Support for arm64 msvc

[talregev]

From our conversion here:
#2607 (comment)

This is ongoing issue.
I posting here a compilation errors and warning while compile flint with arm64 msvc.

Warning:
/FoCMakeFiles\flint.dir\src\generic_files\profiler.c.obj /FdCMakeFiles\flint.dir\ /FS -c D:\b\flint\src\flint-3-44df058a70.clean\src\generic_files\profiler.c
D:\b\flint\src\flint-3-44df058a70.clean\src\profiler.h(20): warning C4163: '__rdtsc': not available as an intrinsic function

Error:
C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Tools\MSVC\14.44.35207\include\immintrin.h(15): fatal error C1189: #error:  This header is specific to X86, X64, ARM64, and ARM64EC targets

This error occur after I patch flint that it will not include immintrin.h in arm64 msvc:

[175/5070] C:\PROGRA~1\MICROS~1\2022\ENTERP~1\VC\Tools\MSVC\1444~1.352\bin\Hostx64\arm64\cl.exe   -DFLINT_BUILD_DLL -Dflint_EXPORTS -ID:\b\flint\src\flint-3-88d753b0f5.clean\src -ID:\b\flint\arm64-windows-dbg -external:ID:\installed\arm64-windows\include -external:W0 /nologo /DWIN32 /D_WINDOWS /utf-8 /MP  /MDd /Z7 /Ob0 /Od /RTC1  -std:c11 -MDd /experimental:c11atomics /showIncludes /FoCMakeFiles\flint.dir\src\mpn_extras\mul.c.obj /FdCMakeFiles\flint.dir\ /FS -c D:\b\flint\src\flint-3-88d753b0f5.clean\src\mpn_extras\mul.c
FAILED: [code=2] CMakeFiles/flint.dir/src/mpn_extras/mul.c.obj 
C:\PROGRA~1\MICROS~1\2022\ENTERP~1\VC\Tools\MSVC\1444~1.352\bin\Hostx64\arm64\cl.exe   -DFLINT_BUILD_DLL -Dflint_EXPORTS -ID:\b\flint\src\flint-3-88d753b0f5.clean\src -ID:\b\flint\arm64-windows-dbg -external:ID:\installed\arm64-windows\include -external:W0 /nologo /DWIN32 /D_WINDOWS /utf-8 /MP  /MDd /Z7 /Ob0 /Od /RTC1  -std:c11 -MDd /experimental:c11atomics /showIncludes /FoCMakeFiles\flint.dir\src\mpn_extras\mul.c.obj /FdCMakeFiles\flint.dir\ /FS -c D:\b\flint\src\flint-3-88d753b0f5.clean\src\mpn_extras\mul.c
D:\b\flint\src\flint-3-88d753b0f5.clean\src\machine_vectors.h(988): error C2088: built-in operator '[' cannot be applied to an operand of type 'vec2d'
D:\b\flint\src\flint-3-88d753b0f5.clean\src\machine_vectors.h(988): warning C4033: 'vec2d_get_index' must return a value
D:\b\flint\src\flint-3-88d753b0f5.clean\src\machine_vectors.h(1104): error C2088: built-in operator '-' cannot be applied to an operand of type '__n128'
D:\b\flint\src\flint-3-88d753b0f5.clean\src\machine_vectors.h(1104): warning C4033: 'vec2d_fmsub' must return a value
D:\b\flint\src\flint-3-88d753b0f5.clean\src\machine_vectors.h(1108): error C2088: built-in operator '-' cannot be applied to an operand of type '__n128'
D:\b\flint\src\flint-3-88d753b0f5.clean\src\machine_vectors.h(1108): warning C4033: 'vec2d_fnmsub' must return a value
D:\b\flint\src\flint-3-88d753b0f5.clean\src\machine_vectors.h(1434): error C2088: built-in operator '[' cannot be applied to an operand of type 'vec2d'
D:\b\flint\src\flint-3-88d753b0f5.clean\src\machine_vectors.h(1434): error C2088: built-in operator '[' cannot be applied to an operand of type 'vec2d'
D:\b\flint\src\flint-3-88d753b0f5.clean\src\machine_vectors.h(1434): error C2088: built-in operator '[' cannot be applied to an operand of type 'vec2d'
D:\b\flint\src\flint-3-88d753b0f5.clean\src\machine_vectors.h(1434): error C2088: built-in operator '[' cannot be applied to an operand of type 'vec2d'
D:\b\flint\src\flint-3-88d753b0f5.clean\src\machine_vectors.h(1434): warning C4033: 'vec2d_same' must return a value

[talregev]

@albinahlback FYI

[albinahlback]

D:\b\flint\src\flint-3-44df058a70.clean\src\profiler.h(20): warning C4163: '__rdtsc': not available as an intrinsic function

Seems to be a fault on our end here. Should be easily fixed.

