JIT: make funclet order correspond to EH clause order by AndyAyersMS · Pull Request #127590 · dotnet/runtime

AndyAyersMS · 2026-04-30T00:33:54Z

For Wasm the JIT host will need to match up funclets with EH clauses without relying on the offset data in the EH clause.

To allow this, set things up so that the host can infer which funclets are associated with an EH clause by making the ordering of EH clauses and the order of funclets (reported via unwind info) correspond.

Note this is not 1-1 as EH clauses with filters will inspire two funclets. In those cases the JIT will always put the filter funclet before the catch funclet.

For Wasm the JIT host will need to match up funclets with EH clauses without relying on the offset data in the EH clause. To allow this, set things up so that the host can infer which funclets are associated with an EH clause by making the ordering of EH clauses and the order of funclets (reported via unwind info) correspond. Note this is not 1-1 as EH clauses with filters will inspire two funclets. In those cases the JIT will always put the filter funclet before the catch funclet.

AndyAyersMS · 2026-04-30T00:35:01Z

@dotnet/jit-contrib PTAL

A handful of (zero size) diffs for methods where there are mutual protect trys where some catch clauses have EH in them -- those are the cases where the JIT's EH table and the EH info reported to the VM end up differently ordered.

dotnet-policy-service · 2026-04-30T00:35:21Z

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.

Copilot

Pull request overview

This PR aims to make EH funclet ordering match the order that EH clauses are reported to the VM (specifically, the “VM EH clause order” derived from try-region ordering), so Wasm can associate funclets with EH clauses without depending on EH offset data. The implementation introduces a shared “EH clause order” mapping computed during funclet creation and reuses it during EH reporting and Wasm virtual-IP clause construction.

Changes:

Compute and store a VM-EH-clause-order mapping in fgCreateFunclets and publish it via a new Compiler::compEHorderTab.
Update EH clause reporting (genReportEH / genReportEHClauses) to use the precomputed clause order instead of sorting at report time.
Update Wasm EH virtual-IP clause prefill/update logic to (attempt to) respect VM clause order.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 8 comments.

Show a summary per file

File	Description
src/coreclr/jit/flowgraph.cpp	Builds and stores the EH-clause ordering map while creating funclets.
src/coreclr/jit/fgwasm.cpp	Adjusts Wasm EH virtual-IP clause construction to use the new ordering map.
src/coreclr/jit/fgbasic.cpp	Removes comment notes from `fgRelocateEHRange` header.
src/coreclr/jit/compiler.h	Adds `compEHorderTab` to persist the ordering map.
src/coreclr/jit/codegencommon.cpp	Switches EH reporting to use `compEHorderTab` rather than sorting clauses at report time.

AndyAyersMS · 2026-04-30T01:40:35Z

Looks like we should keep both jit->vm and vm->jit mappings. Will revise this.

Copilot

Pull request overview

Copilot reviewed 5 out of 5 changed files in this pull request and generated 3 comments.

AndyAyersMS · 2026-04-30T20:58:57Z

@adamperlin want to review this?

adamperlin · 2026-04-30T21:47:48Z

+        // This is not the same as the order of entries in the EH table, so build a mapping from
+        // So, build mappings from vm order to table order and vice versa.


Suggested change

// This is not the same as the order of entries in the EH table, so build a mapping from

// So, build mappings from vm order to table order and vice versa.

// This is not the same as the order of entries in the EH table, so build a mapping from vm order to table order and vice versa.

adamperlin · 2026-04-30T22:43:33Z

 // Return Value:
 //    The last block that was relocated, or nullptr on failure.
 //
-// Notes:


Intentional?

Yes, this method no longer does the bad things the comment said it can do.

adamperlin · 2026-04-30T22:45:38Z

+            vmClauseOrderToEHTabOrder[XTnum] = (unsigned short)XTnum;
+        }
+
+        // The JIT's ordering of EH clauses does not guarantee that clauses covering the same try region are contiguous.


Could we add a comment here about why this is and why the JIT's ordering can't be changed here -- maybe a reference to jiteh.cpp where fgSortEHTable is?

I can add a comment, sure.

There are quite a few places in the JIT that depend on the order of entries in the EH table.

adamperlin · 2026-04-30T23:12:52Z

+            {
+                // We have two clauses mapped to the same try region.
+                // Make sure we order the clause with the smaller index first.
+                return leftIndex < rightIndex;


If two clauses share the same try index, we need to tie break by original eh table index, since i < j in the JIT eh table (with equal try index) implies i more nested than j? And this is then sufficient to be able to determine if we set CORINFO_EH_CLAUSE_SAMETRY , since we're differentiating between:

try { try { } catch (e0) {...} // SAMETRY = false } catch(e1) {...} // SAMETRY = false

try { } catch(e0) {...} // SAMETRY = true catch (e1) {...} // SAMETRY = true

?

