Main Compliance testing for SCATTER and GATHER

Added indices shuffling and random INT32 support to generate lib
with testing of these new random generator modes

Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: I058d8b092470228075e8fe69c2ededa639163003

6 files changed, 218 insertions, 76 deletions

diff --git a/verif/conformance/tosa_base_profile_ops_info.json b/verif/conformance/tosa_base_profile_ops_info.json
index 3a8622b..b186b06 100644
--- a/verif/conformance/tosa_base_profile_ops_info.json
+++ b/verif/conformance/tosa_base_profile_ops_info.json
@@ -1502,11 +1502,15 @@
                         "--target-dtype",
                         "int32",
                         "--tensor-dim-range",
-                        "4,64"
+                        "4,64",
+                        "--max-batch-size",
+                        "64"
                     ],
                     [
                         "--target-dtype",
                         "int8",
+                        "--tensor-dim-range",
+                        "5,20",
                         "--target-shape",
                         "2,65533,1",
                         "--target-shape",
@@ -3229,7 +3233,9 @@
                         "--target-dtype",
                         "int32",
                         "--tensor-dim-range",
-                        "4,64"
+                        "4,64",
+                        "--max-batch-size",
+                        "64"
                     ],
                     [
                         "--target-dtype",
diff --git a/verif/conformance/tosa_main_profile_ops_info.json b/verif/conformance/tosa_main_profile_ops_info.json
index c3bd6ee..fb25622 100644
--- a/verif/conformance/tosa_main_profile_ops_info.json
+++ b/verif/conformance/tosa_main_profile_ops_info.json
@@ -1133,6 +1133,7 @@
         "profile": [
             "tosa-mi"
         ],
+        "support_for": [ "lazy_data_gen" ],
         "generation": {
             "standard": {
                 "negative_dim_range": "1,10",
@@ -1145,15 +1146,19 @@
                         "--target-dtype",
                         "bf16",
                         "--fp-values-range",
-                        "-2.0,2.0",
+                        "-max,max",
                         "--tensor-dim-range",
-                        "5,72"
+                        "5,72",
+                        "--max-batch-size",
+                        "72"
                     ],
                     [
                         "--target-dtype",
                         "fp32",
                         "--fp-values-range",
-                        "-2.0,2.0",
+                        "-max,max",
+                        "--tensor-dim-range",
+                        "5,20",
                         "--target-shape",
                         "2,65536,1",
                         "--target-shape",
@@ -1992,6 +1997,7 @@
         "profile": [
             "tosa-mi"
         ],
+        "support_for": [ "lazy_data_gen" ],
         "generation": {
             "standard": {
                 "negative_dim_range": "1,10",
@@ -2004,15 +2010,17 @@
                         "--target-dtype",
                         "bf16",
                         "--fp-values-range",
-                        "-2.0,2.0",
+                        "-max,max",
                         "--tensor-dim-range",
-                        "5,56"
+                        "5,56",
+                        "--max-batch-size",
+                        "56"
                     ],
                     [
                         "--target-dtype",
                         "fp32",
                         "--fp-values-range",
-                        "-2.0,2.0",
+                        "-max,max",
                         "--target-shape",
                         "1,65541,1",
                         "--target-shape",
diff --git a/verif/generator/datagenerator.py b/verif/generator/datagenerator.py
index 0dd60e5..b5ef35d 100644
--- a/verif/generator/datagenerator.py
+++ b/verif/generator/datagenerator.py
@@ -78,6 +78,10 @@ class GenerateLibrary:
             size_bytes = size * 2
             # Create buffer of bytes and initialize to zero
             buffer = (ct.c_ubyte * size_bytes)(0)
+        elif dtype == "INT32":
+            # Create buffer and initialize to zero
+            buffer = (ct.c_int32 * size)(0)
+            size_bytes = size * 4
         else:
             raise GenerateError(f"Unsupported data type {dtype}")
 
diff --git a/verif/generator/tosa_arg_gen.py b/verif/generator/tosa_arg_gen.py
index 35253e0..50811ac 100644
--- a/verif/generator/tosa_arg_gen.py
+++ b/verif/generator/tosa_arg_gen.py
@@ -204,7 +204,7 @@ class TosaTensorGen:
         return shape_list
 
