allocs profiler: extend to record Types for all allocations - (fix `UnknownType` in allocations results) #43688

vilterp · 2022-01-06T20:56:30Z

The allocation profiler in #42768 doesn't provide Types for all allocations. There are two types of codepaths which allocations go through:

generated code calling jl_gc_alloc_(jl_ptls_t ptls, size_t sz, void *ty) or jl_gc_alloc_string in julia_internal.h: these functions know the type of the object being allocated.
- which then (sometimes) calls jl_gc_pool_alloc / jl_gc_big_alloc
generated code calling jl_gc_pool_alloc or jl_gc_big_alloc in gc.c directly: these functions do not know the type of the object being allocated.

The latter paths make it difficult to record the type of the allocation, since the type is not passed in — code is generated which calls it with just the size, and then sets the type later.

Another previous profiler PR, #33467, refactored the second codepath to take the type, and modifies the codegen to call the modified codepath, but hasn't been finished. For now, julia's Profile.Allocs functionality (introduced in #42768, #43868, #44043) doesn't do this.

Instead, for those types, we emit the type as Profile.Allocs.UnknownType in the allocations profile.

Experiments on a large codebase seem to show around ~40% of allocations have UnknownType, so the tool was still mostly useful. Still, getting the rest would be great!

Fixing this would require either passing the type of the allocation through to those functions by changing the codegen, or removing the instrumentation from those functions and instead emitting the instrumentation directly in the codegen.

The text was updated successfully, but these errors were encountered:

NHDaly · 2022-01-18T02:10:55Z

Just as an update: we are currently thinking that there would already be a lot of value in getting just the stacktraces, even if we can't yet get the types for those allocations. So @vilterp has been giving this a stab in vilterp#17.

Even if we solve that, though, we'll probably leave this issue open to cover future work to add the types for all allocation paths.

…cations Profiler Add _maybe_record_alloc_to_profile() call into jl_gc_big_alloc() to allow us to instrument it in the Allocations Profiler. Followup to #43868 Finishes fixing #43688 (gets the stacks, though not the types, of big allocs).

…cations Profiler Add _maybe_record_alloc_to_profile() call into jl_gc_big_alloc() to allow us to instrument it in the Allocations Profiler. Followup to #43868 Finishes fixing #43688 (gets the stacks, though not the types, of big allocs). ----- - Rename jl_gc_big_alloc to jl_gc_big_alloc_inner, make it inlined - add jl_gc_pool_alloc_noinline that calls jl_gc_big_alloc_inner - replace all internal calls to jl_gc_big_alloc to call this instead - add a new jl_gc_pool_alloc that calls jl_gc_big_alloc_inner - This one is instrumented, and is meant to be called by generated code.

NHDaly · 2022-02-07T17:44:54Z

I'm going to adjust this Issue to reflect that the Types are missing for some codepaths still.
(And some stack traces are still missing until #44043 is merged)

…cations Profiler (#44043) * Instrument jl_gc_big_alloc() when called from generated code for Allocations Profiler Add _maybe_record_alloc_to_profile() call into jl_gc_big_alloc() to allow us to instrument it in the Allocations Profiler. Followup to #43868 Finishes fixing #43688 (gets the stacks, though not the types, of big allocs). ----- - Rename jl_gc_big_alloc to jl_gc_big_alloc_inner, make it inlined - add jl_gc_pool_alloc_noinline that calls jl_gc_big_alloc_inner - replace all internal calls to jl_gc_big_alloc to call this instead - add a new jl_gc_pool_alloc that calls jl_gc_big_alloc_inner - This one is instrumented, and is meant to be called by generated code. Co-authored-by: Pete Vilter <[email protected]>

…cations Profiler (JuliaLang#44043) * Instrument jl_gc_big_alloc() when called from generated code for Allocations Profiler Add _maybe_record_alloc_to_profile() call into jl_gc_big_alloc() to allow us to instrument it in the Allocations Profiler. Followup to JuliaLang#43868 Finishes fixing JuliaLang#43688 (gets the stacks, though not the types, of big allocs). ----- - Rename jl_gc_big_alloc to jl_gc_big_alloc_inner, make it inlined - add jl_gc_pool_alloc_noinline that calls jl_gc_big_alloc_inner - replace all internal calls to jl_gc_big_alloc to call this instead - add a new jl_gc_pool_alloc that calls jl_gc_big_alloc_inner - This one is instrumented, and is meant to be called by generated code. Co-authored-by: Pete Vilter <[email protected]>

…cations Profiler (#44043) * Instrument jl_gc_big_alloc() when called from generated code for Allocations Profiler Add _maybe_record_alloc_to_profile() call into jl_gc_big_alloc() to allow us to instrument it in the Allocations Profiler. Followup to #43868 Finishes fixing #43688 (gets the stacks, though not the types, of big allocs). ----- - Rename jl_gc_big_alloc to jl_gc_big_alloc_inner, make it inlined - add jl_gc_pool_alloc_noinline that calls jl_gc_big_alloc_inner - replace all internal calls to jl_gc_big_alloc to call this instead - add a new jl_gc_pool_alloc that calls jl_gc_big_alloc_inner - This one is instrumented, and is meant to be called by generated code. Co-authored-by: Pete Vilter <[email protected]>

ojwoodford · 2022-11-10T16:48:42Z

I'm having trouble understanding if this has been addressed yet. I'm using Julia 1.8.2. I've been profiling the allocations of my code, and now 100% (of 160k allocations totalling 2.5MB) of them are UnknownType. Is this expected or not?

vilterp · 2022-11-10T16:52:47Z

Nope, hasn't been addressed yet. Need help from someone who knows the LLVM codegen stuff better. Will close the issue when it is addressed.

