FP16 support in serialization

* Allow serialization of fp16 data
* Add package to support integrated half data-type (half_float::half), independent of native float: http://half.sourceforge.net/
* Allow passing of accumulate data-type in serialization

Signed-off-by: James Ward <james.ward@arm.com>
Change-Id: I54357f02e3776d81958228f699ea5044f2014f4b

7 files changed, 97 insertions, 10 deletions

diff --git a/python/serializer/tosa_serializer.py b/python/serializer/tosa_serializer.py
index acec4b7..fb89563 100644
--- a/python/serializer/tosa_serializer.py
+++ b/python/serializer/tosa_serializer.py
@@ -58,6 +58,7 @@ DTypeNames = [
     "INT48",
     "FLOAT",
     "UINT16",
+    "FP16",
 ]
 
 ByteMask = np.uint64(0xFF)
@@ -145,7 +146,15 @@ class TosaSerializerAttribute(TosaSerializerUnion):
     def __init__(self):
         super().__init__()
 
-    def PoolAttribute(self, kernel, stride, pad, input_zp, output_zp):
+    def PoolAttribute(
+        self,
+        kernel,
+        stride,
+        pad,
+        input_zp,
+        output_zp,
+        accum_dtype,
+    ):
         from tosa import PoolAttribute as a, Attribute
 
         self.utype = Attribute.Attribute().PoolAttribute
@@ -156,8 +165,9 @@ class TosaSerializerAttribute(TosaSerializerUnion):
         self.intvecs.append((a.AddStride, stride))
         self.ints.append((a.AddInputZp, input_zp))
         self.ints.append((a.AddOutputZp, output_zp))
+        self.ints.append((a.AddAccumDtype, accum_dtype))
 
-    def ConvAttribute(self, pad, stride, dilation, input_zp, weight_zp):
+    def ConvAttribute(self, pad, stride, dilation, input_zp, weight_zp, accum_dtype):
         from tosa import ConvAttribute as a, Attribute
 
         self.utype = Attribute.Attribute().ConvAttribute
@@ -168,8 +178,11 @@ class TosaSerializerAttribute(TosaSerializerUnion):
         self.intvecs.append((a.AddDilation, dilation))
         self.ints.append((a.AddInputZp, input_zp))
         self.ints.append((a.AddWeightZp, weight_zp))
+        self.ints.append((a.AddAccumDtype, accum_dtype))
 
-    def TransposeConvAttribute(self, outpad, stride, output_shape, input_zp, weight_zp):
+    def TransposeConvAttribute(
+        self, outpad, stride, output_shape, input_zp, weight_zp, accum_dtype
+    ):
         from tosa import TransposeConvAttribute as a, Attribute
 
         self.utype = Attribute.Attribute().TransposeConvAttribute
@@ -180,6 +193,7 @@ class TosaSerializerAttribute(TosaSerializerUnion):
         self.intvecs.append((a.AddOutputShape, output_shape))
         self.ints.append((a.AddInputZp, input_zp))
         self.ints.append((a.AddWeightZp, weight_zp))
+        self.ints.append((a.AddAccumDtype, accum_dtype))
 
     def PadAttribute(self, padding, pad_const_int, pad_const_fp):
         from tosa import PadAttribute as a, Attribute
@@ -316,7 +330,7 @@ class TosaSerializerAttribute(TosaSerializerUnion):
 
         self.intvecs.append((a.AddTable, table))
 
-    def MatMulAttribute(self, A_zp, B_zp):
+    def MatMulAttribute(self, A_zp, B_zp, accum_dtype):
         from tosa import MatMulAttribute as a, Attribute
 
         self.utype = Attribute.Attribute().MatMulAttribute
@@ -324,8 +338,9 @@ class TosaSerializerAttribute(TosaSerializerUnion):
 
         self.ints.append((a.AddAZp, A_zp))
         self.ints.append((a.AddBZp, B_zp))
+        self.ints.append((a.AddAccumDtype, accum_dtype))
 
