[Feat] Add partial support for loops (ForStatement) inside action bodies by Vineet1101 · Pull Request #5552 · p4lang/p4c

Vineet1101 · 2026-03-20T00:11:43Z

This PR provides partial implementation for p4lang/project-ideas#33 by adding support for ForStatement execution inside BMv2 action bodies.

Proposed Changes

backends/bmv2/common/check_unsupported.h:
- Removed the ForStatement rejection block.
- Updated the ForInStatement error message to be more specific (since for-in loops still rely on midend lowering that is not yet fully supported).
backends/bmv2/common/action.cpp:
- Added ForStatement preorder handling to convertActionBody().
- Loops compile cleanly to JSON using a condition label check and a backward unconditional _jump.
testdata/p4_16_samples/forloop-bmv2.p4:
- Added a rigorous test asserting the behavior of standard for-loops, nested for-loops, and conditionals wrapped inside loops.

Verification & Proof of Compilation

The changes successfully lower P4 loop statements into valid BMv2 JSON structures. Below is the generated primitive stack from our test program forloop-bmv2.p4:

1. Simple Loop (action sum_loop)

for (bit<8> i = 1; i <= h.h.count; i = i + 1) {
    h.h.sum = h.h.sum + (bit<32>)i;
}

Compiled BMv2 JSON Primitives:

  [0] assign(h.sum, 0)
  [1] assign(scalars.i_0, 1)          <--- Init
  [2] _jump_if_zero(expr, jump_to=6)  <--- Condition (if false, jump to end)
  [3] assign(h.sum, expr)             <--- Body
  [4] assign(scalars.i_0, expr)       <--- Update
  [5] _jump(jump_to=2)                <--- Jump back to Condition
  [6] assign(standard_metadata.egress_spec, 0)

2. Nested Loops (action nested_loop)

for (bit<8> i = 0; i < h.h.count; i = i + 1) {
    for (bit<8> j = 0; j < h.h.count; j = j + 1) {
        h.h.product = h.h.product + 1;
    }
}

Compiled BMv2 JSON Primitives:

  [0] assign(h.product, 0)
  [1] assign(scalars.i_1, 0)          <--- Outer Init
  [2] _jump_if_zero(expr, jump_to=10) <--- Outer Condition
  [3] assign(scalars.j_0, 0)          <--- Inner Init
  [4] _jump_if_zero(expr, jump_to=8)  <--- Inner Condition
  [5] assign(h.product, expr)         <--- Inner Body
  [6] assign(scalars.j_0, expr)       <--- Inner Update
  [7] _jump(jump_to=4)                <--- Inner Jump back
  [8] assign(scalars.i_1, expr)       <--- Outer Update
  [9] _jump(jump_to=2)                <--- Outer Jump back

All existing BMv2 tests and validations pass with P4TEST_REPLACE=1.

Vineet1101 · 2026-03-20T00:12:23Z

@fruffy @jafingerhut can you please review it

fruffy · 2026-03-20T01:08:03Z

Could you supply some stf test cases with the program to demonstrate various kind of packet input/output scenarios? Your example calculates sums based on the table entries and input header values, we can use that

jafingerhut · 2026-03-20T01:17:19Z

Do these changes support break or continue statements?

Vineet1101 · 2026-03-20T04:14:23Z

Do these changes support break or continue statements?

I was trying to implement forstatement first if this gets merged I will implement other two statements. If you want I can implement them as well in this PR

Vineet1101 · 2026-03-20T04:15:07Z

Could you supply some stf test cases with the program to demonstrate various kind of packet input/output scenarios? Your example calculates sums based on the table entries and input header values, we can use that

ok let me try

Vineet1101 · 2026-03-20T06:15:42Z

Do these changes support break or continue statements?

now it supports break and continue as well

Vineet1101 · 2026-03-20T06:17:07Z

Could you supply some stf test cases with the program to demonstrate various kind of packet input/output scenarios? Your example calculates sums based on the table entries and input header values, we can use that

done can you review it

jafingerhut · 2026-03-20T19:31:06Z

Looks like several test failures in the latest version of the changes.

jafingerhut · 2026-03-20T19:42:17Z

Looks like several test failures in the latest version of the changes.

OK, the C++ lint / formatting checks should be something you can fix with clang-format, but the testgen failures might be due to p4testgen not supporting for loops?

@fruffy - I am not here asking you to support for loops in p4testgen any time soon, as that sounds like too big of a request. Would you recommend marking test programs using for loops with XFAIL for now?

jafingerhut · 2026-03-20T19:55:38Z

Looks like several test failures in the latest version of the changes.

OK, the C++ lint / formatting checks should be something you can fix with clang-format, but the testgen failures might be due to p4testgen not supporting for loops?

