bpo-45367: Specialize BINARY_MULTIPLY#28727

sweeneyde · 2021-10-05T03:45:01Z

https://bugs.python.org/issue45367

bedevere-bot · 2021-10-05T04:39:09Z

🤖 New build scheduled with the buildbot fleet by @sweeneyde for commit 300a0ca 🤖

If you want to schedule another build, you need to add the ":hammer: test-with-buildbots" label again.

sweeneyde · 2021-10-05T05:03:09Z

The failing Tests / Windows (x86) check already has an open ticket at bpo-41682.

sweeneyde · 2021-10-05T05:08:13Z

On -m test test_long test_float test_math test_list test_dict I got

Specialization stats: binary_multiply.specialization_success : 87 binary_multiply.specialization_failure : 1635 binary_multiply.hit : 1332363 binary_multiply.deferred : 103712 binary_multiply.miss : 15 binary_multiply.deopt : 3 binary_multiply.unquickened : 482

and on bm_nbody.py I got

Specialization stats: binary_multiply.specialization_success : 38 binary_multiply.specialization_failure : 8 binary_multiply.hit : 10800672 binary_multiply.deferred : 336 binary_multiply.miss : 0 binary_multiply.deopt : 0 binary_multiply.unquickened : 237

Include/internal/pycore_code.h

markshannon · 2021-10-05T10:11:58Z

Looks promising.

Apologies for the merge conflicts from #28723.
Once you've fixed those and the above comment, I will benchmark it.

Fidget-Spinner

Thanks Dennis, this looks good and I think it's almost ready to merge :). Just needs some stable benchmarks on non-Windows platforms.

Python/ceval.c

Python/specialize.c

markshannon · 2021-10-05T13:48:07Z

Python/specialize.c

+ goto success;
+ }
+else{
+SPECIALIZATION_FAIL(BINARY_MULTIPLY, SPEC_FAIL_OTHER);


Maybe this should be refined further. In another PR, though.

I agree. I don't think it will touch pyperformance, but it seems that in our own test suite almost 10% aren't hits #28727 (comment).
~~This probably varies wildly, I'd imagine test_string has lots of str * num too.~~
Oh gosh I just realized I'd read the entire PR as BINARY_ADD not BINARY_MULTIPLY. Please ignore my delusional ramblings.

Most of the non-hits are deferred, not specialization failure, so I think that's because the nature of a lot of test code is to only be run once, without many tight loops.

Thanks for pointing that out Dennis. I can't believe I missed that :). That means nearly 100% actual specialization on hot code. Hooray!
Off-topic: I've recently wondered if pyperformance is somewhat out of touch (no offence intended to its contributors). Many of its benchmarks (nbody, nqueens, etc.) have been found by the JS folks to not be realistic. The Pyston benchmark suite seems way more comprehensive.

"deferred" means that the ADAPTIVE instruction is counting down until its next specialization attempt.
This only happens after a specialization failure.
Look at the numbers. deferred ≈ sum(specialization-failures) * ADAPTIVE_CACHE_BACKOFF

markshannon · 2021-10-05T15:11:38Z

Significant but small speedup

It does show speedups for the benchmarks we would expect, but there are significant slowdowns for some other benchmarks; notably the pickle benchmarks.

Fidget-Spinner · 2021-10-05T15:27:25Z

Significant but small speedup

I've personally disregarded anything from pyperformance in the range of +-1.05x. That said, I wholly believe nbody sped up, I just don't trust pickle slowing down. There's no BINARY_MULTIPLY there and it's a call to a C function (unless we snagged some PGO problem again :( ). I will try to bench soon.

sweeneyde · 2021-10-05T18:29:26Z

I ran -m test test_dict test_math test_long test_heapq test_bisect test_list test_dict with some more detailed specialization failure stats.