C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Tools\MSVC\14.44.35207\include\immintrin.h(15): fatal error C1189: #error:  This header is specific to X86, X64, ARM64, and ARM64EC targets

This is an aarch64 target, no? Why does it complain then?

machine_vectors.h errors

So you are saying that we cannot ask for index on an float64x2_t in MSVC?

[talregev]

I don't know how to solve these errors. I just present them for you without a solution.

[talregev]

This is an aarch64 target, no? Why does it complain then?

I really don't know why it complain. I try to solve them with AI, but it take time.

[talregev]

I let the AI create a solution. He create a patch that compile flint for arm64 msvc.
I can make a PR if you interested.

[talregev]

Seems to be a fault on our end here. Should be easily fixed.

Can you create a PR that fix it?

[oscarbenjamin]

I was seeing the same issue in flintlib/python-flint#379 and found that just removing the #include fixed it there:

diff --git a/src/longlong_msc_arm64.h b/src/longlong_msc_arm64.h
index 44d96f03f..6ade48289 100644
--- a/src/longlong_msc_arm64.h
+++ b/src/longlong_msc_arm64.h
@@ -14,7 +14,6 @@
 
 #include <stdlib.h>
 #include <intrin.h>
-#include <immintrin.h>
 
 /* Trailing and leading zeros */
 # define flint_clz _CountLeadingZeros64

The situation there is a bit different though as it uses msys2's clangarm64 toolchain to build FLINT and then uses MSVC to build Python extension modules (DLLs) that link with libflint.dll. It was the MSVC compile step that failed with

immintrin.h(15): fatal error C1189: #error:  This header is specific to X86, X64, ARM64, and ARM64EC targets

The patch results in both FLINT and python-flint building successfully and passing all tests on Windows arm64. Or at least it seems to be working in CI (I don't have direct access to a Windows arm64 machine).

Note that that is only tested with FLINT 3.3.1. I will update to FLINT 3.4.0 once the win-arm64 PR is finnished.

[talregev]

@oscarbenjamin That is a nice news.
There is also these errors:

[175/5070] C:\PROGRA~1\MICROS~1\2022\ENTERP~1\VC\Tools\MSVC\1444~1.352\bin\Hostx64\arm64\cl.exe   -DFLINT_BUILD_DLL -Dflint_EXPORTS -ID:\b\flint\src\flint-3-88d753b0f5.clean\src -ID:\b\flint\arm64-windows-dbg -external:ID:\installed\arm64-windows\include -external:W0 /nologo /DWIN32 /D_WINDOWS /utf-8 /MP  /MDd /Z7 /Ob0 /Od /RTC1  -std:c11 -MDd /experimental:c11atomics /showIncludes /FoCMakeFiles\flint.dir\src\mpn_extras\mul.c.obj /FdCMakeFiles\flint.dir\ /FS -c D:\b\flint\src\flint-3-88d753b0f5.clean\src\mpn_extras\mul.c
FAILED: [code=2] CMakeFiles/flint.dir/src/mpn_extras/mul.c.obj 
C:\PROGRA~1\MICROS~1\2022\ENTERP~1\VC\Tools\MSVC\1444~1.352\bin\Hostx64\arm64\cl.exe   -DFLINT_BUILD_DLL -Dflint_EXPORTS -ID:\b\flint\src\flint-3-88d753b0f5.clean\src -ID:\b\flint\arm64-windows-dbg -external:ID:\installed\arm64-windows\include -external:W0 /nologo /DWIN32 /D_WINDOWS /utf-8 /MP  /MDd /Z7 /Ob0 /Od /RTC1  -std:c11 -MDd /experimental:c11atomics /showIncludes /FoCMakeFiles\flint.dir\src\mpn_extras\mul.c.obj /FdCMakeFiles\flint.dir\ /FS -c D:\b\flint\src\flint-3-88d753b0f5.clean\src\mpn_extras\mul.c
D:\b\flint\src\flint-3-88d753b0f5.clean\src\machine_vectors.h(988): error C2088: built-in operator '[' cannot be applied to an operand of type 'vec2d'
D:\b\flint\src\flint-3-88d753b0f5.clean\src\machine_vectors.h(988): warning C4033: 'vec2d_get_index' must return a value
D:\b\flint\src\flint-3-88d753b0f5.clean\src\machine_vectors.h(1104): error C2088: built-in operator '-' cannot be applied to an operand of type '__n128'
D:\b\flint\src\flint-3-88d753b0f5.clean\src\machine_vectors.h(1104): warning C4033: 'vec2d_fmsub' must return a value
D:\b\flint\src\flint-3-88d753b0f5.clean\src\machine_vectors.h(1108): error C2088: built-in operator '-' cannot be applied to an operand of type '__n128'
D:\b\flint\src\flint-3-88d753b0f5.clean\src\machine_vectors.h(1108): warning C4033: 'vec2d_fnmsub' must return a value
D:\b\flint\src\flint-3-88d753b0f5.clean\src\machine_vectors.h(1434): error C2088: built-in operator '[' cannot be applied to an operand of type 'vec2d'
D:\b\flint\src\flint-3-88d753b0f5.clean\src\machine_vectors.h(1434): error C2088: built-in operator '[' cannot be applied to an operand of type 'vec2d'
D:\b\flint\src\flint-3-88d753b0f5.clean\src\machine_vectors.h(1434): error C2088: built-in operator '[' cannot be applied to an operand of type 'vec2d'
D:\b\flint\src\flint-3-88d753b0f5.clean\src\machine_vectors.h(1434): error C2088: built-in operator '[' cannot be applied to an operand of type 'vec2d'
D:\b\flint\src\flint-3-88d753b0f5.clean\src\machine_vectors.h(1434): warning C4033: 'vec2d_same' must return a value

