Thanks for the relpy. I have opened an issue on github: 
https://github.com/golang/go/issues/61730

On Thursday, 3 August 2023 at 17:49:25 UTC+8 Michael Knyszek wrote:

> That line (the full sentence is "The garbage collector now includes 
> non-heap sources of garbage collector work (e.g., stack scanning) when 
> determining how frequently to run.") is unrelated. It only refers to a 
> change in accounting for what gets included in the GOGC calculation, not a 
> change in what was marked and scanned by the GC.
>
> On Thursday, August 3, 2023 at 1:41:35 AM UTC-4 Qingwei Li wrote:
>
>> I notice that Go1.17.7 still allocates the array on heap by calling 
>> newobject while Go1.18 allocates the array on stack. I also notice that in 
>> the release note of Go1.18 that "The garbage collector now includes 
>> non-heap sources of garbage collector work". Does the GC in 1.18 and 
>> following versions of Go ignore some global memory area when marking?
>>
>> On Thursday, August 3, 2023 at 1:03:31 AM UTC+8 Jinbao Chen wrote:
>>
>>> I use go1.20.5 to compile the following code. 
>>> package main
>>>
>>> func use(...interface{}) {
>>>   
>>> }
>>>
>>> func main() {
>>>     testCases := [...][][]int{
>>>         {{42}},
>>>         {{1, 2}},
>>>         {{3, 4, 5}},
>>>         {{}},
>>>         {{1, 2}, {3, 4, 5}, {}, {7}},
>>>     }
>>>     for _, testCase := range testCases {
>>>         use(testCase)
>>>     }
>>> }
>>> In the generated SSA and assembly code, I notice that the Go compiler 
>>> generates some instructions that store a stack pointer(point to the 
>>> stack-allocated array) into a global slice header.
>>>
>>> Just like the assembly code below, the MOV instruction at 0x4585bf 
>>> stores a stack pointer into a global object: 
>>>   0x458589 48c744240800000000       MOVQ $0x0, 0x8(SP) 
>>>   0x458592 48c74424082a000000 MOVQ $0x2a, 0x8(SP) 
>>> testCases := [...][][]int{
>>>   0x45859b 48c705c28e060001000000 MOVQ $0x1, 0x68ec2(IP) 
>>>   0x4585a6 48c705bf8e060001000000 MOVQ $0x1, 0x68ebf(IP) 
>>>   0x4585b1 833d988d090000 CMPL $0x0, runtime.writeBarrier(SB) 
>>>   0x4585b8 750e JNE 0x4585c8 
>>>   0x4585ba 488d442408 LEAQ 0x8(SP), AX 
>>>   0x4585bf 4889059a8e0600 MOVQ AX, 0x68e9a(IP) 
>>>   0x4585c6 eb11 JMP 0x4585d9 
>>>   0x4585c8 488d3d918e0600 LEAQ 0x68e91(IP), DI 
>>>   0x4585cf 488d442408 LEAQ 0x8(SP), AX 
>>>   0x4585d4 e8e7cfffff CALL runtime.gcWriteBarrier(SB) 
>>>
>>> I have read the comments in slicelit 
>>> <https://github.com/golang/go/blob/fb6f38dda15d4155b500f6b3e1a311a951a22b69/src/cmd/compile/internal/walk/complit.go#L288>,
>>>   
>>> but I didn't find any operations that can generate such stores. As far as I 
>>> know, pointers to stack objects cannot be stored in global objects. So is 
>>> this a compiler bug? Or the Go compiler does this on purpose to achieve 
>>> some optimization I don't know yet?
>>>
>>> Thanks
>>>
>>

-- 
You received this message because you are subscribed to the Google Groups 
"golang-nuts" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to golang-nuts+unsubscr...@googlegroups.com.
To view this discussion on the web visit 
https://groups.google.com/d/msgid/golang-nuts/e7fc85da-8df8-43ee-b611-c57cbc142fa9n%40googlegroups.com.

Reply via email to