     @staticmethod
-    def tgScatter(testGen, opName, rank, error_name=None):
+    def tgGather(testGen, opName, rank, error_name=None):
         pl, const = opName["operands"]
 
         assert pl == 2
@@ -212,12 +212,31 @@ class TosaTensorGen:
         if error_name != ErrorIf.WrongRank:
             assert rank == 3
 
+        values_shape = testGen.makeShape(rank)
+        values_shape = testGen.constrictBatchSize(values_shape)
+
+        N = values_shape[0]
+        W = testGen.makeDimension()
+        indices_shape = [N, W]
+
+        shape_list = [values_shape, indices_shape]
+        return shape_list
+
+    @staticmethod
+    def tgScatter(testGen, opName, rank, error_name=None):
+        pl, const = opName["operands"]
+
+        assert pl == 3
+        assert const == 0
+        if error_name != ErrorIf.WrongRank:
+            assert rank == 3
+
         values_in_shape = testGen.makeShape(rank)
-        K = values_in_shape[1]
+        values_in_shape = testGen.constrictBatchSize(values_in_shape)
 
-        # ignore max batch size if target shape is set
-        if testGen.args.max_batch_size and not testGen.args.target_shapes:
-            values_in_shape[0] = min(values_in_shape[0], testGen.args.max_batch_size)
+        N = values_in_shape[0]
+        K = values_in_shape[1]
+        C = values_in_shape[2]
 
         # Make sure W is not greater than K, as we can only write each output index
         # once (having a W greater than K means that you have to repeat a K index)
@@ -225,11 +244,12 @@ class TosaTensorGen:
         W_max = min(testGen.args.tensor_shape_range[1], K)
         W = testGen.randInt(W_min, W_max) if W_min < W_max else W_min
 
-        input_shape = [values_in_shape[0], W, values_in_shape[2]]
+        input_shape = [N, W, C]
 
         shape_list = []
-        shape_list.append(values_in_shape.copy())
-        shape_list.append(input_shape.copy())
+        shape_list.append(values_in_shape)
+        shape_list.append([N, W])  # indices
+        shape_list.append(input_shape)
 
         return shape_list
 
@@ -695,6 +715,13 @@ class TosaTensorValuesGen:
                         "round" in argsDict["data_range_list"][idx]
                         and argsDict["data_range_list"][idx]["round"] is True
                     )
+                if data_range is not None and dtype not in (
+                    DType.FP16,
+                    DType.FP32,
+                    DType.BF16,
+                ):
+                    # Change from inclusive to exclusive range
+                    data_range = (data_range[0], data_range[1] + 1)
                 # Ignore lazy data gen option and create data array using any range limits
                 arr = testGen.getRandTensor(shape, dtype, data_range)
                 if roundMode:
@@ -732,13 +759,15 @@ class TosaTensorValuesGen:
                 # TODO - generate seed for this generator based on test
                 info["rng_seed"] = 42
 
+                data_range = None
                 if "data_range_list" in argsDict:
                     data_range = argsDict["data_range_list"][idx]["range"]
                     if "round" in argsDict["data_range_list"][idx]:
                         info["round"] = argsDict["data_range_list"][idx]["round"]
                 elif "data_range" in argsDict:
                     data_range = argsDict["data_range"]
-                else:
+
+                if data_range is None:
                     data_range = testGen.getDTypeRange(
                         dtypeList[idx], high_inclusive=True
                     )
@@ -1455,6 +1484,109 @@ class TosaTensorValuesGen:
             testGen, opName, dtypeList, shapeList, argsDict, error_name
         )
 
