Merge feature/backport_ea1_ops into HEAD

lib/Conversion/TorchOnnxToTorch/Patterns.cpp

@@ -8,6 +8,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "torch-mlir/Conversion/TorchOnnxToTorch/Patterns.h"
+#include "mlir/IR/BuiltinAttributes.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/raw_ostream.h"
 
@@ -24,12 +25,28 @@ LogicalResult OnnxCustomOpConversionPattern::matchAndRewrite(
   auto foundIt = namedHandlers.find(op.getNameAttr());
   if (foundIt == namedHandlers.end())
     return failure();
+  // The domainVersion comes from the function attribute
+  // torch.onnx_meta.opset_version and defines the opset for all ONNX ops the
+  // function contains. Absent this attribute, domainVersion is 0.
+  int64_t opDomainVersion = domainVersion;
+  // If the op has an individual version (torch.onnx_meta.version attribute), it
+  // overrides the function's domainVersion and will be used for matching later
+  // here.
+  if (auto attr = op->getAttrOfType<IntegerAttr>("torch.onnx_meta.version")) {
+    if (auto type = dyn_cast<IntegerType>(attr.getType())) {
+      if (type.isSigned()) {
+        opDomainVersion =
+            op->getAttrOfType<IntegerAttr>("torch.onnx_meta.version").getSInt();
+      }
+    }
+  }
   auto &reggies = foundIt->second;
   for (const HandlerReg &reg : reggies) {
-    if (domainVersion < reg.sinceVersion) {
+    if (opDomainVersion < reg.sinceVersion) {
       LLVM_DEBUG(dbgs() << ": skipping conversion " << foundIt->first
                         << ", sinceVersion=" << reg.sinceVersion
-                        << ", for domainVersion=" << domainVersion << "\n");
+                        << ", for domainVersion=" << domainVersion
+                        << ", opDomainVersion=" << opDomainVersion << "\n");
       continue;
     }
     if (succeeded(reg.callback(OpBinder(op), rewriter))) {

lib/Conversion/TorchOnnxToTorch/TorchOnnxToTorch.cpp

@@ -45,12 +45,6 @@ class ConvertTorchOnnxToTorch
 
     // Populate our patterns for each handled domain.
     int64_t defaultOpsetVersion = getDefaultOpsetVersion(getOperation());
-    if (defaultOpsetVersion == 0) {
-      emitError(getOperation().getLoc())
-          << "function is missing onnx opset version attribute "
-             "(torch.onnx_meta.opset_version)";
-      return signalPassFailure();
-    }
 
     auto defaultDomainPatterns =
         std::make_unique<OnnxCustomOpConversionPattern>(

lib/Conversion/TorchToTosa/TorchToTosa.cpp

@@ -3066,6 +3066,7 @@ static Value buildUnitNormalCdf(ConversionPatternRewriter &rewriter,
                                 Operation *op, Value x, Type dtype) {
   auto zero = tosa::getConstTensor<float>(rewriter, op, 0, {}, dtype).value();
   auto one = tosa::getConstTensor<float>(rewriter, op, 1, {}, dtype).value();
+
   auto loc = op->getLoc();
 
   // buildNormalCdf, mean = zero, sigma = one

lib/Conversion/TorchToTosa/TosaLegalizeUtils.cpp

@@ -240,6 +240,7 @@ std::optional<Value> getConstTensor<APInt>(PatternRewriter &rewriter,
 
   auto const_op =
       rewriter.create<tosa::ConstOp>(op->getLoc(), const_type, const_attr);
+
   if (dtype) {
     return rewriter.createOrFold<tosa::CastOp>(
         op->getLoc(), RankedTensorType::get(shape, *dtype), const_op);

lib/Dialect/Torch/Transforms/DecomposeComplexOps.cpp

@@ -7538,8 +7538,8 @@ class DecomposeComplexOpsPass
     addPatternIfTargetOpIsIllegal<DecomposeAtenVarMeanDimOp>(patterns);
     addPatternIfTargetOpIsIllegal<DecomposeAtenTopkOp>(patterns);
     addPatternIfTargetOpIsIllegal<DecomposeAtenScalarTensor>(patterns);
-    addPatternIfTargetOpIsIllegal<DecomposeAtenSignOp>(patterns);
     addPatternIfTargetOpIsIllegal<DecomposeAtenScatterValueOp>(patterns);
+    addPatternIfTargetOpIsIllegal<DecomposeAtenSignOp>(patterns);
     addPatternIfTargetOpIsIllegal<DecomposeAtenArcSinCosOp<AtenAsinOp>>(
         patterns);
     addPatternIfTargetOpIsIllegal<DecomposeAtenArcSinCosOp<AtenAcosOp>>(

