[TorchToLinalg] Add lowering of torch.aten.pixel_unshuffle op

include/torch-mlir/Dialect/Torch/IR/GeneratedTorchOps.td

@@ -8668,6 +8668,30 @@ def Torch_AtenPixelShuffleOp : Torch_Op<"aten.pixel_shuffle", [
   }];
 }
 
+def Torch_AtenPixelUnshuffleOp : Torch_Op<"aten.pixel_unshuffle", [
+    AllowsTypeRefinement,
+    HasValueSemantics,
+    ReadOnly
+  ]> {
+  let summary = "Generated op for `aten::pixel_unshuffle : (Tensor, int) -> (Tensor)`";
+  let arguments = (ins
+    AnyTorchTensorType:$self,
+    Torch_IntType:$downscale_factor
+  );
+  let results = (outs
+    AnyTorchOptionalTensorType:$result
+  );
+  let hasCustomAssemblyFormat = 1;
+  let extraClassDefinition = [{
+    ParseResult AtenPixelUnshuffleOp::parse(OpAsmParser &parser, OperationState &result) {
+      return parseDefaultTorchOp(parser, result, 2, 1);
+    }
+    void AtenPixelUnshuffleOp::print(OpAsmPrinter &printer) {
+      printDefaultTorchOp(printer, *this, 2, 1);
+    }
+  }];
+}
+
 def Torch_AtenPermuteOp : Torch_Op<"aten.permute", [
     AllowsTypeRefinement,
     ReadOnly

