Merge pull request #518 from Xilinx/bump_to_c1892de6

[AutoBump] Merge with fixes of c1892de (Dec 05) (128)
Xilinx · Feb 10, 2025 · d18df3d · d18df3d
2 parents 1942116 + b508995
commit d18df3d
Show file tree

Hide file tree

Showing 10 changed files with 479 additions and 1 deletion.
diff --git a/include/torch-mlir/Dialect/Torch/IR/GeneratedTorchOps.td b/include/torch-mlir/Dialect/Torch/IR/GeneratedTorchOps.td
@@ -6684,6 +6684,35 @@ def Torch_AtenConv3dOp : Torch_Op<"aten.conv3d", [
   }];
 }
 
+def Torch_AtenConv3dPaddingOp : Torch_Op<"aten.conv3d.padding", [
+    AllowsTypeRefinement,
+    HasValueSemantics,
+    ReadOnly
+  ]> {
+  let summary = "Generated op for `aten::conv3d.padding : (Tensor, Tensor, Tensor?, int[], str, int[], int) -> (Tensor)`";
+  let arguments = (ins
+    AnyTorchTensorType:$input,
+    AnyTorchTensorType:$weight,
+    AnyTorchOptionalTensorType:$bias,
+    AnyTorchListOfTorchIntType:$stride,
+    Torch_StringType:$padding,
+    AnyTorchListOfTorchIntType:$dilation,
+    Torch_IntType:$groups
+  );
+  let results = (outs
+    AnyTorchOptionalTensorType:$result
+  );
+  let hasCustomAssemblyFormat = 1;
+  let extraClassDefinition = [{
+    ParseResult AtenConv3dPaddingOp::parse(OpAsmParser &parser, OperationState &result) {
+      return parseDefaultTorchOp(parser, result, 7, 1);
+    }
+    void AtenConv3dPaddingOp::print(OpAsmPrinter &printer) {
+      printDefaultTorchOp(printer, *this, 7, 1);
+    }
+  }];
+}
+
 def Torch_AtenConv2dOp : Torch_Op<"aten.conv2d", [
     AllowsTypeRefinement,
     HasValueSemantics,
@@ -6713,6 +6742,35 @@ def Torch_AtenConv2dOp : Torch_Op<"aten.conv2d", [
   }];
 }
 
+def Torch_AtenConv2dPaddingOp : Torch_Op<"aten.conv2d.padding", [
+    AllowsTypeRefinement,
+    HasValueSemantics,
+    ReadOnly
+  ]> {
+  let summary = "Generated op for `aten::conv2d.padding : (Tensor, Tensor, Tensor?, int[], str, int[], int) -> (Tensor)`";
+  let arguments = (ins
+    AnyTorchTensorType:$input,
+    AnyTorchTensorType:$weight,
+    AnyTorchOptionalTensorType:$bias,
+    AnyTorchListOfTorchIntType:$stride,
+    Torch_StringType:$padding,
+    AnyTorchListOfTorchIntType:$dilation,
+    Torch_IntType:$groups
+  );
+  let results = (outs
+    AnyTorchOptionalTensorType:$result
+  );
+  let hasCustomAssemblyFormat = 1;
+  let extraClassDefinition = [{
+    ParseResult AtenConv2dPaddingOp::parse(OpAsmParser &parser, OperationState &result) {
+      return parseDefaultTorchOp(parser, result, 7, 1);
+    }
+    void AtenConv2dPaddingOp::print(OpAsmPrinter &printer) {
+      printDefaultTorchOp(printer, *this, 7, 1);
+    }
+  }];
+}
+
 def Torch_AtenConv1dOp : Torch_Op<"aten.conv1d", [
     AllowsTypeRefinement,
     HasValueSemantics,
@@ -6742,6 +6800,35 @@ def Torch_AtenConv1dOp : Torch_Op<"aten.conv1d", [
   }];
 }
 