 binary_multiply.specialization_success : 96 binary_multiply.specialization_failure : 1650 binary_multiply.hit : 1584206 binary_multiply.deferred : 104756 binary_multiply.miss : 15 binary_multiply.deopt : 3 binary_multiply.unquickened : 485 binary_multiply.specialization_failure_kinds[12] : 56 # different types, other binary_multiply.specialization_failure_kinds[13] : 8 # str * int binary_multiply.specialization_failure_kinds[14] : 0 # int * str binary_multiply.specialization_failure_kinds[15] : 1459 # float * int binary_multiply.specialization_failure_kinds[16] : 135 # int * float

I'm guessing specializing int/float mixtures would be beneficial. I wonder if it's better to have a single slightly-more-complex opcode or add more opcodes. I was thinking something like

TARGET(BINARY_MULTIPLY_FLOAT){PyObject*left=SECOND(); PyObject*right=TOP(); doubledleft, dright; if (PyFloat_CheckExact(left)){dleft= ((PyFloatObject*)left)->ob_fval; if (PyFloat_CheckExact(right)){dright= ((PyFloatObject*)right)->ob_fval} elseif (PyLong_CheckExact(right) &&IS_MEDIUM_VALUE(right)){dright= (double)medium_value(right)} else{DEOPT_IF(1, BINARY_MULTIPLY)} } elseif (PyLong_CheckExact(left) &&IS_MEDIUM_VALUE(left)){DEOPT_IF(!PyFloat_CheckExact(right), BINARY_MULTIPLY); dleft= (double)medium_value(left); dright= ((PyFloatObject*)right)->ob_fval} else{DEOPT_IF(1, BINARY_MULTIPLY)} STAT_INC(BINARY_MULTIPLY, hit); record_hit_inline(next_instr, oparg); PyObject*prod=PyFloat_FromDouble(dleft*dright); SET_SECOND(prod); Py_DECREF(right); Py_DECREF(left); STACK_SHRINK(1); if (prod==NULL){goto error} DISPATCH()}

sweeneyde · 2021-10-05T19:11:28Z

The most recent change gets the results of -m test test_dict test_math test_long test_heapq test_bisect test_list test_dict down to

 binary_multiply.specialization_success : 108 binary_multiply.specialization_failure : 100 binary_multiply.hit : 1679780 binary_multiply.deferred : 5962 binary_multiply.miss : 15 binary_multiply.deopt : 3 binary_multiply.unquickened : 485

markshannon · 2021-10-06T08:20:40Z

What types are you specializing for, and why?

Please read https://github.com/python/cpython/blob/main/Python/adaptive.md, specifically https://github.com/python/cpython/blob/main/Python/adaptive.md#gathering-data

sweeneyde · 2021-10-06T19:48:59Z

My thinking was that two DEOPT_IFs is the same number of branches as two if-statements in the float-float case, and if we're going to deoptimize we might as well check if it's actually an int that we're multiplying by so we can quickly recover and avoid deoptimizing. It seems common enough for an int to slip into places where floats can be, or to have multiplications of the form pi * r * r where r is an integer. Or as you mentioned on the bpo issue, there might be an instruction that is int*float for the first iteration but becomes float*float on subsequent iterations.

To clarify, are you advising that the increased complexity of the opcode is not worth the increased percentage of instructions specialized? Is the solution to

Use something like commit c64540a with only strictly float*float and int*int specialized
Use commit aea424e with float*(int_or_float) rolled into one opcode
Add separate specializations for float*int and int*float*?

I am struggling to run the whole pyperformance suite at once, but as an example, bm_spectral_norm goes from

commit c64540af47ff336b074611d33e130e06cc06e5c2 binary_multiply.specialization_success : 20 binary_multiply.specialization_failure : 1062 binary_multiply.hit : 5341671 binary_multiply.deferred : 67921 binary_multiply.miss : 24 binary_multiply.deopt : 3 binary_multiply.unquickened : 111

to

commit aea424ead30756f11d29e10643ccbfda82567676 binary_multiply.specialization_success : 17 binary_multiply.specialization_failure : 9 binary_multiply.hit : 5409279 binary_multiply.deferred : 337 binary_multiply.miss : 0 binary_multiply.deopt : 0 binary_multiply.unquickened : 111

when incorporating int*float and float*int into BINARY_MULTIPLY_FLOAT.

