Import lzcnt/tzcnt/popcnt as returning int so rngchk works by tannergooding · Pull Request #128658 · dotnet/runtime

tannergooding · 2026-05-27T18:16:09Z

No description provided.

Copilot

Pull request overview

This PR changes JIT import/codegen for count-related intrinsics so their IR result can be TYP_INT while still preserving operand width where needed for correct range-check handling.

Changes:

Imports LZCNT/TZCNT/POPCNT results as TYP_INT and casts back to the signature return type when needed.
Adds xarch metadata/codegen plumbing to preserve operand width for scalar count instructions.
Adds ARM64 primitive codegen/LSRA handling for PopCount and TrailingZeroCount.

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 5 comments.

Show a summary per file

File	Description
`src/coreclr/jit/importercalls.cpp`	Updates primitive intrinsic import for LZCNT, TZCNT, POPCNT, and Log2.
`src/coreclr/jit/hwintrinsiclistxarch.h`	Marks selected scalar xarch intrinsics as deriving base type from the first argument.
`src/coreclr/jit/hwintrinsiccodegenxarch.cpp`	Adjusts xarch count intrinsic codegen to accept an explicit emit size.
`src/coreclr/jit/hwintrinsic.cpp`	Adds cast wrapping for selected direct hardware intrinsics returning long-sized values.
`src/coreclr/jit/codegenarmarch.cpp`	Adds ARM64 codegen for primitive PopCount and TrailingZeroCount.
`src/coreclr/jit/lsraarm64.cpp`	Adds ARM64 LSRA internal SIMD temp allocation for primitive PopCount.
`src/coreclr/jit/codegen.h`	Updates the `genXCNTIntrinsic` declaration.

Copilot

Pull request overview

Copilot reviewed 7 out of 7 changed files in this pull request and generated 3 comments.

Copilot

Pull request overview

Copilot reviewed 7 out of 7 changed files in this pull request and generated no new comments.

EgorBo · 2026-05-27T20:03:29Z

what kind of rngchk?

tannergooding · 2026-05-27T20:09:10Z

what kind of rngchk?

@EgorBo, Range check only works with TYP_INT which means x64 and Arm64 alike are being pessimized for the lzcnt/tzcnt/popcnt nodes that produce TYP_LONG. This canonicalizes the form so that we get CAST(LZCNT(long)) instead (doing int -> long in the cast) which allows us to properly see [0, 64] instead of keUnknown or [INT32_MIN, INT32_MAX]

I'm still waiting to see SPMI diffs and expect we'll also need to get diffs after #128620 goes in (which ensures range check actually handles lzcnt/tzcnt/popcnt for x64/arm64).

But initial investigations showed that this is why Arm64 had -14k bytes of codegen and x64 only had -87 bytes in the above PR (because Arm64 imported as TYP_INT in more cases for TYP_LONG based nodes).

EgorBo · 2026-05-27T20:11:40Z

what kind of rngchk?
@EgorBo, Range check only works with TYP_INT which means x64 and Arm64 alike are being pessimized for the lzcnt/tzcnt/popcnt nodes that produce TYP_LONG. This canonicalizes the form so that we get CAST(LZCNT(long)) instead (doing int -> long in the cast) which allows us to properly see [0, 64] instead of keUnknown or [INT32_MIN, INT32_MAX]

Is there a minimal repro? pretty sure it can be easily fixed in the GetRangeFromAssertions or GetRange vs manual expansions, new QMARKs, etc.

E.g. GetRangeFromAssertions doesn't have to bail on TYP_LONG if known range fits into TYP_INT

tannergooding · 2026-05-27T20:26:24Z

vs manual expansions, new QMARKs, etc.

The only new qmark is from handling int.Log2 (i.e. signed log2) otherwise we never produce an intrinsic and never get IsNeverNegative, general constant folding, etc from it.

This also then ensures that popcnt is properly intrinsic on Arm64 (it wasn't before) and that tzcnt can be handled on Arm64 without pessimizing the IR (it was two nodes, one of which we didn't have good handling for).

Is there a minimal repro?

You can take your pick from the SPMI diffs in the other PR. It's clearly a broadly applicable scenario

E.g. GetRangeFromAssertions doesn't have to bail on TYP_LONG if known range fits into TYP_INT

Local testing had shown this to be much more expensive and error prone as basically all of range check presumes TYP_LONG is not going to exist today. So I put up a draft with the simpler change that handles some xplat differences and pessimizations, ensuring both are equally treated

Copilot

Pull request overview

Copilot reviewed 7 out of 7 changed files in this pull request and generated 2 comments.

