Main Compliance testing support for CAST

Limit CAST input tensor to maximums of output type to avoid
saturation and infinity.

Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I33350a4ce0ec828da7d2e7aa8cd3183a89a97431

4 files changed, 68 insertions, 12 deletions

diff --git a/reference_model/src/generate/generate_utils.cc b/reference_model/src/generate/generate_utils.cc
index 58a3d33..d1758fc 100644
--- a/reference_model/src/generate/generate_utils.cc
+++ b/reference_model/src/generate/generate_utils.cc
@@ -42,6 +42,7 @@ NLOHMANN_JSON_SERIALIZE_ENUM(Op,
                                  { Op::Op_ADD, "ADD" },
                                  { Op::Op_ARGMAX, "ARGMAX" },
                                  { Op::Op_AVG_POOL2D, "AVG_POOL2D" },
+                                 { Op::Op_CAST, "CAST" },
                                  { Op::Op_CEIL, "CEIL" },
                                  { Op::Op_CLAMP, "CLAMP" },
                                  { Op::Op_CONV2D, "CONV2D" },
diff --git a/verif/conformance/tosa_main_profile_ops_info.json b/verif/conformance/tosa_main_profile_ops_info.json
index 0b20e4f..960ad27 100644
--- a/verif/conformance/tosa_main_profile_ops_info.json
+++ b/verif/conformance/tosa_main_profile_ops_info.json
@@ -253,6 +253,7 @@
         "profile": [
             "tosa-mi"
         ],
+        "support_for": [ "lazy_data_gen" ],
         "generation": {
             "standard": {
                 "negative_dim_range": "1,10",
@@ -271,7 +272,7 @@
                         "--target-dtype",
                         "int32",
                         "--fp-values-range",
-                        "-2.0,2.0",
+                        "-max,max",
                         "--tensor-dim-range",
                         "16,64",
                         "--target-rank",
@@ -295,7 +296,7 @@
                         "--target-dtype",
                         "int32",
                         "--fp-values-range",
-                        "-2.0,2.0",
+                        "-max,max",
                         "--tensor-dim-range",
                         "1,16",
                         "--target-rank",
@@ -306,6 +307,8 @@
                     [
                         "--target-dtype",
                         "fp16",
+                        "--fp-values-range",
+                        "-max,max",
                         "--target-shape",
                         "1,1,1,65533,1",
                         "--target-shape",
diff --git a/verif/generator/tosa_arg_gen.py b/verif/generator/tosa_arg_gen.py
index 3057963..c557207 100644
--- a/verif/generator/tosa_arg_gen.py
+++ b/verif/generator/tosa_arg_gen.py
@@ -1434,6 +1434,27 @@ class TosaTensorValuesGen:
             testGen, opName, dtypeList, shapeList, argsDict, error_name
         )
 
+    @staticmethod
+    def tvgCast(testGen, opName, dtypeList, shapeList, argsDict, error_name=None):
+        in_dtype = dtypeList[0]
+        out_dtype = argsDict["out_type"]
+        # Create look up to limit input tensor to output type maximums to avoid
+        # FP infinities and saturation of integers
+        out_range = testGen.getDTypeRange(out_dtype, high_inclusive=True)
+        highval_lookup = {in_dtype: out_range[1]}
+        data_range = TosaTensorValuesGen._get_data_range(
+            testGen,
+            in_dtype,
+            highval_lookup,
+        )
+
+        assert data_range is not None
+        argsDict["data_range"] = data_range
+
+        return TosaTensorValuesGen.tvgLazyGenDefault(
+            testGen, opName, dtypeList, shapeList, argsDict, error_name
+        )
+
 
 class TosaArgGen:
     """Argument generators create exhaustive or random lists of attributes for
@@ -2350,7 +2371,18 @@ class TosaArgGen:
             raise Exception("Unexpected input dtype: {}".format(inDtype))
 
