Hi Paul, Perhaps my answer was a bit simplistic, and I admit that I answered without looking too much into it.
Basically, in your index1, the opts.isDigit passed to IndexFunc is > syntactic sugar for (&opts).isDigit -> then the compiler needs to move opts > on the heap. "*Why? Taking a reference to a function argument shouldn't automatically move it to the heap. As long as the function doesn't store or return the reference, it doesn't exist beyond the scope of that function. Escape analysis should be able to figure that out and leave the value on the stack. My real code passes &opts to several functions without issue. The only ones that cause a problem are the IndexFunc calls.*" I agree with you re: the escape analysis ought to able to detect that scenario - my initial reasoning here was that the compiler could be taking the safest approach for the following scenario: - the (&opts).isDigit function is passed to IndexFunc (or any other function). - that IndexFunc function may or may not diverge into concurrent execution (using go ... ) - while those detached execution flows are still running, the opts variable runs the risk of being popped out of its stack existence when index1 returns, so (&opts).isDigit would not be addressable from the other execution threads. That was my reasoning at the time I replied. But in the meanwhile, out of curiosity, I took IndexFunc and its private sister function, indexFunc out of the strings package, and added a convoluted go routine inside that private indexFunc, that takes the function as a parameter, and sleeps for a number of seconds. In the meanwhile the main execution flow returns. From what I'm seeing, from that moment on, index2 -> opts also (correctly) escapes to heap. This leaves me with agreeing with you and Roger that there's room for improvement. For your index2 function question, I think it's because Go maintains the > variables from a parent function that are referred in an enclosed function > (closure) for as long as both are alive. > I'm guessing that opts.IsDigit(r) is alive and kicking till it returns. "*Sorry, I'm not sure what you mean by this. Could you expand on it?*" What I meant is that the closure, that wrapper func you created inside index2 is aware of the surrounding scope, and stateful. It will have access to &opts for as long as it's alive. cheers silviu On Wed, Sep 5, 2018 at 8:15 AM <pauld...@gmail.com> wrote: > > FYI, here is the escape analysis output. > > > $ go build -gcflags="-l -m=2" > > ./escape.go:20:10: <S> capturing by ref: opts (addr=true assign=false > width=4) > ./escape.go:9:38: (*options).isDigit opts does not escape > ./escape.go:15:34: opts escapes to heap > ./escape.go:15:34: from opts.isDigit (call part) at ./escape.go:15:34 > ./escape.go:14:37: moved to heap: opts > ./escape.go:14:37: index1 s does not escape > ./escape.go:15:34: index1 opts.isDigit does not escape > ./escape.go:18:37: index2 s does not escape > ./escape.go:19:30: index2 func literal does not escape > ./escape.go:20:14: index2.func1 opts does not escape > > > Line 15 is the call to IndexFunc from the function index1. I think "call > part" is supposed to be the reason that opts escapes to the heap but I > couldn't find any information about what that means. > > > On Wednesday, September 5, 2018 at 7:03:12 AM UTC-5, Paul D wrote: >> >> Hi Silviu, >> >> Thanks for your reply. I'm not sure about the points you raise though. >> >> Basically, in your index1, the opts.isDigit passed to IndexFunc is >>> syntactic sugar for (&opts).isDigit -> then the compiler needs to move opts >>> on the heap. >> >> >> Why? Taking a reference to a function argument shouldn't automatically >> move it to the heap. As long as the function doesn't store or return the >> reference, it doesn't exist beyond the scope of that function. Escape >> analysis should be able to figure that out and leave the value on the >> stack. My real code passes &opts to several functions without issue. The >> only ones that cause a problem are the IndexFunc calls. >> >> For your index2 function question, I think it's because Go maintains the >>> variables from a parent function that are referred in an enclosed function >>> (closure) for as long as both are alive. >>> I'm guessing that opts.IsDigit(r) is alive and kicking till it returns. >> >> >> Sorry, I'm not sure what you mean by this. Could you expand on it? >> >> Thanks, >> Paul >> >> >> On Tuesday, September 4, 2018 at 11:05:39 PM UTC-5, Silviu Capota Mera >> wrote: >>> >>> Hi Paul, >>> >>> Check out: https://golang.org/doc/effective_go.html#pointers_vs_values >>> >>> Basically, in your index1, the opts.isDigit passed to IndexFunc is >>> syntactic sugar for (&opts).isDigit -> then the compiler needs to move opts >>> on the heap. >>> >>> You can either >>> a) define an alternate isDigitVal method, e.g. func (opts options) >>> isDigitMethodByVal(r rune) bool >>> or >>> b) you can just pass the reference to an original *options from the get >>> go: >>> >>> func index1Pointer(s string, optsPointer *options) int { >>> return strings.IndexFunc(s, optsPointer.isDigit) >>> } >>> >>> Not sure which one will be faster - you'll need to benchmark. Probably >>> both will be about the same, since that single *options allocation tends to >>> zero. >>> >>> For your index2 function question, I think it's because Go maintains the >>> variables from a parent function that are referred in an enclosed function >>> (closure) for as long as both are alive. >>> I'm guessing that opts.IsDigit(r) is alive and kicking till it returns. >>> >>> cheers, >>> silviu >>> >>> On Tuesday, 4 September 2018 17:46:09 UTC-4, Paul D wrote: >>>> >>>> I'm trying to reduce allocations (and improve performance) in some Go >>>> code. There's a recurring pattern in the code where a struct is passed to a >>>> function, and the function passes one of the struct's methods to >>>> strings.IndexFunc. For some reason, this causes the entire struct to escape >>>> to the heap. If I wrap the method call in an anonymous function, the struct >>>> does not escape and the benchmarks run about 30% faster. >>>> >>>> Here is a minimal example. In the actual code, the struct has more >>>> fields/methods and the function in question actually does something. But >>>> this sample code illustrates the problem. Why does the opts argument escape >>>> to the heap in index1 but not in the functionally equivalent index2? And is >>>> there a robust way to ensure that it stays on the stack? >>>> >>>> >>>> type options struct { >>>> zero rune >>>> } >>>> >>>> func (opts *options) isDigit(r rune) bool { >>>> r -= opts.zero >>>> return r >= 0 && r <= 9 >>>> } >>>> >>>> // opts escapes to heap >>>> func index1(s string, opts options) int { >>>> return strings.IndexFunc(s, opts.isDigit) >>>> } >>>> >>>> // opts does not escape to heap >>>> func index2(s string, opts options) int { >>>> return strings.IndexFunc(s, func(r rune) bool { >>>> return opts.isDigit(r) >>>> }) >>>> } >>>> >>>> >>>> FYI I'm running Go 1.10.3 on Linux. Thanks... >>>> >>>> >>>> -- > You received this message because you are subscribed to a topic in the > Google Groups "golang-nuts" group. > To unsubscribe from this topic, visit > https://groups.google.com/d/topic/golang-nuts/x9ZV-_XVaUk/unsubscribe. > To unsubscribe from this group and all its topics, send an email to > golang-nuts+unsubscr...@googlegroups.com. > For more options, visit https://groups.google.com/d/optout. > -- 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. For more options, visit https://groups.google.com/d/optout.
