BladeDISC related patches

* fix float width * fix divide_floor & export promoteTypes api (#9) * To comply with the old pytorch versions * Add native_dropout_backward & native_layer_norm_backward decomposition (#15) * add native_dropout and related ops pattern (#1211) * [MHLO] fix dot general contract * Fix batch_norm, div.Tensor_mode and folder (#21) * reimplement linear lowering * reimplement 2-D rhs for mutmul * add torchdynamo
llvm · Feb 2, 2023 · 3528c3a · 3528c3a
1 parent 9536174
commit 3528c3a
Show file tree

Hide file tree

Showing 23 changed files with 782 additions and 152 deletions.
diff --git a/include/torch-mlir/Dialect/Torch/IR/GeneratedTorchOps.td b/include/torch-mlir/Dialect/Torch/IR/GeneratedTorchOps.td
@@ -6596,9 +6596,10 @@ def Torch_AtenOnesLikeOp : Torch_Op<"aten.ones_like", [
 }
 
 def Torch_AtenEmptyMemoryFormatOp : Torch_Op<"aten.empty.memory_format", [
+    Pure,
     AllowsTypeRefinement,
     HasValueSemantics,
-    ReadOnly
+    ReadOnly,
   ]> {
   let summary = "Generated op for `aten::empty.memory_format : (int[], int?, int?, Device?, bool?, int?) -> (Tensor)`";
   let arguments = (ins
@@ -7129,6 +7130,31 @@ def Torch_AtenMaxOp : Torch_Op<"aten.max", [
   }];
 }
 
+def Torch_AtenAmaxOp : Torch_Op<"aten.amax", [
+    AllowsTypeRefinement,
+    HasValueSemantics,
+    ReadOnly
+  ]> {
+  let summary = "Generated op for `aten::amax : (Tensor, int[]?, bool) -> Tensor`";
+  let arguments = (ins
+    AnyTorchTensorType:$self,
+    AnyTorchOptionalListOfTorchIntType:$dim,
+    Torch_BoolType:$keepdim
+  );
+  let results = (outs
+    AnyTorchTensorType:$results
+  );
+  let hasCustomAssemblyFormat = 1;
+  let extraClassDefinition = [{
+    ParseResult AtenAmaxOp::parse(OpAsmParser &parser, OperationState &result) {
+      return parseDefaultTorchOp(parser, result, 3, 1);
+    }
+    void AtenAmaxOp::print(OpAsmPrinter &printer) {
+      printDefaultTorchOp(printer, *this, 3, 1);
+    }
+  }];
+}
+
 def Torch_AtenMaxDimOp : Torch_Op<"aten.max.dim", [
     AllowsTypeRefinement,
     HasValueSemantics,

diff --git a/include/torch-mlir/Dialect/Torch/Utils/TorchUpstream.h b/include/torch-mlir/Dialect/Torch/Utils/TorchUpstream.h
@@ -160,6 +160,7 @@ enum Layout { Strided, Sparse, SparseCsr, Mkldnn, NumOptions };
 //===-----------------------------------------------------------------------===//
 enum EmbeddingBagMode { MODE_SUM, MODE_MEAN, MODE_MAX };
 
+ScalarType promoteTypes(ScalarType a, ScalarType b);
 } // namespace torch_upstream
 } // namespace torch
 } // namespace mlir

diff --git a/include/torch-mlir/Dialect/TorchConversion/IR/TorchConversionOps.td b/include/torch-mlir/Dialect/TorchConversion/IR/TorchConversionOps.td
@@ -42,6 +42,7 @@ def TorchConversion_ToBuiltinTensorOp : TorchConversion_Op<"to_builtin_tensor",
   let assemblyFormat = [{
     $operand attr-dict `:` qualified(type($operand)) `->` qualified(type($result))
   }];
+  let hasCanonicalizer = 1;
   let hasVerifier = 1;
 }
 
@@ -61,6 +62,7 @@ def TorchConversion_FromBuiltinTensorOp : TorchConversion_Op<"from_builtin_tenso
   let assemblyFormat = [{
     $operand attr-dict `:` qualified(type($operand)) `->` qualified(type($result))
   }];
+  let hasCanonicalizer = 1;
   let hasVerifier = 1;
 }
 