lib/Dialect/Torch/Transforms/AbstractInterpLibrary.cpp

@@ -7613,6 +7613,56 @@ StringRef mlir::torch::Torch::getAbstractInterpLibrary() {
 "    %15 = torch.aten.append.t %6, %14 : !torch.list<int>, !torch.int -> !torch.list<int>\n"
 "    return %6 : !torch.list<int>\n"
 "  }\n"
+"  func.func @\"__torch_mlir_shape_fn.aten.pixel_unshuffle\"(%arg0: !torch.list<int>, %arg1: !torch.int) -> !torch.list<int> {\n"
+"    %int1 = torch.constant.int 1\n"
+"    %int-3 = torch.constant.int -3\n"
+"    %str = torch.constant.str \"AssertionError: width  must be divisible by downscale_factor in pixel_unshuffle\"\n"
+"    %int-1 = torch.constant.int -1\n"
+"    %str_0 = torch.constant.str \"AssertionError: height  must be divisible by downscale_factor in pixel_unshuffle\"\n"
+"    %int-2 = torch.constant.int -2\n"
+"    %none = torch.constant.none\n"
+"    %str_1 = torch.constant.str \"AssertionError: input must be at least rank-3 in pixel_unshuffle\"\n"
+"    %int3 = torch.constant.int 3\n"
+"    %int0 = torch.constant.int 0\n"
+"    %0 = torch.aten.len.t %arg0 : !torch.list<int> -> !torch.int\n"
+"    %1 = torch.aten.ge.int %0, %int3 : !torch.int, !torch.int -> !torch.bool\n"
+"    torch.prim.If %1 -> () {\n"
+"      torch.prim.If.yield\n"
+"    } else {\n"
+"      torch.prim.RaiseException %str_1, %none : !torch.str, !torch.none\n"
+"      torch.prim.If.yield\n"
+"    }\n"
+"    %2 = torch.aten.mul.int %arg1, %arg1 : !torch.int, !torch.int -> !torch.int\n"
+"    %3 = torch.aten.__getitem__.t %arg0, %int-2 : !torch.list<int>, !torch.int -> !torch.int\n"
+"    %4 = torch.aten.remainder.int %3, %arg1 : !torch.int, !torch.int -> !torch.int\n"
+"    %5 = torch.aten.eq.int %4, %int0 : !torch.int, !torch.int -> !torch.bool\n"
+"    torch.prim.If %5 -> () {\n"
+"      torch.prim.If.yield\n"
+"    } else {\n"
+"      torch.prim.RaiseException %str_0, %none : !torch.str, !torch.none\n"
+"      torch.prim.If.yield\n"
+"    }\n"
+"    %6 = torch.aten.__getitem__.t %arg0, %int-1 : !torch.list<int>, !torch.int -> !torch.int\n"
+"    %7 = torch.aten.remainder.int %6, %arg1 : !torch.int, !torch.int -> !torch.int\n"
+"    %8 = torch.aten.eq.int %7, %int0 : !torch.int, !torch.int -> !torch.bool\n"
+"    torch.prim.If %8 -> () {\n"
+"      torch.prim.If.yield\n"
+"    } else {\n"
+"      torch.prim.RaiseException %str, %none : !torch.str, !torch.none\n"
+"      torch.prim.If.yield\n"
+"    }\n"
+"    %9 = torch.aten.slice.t %arg0, %int0, %int-3, %int1 : !torch.list<int>, !torch.int, !torch.int, !torch.int -> !torch.list<int>\n"
+"    %10 = torch.aten.__getitem__.t %arg0, %int-3 : !torch.list<int>, !torch.int -> !torch.int\n"
+"    %11 = torch.aten.mul.int %10, %2 : !torch.int, !torch.int -> !torch.int\n"
+"    %12 = torch.aten.append.t %9, %11 : !torch.list<int>, !torch.int -> !torch.list<int>\n"
+"    %13 = torch.aten.__getitem__.t %arg0, %int-2 : !torch.list<int>, !torch.int -> !torch.int\n"
+"    %14 = torch.aten.floordiv.int %13, %arg1 : !torch.int, !torch.int -> !torch.int\n"
+"    %15 = torch.aten.append.t %9, %14 : !torch.list<int>, !torch.int -> !torch.list<int>\n"
+"    %16 = torch.aten.__getitem__.t %arg0, %int-1 : !torch.list<int>, !torch.int -> !torch.int\n"
+"    %17 = torch.aten.floordiv.int %16, %arg1 : !torch.int, !torch.int -> !torch.int\n"
+"    %18 = torch.aten.append.t %9, %17 : !torch.list<int>, !torch.int -> !torch.list<int>\n"
+"    return %9 : !torch.list<int>\n"
+"  }\n"
 "  func.func @\"__torch_mlir_shape_fn.aten.permute\"(%arg0: !torch.list<int>, %arg1: !torch.list<int>) -> !torch.list<int> {\n"
 "    %0 = call @__torch__.torch.jit._shape_functions.permute(%arg0, %arg1) : (!torch.list<int>, !torch.list<int>) -> !torch.list<int>\n"
 "    return %0 : !torch.list<int>\n"
@@ -12305,6 +12355,10 @@ StringRef mlir::torch::Torch::getAbstractInterpLibrary() {
 "    %0:2 = torch.prim.TupleUnpack %arg0 : !torch.tuple<int, int> -> !torch.int, !torch.int\n"
 "    return %0#1 : !torch.int\n"
 "  }\n"
+"  func.func @\"__torch_mlir_dtype_fn.aten.pixel_unshuffle\"(%arg0: !torch.tuple<int, int>, %arg1: !torch.int) -> !torch.int {\n"
+"    %0:2 = torch.prim.TupleUnpack %arg0 : !torch.tuple<int, int> -> !torch.int, !torch.int\n"
+"    return %0#1 : !torch.int\n"
+"  }\n"
 "  func.func @\"__torch_mlir_dtype_fn.aten.avg_pool1d\"(%arg0: !torch.tuple<int, int>, %arg1: !torch.list<int>, %arg2: !torch.list<int>, %arg3: !torch.list<int>, %arg4: !torch.bool, %arg5: !torch.bool) -> !torch.int {\n"
 "    %none = torch.constant.none\n"
 "    %str = torch.constant.str \"AssertionError: \"\n"

lib/Dialect/Torch/Transforms/DecomposeComplexOps.cpp

@@ -23,6 +23,7 @@
 #include "llvm/ADT/StringExtras.h"
 #include "llvm/ADT/StringSet.h"
 #include <cstdint>
+#include <iostream>
 #include <set>
 
 using namespace mlir;
@@ -3708,6 +3709,177 @@ class DecomposeAtenPixelShuffleOp
 };
 } // namespace
 
