libjit

Commit Graph

Author	SHA1	Message	Date
Aleksey Demakov	604d64b0b8	Merge pull request #22 from M4GNV5/cherry-misc add inversions of OP_BR_IFALSE and OP_BR_ITRUE to _jit_invert_condition	5 years ago
Jakob Löw	1eba11d8d5	add inversions of OP_BR_IFALSE and OP_BR_ITRUE to _jit_invert_condition	5 years ago
Aleksey Demakov	b27000841c	Make all _jit_cache_* functions static and remove the prefix; remove some commented out code	5 years ago
Aleksey Demakov	942c988db1	Fix last ChangeLog entry	5 years ago
Aleksey Demakov	3997a29821	Include jit-dump.h and jit-memory.h from jit.h	5 years ago
Aleksey Demakov	718931dcdb	Merge pull request #21 from tromey/declare-functions Check for functions that are documented but not declared	5 years ago
Tom Tromey	c608b30f38	Check for functions that are documented but not declared This adds some code to "make check" to look for functions that are documented but not declared in a public header. This found a few such missing functions; this is also fixed in this patch. Fixes #20	5 years ago
Aleksey Demakov	77fbfeceaa	Merge pull request #19 from tromey/branch-optimize Optimize a jump-around-branch case	5 years ago
Tom Tromey	6a7e6d9eb3	Add unit test for CFG transformation This adds a unit test for the CFG transformation patch.	5 years ago
Tom Tromey	06149d7834	Optimize a jump-around-branch case In my Emacs JIT I noticed that sometimes libjit would emit a conditional jump to branch around an unconditional jump, like: 7ffff7fe5160: 0f 85 05 00 00 00 jne 0x7ffff7fe516b 7ffff7fe5166: e9 0e 00 00 00 jmpq 0x7ffff7fe5179 7ffff7fe516b: 41 bf 17 00 00 00 mov $0x17,%r15d While this can be worked around by constructing the IL a bit differently, I was curious about the difficulty of fixing this in the libjit optimizer. This patch notices the specific case where a conditional branch simply jumps around an unconditional branch. In this case, the condition is inverted and the extra jump is eliminated.	7 years ago
Tom Tromey	d2aa30874c	Add tests/unit subdirectory This adds a tests/unit subdirectory. This is just a bit of infrastructure; future patches will add actual test cases here.	5 years ago
Aleksey Demakov	5fba3155b0	Merge pull request #18 from tromey/fix-some-warnings Fix some warnings	5 years ago
Tom Tromey	757576f7f6	Fix some warnings This fixes a few simple warnings I ran across while building on x86-64 Fedora 28.	5 years ago
Aleksey Demakov	7016d5478e	Merge pull request #17 from M4GNV5/cherry-misc Misc fixes & improvements	5 years ago
Jakob Löw	39c7c2ea38	add a changelog entry for `54e7090`, `e385fd1` and `f75ded9`	5 years ago
Jakob Löw	f75ded9ad5	in the x86-64 backend use lea, inc and dec where applicable in add and sub opcodes	5 years ago
Jakob Löw	e385fd1101	fix possible SIGSEGV on conversion when converting byte and short values to larger values the conversion becomes a simple copy opcode. The copy opcode however isn't included in the conversion instrinsics array, causing an out of bounds read.	5 years ago
Jakob Löw	54e709009b	add a simple unwinding check which makes the unwinder stop when the next frame is below the current one	6 years ago
Aleksey Demakov	b9cf37655d	Fix out-of-tree builds	5 years ago
Aleksey Demakov	7f7439dc04	Cosmetic changes	5 years ago
Aleksey Demakov	b1b8182852	Remove JITINT64CXX macro, move jit_size_t definition.	5 years ago
Aleksey Demakov	0516c2695d	Merge pull request #15 from M4GNV5/nesting Implementing Nesting	5 years ago
Jakob Löw	d6c0eb9d60	make sure values don't change their address the current implementations of incoming_frame_posn and outgoing_frame_posn tend to break addressable variables, as they simply change a values frame_offset. For incoming_frame_posn this is fine, as it is only used at the start of a function. This commit changes the implementation of outgoing_frame_posn to store the value at the target frame offset.	6 years ago
Jakob Löw	4740b2753e	make sure struct return pointer arguments are always before parent frame pointer arguments	6 years ago
Jakob Löw	e6dcb55fce	remove all calls to _jit_gen_fix_value from jit-insn.c, re-adding the import opcode, which is transformed into an add_relative in jit-compile.c	6 years ago
Jakob Löw	5281baec45	merge the latest commits from upstream	6 years ago
Aleksey Demakov	554c9f5c75	Fix _jit_load_opcode() invocation in interpreter (thanks Jakob Löw <jakob@m4gnus.de>)	6 years ago