lib/Dialect/Torch/Transforms/RecomposeComplexOps.cpp

@@ -70,7 +70,6 @@ class RecomposeSliceCopy_ : public OpRewritePattern<AtenCopy_Op> {
       newEnd =
           rewriter.create<AtenAddIntOp>(op.getLoc(), dimSize, sliceOp.getEnd());
     }
-    newEnd = rewriter.create<PrimMinIntOp>(op.getLoc(), newEnd, dimSize);
 
     newStart = rewriter.create<PrimMinIntOp>(op.getLoc(), newStart, dimSize);
     newEnd = rewriter.create<PrimMinIntOp>(op.getLoc(), newEnd, dimSize);

projects/pt1/e2e_testing/xfail_sets.py

@@ -2328,7 +2328,6 @@
     "ElementwiseAcosTensorIntModule_basic",
     "ElementwiseAsinTensorIntModule_basic",
     "FakeQuantizePerTensorAffineCachemaskModule_basic",
-    "Im2ColModule_basic",
     "IndexPutImpl2DNoneIndexBroadcastStaticModule_basic",
     "PrimsSumFloatModule_basic",
     "RepeatInterleaveFillModule_basic",

projects/pt1/python/torch_mlir/dynamo.py

@@ -65,10 +65,7 @@ def _get_decomposition_table():
         aten._native_batch_norm_legit,
         aten.squeeze,
         aten.cumsum,
-        aten.im2col,
         aten.index_select,
-        aten.linalg_vector_norm,
-        aten.eye,
     ]
     # TODO: enable test once 2.1.0 is stable
     if torch_version_for_comparison() >= version.parse("2.1.0.dev"):

projects/pt1/python/torch_mlir_e2e_test/test_suite/__init__.py

@@ -14,7 +14,6 @@
     "QuantizedMLP_basic",
     "ReduceMaxAlongDimUnsignedInt_basic",
     "RepeatInterleaveModule_basic",
-    "Im2ColModule_basic",
     "ReduceMinAlongDimUnsignedInt_basic",
     "ElementwiseToDtypeI64ToUI8Module_basic",
 }

projects/pt1/python/torch_mlir_e2e_test/test_suite/basic.py

@@ -5141,26 +5141,6 @@ def forward(self, x):
 def Add_Module_basic(module, tu: TestUtils):
     module.forward(tu.rand(2, 3))
 
-# ==============================================================================
-
-class Im2Col_Module(torch.nn.Module):
-
-    def __init__(self):
-        super().__init__()
-        self.tensor = torch.ones(2, 3)
-
-    @export
-    @annotate_args([
-        None,
-        ([-1, -1, -1, -1], torch.float32, True),
-    ])
-    def forward(self, x):
-        return torch.ops.aten.im2col(x, [9, 1], [1, 1], [4, 0], [1, 1]);
-
-@register_test_case(module_factory=lambda: Im2Col_Module())
-def Im2ColModule_basic(module, tu: TestUtils):
-    module.forward(tu.rand(3,4,5,2))
-
 
 # ==============================================================================
 

projects/pt1/python/torch_mlir_e2e_test/test_suite/constant_alloc.py

@@ -1850,22 +1850,6 @@ def NewEmptyStridedModuleDefaultDtype_basic(module, tu: TestUtils):
 # ==============================================================================
 
 
-class EyeStaticModule(torch.nn.Module):
-    @export
-    @annotate_args([
-        None,
-    ])
-    def forward(self):
-        return torch.ops.aten.eye(3, 5)
-
-
-@register_test_case(module_factory=lambda: EyeStaticModule())
-def EyeStaticModule_basic(module, tu: TestUtils):
-    module.forward()
-
-# ==============================================================================
-
-
 class EmptyStridedModule(torch.nn.Module):
 
     def __init__(self):

projects/pt1/python/torch_mlir_e2e_test/test_suite/matmul.py

@@ -419,4 +419,4 @@ def forward(self, a, b):
 
