Commit Graph

222 Commits

Author SHA1 Message Date
Mike Pall
1cda54a3ab Add compile errors for 64 bit OS without an equivalent of MAP_32BIT. 2010-03-02 19:41:56 +01:00
Mike Pall
a1bacd8d68 Fix amalgamated build. 2010-03-02 01:48:37 +01:00
Mike Pall
5d0b65519c Enable tracing of recursion.
Now compiles tail-recursion, up-recursion and down-recursion.
Benchmarks vs. Lua: fib 20x, ack 23x, binary-trees 4.7x.
2010-03-01 06:54:35 +01:00
Mike Pall
e7b737aa12 Implement down-recursion. 2010-03-01 06:45:30 +01:00
Mike Pall
69ea553024 Fix pc to line number translation for GDB JIT API. 2010-03-01 06:33:48 +01:00
Mike Pall
7e696124ec Resize stack up to the true limit. 2010-03-01 05:51:04 +01:00
Mike Pall
89c90f0955 Fix error message generation for OOM error. 2010-02-28 22:22:45 +01:00
Mike Pall
8cc50cf6b1 Conditionally compile functions that are unused with JIT disabled. 2010-02-28 21:58:40 +01:00
Mike Pall
2e22d33d9d Enable JIT compiler for x64.
Only works on Linux/x64 and Windows/x64 right now.
Force an x64 build on Linux/x64 with: make CC="gcc -m64"
NYI: handle on-trace OOM errors.
NYI: improve register allocation for x64.
2010-02-28 21:51:01 +01:00
Mike Pall
956065fd04 Add frame info for x64 GDB JIT API. 2010-02-28 17:52:11 +01:00
Mike Pall
2144d55eb8 Save/restore remaining callee-save regs on trace<->interp transitions. 2010-02-27 21:40:40 +01:00
Mike Pall
ec78d3177e Correctly align and free allocated machine code areas.
Bump default mcode area size to 64K for x64.
2010-02-27 20:18:11 +01:00
Mike Pall
52b7651327 Place dynamically generated code near static code on x64. 2010-02-26 18:20:24 +01:00
Mike Pall
a0fbb05bf0 Make penalty PRNG available for general use by compiler. 2010-02-26 17:01:33 +01:00
Mike Pall
3c6cec0846 Add x64 call argument setup. More 32/64 bit cleanups in assembler. 2010-02-25 03:35:07 +01:00
Mike Pall
b95294572c Move SIMD constants to jit_State to keep them in the low 4GB. 2010-02-24 23:17:17 +01:00
Mike Pall
b32f4f4549 Handle negative continuation offsets (WIN64 with debug). 2010-02-24 20:39:41 +01:00
Mike Pall
4e73488985 Fix 64 bit conversion warnings. 2010-02-24 20:02:20 +01:00
Mike Pall
4c9f71be5d Major 32/64 bit cleanups in assembler and exit handling.
Add 64 bit lightuserdata handling. Keep the tagged 64 bit value.
Allocate/save/restore 64 bit spill slots for 64 bit lightuserdata.
Fix code generation for 64 bit loads/stores/moves/compares.
Fix code generation for stack pointer adjustments.
Add fixed spill slot definitions for x64. Reduce reserved spill slots.
Disable STRREF + ADD fusion in 64 bit mode (avoid negative 32 bit ofs).
2010-02-24 07:09:34 +01:00
Mike Pall
e46f4c8a11 Fix 64 bit portability problem in rec_ret(). 2010-02-24 05:29:46 +01:00
Mike Pall
13a9d7e135 Add trace exit helper for x64. Fix trace entry for x64. 2010-02-24 01:32:44 +01:00
Mike Pall
2f3a917f2f Check relative jump distances for x64. 2010-02-24 01:29:11 +01:00
Mike Pall
47da0d8018 Fix type() implementation for x64 lightuserdata. 2010-02-24 01:23:52 +01:00
Mike Pall
857f538eac Fix exit state for 64 bit mode. 2010-02-24 01:18:49 +01:00
Mike Pall
8ae2f9feaa Randomize penalties for aborts and add blacklisting. 2010-02-23 19:41:32 +01:00
Mike Pall
d5c8fe4b90 Mark renamed registers as modified. 2010-02-23 17:22:12 +01:00
Mike Pall
c1362dcac9 Don't eliminate SLOAD restores across RETF.
