MNT: Update windows-2019 to windows-2022[wheel build] #28955

abhishek-iitmadras · 2025-05-14T09:49:00Z

As per info in : actions/runner-images#12045
We need to change window runner.

charris · 2025-05-26T13:47:58Z

close/reopen

charris · 2025-05-26T13:54:22Z

Put a triage label on this, as the deadline is coming up and there are problems.

charris · 2025-05-26T13:56:04Z

@h-vetinari Ping, because ISTR that you were involved last time we updated the windows version.

charris · 2025-05-26T14:25:43Z

The failures seem to mostly come from test_unary_spurious_fpexception and accuracy in some ufuncs.

charris · 2025-05-26T14:32:22Z

The failures seem to mostly come from test_unary_spurious_fpexception and accuracy in some ufuncs.

The test_unary_spurious_fpexception seems to come from OpenBLAS
The accuracy failures are in the no blas builds.

h-vetinari · 2025-05-27T03:37:03Z

@h-vetinari Ping, because ISTR that you were involved last time we updated the windows version.

Thanks for the ping. The only place where this really becomes relevant is for linking to npymath.lib. Anyone doing that will most likely need to update to VS2022.

Given that VS2019 is EOL for over a year, and that VS2022 is an ABI-compatible update, this should not be an undue burden. The timing matches well, I've just picked up the proposal to move to VS2022 in conda-forge again.

mattip · 2025-05-27T05:03:41Z

The only place where this really becomes relevant is for linking to npymath.lib

Isn't this snippet meant to make a npymath.lib that can be used by older linkers?

numpy/numpy/_core/meson.build

Lines 569 to 577 in 3c995e7

    
           if cc.get_id() == 'msvc' 
        
             # Disable voltbl section for vc142 to allow link using mingw-w64; see: 
        
             # https://github.com/matthew-brett/dll_investigation/issues/1#issuecomment-1100468171 
        
             # Needs to be added to static libraries that are shipped for reuse (i.e., 
        
             # libnpymath and libnpyrandom) 
        
             if cc.has_argument('-d2VolatileMetadata-') 
        
                staticlib_cflags +=  '-d2VolatileMetadata-' 
        
             endif 
        
           endif

mattip · 2025-05-27T05:05:29Z

The test_unary_spurious_fpexception seems to come from OpenBLAS

Maybe related to #29039 which updates OpenBLAS

h-vetinari · 2025-05-27T05:20:28Z

Isn't this snippet meant to make a npymath.lib that can be used by older linkers?

Not AFAIU. That's about reusing/sharing the objects in the static library between MSVC and mingw, and the comment from the upstream devs at the time was

Mixing static libraries or object files between mingw and msvc is not
supported, and not expected to work, in general. You might have been lucky
and your library might have been simple enough not to hit any problematic
case though.

The virality of the msvc toolset concerns being able to link to something built with VS2022 (which - at least on paper - then also requires VS2022)

charris · 2025-05-28T16:50:49Z

Rather than latest, it would be better to explicitly use vs2022.

charris · 2025-05-28T19:16:15Z

@seiko2plus The test_unary_spurious_fpexception are in ufunc = <ufunc 'spacing'> for f and d. I see that test is already skipped for e. Should we just skip it for f and d as well? Is there something we should do for SIMD instead?

rgommers · 2025-05-30T16:49:33Z

The only place where this really becomes relevant is for linking to npymath.lib. Anyone doing that will most likely need to update to VS2022.

Agreed, which means we need to update to the 2022 image in SciPy first before merging this.

We should probably do that asap and then merge this and backport it to the 2.3.x branch. before releasing 2.3.0. Otherwise it'll become a potentially risky change in 2.3.1 or 2.3.2.

abhishek-iitmadras · 2025-06-01T16:55:40Z

Look like i need to skip spacing test for window

charris · 2025-06-01T17:33:12Z

vs2022 changes: https://devblogs.microsoft.com/cppblog/the-fpcontract-flag-and-changes-to-fp-modes-in-vs2022/

abhishek-iitmadras · 2025-06-01T17:35:40Z

vs2022 changes: https://devblogs.microsoft.com/cppblog/the-fpcontract-flag-and-changes-to-fp-modes-in-vs2022/

Windows Server 2019: Uses older MSVC that generates FMA contractions by default
Windows Server 2022: Uses VS2022's MSVC which does NOT generate FMA contractions by default