+def Torch_AtenConv1dPaddingOp : Torch_Op<"aten.conv1d.padding", [
+    AllowsTypeRefinement,
+    HasValueSemantics,
+    ReadOnly
+  ]> {
+  let summary = "Generated op for `aten::conv1d.padding : (Tensor, Tensor, Tensor?, int[], str, int[], int) -> (Tensor)`";
+  let arguments = (ins
+    AnyTorchTensorType:$input,
+    AnyTorchTensorType:$weight,
+    AnyTorchOptionalTensorType:$bias,
+    AnyTorchListOfTorchIntType:$stride,
+    Torch_StringType:$padding,
+    AnyTorchListOfTorchIntType:$dilation,
+    Torch_IntType:$groups
+  );
+  let results = (outs
+    AnyTorchOptionalTensorType:$result
+  );
+  let hasCustomAssemblyFormat = 1;
+  let extraClassDefinition = [{
+    ParseResult AtenConv1dPaddingOp::parse(OpAsmParser &parser, OperationState &result) {
+      return parseDefaultTorchOp(parser, result, 7, 1);
+    }
+    void AtenConv1dPaddingOp::print(OpAsmPrinter &printer) {
+      printDefaultTorchOp(printer, *this, 7, 1);
+    }
+  }];
+}
+
 def Torch_AtenConvTranspose1dOp : Torch_Op<"aten.conv_transpose1d", [
     AllowsTypeRefinement,
     HasValueSemantics,

diff --git a/lib/Conversion/TorchToLinalg/Linear.cpp b/lib/Conversion/TorchToLinalg/Linear.cpp
@@ -1007,6 +1007,12 @@ class ConvertAtenConvolutionOp : public OpConversionPattern<AtenConvolutionOp> {
       strideInts.clear();
       strideInts.append(numSpatialDims, 1);
     } else {
+      if ((int64_t)paddingIntValues.size() + 2 !=
+          cast<RankedTensorType>(input.getType()).getRank()) {
+        // pytorch 2.5 generates one element padding = {0} for
+        // Conv2dWithValidPaddingModule
+        return rewriter.notifyMatchFailure(op, "unexpected number of padding");
+      }
       // Pad input
       paddedInput = torch_to_linalg::getDynamicZeroPaddedTensor(
           op, rewriter, input, paddingIntValues, /*unpaddedDims=*/2, pad);

diff --git a/lib/Conversion/TorchToTosa/TorchToTosa.cpp b/lib/Conversion/TorchToTosa/TorchToTosa.cpp
@@ -2385,6 +2385,12 @@ LogicalResult ConvertAtenOp<AtenConvolutionOp>::matchAndRewrite(
     return rewriter.notifyMatchFailure(op,
                                        "non-const padding list unsupported");
 
+  if (padding_2d.size() != 2) {
+    // pytorch 2.5 generates one element padding = {0} for
+    // Conv2dWithValidPaddingModule
+    return rewriter.notifyMatchFailure(op, "unexpected number of paddings");
+  }
+
   // TOSA uses 4D padding {t, b, l, r} while Torch defines 2D padding {t, l}.
   // The Torch OFM computation uses 2*pad in each spatial direction, implying
   // the same t=b and l=r values for TOSA.

