Annotating JITed code
[
Date Prev][
Date Next][
Thread Prev][
Thread Next]
[
Date Index]
[
Thread Index]
- Subject: Annotating JITed code
- From: Geoff Leyland <geoff_leyland@...>
- Date: Mon, 5 Sep 2011 08:16:55 +1200
Hi List,
luatrace [1] now includes a -jannotate option for LuaJIT. -jannotate offers less information than -jdump, but formats it differently in manner that I find easier to comprehend.
First, it tells you how many traces were compiled, how many were aborted and why (sorry, this will look awful if you're not reading your mail in a monospaced font):
Trace Status Traces Bytecodes Lines
------------ ------ --------- -----
Success 132 ( 85%) 1956 ( 89%) 540 ( 87%)
NYI: bytecode CAT 4 ( 2%) 112 ( 5%) 32 ( 5%)
NYI: FastFunc string.match 8 ( 5%) 45 ( 2%) 13 ( 2%)
NYI: unsupported C type conversion 1 ( 0%) 39 ( 1%) 12 ( 1%)
inner loop in root trace 1 ( 0%) 16 ( 0%) 5 ( 0%)
NYI: FastFunc os.clock 1 ( 0%) 11 ( 0%) 5 ( 0%)
NYI: return to lower frame 2 ( 1%) 8 ( 0%) 3 ( 0%)
leaving loop in root trace 4 ( 2%) 5 ( 0%) 4 ( 0%)
NYI: FastFunc string.upper 1 ( 0%) 4 ( 0%) 1 ( 0%)
error thrown or hook called during recording 1 ( 0%) 0 ( 0%) 0 ( 0%)
-------------------------------------------- --------------- --------------- ---------------
Total 155 (100%) 2196 (100%) 615 (100%)
============================================ =============== =============== ===============
then the lines that aborted the traces:
Trace Status Traces Bytecodes Lines Line
------------ ------ --------- ----- ----
Success 132 ( 85%) 1956 ( 89%) 540 ( 87%)
./luasimplex/mps.lua:189 (NYI: bytecode CAT) 3 ( 1%) 68 ( 3%) 24 ( 3%) local name = r.name.."_SLACK"
./luasimplex/rsm.lua:223 (NYI: bytecode CAT) 1 ( 0%) 44 ( 2%) 8 ( 1%) M.variable_names[k] = M.constraint_names[i].."_ARTIFICIAL"
./luasimplex.lua:85 (NYI: unsupported C type conversion) 1 ( 0%) 39 ( 1%) 12 ( 1%) M.indexes = luasimplex.iarray(nonzeroes)
...
and finally shows you the bytecode of the traces side-by-side with your source code (be careful here - bytecode order isn't always the same as source order. I've tried to prefer source order):
Trace #5 (2 lines, 10 bytecodes, 1 attempts)
| ./luasimplex/rsm.lua:210-211
0025 FORL 14 => 0019 | 210 | for j = row_starts[i], row_starts[i+1]-1 do
0019 TGETV 18 6 17 | 211 | z = z - elements[j] * I.x[indexes[j]]
0000 . . FUNCC ; ffi.meta.__index | . |
0020 TGETS 19 1 6 ; "x" | . |
0021 TGETV 20 5 17 | . |
0000 . . FUNCC ; ffi.meta.__index | . |
0022 TGETV 19 19 20 | . |
0000 . . FUNCC ; ffi.meta.__index | . |
0023 MULVV 18 18 19 | . |
0024 SUBVV 13 13 18 | . |
Cheers,
Geoff
[1] https://github.com/geoffleyland/luatrace