Compute effects for indirect calls in GlobalEffects by stevenfontanella · Pull Request #8609 · WebAssembly/binaryen

stevenfontanella · 2026-04-15T22:59:21Z

When running in --closed-world, compute effects for indirect calls by unioning the effects of all potential functions of that type. In --closed-world, we assume that all references originate in our module, so the only possible functions that we don't know about are imports. Previously we gave up on effects analysis for indirect calls.

Yields a very small byte count reduction in calcworker (3799354 - 3799297 = 57 bytes). Also shows no significant difference in Binaryen runtime: (0.1346069 -> 0.13375045 = <1% improvement, probably within noise). We expect more benefits after we're able to share indirect call effects with other passes, since currently they're only seen one layer up for callers of functions that indirectly call functions (see the newly-added tests for examples).

Followups:

Share effect information per type with other passes besides just via Function::effects (Use effects for indirect call expressions #8625)
Exclude functions that don't have an address (i.e. functions that aren't the target of ref.func) from effect analysis ()
Compute effects more precisely for exact + nullable/non-nullable references

Part of #8615.

kripken · 2026-04-17T19:53:49Z

+    ;; There's no function with this type, so it's impossible to create a ref to
+    ;; call this function with and there are no effects to aggregate.
+    ;; Remove this call.
+    (call $calls-uninhabited (local.get $ref))


This is replacing a trap with a nop, which is generally something we don't want to do, see #8387 - can we avoid this?

This can't trap since the refs are non-nullable. Added a test case in the first module showing that a nullable ref won't be optimized out even if we determine that its potential function targets don't have effects. This is handled in the initial effects calculation, we only clear the calls and throws effects and replace them with the effects of function bodies, but the trap effect stays.

Sounds good. Please also add a testcase with a nullable uninstantiated type - that one must trap, so (1) we should not remove it, and (2) we can add a TODO to turn it into an unreachable if we don't already (though perhaps other passes do that, I'm not sure offhand).

(Specifically, I mean to add a function right after this one, with (ref null $uninhabited))

It would still be good to add the TODO suggested above. Also, even in the case with non-nullable references, it's generally better to replace code that cannot possibly be executed with unreachable rather than nop because that lets us remove more code as a follow-on. So it would be good to have a TODO about that as well.

I put the TODO above on $calls-nullable-uninhabited, but on second thought that might be the wrong place for it. With unreachable, is there a distinction between "impossible" and "always traps"? Does the value of --traps-never-happen affect that? $calls-uninhabited is the latter case while $calls-nullable-uninhabited is the former case.

Without --traps-never-happen, there's no good way to encode "impossible" in the type system¹, but we can still remove all the code that comes after an unreachable. With --traps-never-happen, we can additionally remove all the code that comes before the unreachable.

Footnotes

We could in principle use an uninhabitable type like (ref none) to represent impossibility, but we don't use it in that way today. Most toolchains do use --traps-never-happen, so it doesn't matter much in practice. ↩

So both $calls-nullable-uninhabited and $calls-uninhabited should have TODOs for optimizing to unreachable, but for different reasons. The former because we know it must trap, and the latter because we know it is impossible to reach.

stevenfontanella · 2026-04-19T08:36:56Z

Also added some tests exercising call_indirect.

We don't need to walk up the supertype chain for type targets of indirect calls, we already do that for function implementations which would reach the type of the indirect call anyway

stevenfontanella · 2026-04-20T21:04:55Z

Added a test showing two different effect types being aggregated into the same function (global-effects-closed-world-simplify-locals.wast).

stevenfontanella · 2026-04-21T00:07:13Z

Added a test showing two different effect types being aggregated into the same function (global-effects-closed-world-simplify-locals.wast).

The test went wrong at some point, let me update.

stevenfontanella · 2026-04-21T00:52:56Z

Added a test showing two different effect types being aggregated into the same function (global-effects-closed-world-simplify-locals.wast).

