[TOSA] Add lowering for aten.expm1 (#3949)

* Add Torch to TOSA legalization for aten.expm1
* Update xfail_sets with new test results
* Add new LIT tests


Change-Id: I834d0c7416341f884612053aebf9fcc90bcb3b53

Signed-off-by: Justin Ngo <[email protected]>

Author: justin-ngo-arm

lib/Conversion/TorchToTosa/TorchToTosa.cpp

@@ -8212,6 +8212,47 @@ LogicalResult ConvertAtenOp<AtenLog10Op>::matchAndRewrite(
   return success();
 }
 
+// Legalization for aten.expm1
+template <>
+LogicalResult ConvertAtenOp<AtenExpm1Op>::matchAndRewrite(
+    AtenExpm1Op op, OpAdaptor adaptor,
+    ConversionPatternRewriter &rewriter) const {
+  // expm1 formula:
+  // yi = exp(x) - 1
+  // Note: This lowering might not provide as great precision as aten.expm1
+  // since TOSA doesn't have a built-in expm1 op.
+  auto self = adaptor.getSelf();
+
+  auto selfType = dyn_cast<TensorType>(self.getType());
+  if (!selfType)
+    return rewriter.notifyMatchFailure(op, "Only tensor types are supported");
+
+  auto resultType =
+      dyn_cast<TensorType>(typeConverter->convertType(op.getType()));
+  auto resultElemTy = resultType.getElementType();
+
+  if (!isa<mlir::FloatType>(resultElemTy))
+    return rewriter.notifyMatchFailure(
+        op, "Only floating-point datatype result types are supported");
+
+  // If input is not a float type then cast it to result element type
+  auto selfElemTy = selfType.getElementType();
+  if (!isa<mlir::FloatType>(selfElemTy))
+    self = tosa::promoteType(rewriter, self, resultType);
+
+  auto one =
+      tosa::getConstTensor<float>(rewriter, op, 1.0f, {}, resultElemTy).value();
+
+  auto expOp = rewriter.create<tosa::ExpOp>(op->getLoc(), resultType, self);
+
+  auto result = rewriter.create<tosa::SubOp>(op->getLoc(), resultType,
+                                             expOp.getResult(), one);
+
+  rewriter.replaceOp(op, {result.getResult()});
+
+  return success();
+}
+
 // Legalization for aten.tan
 template <>
 LogicalResult ConvertAtenOp<AtenTanOp>::matchAndRewrite(
@@ -8805,6 +8846,7 @@ std::set<StringRef> torch::populateTorchToTosaConversionPatternsAndIllegalOps(
   INSERT_ATENOP_PATTERN(AtenLogitOp);
   INSERT_ATENOP_PATTERN(AtenLog1pOp);
   INSERT_ATENOP_PATTERN(AtenLog10Op);
+  INSERT_ATENOP_PATTERN(AtenExpm1Op);
   INSERT_ATENOP_PATTERN(AtenTanOp);
   INSERT_ATENOP_PATTERN(AtenUnfoldOp);
 #undef INSERT_ATENOP_PATTERN

projects/pt1/e2e_testing/xfail_sets.py

@@ -1709,8 +1709,12 @@
     "Unfold_Module_Rank_Zero_basic",
     "Unfold_Module_basic",
     "ElementwiseErfIntModule_basic",
+    "ElementwiseExpm1IntModule_basic",
+    "ElementwiseExpm1Module_basic",
     "ElementwiseIntTensorLtFloatScalarModule_basic",
     "ElementwiseSigmoidIntModule_basic",
+    "ElementwiseSpecialExpm1IntModule_basic",
+    "ElementwiseSpecialExpm1Module_basic",
     "ElementwiseTanIntModule_basic",
     "ElementwiseTanModule_basic",
     "ElementwiseUnaryIntModule_basic",
@@ -3668,16 +3672,12 @@
     "ElementwiseCoshModule_basic",
     "ElementwiseDequantizePerChannelModule_basic",
     "ElementwiseDequantizePerTensorModule_basic",
-    "ElementwiseExpm1IntModule_basic",
-    "ElementwiseExpm1Module_basic",
     "ElementwiseMulTensorComplexDiffModule_basic",
     "ElementwiseMulTensorComplexModule_basic",
     "ElementwiseQuantizePerTensorModule_basic",
     "ElementwiseQuantizePerTensorUIntModule_basic",
     "ElementwiseSinhIntModule_basic",
     "ElementwiseSinhModule_basic",
-    "ElementwiseSpecialExpm1IntModule_basic",
-    "ElementwiseSpecialExpm1Module_basic",
     "ElementwiseToDtypeF32ToI64Module_basic",
     "ElementwiseToDtypeI64ToUI8Module_basic",
     "ElementwiseWhereScalarOtherStaticModule_basic",

