to offsets in the generated jit.
* include/lightning.h, lib/lightning.c: Update for the new
note code.
Add an extra mandatory argument to init_jit, that is used
as argument to bfd_openr.
Change from generic void* to char* the argument to jit_note
and add an extra integer argument, to map to filename and
line number.
* check/ccall.c, check/lightning.c, include/lightning/jit_private.h,
lib/jit_arm.c, lib/jit_disasm.c, lib/jit_mips.c, lib/jit_ppc.c,
lib/jit_print.c, lib/jit_x86.c: lib/Makefile.am: Update for the
new annotation code.
* configure.ac, check/Makefile.am: Update to work with latest
automake.
* check/cccall.c, check/ccall.ok: New test case to validate
interleaved calls from/to C code and jit.
* check/Makefile.am: Update for the new ccall test case.
* include/lightning.h, lib/lightning.c: Add the new jit_address
call that returns the real/final address of a "note" in the
generated jit. It requires a jit_node_t as returned by the
jit_note call, and is only valid after calling jit_emit.
Add an intermediate solution to properly handle arm
soft and softfp modes that move a double to an integer register
pair. Currently it just adds extra tests for the condition,
but the proper solution should be to have extra lightning
codes for these conditions, codes which should be only used
by the backends that need it, and merged with the existing
jit_pusharg*_{f,d}.
* include/lightning/jit_private.h: Add new jit_state_t flag
to know it finished jit_emit, so that calls to jit_address
are valid.
* lib/jit_mips.c: Correct abi implementation so that the
new ccall test case pass. Major problem was using
_jit->function.self.arg{i,f} as boolean values, but that
would cause lightning.c:patch_registers() to incorrectly
assume only one register was used as argument when calling
jit_regarg_p(); _jit->function.self.arg{i,f} must be the
number of registers used as arguments (in all backends).
* lib/jit_x86.c: Add workaround, by marking %rax as used,
to a special condition, when running out of registers and the
allocator trying to spill and reload %rax, but %rax was used
as a pointer to a function, what would cause the reload to
destroy the return value. This condition can be better
generalized, but the current solution is good enough.
* include/lightning/jit_ppc.h, lib/jit_ppc-cpu.c, lib/jit_ppc.c:
Rewrite logic to handle arguments, as the original code was
written based on a SysV pdf about the generic powerpc ABI,
what did "invent" a new abi for the previous test cases, but
failed in the new ccall test in Darwin PPC. Now it properly
handles 13 float registers for arguments, as well as proper
computation of stack offsets when running out of registers
for arguments.
* lib/lightning.c: Correct both, wrong and confusing logic
to compute the reverse of a jump. Now it properly matches
C semantics for "eq" (==) and "ne" (!=) and correct computation
of reverse of "uneq" as "gt".
* check/branch.tst: Update "ne" float branch check that
previously happened to be wrongly tested with a NaN argument.
* check/lightning.c, include/lightning.h, lib/jit_arm.c,
lib/jit_mips.c, lib/jit_ppc.c, lib/jit_print.c, lib/jit_x86.c,
lib/lightning.c: Change return value of jit_arg{,_f,_d} to
a jit_node_t* object, that should be used as argument to
jit_getarg_{c,uc,s,us,i,ui,l,f,d}. This just requires changing
from jit_int32_t to jit_pointer_t (or jit_node_t*) the "handle"
for the getarg calls, with the benefit that it makes it easy
to implement patching of the stack address of non register
arguments, this way allowing to implement variable size stack
frames if applicable; useful if there are too many registers and
jit functions uses only a few callee save registers.
* lib/lightning.c: Partially rewrite/revert code to compute
initial register live state at the start of a basic block.
The original logic was corrupted when adding optimizations
to do as few computations as possible in jit_update. The
reglive field must be always a known set of live registers
at the start of a basic block. The value that was incorrect
was the regmask field, that must be the set of registers
that are in unknown state, because they are not known live,
neither set (or possibly not set) in the basic block, and
*must* store the state at the start of the basic block.
* check/stack.ok, check/stack.tst: New files to test data
integrity on a deep chain of stack frames.