Move restore-elimination logic into snapshot_slots().
2010-02-23 03:08:49 +01:00
Mike Pall
f751cd1d6f Do not patch parent exit for a stack check. 2010-02-22 21:21:52 +01:00
Mike Pall
659ea9de7b Ensure function and all args have a reference for call recording.
In practice this is only needed after a return to a lower frame.
2010-02-22 17:37:26 +01:00
Mike Pall
3336434aa0 Fix TRef for (dummy) 2nd arg of __len metamethod. 2010-02-22 17:33:39 +01:00
Mike Pall
86494c783d Back out history buffer for tailcall counts.
Use an aggregate counter independent of frame depth.
2010-02-22 16:57:59 +01:00
Mike Pall
19af483166 Add region selection for up-recursion and tail-recursion. 2010-02-22 14:35:47 +01:00
Mike Pall
e76bcd2914 Refactor Lua stack handling in lj_asm.c. 2010-02-21 17:26:21 +01:00
Mike Pall
c1658ddcf1 Eliminate redundant stack checks. Add checks for growing root traces. 2010-02-21 16:47:43 +01:00
Mike Pall
c52da1f2da Rethrow errors from trace exit handling from the right C frame. 2010-02-19 03:13:48 +01:00
Mike Pall
bbe7d818d9 Allow linking to already compiled functions. 2010-02-18 19:37:30 +01:00
Mike Pall
b11eeab906 Use a limited history buffer for tailcall counts while recording. 2010-02-18 19:32:13 +01:00
Mike Pall
65586ca4dc Adapt bytecode patching/unpatching for hot calls. 2010-02-18 17:17:50 +01:00
Mike Pall
59f54be9d3 Update trace recorder infrastructure for hot calls. 2010-02-18 06:07:31 +01:00
Mike Pall
2f300fb66a Fix some uses of tref_isint(). 2010-02-18 03:24:18 +01:00
Mike Pall
2b22b9e710 Drop obsolete shadow frame link stack. 2010-02-18 03:19:46 +01:00
Mike Pall
265a24b978 Add more assertions to compare the recorder state and the VM state. 2010-02-18 02:49:08 +01:00
Mike Pall
00ad80c9e4 Clear frame gaps in recorder to avoid resurrecting previous refs. 2010-02-18 02:45:03 +01:00
Mike Pall
8060f5b531 Just disable JIT compiler for non-SSE2 CPUs instead of aborting. 2010-02-16 23:39:24 +01:00
Mike Pall
c225ee8db4 Split CALL/FUNC recording.
Record __call resolving and specialization for CALL* bytecodes.
Record argument adjustment and fast functions for FUNC* bytecodes.
Avoids all pending/immediate decisions for chained fast functions.
Cleaner semantics for pcall(), xpcall() and __tostring metamethod.
Prerequisite to drop the shadow frame link stack again.
2010-02-16 04:04:16 +01:00
Mike Pall
afa07e0c46 Improve error reporting for traces aborted in non-Lua functions. 2010-02-15 22:44:23 +01:00
Mike Pall
e664945e0b Change record vmevent parameters. Dump function id on FUNC*. 2010-02-15 22:26:33 +01:00
Mike Pall
e1905f498a Use a different marker for hot calls. 2010-02-15 18:04:06 +01:00
Mike Pall
3452bfcf8c Add generic function handling for debug modules.
Don't call record vmevent for non-Lua functions.