I suppose we could gather specialization stats for what happens if int*float and float*int are added, and if those stats are just as good then the first-iteration-versus-the-rest type instability is not an issue.

Fidget-Spinner · 2021-10-06T22:33:46Z

nbody sped up, everything else looks in the realm of noise for pyperformance: https://gist.github.com/Fidget-Spinner/6fd3149fc82497d028b02046765ba8d8

markshannon · 2021-10-07T10:56:07Z

To clarify, are you advising that the increased complexity of the opcode is not worth the increased percentage of instructions specialized? Is the solution to
* Use something like commit c64540a with only strictly `float*float` and `int*int` specialized 

Yes.

* Use commit [aea424e](https://github.com/python/cpython/commit/aea424ead30756f11d29e10643ccbfda82567676) with `float*(int_or_float)` rolled into one opcode * Add separate specializations for `float*int` and `int*float*`?

I don't think there is sufficient evidence that these are useful. At least, not yet.

I am struggling to run the whole pyperformance suite at once, but as an example, bm_spectral_norm goes from

commit c64540af47ff336b074611d33e130e06cc06e5c2 binary_multiply.specialization_success : 20 binary_multiply.specialization_failure : 1062 binary_multiply.hit : 5341671 binary_multiply.deferred : 67921 binary_multiply.miss : 24 binary_multiply.deopt : 3 binary_multiply.unquickened : 111

The deferred operations only represent about 1%. Not worth worrying about.

markshannon · 2021-10-07T11:00:48Z

In general, it is not the number of branches that matters, as much as their predictability.
Although there about ~1% deferred, in the numbers for bm_spectral_norm that you give, the number of misses is very low (~4 per million) meaning that the branches are extremely predictable.

sweeneyde · 2021-10-10T08:00:31Z

I am beginning to think this PR may not be worth it at all: even on microbenchmarks, after PGO, there's not that much benefit:

Program:

frompyperfimportRunnerrunner=Runner() runner.timeit( "int: x*x", setup="from itertools import repeat; it = repeat(1, 1_000_000)", stmt="for x in it: x*x") runner.timeit( "float: x*x", setup="from itertools import repeat; it = repeat(1.0, 1_000_000)", stmt="for x in it: x*x") runner.timeit( "int: x*...*x", setup="from itertools import repeat; it = repeat(1, 1_000_000)", stmt="for x in it: x*x*x*x*x*x*x*x*x*x*x*x*x*x*x*x*x*x*x*x*x") runner.timeit( "float: x*...*x", setup="from itertools import repeat; it = repeat(1.0, 1_000_000)", stmt="for x in it: x*x*x*x*x*x*x*x*x*x*x*x*x*x*x*x*x*x*x*x*x")

Results from PGO on WSL:

Benchmark	./floatbench.json	../multiply_worktree/floatbench.json
int: x...x	205 ms	174 ms: 1.17x faster
float: x...x	238 ms	179 ms: 1.33x faster
Geometric mean	(ref)	1.12x faster

Benchmark hidden because not significant (2): int: x*x, float: x*x

Results from pgo using build.bat (MSVC):

Benchmark	./floatbench.json	../multiply/floatbench.json
int: x*x	28.5 ns	27.7 ns: 1.03x faster
float: x*x	28.4 ns	27.9 ns: 1.02x faster
int: x...x	262 ms	220 ms: 1.19x faster
float: x...x	260 ms	253 ms: 1.03x faster
Geometric mean	(ref)	1.06x faster

markshannon · 2021-10-11T14:41:55Z

@sweeneyde
What sort of speedups were you expecting?

The speedups are worthwhile for less than 100 additional lines of code, IMO.

The speedups on the micro benchmark shows that it works.
The overall speedup, while modest, shows that it works in a wider context.