 @register_test_case(module_factory=lambda: AtenLinalgCrossDynamic())
 def AtenLinalgCrossDynamic_basic(module, tu: TestUtils):
-    module.forward(tu.rand(4, 3, 1, 6), tu.rand(4, 3, 7, 1))
+    module.forward(tu.rand(4, 3, 1, 6), tu.rand(4, 3, 7, 1))

test/Conversion/TorchOnnxToTorch/op_wise_version.mlir

@@ -0,0 +1,17 @@
+// RUN: torch-mlir-opt <%s --split-input-file -convert-torch-onnx-to-torch | FileCheck %s
+
+// CHECK-LABEL: @test_quantizelinear_opset_16_op_19
+func.func @test_quantizelinear_opset_16_op_19(%arg0: !torch.vtensor<[6],f32>, %arg1: !torch.vtensor<[],f32>, %arg2: !torch.vtensor<[],si8>) -> !torch.vtensor<[6],si8> attributes {torch.onnx_meta.ir_version = 9 : si64, torch.onnx_meta.opset_version = 16 : si64} {
+  // CHECK-NOT: torch.operator
+  %0 = torch.operator "onnx.QuantizeLinear"(%arg0, %arg1, %arg2) {torch.onnx_meta.version = 19 : si64} : (!torch.vtensor<[6],f32>, !torch.vtensor<[],f32>, !torch.vtensor<[],si8>) -> !torch.vtensor<[6],si8>
+  return %0 : !torch.vtensor<[6],si8>
+}
+
+// -----
+
+// CHECK-LABEL: @test_quantizelinear_no_opset_op_19
+func.func @test_quantizelinear_no_opset_op_19(%arg0: !torch.vtensor<[6],f32>, %arg1: !torch.vtensor<[],f32>, %arg2: !torch.vtensor<[],si8>) -> !torch.vtensor<[6],si8> attributes {torch.onnx_meta.ir_version = 9 : si64} {
+  // CHECK-NOT: torch.operator
+  %0 = torch.operator "onnx.QuantizeLinear"(%arg0, %arg1, %arg2) {torch.onnx_meta.version = 19 : si64} : (!torch.vtensor<[6],f32>, !torch.vtensor<[],f32>, !torch.vtensor<[],si8>) -> !torch.vtensor<[6],si8>
+  return %0 : !torch.vtensor<[6],si8>
+}

test/Conversion/TorchOnnxToTorch/unsupported_simple_ops.mlir

@@ -16,3 +16,21 @@ func.func @test_argmin_no_keepdims_example_select_last_index(%arg0: !torch.vtens
   %0 = torch.operator "onnx.ArgMin"(%arg0) {torch.onnx.axis = 1 : si64, torch.onnx.keepdims = 0 : si64, torch.onnx.select_last_index = 1 : si64} : (!torch.vtensor<[2,2],f32>) -> !torch.vtensor<[2],si64>
   return %0 : !torch.vtensor<[2],si64>
 }
+
+// -----
+
+// Less is supported starting from v13, so although this Less is legal, it will not be accepted.
+
+func.func @test_earlier_version_than_supported(%arg0: !torch.vtensor<[3,4,5],f32>, %arg1: !torch.vtensor<[3,4,5],f32>) -> !torch.vtensor<[3,4,5],i1> attributes {torch.onnx_meta.ir_version = 7 : si64, torch.onnx_meta.opset_version = 13 : si64, torch.onnx_meta.producer_name = "backend-test", torch.onnx_meta.producer_version = ""} {
+  // expected-error @+1 {{failed to legalize operation 'torch.operator'}}
+  %0 = torch.operator "onnx.Less"(%arg0, %arg1) { torch.onnx_meta.version = 7 : si64 } : (!torch.vtensor<[3,4,5],f32>, !torch.vtensor<[3,4,5],f32>) -> !torch.vtensor<[3,4,5],i1>
+  return %0 : !torch.vtensor<[3,4,5],i1>
+}
+
+// -----
+
+func.func @test_no_version(%arg0: !torch.vtensor<[3,4,5],f32>, %arg1: !torch.vtensor<[3,4,5],f32>) -> !torch.vtensor<[3,4,5],i1> attributes {torch.onnx_meta.ir_version = 7 : si64, torch.onnx_meta.producer_name = "backend-test", torch.onnx_meta.producer_version = ""} {
+  // expected-error @+1 {{failed to legalize operation 'torch.operator'}}
+  %0 = torch.operator "onnx.Less"(%arg0, %arg1) : (!torch.vtensor<[3,4,5],f32>, !torch.vtensor<[3,4,5],f32>) -> !torch.vtensor<[3,4,5],i1>
+  return %0 : !torch.vtensor<[3,4,5],i1>
+}