Ken Brown	d09676d2f3	Remove reference to __CYGWIN32__	6 years ago
Ken Brown	22d66c3c4d	Don't use Win32 API on Cygwin	6 years ago
Ken Brown	59340183e8	Remove obsolete compiler option on Cygwin	6 years ago
Aleksey Demakov	a79b4c421d	gitignore minimal binary	6 years ago
Jakob Löw	1c5c925b47	add a chanelog entry for the new nesting system	6 years ago
Jakob Löw	ec029283de	fix importing parameter in the interpreter as they reside in a custom memory location this commit also makes sure that the frame offset of parameters is available when the function is built, so they dont get assigned a different offset when being imported.	6 years ago
Jakob Löw	00f8f12fb8	Add a special OP for retrieving the frame pointer previousely this was handeled by incoming_reg however the register allocator doesn't allow to move registers from different reg classes. Nesting is tested on x86-86, x86 and the interpreter.	6 years ago
Aleksey Demakov	7da475e60b	jit-rules-x86-64.c: dont declare var in for initializer	6 years ago
Aleksey Demakov	d109997f50	Fix a typo	6 years ago
Aleksey Demakov	92a44c67e7	fix a warning	6 years ago
Aleksey Demakov	1f4e7b1d86	Fix libjitdynamic.p.in (thanks Eli Zaretskii).	6 years ago
Jakob Löw	217a496c9f	implement nesting in the interpreter backend	6 years ago
Aleksey Demakov	9ad05b4c05	Bump version to 0.1.4	6 years ago
Aleksey Demakov	ad84940ad9	edit TODO	6 years ago
Aleksey Demakov	d436212b0c	Fix jit_vmem_commit() return value	6 years ago
Jakob Löw	cffe11af0c	retrieve the frame pointer by using incoming_reg rather than incoming_frame	6 years ago
Jakob Löw	13eaac6ce0	add an instruction for calling a nested function via a function and a frame pointer currently nested functions can only be called using jit_insn_call. This requires the caller to have the jit_function_t object available. The new function jit_insn_call_nested_indirect is similar to jit_insn_call_indirect - allowing to call a function using a jit_value_t - but it has an additional parameter for passing a pointer to the parent frame of the callee. Using jit_insn_get_parent_frame_pointer_of and jit_function_to_closure one can now create both the closure and the parent frame and pass it around together as jit values, calling them does not require the caller to know the actual jit_function_t he is calling. This makes it possible to implement what C# calls delegates.	7 years ago
Jakob Löw	ba1ebf2746	fix some small bugs with importing nested values	7 years ago
Jakob Löw	87735cf139	remove the unused opcodes setup_for_sibling, setup_for_nested and import	7 years ago
Jakob Löw	25c88f5e1c	automatically find and pass parent frame pointers when calling nested functions when calling a nested function the function needs a pointer to the parent frame to import values. With the new nested importing system the work of importing values was taken away from the backends, so this commit takes away setting up calls to nested functions from the backend. Instead this is done in jit-rules.c (and architecture specific versions of jit-rule.c).	7 years ago
Jakob Löw	780b4b816a	Fix some issues with importing values The previous commit intrduced a couple of bugs. While importing values from the parent worked importing values from a higher ancestor or calling siblings was broken. This commit fixes these issues	7 years ago
Jakob Löw	a40e0bc422	rework how nested values are accessed The current way on how importing values from ancestors works mainly aims at hiding how it works from the outside. The main problem was that it required a complex implementation in each backend which only one of the backends (x86) implemented. This commit changes accessing nested values, making it completely frontend based. Furthermore it allows the library users to retrieve the frame pointer of a parent and pass it in any way they like instead of forcing them to use jit_insn_call. This means the new way allows implementing things such as lambdas as function pointer (closure) together with a pointer to the parent frame and then call this lambda using jit_insn_call_indirect.	7 years ago
Aleksey Demakov	825a982ce2	Undo incorrect commit out of the last merge	7 years ago

1 2 3 4 5 ...

718 Commits (master) All Branches Search

718 Commits (master)

All Branches