@@ -80,6 +82,7 @@ def TorchConversion_ToI1Op : TorchConversion_Op<"to_i1", [
   let assemblyFormat = [{
     $operand attr-dict
   }];
+  let hasFolder = 1;
 }
 
 def TorchConversion_FromI1Op : TorchConversion_Op<"from_i1", [
@@ -98,6 +101,7 @@ def TorchConversion_FromI1Op : TorchConversion_Op<"from_i1", [
   let assemblyFormat = [{
     $operand attr-dict
   }];
+  let hasFolder = 1;
 }
 
 def TorchConversion_ToI64Op : TorchConversion_Op<"to_i64", [

diff --git a/lib/Conversion/TorchToArith/TorchToArith.cpp b/lib/Conversion/TorchToArith/TorchToArith.cpp
@@ -383,13 +383,17 @@ class ConvertTorchToArith : public ConvertTorchToArithBase<ConvertTorchToArith>
     target.addIllegalOp<Torch::ConstantIntOp>();
     patterns.add<ConvertTorchConstantOp<Torch::ConstantIntOp>>(typeConverter,
                                                                context);
-    target.addIllegalOp<AtenAddIntOp, AtenSubIntOp, AtenMulIntOp>();
+    target.addIllegalOp<AtenAddIntOp, AtenSubIntOp, AtenMulIntOp,
+                        AtenRemainderIntOp>();
     patterns.add<ConvertAtenBinaryOp<AtenAddIntOp, arith::AddIOp>>(
         typeConverter, context);
     patterns.add<ConvertAtenBinaryOp<AtenSubIntOp, arith::SubIOp>>(
         typeConverter, context);
     patterns.add<ConvertAtenBinaryOp<AtenMulIntOp, arith::MulIOp>>(
         typeConverter, context);
+    patterns.add<ConvertAtenBinaryOp<AtenRemainderIntOp, arith::RemSIOp>>(
+        typeConverter, context);
+
     target.addIllegalOp<AtenSubFloatOp>();
     patterns.add<ConvertAtenBinaryOp<AtenSubFloatOp, arith::SubFOp>>(
         typeConverter, context);

diff --git a/lib/Conversion/TorchToMhlo/Basic.cpp b/lib/Conversion/TorchToMhlo/Basic.cpp
@@ -29,6 +29,7 @@
 using namespace mlir;
 using namespace mlir::torch;
 using namespace mlir::torch::Torch;
+using namespace mlir::torch::TorchConversion;
 using namespace mlir::torch::torch_to_mhlo;
 
 LogicalResult broadcastRanks(PatternRewriter &rewriter, Operation *op,
@@ -166,16 +167,19 @@ class ConvertAtenUnaryFPOnlyOp : public OpConversionPattern<AtenOpT> {
     if (!selfTy)
       return op.emitError("only Tensor types supported in MHLO");
 
-    if (selfTy.getElementType().isa<mlir::FloatType>()) {
+    auto outTy = OpConversionPattern<AtenOpT>::getTypeConverter()->convertType(
+        op.getType());
+    if (selfTy != outTy) {
+      auto out = rewriter.create<MhloOpT>(op.getLoc(), selfTy, self);
+      rewriter.replaceOpWithNewOp<tensor::CastOp>(op, outTy, out);
+      return success();
+    } else {
       rewriter.replaceOpWithNewOp<MhloOpT>(
           op,
           OpConversionPattern<AtenOpT>::getTypeConverter()->convertType(
               op.getType()),
           self);
       return success();
-    } else {
-      return op.emitError(
-          "only floating-point datatype legalization supported");
     }
   }
 };
@@ -345,15 +349,10 @@ class ConvertAtenMulDivOp : public OpConversionPattern<AtenOpT> {
     } else if (!rhsType) {
       rhs = mhlo::scalarToMhloTensor(rewriter, op, adaptor.getOther(), outElemTy);
     }
-    DenseIntElementsAttr bcastDimensions;
-    lhs = mhlo::promoteType(rewriter, lhs, outType);
-    rhs = mhlo::promoteType(rewriter, rhs, outType);
-    auto loc = op.getLoc();
-    Value result =
-        rewriter.create<ChloOpT>(loc, outType, lhs, rhs, bcastDimensions);
-
     if (!isa<AtenDivTensorModeOp>(op)) {
-      rewriter.replaceOp(op, result);
+      lhs = mhlo::promoteType(rewriter, lhs, outType);
+      rhs = mhlo::promoteType(rewriter, rhs, outType);
+      rewriter.replaceOpWithNewOp<ChloOpT>(op, outType, lhs, rhs, nullptr);
       return success();
     }
 