test/Conversion/TorchToTosa/basic.mlir

@@ -3024,3 +3024,36 @@ func.func @torch.aten.unfold$rank_zero(%arg0: !torch.vtensor<[],f32>) -> !torch.
 }
 
 // -----
+
+// CHECK-LABEL:   func.func @torch.aten.expm1$basic(
+// CHECK-SAME:                                      %[[VAL_0:.*]]: !torch.vtensor<[3,4],f32>) -> !torch.vtensor<[3,4],f32> {
+// CHECK:           %[[VAL_1:.*]] = torch_c.to_builtin_tensor %[[VAL_0]] : !torch.vtensor<[3,4],f32> -> tensor<3x4xf32>
+// CHECK:           %[[VAL_2:.*]] = "tosa.const"() <{value = dense<1.000000e+00> : tensor<f32>}> : () -> tensor<f32>
+// CHECK:           %[[VAL_3:.*]] = tosa.exp %[[VAL_1]] : (tensor<3x4xf32>) -> tensor<3x4xf32>
+// CHECK:           %[[VAL_4:.*]] = tosa.sub %[[VAL_3]], %[[VAL_2]] : (tensor<3x4xf32>, tensor<f32>) -> tensor<3x4xf32>
+// CHECK:           %[[VAL_5:.*]] = torch_c.from_builtin_tensor %[[VAL_4]] : tensor<3x4xf32> -> !torch.vtensor<[3,4],f32>
+// CHECK:           return %[[VAL_5]] : !torch.vtensor<[3,4],f32>
+// CHECK:         }
+func.func @torch.aten.expm1$basic(%arg0: !torch.vtensor<[3,4],f32>) -> !torch.vtensor<[3,4],f32> {
+  %0 = torch.aten.expm1 %arg0 : !torch.vtensor<[3,4],f32> -> !torch.vtensor<[3,4],f32>
+  return %0 : !torch.vtensor<[3,4],f32>
+}
+
+// -----
+
+// CHECK-LABEL:   func.func @torch.aten.expm1$int(
+// CHECK-SAME:                                    %[[VAL_0:.*]]: !torch.vtensor<[3,4],si32>) -> !torch.vtensor<[3,4],f32> {
+// CHECK:           %[[VAL_1:.*]] = torch_c.to_builtin_tensor %[[VAL_0]] : !torch.vtensor<[3,4],si32> -> tensor<3x4xi32>
+// CHECK:           %[[VAL_2:.*]] = tosa.cast %[[VAL_1]] : (tensor<3x4xi32>) -> tensor<3x4xf32>
+// CHECK:           %[[VAL_3:.*]] = "tosa.const"() <{value = dense<1.000000e+00> : tensor<f32>}> : () -> tensor<f32>
+// CHECK:           %[[VAL_4:.*]] = tosa.exp %[[VAL_2]] : (tensor<3x4xf32>) -> tensor<3x4xf32>
+// CHECK:           %[[VAL_5:.*]] = tosa.sub %[[VAL_4]], %[[VAL_3]] : (tensor<3x4xf32>, tensor<f32>) -> tensor<3x4xf32>
+// CHECK:           %[[VAL_6:.*]] = torch_c.from_builtin_tensor %[[VAL_5]] : tensor<3x4xf32> -> !torch.vtensor<[3,4],f32>
+// CHECK:           return %[[VAL_6]] : !torch.vtensor<[3,4],f32>
+// CHECK:         }
+func.func @torch.aten.expm1$int(%arg0: !torch.vtensor<[3,4],si32>) -> !torch.vtensor<[3,4],f32> {
+  %0 = torch.aten.expm1 %arg0 : !torch.vtensor<[3,4],si32> -> !torch.vtensor<[3,4],f32>
+  return %0 : !torch.vtensor<[3,4],f32>
+}
+
+// -----