* lib/jit_arm.c, lib/jit_arm-cpu.c, lib/jit_mips.c,
lib/jit_mips-cpu.c, lib/jit_ppc.c, lib/jit_ppc-cpu.c,
lib/jit_x86.c, lib/jit_x86-cpu.c: Calculate _jit->function->stack
in the emit stage, otherwise it will calculate it wrong if
need to jit_allocai space to spill registers.
* lib/lightning.c: Correct wrong offset when updating the
"current" jit function pointer in the code that may need to
allocate stack space to spill registers.
* check/lightning.c: Correct off by one data space check.
* check/Makefile.am: Update for new test case.
* include/lightning.h, include/lightning/jit_private.h,
lib/jit_arm.c, lib/jit_mips.c, lib/jit_ppc.c, lib/jit_x86.c,
lib/lightning.c: Make jit_ellipsis implementation not
backend specific. It is not intended to handle va_list
like objects at runtime, as jit_arg* and jit_getarg*
return constant values resolved at parse time, so, effectively
it is not possible to create printf like jit functions, as
there is no va_start, va_arg, va_end, etc, abstraction. This
limitation should be kept for the sake of making new ports
easier.
* include/lightning.h, lib/jit_arm.c, lib/jit_mips.c,
lib/jit_ppc.c, lib/jit_x86.c, lib/lightning.c: Change jit_prepare
to no longer receive an argument. If receiving an argument, it
should be an ABI specifier, not a boolean if varargs or not,
and add the new jit_ellipsis call, to specify where the
ellipsis is in the C prototype of the function being called.
Note that currently it is not supported to define varargs
functions and it will be ignored if calling jit_ellipsis not
in a prepare/finish* block, but this should be addressed.
* check/allocai.tst, check/alu_add.tst, check/alu_and.tst,
check/alu_com.tst, check/alu_div.tst, check/alu_lsh.tst,
check/alu_mul.tst, check/alu_neg.tst, check/alu_or.tst,
check/alu_rem.tst, check/alu_rsh.tst, check/alu_sub.tst,
check/alu_xor.tst, check/alux_add.tst, check/alux_sub.tst,
check/bp.tst, check/branch.tst, check/cvt.tst, check/divi.tst,
check/fib.tst, check/ldsti.tst, check/ldstr-c.tst,
check/ldstr.tst, check/ldstxi-c.tst, check/ldstxi.tst,
check/ldstxr-c.tst, check/ldstxr.tst, check/rpn.tst,
check/lightning.c: Update for the change to jit_prepare and
addition of jit_ellipsis.
* lib/jit_ppc-cpu.c: Make movr a function that checks arguments
so that other code can safely assume it is a noop if src and dst
are the same register.
Implement rem{r,i}{,_u} as a div{,u}/mul/sub.
Correct ANDIS, ORIS and XORIS calls to cast the argument to
unsigned before the shift to avoid an assertion if the argument
had the topmost bit set.
Implement lshi, rshi and rshi_u as functions to test for a
zero argument, that would otherwise trigger an assertion when
computing the shift value.
Do a simple implementation of bm{s,c}{r,i} with a temporary,
"andr" of arguments and jump based on comparison with zero.
Correct typo in ldxi_c.
* lib/jit_ppc-fpu.c: Correct wrong arguments to FDIV* and STF*.
* lib/jit_ppc.c: Correct wrong check for 6 instead of 8 integer
arguments in registers. If calling a varargs function and
passing a float or double argument, also either store the
value in the stack or in integer registers, as varargs functions
do not fetch it from float registers.
Add "case" for new functions and incorrectly missing ones.
Call libgcc's __clear_cache, that should know what to do
if the hardware needs flushing cache before execution.
* lib/lightning.c: Do a simple/trivial logic in jit_regset_scan1,
that should make it easier for the compiler to optimize it, and
that also corrects the previously wrong code for big endian, and
that was causing problems in ppc due to not saving all callee save
registers as it was not "finding" them in the regset due to the
little endian assumption bug.
* configure.ac: Only default to using the builtin disassembler
if on GNU/Linux. This should be temporary, due to requiring
/proc/self/exe.
