gh-mirror/mikepaul-LuaJIT
1
0
Fork 0
mirror of https://github.com/LuaJIT/LuaJIT.git synced 2025-02-08 07:34:07 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity
1,256 Commits 3 Branches 28 Tags 16 MiB
653facd0d5
Commit Graph

81 Commits

Author SHA1 Message Date
Mike Pall
f2479a9654 Minor fixes to FOLD rules. 2012-08-20 18:45:50 +02:00
Mike Pall
0dd175d9e7 Fix last commit. 2012-08-15 21:08:00 +02:00
Mike Pall
ff0a1f3f4e FOLD x / 2^k ==> x * 2^-k. 2012-08-15 16:17:34 +02:00
Mike Pall
e8af6e9da4 Don't treat all constified cdata content as constant. 2012-07-24 14:56:29 +02:00
Mike Pall
4d9c29a78c FFI: Box all accessed or returned enums. 2012-07-17 22:22:07 +02:00
Mike Pall
2139c6791f Fix FOLD rule (i-j)-i => 0-j. 
Thanks to Thomas Schilling.
 2012-07-17 21:47:45 +02:00
Mike Pall
c172e7aa30 Workaround for MSVC conversion bug (double -> uint32_t -> int32_t). 2012-07-17 21:45:49 +02:00
Mike Pall
17d3fc47f3 Avoid pesky compiler warnings about C++ keywords (eh?). 2012-07-03 13:19:32 +02:00
Mike Pall
a9bf455b96 Add HREFK forwarding. Eliminate HREFK guard for TDUP refs. 2012-06-28 15:10:52 +02:00
Mike Pall
5032e6d87f FFI: Use CNEWI/FLOAD to box/access ffi.new("int", x) (for varargs). 2012-06-04 15:39:16 +02:00
Mike Pall
10ef109eef Bump copyright date to 2012. 2012-01-23 22:42:42 +01:00
Mike Pall
455917fa0a Add more FOLD rules for shift + BAND simplifications. 2011-10-25 21:16:27 +02:00
Mike Pall
aa502203a6 Add missing FOLD rule for CONV.u64.int + sign-extension. 2011-10-22 02:03:07 +02:00
Mike Pall
a39aac0445 Fix FOLD rule for CONV.flt.num(CONV.num.flt(x)) => x. 2011-10-19 01:52:50 +02:00
Mike Pall
270de2d6cc Add missing FOLD rule for integer NEG of constants. 2011-10-17 17:21:09 +02:00
Mike Pall
f34ac81e3a Remove unneeded PHI barrier for reassociation of duplicate ops. 2011-10-10 20:35:42 +02:00
Mike Pall
8c5935000d Add narrowing of modulo operator. 2011-06-03 18:48:02 +02:00
Mike Pall
fff2fb31f9 ARM: Don't generate unaligned XLOADs. 2011-06-02 01:46:23 +02:00
Mike Pall
7683f39786 ARM: Fix math.ldexp() recording and folding. 2011-05-27 02:03:02 +02:00
Mike Pall
185554b682 ARM: Unify rotates to simplify assembler backend. 2011-05-23 20:20:34 +02:00
Mike Pall
28e87d33e9 Use common helper functions for unaligned loads. 2011-05-09 18:09:29 +02:00
Mike Pall
89022b4c3e Use lj_vm_tobit() on targets without FPU. 2011-04-10 16:58:22 +02:00
Mike Pall
3acd4892e9 Add FOLD rule for FLOAT->NUM->FLOAT conversions. 2011-04-05 16:31:48 +02:00
Mike Pall
dcbae09b1d DUALNUM: Narrow unary minus. 2011-03-15 18:14:18 +01:00
Mike Pall
7fb1075903 DUALNUM: Add integer variant of MIN/MAX. 2011-03-11 23:28:46 +01:00
Mike Pall
889368e921 Get rid of the remaining silly cast macros from Lua. 2011-03-10 02:13:43 +01:00
Mike Pall
bfce3c1127 DUALNUM: Handle integer type in JIT compiler. 2011-03-10 01:57:24 +01:00
Mike Pall
3f26e3a89d FFI: Fix and optimize recording of cdata[cdata]. 2011-03-10 01:41:58 +01:00
Mike Pall
54978e481b DUALNUM: Make overflow guards weak. Add IR_USE and IR_MULOV. 2011-03-07 20:21:38 +01:00
Mike Pall
e1aa8d0d97 FFI: Fix compiled ffi.string() semantics. 2011-02-28 19:47:51 +01:00
Mike Pall
97d0874ff5 Add IR_XBAR, a barrier against XLOAD/XSTORE optimizations. 2011-02-07 21:59:25 +01:00
Mike Pall
4c74f70464 Strength-reduce 32 to 64 bit widening for XLOAD U8/U16 inputs. 2011-02-06 00:45:39 +01:00
Mike Pall
26fadcd25b FFI: Record simple C function calls. 
