[mlir][linalg] Use ub.poison in data layout propagation if a packed operand requires padding. (llvm#159467)

In the past, it was hard to set padding values because we did not have
ub.poison. It is not always correct if we set zeros as padding values.
Now we can use `ub.poison` in this case. The revision adds the support
for setting padding value using `ub.poison` when padding is required in
the propagation. Otherwise, it creates an invalid pack op.

Additionally the revision adds a control option for allowing padding in
the pattern which is false by default. To correctly do this, a new
`requirePaddingValueStrict` method is added which assumes dynamic dims
would mean padding is required.

The revision also removes trailing white space in the lit test file.

Co-authored-by : Nirvedh Meshram <[email protected]>

---------

Signed-off-by: hanhanW <[email protected]>
Signed-off-by: Nirvedh Meshram <[email protected]>
Co-authored-by: Nirvedh Meshram <[email protected]>

Author: hanhanW

mlir/include/mlir/Dialect/Linalg/IR/LinalgRelayoutOps.td

@@ -239,6 +239,14 @@ def Linalg_PackOp : Linalg_RelayoutOp<"pack", [
                                     ArrayRef<int64_t> outerDimsPerm,
                                     ArrayRef<OpFoldResult> innerTiles);
 
+    // Same as above function but here dynamic dimensions are assumed
+    // to require padding.
+    static bool requirePaddingValueStrict(ArrayRef<int64_t> inputShape,
+                                          ArrayRef<int64_t> innerDimsPos,
+                                          ArrayRef<int64_t> outputShape,
+                                          ArrayRef<int64_t> outerDimsPerm,
+                                          ArrayRef<OpFoldResult> innerTiles);
+
     static Value createDestinationTensor(OpBuilder &b, Location loc,
         Value source, ArrayRef<OpFoldResult> innerTileSizes,
         ArrayRef<int64_t> innerDimsPos, ArrayRef<int64_t> outerDimsPerm);

mlir/include/mlir/Dialect/Linalg/Transforms/Transforms.h

@@ -1914,9 +1914,12 @@ void populateElementwiseOpsFusionPatterns(
 using ControlPropagationFn = std::function<bool(OpOperand *opOperand)>;
 
 /// Patterns to bubble up or down data layout ops across other operations.
+/// The function also has an option to allow the patterns to propagate with
+/// poison padding if requested by the caller.
 void populateDataLayoutPropagationPatterns(
     RewritePatternSet &patterns,
-    const ControlPropagationFn &controlPackUnPackPropagation);
+    const ControlPropagationFn &controlPackUnPackPropagation,
+    bool PoisonPaddingOk = false);
 
 /// Patterns to sink extract slice across other operations.
 void populateExtractSliceSinkingPatterns(

mlir/lib/Dialect/Linalg/IR/LinalgOps.cpp

@@ -5310,6 +5310,32 @@ bool PackOp::requirePaddingValue(ArrayRef<int64_t> inputShape,
   return false;
 }
 
+bool PackOp::requirePaddingValueStrict(ArrayRef<int64_t> inputShape,
+                                       ArrayRef<int64_t> innerDimsPos,
+                                       ArrayRef<int64_t> outputShape,
+                                       ArrayRef<int64_t> outerDimsPerm,
+                                       ArrayRef<OpFoldResult> innerTiles) {
+  SmallVector<int64_t> outputTileSizes(
+      outputShape.take_front(inputShape.size()));
+  if (!outerDimsPerm.empty()) {
+    assert(outerDimsPerm.size() == outputTileSizes.size() &&
+           "expected output and outer_dims_perm to have same size");
+    applyPermutationToVector(outputTileSizes,
+                             invertPermutationVector(outerDimsPerm));
+  }
+  for (auto [pos, tileSize] : llvm::zip_equal(innerDimsPos, innerTiles)) {
+    if (ShapedType::isDynamic(inputShape[pos]) ||
+        ShapedType::isDynamic(outputTileSizes[pos]))
+      return true;
+    std::optional<int64_t> constantTile = getConstantIntValue(tileSize);
+    if (!constantTile)
+      return true;
+    if (inputShape[pos] % (*constantTile) != 0)
+      return true;
+  }
+  return false;
+}
+
 LogicalResult PackOp::verify() {
   if (failed(commonVerifierPackAndUnPackOp(*this)))
     return failure();