The test went wrong at some point, let me update.

It turns out that the effects were pessimized when changing the features from --enable-gc --enable-reference-types to -all. Changed the features of the test back and looking into why this is now.

stevenfontanella · 2026-04-21T21:07:43Z

Added a test showing two different effect types being aggregated into the same function (global-effects-closed-world-simplify-locals.wast).

The test went wrong at some point, let me update.

It turns out that the effects were pessimized when changing the features from --enable-gc --enable-reference-types to -all. Changed the features of the test back and looking into why this is now.

Looks like this was due to LinearExecutionWalker (used in SimplifyLocals) not taking global effects into account and unconditionally assuming that functions may throw even if we determined that they won't. I tested with #8637 and found that the test works correctly with -all / --enable-exception-handling. Will land that PR after this one.

stevenfontanella · 2026-04-23T17:56:11Z

(Messed up merging in JJ, let me try again)

This now works after #8637

stevenfontanella · 2026-04-23T18:27:11Z

(Messed up merging in JJ, let me try again)

Done. Merged in main with #8637 and changed the test to run with -all to ensure it's working as intended with all features in case something else changes.

tlively

LGTM % final comments.

tlively · 2026-04-23T23:37:44Z

+              } else if (auto* callIndirect = curr->dynCast<CallIndirect>()) {
+                type = callIndirect->heapType;
+              } else {
+                WASM_UNREACHABLE("Unexpected call type");


I think we model a few of the stack switching instructions as having the calls effect, so we might want to handle this more gracefully. Does this pass the fuzzer as-is?

tlively · 2026-04-24T00:10:54Z

+    ;; There's no function with this type, so it's impossible to create a ref to
+    ;; call this function with and there are no effects to aggregate.
+    ;; Remove this call.
+    (call $calls-uninhabited (local.get $ref))


It would still be good to add the TODO suggested above. Also, even in the case with non-nullable references, it's generally better to replace code that cannot possibly be executed with unreachable rather than nop because that lets us remove more code as a follow-on. So it would be good to have a TODO about that as well.

tlively · 2026-04-24T00:13:21Z

+;; Same as above but this time our reference is the exact supertype
+;; so we know not to aggregate effects from the subtype.
+;; TODO: this case doesn't optimize today. Add exact ref support in the pass.
+(module


This could be tested in the previous module to avoid some duplication of the setup.

stevenfontanella changed the base branch from main to effects-scc April 16, 2026 16:28

stevenfontanella force-pushed the indirect-effects-scc branch 8 times, most recently from b7ce0b6 to 9fcf76b Compare April 16, 2026 20:59

stevenfontanella changed the title ~~Indirect effects scc~~ Compute effects for indirect calls in GlobalEffects Apr 16, 2026

stevenfontanella force-pushed the indirect-effects-scc branch from 9fcf76b to 0e28a7a Compare April 16, 2026 21:42

stevenfontanella marked this pull request as ready for review April 16, 2026 21:43

stevenfontanella requested a review from a team as a code owner April 16, 2026 21:43

stevenfontanella requested review from kripken and removed request for a team April 16, 2026 21:43

stevenfontanella force-pushed the indirect-effects-scc branch from 0e28a7a to 83f4b3b Compare April 16, 2026 22:32

Base automatically changed from effects-scc to main April 16, 2026 22:36

Cherrypick SCC with indirect effects

3a25c2e

stevenfontanella force-pushed the indirect-effects-scc branch from 83f4b3b to 3a25c2e Compare April 16, 2026 22:40