Only handles cdecl and fixarg C functions.
Doesn't handle pass-by-value aggregates.
Doesn't handle 64 bit args/returns on 32 bit CPUs.
 2011-02-05 21:50:15 +01:00
Mike Pall
14f5103764 FFI: Record C library namespace lookups. 2011-02-05 14:16:16 +01:00
Mike Pall
1abf5d4d26 FFI: Disable MUL => BSHL FOLD rule on 32 bit. 2011-02-03 20:10:13 +01:00
Mike Pall
df65b8b419 FFI: Rename IR_CNEWP to IR_CNEWI and use it to box 64 bit integers. 
Generates smaller IR and DCE eliminates many intermediate boxes.
Needs allocation sinking to eliminate the boxes kept alive by PHIs.
 2011-02-03 04:22:27 +01:00
Mike Pall
1027018b21 Rename IR_POWI to IR_POW. 2011-02-02 21:33:11 +01:00
Mike Pall
433fd18f6d FFI: Add basic FOLD rules for 64 bit integer DIV, MOD and POWI. 2011-02-02 21:25:43 +01:00
Mike Pall
b613216efc Add SPLIT pass to split 64 bit IR instructions for 32 bit CPUs. 
Add generic HIOP instruction for extra backend functionality.
Add support for HIOP to x86 backend.
Use POWI for 64 bit integer x^k, too.
POWI is lowered to a call by SPLIT or the x64 backend.
 2011-02-02 02:29:37 +01:00
Mike Pall
6fee0002b9 Fix 64 bit case of (SUB x x) and (BXOR x x) FOLD rules. 2011-01-28 02:17:35 +01:00
Mike Pall
925050fe3f Differentiate between IR_KPTR and IR_KKPTR. 
IR_KPTR holds a const pointer to possibly non-const content.
IR_KKPTR holds a const pointer to definitely const content.
Note that only content known by the VM to be const qualifies.
Content tagged as const by users (e.g. const char *) doesn't.
 2011-01-19 00:40:03 +01:00
Mike Pall
59643c86fb Add FOLD rule for CONV.num.u32 KINT. 2011-01-13 17:18:08 +01:00
Mike Pall
06f99fc3df Bump copyright date to 2011. 2011-01-09 17:12:53 +01:00
Mike Pall
b2c74c52cc FFI: Add support for cdata constants to IR. 2011-01-02 22:21:10 +01:00
Mike Pall
783dbd335f Refactoring of conversion ops, part 5: drop TOINT/TOI64/TONUM. 2010-12-31 04:09:32 +01:00
Mike Pall
1716540c55 Refactoring of conversion ops, part 4: use CONV instead of TOINT/TONUM. 
Also narrow CONV.int.num and CONV.i64.num.
 2010-12-31 03:56:30 +01:00
Mike Pall
1f26961092 Refactoring of conversion ops, part 3: add FOLD rules for IR_CONV. 2010-12-31 03:47:30 +01:00
Mike Pall
65b194a2f8 Copy destination type for CONV from ir->t to op2, too. 2010-12-31 01:00:54 +01:00
Mike Pall
2ca2de7f0e Refactoring of conversion ops, part 1: add IR_CONV. 2010-12-30 00:35:32 +01:00
Mike Pall
255c6e8c87 FFI: Drop IR_CNEWI. Add IR_CNEWP only for pointers/refs. 2010-12-17 17:20:04 +01:00