-    def FullyConnectedAttribute(self, input_zp, weight_zp):
+    def FullyConnectedAttribute(self, input_zp, weight_zp, accum_dtype):
         from tosa import FullyConnectedAttribute as a, Attribute
 
         self.utype = Attribute.Attribute().FullyConnectedAttribute
@@ -333,6 +348,7 @@ class TosaSerializerAttribute(TosaSerializerUnion):
 
         self.ints.append((a.AddInputZp, input_zp))
         self.ints.append((a.AddWeightZp, weight_zp))
+        self.ints.append((a.AddAccumDtype, accum_dtype))
 
     def NegateAttribute(self, input1_zp, output_zp):
         from tosa import NegateAttribute as a, Attribute
@@ -364,6 +380,8 @@ class TosaSerializerTensor:
 
         if dtype == DType.FLOAT:
             fntype = np.float32
+        elif dtype == DType.FP16:
+            fntype = np.float16
         else:
             fntype = int
 
@@ -445,10 +463,18 @@ class TosaSerializerTensor:
                     b4 = (val_u64 >> np.uint64(32)) & ByteMask
                     b5 = (val_u64 >> np.uint64(40)) & ByteMask
                     u8_data.extend([b0, b1, b2, b3, b4, b5])
+            elif self.dtype == DType.FP16:
+                np_arr = np.array(self.data, dtype=np.float16)
+                u8_data.extend(np_arr.view(np.uint8))
             elif self.dtype == DType.FLOAT:
                 for val in self.data:
                     b = struct.pack("!f", val)
                     u8_data.extend([b[3], b[2], b[1], b[0]])
+            elif self.dtype == TosaDType.DType:
+                # Serialize DType enum data as uint8 bytes
+                for val in self.data:
+                    np_arr = np.array(self.data, dtype=np.uint32)
+                    u8_data.extend(np_arr.view(np.uint8))
             else:
                 raise Exception(
                     "unsupported data type {}".format(DTypeNames[self.dtype])
@@ -873,6 +899,7 @@ class TosaSerializer:
             )
             ConvAttribute.AddInputZp = ConvAttribute.ConvAttributeAddInputZp
             ConvAttribute.AddWeightZp = ConvAttribute.ConvAttributeAddWeightZp
+            ConvAttribute.AddAccumDtype = ConvAttribute.ConvAttributeAddAccumDtype
             ConvAttribute.End = ConvAttribute.ConvAttributeEnd
         from tosa import FullyConnectedAttribute
 
@@ -886,6 +913,9 @@ class TosaSerializer:
             FullyConnectedAttribute.AddWeightZp = (
                 FullyConnectedAttribute.FullyConnectedAttributeAddWeightZp
             )
+            FullyConnectedAttribute.AddAccumDtype = (
+                FullyConnectedAttribute.FullyConnectedAttributeAddAccumDtype
+            )
             FullyConnectedAttribute.End = (
                 FullyConnectedAttribute.FullyConnectedAttributeEnd
             )
@@ -895,6 +925,7 @@ class TosaSerializer:
             MatMulAttribute.Start = MatMulAttribute.MatMulAttributeStart
             MatMulAttribute.AddAZp = MatMulAttribute.MatMulAttributeAddAZp
             MatMulAttribute.AddBZp = MatMulAttribute.MatMulAttributeAddBZp
+            MatMulAttribute.AddAccumDtype = MatMulAttribute.MatMulAttributeAddAccumDtype
             MatMulAttribute.End = MatMulAttribute.MatMulAttributeEnd
         from tosa import PoolAttribute
 