+    @staticmethod
+    def tvgGather(testGen, opName, dtypeList, shapeList, argsDict, error_name=None):
+        K = shapeList[0][1]
+
+        # Fix the type of the indices tensor
+        dtypeList[1] = DType.INT32
+
+        dtype = dtypeList[0]
+        if not gtu.dtypeIsSupportedByCompliance(dtype):
+            # Test unsupported by data generator
+            op = testGen.TOSA_OP_LIST[opName]
+            pCount, cCount = op["operands"]
+            assert (
+                pCount == 2 and cCount == 0
+            ), "Op.GATHER must have 2 placeholders, 0 consts"
+
+            tens_ser_list = []
+            for idx, shape in enumerate(shapeList):
+                dtype = dtypeList[idx]
+                if idx != 1:
+                    arr = testGen.getRandTensor(shape, dtype)
+                    tens_ser_list.append(testGen.ser.addPlaceholder(shape, dtype, arr))
+                else:
+                    # Limit data range of indices tensor upto K (exclusive)
+                    arr = testGen.getRandTensor(shape, dtype, (0, K))
+                    # To match old functionality - create indices as CONST
+                    tens_ser_list.append(testGen.ser.addConst(shape, dtype, arr))
+
+            return TosaTensorValuesGen.TVGInfo(tens_ser_list, None)
+
+        else:
+            # ERROR_IF or floating point test
+            # Use inclusive values upto index K for indices tensor
+            data_range_list = (
+                {"range": None},
+                {"range": (0, K - 1)},
+            )
+            argsDict["data_range_list"] = data_range_list
+
+            return TosaTensorValuesGen.tvgLazyGenDefault(
+                testGen, opName, dtypeList, shapeList, argsDict, error_name
+            )
+
+    @staticmethod
+    def tvgScatter(testGen, opName, dtypeList, shapeList, argsDict, error_name=None):
+        K = shapeList[0][1]
+        W = shapeList[2][1]
+
+        # Work out an indices tensor here with data that doesn't exceed the
+        # dimension K of the values_in tensor and does NOT repeat the same K
+        # location as needed by the spec:
+        # "It is not permitted to repeat the same output index within a single
+        # SCATTER operation and so each output index occurs at most once."
+        assert K >= W, "Op.SCATTER W must be smaller or equal to K"
+
+        # Fix the type of the indices tensor
+        dtypeList[1] = DType.INT32
+
+        dtype = dtypeList[0]
+        if not gtu.dtypeIsSupportedByCompliance(dtype):
+            # Test unsupported by data generator
+            op = testGen.TOSA_OP_LIST[opName]
+            pCount, cCount = op["operands"]
+            assert (
+                pCount == 3 and cCount == 0
+            ), "Op.SCATTER must have 3 placeholders, 0 consts"
+
+            tens_ser_list = []
+            for idx, shape in enumerate(shapeList):
+                dtype = dtypeList[idx]
+                if idx != 1:
+                    arr = testGen.getRandTensor(shape, dtype)
+                    tens_ser_list.append(testGen.ser.addPlaceholder(shape, dtype, arr))
+                else:
+                    # Create the indices array
+                    assert dtype == DType.INT32, "Op.SCATTER unexpected indices type"
+                    arr = []
+                    for n in range(shape[0]):
+                        # Get a shuffled list of output indices (0 to K-1) and
+                        # limit length to W
+                        arr.append(testGen.rng.permutation(K)[:W])
+                    indices_arr = np.array(arr, dtype=np.int32)  # (N, W)
+                    # To match old functionality - create indices as CONST
+                    tens_ser_list.append(
+                        testGen.ser.addConst(shape, dtype, indices_arr)
+                    )
+
+            return TosaTensorValuesGen.TVGInfo(tens_ser_list, None)
+
+        else:
+            # ERROR_IF or floating point test
+            # Use inclusive values upto index K for indices tensor
+            data_range_list = (
+                {"range": None},
+                {"range": (0, K - 1)},
+                {"range": None},
+            )
+            argsDict["data_range_list"] = data_range_list
+
+            return TosaTensorValuesGen.tvgLazyGenDefault(
+                testGen, opName, dtypeList, shapeList, argsDict, error_name
+            )
+
 
