[MLIR][TORCH] Add decomposition for prims.var and prims.sqrt op

Signed-Off By: Vivek Khandelwal <vivek@nod-labs.com>

e2e_testing/xfail_sets.py

@@ -83,6 +83,8 @@ TORCHDYNAMO_XFAIL_SET = {
     # error: unsupported by backend contract: tensor with unknown rank
     # note: see current operation: %1 = "torch.tensor_static_info_cast"(%arg0) : (!torch.vtensor<[5,4,3,2,1],f32>) -> !torch.vtensor<*,f32>
     "ElementwisePreluModule_basic",
+    # error: op lowering missing. Issue: https://github.com/llvm/torch-mlir/issues/1792
+    "StdCorrectionKeepDimModule_basic",
 }
 
 MHLO_PASS_SET = {

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

@@ -10881,6 +10881,55 @@ def Torch_PrimsConvertElementTypeOp : Torch_Op<"prims.convert_element_type", [
   }];
 }
 
+def Torch_PrimsVarOp : Torch_Op<"prims.var", [
+    AllowsTypeRefinement,
+    HasValueSemantics,
+    ReadOnly
+  ]> {
+  let summary = "Generated op for `prims::var : (Tensor, int[]?, int, int?) -> (Tensor)`";
+  let arguments = (ins
+    AnyTorchTensorType:$inp,
+    AnyTorchOptionalListOfTorchIntType:$dims,
+    Torch_IntType:$correction,
+    AnyTorchOptionalIntType:$output_dtype
+  );
+  let results = (outs
+    AnyTorchTensorType:$result
+  );
+  let hasCustomAssemblyFormat = 1;
+  let extraClassDefinition = [{
+    ParseResult PrimsVarOp::parse(OpAsmParser &parser, OperationState &result) {
+      return parseDefaultTorchOp(parser, result, 4, 1);
+    }
+    void PrimsVarOp::print(OpAsmPrinter &printer) {
+      printDefaultTorchOp(printer, *this, 4, 1);
+    }
+  }];
+}
+
+def Torch_PrimsSqrtOp : Torch_Op<"prims.sqrt", [
+    AllowsTypeRefinement,
+    HasValueSemantics,
+    ReadOnly
+  ]> {
+  let summary = "Generated op for `prims::sqrt : (Tensor) -> (Tensor)`";
+  let arguments = (ins
+    AnyTorchTensorType:$self
+  );
+  let results = (outs
+    AnyTorchTensorType:$result
+  );
+  let hasCustomAssemblyFormat = 1;
+  let extraClassDefinition = [{
+    ParseResult PrimsSqrtOp::parse(OpAsmParser &parser, OperationState &result) {
+      return parseDefaultTorchOp(parser, result, 1, 1);
+    }
+    void PrimsSqrtOp::print(OpAsmPrinter &printer) {
+      printDefaultTorchOp(printer, *this, 1, 1);
+    }
+  }];
+}
+
 def Torch_QuantizedLinearOp : Torch_Op<"quantized.linear", [
     HasValueSemantics,
     AllowsTypeRefinement,