@fruffy - I am not here asking you to support for loops in p4testgen any time soon, as that sounds like too big of a request. Would you recommend marking test programs using for loops with XFAIL for now?

FYI, there are already P4 test programs that have for loops in them that are exercised by p4testgen, but I believe all of the existing ones are ones that a p4c pass written by Chris Dodd completely unroll at compile time, so there are no for lops in the IR by the time the back end sees it. Perhapt p4testgen can unroll or otherwise handle the same loops that p4c pass unrolls?

I suspect this PR has P4 test programs that are not unrolled by that pass, because the number of iterations is not possible (or not easy enough) to determine at compile time.

fruffy · 2026-03-21T16:57:21Z

Perhapt p4testgen can unroll or otherwise handle the same loops that p4c pass unrolls?

Yes, the easiest approach is to just add the pass that unrolls loops and things should work as expected. At some point, if we want to support differential testing, we can support stepping through loops in the interpreter.

@Vineet1101 You can either try to add the loops unrolling pass to P4Testgen's BMv2 mid end here: https://github.com/p4lang/p4c/blob/main/backends/p4tools/modules/testgen/targets/bmv2/target.cpp#L112

Or add the failing tests to XFails.

jafingerhut · 2026-03-24T19:48:26Z

@Vineet1101 Do you know how to add the failing tests to XFAILs, so that they are expected to fail, and when they do, CI tests overall succeed? Let me know if you need help with that.

Vineet1101 · 2026-03-24T21:09:47Z

@fruffy @jafingerhut I have addressed your reviews in recent commits. For now, I have added tests to XFAILs

Vineet1101 · 2026-03-24T21:10:27Z

@Vineet1101 Do you know how to add the failing tests to XFAILs, so that they are expected to fail, and when they do, CI tests overall succeed? Let me know if you need help with that.

Not needed just went through the docs. Thanks for asking @jafingerhut :)

fruffy

Great! Which model are you using if I may ask :)

fruffy · 2026-03-24T23:56:36Z

backends/bmv2/common/action.cpp

+            int labelIdEndOfLoop = jumpInfo->numLabels;
+            jumpInfo->numLabels += 1;
+
+            // Step 1: Emit init statements


Nit: Gemini likes to do this for some reason. These comments go out of sync quite quickly. Would remove the Step n.

so should I remove all these comments??

The general convention of "nit" is that it's a "nice-to-have". Not required for merging and you can use your own judgement whether you want to address it or not.

https://onetwobytes.com/2021/11/09/nit-code-review/

Thanks for sharing this, I didn't know it before. I have removed the unnecessary steps comment

Vineet1101 · 2026-03-25T03:40:23Z

Great! Which model are you using if I may ask :)

I use Antigravity as an IDE and Claude Opus 4.6 for difficult tasks. To be clear, can one use AI as long as they know what it is doing?.

fruffy · 2026-03-26T01:51:07Z

To be clear, can one use AI as long as they know what it is doing?.

Yes! But always question what the LLM does. https://neilkakkar.com/agentic-debt.html

Add ForStatement compilation to convertActionBody() in the BMv2 backend. The loop structure is compiled as: [init statements] labelCondition: _jump_if_zero(cond, labelEndOfLoop) [body] [update statements] _jump(labelCondition) labelEndOfLoop: This leverages the existing JumpLabelInfo infrastructure and the _jump/_jump_if_zero BMv2 primitives that already support backward jumps. Also remove the ForStatement block from CheckUnsupported and update the ForInStatement error message to be more specific. Signed-off-by: Vineet1101 <vineetgoel692@gmail.com>

Adds a test program that exercises a simple for-loop, nested for-loops, and conditionals within a for-loop inside a BMv2 action. Also includes the expected frontend, midend, and JSON compiler outputs. Signed-off-by: Vineet1101 <vineetgoel692@gmail.com>

Add four STF scenarios exercising for-loop actions in forloop-bmv2.p4: 1. Default action (sum_loop): count=5, expects sum=15 2. Nested loop (nested_loop): count=4, expects product=16 3. Conditional loop (conditional_loop): count=10, expects sum=5 4. Zero iterations (sum_loop): count=0, expects sum=0 Tests cover simple loops, nested loops, loops with conditionals, and the edge case of zero iterations. Signed-off-by: Vineet1101 <vineetgoel692@gmail.com>

Extend the ForStatement handler in action body compilation to support break and continue statements: - Add labelIdBreak and labelIdContinue fields to JumpLabelInfo to carry the innermost loop's label context through recursive calls. - Handle BreakStatement by emitting _jump to labelIdBreak (end of loop). - Handle ContinueStatement by emitting _jump to labelIdContinue (update section of the loop). - Add a separate labelIdUpdate to the ForStatement handler so continue jumps to the update statements rather than the condition check. - Save/restore loop context around body compilation to support nested loops correctly. Signed-off-by: Vineet1101 <vineetgoel692@gmail.com>

