cmd/asm: shift operands unsupported on arm64 (but supported on arm)

[williamweixiao]

Please answer these questions before submitting your issue. Thanks!

What version of Go are you using (go version)?

go version go1.7.3 linux/arm64

What operating system and processor architecture are you using (go env)?

GOARCH="arm64"
GOBIN=""
GOEXE=""
GOHOSTARCH="arm64"
GOHOSTOS="linux"
GOOS="linux"
GOPATH="/home/weixia01/workspace/golang/gocode"
GORACE=""
GOROOT="/home/weixia01/workspace/go-go1.7.3"
GOTOOLDIR="/home/weixia01/workspace/go-go1.7.3/pkg/tool/linux_arm64"
CC="gcc"
GOGCCFLAGS="-fPIC -pthread -fmessage-length=0 -fdebug-prefix-map=/tmp/go-build695199653=/tmp/go-build -gno-record-gcc-switches"
CXX="g++"
CGO_ENABLED="1"

What did you do?

Assemble following assembly code (shiftopd.s):

1
2TEXT test?shiftopd(SB),0,$0-0
3 ADD R1>>1, R2, R0
4 RET

with command:
go tool asm shiftopd.s

What did you expect to see?

Assemble successfully

What did you see instead?

shiftopd.s:3: expected end of operand, found >>
asm: assembly of shiftopd.s failed

[davecheney]

I'm pretty sure the ADD instruction in arm64 does not a register as a shifted operand. You can use a small immediate constant, but not a register, and certainly not as the first operand.

…

On Mon, Nov 28, 2016 at 1:15 PM, Wei Xiao ***@***.***> wrote: Please answer these questions before submitting your issue. Thanks! What version of Go are you using (go version)? go version go1.7.3 linux/arm64 What operating system and processor architecture are you using (go env)? GOARCH="arm64" GOBIN="" GOEXE="" GOHOSTARCH="arm64" GOHOSTOS="linux" GOOS="linux" GOPATH="/home/weixia01/workspace/golang/gocode" GORACE="" GOROOT="/home/weixia01/workspace/go-go1.7.3" GOTOOLDIR="/home/weixia01/workspace/go-go1.7.3/pkg/tool/linux_arm64" CC="gcc" GOGCCFLAGS="-fPIC -pthread -fmessage-length=0 -fdebug-prefix-map=/tmp/go-build695199653=/tmp/go-build -gno-record-gcc-switches" CXX="g++" CGO_ENABLED="1" What did you do? Assemble following assembly code (shiftopd.s): 1 2TEXT test?shiftopd(SB),0,$0-0 3 ADD R1>>1, R2, R0 4 RET with command: go tool asm shiftopd.s What did you expect to see? Assemble successfully What did you see instead? shiftopd.s:3: expected end of operand, found >> asm: assembly of shiftopd.s failed — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#18070>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAAcA3aQ4_z-7e9MKhXGHo3cU6TVO2_8ks5rCjlYgaJpZM4K9WAN> .

[davecheney]

My apologies, after studying the arm64 instruction manuals I did find ADD Wd, Wn, Wm{, ashift #imm} // Add (register): Wd = Wn + ashift(Wm, imm). Can you pleaes a. you said this was a regression, which versions of Go did this work on b. Upload your source file, the text pasted into the issue has some emojibake corruption. Thanks

…

On Mon, Nov 28, 2016 at 1:22 PM, Dave Cheney ***@***.***> wrote: I'm pretty sure the ADD instruction in arm64 does not a register as a shifted operand. You can use a small immediate constant, but not a register, and certainly not as the first operand. On Mon, Nov 28, 2016 at 1:15 PM, Wei Xiao ***@***.***> wrote: > Please answer these questions before submitting your issue. Thanks! > What version of Go are you using (go version)? > > go version go1.7.3 linux/arm64 > What operating system and processor architecture are you using (go env)? > > GOARCH="arm64" > GOBIN="" > GOEXE="" > GOHOSTARCH="arm64" > GOHOSTOS="linux" > GOOS="linux" > GOPATH="/home/weixia01/workspace/golang/gocode" > GORACE="" > GOROOT="/home/weixia01/workspace/go-go1.7.3" > GOTOOLDIR="/home/weixia01/workspace/go-go1.7.3/pkg/tool/linux_arm64" > CC="gcc" > GOGCCFLAGS="-fPIC -pthread -fmessage-length=0 > -fdebug-prefix-map=/tmp/go-build695199653=/tmp/go-build > -gno-record-gcc-switches" > CXX="g++" > CGO_ENABLED="1" > What did you do? > > Assemble following assembly code (shiftopd.s): > > 1 > 2TEXT test?shiftopd(SB),0,$0-0 > 3 ADD R1>>1, R2, R0 > 4 RET > > with command: > go tool asm shiftopd.s > What did you expect to see? > > Assemble successfully > What did you see instead? > > shiftopd.s:3: expected end of operand, found >> > asm: assembly of shiftopd.s failed > > — > You are receiving this because you are subscribed to this thread. > Reply to this email directly, view it on GitHub > <#18070>, or mute the thread > <https://github.com/notifications/unsubscribe-auth/AAAcA3aQ4_z-7e9MKhXGHo3cU6TVO2_8ks5rCjlYgaJpZM4K9WAN> > . >

[gopherbot]

CL https://golang.org/cl/33595 mentions this issue.

[williamweixiao]

a. according to go source code, shift operand is supported on arm but fail to work on armv8
b. i have added a test in: CL https://golang.org/cl/33595

[minux]

If it has never been supported by cmd/asm before, then it's not a regression. arm64 and arm are two completely different ISAs, so we certainly can't say this arm64 bug is regression from the arm support. It's just cmd/asm lacks shift operand support in arm64.

[williamweixiao]

There is a discussion about the issue: https://groups.google.com/forum/#!msg/golang-nuts/yKmWV0WZZiw/n3JkTTQhBQAJ;context-place=forum/golang-nuts
Aram said

They used to be supported but I think I dropped them with the old C assembler.

If we just consider the issue since golang implementation switching from c to go, lack shift operand support on arm64 maybe a more suitable title

[rsc]

I would like to wait until Go 1.9 for this, to give us more time to set up a proper test suite. This is the kind of thing that is easy to get wrong and hard to correct.

[rsc]

Ideally, the way this would proceed for Go 1.9 is the following. One thing that is missing from the arm64 port is a disassembler generated from a table of the instructions. See golang.org/x/arch's top-level directories for other examples. Once we have that, then we can generate a table of tests, commenting out the instructions that we don't have yet. Then when each new CL adds instructions it just uncomments the already-generated test, which gives us some confidence that the test is not just repeating what the CL implemented.

go/src/cmd/asm/internal/asm/testdata/amd64enc.s is an example of this.

[williamweixiao]

We have started developing arm64 disassembler and expect to finish it within about three months.
As for using disassembler to generate table of tests, do you know where is the "x86test"?