diff --git a/lib/Dialect/Torch/Transforms/AbstractInterpLibrary.cpp b/lib/Dialect/Torch/Transforms/AbstractInterpLibrary.cpp
@@ -10045,10 +10045,65 @@ StringRef mlir::torch::Torch::getAbstractInterpLibrary() {
 "    %0 = call @__torch__.torch.jit._shape_functions.conv2d(%arg0, %arg1, %arg2, %arg3, %arg4, %arg5, %arg6) : (!torch.list<int>, !torch.list<int>, !torch.optional<list<int>>, !torch.list<int>, !torch.list<int>, !torch.list<int>, !torch.int) -> !torch.list<int>\n"
 "    return %0 : !torch.list<int>\n"
 "  }\n"
+"  func.func @\"__torch_mlir_shape_fn.aten.conv2d.padding\"(%arg0: !torch.list<int>, %arg1: !torch.list<int>, %arg2: !torch.optional<list<int>>, %arg3: !torch.list<int>, %arg4: !torch.str, %arg5: !torch.list<int>, %arg6: !torch.int) -> !torch.list<int> {\n"
+"    %0 = call @__torch__._conv_padding(%arg1, %arg5, %arg4) : (!torch.list<int>, !torch.list<int>, !torch.str) -> !torch.list<int>\n"
+"    %1 = call @__torch__.torch.jit._shape_functions.conv2d(%arg0, %arg1, %arg2, %arg3, %0, %arg5, %arg6) : (!torch.list<int>, !torch.list<int>, !torch.optional<list<int>>, !torch.list<int>, !torch.list<int>, !torch.list<int>, !torch.int) -> !torch.list<int>\n"
+"    return %1 : !torch.list<int>\n"
+"  }\n"
+"  func.func @__torch__._conv_padding(%arg0: !torch.list<int>, %arg1: !torch.list<int>, %arg2: !torch.str) -> !torch.list<int> {\n"
+"    %true = torch.constant.bool true\n"
+"    %int-1 = torch.constant.int -1\n"
+"    %str = torch.constant.str \"same\"\n"
+"    %none = torch.constant.none\n"
+"    %str_0 = torch.constant.str \"AssertionError: conv: weight must be at least 3 dimensional.\"\n"
+"    %int2 = torch.constant.int 2\n"
+"    %int0 = torch.constant.int 0\n"
+"    %int1 = torch.constant.int 1\n"
+"    %0 = torch.aten.len.t %arg0 : !torch.list<int> -> !torch.int\n"
+"    %1 = torch.aten.gt.int %0, %int2 : !torch.int, !torch.int -> !torch.bool\n"
+"    torch.prim.If %1 -> () {\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"
+"    %2 = torch.aten.sub.int %0, %int2 : !torch.int, !torch.int -> !torch.int\n"
+"    %3 = torch.prim.ListConstruct %int0 : (!torch.int) -> !torch.list<int>\n"
+"    %4 = torch.aten.mul.left_t %3, %2 : !torch.list<int>, !torch.int -> !torch.list<int>\n"
+"    %5 = torch.aten.eq.str %arg2, %str : !torch.str, !torch.str -> !torch.bool\n"
+"    torch.prim.If %5 -> () {\n"
+"      %6 = torch.aten.sub.int %2, %int1 : !torch.int, !torch.int -> !torch.int\n"
+"      %7 = torch.aten.len.t %arg1 : !torch.list<int> -> !torch.int\n"
+"      %8 = torch.aten.__range_length %6, %int-1, %int-1 : !torch.int, !torch.int, !torch.int -> !torch.int\n"
+"      %9 = torch.prim.ListConstruct %7, %8 : (!torch.int, !torch.int) -> !torch.list<int>\n"
+"      %10 = torch.prim.min.self_int %9 : !torch.list<int> -> !torch.int\n"
+"      torch.prim.Loop %10, %true, init() {\n"
+"      ^bb0(%arg3: !torch.int):\n"
+"        %11 = torch.aten.__getitem__.t %arg1, %arg3 : !torch.list<int>, !torch.int -> !torch.int\n"
+"        %12 = torch.aten.__derive_index %arg3, %6, %int-1 : !torch.int, !torch.int, !torch.int -> !torch.int\n"
+"        %13 = torch.aten.add.int %int2, %12 : !torch.int, !torch.int -> !torch.int\n"
+"        %14 = torch.aten.__getitem__.t %arg0, %13 : !torch.list<int>, !torch.int -> !torch.int\n"
+"        %15 = torch.aten.sub.int %14, %int1 : !torch.int, !torch.int -> !torch.int\n"
+"        %16 = torch.aten.mul.int %11, %15 : !torch.int, !torch.int -> !torch.int\n"
+"        %17 = torch.aten.floordiv.int %16, %int2 : !torch.int, !torch.int -> !torch.int\n"
+"        %18 = torch.aten._set_item.t %4, %12, %17 : !torch.list<int>, !torch.int, !torch.int -> !torch.list<int>\n"
+"        torch.prim.Loop.condition %true, iter()\n"
+"      } : (!torch.int, !torch.bool) -> ()\n"
+"      torch.prim.If.yield\n"
+"    } else {\n"