Don't forget that BINARY_MULTIPLY only represents a fairly small fraction of the total executed for the benchmark suite as a whole. This PR was never going to speed things up that much, but it is still a useful contribution.

sweeneyde · 2021-10-11T20:15:37Z

Okay, that makes sense, thanks. Re-opened.

bedevere-bot · 2021-10-13T16:39:58Z

🤖 New build scheduled with the buildbot fleet by @markshannon for commit 31c3bb9 🤖

If you want to schedule another build, you need to add the ":hammer: test-with-buildbots" label again.

markshannon · 2021-10-14T14:53:28Z

Buildbot failures seem to be the usual suspects.

bedevere-bot · 2021-10-14T15:33:45Z

⚠️⚠️⚠️ Buildbot failure ⚠️⚠️⚠️

Hi! The buildbot AMD64 FreeBSD Shared 3.x has failed when building commit 3b3d30e.

What do you need to do:

Don't panic.
Check the buildbot page in the devguide if you don't know what the buildbots are or how they work.
Go to the page of the buildbot that failed (https://buildbot.python.org/all/#builders/483/builds/974) and take a look at the build logs.
Check if the failure is related to this commit (3b3d30e) or if it is a false positive.
If the failure is related to this commit, please, reflect that on the issue and make a new Pull Request with a fix.

You can take a look at the buildbot page here:

https://buildbot.python.org/all/#builders/483/builds/974

Failed tests:

test_concurrent_futures
test_gdb

Failed subtests:

test_pyup_command - test.test_gdb.StackNavigationTests
test_idle_process_reuse_multiple - test.test_concurrent_futures.ProcessPoolForkserverProcessPoolExecutorTest
test_basic_command - test.test_gdb.PyLocalsTests
test_locals_after_up - test.test_gdb.PyLocalsTests
test_basic_command - test.test_gdb.PyPrintTests
test_up_then_down - test.test_gdb.StackNavigationTests
test_bt_full - test.test_gdb.PyBtTests
test_one_abs_arg - test.test_gdb.PyListTests
test_bt - test.test_gdb.PyBtTests
test_idle_process_reuse_one - test.test_concurrent_futures.ProcessPoolForkserverProcessPoolExecutorTest
test_basic_command - test.test_gdb.PyListTests
test_print_after_up - test.test_gdb.PyPrintTests

Summary of the results of the build (if available):

== Tests result: FAILURE then FAILURE ==

406 tests OK.

10 slowest tests:

test_multiprocessing_spawn: 7 min 57 sec
test_tokenize: 7 min 46 sec
test_gdb: 6 min 52 sec
test_concurrent_futures: 5 min 19 sec
test_unparse: 4 min 20 sec
test_unicodedata: 4 min 9 sec
test_multiprocessing_forkserver: 3 min 40 sec
test_subprocess: 3 min 33 sec
test_lib2to3: 3 min 27 sec
test_asyncio: 3 min 10 sec

1 test failed:
test_gdb

20 tests skipped:
test_ctypes test_dbm_gnu test_devpoll test_epoll test_idle
test_ioctl test_msilib test_readline test_spwd test_startfile
test_tcl test_tix test_tk test_ttk_guionly test_ttk_textonly
test_turtle test_winconsoleio test_winreg test_winsound
test_zipfile64