+// Decompose aten.pixel_unshuffle into: prims.split_dim, aten.permute, and
+// prims.collapse operations.
+//
+// We want to do the exact opposite of aten.pixel_shuffle
+//
+// If input is a tensor of shape
+//     (*leading_dims, C, H*r, W*r),
+//
+// where leading_dims is of size N, then
+//    X = pixel_unshuffle(input, downscale_factor)
+//
+// gets replaced with
+//    X = input.split_dim(...)  # shape (*leading_dims, C, H, r, W*r)
+//    X = X.split_dim(...)      # shape (*leading_dims, C, H, r, W, r)
+//    X = X.permute(0, ..., N, N+2, N+4, N+1, N+3)
+//                              # shape (*leading_dims, C, r, r, H, W)
+//    X = X.collapse(...)       # shape (*leading_dims, C, r*r, H, W)
+//    X = X.collapse(...)       # shape (*leading_dims, C*r*r, H, W)
+//
+// 'r' above is referred to as the 'downscale factor' or just 'factor' below.
+namespace {
+class DecomposeAtenPixelUnshuffleOp
+    : public OpRewritePattern<AtenPixelUnshuffleOp> {
+public:
+  using OpRewritePattern::OpRewritePattern;
+  LogicalResult matchAndRewrite(AtenPixelUnshuffleOp op,
+                                PatternRewriter &rewriter) const override {
+
+    Location loc = op.getLoc();
+    Value inValue = op.getSelf();
+    auto inType = cast<BaseTensorType>(inValue.getType());
+    auto maybeSizes = inType.getOptionalSizes();
+    if (!maybeSizes) {
+      return rewriter.notifyMatchFailure(
+          op, "Expected input tensor to have known rank.");
+    }
+    auto inShape = maybeSizes.value();
+    auto inRank = inShape.size();
+
+    // The input tensor must have at least 3 dimensions: (1) the channel
+    // dimension which gets bigger by 'factor*factor', (2) the H channel which
+    // gets smaller by 'factor' and (3) the W channel which get smaller by
+    // 'factor'. The total number of dimensions is 3 + N, where N is the number
+    // of leading dimensions, and N >= 0 so the input must have rank at least 3.
+    if (inRank < 3)
+      return rewriter.notifyMatchFailure(
+          op, "Expected input tensor to have rank greater than 2.");
+
+    const auto inOptionalDType = inType.getOptionalDtype();
+
+    auto getTypeFromShape = [inOptionalDType](auto &&vals) {
+      // Get a vector of integers from a vector of Values.
+      auto getIntShape = [](auto &&vals) {
+        SmallVector<int64_t> shape;
+        shape.reserve(vals.size());
+        for (auto v : vals) {
+          int64_t cst_val;
+          if (matchPattern(v, m_TorchConstantInt(&cst_val))) {
+            shape.push_back(cst_val);
+          } else {
+            shape.push_back(kUnknownSize);
+          }
+        }
+        return shape;
+      };
+
+      const auto intShape = getIntShape(vals);
+      return ValueTensorType::get(vals[0].getContext(),
+                                  llvm::ArrayRef(intShape), inOptionalDType);
+    };
+
+    auto nLeadingDims = inRank - 3;
+
+    // Get the size of the dimension 'i'. Note the use of 'createOrFold' instead
+    // of 'create': if the dimension size is known, then the AtenSizeIntOp is
+    // folded to a ConstantOp.
+    auto getDimSize = [&](uint64_t i) -> Value {
+      Value dim =
+          rewriter.create<ConstantIntOp>(loc, rewriter.getI64IntegerAttr(i));
+      return rewriter.createOrFold<AtenSizeIntOp>(loc, inValue, dim);
+    };
+
+    auto inC = getDimSize(inRank - 3);
+    auto inH = getDimSize(inRank - 2);
+    auto inW = getDimSize(inRank - 1);
+
+    auto factor = op.getDownscaleFactor();
+
+    Value factorSquared =
+        rewriter.createOrFold<AtenMulIntOp>(loc, factor, factor);
+
+    Value outC = rewriter.createOrFold<AtenMulIntOp>(loc, inC, factorSquared);
+
+    Value outH = rewriter.createOrFold<AtenFloordivIntOp>(loc, inH, factor);
