- Notifications
You must be signed in to change notification settings - Fork288
Enable compiling arm/neon with MSVC for windows on arm64#612
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to ourterms of service andprivacy statement. We’ll occasionally send you account related emails.
Already on GitHub?Sign in to your account
base:master
Are you sure you want to change the base?
Uh oh!
There was an error while loading.Please reload this page.
Conversation
JohanMabille commentedOct 19, 2021
Can you rebase your PR on the master branch please? This would make the diff easier to read (and would solve the conflicts). |
8e7a023 toa82ed95Compareniyas-sait commentedOct 19, 2021
Yes sure |
niyas-sait commentedOct 20, 2021
The change might be a bit invasive. Let me know if you need help with the review. Another approach to support MSVC would be to wrap the vector types in a custom class so that all types like float32x4_t and int32x4_t can be distinguished by the dispatcher but it would require too many conversions from the wrapper class to the native class before passing onto intrinsic functions etc which can be done with the user-defined operator but I guess performance cost will be there. |
serge-sans-paille commentedOct 23, 2021
We cannot do such a change while not setting up CI for that platform + arch combination. i'll try to prepare that. |
| #else | ||
| #defineXSIMD_WITH_NEON640 | ||
| #endif | ||
| #elif defined(_MSC_VER) && defined(_M_ARM64) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others.Learn more.
There's a specific section for MSVC workarounds at the bottom. Also, fromMSVC docs it is implied that NEON support is always available.
| #elif defined(_MSC_VER) && defined(_M_ARM64) | |
| #elif defined(_MSC_VER) | |
| #if defined(_M_ARM64) | |
| #defineXSIMD_WITH_NEON641 | |
| #else | |
| #defineXSIMD_WITH_NEON1 | |
| #endif |
This patch enables building arm/neon with MSVC compiler for windows on arm64 target and contains following changes,
Replace the dispatcher mechanism and use an explicit function selection using scalar types. This is required as MSVC intrinsics uses the same underlying type for multiple neon vector types and function selection using target vector type causes the wrong function to be called.
Add a function to convert
Initializer_list<batch<T>>to neon vector type as there are no constructors provided for the same operation in MSVC.NEON/NEON64 identification using MSVC specific flags
Add a
_to intrinsics wrapper functions. MSVC defines some intrinsics using macros and without the prefix the wrapper function names get replaced by the pre-processor.