Implement computed goto dispatch for the CoreCLR interpreter by BrzVlad · Pull Request #129216 · dotnet/runtime

BrzVlad · 2026-06-10T05:43:50Z

Instead of making each opcode dispatch to the loop start where we switch on the opcode, we transform every switch case into a label, we let the compiler statically populate the s_dispatchTable which maps each opcode to the label address. This makes opcode dispatch a simple load + branch to the label address from this table. This makes the interpreter 25% faster.

It is unclear whether this has any impact on wasm. Likely not, because wasm has control flow limitations that make random branches impossible.

As an additional optimization, we avoid saving pFrame->ip for each opcode. Removing this can improve execution speed by around 3%. Instead we explicitly save pFrame->ip in opcodes that can trigger GC or throw exception. Add InterpThrow helper to ensure pFrame->ip is set before all throws.

Copilot

Copilot wasn't able to review any files in this pull request.

BrzVlad · 2026-06-10T05:44:34Z

/azp run runtime-interpreter

azure-pipelines · 2026-06-10T05:44:43Z

Azure Pipelines successfully started running 1 pipeline(s).

BrzVlad · 2026-06-10T05:45:25Z

/azp run runtime-libraries-interpreter

azure-pipelines · 2026-06-10T05:45:34Z

Azure Pipelines successfully started running 1 pipeline(s).

BrzVlad · 2026-06-11T06:57:16Z

Assembly before

// INTOP_NEXT
0x105260a0c: ldr w23, [x28, #0xc]! // x28 = ip, load the next opcode
0x105260a10: cmp w23, #0x15f // check for the default branch
0x105260a14: b.ls 0x10525e034
0x105260a18: b 0x10526295c
// Loop start
0x10525e034: mov w8, w23
0x10525e038: adrp x11, 414 // rather long switch dispatch
0x10525e03c: add x11, x11, #0x624
0x10525e040: adr x9, 0x10525e008
0x10525e044: ldrh w10, [x11, x8, lsl #1]
0x10525e048: add x9, x9, x10, lsl #2
0x10525e04c: br x9

Assembly after

// INTOP_NEXT
0x1051f63e0: ldr w23, [x27, #0x8]! // x27 = ip, load the next opcode
0x1051f63e4: ldr x8, [x22, x23, lsl #3] // x22 = s_dispatchTable, load label of next
0x1051f63e8: br x8 // dispatch

Instead of making each opcode dispatch to the loop start where we switch on the opcode, we transform every switch case into a label, we let the compiler statically populate the s_dispatchTable which maps each opcode to the label address. This makes opcode dispatch a simple load + branch to the label address from this table. This makes the interpreter 25% faster. It is unclear whether this has any impact on wasm. Likely not, because wasm has control flow limitations that make random branches impossible.

Removing this can improve execution speed by around 3%. Instead we explicitly save pFrame->ip in opcodes that can trigger GC or throw exception. Add InterpThrow helper to ensure pFrame->ip is set before all throws.

BrzVlad · 2026-06-17T15:44:27Z

/azp run runtime-interpreter
/azp run runtime-libraries-interpreter

azure-pipelines · 2026-06-17T15:44:34Z

No pipelines are associated with this pull request.

BrzVlad · 2026-06-17T15:46:26Z

/azp run runtime-interpreter

azure-pipelines · 2026-06-17T15:46:42Z

Azure Pipelines successfully started running 1 pipeline(s).

BrzVlad · 2026-06-17T15:46:48Z

/azp run runtime-libraries-interpreter

azure-pipelines · 2026-06-17T15:46:58Z

Azure Pipelines successfully started running 1 pipeline(s).

Copilot

Pull request overview

Copilot reviewed 1 out of 2 changed files in this pull request and generated 3 comments.

BrzVlad · 2026-06-18T20:09:31Z

/ba-g unrelated timeouts in non-interpreter jobs

We disable this on windows where vc compiler doesn't supported it.

If we ever get to interpreting an invalid opcode, the most likely value would be 0. Having this opcode do random operations (like returning from the frame as it was the case before with INTOP_RET having the 0 value) could make it harder to diagnose this types of bugs if they will ever appear in the future. On mono we've had multiple cases where we were trying to execute code from invalid memory, hitting this case.

BrzVlad · 2026-06-23T05:01:10Z

Could I get a final review on this ? @janvorli your previous review got stale after a rebase.