Copilot

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 1 comment.

tannergooding · 2026-05-28T04:09:24Z

Pulled a couple of the changes out into their own PRs

Ensure Log2 can actually be imported as intrinsic #128678 - Ensure Log2 can actually be imported as intrinsic
Improve the handling of popcnt and tzcnt on arm64 #128677 - Improve the handling of popcnt and tzcnt on arm64

And then put up a PR that has RangeCheck::ComputeRange handle some TYP_LONG cases: #128676

Will see if I can also get RangeCheck::GetRangeFromAssertions to handle some TYP_LONG scenarios as suggested, but that one is a bit more complex I think, namely because it has the check for if a VN exists and so the type needs to be passed through as well.

Going to keep this PR in draft and aim to get the others merged first.

Copilot

Pull request overview

Copilot reviewed 9 out of 9 changed files in this pull request and generated 2 comments.

Copilot

Pull request overview

Copilot reviewed 9 out of 9 changed files in this pull request and generated 1 comment.

Copilot

Pull request overview

Copilot reviewed 9 out of 9 changed files in this pull request and generated no new comments.

Copilot

Pull request overview

Copilot reviewed 9 out of 9 changed files in this pull request and generated no new comments.

tannergooding · 2026-06-12T00:58:04Z

I'm expecting the other work done allows this to be closed now, but resolved conflicts and want to run SPMI again to confirm

Copilot

Pull request overview

Copilot reviewed 7 out of 7 changed files in this pull request and generated 5 comments.

+ // rangecheck requires the type to be TYP_INT to handle this, but the operation may expect TYP_LONG
+ // so we fixup the baseType here and in remove the cast in lowering if its unnecessary
+ baseType = TYP_INT;


+ // rangecheck requires the type to be TYP_INT to handle this, but the operation may expect TYP_LONG
+ // so we fixup the baseType here and in remove the cast in lowering if its unnecessary
+ baseType = TYP_INT;


+ // rangecheck requires the type to be TYP_INT to handle this, but the operation may expect TYP_LONG
+ // so we fixup the baseType here and in remove the cast in lowering if its unnecessary
+ baseType = TYP_INT;


+ if (HWIntrinsicInfo::BaseTypeFromFirstArg(intrinsicId))
+ {
+ assert(op1 != nullptr);
+ targetType = genActualType(op1->TypeGet());
+ }
+ else
+ {
+ assert(!HWIntrinsicInfo::BaseTypeFromFirstArg(intrinsicId));
+ }


+ // rangecheck requires the type to be TYP_INT to handle this, but the operation may expect
+ // TYP_LONG so we fixup the baseType here and in remove the cast in lowering if its
+ // unnecessary


CopilotAI review requested due to automatic review settings May 27, 2026 18:16

Copilotstarted reviewing on behalf of tannergooding May 27, 2026 18:16 View session

github-actionsBot added the area-CodeGen-coreclr CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI label May 27, 2026

CopilotAI reviewed May 27, 2026
View reviewed changes

Comment threadsrc/coreclr/jit/hwintrinsiccodegenxarch.cpp
Comment threadsrc/coreclr/jit/importercalls.cpp
Comment threadsrc/coreclr/jit/importercalls.cpp Outdated
Comment threadsrc/coreclr/jit/importercalls.cpp
Comment threadsrc/coreclr/jit/importercalls.cpp Outdated

CopilotAI review requested due to automatic review settings May 27, 2026 19:10

tannergooding force-pushed the better-lzcnt branch from 7182c17 to e9f3b57Compare May 27, 2026 19:10

Copilotstarted reviewing on behalf of tannergooding May 27, 2026 19:10 View session

CopilotAI reviewed May 27, 2026
View reviewed changes

Comment threadsrc/coreclr/jit/importercalls.cpp Outdated
Comment threadsrc/coreclr/jit/importercalls.cpp Outdated
Comment threadsrc/coreclr/jit/importercalls.cpp

tannergooding force-pushed the better-lzcnt branch from fc3e9f8 to beb6368Compare May 27, 2026 19:21

CopilotAI review requested due to automatic review settings May 27, 2026 19:31

tannergooding force-pushed the better-lzcnt branch from beb6368 to 2126914Compare May 27, 2026 19:31

Copilotstarted reviewing on behalf of tannergooding May 27, 2026 19:32 View session