Correctly check $target_cpu for powerpc.
* include/lightning/jit_ppc.h: Correctly implement jit_v_num.
* include/lightning/jit_private.h: Declare proper prototype
for jit_init_debug and jit_finish_debug.
* lib/jit_ppc-cpu.c: Remove code to save/restore callee save
float registers, as it is not required since those float
registers are not usable currently.
Change prolog and epilog generation to, at least comparing
code, match what gcc generates in "gcc -O0", but it is still
failing in Darwin PPC, apparently due to the __clear_cache
call not being enough, as frequently it will also fail to
execute, and the code buffer is all zeroes.
* lib/lightning.c: Do not fail in jit_regset_scan1 calls due
to passing 64 as argument on computers with 64 registers.
* check/alu.inc, check/alu_add.ok, check/alu_add.tst,
check/alu_and.ok, check/alu_and.tst, check/alu_com.ok,
check/alu_com.tst, check/alu_div.ok, check/alu_div.tst,
check/alu_lsh.ok, check/alu_lsh.tst, check/alu_mul.ok,
check/alu_mul.tst, check/alu_neg.ok, check/alu_neg.tst,
check/alu_or.ok, check/alu_or.tst, check/alu_rem.ok,
check/alu_rem.tst, check/alu_rsh.ok, check/alu_rsh.tst,
check/alu_sub.ok, check/alu_sub.tst, check/alu_xor.ok,
check/alu_xor.tst, check/alux_add.ok, check/alux_add.tst,
check/alux_sub.ok, check/alux_sub.tst, check/branch.ok,
check/branch.tst: New test cases for arithmetic and branch
tests.
* check/Makefile.am: Update for new test cases.
* include/lightning/jit_private.h: Make the jit_reg_free_p
macro shared by all backends. Previously was added for the
arm backend, but is useful in the x86_64 backend when checking
state of "special purpose register".
Also add the new jit_class_named register class, that must be
or'ed with the register value if calling jit_get_reg expecting
an specific value, because the specific register value may be
zero, that previously was treated as no register requested.
* lib/jit_arm-cpu.c: Correct argument order for T2_MVN.
* lib/jit_arm-swf.c: Call the proper function for double
divide. The "software float" implementation just calls
libgcc functions.
* lib/jit_arm.c: Return float/double values in the float
register if using the hard float ABI.
* lib/jit_x86-cpu.c: Change the can_sign_extend_int_p macro
to not include -0x80000000L, because there is code that
"abuses" it and thinks it can negate the immediate value
after calling that macro.
Correct implementation of jit_subi that had a wrong code
patch logic doing subtraction with reversed arguments.
Correct REX prefix calculation in the jit_muli implementation.
Correct logic to get/unget %*ax and %*dx registers in divremr
and divremi.
Correct divremi that was using the symbolic, unique %*ax
value in on place (not using the _REGNO name suffix).
Correct cut&paste error causing it to use "xor" instead of
"or" in one code path of the jit_ori implementation.
Correct several flaws when clobbering registers and/or when
one of the arguments was %*cx in the rotshr wrapper function
implementing most shift operations.
* lib/lightning.c: No longer expect that the backend be smart
enough to know what to do when asking for a named register
if that register is already an argument or is live. It fails
if it is an argument, or if register is live, fails if cannot
spill.
No longer incorrectly assume that eqr_{f,d} and ltgr_{f,d} are
safe to inverse value tests in jump thread optimization.
* include/lightning/jit_mips.h, lib/jit_mips-cpu.c,
lib/jit_mips-fpu.c, lib/jit_mips.c: Correct float/double
argument handling and make the mips backend pass the initial
test cases.
* include/lightning.h, ib/jit_print.c, lib/lightning.c:
Add extra enum values for argument handling functions that
could not be abstracted to the current codes, that is, when
float values need to move from/to gpr from/to fpr. It would
be more tempting to add such primitives, but they would have
wordsize limitations, and it is not expected to add codes
with one gpr argument for 64 bit and two for 32 bit.