lib/Dialect/Torch/Transforms/AbstractInterpLibrary.cpp

@@ -5768,6 +5768,10 @@ StringRef mlir::torch::Torch::getAbstractInterpLibrary() {
 "    %0 = call @__torch__.torch.jit._shape_functions.unary(%arg0) : (!torch.list<int>) -> !torch.list<int>\n"
 "    return %0 : !torch.list<int>\n"
 "  }\n"
+"  func.func @\"__torch_mlir_shape_fn.prims.sqrt\"(%arg0: !torch.list<int>) -> !torch.list<int> {\n"
+"    %0 = call @__torch__.torch.jit._shape_functions.unary(%arg0) : (!torch.list<int>) -> !torch.list<int>\n"
+"    return %0 : !torch.list<int>\n"
+"  }\n"
 "  func.func @\"__torch_mlir_shape_fn.aten.neg\"(%arg0: !torch.list<int>) -> !torch.list<int> {\n"
 "    %0 = call @__torch__.torch.jit._shape_functions.unary(%arg0) : (!torch.list<int>) -> !torch.list<int>\n"
 "    return %0 : !torch.list<int>\n"
@@ -6032,6 +6036,13 @@ StringRef mlir::torch::Torch::getAbstractInterpLibrary() {
 "    %0 = torch.prim.ListConstruct  : () -> !torch.list<int>\n"
 "    return %0 : !torch.list<int>\n"
 "  }\n"
+"  func.func @\"__torch_mlir_shape_fn.prims.var\"(%arg0: !torch.list<int>, %arg1: !torch.optional<list<int>>, %arg2: !torch.int, %arg3: !torch.optional<int>) -> !torch.list<int> {\n"
+"    %none = torch.constant.none\n"
+"    %false = torch.constant.bool false\n"
+"    %0 = torch.derefine %none : !torch.none to !torch.any\n"
+"    %1 = call @__torch__.torch.jit._shape_functions.sum_mean_dim(%arg0, %arg1, %false, %0) : (!torch.list<int>, !torch.optional<list<int>>, !torch.bool, !torch.any) -> !torch.list<int>\n"
+"    return %1 : !torch.list<int>\n"
+"  }\n"
 "  func.func @\"__torch_mlir_shape_fn.aten.var.dim\"(%arg0: !torch.list<int>, %arg1: !torch.optional<list<int>>, %arg2: !torch.bool, %arg3: !torch.bool) -> !torch.list<int> {\n"
 "    %none = torch.constant.none\n"
 "    %0 = torch.derefine %none : !torch.none to !torch.any\n"

lib/Dialect/Torch/Transforms/DecomposeComplexOps.cpp

@@ -3425,6 +3425,38 @@ public:
 };
 } // namespace
 
+namespace {
+// Decompose `prims.var` op into `aten.var.correction` op.
+class DecomposePrimsVarOp : public OpRewritePattern<PrimsVarOp> {
+public:
+  using OpRewritePattern::OpRewritePattern;
+  LogicalResult matchAndRewrite(PrimsVarOp op,
+                                PatternRewriter &rewriter) const override {
+    if (!op.getOutputDtype().getType().isa<Torch::NoneType>())
+      return rewriter.notifyMatchFailure(
+          op, "Unimplemented non-None dtype for prims::var op");
+    Value cstFalse = rewriter.create<Torch::ConstantBoolOp>(op.getLoc(), false);
+    rewriter.replaceOpWithNewOp<AtenVarCorrectionOp>(
+        op, op.getType(), op.getInp(), op.getDims(), op.getCorrection(),
+        /*keepdim=*/cstFalse);
+    return success();
+  }
+};
+} // namespace
+
+namespace {
+// Decompose `prims.sqrt` op into `aten.sqrt` op.
+class DecomposePrimsSqrtOp : public OpRewritePattern<PrimsSqrtOp> {
+public:
+  using OpRewritePattern::OpRewritePattern;
+  LogicalResult matchAndRewrite(PrimsSqrtOp op,
+                                PatternRewriter &rewriter) const override {
+    rewriter.replaceOpWithNewOp<AtenSqrtOp>(op, op.getType(), op.getSelf());
+    return success();
+  }
+};
+} // namespace
+
 namespace {
 // The op is decomposed using the Box-Muller transform.
 // Refer: https://en.wikipedia.org/wiki/Box-Muller_transform
@@ -3659,6 +3691,8 @@ public:
     addPatternIfTargetOpIsIllegal<DecomposeAtenRandintLowOp>(patterns);
     addPatternIfTargetOpIsIllegal<DecomposeAtenVarMeanCorrectionOp>(patterns);
     addPatternIfTargetOpIsIllegal<DecomposePrimsConvertElementTypeOp>(patterns);
+    addPatternIfTargetOpIsIllegal<DecomposePrimsVarOp>(patterns);
+    addPatternIfTargetOpIsIllegal<DecomposePrimsSqrtOp>(patterns);
     addPatternIfTargetOpIsIllegal<DecomposeAtenRandnOp>(patterns);
     addPatternIfTargetOpIsIllegal<DecomposeAtenRandnGeneratorOp>(patterns);
     addPatternIfTargetOpIsIllegal<DecomposeAtenVarMeanOp>(patterns);