This change significantly affects floating-point precision, especially for transcendental functions

So we need to add the /fp:contract flag to MSVC builds.

charris · 2025-06-02T12:22:46Z

Windows Server 2022: Uses VS2022's MSVC which does NOT generate FMA contractions by default

Note that meson shows FMA3 enabled, but evidently a flag is required for msvc in vs2022. I'm not clear on whether meson should handle that or not. Unfortunately, the /fp:contract flag is new. Maybe disable FMA3 if the flag is missing for recent msvc? @rgommers Thoughts?

charris · 2025-06-02T12:49:51Z

.github/workflows/windows.yml

@@ -119,7 +119,7 @@ jobs:

      - name: Build and install
        run: |
-          python -m pip install . -v -Ccompile-args="-j2" -Csetup-args="-Dallow-noblas=true"
+          python -m pip install . -v -Ccompile-args="-j2" -Csetup-args="-Dallow-noblas=true" -Csetup-args="-Dc_args=/fp:contract" -Csetup-args="-Dcpp_args=/fp:contract"


This would read better with a line continuation \ to break the line. Probably for the previous use as well.

This should be part of the meson flags when building SIMD code. I suggest moving it here:

numpy/meson_cpu/x86/meson.build

Line 214 in b353a18

FMA3.update(args: {'val': '/arch:AVX2', 'match': clear_arch})

numpy/_core/tests/test_umath.py

charris · 2025-06-02T17:41:57Z

.github/workflows/wheels.yml

+          # Add compiler flags for 32-bit builds
+          if ("${{ matrix.buildplat[1] }}" -eq "win32") {
+            echo 'CIBW_CONFIG_SETTINGS=setup-args=-Dc_args="/fp:contract" setup-args=-Dcpp_args="/fp:contract"' >> $env:GITHUB_ENV
+          }


The extra flags should not be required after updating meson_cpu.

charris · 2025-06-02T18:56:29Z

You need to do the update this way:

  FMA3.update(args: {'val': '/arch:AVX2', 'match': clear_arch})
  FMA3.update(args: {'val': '/fp:contract'})

abhishek-iitmadras · 2025-06-02T19:00:28Z

You need to do the update this way:

  FMA3.update(args: {'val': '/arch:AVX2', 'match': clear_arch})
  FMA3.update(args: {'val': '/fp:contract'})

isn't it overriding the args for FMA3 ?Correct me if i am wrong

abhishek-iitmadras · 2025-06-02T19:41:30Z

Hi @charris @r-devulap

Now floating-point contract flag for 32-bit MSVC builds only is added to meson build .
Skip spacing tests with NaN on Windows MSVC (all dtypes) only.

Please have a review to this patch.
Thanks

r-devulap · 2025-06-02T19:48:28Z

meson_cpu/x86/meson.build

+
+  # Add floating-point contract flag for 32-bit builds
+  # Fixes transcendental function accuracy on Windows Server 2022
+  if cpu_family == 'x86'


This file is already under meson_cpu/x86/meson.build, you don't need the if condition here.

Don't we need the flag for 64-bit as well?

Don't we need the flag for 64-bit as well?

It doesn't seem required by us, but we probably do want it.

charris · 2025-06-02T19:57:37Z

#29116 is working now. There is a bogus circleci failure that can be ignored. I'm curious if @r-devulap suggestion of putting the args in a list works.

charris · 2025-06-02T22:28:56Z

The PyPy failure can be ignored.

abhishek-iitmadras · 2025-06-03T11:16:32Z

@rgommers @charris @r-devulap
This patch seems ok, please have a review and merge.

meson_cpu/x86/meson.build

charris · 2025-06-03T14:54:13Z

meson_cpu/x86/meson.build

+
+  # Add floating-point contract flag for 32-bit builds
+  # Fixes transcendental function accuracy on Windows Server 2022
+  add_project_arguments('/fp:contract', language: ['c', 'cpp'])


This should depend on the existence of FMA. Note that add_project_arguments is not used elsewhere in cpu specific contexts.

