Commit Graph

72 Commits

Author SHA1 Message Date
Mark Shannon 0e71a295e9
GH-113710: Add a "globals to constants" pass (GH-114592)
Converts specializations of `LOAD_GLOBAL` into constants during tier 2 optimization.
2024-02-02 12:14:34 +00:00
Brandt Bucher f6d9e5926b
GH-113464: Add a JIT backend for tier 2 (GH-113465)
Add an option (--enable-experimental-jit for configure-based builds
or --experimental-jit for PCbuild-based ones) to build an
*experimental* just-in-time compiler, based on copy-and-patch (https://fredrikbk.com/publications/copy-and-patch.pdf).

See Tools/jit/README.md for more information on how to install the required build-time tooling.
2024-01-28 18:48:48 -08:00
Mark Shannon 981d172f7f
GH-112354: `END_FOR` instruction to only pop one value. (GH-114247)
* Compiler emits END_FOR; POP_TOP instead of END_FOR. To support tier 2 side exits in loops.
2024-01-24 15:10:17 +00:00
Mark Shannon 384429d1c0
GH-113710: Add a tier 2 peephole optimization pass. (GH-114487)
* Convert _LOAD_CONST to inline versions

* Remove PEP 523 checks
2024-01-24 12:08:31 +00:00
Mark Shannon ac10947ba7
GH-112354: `_GUARD_IS_TRUE_POP` side-exits to target the next instruction, not themselves. (GH-114078) 2024-01-15 11:41:06 +00:00
Brandt Bucher 30e6cbdba2
GH-113860: Get rid of `_PyUOpExecutorObject` (GH-113954) 2024-01-12 11:58:23 +00:00
Mark Shannon 55824d01f8
GH-113853: Guarantee forward progress in executors (GH-113854) 2024-01-11 18:20:42 +00:00
Mark Shannon a0c9cf9456
GH-113860: All executors are now defined in terms of micro ops. Convert counter executor to use uops. (GH-113864) 2024-01-10 15:44:34 +00:00
Guido van Rossum 65f8eb7119
Fix opcode name printing in debug mode (#113870)
Fix a few places where the lltrace debug output printed ``(null)`` instead of an opcode name, because it was calling ``_PyUOpName()`` on a Tier-1 opcode.
2024-01-09 18:18:11 +00:00
Mark Shannon 9a35794fcb
GH-111485: Fix handling of FOR_ITER in Tier 2 (GH-113394) 2023-12-24 10:07:34 -08:00
Mark Shannon e96f26083b
GH-111485: Generate instruction and uop metadata (GH-113287) 2023-12-20 14:27:25 +00:00
Guido van Rossum 7316dfb0eb
gh-112320: Implement on-trace confidence tracking for branches (#112321)
We track the confidence as a scaled int.
2023-12-12 21:43:08 +00:00
Mark Shannon 956023826a
GH-108866: Guarantee forward progress in executors. (GH-113006) 2023-12-12 19:02:24 +00:00
Guido van Rossum e723700190
Rename ...Uop... to ...UOp... (uppercase O) for consistency (#112327)
* Rename _PyUopExecute to _PyUOpExecute (uppercase O) for consistency
* Also rename _PyUopName and _PyUOp_Replacements, and some output strings
2023-11-28 17:10:11 -08:00
Guido van Rossum c4c63211e8
gh-111848: Clean up RESERVE() macro (#112274)
Also avoid compiler warnings about unused 'reserved' variable.
2023-11-20 10:45:42 -08:00
Guido van Rossum 1995955173
gh-106529: Make FOR_ITER a viable uop (#112134)
This uses the new mechanism whereby certain uops
are replaced by others during translation,
using the `_PyUop_Replacements` table.
We further special-case the `_FOR_ITER_TIER_TWO` uop
to update the deoptimization target to point
just past the corresponding `END_FOR` opcode.

Two tiny code cleanups are also part of this PR.
2023-11-20 10:08:53 -08:00
Guido van Rossum 7405745817
Various small improvements to uop debug output (#112218)
- Show uop name in Error/DEOPT messages
- Add target to some messages
- Expose uop_name() as _PyUopName()
2023-11-17 22:25:57 +00:00
Guido van Rossum be0bd54c6b
gh-106529: Cleanups split off gh-112134 (#112214)
- Double max trace size to 256
- Add a dependency on executor_cases.c.h for ceval.o
- Mark `_SPECIALIZE_UNPACK_SEQUENCE` as `TIER_ONE_ONLY`
- Add debug output back showing the optimized trace
- Bunch of cleanups to Tools/cases_generator/
2023-11-17 11:49:42 -08:00
Mark Shannon 4bbb367ba6
GH-111848: Set the IP when de-optimizing (GH-112065)
* Replace jumps with deopts in tier 2