kripken reviewed Apr 16, 2026

View reviewed changes

Comment thread src/passes/GlobalEffects.cpp Outdated

Comment thread src/passes/GlobalEffects.cpp

Comment thread src/passes/GlobalEffects.cpp Outdated

Comment thread src/passes/GlobalEffects.cpp Outdated

Comment thread src/passes/GlobalEffects.cpp

Comment thread src/passes/GlobalEffects.cpp

PR updates

d764f59

tlively reviewed Apr 17, 2026

View reviewed changes

Comment thread src/passes/GlobalEffects.cpp Outdated

Comment thread src/passes/GlobalEffects.cpp Outdated

Comment thread src/passes/GlobalEffects.cpp Outdated

Comment thread src/passes/GlobalEffects.cpp Outdated

stevenfontanella added 3 commits April 17, 2026 19:16

PR updates

ff72d4f

Remove dead code

20fd3cd

Add explicit deduction guide to resolve CI error

6bf6f2a

kripken reviewed Apr 17, 2026

View reviewed changes

stevenfontanella added 3 commits April 17, 2026 21:17

PR updates

8a0e725

Add test for unreachable ref

ac59711

Add comment

4a440f9

tlively reviewed Apr 17, 2026

View reviewed changes

Comment thread src/passes/GlobalEffects.cpp Outdated

Comment thread src/passes/GlobalEffects.cpp Outdated

Comment thread src/passes/GlobalEffects.cpp

kripken reviewed Apr 17, 2026

View reviewed changes

Comment thread test/lit/passes/global-effects-closed-world.wast

kripken reviewed Apr 17, 2026

View reviewed changes

Comment thread out.txt Outdated

PR updates

3192366

tlively reviewed Apr 18, 2026

View reviewed changes

Comment thread test/lit/passes/global-effects-closed-world.wast Outdated

Comment thread test/lit/passes/global-effects-closed-world.wast

Comment thread src/passes/GlobalEffects.cpp Outdated

PR updates

512dad6

stevenfontanella added 5 commits April 19, 2026 16:31

Use type alias

0890c8f

Small fix to graph building

4994a9d

We don't need to walk up the supertype chain for type targets of indirect calls, we already do that for function implementations which would reach the type of the indirect call anyway

Optimize logic for walking up supertype heirarchy

7e09feb

Add more tests demonstrating different effects beings aggregated

f5d9926

Try reformatting

3567c0e

stevenfontanella added 2 commits April 20, 2026 21:24

Reformat again

ac58ee6

Lift out lambda

68971e4

tlively reviewed Apr 21, 2026

View reviewed changes

Comment thread src/passes/GlobalEffects.cpp

Comment thread src/support/graph_traversal.h Outdated

stevenfontanella force-pushed the indirect-effects-scc branch from 5086816 to 79f3ead Compare April 21, 2026 18:53

stevenfontanella added 3 commits April 21, 2026 20:15

Fix test

ffb73c5

Add unit test for DFS

48a9cab

Fix test comment

2630eff

stevenfontanella force-pushed the indirect-effects-scc branch from 189d543 to 2630eff Compare April 21, 2026 20:15

stevenfontanella added 2 commits April 23, 2026 17:58

Merge main

39d2d31

Use -all in test

259b234

This now works after #8637

stevenfontanella force-pushed the indirect-effects-scc branch from 467a5aa to 259b234 Compare April 23, 2026 18:04

stevenfontanella requested a review from tlively April 23, 2026 20:07

tlively approved these changes Apr 24, 2026

View reviewed changes

Conversation

stevenfontanella commented Apr 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Footnotes

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

stevenfontanella commented Apr 19, 2026

Uh oh!

stevenfontanella commented Apr 20, 2026

Uh oh!

stevenfontanella commented Apr 21, 2026

Uh oh!

stevenfontanella commented Apr 21, 2026

Uh oh!

Uh oh!

Uh oh!

stevenfontanella commented Apr 21, 2026

Uh oh!

stevenfontanella commented Apr 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

stevenfontanella commented Apr 23, 2026

Uh oh!

tlively left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

stevenfontanella commented Apr 15, 2026 •

edited

Loading

stevenfontanella commented Apr 23, 2026 •

edited

Loading