[mlir][tosa] Check for isolated regions in tosa.while_loop
#144865
+146
−44
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This commit fixes a check in the validation pass which intended to validate whether a `tosa.cond_if` operation was conformant to the specification. The specification requires all values used in the then/else regions are explicitly declared within the regions. This change checks that these regions are 'isolated from above', to ensure this requirement is true. Change-Id: I1b6eac1ed571e6b1eda4a58f0677c80e22977e58
Similarly to `tosa.cond_if`, this patch checks that the cond/body regions of `tosa.while_loop` are isolated from above. This is required since the specification requires all values used in the cond/body regions are explicitly declared within the regions. Change-Id: Ia7396b9811db54805ec33befd24ab97d1b605905
FranklandJack
requested changes
Jul 16, 2025
| bool isElseGraphNullaryRegion = isNullaryRegion(elseGraph); | ||
| bool isInputListEmpty = ifOp.getInputList().size() == 0; | ||
| bool checkErrorIfWhileLoop(Operation *op) { | ||
| auto whileOp = dyn_cast<tosa::WhileOp>(op); |
There was a problem hiding this comment.
Could this be const?
|
|
||
| if ((isInputListEmpty != isThenGraphNullaryRegion) || | ||
| (isInputListEmpty != isElseGraphNullaryRegion)) { | ||
| Region &condGraph = whileOp.getCondGraph(); |
| (isInputListEmpty != isElseGraphNullaryRegion)) { | ||
| Region &condGraph = whileOp.getCondGraph(); | ||
| Region &bodyGraph = whileOp.getBodyGraph(); | ||
| bool isCondGraphIsolatedRegion = isRegionIsolatedFromAbove(condGraph); |
| if ((isInputListEmpty != isThenGraphNullaryRegion) || | ||
| (isInputListEmpty != isElseGraphNullaryRegion)) { | ||
| Region &condGraph = whileOp.getCondGraph(); | ||
| Region &bodyGraph = whileOp.getBodyGraph(); |
| Region &condGraph = whileOp.getCondGraph(); | ||
| Region &bodyGraph = whileOp.getBodyGraph(); | ||
| bool isCondGraphIsolatedRegion = isRegionIsolatedFromAbove(condGraph); | ||
| bool isBodyGraphIsolatedRegion = isRegionIsolatedFromAbove(bodyGraph); |
| op->emitOpError() | ||
| << "the current simplified form is not strictly conformant to the " | ||
| "spec, please use the generic format\n"; | ||
| << "is not conformant to the TOSA specification. It requires the " |
There was a problem hiding this comment.
Similarly to the if case, we could provide more information here if we were to split the checks into two. Only a week preference for this though.
| %0 = "tosa.const"() {values = dense<0> : tensor<i32>} : () -> tensor<i32> | ||
| // expected-error@+1 {{'tosa.while_loop' op is not conformant to the TOSA specification. It requires the cond/body regions are isolated from above.}} | ||
| %1 = "tosa.while_loop"(%0) ({ | ||
| ^bb0(%arg3: tensor<i32>): |
There was a problem hiding this comment.
Should the blocks have the same ID bb0?
| func.func @test_while_loop_isolated_from_above(%arg0: tensor<f32>, %arg1: tensor<i32>) { | ||
| %0 = "tosa.const"() {values = dense<0> : tensor<i32>} : () -> tensor<i32> | ||
| %1:3 = "tosa.while_loop"(%0, %arg0, %arg1) ({ | ||
| ^bb0(%arg3: tensor<i32>, %arg4: tensor<f32>, %arg5: tensor<i32>): |
There was a problem hiding this comment.
Should the blocks have the same ID bb0?
|
|
||
| // ----- | ||
|
|
||
| // COM: Check isolated while_loops are valid |
There was a problem hiding this comment.
Do we need to use the COM directive here if there are no strings in the comment that could be interpreted as a FileCheck directive?
| %3 = "tosa.logical_not"(%2) : (tensor<i1>) -> tensor<i1> | ||
| tosa.yield %3 : tensor<i1> | ||
| }, { | ||
| ^bb0(%arg3: tensor<i32>): |
There was a problem hiding this comment.
Here it looks like the body graph is isolated from above, and it is only the condition graph that isn't. Should we add another test to check the error is triggered when the body graph isn't isolated from above?
udaya-ranga
reviewed
Jul 16, 2025
| isa<tosa::YieldOp>(op) || isa<tosa::VariableOp>(op)) | ||
| return true; | ||
| return false; | ||
| static bool isOpIsolatedFromAbove(Operation *op, Region ®ion) { |
There was a problem hiding this comment.
Will there just be 1 copy after merging this and the other review for cond.if or is this patch built on top of the other so that it appears here as well ? I guess the latter ?
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Similarly to
tosa.cond_if, this patch checks that the cond/body regions oftosa.while_loopare isolated from above. This is required since the specification requires all values used in the cond/body regions are explicitly declared within the regions.Note: this change is dependent on #143772