* Fewer special cases of uop names

* Add target field to uop IR

* Remove more redundant SET_IP and _CHECK_VALIDITY micro-ops

* Extend whitelist of non-escaping API functions.
2023-11-15 15:48:58 +00:00
Mark Shannon a519b87958
GH-111848: Convert remaining jumps to deopts into tier 2 code. (GH-112045) 2023-11-14 15:30:33 +00:00
Mark Shannon 34a03e951b
GH-111843: Tier 2 exponential backoff (GH-111850) 2023-11-09 13:49:51 +00:00
Mark Shannon 25c4956488
GH-109369: Exit tier 2 if executor is invalid (GH-111657) 2023-11-09 11:19:51 +00:00
Mark Shannon 06efb60264
GH-111848: Tidy up tier 2 handling of FOR_ITER specialization by using DEOPT_IF instead of jumps. (GH-111849) 2023-11-08 13:31:55 +00:00
Mark Shannon d78c872e0d
GH-111646: Simplify optimizer, by compacting uops when making executor. (GH-111647) 2023-11-06 11:28:52 +00:00
Guido van Rossum 7e135a48d6
gh-111520: Integrate the Tier 2 interpreter in the Tier 1 interpreter (#111428)
- There is no longer a separate Python/executor.c file.
- Conventions in Python/bytecodes.c are slightly different -- don't use `goto error`,
  you must use `GOTO_ERROR(error)` (same for others like `unused_local_error`).
- The `TIER_ONE` and `TIER_TWO` symbols are only valid in the generated (.c.h) files.
- In Lib/test/support/__init__.py, `Py_C_RECURSION_LIMIT` is imported from `_testcapi`.
- On Windows, in debug mode, stack allocation grows from 8MiB to 12MiB.
- **Beware!** This changes the env vars to enable uops and their debugging
  to `PYTHON_UOPS` and `PYTHON_LLTRACE`.
2023-11-01 13:13:02 -07:00
Savannah Ostrowski 4a929d432b
GH-111339: Fix initialization and finalization of static optimizer types (GH-111430) 2023-10-29 13:53:25 -07:00
Irit Katriel 67a91f78e4
gh-109094: replace frame->prev_instr by frame->instr_ptr (#109095) 2023-10-26 13:43:10 +00:00
Mark Shannon 5c9d4497ab
GH-111339: Change `valid` property of executors to `is_valid()` method (GH-111350) 2023-10-26 11:31:51 +01:00
Brandt Bucher e5168ff3f8
GH-109214: _SET_IP before _PUSH_FRAME (but not _POP_FRAME) (GH-111001) 2023-10-24 13:27:42 -07:00
Mark Shannon 52e902ccf0
GH-109369: Add machinery for deoptimizing tier2 executors, both individually and globally. (GH-110384) 2023-10-23 14:49:09 +01:00
Mark Shannon 19b7ead5eb
GH-109214: Convert _SAVE_CURRENT_IP to _SET_IP in tier 2 trace creation. (GH-110755) 2023-10-12 10:34:32 +01:00
Michael Droettboom 9eb2489266
gh-109329: Add stat for "trace too short" (GH-110402) 2023-10-05 16:12:06 +01:00
Michael Droettboom e561e98058
GH-109329: Add tier 2 stats (GH-109913) 2023-10-04 14:52:28 -07:00
Brandt Bucher 6c13e13b13
GH-104584: Don't call executors from JUMP_BACKWARD (GH-109347) 2023-09-13 10:26:50 -07:00
Guido van Rossum fbaf77eb9b
gh-109214: Rename SAVE_IP to _SET_IP, and similar (#109285)
* Rename SAVE_IP to _SET_IP
* Rename EXIT_TRACE to _EXIT_TRACE
* Rename SAVE_CURRENT_IP to _SAVE_CURRENT_IP
* Rename INSERT to _INSERT (This is for Ken Jin's abstract interpreter)
* Rename IS_NONE to _IS_NONE
* Rename JUMP_TO_TOP to _JUMP_TO_TOP
2023-09-11 15:39:19 -07:00
Guido van Rossum bcce5e2718
gh-109039: Branch prediction for Tier 2 interpreter (#109038)
This adds a 16-bit inline cache entry to the conditional branch instructions POP_JUMP_IF_{FALSE,TRUE,NONE,NOT_NONE} and their instrumented variants, which is used to keep track of the branch direction.

Each time we encounter these instructions we shift the cache entry left by one and set the bottom bit to whether we jumped.

