some updates and fixes to inlining cost #27857

JeffBezanson · 2018-06-29T04:23:42Z

Instead of always inlining functions marked @inline, increase the cost threshold ~~10x~~ 20x
Don't inline functions inferred not to return
statement_cost no longer needs to look at nested Exprs in general
Fix cost of :copyast

This also fixes the regression in mapslices (#27417). It was due to inlining a function marked @inline, even though the argument types were abstract and the body contained many dynamic calls. I believe most compilers just increase the threshold (by a lot) for functions marked inline, and have a separate always_inline to truly force inlining. We should do the same thing; otherwise we end up inlining silly things the user is unlikely to have intended.

JeffBezanson · 2018-06-29T04:49:10Z

@nanosoldier runbenchmarks(ALL, vs = ":master")

KristofferC · 2018-06-29T07:20:51Z

I believe most compilers just increase the threshold (by a lot) for functions marked inline, and have a separate always_inline to truly force inlining.

Can we get that always_inline then? Because when the inliner gets it "wrong", the effect can be dramatic and not having an escape hatch seems like a step backwards.

nanosoldier · 2018-06-29T10:16:30Z

Your benchmark job has completed - possible performance regressions were detected. A full report can be found here. cc @ararslan

JeffBezanson · 2018-06-29T17:49:25Z

Yes, you can have always_inline, but beware: if we can't infer the types of arguments (f(x::Any)), we will inline the function anyway, with everything in it dynamically dispatched, instead of dispatching to an optimized version. That is a dramatic effect, and probably not what you want.

timholy · 2018-06-29T18:53:08Z

I suppose another alternative is to aggressively eliminate many current @inline annotations, and save the current @inline to mean always_inline. The data in #22210 (comment) seem to suggest that might be feasible.

But there's a lot of sense in this proposal.

timholy · 2018-06-29T18:54:36Z

base/compiler/optimize.jl

        return 0
-    end
-    argcost = 0
-    for a in ex.args


What's the motivation for making this non-recursive?

The structure of the IR has changed such that calls no longer contain other calls, so it's just unnecessary.

JeffBezanson · 2018-06-29T22:02:21Z

Performance summary: some nice speedups from avoiding inlining things with dynamic dispatches, but some big slowdowns in find functions. The slowdowns were mostly related to the use of invoke in deprecations, so I just changed that.

JeffBezanson · 2018-06-29T22:02:42Z

@nanosoldier runbenchmarks(ALL, vs = ":master")

nanosoldier · 2018-06-29T22:47:26Z

Something went wrong when running your job:

NanosoldierError: failed to run benchmarks against primary commit: failed process: Process(`sudo cset shield -e su nanosoldier -- -c ./benchscript.sh`, ProcessExited(1)) [1]

Logs and partial data can be found here
cc @ararslan

JeffBezanson · 2018-06-30T04:24:17Z

@nanosoldier runbenchmarks(ALL, vs = ":master")

nanosoldier · 2018-06-30T10:51:19Z

Your benchmark job has completed - possible performance regressions were detected. A full report can be found here. cc @ararslan

JeffBezanson · 2018-07-05T21:42:51Z

Fixes #26446

JeffBezanson · 2018-07-05T23:21:55Z

@nanosoldier runbenchmarks("collection" && "queries & updates", vs = ":master")

nanosoldier · 2018-07-06T00:34:19Z

Your benchmark job has completed - possible performance regressions were detected. A full report can be found here. cc @ararslan

JeffBezanson · 2018-07-06T03:03:39Z

@nanosoldier runbenchmarks("collection" && "queries & updates", vs = ":master")

nanosoldier · 2018-07-06T04:15:55Z

Your benchmark job has completed - possible performance regressions were detected. A full report can be found here. cc @ararslan

JeffBezanson · 2018-07-06T16:40:25Z

@nanosoldier runbenchmarks("string" && "readuntil", vs = ":master")

nanosoldier · 2018-07-06T21:28:52Z

Your benchmark job has completed - no performance regressions were detected. A full report can be found here. cc @ararslan

vtjnash · 2018-07-06T23:41:08Z

base/compiler/optimize.jl

-                end
-                return plus_saturate(argcost, T_IFUNC_COST[iidx])
+        elseif ftyp isa DataType && isdefined(ftyp, :instance)
+            f = ftyp.instance