@@ -910,6 +941,7 @@ class TosaSerializer:
             PoolAttribute.StartStrideVector = (
                 PoolAttribute.PoolAttributeStartStrideVector
             )
+            PoolAttribute.AddAccumDtype = PoolAttribute.PoolAttributeAddAccumDtype
             PoolAttribute.AddInputZp = PoolAttribute.PoolAttributeAddInputZp
             PoolAttribute.AddOutputZp = PoolAttribute.PoolAttributeAddOutputZp
             PoolAttribute.End = PoolAttribute.PoolAttributeEnd
@@ -944,6 +976,7 @@ class TosaSerializer:
             PoolAttribute.StartStrideVector = (
                 PoolAttribute.PoolAttributeStartStrideVector
             )
+            PoolAttribute.AddAccumDtype = PoolAttribute.PoolAttributeAddAccumDtype
             PoolAttribute.AddInputZp = PoolAttribute.PoolAttributeAddInputZp
             PoolAttribute.AddOutputZp = PoolAttribute.PoolAttributeAddOutputZp
             PoolAttribute.End = PoolAttribute.PoolAttributeEnd
@@ -1123,6 +1156,9 @@ class TosaSerializer:
             TransposeConvAttribute.AddWeightZp = (
                 TransposeConvAttribute.TransposeConvAttributeAddWeightZp
             )
+            TransposeConvAttribute.AddAccumDtype = (
+                TransposeConvAttribute.TransposeConvAttributeAddAccumDtype
+            )
             TransposeConvAttribute.End = (
                 TransposeConvAttribute.TransposeConvAttributeEnd
             )
diff --git a/python/tosa/ConvAttribute.py b/python/tosa/ConvAttribute.py
index fb22c7a..c06e8c7 100644
--- a/python/tosa/ConvAttribute.py
+++ b/python/tosa/ConvAttribute.py
@@ -123,7 +123,14 @@ class ConvAttribute(object):
             return self._tab.Get(flatbuffers.number_types.Int32Flags, o + self._tab.Pos)
         return 0
 
-def ConvAttributeStart(builder): builder.StartObject(5)
+    # ConvAttribute
+    def AccumDtype(self):
+        o = flatbuffers.number_types.UOffsetTFlags.py_type(self._tab.Offset(14))
+        if o != 0:
+            return self._tab.Get(flatbuffers.number_types.Uint32Flags, o + self._tab.Pos)
+        return 0
+
+def ConvAttributeStart(builder): builder.StartObject(6)
 def Start(builder):
     return ConvAttributeStart(builder)
 def ConvAttributeAddPad(builder, pad): builder.PrependUOffsetTRelativeSlot(0, flatbuffers.number_types.UOffsetTFlags.py_type(pad), 0)
@@ -150,6 +157,9 @@ def AddInputZp(builder, inputZp):
 def ConvAttributeAddWeightZp(builder, weightZp): builder.PrependInt32Slot(4, weightZp, 0)
 def AddWeightZp(builder, weightZp):
     return ConvAttributeAddWeightZp(builder, weightZp)
+def ConvAttributeAddAccumDtype(builder, accumDtype): builder.PrependUint32Slot(5, accumDtype, 0)
+def AddAccumDtype(builder, accumDtype):
+    return ConvAttributeAddAccumDtype(builder, accumDtype)
 def ConvAttributeEnd(builder): return builder.EndObject()
 def End(builder):
     return ConvAttributeEnd(builder)
\ No newline at end of file
diff --git a/python/tosa/DType.py b/python/tosa/DType.py
index e6b41ed..27d28c4 100644
--- a/python/tosa/DType.py
+++ b/python/tosa/DType.py
@@ -13,3 +13,4 @@ class DType(object):
     INT48 = 7
     FLOAT = 8
     UINT16 = 9
+    FP16 = 10
diff --git a/python/tosa/FullyConnectedAttribute.py b/python/tosa/FullyConnectedAttribute.py
index 892b0da..546ec60 100644
--- a/python/tosa/FullyConnectedAttribute.py
+++ b/python/tosa/FullyConnectedAttribute.py
@@ -42,7 +42,14 @@ class FullyConnectedAttribute(object):
             return self._tab.Get(flatbuffers.number_types.Int32Flags, o + self._tab.Pos)
         return 0
 