Add break_loop and continue_loop actions to forloop-bmv2.p4: - break_loop: loop from 0..count, break at i==5, verifies early exit - continue_loop: loop 0..9, skip even values via continue, sum odd Add corresponding STF scenarios 5 and 6, and update expected outputs. Signed-off-by: Vineet1101 <vineetgoel692@gmail.com>

Signed-off-by: Vineet1101 <vineetgoel692@gmail.com>

The forloop-bmv2.p4 test program uses runtime-dependent loop bounds (h.h.count) that cannot be unrolled at compile time by the UnrollLoops pass. Mark it as XFAIL across all 5 BMv2 testgen test suites until p4testgen adds ForStatement support. Signed-off-by: Vineet1101 <vineetgoel692@gmail.com>

Vineet1101 · 2026-03-28T14:59:05Z

To be clear, can one use AI as long as they know what it is doing?.

Yes! But always question what the LLM does. https://neilkakkar.com/agentic-debt.html

Will remember this from now on, and also thanks for sharing this insightful resource.

Signed-off-by: Vineet1101 <vineetgoel692@gmail.com>

fruffy · 2026-03-29T17:04:24Z

@jafingerhut Any more comments?

jafingerhut

LGTM. I created my own new test action that had nested loops, and break and continue statements in both outer and inner loops, and examined the BMv2 JSON data file produced for that action, and it looks correct. All STF test cases in this PR also look correct. Nice!

fruffy added the core Topics concerning the core segments of the compiler (frontend, midend, parser) label Mar 21, 2026

Vineet1101 force-pushed the feature/bmv2-forstatement-support branch from 01c5dee to 86a4218 Compare March 24, 2026 21:06

fruffy approved these changes Mar 24, 2026

View reviewed changes

Vineet1101 force-pushed the feature/bmv2-forstatement-support branch from 86a4218 to 0991080 Compare March 28, 2026 14:56

Vineet1101 added 7 commits March 28, 2026 07:58

bmv2: fix clang-format violations in action.cpp

18a67a9

Signed-off-by: Vineet1101 <vineetgoel692@gmail.com>

Vineet1101 and others added 2 commits March 28, 2026 08:02

removed unecessary comments

0991080

Signed-off-by: Vineet1101 <vineetgoel692@gmail.com>

Merge branch 'main' into feature/bmv2-forstatement-support

463ea31

jafingerhut approved these changes Mar 30, 2026

View reviewed changes

jafingerhut added this pull request to the merge queue Mar 30, 2026

Merged via the queue into p4lang:main with commit 4536e01 Mar 30, 2026
19 checks passed

Conversation

Vineet1101 commented Mar 20, 2026

Proposed Changes

Verification & Proof of Compilation

Uh oh!

Vineet1101 commented Mar 20, 2026

Uh oh!

fruffy commented Mar 20, 2026

Uh oh!

jafingerhut commented Mar 20, 2026

Uh oh!

Vineet1101 commented Mar 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Vineet1101 commented Mar 20, 2026

Uh oh!

Vineet1101 commented Mar 20, 2026

Uh oh!

Vineet1101 commented Mar 20, 2026

Uh oh!

jafingerhut commented Mar 20, 2026

Uh oh!

jafingerhut commented Mar 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jafingerhut commented Mar 20, 2026

Uh oh!

fruffy commented Mar 21, 2026

Uh oh!

jafingerhut commented Mar 24, 2026

Uh oh!

Vineet1101 commented Mar 24, 2026

Uh oh!

Vineet1101 commented Mar 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

fruffy left a comment

Choose a reason for hiding this comment

Uh oh!

fruffy Mar 24, 2026

Choose a reason for hiding this comment

Uh oh!

Vineet1101 Mar 25, 2026

Choose a reason for hiding this comment

Uh oh!

fruffy Mar 26, 2026

Choose a reason for hiding this comment

Uh oh!

Vineet1101 Mar 28, 2026

Choose a reason for hiding this comment

Uh oh!

Vineet1101 commented Mar 25, 2026

Uh oh!

fruffy commented Mar 26, 2026

Uh oh!

Vineet1101 commented Mar 28, 2026

Uh oh!

fruffy commented Mar 29, 2026

Uh oh!

jafingerhut left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Vineet1101 commented Mar 20, 2026 •

edited

Loading

jafingerhut commented Mar 20, 2026 •

edited

Loading

Vineet1101 commented Mar 24, 2026 •

edited

Loading