lib/Dialect/Torch/Transforms/LowerToBackendContract.cpp

@@ -438,6 +438,8 @@ static void markDecomposedOpsAsIllegal(MLIRContext *context,
   target.addIllegalOp<AtenRandintLowOp>();
   target.addIllegalOp<AtenVarMeanCorrectionOp>();
   target.addIllegalOp<PrimsConvertElementTypeOp>();
+  target.addIllegalOp<PrimsVarOp>();
+  target.addIllegalOp<PrimsSqrtOp>();
   target.addIllegalOp<AtenRandnOp>();
   target.addIllegalOp<AtenRandnGeneratorOp>();
   target.addIllegalOp<AtenVarMeanOp>();

lib/Dialect/Torch/Transforms/RefineTypes.cpp

@@ -675,7 +675,8 @@ void TypeAnalysis::visitOperation(Operation *op,
   // Dtype is always float32, except for bfloat16, float16, float64 and nullptr.
   if (isa<AtenTanhOp, AtenExpOp, AtenSinOp, AtenCosOp, AtenSigmoidOp,
           AtenReciprocalOp, AtenLogOp, AtenSqrtOp, AtenLog2Op, AtenLog1pOp,
-          AtenRsqrtOp, AtenErfOp, AtenSoftplusOp, AtenFrobeniusNormDimOp>(op)) {
+          AtenRsqrtOp, AtenErfOp, AtenSoftplusOp, AtenFrobeniusNormDimOp,
+          PrimsSqrtOp>(op)) {
     ValueKnowledge knowledge =
         ValueKnowledge::getTensorPessimisticValueState(op->getContext());
     Type dtype = operands[0]->getValue().dtype;
@@ -978,7 +979,7 @@ void TypeAnalysis::visitOperation(Operation *op,
     visitReductionAlongAllDimsOp(op, dtype, operands);
     return;
   } else if (isa<AtenStdOp, AtenStdDimOp, AtenStdCorrectionOp, AtenVarOp,
-                 AtenVarDimOp, AtenVarCorrectionOp>(op)) {
+                 AtenVarDimOp, AtenVarCorrectionOp, PrimsVarOp>(op)) {
     auto input = operands[0]->getValue();
     visitReductionAlongAllDimsOp(op, input.dtype, operands);
     return;

python/torch_mlir/dialects/torch/importer/jit_ir/build_tools/abstract_interp_lib_gen.py

@@ -107,6 +107,9 @@ def aten〇hardtanh〡shape(self: List[int], min_val: float = -1, max_val: float
 def aten〇sqrt〡shape(self: List[int]) -> List[int]:
     return upstream_shape_functions.unary(self)
 
+def prims〇sqrt〡shape(self: List[int]) -> List[int]:
+    return upstream_shape_functions.unary(self)
+
 def aten〇neg〡shape(self: List[int]) -> List[int]:
     return upstream_shape_functions.unary(self)
 
@@ -307,6 +310,9 @@ def aten〇mean〡shape(self: List[int], dtype: Optional[int] = None) -> List[in
 def aten〇var〡shape(self: List[int], unbiased: bool = True) -> List[int]:
     return []
 
+def prims〇var〡shape(inp: List[int], dims: Optional[List[int]], correction: int, output_dtype: Optional[int] = None) -> List[int]:
+    return upstream_shape_functions.sum_mean_dim(inp, dims, False, None)
+
 def aten〇var〇dim〡shape(self: List[int], dim: Optional[List[int]], unbiased: bool = True, keepdim: bool = False) -> List[int]:
     return upstream_shape_functions.sum_mean_dim(self, dim, keepdim, None)
 

python/torch_mlir/dialects/torch/importer/jit_ir/build_tools/torch_ods_gen.py

@@ -674,6 +674,8 @@ def emit_ops(emitter_td: TextEmitter, registry: Registry):
     # ==========================================================================
 
     emit("prims::convert_element_type : (Tensor, int) -> (Tensor)")
+    emit("prims::var : (Tensor, int[]?, int, int?) -> (Tensor)")
+    emit("prims::sqrt : (Tensor) -> (Tensor)")
 
     # ==========================================================================
     # `quantized::` namespace.