GO TIDBITS
Below are a few bits of information I'd like to remember when writing Go programs.
ANALYSIS REPORTING
Sometimes it's difficult to know how the compiler will reason about certain pieces of code. This command will report analysis/optimization decisions made during compilation.
$ go build -gcflags="-m" . <source>:5:6 can inline SomeProcedure <source>:8:4 can inline AnotherProcedure[go.shape.float64] <source>:8:4 ... argument does not escape <source>:10:2 foo escapes to heap <source>:10:5 bar escapes to heap
BOUNDS CHECK REPORTING
It's helpful to know when bounds checks are added to the generated code. This command will report any cases where this happens.
$ go build -gcflags="-d=ssa/check_bce/debug=1" . <source>:10:8: Found IsInBounds <source>:11:4: Found IsInBounds
When bounds checks are removed:
- When the length of a slice and the index are known at compile-time
- When a bounds check has been performed manually ahead of indexing
- When a hint is given ahead of indexing
- When compiling with
gcflags=-B(removes all bounds checks)
Code examples:
// Slice length and index are known at compile-time
func Get0th(data [255]int) int {
return data[0]
}
// Manual bounds check was performed before indexing
func GetNth(data [255]int, idx int) int {
if idx < 0 || idx >= len(data) {
return data[0]
}
return data[idx]
}
// Bounds hint
func Get1st(data []int) int {
_ = data[1]
return data[1]
}
SEE GENERATED ASSEMBLY
Besides using Compiler Explorer, Go can directly output human-readable assembly for a target platform with this command.
$ go tool compile -S -trimpath="$(realpath $1)" file.go main.add STEXT size=16 args=0x10 locals=0x0 funcid=0x0 align=0x0 leaf 0x0000 00000 (file.go:3) TEXT main.add(SB), LEAF|NOFRAME|ABIInternal, $0-16 0x0000 00000 (file.go:3) FUNCDATA $0, gclocals·g2BeySu+wFnoycgXfElmcg==(SB) 0x0000 00000 (file.go:3) FUNCDATA $1, gclocals·g2BeySu+wFnoycgXfElmcg==(SB) 0x0000 00000 (file.go:3) FUNCDATA $5, main.add.arginfo1(SB) 0x0000 00000 (file.go:3) FUNCDATA $6, main.add.argliveinfo(SB) 0x0000 00000 (file.go:3) PCDATA $3, $1 0x0000 00000 (file.go:4) ADD R1, R0, R0 0x0004 00004 (file.go:4) RET (R30)
Outputting assembly for a platform other than the host machine:
$ GOOS=js GOARCH=wasm go tool compile -S -trimpath="$(realpath $1)" file.go main.add STEXT size=2 args=0x18 locals=0x0 funcid=0x0 align=0x0 0x0000 00000 (file.go:3) TEXT main.add(SB), ABIInternal, $0-24 0x0000 00000 (file.go:3) Block 0x0000 00000 (file.go:3) Block 0x0000 00000 (file.go:3) Get PC_B 0x0000 00000 (file.go:3) BrTable 0x0000 00000 (file.go:3) End 0x0000 00000 (file.go:3) FUNCDATA $0, gclocals·g2BeySu+wFnoycgXfElmcg==(SB) 0x0000 00000 (file.go:3) FUNCDATA $1, gclocals·g2BeySu+wFnoycgXfElmcg==(SB) 0x0000 00000 (file.go:3) FUNCDATA $5, main.add.arginfo1(SB) 0x0000 00000 (file.go:4) Get SP 0x0000 00000 (file.go:4) I64ExtendI32U 0x0000 00000 (file.go:4) I64Const $24 0x0000 00000 (file.go:4) I64Add 0x0000 00000 (file.go:4) I32WrapI64 0x0000 00000 (file.go:4) Get SP 0x0000 00000 (file.go:4) I64Load $8 0x0000 00000 (file.go:4) Get SP 0x0000 00000 (file.go:4) I64Load $16 0x0000 00000 (file.go:4) I64Add 0x0000 00000 (file.go:4) I64Store $0 0x0000 00000 (file.go:4) Get SP 0x0000 00000 (file.go:4) I32Const $8 0x0000 00000 (file.go:4) I32Add 0x0000 00000 (file.go:4) Set SP 0x0000 00000 (file.go:4) I32Const $0 0x0000 00000 (file.go:4) Return 0x0001 00001 (file.go:4) End
CLEANING UP ASSEMBLY
If we use go tool -S to generate assembly for a bit of code, there's a few modifications we have to make to its output before the compiler can reuse it.
Starting with a regular Go file:
package main
func asm_add(x, y int) int {
return x + y
}
Update OS and ARCH then run this command:
$ GOOS=OS GOARCH=ARCH go tool compile -S -trimpath="$(realpath $1)" file.go > file_[ARCH].s
This will generate a file_[ARCH].s file (will look different depending on architecture):
main.asm_add STEXT size=16 args=0x10 locals=0x0 funcid=0x0 align=0x0 leaf 0x0000 00000 (asm.go:3) TEXT main.asm_add(SB), LEAF|NOFRAME|ABIInternal, $0-16 0x0000 00000 (asm.go:3) FUNCDATA $0, gclocals·g2BeySu+wFnoycgXfElmcg==(SB) 0x0000 00000 (asm.go:3) FUNCDATA $1, gclocals·g2BeySu+wFnoycgXfElmcg==(SB) 0x0000 00000 (asm.go:3) FUNCDATA $5, main.asm_add.arginfo1(SB) 0x0000 00000 (asm.go:3) FUNCDATA $6, main.asm_add.argliveinfo(SB) 0x0000 00000 (asm.go:3) PCDATA $3, $1 0x0000 00000 (asm.go:4) ADD R1, R0, R0 0x0004 00004 (asm.go:4) RET (R30) 0x0000 00 00 01 8b c0 03 5f d6 00 00 00 00 00 00 00 00 ......_......... go:cuinfo.producer.<unlinkable> SDWARFCUINFO dupok size=0 0x0000 72 65 67 61 62 69 regabi go:cuinfo.packagename.main SDWARFCUINFO dupok size=0 0x0000 6d 61 69 6e main main..inittask SNOPTRDATA size=8 0x0000 00 00 00 00 00 00 00 00 ........ gclocals·g2BeySu+wFnoycgXfElmcg== SRODATA dupok size=8 0x0000 01 00 00 00 00 00 00 00 ........ main.asm_add.arginfo1 SRODATA static dupok size=5 0x0000 00 08 08 08 ff ..... main.asm_add.argliveinfo SRODATA static dupok size=2 0x0000 00 00 ..
Make the following modifications:
- Remove the first line (unnecessary preamble)
- Remove all lines after the final return instruction
- Remove
FUNCDATAandPCDATAinstruction lines - Remove prefix addresses and file locations
- Remove addressing modes (
LEAF,NOFRAME, etc.) from theTEXTline - Change any instance of
main.nameto·name - Change any instance of
namespace.nametonamespace·name
The file should now resemble:
TEXT ·asm_add(SB), $0-16 ADD R1, R0, R0 RET (R30)
Finally, remove the function body from the original Go file:
package main func asm_add(x, y int) int
asm_add can now be called from Go and the assembly version will be used (if compiling for that specific architecture).
GitHub issue around simplifying this process.