Then when it's time to translate such a branch to Tier 2 uops, we use the bit count from the cache entry to decided whether to continue translating the "didn't jump" branch or the "jumped" branch.

The counter is initialized to a pattern of alternating ones and zeros to avoid bias.

The .pyc file magic number is updated. There's a new test, some fixes for existing tests, and a few miscellaneous cleanups.
2023-09-11 18:20:24 +00:00
Brandt Bucher 6971e40c2e
GH-104584: Restore frame->stacktop on optimizer error (GH-108953) 2023-09-06 13:59:50 -07:00
Victor Stinner b298b395e8
gh-108765: Cleanup #include in Python/*.c files (#108977)
Mention one symbol imported by each #include.
2023-09-06 15:56:08 +02:00
Irit Katriel 844f4c2e12
gh-108727: Fix segfault due to missing tp_dealloc definition for CounterOptimizer_Type (GH-108734) 2023-09-01 10:16:09 +01:00
Guido van Rossum 4f22152713
gh-107557: Remove unnecessary SAVE_IP instructions (#108583)
Also remove NOP instructions.

The "stubs" are not optimized in this fashion (their SAVE_IP should always be preserved since it's where to jump next, and they don't contain NOPs by their nature).
2023-08-29 16:51:51 +00:00
Irit Katriel 72119d16a5
gh-105481: remove regen-opcode. Generated _PyOpcode_Caches in regen-cases. (#108367) 2023-08-23 18:39:00 +01:00
Guido van Rossum 61c7249759
gh-106581: Project through calls (#108067)
This finishes the work begun in gh-107760. When, while projecting a superblock, we encounter a call to a short, simple function, the superblock will now enter the function using `_PUSH_FRAME`, continue through it, and leave it using `_POP_FRAME`, and then continue through the original code. Multiple frame pushes and pops are even possible. It is also possible to stop appending to the superblock in the middle of a called function, when running out of space or encountering an unsupported bytecode.
2023-08-17 11:29:58 -07:00
Guido van Rossum dc8fdf5fd5
gh-106581: Split `CALL_PY_EXACT_ARGS` into uops (#107760)
* Split `CALL_PY_EXACT_ARGS` into uops

This is only the first step for doing `CALL` in Tier 2.
The next step involves tracing into the called code object and back.
After that we'll have to do the remaining `CALL` specialization.
Finally we'll have to deal with `KW_NAMES`.

Note: this moves setting `frame->return_offset` directly in front of
`DISPATCH_INLINED()`, to make it easier to move it into `_PUSH_FRAME`.
2023-08-16 16:26:43 -07:00
Ken Jin e28b0dc86d
gh-107557: Setup abstract interpretation (#107847)
Co-authored-by: Guido van Rossum <gvanrossum@users.noreply.github.com>
Co-authored-by: Jules <57632293+juliapoo@users.noreply.github.com>
2023-08-15 18:04:17 +00:00
Guido van Rossum 328d925244
gh-107758: Improvements to lltrace feature (#107757)
- The `dump_stack()` method could call a `__repr__` method implemented in Python,
  causing (infinite) recursion.
  I rewrote it to only print out the values for some fundamental types (`int`, `str`, etc.);
  for everything else it just prints `<type_name @ 0xdeadbeef>`.

- The lltrace-like feature for uops wrote to `stderr`, while the one in `ceval.c` writes to `stdout`;
  I changed the uops to write to stdout as well.
2023-08-07 21:36:25 -07:00
Ivin Lee 4e6fac7fcc
gh-106608: make uop trace variable length (#107531)
Executors are now more like tuples.
2023-08-04 21:10:46 -07:00
Brandt Bucher 5e584eb704
GH-104584: Fix incorrect uoperands (GH-107513) 2023-07-31 21:16:57 +00:00
Brandt Bucher 214a25dd81
GH-104584: Miscellaneous fixes for -Xuops (GH-106908) 2023-07-20 16:35:39 +00:00
Guido van Rossum 8e9a1a0322
gh-106603: Make uop struct a triple (opcode, oparg, operand) (#106794) 2023-07-17 12:12:33 -07:00
Guido van Rossum b2b261ab2a
gh-106529: Generate uops for POP_JUMP_IF_[NOT_]NONE (#106796)
These aren't automatically translated because (ironically)
they are macros deferring to POP_JUMP_IF_{TRUE,FALSE},
which are not viable uops (being manually translated).

The hack is that we emit IS_NONE and then set opcode and
jump to the POP_JUMP_IF_{TRUE,FALSE} translation code.
2023-07-17 10:06:05 -07:00