+"      torch.prim.If.yield\n"
+"    }\n"
+"    return %4 : !torch.list<int>\n"
+"  }\n"
 "  func.func @\"__torch_mlir_shape_fn.aten.conv3d\"(%arg0: !torch.list<int>, %arg1: !torch.list<int>, %arg2: !torch.optional<list<int>>, %arg3: !torch.list<int>, %arg4: !torch.list<int>, %arg5: !torch.list<int>, %arg6: !torch.int) -> !torch.list<int> {\n"
 "    %0 = call @__torch__.torch.jit._shape_functions.conv3d(%arg0, %arg1, %arg2, %arg3, %arg4, %arg5, %arg6) : (!torch.list<int>, !torch.list<int>, !torch.optional<list<int>>, !torch.list<int>, !torch.list<int>, !torch.list<int>, !torch.int) -> !torch.list<int>\n"
 "    return %0 : !torch.list<int>\n"
 "  }\n"
+"  func.func @\"__torch_mlir_shape_fn.aten.conv3d.padding\"(%arg0: !torch.list<int>, %arg1: !torch.list<int>, %arg2: !torch.optional<list<int>>, %arg3: !torch.list<int>, %arg4: !torch.str, %arg5: !torch.list<int>, %arg6: !torch.int) -> !torch.list<int> {\n"
+"    %0 = call @__torch__._conv_padding(%arg1, %arg5, %arg4) : (!torch.list<int>, !torch.list<int>, !torch.str) -> !torch.list<int>\n"
+"    %1 = call @__torch__.torch.jit._shape_functions.conv3d(%arg0, %arg1, %arg2, %arg3, %0, %arg5, %arg6) : (!torch.list<int>, !torch.list<int>, !torch.optional<list<int>>, !torch.list<int>, !torch.list<int>, !torch.list<int>, !torch.int) -> !torch.list<int>\n"
+"    return %1 : !torch.list<int>\n"
+"  }\n"
 "  func.func @\"__torch_mlir_shape_fn.aten.conv_transpose2d.input\"(%arg0: !torch.list<int>, %arg1: !torch.list<int>, %arg2: !torch.optional<list<int>>, %arg3: !torch.list<int>, %arg4: !torch.list<int>, %arg5: !torch.list<int>, %arg6: !torch.int, %arg7: !torch.list<int>) -> !torch.list<int> {\n"
 "    %0 = torch.derefine %arg3 : !torch.list<int> to !torch.optional<list<int>>\n"
 "    %1 = torch.derefine %arg4 : !torch.list<int> to !torch.optional<list<int>>\n"
@@ -10118,6 +10173,14 @@ StringRef mlir::torch::Torch::getAbstractInterpLibrary() {
 "    %1 = call @__torch__.torch.jit._shape_functions.conv_forwards(%arg0, %arg1, %arg2, %arg3, %arg4, %arg5, %false, %0, %int1) : (!torch.list<int>, !torch.list<int>, !torch.optional<list<int>>, !torch.list<int>, !torch.list<int>, !torch.list<int>, !torch.bool, !torch.list<int>, !torch.int) -> !torch.list<int>\n"
 "    return %1 : !torch.list<int>\n"
 "  }\n"
+"  func.func @\"__torch_mlir_shape_fn.aten.conv1d.padding\"(%arg0: !torch.list<int>, %arg1: !torch.list<int>, %arg2: !torch.optional<list<int>>, %arg3: !torch.list<int>, %arg4: !torch.str, %arg5: !torch.list<int>, %arg6: !torch.int) -> !torch.list<int> {\n"
+"    %false = torch.constant.bool false\n"
+"    %int1 = torch.constant.int 1\n"
+"    %0 = call @__torch__._conv_padding(%arg1, %arg5, %arg4) : (!torch.list<int>, !torch.list<int>, !torch.str) -> !torch.list<int>\n"
+"    %1 = torch.prim.ListConstruct  : () -> !torch.list<int>\n"
+"    %2 = call @__torch__.torch.jit._shape_functions.conv_forwards(%arg0, %arg1, %arg2, %arg3, %0, %arg5, %false, %1, %int1) : (!torch.list<int>, !torch.list<int>, !torch.optional<list<int>>, !torch.list<int>, !torch.list<int>, !torch.list<int>, !torch.bool, !torch.list<int>, !torch.int) -> !torch.list<int>\n"
+"    return %2 : !torch.list<int>\n"
+"  }\n"
 "  func.func @\"__torch_mlir_shape_fn.aten.conv_transpose1d\"(%arg0: !torch.list<int>, %arg1: !torch.list<int>, %arg2: !torch.optional<list<int>>, %arg3: !torch.list<int>, %arg4: !torch.list<int>, %arg5: !torch.list<int>, %arg6: !torch.int, %arg7: !torch.list<int>) -> !torch.list<int> {\n"
 "    %true = torch.constant.bool true\n"
 "    %0 = call @\"__torch_mlir_shape_fn.aten.convolution\"(%arg0, %arg1, %arg2, %arg3, %arg4, %arg7, %true, %arg5, %arg6) : (!torch.list<int>, !torch.list<int>, !torch.optional<list<int>>, !torch.list<int>, !torch.list<int>, !torch.list<int>, !torch.bool, !torch.list<int>, !torch.int) -> !torch.list<int>\n"

