Icecream95
11b9d3744e
Always emit veneers for non-bl jumps to ARM code
...
It is unlikely for any ARM code to be close enough to not have needed
a veneer, but it is possible, especially if running in a program with
another JIT library.
2020-04-09 22:31:20 +12:00
Icecream95
8045386a45
Return a function pointer from jit_address
...
This will allow supporting ARM code on armv7 without having to change
any users of Lightening.
2020-04-09 20:26:35 +12:00
Icecream95
aacaa6e38c
Add separate functions for veneer patching
2020-04-06 19:52:29 +12:00
Icecream95
297ae99c3f
Add flag bits to the jit_reloc_kind enum
2020-04-06 16:04:57 +12:00
Andy Wingo
909c0077cc
Correctly handle overflow when emitting literal pools
2019-05-27 11:34:13 +02:00
Andy Wingo
0b3393415e
Allow add_pending_literal to return false
...
This avoids invalidating offsets for the reloc.
2019-05-22 17:29:10 +02:00
Andy Wingo
b67c4ed1e1
Avoid recursive pool emit for aarch64; add another overflow check
2019-05-21 16:05:03 +02:00
Andy Wingo
b7f367165f
Various fixes for as-needed emission of literal pool
2019-05-21 15:25:08 +02:00
Andy Wingo
568fdecc86
Fix literal pool emission on armv7
2019-05-21 14:19:48 +02:00
Andy Wingo
0518651bfd
Expose API to bless function pointers
...
On ARMv7, we need to set the low bit to indicate that we're in
thumb-land.
2019-05-21 14:00:04 +02:00
Andy Wingo
ef7abb568a
Fix typo in previous commit, doh
2019-05-21 12:44:00 +02:00
Andy Wingo
d2c15dd6cd
Clear literal pool more thoroughly
2019-05-21 12:38:53 +02:00
Andy Wingo
723f70ee11
Clear missing bit from literal pool
2019-05-21 12:24:20 +02:00
Andy Wingo
0b723c0401
ARMv7 backend passing all tests!
2019-05-20 15:20:33 +02:00
Andy Wingo
7dd18bddd7
ARMv7 backend compiling without warnings
2019-05-20 11:02:21 +02:00
Andy Wingo
1ea34693cb
Port of arm-cpu.c to current lightening
2019-05-19 22:21:50 +02:00
Andy Wingo
9d4185af2b
Remove jit_nop
...
Instead, jit_align will call nop() internally. You can't nop 3 bytes on
most architectures.
2019-05-16 11:40:24 +02:00
Andy Wingo
9638e4f169
Fix a couple bugs related to aarch64 literal pools
2019-05-16 11:03:49 +02:00
Andy Wingo
9c9389ad2b
Fix enter/leave JIT ABI
2019-05-16 10:45:44 +02:00
Andy Wingo
a643f99d68
Fix compilation on aarch64
2019-05-16 10:19:02 +02:00
Andy Wingo
7e0b0be1ba
Fix GPR temporary selection
2019-05-15 16:26:18 +02:00
Andy Wingo
09e71475f4
Allow a backend to have multiple temporary registers
2019-05-15 16:19:33 +02:00
Andy Wingo
5b8262e804
Rework register saving to avoid push/pop
...
Push and pop are not well supported on AArch64, so we might as well just
bump the stack pointer once and fill in by offset.
2019-05-15 15:41:02 +02:00
Andy Wingo
0bfdcc7016
Refactor to add support for constant tables, shifted relocs
2019-05-14 15:53:25 +02:00
Andy Wingo
fc9b474da6
Refactor some bits from x86 to lightening
2019-05-10 14:14:32 +02:00
Andy Wingo
f7080facb4
Refactor to move temp register acquire to core
2019-05-09 16:02:39 +02:00
Andy Wingo
570f361c6c
Remove Sparc support
...
Sadly, this is a dead architecture, without an official Debian port.
Rest in peace!
2019-05-08 21:44:03 +02:00
Andy Wingo
ced3c42dee
Remove hppa support
...
This is a dead architecture without an official Debian port.
2019-05-08 21:44:03 +02:00
Andy Wingo
d51bd7d592
Remove alpha support
...
This is a dead architecture without an official Debian port.
2019-05-08 21:44:01 +02:00
Andy Wingo
80f79fca93
Remove ia64 (itanium) support
...
This is a dead architecture without an official Debian port.
2019-05-08 21:37:40 +02:00
Andy Wingo
2a5d315ab9
Unget temp xpr after mem->mem addend fixup
2019-04-28 16:21:09 +02:00
Andy Wingo
04cd8874de
Stack alignment takes saved return address into account
2019-04-28 12:37:19 +02:00
Andy Wingo
d617315fdb
Fix jit_leave_jit_abi to pop correct registers
...
Also relax stack alignment on 32-bit x86
2019-04-27 22:27:39 +02:00
Andy Wingo
e0cf51e482
Small bug fixes for ia32
2019-04-26 17:41:43 +02:00
Andy Wingo
42bc762d26
Add facility to enter and leave JIT ABI
...
This allows us to save and restore callee-save temporaries, i.e. RBP on
32-bit x86. Otherwise it's a disaster shuffling stack arguments using
temporaries.
2019-04-26 17:28:29 +02:00
Andy Wingo
e9d1567f7f
Use jit_word_t for register-sized values instead of intmax_t
2019-04-26 15:00:29 +02:00
Andy Wingo
bab1f40b5e
Fix lightening to compile on 32-bit x86
2019-04-26 14:44:09 +02:00
Andy Wingo
89fd69fc00
Account for saved return address when locating args
...
Also, fix alignment to be in bytes rather than bits. Oops!
2019-04-26 14:38:29 +02:00
Andy Wingo
ac972e1dba
Inline some internal definitions into backend-specific files
2019-04-26 14:37:35 +02:00
Andy Wingo
04d89a7ce2
Implement reloc shortening, and remove unused functionality
2019-04-26 12:40:44 +02:00
Andy Wingo
dc28ac03c1
Make jit_align_stack and jit_shrink_stack public
2019-04-26 10:04:04 +02:00
Andy Wingo
c54c248bc9
Track frame size to more reliably align stack before calls
2019-04-26 09:56:46 +02:00
Andy Wingo
a3e044c808
Move jit_move_operands etc to common lightening.c file
2019-04-26 09:09:50 +02:00
Andy Wingo
2602f17fb4
Simplify register representation again
...
There's only one flag, "callee-save". Also the regno range is limited
to 0-63 inclusive, to allow for cheap register sets as uint64_t values.
2019-04-25 19:12:55 +02:00
Andy Wingo
ddd66a2f34
Simplify register representation
...
Instead of JIT_R0 being a wrapped index into a table which then gives
the regno and class, just have JIT_R0 be the wrapped regno and class.
2019-04-25 17:03:46 +02:00
Andy Wingo
0be4f7a2a1
Simplify API for loading call arguments
2019-04-22 09:15:03 +02:00
Andy Wingo
1d6feefa9d
Add assert for jit_address
2019-04-04 11:41:51 +02:00
Andy Wingo
bbfb03ac30
Fix implementation of jit_{gpr,fpr}_is_callee_save
2019-04-04 11:08:54 +02:00
Andy Wingo
f9da599ef5
Fix jit_reset to be useful
2019-04-03 17:37:18 +02:00
Andy Wingo
22d06620ee
Allow users to pass custom allocators
2019-04-03 15:25:21 +02:00