2 re-run tests:
test_concurrent_futures test_gdb

Total duration: 31 min 6 sec

Click to see traceback logs

Traceback (most recent call last): File "/usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/test_gdb.py", line 813, in test_bt_fullself.assertMultilineMatches(bt, ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/test_gdb.py", line 297, in assertMultilineMatchesself.fail(msg='%r did not match %r'% (actual, pattern)) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^AssertionError: 'Breakpoint 1 (builtin_id) pending.\n\nBreakpoint 1, builtin_id (self=<optimized out>, v=42) at Python/bltinmodule.c:1197\n1197\t PyObject *id = PyLong_FromVoidPtr(v);\n#1 <built-in method id of module object at remote 0x8014ac3b0>\n#4 Frame 0x800235110, for file /usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/gdb_sample.py, line 7, in bar (a=1, b=2, c=3)\n baz(a, b, c)\n#9 Frame 0x800235020, for file /usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/gdb_sample.py, line 12, in <module> ()\n foo(1, 2, 3)\n' did not match '^.*\n#[0-9]+ Frame 0x-?[0-9a-f]+, for file .*gdb_sample.py, line 7, in bar \\(a=1, b=2, c=3\\)\n baz\\(a, b, c\\)\n#[0-9]+ Frame 0x-?[0-9a-f]+, for file .*gdb_sample.py, line 4, in foo \\(a=1, b=2, c=3\\)\n bar\\(a=a, b=b, c=c\\)\n#[0-9]+ Frame 0x-?[0-9a-f]+, for file .*gdb_sample.py, line 12, in <module> \\(\\)\n foo\\(1, 2, 3\\)\n' Traceback (most recent call last): File "/usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/test_concurrent_futures.py", line 1038, in test_idle_process_reuse_multipleself.assertLessEqual(len(executor._processes), 2) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^AssertionError: 3 not less than or equal to 2 Traceback (most recent call last): File "/usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/test_gdb.py", line 735, in test_pyup_commandself.assertMultilineMatches(bt, ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/test_gdb.py", line 297, in assertMultilineMatchesself.fail(msg='%r did not match %r'% (actual, pattern)) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^AssertionError: 'Breakpoint 1 (builtin_id) pending.\n\nBreakpoint 1, builtin_id (self=<optimized out>, v=42) at Python/bltinmodule.c:1197\n1197\t PyObject *id = PyLong_FromVoidPtr(v);\n#4 Frame 0x800235110, for file /usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/gdb_sample.py, line 7, in bar (a=1, b=2, c=3)\n baz(a, b, c)\n#9 Frame 0x800235020, for file /usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/gdb_sample.py, line 12, in <module> ()\n foo(1, 2, 3)\n' did not match '^.*\n#[0-9]+ Frame 0x-?[0-9a-f]+, for file .*gdb_sample.py, line 10, in baz \\(args=\\(1, 2, 3\\)\\)\n id\\(42\\)\n#[0-9]+ Frame 0x-?[0-9a-f]+, for file .*gdb_sample.py, line 7, in bar \\(a=1, b=2, c=3\\)\n baz\\(a, b, c\\)\n#[0-9]+ Frame 0x-?[0-9a-f]+, for file .*gdb_sample.py, line 4, in foo \\(a=1, b=2, c=3\\)\n bar\\(a=a, b=b, c=c\\)\n#[0-9]+ Frame 0x-?[0-9a-f]+, for file .*gdb_sample.py, line 12, in <module> \\(\\)\n foo\\(1, 2, 3\\)\n$' Traceback (most recent call last): File "/usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/test_gdb.py", line 696, in test_basic_commandself.assertListing(' 5 \n'^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/test_gdb.py", line 689, in assertListingself.assertEndsWith(actual, expected) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/test_gdb.py", line 291, in assertEndsWithself.assertTrue(actual.endswith(exp_end), ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^AssertionError: False is not true : 'Breakpoint 1 (builtin_id) pending.