+    Value outW = rewriter.createOrFold<AtenFloordivIntOp>(loc, inW, factor);
+
+    SmallVector<Value> dimensionConstants;
+    dimensionConstants.reserve(inRank + 2);
+    for (unsigned i = 0; i < inRank + 2; ++i) {
+      dimensionConstants.push_back(
+          rewriter.create<ConstantIntOp>(loc, rewriter.getI64IntegerAttr(i)));
+    }
+
+    SmallVector<Value> leadingDims;
+    leadingDims.reserve(nLeadingDims);
+    for (unsigned i = 0; i < nLeadingDims; ++i) {
+      Value leadingDimSize = rewriter.createOrFold<AtenSizeIntOp>(
+          loc, inValue, dimensionConstants[i]);
+      leadingDims.push_back(leadingDimSize);
+    }
+
+    SmallVector<Value> partiallyExpandedShape = leadingDims;
+    partiallyExpandedShape.append({inC, outH, factor, inW});
+
+    SmallVector<Value> prePermuteShape = leadingDims;
+    prePermuteShape.append({inC, outH, factor, outW, factor});
+
+    SmallVector<Value> postPermuteShape = leadingDims;
+    postPermuteShape.append({inC, factor, factor, outH, outW});
+
+    SmallVector<Value> partiallyCollapsedShape = leadingDims;
+    partiallyCollapsedShape.append({inC, factorSquared, outH, outW});
+
+    SmallVector<Value> outShape = leadingDims;
+    outShape.append({outC, outH, outW});
+
+    SmallVector<Value> permutation{dimensionConstants.begin(),
+                                   dimensionConstants.begin() + nLeadingDims};
+    SmallVector<uint64_t> permutationTail{0, 2, 4, 1, 3};
+    for (uint64_t d : permutationTail) {
+      permutation.push_back(dimensionConstants[nLeadingDims + d]);
+    }
+
+    Value permuteDimsOrder = rewriter.create<PrimListConstructOp>(
+        loc, Torch::ListType::get(Torch::IntType::get(op->getContext())),
+        permutation);
+
+    // Split input channel inH -> (outH, factor)
+    auto partiallyExpanded =
+        rewriter
+            .create<PrimsSplitDimOp>(
+                loc, getTypeFromShape(partiallyExpandedShape), inValue,
+                dimensionConstants[nLeadingDims + 1], outH)
+            .getResult();
+
+    // Split new dimension inW -> (outW, factor)
+    auto fullyExpanded = rewriter.create<PrimsSplitDimOp>(
+        loc, getTypeFromShape(prePermuteShape), partiallyExpanded,
+        dimensionConstants[nLeadingDims + 3], outW);
+
+    // Perform the permutation
+    auto permuted =
+        rewriter.create<AtenPermuteOp>(loc, getTypeFromShape(postPermuteShape),
+                                       fullyExpanded, permuteDimsOrder);
+
+    // Collapse final 2 dimension
+    auto partiallyCollapsed = rewriter.create<PrimsCollapseOp>(
+        loc, getTypeFromShape(partiallyCollapsedShape), permuted,
+        dimensionConstants[nLeadingDims + 1],
+        dimensionConstants[nLeadingDims + 2]);
+
+    // Collapse back to original rank
+    rewriter.replaceOpWithNewOp<PrimsCollapseOp>(
+        op, op.getType(), partiallyCollapsed, dimensionConstants[nLeadingDims],
+        dimensionConstants[nLeadingDims + 1]);
+
+    return success();
+  }
+};
+} // namespace
+
 // ReLU6(x) = min(max(0, x), 6) = min(Relu(x), 6)
 static Value getRelu6Results(PatternRewriter &rewriter, Location loc,
                              Value input) {
@@ -12518,6 +12690,7 @@ class DecomposeComplexOpsPass
     addPatternIfTargetOpIsIllegal<DecomposeAtenRenormOp>(patterns);
     addPatternIfTargetOpIsIllegal<DecomposeAtenLinalgCrossOp>(patterns);
     addPatternIfTargetOpIsIllegal<DecomposeAtenPixelShuffleOp>(patterns);
+    addPatternIfTargetOpIsIllegal<DecomposeAtenPixelUnshuffleOp>(patterns);
     addPatternIfTargetOpIsIllegal<DecomposeAtenTOp>(patterns);
     addPatternIfTargetOpIsIllegal<DecomposeAten_LogSoftmaxBackwardDataOp>(
         patterns);