* lib/jit_ppc.c: Check _jit->function before calling jit_epilog()
to avoid a runtime exception.
* check/all.tst, check/lightning.c: Only declare or use 64 bit
interfaces on 64 bit builds.
* check/fib.tst: Use simpler logic to not need preprocessor
conditionals for 32 or 64 bit.
* include/lightning.h: Only declare 64 bit macros on a 64 bit
build. Code using lightning must know about wordsize and the
jit generation limitations, also, this way it generates a
compile time failure, not a runtime assertion.
* include/lightning/jit_x86.h: Correct typo in macro name.
* lib/jit_arm.c, lib/jit_arm-cpu.c, lib/jit_mips.c,
lib/jit_mips-cpu.c, lib/jit_ppc.c, lib/jit_ppc-cpu.c,
lib/jit_x86.c, lib/jit_x86-cpu.c: Correct wrong code to get
current jit function pointer.
* lib/lightning.c: Move call to the simplify() optimization
to after register liveness is known. Previous code did work
by accident but now with proper test cases the problem was
noticed.
* lib/jit_disasm.c: Always cast bfd_vma to long long when
passing it as printf argument.
2012-12-02 Paulo Andrade <pcpa@gnu.org>
* tests/Makefile.am, tests/3to2.c, tests/3to2.ok, tests/add.c,
tests/add.ok, tests/allocai.c, tests/allocai.ok, tests/bp.c,
tests/bp.ok, tests/divi.c, tests/divi.ok, tests/fib.c, tests/fib.ok,
tests/fibdelay.c, tests/fibdelay.ok, tests/fibit.c, tests/fibit.ok,
tests/funcfp.c, tests/funcfp.ok, tests/incr.c, tests/incr.ok,
tests/ldst.c, tests/ldst.ok, tests/ldxi.c, tests/ldxi.ok,
tests/modi.c, tests/modi.ok, tests/movi.c, tests/movi.ok,
tests/printf.c, tests/printf.ok, tests/printf2.c, tests/printf2.ok,
tests/ret.c, tests/ret.ok, tests/rpn.c, tests/rpn.ok, tests/rpnfp.c,
tests/rpnfp.ok, tests/sete.c, tests/sete.ok, tests/testfp.c,
tests/testfp.ok, tests-run-test: Removed previous test suite, in
favor of a newer one in the check subdirectory.
* check/3to2.ok, check/3to2.tst, check/add.ok, check/add.tst,
check/allocai.ok, check/allocai.tst, check/bp.ok, check/bp.tst,
check/divi.ok, check/divi.tst, check/fib.ok, check/fib.tst:
New sample input for the new test program, loosely matching
several of the previous test cases.
* check/Makefile.am: New test suite makefile.
* check/check.sh, check/run-test: New wrapper files for the
new test suite.
* check/lightning.c: New file. The main driver of the new test
suite, that compiles to a parser of a very simple assembly like
language, generates jit and executes it.
* check/all.tst: New file. A generic debug and sample test file
with a directive to prevent it from being executed, and useful to
read disassembly of all possible instructions, using a fixed set
of registers.
* include/Makefile.am, include/lightning.h,
include/lightning/Makefile.am, include/lightning/jit_arm.h,
include/lightning/jit_mips.h, include/lightning/jit_ppc.h,
include/lightning/jit_private.h, include/lightning/jit_x86.h,
lib/Makefile.am, lib/jit_disasm.c, lib/jit_print.c,
lib/jit_x86-cpu.c, lib/jit_x86-sse.c, lib/jit_x86-x87.c,
lib/jit_x86.c, lib/lightning.c: New files. These files are
written from scratch, only by <pcpa@gnu.org>, and have now
copyright assignment to the FSF. This is the core of the new
lightning rework. Previously it was integrated in code with
a garbage collector and several custom types like vectors and
hash tables, so this first code merge with lightning converts
that code into a library extracting only the jit bits, and at
first only for x86_64 GNU/Linux.
* lightning.h, m4/lightning.m4: Removed. These are no longer
required in the new lightning code.
.gitignore, Makefile.am, configure.ac: Update for the new
lightning code.