r-devulap · 2025-06-03T16:27:46Z

meson_cpu/x86/meson.build

@@ -211,7 +211,8 @@ if compiler_id == 'msvc'
      fet.update(args: '')
    endif
  endforeach
-  FMA3.update(args: {'val': '/arch:AVX2', 'match': clear_arch})
+  # Add floating-point contract flag to fixes transcendental function accuracy on Windows Server 2022
+  FMA3.update(args: [{'val': '/arch:AVX2', 'match': clear_arch},{'val': '/fp:contract', 'match': clear_arch}])


Lets stick to the same way @charris fixed it in #29116:

Suggested change

FMA3.update(args: [{'val': '/arch:AVX2', 'match': clear_arch},{'val': '/fp:contract', 'match': clear_arch}])

FMA3.update(args: {'val': '/arch:AVX2', 'match': clear_arch})

FMA3.update(args: {'val': '/fp:contract'})

Tried it locally and confirmed that it picks up the flag correctly:

"command": "\"cl\" \"-Inumpy\\_core\\libloops_trigonometric.dispatch.h_AVX2.a.p\" \"-Inumpy\\_core\" \"-I..\\numpy\\_core\" \"-Inumpy\\_core\\include\" \"-I..\\numpy\\_core\\include\" \"-I..\\numpy\\_core\\src\\common\" \"-I..\\numpy\\_core\\src\\multiarray\" \"-I..\\numpy\\_core\\src\\npymath\" \"-I..\\numpy\\_core\\src\\umath\" \"-I..\\numpy\\_core\\src\\highway\" \"-IC:\\Program Files\\WindowsApps\\PythonSoftwareFoundation.Python.3.11_3.11.2544.0_x64__qbz5n2kfra8p0\\Include\" \"-IC:\\Users\\RDEVULAP\\source\\repos\\numpy\\build\\meson_cpu\" \"/MD\" \"/nologo\" \"/showIncludes\" \"/utf-8\" \"/Zc:__cplusplus\" \"/W2\" \"/EHsc\" \"/std:c++17\" \"/permissive-\" \"/O2\" \"/Zi\" \"-DMS_WIN64=\" \"-DNPY_INTERNAL_BUILD\" \"-DHAVE_NPY_CONFIG_H\" \"-D_FILE_OFFSET_BITS=64\" \"-D_LARGEFILE_SOURCE=1\" \"-D_LARGEFILE64_SOURCE=1\" \"/O2\" \"-DNPY_HAVE_SSE2\" \"-DNPY_HAVE_SSE\" \"-DNPY_HAVE_SSE3\" \"-DNPY_HAVE_SSSE3\" \"-DNPY_HAVE_SSE41\" \"-DNPY_HAVE_POPCNT\" \"-DNPY_HAVE_SSE42\" \"-DNPY_HAVE_AVX\" \"-DNPY_HAVE_F16C\" \"-DNPY_HAVE_FMA3\" \"-DNPY_HAVE_AVX2\" \"/fp:contract\" \"/arch:AVX2\" \"-DNPY_MTARGETS_CURRENT=AVX2\" \"/Fdnumpy\\_core\\libloops_trigonometric.dispatch.h_AVX2.a.p\\src_umath_loops_trigonometric.dispatch.cpp.pdb\" /Fonumpy/_core/libloops_trigonometric.dispatch.h_AVX2.a.p/src_umath_loops_trigonometric.dispatch.cpp.obj \"/c\" ../numpy/_core/src/umath/loops_trigonometric.dispatch.cpp",

Thanks @r-devulap for confirmation .

Co-authored-by: Charles Harris <charlesr.harris@gmail.com>

r-devulap

LGTM.

charris · 2025-06-03T18:10:27Z

Thanks @abhishek-iitmadras .

abhishek-iitmadras changed the title ~~Update windows-2019 to windows-latest~~ MNT: Update windows-2019 to windows-latest May 15, 2025

charris closed this May 26, 2025

charris reopened this May 26, 2025

charris added the 03 - Maintenance label May 26, 2025

charris added this to the 2.3.0 release milestone May 26, 2025

charris added the triage review Issue/PR to be discussed at the next triage meeting label May 26, 2025

rgommers mentioned this pull request May 30, 2025

CI: update windows-2019 runner image to windows-2022 scipy/scipy#23086

Merged

abhishek-iitmadras force-pushed the abhishek_window branch from 8aa40c7 to 75079b9 Compare May 31, 2025 16:04

abhishek-iitmadras changed the title ~~MNT: Update windows-2019 to windows-latest~~ MNT: Update windows-2019 to windows-2022 May 31, 2025