2010-02-15 17:36:29 +01:00
Mike Pall
b838cd8dca Minor cleanup of trace event handling. 2010-02-15 16:41:52 +01:00
Mike Pall
1288e04186 Add missing FORI coercions in recorder. 2010-02-15 01:51:41 +01:00
Mike Pall
b6e4fde0dc Improve FOR loop const specialization and integerness checks. 2010-02-15 01:07:30 +01:00
Mike Pall
2a2f8ed6a1 Implement return hooks for Lua functions (zero-cost if disabled). 2010-02-14 20:48:33 +01:00
Mike Pall
8e38231f9e Implement call hooks (zero-cost if disabled). 2010-02-14 17:47:03 +01:00
Mike Pall
c93138b59e Major redesign of function call handling.
Drop call gates. Use function headers, dispatched like bytecodes.
Emit BC_FUNCF/BC_FUNCV bytecode at PC 0 for all Lua functions.
C functions and ASM fast functions get extra bytecodes.
Modify internal calling convention: new base in BASE (formerly in RA).
Can now use better C function wrapper semantics (dynamic on/off).
Prerequisite for call hooks with zero-overhead if disabled.
Prerequisite for compiling recursive calls.
Prerequisite for efficient 32/64 bit prototype guards.
2010-02-13 04:51:56 +01:00
Mike Pall
4f8d7be8ea Extend scope of local vars in debug info beyond final return. 2010-02-12 16:17:42 +01:00
Mike Pall
b2067236c5 Move dispatch tables out of GG_State struct. 2010-02-11 16:21:18 +01:00
Mike Pall
565eb39574 Drop frame clearing in exit handling and JIT compiled code. 2010-02-11 01:50:32 +01:00
Mike Pall
ab90b8fc2b Switch to pre-initialized stacks. Drop frame clearing in interpreter. 2010-02-11 01:21:40 +01:00
Mike Pall
bb0384c366 Replace GCproto reference with bytecode PC in GCfuncL. 2010-02-10 21:45:57 +01:00
Mike Pall
d5a46621f9 Drop bc field in GCproto since the bytecode is colocated. 2010-02-08 20:17:34 +01:00
Mike Pall
2ddea3838d Fix constructor bytecode generation for conditional values. 2010-02-08 16:08:58 +01:00
Mike Pall
f86f9e8eec Reduce whitespace in lj_bcdef.h. 2010-02-08 05:35:18 +01:00
Mike Pall
6194b1c896 Redesign of prototype generation, part 5: colocation of protoype arrays. 2010-02-08 05:30:57 +01:00
Mike Pall
f275a9d7ef Redesign of prototype generation, part 4: late creation of prototype. 2010-02-08 05:29:47 +01:00
Mike Pall
4424027844 Redesign of prototype generation, part 3: bc and lineinfo.
Use a growable, per-chunk bytecode instruction/line stack.
Collect bc/lineinfo for prototype at the end.
2010-02-08 05:28:57 +01:00
Mike Pall
48d93d8c84 Redesign of prototype generation, part 2: late init of chunkname. 2010-02-08 05:27:43 +01:00
Mike Pall
60b5af4422 Redesign of prototype generation, part 1: varinfo and uvname.
Use a growable, per-chunk variable stack.
Collect varinfo/uvname for prototype at the end.
2010-02-08 05:26:52 +01:00
Mike Pall
1307f49137 Major cleanup of bytecode parser. 2010-02-06 08:18:32 +01:00
Mike Pall
1f39cc5c21 Update .gitignore. 2010-02-06 08:18:20 +01:00
Mike Pall
c4dadf1d67 Move bytecode offsets from lj_vm.* to generated header. 2010-02-05 20:15:01 +01:00
Mike Pall
a0914c409b 32/64 bit memory ref cleanup, part 4: GCproto ->varinfo. 2010-02-05 01:35:38 +01:00
Mike Pall
8dcc4364cf 32/64 bit memory ref cleanup, part 3: GCproto ->lineinfo. 2010-02-05 01:16:22 +01:00
Mike Pall
d778982120 32/64 bit memory ref cleanup, part 2: GCproto ->uvname and ->chunkname. 2010-02-05 00:52:21 +01:00
Mike Pall
c8d55e8506 32/64 bit memory ref cleanup, part 1: GCproto ->bc and ->k. 2010-02-05 00:07:32 +01:00
Mike Pall
370c868c31 RETF modifies BASE. Treat it like a store or it gets CSEd. 2010-02-04 21:36:32 +01:00
Mike Pall
5d2690c608 Reset the hotcount table after a JIT off to on transition. 2010-02-04 20:40:00 +01:00
Mike Pall
7256690364 Add shadow frame link stack for trace recorder.
Simplifies snapshots. Prerequisite for pre-call snapshots.