\n\nBreakpoint 1, builtin_id (self=<optimized out>, v=42) at Python/bltinmodule.c:1197\n1197\t PyObject *id = PyLong_FromVoidPtr(v);\n 2 \n 3 def foo(a, b, c):\n 4 bar(a=a, b=b, c=c)\n 5 \n 6 def bar(a, b, c):\n >7 baz(a, b, c)\n 8 \n 9 def baz(*args):\n 10 id(42)\n 11 \n 12 foo(1, 2, 3)\n' did not end with ' 5 \n 6 def bar(a, b, c):\n 7 baz(a, b, c)\n 8 \n 9 def baz(*args):\n >10 id(42)\n 11 \n 12 foo(1, 2, 3)\n' Traceback (most recent call last): File "/usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/test_concurrent_futures.py", line 1029, in test_idle_process_reuse_oneself.assertEqual(len(executor._processes), 1) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^AssertionError: 2 != 1 Traceback (most recent call last): File "/usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/test_gdb.py", line 770, in test_up_then_downself.assertMultilineMatches(bt, ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/test_gdb.py", line 297, in assertMultilineMatchesself.fail(msg='%r did not match %r'% (actual, pattern)) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ AssertionError: 'Breakpoint 1 (builtin_id) pending.\n\nBreakpoint 1, builtin_id (self=<optimized out>, v=42) at Python/bltinmodule.c:1197\n1197\t PyObject *id = PyLong_FromVoidPtr(v);\n#4 Frame 0x800235110, for file /usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/gdb_sample.py, line 7, in bar (a=1, b=2, c=3)\n baz(a, b, c)\n#9 Frame 0x800235020, for file /usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/gdb_sample.py, line 12, in <module> ()\n foo(1, 2, 3)\n#4 Frame 0x800235110, for file /usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/gdb_sample.py, line 7, in bar (a=1, b=2, c=3)\n baz(a, b, c)\n' did not match '^.*\n#[0-9]+ Frame 0x-?[0-9a-f]+, for file .*gdb_sample.py, line 10, in baz \\(args=\\(1, 2, 3\\)\\)\n id\\(42\\)\n#[0-9]+ Frame 0x-?[0-9a-f]+, for file .*gdb_sample.py, line 7, in bar \\(a=1, b=2, c=3\\)\n baz\\(a, b, c\\)\n#[0-9]+ Frame 0x-?[0-9a-f]+, for file .*gdb_sample.py, line 4, in foo \\(a=1, b=2, c=3\\)\n bar\\(a=a, b=b, c=c\\)\n#[0-9]+ Frame 0x-?[0-9a-f]+, for file .*gdb_sample.py, line 12, in <module> \\(\\)\n foo\\(1, 2, 3\\)\n#[0-9]+ Frame 0x-?[0-9a-f]+, for file .*gdb_sample.py, line 10, in baz \\(args=\\(1, 2, 3\\)\\)\n id\\(42\\)\n#[0-9]+ Frame 0x-?[0-9a-f]+, for file .*gdb_sample.py, line 7, in bar \\(a=1, b=2, c=3\\)\n baz\\(a, b, c\\)\n$' Traceback (most recent call last): File "/usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/test_gdb.py", line 990, in test_print_after_upself.assertMultilineMatches(bt, ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/test_gdb.py", line 297, in assertMultilineMatchesself.fail(msg='%r did not match %r'% (actual, pattern)) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^AssertionError: "Breakpoint 1 (builtin_id) pending.\n\nBreakpoint 1, builtin_id (self=<optimized out>, v=42) at Python/bltinmodule.c:1197\n1197\t PyObject *id = PyLong_FromVoidPtr(v);\n#4 Frame 0x800235110, for file /usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/gdb_sample.py, line 7, in bar (a=1, b=2, c=3)\n baz(a, b, c)\n#9 Frame 0x800235020, for file /usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/gdb_sample.py, line 12, in <module> ()\n foo(1, 2, 3)\n'c' not found\n'b' not found\n'a' not found\n" did not match ".*\\nlocal 'c' = 3\\nlocal 'b' = 2\\nlocal 'a' = 1\\n.*" Traceback (most recent call last): File "/usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/test_gdb.py", line 1024, in test_locals_after_upself.assertMultilineMatches(bt, ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/test_gdb.py", line 297, in assertMultilineMatchesself.fail(msg='%r did not match %r'% (actual, pattern)) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^AssertionError: 'Breakpoint 1 (builtin_id) pending.\n\nBreakpoint 1, builtin_id (self=<optimized out>, v=42) at Python/bltinmodule.c:1197\n1197\t PyObject *id = PyLong_FromVoidPtr(v);\n#4 Frame 0x800235110, for file /usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/gdb_sample.py, line 7, in bar (a=1, b=2, c=3)\n baz(a, b, c)\n#9 Frame 0x800235020, for file /usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/gdb_sample.py, line 12, in <module> ()\n foo(1, 2, 3)\nLocals for <module>\n' did not match '^.