BrzVlad · 2026-06-23T06:53:16Z

/ba-g non-interpreter failures

Instead of making each opcode dispatch to the loop start where we switch on the opcode, we transform every switch case into a label, we let the compiler statically populate the s_dispatchTable which maps each opcode to the label address. This makes opcode dispatch a simple load + branch to the label address from this table. This makes the interpreter 25% faster. It is unclear whether this has any impact on wasm. Likely not, because wasm has control flow limitations that make random branches impossible. As an additional optimization, we avoid saving `pFrame->ip` for each opcode. Removing this can improve execution speed by around 3%. Instead we explicitly save `pFrame->ip` in opcodes that can trigger GC or throw exception. Add InterpThrow helper to ensure `pFrame->ip` is set before all throws.

CopilotAI review requested due to automatic review settings June 10, 2026 05:43

BrzVlad requested review from janvorli and kg as code owners June 10, 2026 05:43

CopilotAI reviewed Jun 10, 2026
View reviewed changes

github-actionsBot added the area-CodeGen-Interpreter-coreclr label Jun 10, 2026

jkotas added the tenet-performance Performance related issue label Jun 10, 2026

janvorli approved these changes Jun 17, 2026
View reviewed changes

BrzVlad added 2 commits June 17, 2026 18:42

Stop saving pFrame->ip for each opcodes
e39e619
Removing this can improve execution speed by around 3%. Instead we explicitly save pFrame->ip in opcodes that can trigger GC or throw exception. Add InterpThrow helper to ensure pFrame->ip is set before all throws.

BrzVlad force-pushed the feature-clrinterp-computed-goto branch from 1f2bd0a to e39e619Compare June 17, 2026 15:43

Fix build
0c8750f

CopilotAI review requested due to automatic review settings June 18, 2026 12:20

Copilotstarted reviewing on behalf of BrzVlad June 18, 2026 12:21 View session

CopilotAI reviewed Jun 18, 2026
View reviewed changes

Comment threadsrc/coreclr/vm/interpexec.cpp
Comment threadsrc/coreclr/vm/interpexec.cpp
Comment threadsrc/coreclr/interpreter/inc/intops.def

BrzVlad requested a review from janvorli June 18, 2026 20:10

jkotas reviewed Jun 18, 2026
View reviewed changes

Comment threadsrc/coreclr/interpreter/inc/intops.def

am11 reviewed Jun 18, 2026
View reviewed changes

Comment threadsrc/coreclr/vm/interpexec.cpp Outdated

BrzVlad added 3 commits June 22, 2026 08:57

Remove definition of opcodes on platforms where they are never used
b6cb768

Tweak computed goto condition
8f7d3af
We disable this on windows where vc compiler doesn't supported it.

build-analysisBot mentioned this pull request Jun 22, 2026
AF: *(_UNCHECKED_OBJECTREF *)handle == NULL (HndCreateHandle called by getJitHandleForObject) #117138
Open

janvorli approved these changes Jun 23, 2026
View reviewed changes

BrzVlad merged commit 461fd2a into dotnet:mainJun 23, 2026
108 of 112 checks passed

dotnet-maestroBot mentioned this pull request Jun 24, 2026
[main] Source code updates from dotnet/runtime dotnet/dotnet#7344
Merged

dotnet-milestone-botBot added this to the 11.0-preview6 milestone Jun 24, 2026

am11 mentioned this pull request Jun 24, 2026
Interpreter Performance #122464
Open
15 tasks

github-actionsBot locked and limited conversation to collaborators Jul 24, 2026

Uh oh!

Conversation

BrzVlad commented Jun 10, 2026• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

CopilotAI left a comment

Choose a reason for hiding this comment

Uh oh!

BrzVlad commented Jun 10, 2026

Uh oh!

azure-pipelinesBot commented Jun 10, 2026

Uh oh!

BrzVlad commented Jun 10, 2026

Uh oh!

azure-pipelinesBot commented Jun 10, 2026

Uh oh!

BrzVlad commented Jun 11, 2026

Uh oh!

BrzVlad commented Jun 17, 2026

Uh oh!

azure-pipelinesBot commented Jun 17, 2026

Uh oh!

BrzVlad commented Jun 17, 2026

Uh oh!

azure-pipelinesBot commented Jun 17, 2026

Uh oh!

BrzVlad commented Jun 17, 2026

Uh oh!

azure-pipelinesBot commented Jun 17, 2026

Uh oh!

CopilotAI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Uh oh!

BrzVlad commented Jun 18, 2026

Uh oh!

Uh oh!

Uh oh!

BrzVlad commented Jun 23, 2026

Uh oh!

BrzVlad commented Jun 23, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

BrzVlad commented Jun 10, 2026•
edited
Loading