 class TosaArgGen:
     """Argument generators create exhaustive or random lists of attributes for
diff --git a/verif/generator/tosa_error_if.py b/verif/generator/tosa_error_if.py
index 86be347..5dd785f 100644
--- a/verif/generator/tosa_error_if.py
+++ b/verif/generator/tosa_error_if.py
@@ -666,12 +666,8 @@ class TosaErrorValidator:
         error_reason = "Op input list does not match expected input"
 
         if check:
-            op = kwargs["op"]
             input_list = kwargs["input_list"]
             num_operands = kwargs["num_operands"]
-            if op["op"] in [Op.SCATTER, Op.GATHER]:
-                # SCATTER/GATHER add an indices input tensor in their build functions
-                num_operands += 1
             if len(input_list) != num_operands:
                 error_result = True
 
diff --git a/verif/generator/tosa_test_gen.py b/verif/generator/tosa_test_gen.py
index 53b0b75..ee935d4 100644
--- a/verif/generator/tosa_test_gen.py
+++ b/verif/generator/tosa_test_gen.py
@@ -66,6 +66,12 @@ class TosaTestGen:
                     v = maxFP
                 elif v == "-max":
                     v = -maxFP
+                elif v < 0:
+                    # Trim to minimum data type value
+                    v = max(v, -maxFP)
+                elif v > 0:
+                    # Trim to maximum data type value
+                    v = min(v, maxFP)
                 vals.append(v)
             return tuple(sorted(vals))
 
@@ -1722,27 +1728,19 @@ class TosaTestGen:
         self.ser.addOperator(op["op"], input_list, output_list, attr)
         return result_tens
 
-    def build_gather(self, op, values, validator_fcns=None, error_name=None):
-
-        # Create a new indicies tensor
-        # here with data that doesn't exceed the dimensions of the values tensor
-
-        K = values.shape[1]  # K
-        W = self.randInt(
-            self.args.tensor_shape_range[0], self.args.tensor_shape_range[1]
-        )  # W
-        indicies_arr = np.int32(
-            self.rng.integers(low=0, high=K, size=[values.shape[0], W])
-        )  # (N, W)
-        indicies = self.ser.addConst(indicies_arr.shape, DType.INT32, indicies_arr)
+    def build_gather(
+        self, op, inputs, args_dict, validator_fcns=None, error_name=None, qinfo=None
+    ):
+        assert len(inputs) == 2
+        values, indices = inputs
 
-        result_tens = OutputShaper.gatherOp(
-            self.ser, self.rng, values, indicies, error_name
+        result_tensor = OutputShaper.gatherOp(
+            self.ser, self.rng, values, indices, error_name
         )
 
         # Invalidate Input/Output list for error if checks.
-        input_list = [values.name, indicies.name]
-        output_list = [result_tens.name]
+        input_list = [values.name, indices.name]
+        output_list = [result_tensor.name]
         pCount, cCount = op["operands"]
         num_operands = pCount + cCount
         input_list, output_list = TosaErrorIfArgGen.eiInvalidateInputOutputList(
@@ -1755,10 +1753,10 @@ class TosaTestGen:
             error_name,
             op=op,
             input_shape=values.shape,
-            output_shape=result_tens.shape,
+            output_shape=result_tensor.shape,
             input_dtype=values.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,
@@ -1767,33 +1765,24 @@ class TosaTestGen:
 
         self.ser.addOperator(op["op"], input_list, output_list)
 
-        return result_tens
+        compliance = self.tensorComplianceMetaData(
+            op, values.dtype, args_dict, result_tensor, error_name
+        )
 