*\nLocals for foo\na = 1\nb = 2\nc = 3\nLocals for <module>\n.*$' Traceback (most recent call last): File "/usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/test_gdb.py", line 793, in test_btself.assertMultilineMatches(bt, ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/test_gdb.py", line 297, in assertMultilineMatchesself.fail(msg='%r did not match %r'% (actual, pattern)) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^AssertionError: 'Breakpoint 1 (builtin_id) pending.\n\nBreakpoint 1, builtin_id (self=<optimized out>, v=42) at Python/bltinmodule.c:1197\n1197\t PyObject *id = PyLong_FromVoidPtr(v);\nTraceback (most recent call first):\n <built-in method id of module object at remote 0x8014ac3b0>\n File "/usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/gdb_sample.py", line 7, in bar\n baz(a, b, c)\n File "/usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/gdb_sample.py", line 12, in <module>\n foo(1, 2, 3)\n' did not match '^.*\nTraceback \\(most recent call first\\):\n <built-in method id of module object .*>\n File ".*gdb_sample.py", line 10, in baz\n id\\(42\\)\n File ".*gdb_sample.py", line 7, in bar\n baz\\(a, b, c\\)\n File ".*gdb_sample.py", line 4, in foo\n bar\\(a=a, b=b, c=c\\)\n File ".*gdb_sample.py", line 12, in <module>\n foo\\(1, 2, 3\\)\n' Traceback (most recent call last): File "/usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/test_gdb.py", line 981, in test_basic_commandself.assertMultilineMatches(bt, ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/test_gdb.py", line 297, in assertMultilineMatchesself.fail(msg='%r did not match %r'% (actual, pattern)) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^AssertionError: "Breakpoint 1 (builtin_id) pending.\n\nBreakpoint 1, builtin_id (self=<optimized out>, v=42) at Python/bltinmodule.c:1197\n1197\t PyObject *id = PyLong_FromVoidPtr(v);\n#4 Frame 0x800235110, for file /usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/gdb_sample.py, line 7, in bar (a=1, b=2, c=3)\n baz(a, b, c)\n'args' not found\n" did not match ".*\\nlocal 'args' = \\(1, 2, 3\\)\\n.*" Traceback (most recent call last): File "/usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/test_gdb.py", line 711, in test_one_abs_argself.assertListing(' 9 def baz(*args):\n'^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/test_gdb.py", line 689, in assertListingself.assertEndsWith(actual, expected) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/test_gdb.py", line 291, in assertEndsWithself.assertTrue(actual.endswith(exp_end), ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^AssertionError: False is not true : 'Breakpoint 1 (builtin_id) pending.\n\nBreakpoint 1, builtin_id (self=<optimized out>, v=42) at Python/bltinmodule.c:1197\n1197\t PyObject *id = PyLong_FromVoidPtr(v);\n 9 def baz(*args):\n 10 id(42)\n 11 \n 12 foo(1, 2, 3)\n' did not end with ' 9 def baz(*args):\n >10 id(42)\n 11 \n 12 foo(1, 2, 3)\n' Traceback (most recent call last): File "/usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/test_gdb.py", line 1015, in test_basic_commandself.assertMultilineMatches(bt, ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/test_gdb.py", line 297, in assertMultilineMatchesself.fail(msg='%r did not match %r'% (actual, pattern)) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^AssertionError: 'Breakpoint 1 (builtin_id) pending.\n\nBreakpoint 1, builtin_id (self=<optimized out>, v=42) at Python/bltinmodule.c:1197\n1197\t PyObject *id = PyLong_FromVoidPtr(v);\n#4 Frame 0x800235110, for file /usr/home/buildbot/python/3.x.koobs-freebsd-564d/build/Lib/test/gdb_sample.py, line 7, in bar (a=1, b=2, c=3)\n baz(a, b, c)\nLocals for bar\na = 1\nb = 2\nc = 3\n' did not match '.*\\nargs = \\(1, 2, 3\\)\\n.*'