The JIT table order reflects the order they appeared in the IL, so in your example above we'd make sure to emit the catch(e0) before the catch(e1).

We didn't always do this sorting; see #104531

Copilot

Pull request overview

Copilot reviewed 5 out of 5 changed files in this pull request and generated 1 comment.

Copilot · 2026-05-01T01:58:12Z

+        for (unsigned XTnum = 0; XTnum < compHndBBtabCount; XTnum++)
        {
-            const unsigned    index = ehGetIndex(dsc);
+            EHblkDsc* const dsc = ehGetDsc(XTnum);
+
            CORINFO_EH_CLAUSE clause;
            clause.ClassToken    = dsc->HasFilter() ? 0 : dsc->ebdTyp;
            clause.Flags         = ToCORINFO_EH_CLAUSE_FLAGS(dsc->ebdHandlerType);
            clause.TryOffset     = 0;
            clause.TryLength     = 0;
            clause.HandlerOffset = 0;
            clause.HandlerLength = 0;
-            clauses[index]       = {clause, dsc};
+
+            unsigned const vmIndex = compEHTabOrderToVMClauseOrder[XTnum];
+
+            clauses[vmIndex] = {clause, dsc};


fgWasmVirtualIP now stores EHClauseInfo entries in VM clause order (via compEHTabOrderToVMClauseOrder), but the DEBUG dump later in this method still indexes clauses using ehGetIndex(dsc) (EH table order). This will print mismatched Try/Handler ranges for the descriptor being dumped. The dump should translate EH table indices to VM indices before indexing into clauses (e.g., use compEHTabOrderToVMClauseOrder[index]).

Copilot AI review requested due to automatic review settings April 30, 2026 00:33

Copilot started reviewing on behalf of AndyAyersMS April 30, 2026 00:34 View session

github-actions Bot added the area-CodeGen-coreclr CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI label Apr 30, 2026

dotnet-policy-service Bot assigned AndyAyersMS Apr 30, 2026

Copilot AI reviewed Apr 30, 2026

View reviewed changes

AndyAyersMS added 2 commits April 29, 2026 19:01

revise; keep mapping in both directions

ff623f1

revise comments

2b42a46

Copilot AI review requested due to automatic review settings April 30, 2026 02:04

Copilot started reviewing on behalf of AndyAyersMS April 30, 2026 02:04 View session

Copilot AI reviewed Apr 30, 2026

View reviewed changes

Comment thread src/coreclr/jit/fgwasm.cpp Outdated

Comment thread src/coreclr/jit/codegencommon.cpp Outdated

Comment thread src/coreclr/jit/flowgraph.cpp Outdated

more revision

ce25891

This was referenced Apr 30, 2026

The Operation will be canceled. The next steps may not contain expected logs. dotnet/dnceng#3008

Open

XHarness package install failure on iOS due to devicectl NSPOSIXErrorDomain error 49 #123796

Open

AndyAyersMS requested a review from adamperlin April 30, 2026 20:59

adamperlin reviewed Apr 30, 2026

View reviewed changes

fix comment

0cb65f0

Copilot AI review requested due to automatic review settings May 1, 2026 01:47

Copilot started reviewing on behalf of AndyAyersMS May 1, 2026 01:47 View session

Copilot AI reviewed May 1, 2026

View reviewed changes

build-analysis Bot mentioned this pull request May 1, 2026

Unable to pull image from mcr.microsoft.com #117164

Open

fix dump after computing wasm virtual IPs

44cc8a3

		// This is not the same as the order of entries in the EH table, so build a mapping from
		// So, build mappings from vm order to table order and vice versa.

Conversation

AndyAyersMS commented Apr 30, 2026

Uh oh!

AndyAyersMS commented Apr 30, 2026

Uh oh!

dotnet-policy-service Bot commented Apr 30, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

AndyAyersMS commented Apr 30, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Uh oh!

AndyAyersMS commented Apr 30, 2026

Uh oh!

adamperlin Apr 30, 2026

Choose a reason for hiding this comment

Uh oh!

adamperlin Apr 30, 2026

Choose a reason for hiding this comment

Uh oh!

AndyAyersMS May 1, 2026

Choose a reason for hiding this comment

Uh oh!

adamperlin Apr 30, 2026

Choose a reason for hiding this comment

Uh oh!

AndyAyersMS May 1, 2026

Choose a reason for hiding this comment

Uh oh!

adamperlin Apr 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

AndyAyersMS May 1, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Copilot AI May 1, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

adamperlin Apr 30, 2026 •

edited

Loading