We usually see less than 100% UnknownType, but it does depend on the code being profiled 🤷‍♂️

In the meantime, hopefully looking at the line of code at the tip of the stack trace for an allocation will show you what type is being allocated.

Pass the types to the allocator functions. ------- Before this PR, we were missing the types for allocations in two cases: 1. allocations from codegen 2. allocations in `gc_managed_realloc_` The second one is easy: those are always used for buffers, right? For the first one: we extend the allocation functions called from codegen, to take the type as a parameter, and set the tag there. I kept the old interfaces around, since I think that they cannot be removed due to supporting legacy code? ------ An example of the generated code: ```julia %ptls_field6 = getelementptr inbounds {}**, {}*** %4, i64 2 %13 = bitcast {}*** %ptls_field6 to i8** %ptls_load78 = load i8*, i8** %13, align 8 %box = call noalias nonnull dereferenceable(32) {}* @ijl_gc_pool_alloc_typed(i8* %ptls_load78, i32 1184, i32 32, i64 4366152144) #7 ``` Fixes #43688. Fixes #45268. Co-authored-by: Valentin Churavy <[email protected]>

Pass the types to the allocator functions. ------- Before this PR, we were missing the types for allocations in two cases: 1. allocations from codegen 2. allocations in `gc_managed_realloc_` The second one is easy: those are always used for buffers, right? For the first one: we extend the allocation functions called from codegen, to take the type as a parameter, and set the tag there. I kept the old interfaces around, since I think that they cannot be removed due to supporting legacy code? ------ An example of the generated code: ```julia %ptls_field6 = getelementptr inbounds {}**, {}*** %4, i64 2 %13 = bitcast {}*** %ptls_field6 to i8** %ptls_load78 = load i8*, i8** %13, align 8 %box = call noalias nonnull dereferenceable(32) {}* @ijl_gc_pool_alloc_typed(i8* %ptls_load78, i32 1184, i32 32, i64 4366152144) #7 ``` Fixes JuliaLang#43688. Fixes JuliaLang#45268. Co-authored-by: Valentin Churavy <[email protected]>

This was referenced Jan 6, 2022

alloc profile: print warning message vilterp/julia#12

Merged

Allocation profiler #42768

Merged

NHDaly added GC Garbage collector profiler feature Indicates new feature / enhancement requests labels Jan 7, 2022

This was referenced Jan 7, 2022

alloc profiler: record pool allocs, even though they'll be missing a type vilterp/julia#14

Closed

alloc profile: print percentage of allocs missed vilterp/julia#15

Merged

vilterp mentioned this issue Jan 19, 2022

allocation profiler: get stacks for pool allocs via wrapper function #43868

Merged

NHDaly mentioned this issue Feb 4, 2022

Instrument jl_gc_big_alloc() when called from generated code for Allocations Profiler #44043

Merged

vtjnash closed this as completed in #43868 Feb 7, 2022

NHDaly reopened this Feb 7, 2022

NHDaly changed the title ~~alloc profiler: extend to cover all allocs~~ allocs profiler: extend to record Types for all allocations Feb 7, 2022

NHDaly changed the title ~~allocs profiler: extend to record Types for all allocations~~ allocs profiler: extend to record Types for all allocations - (fix UnknownType in allocations results) Feb 8, 2022

vilterp mentioned this issue Feb 8, 2022

alloc profiler: adjust warnings in docs; remove logged warning from implementation #44077

Merged

IanButterworth mentioned this issue Sep 13, 2022

Improve documentation for new allocation profiler #45268

Closed

NHDaly mentioned this issue Jul 7, 2023

Allocation Profiler: Types for all allocations #50337

Merged

vchuravy closed this as completed in #50337 Jul 31, 2023

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

allocs profiler: extend to record Types for all allocations - (fix `UnknownType` in allocations results) #43688

allocs profiler: extend to record Types for all allocations - (fix `UnknownType` in allocations results) #43688

vilterp commented Jan 6, 2022 •

edited by NHDaly

Loading

NHDaly commented Jan 18, 2022

NHDaly commented Feb 7, 2022

ojwoodford commented Nov 10, 2022

vilterp commented Nov 10, 2022 •

edited

Loading

allocs profiler: extend to record Types for all allocations - (fix UnknownType in allocations results) #43688

allocs profiler: extend to record Types for all allocations - (fix UnknownType in allocations results) #43688

Comments

vilterp commented Jan 6, 2022 • edited by NHDaly Loading

NHDaly commented Jan 18, 2022

NHDaly commented Feb 7, 2022

ojwoodford commented Nov 10, 2022

vilterp commented Nov 10, 2022 • edited Loading

allocs profiler: extend to record Types for all allocations - (fix `UnknownType` in allocations results) #43688

allocs profiler: extend to record Types for all allocations - (fix `UnknownType` in allocations results) #43688

vilterp commented Jan 6, 2022 •

edited by NHDaly

Loading

vilterp commented Nov 10, 2022 •

edited

Loading