lib/Dialect/Torch/Transforms/LowerToBackendContract.cpp

@@ -421,6 +421,7 @@ static void markDecomposedOpsAsIllegal(MLIRContext *context,
   target.addIllegalOp<Aten_LinalgDetOp>();
   target.addIllegalOp<AtenLinalgSlogdetOp>();
   target.addIllegalOp<AtenPixelShuffleOp>();
+  target.addIllegalOp<AtenPixelUnshuffleOp>();
   target.addIllegalOp<AtenTOp>();
   target.addIllegalOp<Aten_LogSoftmaxBackwardDataOp>();
   target.addDynamicallyLegalOp<AtenMatmulOp>([](AtenMatmulOp op) {

lib/Dialect/Torch/Utils/Utils.cpp

@@ -317,17 +317,17 @@ bool Torch::isViewLikeOp(Operation *op) {
   // correct. We could potentially be more precise and identify the cases
   // that it does not return a view and treat those as having value
   // semantics.
-  return isa<AtenAsStridedOp, AtenBroadcastToOp, AtenContiguousOp, AtenDetachOp,
-             AtenExpandAsOp, AtenExpandOp, AtenFlattenUsingIntsOp,
-             AtenUnflattenIntOp, AtenPermuteOp, AtenReshapeOp,
-             Aten_ReshapeAliasOp, AtenSelectIntOp, AtenSliceTensorOp,
-             AtenSqueezeDimOp, AtenSqueezeOp, AtenTOp, AtenToDtypeOp,
-             AtenTransposeIntOp, AtenUnsqueezeOp, AtenViewOp,
-             TensorStaticInfoCastOp, AtenToDtypeLayoutOp, AtenNumpyTOp,
-             AtenNarrowOp, AtenNarrowTensorOp, AtenToDeviceOp, PrimsSqueezeOp,
-             AtenMovedimIntOp, PrimsViewOfOp, AtenRealOp, AtenImagOp,
-             PrimsSplitDimOp, AtenViewAsComplexOp, AtenViewAsRealOp,
-             AtenPixelShuffleOp, AtenDiagonalOp, AtenUnfoldOp>(op);
+  return isa<
+      AtenAsStridedOp, AtenBroadcastToOp, AtenContiguousOp, AtenDetachOp,
+      AtenExpandAsOp, AtenExpandOp, AtenFlattenUsingIntsOp, AtenUnflattenIntOp,
+      AtenPermuteOp, AtenReshapeOp, Aten_ReshapeAliasOp, AtenSelectIntOp,
+      AtenSliceTensorOp, AtenSqueezeDimOp, AtenSqueezeOp, AtenTOp,
+      AtenToDtypeOp, AtenTransposeIntOp, AtenUnsqueezeOp, AtenViewOp,
+      TensorStaticInfoCastOp, AtenToDtypeLayoutOp, AtenNumpyTOp, AtenNarrowOp,
+      AtenNarrowTensorOp, AtenToDeviceOp, PrimsSqueezeOp, AtenMovedimIntOp,
+      PrimsViewOfOp, AtenRealOp, AtenImagOp, PrimsSplitDimOp,
+      AtenViewAsComplexOp, AtenViewAsRealOp, AtenPixelShuffleOp,
+      AtenPixelUnshuffleOp, AtenDiagonalOp, AtenUnfoldOp>(op);
 }
 