-    def build_scatter(self, op, values_in, input, validator_fcns=None, error_name=None):
-
-        K = values_in.shape[1]  # K
-        W = input.shape[1]  # W
-
-        # Create an indices tensor here with data that doesn't exceed the
-        # dimension K of the values_in tensor and does NOT repeat the same K
-        # location as needed by the spec:
-        # "It is not permitted to repeat the same output index within a single
-        # SCATTER operation and so each output index occurs at most once."
-        assert K >= W
-        arr = []
-        for n in range(values_in.shape[0]):
-            # Get a shuffled list of output indices and limit it to size W
-            arr.append(self.rng.permutation(K)[:W])
-        indices_arr = np.array(arr, dtype=np.int32)  # (N, W)
-        indices = self.ser.addConst(indices_arr.shape, DType.INT32, indices_arr)
-
-        result_tens = OutputShaper.scatterOp(
+        return TosaTestGen.BuildInfo(result_tensor, compliance)
+
+    def build_scatter(
+        self, op, inputs, args_dict, validator_fcns=None, error_name=None, qinfo=None
+    ):
+        assert len(inputs) == 3
+        values_in, indices, input = inputs
+        result_tensor = OutputShaper.scatterOp(
             self.ser, self.rng, values_in, indices, input, error_name
         )
 
         # Invalidate Input/Output list for error if checks.
         input_list = [values_in.name, indices.name, input.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(
@@ -1806,10 +1795,10 @@ class TosaTestGen:
             error_name,
             op=op,
             input_shape=values_in.shape,
-            output_shape=result_tens.shape,
+            output_shape=result_tensor.shape,
             input_dtype=values_in.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,
@@ -1818,7 +1807,11 @@ class TosaTestGen:
 
         self.ser.addOperator(op["op"], input_list, output_list)
 
-        return result_tens
+        compliance = self.tensorComplianceMetaData(
+            op, values_in.dtype, args_dict, result_tensor, error_name
+        )
+
+        return TosaTestGen.BuildInfo(result_tensor, compliance)
 
     def build_resize(
         self,
@@ -4310,14 +4303,13 @@ class TosaTestGen:
         # Scatter/Gather
         "gather": {
             "op": Op.GATHER,
-            # Only specify 'values' tensor here. 'indices' is generated in op building stage
-            "operands": (1, 0),
+            "operands": (2, 0),
             "rank": (3, 3),
             "build_fcn": (
                 build_gather,
-                TosaTensorGen.tgBasic,
-                TosaTensorValuesGen.tvgDefault,
-                None,
+                TosaTensorGen.tgGather,
+                TosaTensorValuesGen.tvgGather,
+                TosaArgGen.agNone,
             ),
             "types": (
                 DType.INT8,
@@ -4334,18 +4326,19 @@ class TosaTestGen:
                 TosaErrorValidator.evWrongOutputList,
                 TosaErrorValidator.evWrongRank,
             ),
+            "data_gen": {
+                "fp": (gtu.DataGenType.PSEUDO_RANDOM,),
+            },
         },
         "scatter": {
             "op": Op.SCATTER,
-            # Only specify 'values_in' tensor here.
-            # 'indices' and 'input' are generated in op building stage
-            "operands": (2, 0),
+            "operands": (3, 0),
             "rank": (3, 3),
             "build_fcn": (
                 build_scatter,
                 TosaTensorGen.tgScatter,
-                TosaTensorValuesGen.tvgDefault,
-                None,
+                TosaTensorValuesGen.tvgScatter,
+                TosaArgGen.agNone,
             ),
             "types": TYPE_INT_FP,
             "error_if_validators": (
@@ -4355,6 +4348,9 @@ class TosaTestGen:
                 TosaErrorValidator.evWrongOutputList,
                 TosaErrorValidator.evWrongRank,
             ),
+            "data_gen": {
+                "fp": (gtu.DataGenType.PSEUDO_RANDOM,),
+            },
         },
         # Image operations
         "resize": {