gh-153740: Optimize PyFloat_Pack/Unpack2 using native _Float16

[skirpichev]

Benchmarks:

• gh-153740: Optimize PyFloat_Pack/Unpack2 using native _Float16 #154796 (comment)
• gh-153740: Optimize PyFloat_Pack/Unpack2 using native _Float16 #154796 (comment)

• Issue: Conditionally support native half-precision IEEE types in struct/array/ctypes modules and memoryview #153740

[skirpichev]

Not sure how to fix this error:

/usr/bin/ld: /usr/bin/ld: DWARF error: invalid or unhandled FORM value: 0x25
Objects/floatobject.o: in function `PyFloat_Pack2':
floatobject.c:(.text.PyFloat_Pack2[PyFloat_Pack2]+0x32): undefined reference to `__truncdfhf2'
clang: error: linker command failed with exit code 1 (use -v to see invocation)

CC @StanFromIreland (fuzzers)

[StanFromIreland]

The configure probe only tests a compile-time _Float16 constant, so it reports "yes" in the OSS-Fuzz Docker container where clang accepts the type but the container's pre-GCC-12 libgcc lacks __truncdfhf2 which is the runtime helper the PyFloat_Pack2 double to _Float16 conversion links against. I suggest you make configure do a runtime double to _Float16 conversion.

[skirpichev]

I suggest you make configure do a runtime double to _Float16 conversion.

Good idea, that works.

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @skirpichev for commit ac38cb8 🤖

Results will be shown at:

https://buildbot.python.org/all/#/grid?branch=refs%2Fpull%2F154796%2Fmerge

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

[vstinner]

@skirpichev: Did you close your PR on purpose?

[skirpichev]

Did you close your PR on purpose?

Yes, this lacks benchmarks. I hope someone will continue this.

[skirpichev]

@hpkfft, maybe you can check performance for a simple pyperf-based benchmark?

import pyperf
from struct import pack, unpack

x = 3.140625
bx = struct.pack('e', x)

runner = pyperf.Runner()
runner.bench_func('pack("e", 3.140625)', pack, "e", x)
runner.bench_func('unpack("e", b"HB")', unpack, "e", bx)

[hpkfft]

Using gcc 14.2.0 and ./configure:

main:
pack("e", 3.140625): Mean +- std dev: 56.2 ns +- 1.1 ns
unpack("e", b"HB"): Mean +- std dev: 64.8 ns +- 1.2 ns

patch:
pack("e", 3.140625): Mean +- std dev: 51.9 ns +- 0.8 ns
unpack("e", b"HB"): Mean +- std dev: 59.0 ns +- 1.0 ns

Using ./configure --enable-optimizations:

main:
pack("e", 3.140625): Mean +- std dev: 37.7 ns +- 0.2 ns
unpack("e", b"HB"): Mean +- std dev: 46.1 ns +- 0.4 ns

patch:
pack("e", 3.140625): Mean +- std dev: 38.2 ns +- 0.3 ns
unpack("e", b"HB"): Mean +- std dev: 43.6 ns +- 0.5 ns

Interesting. Does the training input for PGO have NaNs in it? If so, maybe it shouldn't....

[skirpichev]

Does the training input for PGO have NaNs in it?

Yes, PGO set of tests lacks test_capi, but test_struct.py is selected (see Lib/test/libregrtest/pgo.py). That one has various tests, including half-float NaNs. You can filter out some tests for PGO by @skip_if_pgo_task. (Usually we do this for expensive tests.)

Apparently, CPython runtime overhead dominated in the above benchmark. For C code I got ~2x speedup, like you above. I'll try to repeat that with some benchmarking framework.

Anyway, lets reopen this.

[vstinner]

Ah good, the new code (since my previous review) is way shorter. Its size is now more acceptable to me.

[vstinner]

LGTM

[vstinner]

Merged, thanks.

[vstinner]

I didn't close #153740 since it seems like more changes are planned.

[bedevere-bot]

⚠️⚠️⚠️ Buildbot failure ⚠️⚠️⚠️

Hi! The buildbot wasm32-wasi 3.x (tier-2) has failed when building commit e7ef46a.

What do you need to do:

1. Don't panic.
2. Check the buildbot page in the devguide if you don't know what the buildbots are or how they work.
3. Go to the page of the buildbot that failed (https://buildbot.python.org/#/builders/1046/builds/12907) and take a look at the build logs.
4. Check if the failure is related to this commit (e7ef46a) or if it is a false positive.
5. 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/#/builders/1046/builds/12907

Failed tests:

• test_zipfile

Failed subtests:

• test_append_to_concatenated_zip_file - test.test_zipfile.test_core.StoredTestsWithSourceFile.test_append_to_concatenated_zip_file
• test_write_filtered_python_package - test.test_zipfile.test_core.PyZipFileTests.test_write_filtered_python_package

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

==

Click to see traceback logs

Traceback (most recent call last):
  File "/Lib/test/test_zipfile/test_core.py", line 3649, in test_write_filtered_python_package
    with TemporaryFile() as t, zipfile.PyZipFile(t, "w") as zipfp:
                               ~~~~~~~~~~~~~~~~~^^^^^^^^
  File "/Lib/zipfile/__init__.py", line 2026, in __exit__
    self.close()
    ~~~~~~~~~~^^
  File "/Lib/zipfile/__init__.py", line 2667, in close
    self.fp.seek(self.start_dir)
    ~~~~~~~~~~~~^^^^^^^^^^^^^^^^
OSError: [Errno 51] No space left on device


Traceback (most recent call last):
  File "/Lib/test/test_zipfile/test_core.py", line 3654, in test_write_filtered_python_package
    zipfp.writepy(packagedir)
    ~~~~~~~~~~~~~^^^^^^^^^^^^
  File "/Lib/zipfile/__init__.py", line 2830, in writepy
    self.writepy(path, basename,
    ~~~~~~~~~~~~^^^^^^^^^^^^^^^^
                 filterfunc=filterfunc)  # Recursive call
                 ^^^^^^^^^^^^^^^^^^^^^^
  File "/Lib/zipfile/__init__.py", line 2840, in writepy
    self.writestr(arcname, bytecode)
    ~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^
  File "/Lib/zipfile/__init__.py", line 2605, in writestr
    with self.open(zinfo, mode='w') as dest:
         ~~~~~~~~~^^^^^^^^^^^^^^^^^
  File "/Lib/zipfile/__init__.py", line 1423, in close
    self._fileobj.seek(self._zinfo.header_offset)
    ~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^
OSError: [Errno 51] No space left on device


Traceback (most recent call last):
  File "/Lib/test/test_zipfile/test_core.py", line 3649, in test_write_filtered_python_package
    with TemporaryFile() as t, zipfile.PyZipFile(t, "w") as zipfp:
         ~~~~~~~~~~~~~^^
OSError: [Errno 51] No space left on device


Traceback (most recent call last):
  File "/Lib/test/test_zipfile/test_core.py", line 572, in test_append_to_concatenated_zip_file
    f.write(data)
    ~~~~~~~^^^^^^
OSError: [Errno 51] No space left on device