This branch should be dead-code

(also, it's available as the singleton_type function)

This might indicate a problem elsewhere, but it's not dead code --- in fact I needed this to fix a regression in an earlier version of the PR. A call to Core.sizeof had typeof(Core.sizeof) as the type of the function.

vtjnash · 2018-07-06T23:41:52Z

base/compiler/optimize.jl

+        end
+        a = ex.args[2]
+        if a isa Expr
+            cost += statement_cost(a, -1, src, spvals, slottypes, params)


cost = plus_saturate(cost, ...)

- Instead of always inlining functions marked at-inline, increase the cost threshold 20x - Don't inline functions inferred not to return - statement_cost no longer needs to look at nested Exprs in general - Fix cost of `:copyast`

KristofferC · 2018-09-13T17:44:40Z

Bump for @forceinline...

Issues like JuliaArrays/StaticArrays.jl#494 are quite frustrating when you always have to worry that the threshold might have gotten reached.

JeffBezanson · 2018-09-13T18:17:24Z

There might be something wrong with our cost model in this case --- a few dozen arithmetic ops should not be a problem. I'm worried about dynamic dispatches that take thousands of times longer than that. We can add forceinline as well though.

KristofferC · 2018-09-13T18:22:49Z

The example there has a SIMD loop though so might be why it doesn't inline. Anyway, having a way to force would be good in the inevitable cases where the cost model gets it wrong.

vtjnash · 2018-09-13T18:38:59Z

If the function type is maximally typed (isdispatchtuple(specTypes)), we could simply always obey the user annotation. In that case, you're already then at least guaranteed that keeping the dispatch barrier won't improve type information.

This implements the suggestion in #27857 (comment)

…29258) This implements the suggestion in #27857 (comment)

…29258) This implements the suggestion in #27857 (comment) (cherry picked from commit 8d6c1ce)

JeffBezanson added the optimizer Optimization passes (mostly in base/compiler/ssair/) label Jun 29, 2018

martinholters mentioned this pull request Jun 29, 2018

Let tmerge form a Union more often #27843

Merged

timholy reviewed Jun 29, 2018

View reviewed changes

JeffBezanson force-pushed the jb/inliningcostfixes branch from 5acca3f to 6cb12f0 Compare June 29, 2018 21:57

JeffBezanson force-pushed the jb/inliningcostfixes branch from 6cb12f0 to 19d1f8b Compare June 30, 2018 04:23

JeffBezanson force-pushed the jb/inliningcostfixes branch from 19d1f8b to 514767a Compare July 5, 2018 20:09

JeffBezanson force-pushed the jb/inliningcostfixes branch from 514767a to 8f2e0fb Compare July 6, 2018 03:03

vtjnash reviewed Jul 6, 2018

View reviewed changes

some updates and fixes to inlining cost

e246245

- Instead of always inlining functions marked at-inline, increase the cost threshold 20x - Don't inline functions inferred not to return - statement_cost no longer needs to look at nested Exprs in general - Fix cost of `:copyast`

JeffBezanson force-pushed the jb/inliningcostfixes branch from 8f2e0fb to e246245 Compare July 7, 2018 16:40

JeffBezanson merged commit 9277d3a into master Jul 7, 2018

JeffBezanson deleted the jb/inliningcostfixes branch July 7, 2018 19:47

JeffBezanson mentioned this pull request Jul 10, 2018

performance regression in findfirst due to inlining #26446

Closed

maleadt mentioned this pull request Jul 12, 2018

Codegen regression with llvmcall/inlining #28078

Closed

KristofferC mentioned this pull request Jul 22, 2018

View allocation elimination stumped by conditional? #28226

Closed

JeffBezanson added a commit that referenced this pull request Sep 18, 2018

always obey inline declarations if the calling signature is concrete

b9c18a2

This implements the suggestion in #27857 (comment)