rgommers mentioned this pull request May 31, 2025

WIP: Use vs2022 in NumPy 2.3.x [wheel build] #29103

Closed

abhishek-iitmadras force-pushed the abhishek_window branch from 75079b9 to 755cc38 Compare June 1, 2025 16:42

abhishek-iitmadras force-pushed the abhishek_window branch from 087ca33 to f33423f Compare June 1, 2025 17:13

charris reviewed Jun 2, 2025

View reviewed changes

numpy/_core/tests/test_umath.py Show resolved Hide resolved

charris reviewed Jun 2, 2025

View reviewed changes

abhishek-iitmadras force-pushed the abhishek_window branch from 654c62e to 4a09c91 Compare June 2, 2025 19:18

r-devulap requested changes Jun 2, 2025

View reviewed changes

charris removed the triage review Issue/PR to be discussed at the next triage meeting label Jun 2, 2025

charris modified the milestones: 2.3.0 release, 2.4.0 release Jun 2, 2025

abhishek-iitmadras force-pushed the abhishek_window branch 6 times, most recently from 3a5335f to 4490065 Compare June 3, 2025 11:14

charris reviewed Jun 3, 2025

View reviewed changes

meson_cpu/x86/meson.build Show resolved Hide resolved

charris reviewed Jun 3, 2025

View reviewed changes

abhishek-iitmadras force-pushed the abhishek_window branch 3 times, most recently from 8f2916c to 16a8390 Compare June 3, 2025 16:03

r-devulap requested changes Jun 3, 2025

View reviewed changes

update windows-2019 to windows-2022 and meson flag[wheel build]

2306bce

Co-authored-by: Charles Harris <charlesr.harris@gmail.com>

abhishek-iitmadras force-pushed the abhishek_window branch from 16a8390 to 2306bce Compare June 3, 2025 16:35

r-devulap approved these changes Jun 3, 2025

View reviewed changes

charris merged commit ff1d6cc into numpy:main Jun 3, 2025
140 checks passed

abhishek-iitmadras deleted the abhishek_window branch June 3, 2025 18:33

	FMA3.update(args: [{'val': '/arch:AVX2', 'match': clear_arch},{'val': '/fp:contract', 'match': clear_arch}])
	FMA3.update(args: {'val': '/arch:AVX2', 'match': clear_arch})
	FMA3.update(args: {'val': '/fp:contract'})

Uh oh!

MNT: Update windows-2019 to windows-2022[wheel build] #28955

MNT: Update windows-2019 to windows-2022[wheel build] #28955

Uh oh!

Conversation

abhishek-iitmadras commented May 14, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

charris commented May 26, 2025

Uh oh!

charris commented May 26, 2025

Uh oh!

charris commented May 26, 2025

Uh oh!

charris commented May 26, 2025

Uh oh!

charris commented May 26, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

h-vetinari commented May 27, 2025

Uh oh!

mattip commented May 27, 2025

Uh oh!

mattip commented May 27, 2025

Uh oh!

h-vetinari commented May 27, 2025

Uh oh!

charris commented May 28, 2025

Uh oh!

charris commented May 28, 2025

Uh oh!

rgommers commented May 30, 2025

Uh oh!

abhishek-iitmadras commented Jun 1, 2025

Uh oh!

charris commented Jun 1, 2025

Uh oh!

abhishek-iitmadras commented Jun 1, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

charris commented Jun 2, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

charris commented Jun 2, 2025

Uh oh!

abhishek-iitmadras commented Jun 2, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

abhishek-iitmadras commented Jun 2, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

charris commented Jun 2, 2025

Uh oh!

charris commented Jun 2, 2025

Uh oh!

abhishek-iitmadras commented Jun 3, 2025

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

abhishek-iitmadras Jun 3, 2025 • 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!

r-devulap left a comment

Choose a reason for hiding this comment

abhishek-iitmadras commented May 14, 2025 •

edited

Loading

charris commented May 26, 2025 •

edited

Loading

abhishek-iitmadras commented Jun 1, 2025 •

edited

Loading

abhishek-iitmadras commented Jun 2, 2025 •

edited

Loading

abhishek-iitmadras commented Jun 2, 2025 •

edited

Loading

abhishek-iitmadras Jun 3, 2025 •

edited

Loading