@@ -365,6 +364,17 @@ class ConvertAtenMulDivOp : public OpConversionPattern<AtenOpT> {
       return rewriter.notifyMatchFailure(
           op, "only support constant str rounding mode");
 
+    auto computeTy = outType;
+    if (outElemTy.isIntOrIndex()) {
+      computeTy =
+          RankedTensorType::get(outType.getShape(), rewriter.getF32Type());
+    }
+    lhs = mhlo::promoteType(rewriter, lhs, computeTy);
+    rhs = mhlo::promoteType(rewriter, rhs, computeTy);
+    auto loc = op.getLoc();
+    auto result =
+        rewriter.create<ChloOpT>(loc, computeTy, lhs, rhs, nullptr).getResult();
+
     if (roundingMode == "trunc") {
       // "trunc" - rounds the results of the division towards zero. Equivalent
       // to C-style integer division.
@@ -378,7 +388,7 @@ class ConvertAtenMulDivOp : public OpConversionPattern<AtenOpT> {
       // floor division in Python (the // operator)
       result = rewriter.create<mhlo::FloorOp>(loc, result).getResult();
     }
-    rewriter.replaceOp(op, result);
+    rewriter.replaceOpWithNewOp<mhlo::ConvertOp>(op, outType, result);
     return success();
   }
 };
@@ -836,7 +846,11 @@ LogicalResult ConvertAtenOp<AtenReluOp>::matchAndRewrite(
       APFloat::getZero(lhsElemTy.cast<mlir::FloatType>().getFloatSemantics(),
                        false),
       lhs);
-  rewriter.replaceOpWithNewOp<mhlo::MaxOp>(op, lhs, zeroTensor);
+  auto outType = getTypeConverter()
+                     ->convertType(op.getType())
+                     .template dyn_cast<TensorType>();
+
+  rewriter.replaceOpWithNewOp<mhlo::MaxOp>(op, outType, lhs, zeroTensor);
   return success();
 }
 
@@ -862,7 +876,11 @@ LogicalResult ConvertAtenOp<AtenGeluOp>::matchAndRewrite(
   auto erf = rewriter.create<mlir::chlo::ErfOp>(loc, erfElement);
   auto erfAdd = rewriter.create<mhlo::AddOp>(loc, erf, one);
   auto halfMul = rewriter.create<mhlo::MulOp>(loc, erfAdd, half);
-  rewriter.replaceOpWithNewOp<mhlo::MulOp>(op, input, halfMul);
+  auto outType = getTypeConverter()
+                     ->convertType(op.getType())
+                     .template dyn_cast<TensorType>();
+
+  rewriter.replaceOpWithNewOp<mhlo::MulOp>(op, outType, input, halfMul);
   return success();
 }
 
@@ -1463,7 +1481,6 @@ void mlir::torch::torch_to_mhlo::populateBasicOpPatternsAndLegality(
   INSERT_ATENOP_PATTERN(ValueTensorLiteralOp);
   INSERT_ATENOP_PATTERN(AtenReciprocalOp);
   INSERT_ATENOP_PATTERN(PrimNumToTensorScalarOp);
-  INSERT_ATENOP_PATTERN(AtenContiguousOp);
 
   INSERT_ATENOP_PATTERN(AtenReluOp);
   INSERT_ATENOP_PATTERN(AtenGeluOp);