I try to solve them with AI in here:
#2609

If you can take a look and let me know, or other ideas how to solve it.
Thank you!

[oscarbenjamin]

There is also these errors:

Yes, so I think those are because you are actually trying to build FLINT itself with MSVC which is different from what I am doing.

I could get away with just removing the include from longlong_msc_arm64.h because I am not using that header when building FLINT with the clangarm64 toolchain which will use longlong_asm_clang.h. It is after that that I use MSVC to build DLLs that depend on FLINT and then longlong_msc_arm64.h was included and it includes immintrin.h which for some reason fails with

immintrin.h(15): fatal error C1189: #error: This header is specific to X86, X64, ARM64, and ARM64EC targets

This is a very confusing error message, given that we on a ARM64 system compiling for an ARM64 target. I don't know how to see the contents of that header to see the logic leading up to line 15. Since the code I am compiling doesn't need any of those intrinsics I can get away with just patching FLINT not to include that header.

When you use MSVC to build FLINT itself that won't work because some of the code being compiled actually needs the contents of that header to compile. I don't know enough about MSVC or the intrinsics in use here to identify exactly what the problem is. It seems odd that just including the header leads to this error but maybe there is some preprocessor logic around that line 15. The error code C1189 is just the generic code that comes from a preprocessor #error.

I think that the header just has something like

#if !defined(_M_X86) && !defined(_M_X64) && ...
#error This header is specific to X86, X64 ...
#endif

So I guess the issue here is that _M_ARM64 is not defined but I'm not sure why it would not be defined. The MSVC docs say it is "Defined as 1 for compilations that target ARM64. Otherwise, undefined." but in that case it should be defined as 1 for my build.

Maybe the wrong arguments are being passed to MSVC somehow but at least without that include it seems to generate working code so it shouldn't be that it is just compiling for the wrong architecture or something like that. Your build log has:

[175/5070] C:\PROGRA~1\MICROS~1\2022\ENTERP~1\VC\Tools\MSVC\1444~1.352\bin\Hostx64\arm64\cl.exe   -DFLINT_BUILD_DLL -Dflint_EXPORTS -ID:\b\flint\src\flint-3-88d753b0f5.clean\src -ID:\b\flint\arm64-windows-dbg -external:ID:\installed\arm64-windows\include -external:W0 /nologo /DWIN32 /D_WINDOWS /utf-8 /MP  /MDd /Z7 /Ob0 /Od /RTC1  -std:c11 -MDd /experimental:c11atomics /showIncludes /FoCMakeFiles\flint.dir\src\mpn_extras\mul.c.obj /FdCMakeFiles\flint.dir\ /FS -c D:\b\flint\src\flint-3-88d753b0f5.clean\src\mpn_extras\mul.c
FAILED: [code=2] CMakeFiles/flint.dir/src/mpn_extras/mul.c.obj 

Mine looks like

"cl" "-Isrc/flint\pyflint.cp311-win_arm64.pyd.p" "-Isrc/flint" "-I..\src\flint" "-IC:\Users\runneradmin\AppData\Local\pypa\cibuildwheel\Cache\nuget-cpython\pythonarm64.3.11.9\tools\Include" "-IC:/a/python-flint/python-flint/.local/include" "-DNDEBUG" "/MD" "/nologo" "/showIncludes" "/utf-8" "/W2" "/std:c11" "/O2" "/Gw" "-DMS_WIN64=" "/Fdsrc/flint\pyflint.cp311-win_arm64.pyd.p\meson-generated_src_flint_pyflint.pyx.c.pdb" /Fosrc/flint/pyflint.cp311-win_arm64.pyd.p/meson-generated_src_flint_pyflint.pyx.c.obj "/c" src/flint/pyflint.cp311-win_arm64.pyd.p/src/flint/pyflint.pyx.c

I don't know enough about those command line options to say if any of them should affect this. In your command it looks you are running Hostx64\arm64\cl.exe. I'm not sure whether you are cross-compiling for Windows on ARM or if that is the path to the correct compiler for that. I was running the build as a native build on a Windows arm64 CI runner.