@zhangfannie

cc @golang/arm @golang/compiler

This doesn't need to be a proposal. Feel free to just send a CL adding the instruction. Thanks.

If I could, then I would @cherrymui. I would do the entire RISC set and get it over with actually.

I'm refactoring the arm64 assembler, which is expected to support the vast majority of missing instructions.

By the way, VTST is an Armv7 instruction, not an Arm64 instruction, I want to confirm with you that it is the instruction you want? Armv8 has a similar instruction CMTST, are you looking for it?

I was wondering why none of the opcodes from the "Alphabetical list of A64 Advanced SIMD and floating-point instructions" of the "Arm Architecture Reference Manual for A-profile architecture" match func oprrr. Starting to make sense now—wrong document. 😄

For the zig-zag usecase, I need arithmetic shifts int64(x) >> 63, int32(x) >> 31 and int16(x) >> 15. From the documentation, it seems both CMTST and VTST could achieve the same results, yes. 👍

Also happy to to help with the refactoring. Could start with some basic comments. Refer to the correct manual and such. 🤭 Make func asmout append (the right amount of) bytes perhaps? Let me know if that would be of any use. No is a perfectly fine answer. pascal@quies.net

For the zig-zag usecase, I need arithmetic shifts int64(x) >> 63, int32(x) >> 31 and int16(x) >> 15. From the documentation, it seems both CMTST and VTST could achieve the same results, yes. 👍

Glad to hear this. CMTST has been supported already, it's called VCMTST in go, here are two use cases:

	VCMTST	V2.B8, V29.B8, V2.B8            // a28f220e
	VCMTST	V2.D2, V23.D2, V3.D2            // e38ee24e

I'll close this issue, feel free to reopen it if there's any further comment, thanks.

Any luck with the refactoring yet @erifan?
I managed to produce something consistent now with the right manual 😁 yet something is still missing to come to live. https://go-review.googlesource.com/c/go/+/509575

@pascaldekloe The refactoring work is still in progress, and I'm sorry that the instructions you want are still not supported yet. If you are eager to use them, I suggest you temporarily use "DWORD <32-bit instruction binary>" to work around, because now it is the release cycle, we can't merge such changes into the master branch.

The refactoring changes are here https://go-review.googlesource.com/c/go/+/424137

I'm not here to put any pressure on you Eric. Better get this right than fast. B.t.w., Go's 0b notation, including _ separation works really nice with opcodes I think.