JeffBezanson mentioned this pull request Sep 18, 2018

always obey inline declarations if the calling signature is concrete #29258

Merged

JeffBezanson added a commit that referenced this pull request Sep 25, 2018

always obey inline declarations if the calling signature is concrete (#…

8d6c1ce

…29258) This implements the suggestion in #27857 (comment)

KristofferC pushed a commit that referenced this pull request Sep 30, 2018

always obey inline declarations if the calling signature is concrete (#…

2f9e279

…29258) This implements the suggestion in #27857 (comment) (cherry picked from commit 8d6c1ce)

KristofferC pushed a commit that referenced this pull request Feb 11, 2019

always obey inline declarations if the calling signature is concrete (#…

65cb309

…29258) This implements the suggestion in #27857 (comment) (cherry picked from commit 8d6c1ce)

KristofferC pushed a commit that referenced this pull request Feb 20, 2020

always obey inline declarations if the calling signature is concrete (#…

4afbacb

…29258) This implements the suggestion in #27857 (comment) (cherry picked from commit 8d6c1ce)

kpamnany mentioned this pull request Jul 27, 2021

Perf: Specialize fn in rmap and rmaptype CliMA/ClimaCore.jl#80

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

some updates and fixes to inlining cost #27857

some updates and fixes to inlining cost #27857

JeffBezanson commented Jun 29, 2018 •

edited

JeffBezanson commented Jun 29, 2018

KristofferC commented Jun 29, 2018 •

edited

nanosoldier commented Jun 29, 2018

JeffBezanson commented Jun 29, 2018

timholy commented Jun 29, 2018

timholy Jun 29, 2018

JeffBezanson Jun 29, 2018

JeffBezanson commented Jun 29, 2018

JeffBezanson commented Jun 29, 2018

nanosoldier commented Jun 29, 2018

JeffBezanson commented Jun 30, 2018

nanosoldier commented Jun 30, 2018

JeffBezanson commented Jul 5, 2018

JeffBezanson commented Jul 5, 2018

nanosoldier commented Jul 6, 2018

JeffBezanson commented Jul 6, 2018

nanosoldier commented Jul 6, 2018

JeffBezanson commented Jul 6, 2018

nanosoldier commented Jul 6, 2018

vtjnash Jul 6, 2018

JeffBezanson Jul 6, 2018

vtjnash Jul 6, 2018

KristofferC commented Sep 13, 2018

JeffBezanson commented Sep 13, 2018

KristofferC commented Sep 13, 2018 •

edited

vtjnash commented Sep 13, 2018

some updates and fixes to inlining cost #27857

some updates and fixes to inlining cost #27857

Conversation

JeffBezanson commented Jun 29, 2018 • edited

JeffBezanson commented Jun 29, 2018

KristofferC commented Jun 29, 2018 • edited

nanosoldier commented Jun 29, 2018

JeffBezanson commented Jun 29, 2018

timholy commented Jun 29, 2018

timholy Jun 29, 2018

Choose a reason for hiding this comment

JeffBezanson Jun 29, 2018

Choose a reason for hiding this comment

JeffBezanson commented Jun 29, 2018

JeffBezanson commented Jun 29, 2018

nanosoldier commented Jun 29, 2018

JeffBezanson commented Jun 30, 2018

nanosoldier commented Jun 30, 2018

JeffBezanson commented Jul 5, 2018

JeffBezanson commented Jul 5, 2018

nanosoldier commented Jul 6, 2018

JeffBezanson commented Jul 6, 2018

nanosoldier commented Jul 6, 2018

JeffBezanson commented Jul 6, 2018

nanosoldier commented Jul 6, 2018

vtjnash Jul 6, 2018

Choose a reason for hiding this comment

JeffBezanson Jul 6, 2018

Choose a reason for hiding this comment

vtjnash Jul 6, 2018

Choose a reason for hiding this comment

KristofferC commented Sep 13, 2018

JeffBezanson commented Sep 13, 2018

KristofferC commented Sep 13, 2018 • edited

vtjnash commented Sep 13, 2018

JeffBezanson commented Jun 29, 2018 •

edited

KristofferC commented Jun 29, 2018 •

edited

KristofferC commented Sep 13, 2018 •

edited