CopilotAI reviewed May 27, 2026
View reviewed changes

CopilotAI review requested due to automatic review settings May 28, 2026 00:41

tannergooding force-pushed the better-lzcnt branch from 77def0d to bb87972Compare May 28, 2026 00:41

Copilotstarted reviewing on behalf of tannergooding May 28, 2026 00:41 View session

CopilotAI reviewed May 28, 2026
View reviewed changes

Comment threadsrc/coreclr/jit/codegenarmarch.cpp Outdated
Comment threadsrc/coreclr/jit/hwintrinsic.cpp Outdated

CopilotAI review requested due to automatic review settings May 28, 2026 03:35

Copilotstarted reviewing on behalf of tannergooding May 28, 2026 03:35 View session

CopilotAI reviewed May 28, 2026
View reviewed changes

Comment threadsrc/coreclr/jit/hwintrinsic.cpp Outdated

CopilotAI review requested due to automatic review settings May 28, 2026 04:13

tannergooding force-pushed the better-lzcnt branch from 37dda6c to 51a44f6Compare May 28, 2026 04:13

Copilotstarted reviewing on behalf of tannergooding May 28, 2026 04:14 View session

xtqqczze mentioned this pull request May 28, 2026
Unnecessary sign-extension for LeadingZeroCount(UInt64)#119699
Open

CopilotAI reviewed May 28, 2026
View reviewed changes

Comment threadsrc/coreclr/jit/codegenarmarch.cpp Outdated
Comment threadsrc/coreclr/jit/hwintrinsic.cpp Outdated

tannergooding force-pushed the better-lzcnt branch from 51a44f6 to 73da56bCompare May 28, 2026 04:21

CopilotAI review requested due to automatic review settings May 28, 2026 04:23

tannergooding force-pushed the better-lzcnt branch from 73da56b to 36f40aaCompare May 28, 2026 04:23

Copilotstarted reviewing on behalf of tannergooding May 28, 2026 04:24 View session

tannergooding force-pushed the better-lzcnt branch from 36f40aa to 0c61e06Compare May 28, 2026 04:28

CopilotAI reviewed May 28, 2026
View reviewed changes

Comment threadsrc/coreclr/jit/codegenarmarch.cpp Outdated

tannergooding force-pushed the better-lzcnt branch from 0c61e06 to 947df77Compare May 28, 2026 04:29

CopilotAI review requested due to automatic review settings May 28, 2026 04:46

Copilotstarted reviewing on behalf of tannergooding May 28, 2026 04:46 View session

CopilotAI reviewed May 28, 2026
View reviewed changes

CopilotAI review requested due to automatic review settings May 28, 2026 05:22

tannergooding force-pushed the better-lzcnt branch from 14f48bd to d4b757aCompare May 28, 2026 05:22

Copilotstarted reviewing on behalf of tannergooding May 28, 2026 05:22 View session

CopilotAI reviewed May 28, 2026
View reviewed changes

Import lzcnt/tzcnt/popcnt as returning int so rngchk works
9360551

tannergooding force-pushed the better-lzcnt branch from d4b757a to 9360551Compare May 28, 2026 06:20

Merge branch 'main' into better-lzcnt
2236c7a

CopilotAI review requested due to automatic review settings June 12, 2026 00:57

Copilotstarted reviewing on behalf of tannergooding June 12, 2026 00:57 View session

CopilotAI reviewed Jun 12, 2026
View reviewed changes

This was referenced Jun 12, 2026
Multiple Helix work items fail on maccatalyst/tvos CoreCLR Release #126460
Closed
[browser][coreCLR] Wasm.Console.Node.Sample - undefined symbol: SystemInteropJS_GetManagedStackTrace #129229
Closed

tannergooding closed this Jun 12, 2026

github-actionsBot locked and limited conversation to collaborators Jul 13, 2026

Uh oh!

Conversation

tannergooding commented May 27, 2026

Uh oh!

CopilotAI 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!

CopilotAI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Uh oh!

CopilotAI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

EgorBo commented May 27, 2026

Uh oh!

tannergooding commented May 27, 2026• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

EgorBo commented May 27, 2026• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

tannergooding commented May 27, 2026

Uh oh!

CopilotAI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

CopilotAI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

tannergooding commented May 28, 2026

Uh oh!

CopilotAI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

CopilotAI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

CopilotAI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

CopilotAI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

tannergooding commented Jun 12, 2026

Uh oh!

CopilotAI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

tannergooding commented May 27, 2026•
edited
Loading

EgorBo commented May 27, 2026•
edited
Loading