diff --git a/lib/Dialect/Torch/Transforms/DecomposeComplexOps.cpp b/lib/Dialect/Torch/Transforms/DecomposeComplexOps.cpp
@@ -5175,6 +5175,82 @@ class DecomposeAtenConv2dOp : public OpRewritePattern<AtenConv2dOp> {
 };
 } // namespace
 
+// Decompose aten.conv(1/2/3)d.padding to aten.convolution
+namespace {
+template <typename ConvPaddingOp>
+class DecomposeAtenConvPaddingOp : public OpRewritePattern<ConvPaddingOp> {
+public:
+  using OpRewritePattern<ConvPaddingOp>::OpRewritePattern;
+  LogicalResult matchAndRewrite(ConvPaddingOp op,
+                                PatternRewriter &rewriter) const override {
+
+    Location loc = op.getLoc();
+
+    Value weight = op.getWeight();
+    std::optional<unsigned> maybeRank = getTensorRank(weight);
+    if (!maybeRank) {
+      return rewriter.notifyMatchFailure(op, "expected weight to have a rank");
+    }
+    unsigned rank = *maybeRank;
+    // first 2 dimensions of weight are out_channels and in_channels / groups
+    if (rank < 3)
+      return rewriter.notifyMatchFailure(
+          op, "ConvPaddingOp weight must be at least 3 dimensional.");
+
+    std::string padding_str;
+    if (!matchPattern(op.getPadding(), m_TorchConstantStr(padding_str)))
+      return rewriter.notifyMatchFailure(op,
+                                         "padding must be a constant string");
+
+    Value zero = rewriter.create<Torch::ConstantIntOp>(
+        loc, rewriter.getI64IntegerAttr(0));
+
+    SmallVector<Value> paddingValues;
+    if (padding_str == "valid") {
+      // valid means no padding
+      for (unsigned iRank = 2; iRank < rank; iRank++) {
+        paddingValues.push_back(zero);
+      }
+    } else {
+
+      SmallVector<Value> dilation;
+      getListConstructElements(op.getDilation(), dilation);
+
+      Value one =
+          rewriter.create<ConstantIntOp>(loc, rewriter.getI64IntegerAttr(1));
+      Value two =
+          rewriter.create<ConstantIntOp>(loc, rewriter.getI64IntegerAttr(2));
+      for (unsigned iRank = 2; iRank < rank; iRank++) {
+        Value dim = rewriter.create<Torch::ConstantIntOp>(
+            loc, rewriter.getI64IntegerAttr(iRank));
+        Value kernelSize =
+            rewriter.create<Torch::AtenSizeIntOp>(loc, weight, dim);
+        Value kernelSizeMinusOne =
+            rewriter.create<Torch::AtenSubIntOp>(loc, kernelSize, one);
+        Value padding = rewriter.create<Torch::AtenMulIntOp>(
+            loc, dilation[iRank - 2], kernelSizeMinusOne);
+        padding = rewriter.create<AtenFloordivIntOp>(loc, padding, two);
+        paddingValues.push_back(padding);
+      }
+    }
+
+    Value emptyList = rewriter.create<PrimListConstructOp>(
+        op.getLoc(), Torch::ListType::get(Torch::IntType::get(op.getContext())),
+        SmallVector<Value>());
+    Value cstFalse = rewriter.create<Torch::ConstantBoolOp>(op.getLoc(), false);
+    Value padding = rewriter.create<PrimListConstructOp>(
+        op.getLoc(), Torch::ListType::get(Torch::IntType::get(op.getContext())),
+        paddingValues);
+    rewriter.replaceOpWithNewOp<AtenConvolutionOp>(
+        op, op->getResultTypes(), op.getInput(), op.getWeight(), op.getBias(),
+        op.getStride(), padding, op.getDilation(), cstFalse, emptyList,
+        op.getGroups());
+
+    return success();
+  }
+};
+} // namespace
+
 // Decompose aten.conv3d to aten.convolution
 namespace {
 class DecomposeAtenConv3dOp : public OpRewritePattern<AtenConv3dOp> {
@@ -11473,6 +11549,12 @@ class DecomposeComplexOpsPass
     addPatternIfTargetOpIsIllegal<DecomposeAtenConv1dOp>(patterns);
     addPatternIfTargetOpIsIllegal<DecomposeAtenConv2dOp>(patterns);
     addPatternIfTargetOpIsIllegal<DecomposeAtenConv3dOp>(patterns);
+    addPatternIfTargetOpIsIllegal<
+        DecomposeAtenConvPaddingOp<AtenConv1dPaddingOp>>(patterns);
+    addPatternIfTargetOpIsIllegal<
+        DecomposeAtenConvPaddingOp<AtenConv2dPaddingOp>>(patterns);
+    addPatternIfTargetOpIsIllegal<
+        DecomposeAtenConvPaddingOp<AtenConv3dPaddingOp>>(patterns);
     addPatternIfTargetOpIsIllegal<DecomposeAtenThresholdOp>(patterns);
     addPatternIfTargetOpIsIllegal<DecomposeAtenFloatPowerTensorTensorOp>(
         patterns);