 Value Torch::getConstantWithGivenDtypeAndValue(PatternRewriter &rewriter,

projects/pt1/e2e_testing/xfail_sets.py

@@ -3122,6 +3122,11 @@
     "PixelShuffleModuleSpatiallyDynamic_basic",
     "PixelShuffleModuleSpatiallyStatic_basic",
     "PixelShuffleModuleStaticRank3Int64_basic",
+    "PixelUnshuffleModuleStaticRank5Float32_basic",
+    "PixelUnshuffleModuleStaticRank3Int64_basic",
+    "PixelUnshuffleModuleFullDynamic_basic",
+    "PixelUnshuffleModuleSpatiallyDynamic_basic",
+    "PixelUnshuffleModuleSpatiallyStatic_basic",
     "PowIntIntModule_basic",
     "PrimMaxIntModule_basic",
     "PrimMinIntDynamicModule_basic",
@@ -4710,6 +4715,11 @@
     "PixelShuffleModuleSpatiallyStatic_basic",
     "PixelShuffleModuleStaticRank3Int64_basic",
     "PixelShuffleModuleStaticRank4Float32_basic",
+    "PixelUnshuffleModuleStaticRank5Float32_basic",
+    "PixelUnshuffleModuleStaticRank3Int64_basic",
+    "PixelUnshuffleModuleFullDynamic_basic",
+    "PixelUnshuffleModuleSpatiallyDynamic_basic",
+    "PixelUnshuffleModuleSpatiallyStatic_basic",
     "PrimMaxIntModule_basic",
     "PrimMinIntDynamicModule_basic",
     "PrimMinIntModule_basic",

projects/pt1/python/torch_mlir/jit_ir_importer/build_tools/abstract_interp_lib_gen.py

@@ -839,6 +839,19 @@ def aten〇pixel_shuffle〡shape(self: List[int], upscale_factor: int) -> List[i
     out.append(self[-1] * upscale_factor)
     return out
 
+def aten〇pixel_unshuffle〡shape(self: List[int], downscale_factor: int) -> List[int]:
+
+    assert len(self) >= 3, "input must be at least rank-3 in pixel_unshuffle"
+    downscale_factor_squared = downscale_factor * downscale_factor
+    assert self[-2] % (downscale_factor) == 0, "height  must be divisible by downscale_factor in pixel_unshuffle"
+    assert self[-1] % (downscale_factor) == 0, "width  must be divisible by downscale_factor in pixel_unshuffle"
+
+    out = self[0:-3]
+    out.append(self[-3] * downscale_factor_squared)
+    out.append(self[-2] // downscale_factor)
+    out.append(self[-1] // downscale_factor)
+    return out
+
 
 
 def aten〇permute〡shape(self: List[int], dims: List[int]) -> List[int]:
@@ -3058,6 +3071,11 @@ def aten〇pixel_shuffle〡dtype(self_rank_dtype: Tuple[int, int], upscale_facto
     self_rank, self_dtype = self_rank_dtype
     return self_dtype
 
+@check_dtype_function(_check_tensors_with_the_same_dtype(tensor_shapes=[(1, 2, 2)], downscale_factor = 2))
+def aten〇pixel_unshuffle〡dtype(self_rank_dtype: Tuple[int, int], downscale_factor: int) -> int:
+    self_rank, self_dtype = self_rank_dtype
+    return self_dtype
+
 @check_dtype_function(_check_tensors_with_the_same_dtype(tensor_shapes=[(2, 3, 7)], kernel_size=[2], error_types={torch.uint8}))
 def aten〇avg_pool1d〡dtype(self_rank_dtype: Tuple[int, int], kernel_size: List[int], stride: List[int] = (), padding: List[int] = (0,), ceil_mode: bool = False, count_include_pad: bool = True) -> int:
     self_rank, self_dtype = self_rank_dtype

projects/pt1/python/torch_mlir/jit_ir_importer/build_tools/torch_ods_gen.py

@@ -719,6 +719,7 @@ def emit_with_mutating_variants(key, **kwargs):
     emit("aten::topk : (Tensor, int, int, bool, bool) -> (Tensor, Tensor)")
     emit("aten::transpose.int : (Tensor, int, int) -> (Tensor)", has_folder=True)
     emit("aten::pixel_shuffle : (Tensor, int) -> (Tensor)")
+    emit("aten::pixel_unshuffle : (Tensor, int) -> (Tensor)")
     emit("aten::permute : (Tensor, int[]) -> (Tensor)", has_verifier=True)
     emit("aten::movedim.int : (Tensor, int, int) -> (Tensor)")
     emit("aten::bmm : (Tensor, Tensor) -> (Tensor)")