         for dtype in dtypeList:
-            arg_list.append(("out{}".format(testGen.typeStr(dtype)), [dtype]))
+            arg_list.append(
+                ("out{}".format(testGen.typeStr(dtype)), {"out_type": dtype})
+            )
+
+        # Now add data generator types
+        arg_list = TosaArgGen._add_data_generators(
+            testGen,
+            opName,
+            dtype,
+            arg_list,
+            error_name,
+        )
 
         return arg_list
 
diff --git a/verif/generator/tosa_test_gen.py b/verif/generator/tosa_test_gen.py
index 63958a9..1602109 100644
--- a/verif/generator/tosa_test_gen.py
+++ b/verif/generator/tosa_test_gen.py
@@ -307,10 +307,15 @@ class TosaTestGen:
     def tensorComplianceMetaData(
         self, op, inputType, argsDict, outputTensor, errorName
     ):
+        # TODO - Dot product Ops with FP16 or BF16 inputs that produce FP32 outputs are not supported yet
+        UNSUPPORTED_NON_FP32_INPUT_OPS = (Op.MATMUL, Op.CONV2D, Op.FULLY_CONNECTED)
         if (
             errorName
             or not gtu.dtypeIsSupportedByCompliance(outputTensor.dtype)
-            or not gtu.dtypeIsSupportedByCompliance(inputType)
+            or (
+                not gtu.dtypeIsSupportedByCompliance(inputType)
+                and op["op"] in UNSUPPORTED_NON_FP32_INPUT_OPS
+            )
         ):
             # No compliance for error tests or unsupported types currently
             return None
@@ -1874,14 +1879,20 @@ class TosaTestGen:
         return val
 
     # Type Conversion
-    def build_cast(self, op, val, out_dtype, validator_fcns=None, error_name=None):
-        result_tens = OutputShaper.typeConversionOp(
+    def build_cast(
+        self, op, inputs, args_dict, validator_fcns=None, error_name=None, qinfo=None
+    ):
+        assert len(inputs) == 1
+        val = inputs[0]
+        out_dtype = args_dict["out_type"]
+
+        result_tensor = OutputShaper.typeConversionOp(
             self.ser, self.rng, val, out_dtype, error_name
         )
 
         # Invalidate Input/Output list for error if checks.
         input_list = [val.name]
-        output_list = [result_tens.name]
+        output_list = [result_tensor.name]
         pCount, cCount = op["operands"]
         num_operands = pCount + cCount
         input_list, output_list = TosaErrorIfArgGen.eiInvalidateInputOutputList(
@@ -1894,10 +1905,10 @@ class TosaTestGen:
             error_name,
             op=op,
             input_shape=val.shape,
-            output_shape=result_tens.shape,
+            output_shape=result_tensor.shape,
             input_dtype=val.dtype,
-            output_dtype=result_tens.dtype,
-            result_tensors=[result_tens],
+            output_dtype=result_tensor.dtype,
+            result_tensors=[result_tensor],
             input_list=input_list,
             output_list=output_list,
             num_operands=num_operands,
@@ -1905,7 +1916,12 @@ class TosaTestGen:
             return None
 
         self.ser.addOperator(op["op"], input_list, output_list)
-        return result_tens
+
+        compliance = self.tensorComplianceMetaData(
+            op, val.dtype, args_dict, result_tensor, error_name
+        )
+
+        return TosaTestGen.BuildInfo(result_tensor, compliance)
 
     def build_rescale(
         self,
@@ -4365,7 +4381,7 @@ class TosaTestGen:
             "build_fcn": (
                 build_cast,
                 TosaTensorGen.tgBasic,
-                TosaTensorValuesGen.tvgDefault,
+                TosaTensorValuesGen.tvgCast,
                 TosaArgGen.agCast,
             ),
             "types": (
@@ -4383,6 +4399,10 @@ class TosaTestGen:
                 TosaErrorValidator.evWrongInputList,
                 TosaErrorValidator.evWrongOutputList,
             ),
+            "data_gen": {
+                "fp": (gtu.DataGenType.PSEUDO_RANDOM,),
+            },
+            "compliance": {"ulp": 0.5},
         },
         "rescale": {
             "op": Op.RESCALE,