Increases consistency for fast function calls, too.
2010-02-04 03:08:29 +01:00
Mike Pall
78f5e2ffd3 Fix unroll limit checks and frame depth adjustment for pcall. 2010-02-03 16:48:25 +01:00
Mike Pall
318e86c7eb Clean up frame depth checks and loop detection. 2010-02-03 14:55:56 +01:00
Mike Pall
c1c9abeab7 Add extra check to suppress hotcall event during recording. 2010-02-03 14:34:30 +01:00
Mike Pall
305ecbbb84 Another loop formation test must check for return to lower frame. 2010-02-03 14:31:42 +01:00
Mike Pall
61c5a60dc3 Improve coalescing of BASE register in side traces. 2010-02-01 23:32:26 +01:00
Mike Pall
9a682f341d Revise hardcoded inlining in lj_asm.c. Saves 1-2K. 2010-01-30 14:33:08 +01:00
Mike Pall
02e58f5e56 Add support for weak IR references to register allocator.
Spilling a weak ref forces a spill slot, but omits the restore.
Spill slots for snapshot refs override the register, anyway.
Marking snapshot refs weak avoids pointless restores.
2010-01-30 06:50:39 +01:00
Mike Pall
96e8a56260 Loop formation test must check for return to lower frame. 2010-01-29 15:07:40 +01:00
Mike Pall
8681b7330f Compile return to lower frame. Only for Lua frames right now. 2010-01-29 03:32:37 +01:00
Mike Pall
272b2f7368 Followup fix: set maxslot for continuation return. 2010-01-28 04:29:15 +01:00
Mike Pall
2ed0bb7f1e Drop obsolete frame shrinking after continuation return. 2010-01-28 04:20:20 +01:00
Mike Pall
a6565d21fa Fix recording of metamethod result adjustment. 2010-01-28 04:02:32 +01:00
Mike Pall
725da9224d Don't modify jit_State and exit counters while in vmevent.
Fixes crash with hot loop in TEXIT callback which cleared J->parent.
2010-01-28 02:30:12 +01:00
Mike Pall
21d6709271 Avoid reuse of PHI registers, even for duplicate right PHIs. 2010-01-27 22:09:43 +01:00
Mike Pall
09b8f67cd8 Expose compressed snapshot map to reflection API.
Update jit.dump module and restore printing of frame separators.
2010-01-27 20:06:03 +01:00
Mike Pall
3aeb892262 Eliminate IR_FRAME. Replace with KGC and TRef/SnapEntry flags. 2010-01-27 03:50:29 +01:00
Mike Pall
d216cdb2b9 Add frame and continuation flags to TRef and SnapEntry. 2010-01-27 02:17:56 +01:00
Mike Pall
05d67cf566 Add missing check for return to lower frame. 2010-01-27 01:57:15 +01:00
Mike Pall
c8a80fa989 Cleanup types for narrowing stack machine. 2010-01-27 00:39:22 +01:00
Mike Pall
67ca399a30 Compress snapshots using a simple, extensible 1D-compression.
Typically reduces storage overhead for snapshot maps by 60%.
The extensible format is a prerequisite for the next redesign steps:
Eliminate IR_FRAME and implement return-to-lower-frame.
2010-01-26 21:49:04 +01:00
Mike Pall
e058714a2e Add missing eviction in asm_obar(). 2010-01-26 02:56:00 +01:00
Mike Pall
34d84f8836 Fill gaps in frames (caused by metamethod calls) with nil.
Simplifies storing snapshots to stack.
2010-01-26 00:45:30 +01:00