Implement specialized opcodes for BINARY_MULTIPLY
3c85756

sweeneyde requested a review from markshannon as a code owner October 5, 2021 03:45

bedevere-bot added the awaiting review label Oct 5, 2021

the-knights-who-say-ni added the CLA signed label Oct 5, 2021

📜🤖 Added by blurb_it.
300a0ca

sweeneyde added the performance Performance or resource usage label Oct 5, 2021

sweeneyde closed this Oct 5, 2021

sweeneyde reopened this Oct 5, 2021

sweeneyde added the 🔨 test-with-buildbots Test PR w/ buildbots; report in status section label Oct 5, 2021

bedevere-bot removed the 🔨 test-with-buildbots Test PR w/ buildbots; report in status section label Oct 5, 2021

markshannon reviewed Oct 5, 2021
View reviewed changes

Include/internal/pycore_code.h Outdated Show resolvedHide resolved

Fidget-Spinner reviewed Oct 5, 2021
View reviewed changes

Python/ceval.c Outdated Show resolvedHide resolved
Python/specialize.cShow resolvedHide resolved
Python/specialize.c Outdated Show resolvedHide resolved

Resolve merge conflicts
d266912

markshannon self-assigned this Oct 5, 2021

Apply changes from code review
c64540a

markshannon reviewed Oct 5, 2021
View reviewed changes

Allow float/int mixtures is BINARY_MULTIPLY_FLOAT
5a44ecc

Minor cleanups and comments
aea424e

sweeneyde added 3 commits October 7, 2021 10:29

Revert the float/int mixtures
37d3716

typo: sum --> prod
903ff9c

Merge branch 'main' into multiply
7aecc97

sweeneyde closed this Oct 10, 2021

sweeneyde reopened this Oct 11, 2021

Merge main into multiply
31c3bb9

markshannon added the 🔨 test-with-buildbots Test PR w/ buildbots; report in status section label Oct 13, 2021

bedevere-bot removed the 🔨 test-with-buildbots Test PR w/ buildbots; report in status section label Oct 13, 2021

markshannon merged commit 3b3d30e into python:mainOct 14, 2021

bedevere-bot removed the awaiting review label Oct 14, 2021

Uh oh!

bpo-45367: Specialize BINARY_MULTIPLY#28727

bpo-45367: Specialize BINARY_MULTIPLY #28727

Uh oh!

Conversation

sweeneyde commented Oct 5, 2021• edited by bedevere-bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

bedevere-bot commented Oct 5, 2021

Uh oh!

sweeneyde commented Oct 5, 2021• edited by bedevere-bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

sweeneyde commented Oct 5, 2021

Uh oh!

Uh oh!

markshannon commented Oct 5, 2021

Uh oh!

Fidget-Spinner left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

markshannonOct 5, 2021

Choose a reason for hiding this comment

Uh oh!

Fidget-SpinnerOct 5, 2021• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

sweeneydeOct 5, 2021

Choose a reason for hiding this comment

Uh oh!

Fidget-SpinnerOct 5, 2021• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

markshannonOct 5, 2021• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

markshannon commented Oct 5, 2021

Uh oh!

Fidget-Spinner commented Oct 5, 2021

Uh oh!

sweeneyde commented Oct 5, 2021

Uh oh!

sweeneyde commented Oct 5, 2021

Uh oh!

markshannon commented Oct 6, 2021

Uh oh!

sweeneyde commented Oct 6, 2021

Uh oh!

Fidget-Spinner commented Oct 6, 2021• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

markshannon commented Oct 7, 2021

Uh oh!

markshannon commented Oct 7, 2021

Uh oh!

sweeneyde commented Oct 10, 2021

Uh oh!

markshannon commented Oct 11, 2021

Uh oh!

sweeneyde commented Oct 11, 2021

Uh oh!

bedevere-bot commented Oct 13, 2021

Uh oh!

markshannon commented Oct 14, 2021

Uh oh!

bedevere-bot commented Oct 14, 2021

⚠️⚠️⚠️ Buildbot failure ⚠️⚠️⚠️

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

sweeneyde commented Oct 5, 2021•
edited by bedevere-bot
Loading

sweeneyde commented Oct 5, 2021•
edited by bedevere-bot
Loading

Fidget-SpinnerOct 5, 2021•
edited
Loading

Fidget-SpinnerOct 5, 2021•
edited
Loading

markshannonOct 5, 2021•
edited
Loading

Fidget-Spinner commented Oct 6, 2021•
edited
Loading