-def FullyConnectedAttributeStart(builder): builder.StartObject(2)
+    # FullyConnectedAttribute
+    def AccumDtype(self):
+        o = flatbuffers.number_types.UOffsetTFlags.py_type(self._tab.Offset(8))
+        if o != 0:
+            return self._tab.Get(flatbuffers.number_types.Uint32Flags, o + self._tab.Pos)
+        return 0
+
+def FullyConnectedAttributeStart(builder): builder.StartObject(3)
 def Start(builder):
     return FullyConnectedAttributeStart(builder)
 def FullyConnectedAttributeAddInputZp(builder, inputZp): builder.PrependInt32Slot(0, inputZp, 0)
@@ -51,6 +58,9 @@ def AddInputZp(builder, inputZp):
 def FullyConnectedAttributeAddWeightZp(builder, weightZp): builder.PrependInt32Slot(1, weightZp, 0)
 def AddWeightZp(builder, weightZp):
     return FullyConnectedAttributeAddWeightZp(builder, weightZp)
+def FullyConnectedAttributeAddAccumDtype(builder, accumDtype): builder.PrependUint32Slot(2, accumDtype, 0)
+def AddAccumDtype(builder, accumDtype):
+    return FullyConnectedAttributeAddAccumDtype(builder, accumDtype)
 def FullyConnectedAttributeEnd(builder): return builder.EndObject()
 def End(builder):
     return FullyConnectedAttributeEnd(builder)
\ No newline at end of file
diff --git a/python/tosa/MatMulAttribute.py b/python/tosa/MatMulAttribute.py
index b42ebfa..af6ba0b 100644
--- a/python/tosa/MatMulAttribute.py
+++ b/python/tosa/MatMulAttribute.py
@@ -42,7 +42,14 @@ class MatMulAttribute(object):
             return self._tab.Get(flatbuffers.number_types.Int32Flags, o + self._tab.Pos)
         return 0
 
-def MatMulAttributeStart(builder): builder.StartObject(2)
+    # MatMulAttribute
+    def AccumDtype(self):
+        o = flatbuffers.number_types.UOffsetTFlags.py_type(self._tab.Offset(8))
+        if o != 0:
+            return self._tab.Get(flatbuffers.number_types.Uint32Flags, o + self._tab.Pos)
+        return 0
+
+def MatMulAttributeStart(builder): builder.StartObject(3)
 def Start(builder):
     return MatMulAttributeStart(builder)
 def MatMulAttributeAddAZp(builder, aZp): builder.PrependInt32Slot(0, aZp, 0)
@@ -51,6 +58,9 @@ def AddAZp(builder, aZp):
 def MatMulAttributeAddBZp(builder, bZp): builder.PrependInt32Slot(1, bZp, 0)
 def AddBZp(builder, bZp):
     return MatMulAttributeAddBZp(builder, bZp)
+def MatMulAttributeAddAccumDtype(builder, accumDtype): builder.PrependUint32Slot(2, accumDtype, 0)
+def AddAccumDtype(builder, accumDtype):
+    return MatMulAttributeAddAccumDtype(builder, accumDtype)
 def MatMulAttributeEnd(builder): return builder.EndObject()
 def End(builder):
     return MatMulAttributeEnd(builder)
\ No newline at end of file
diff --git a/python/tosa/PoolAttribute.py b/python/tosa/PoolAttribute.py
index 8256a6d..4307114 100644
--- a/python/tosa/PoolAttribute.py
+++ b/python/tosa/PoolAttribute.py
@@ -123,7 +123,14 @@ class PoolAttribute(object):
             return self._tab.Get(flatbuffers.number_types.Int32Flags, o + self._tab.Pos)
         return 0
 
-def PoolAttributeStart(builder): builder.StartObject(5)
+    # PoolAttribute
+    def AccumDtype(self):
+        o = flatbuffers.number_types.UOffsetTFlags.py_type(self._tab.Offset(14))
+        if o != 0:
+            return self._tab.Get(flatbuffers.number_types.Uint32Flags, o + self._tab.Pos)
+        return 0
+
+def PoolAttributeStart(builder): builder.StartObject(6)
 def Start(builder):
     return PoolAttributeStart(builder)
 def PoolAttributeAddPad(builder, pad): builder.PrependUOffsetTRelativeSlot(0, flatbuffers.number_types.UOffsetTFlags.py_type(pad), 0)
@@ -150,6 +157,9 @@ def AddInputZp(builder, inputZp):
 def PoolAttributeAddOutputZp(builder, outputZp): builder.PrependInt32Slot(4, outputZp, 0)
 def AddOutputZp(builder, outputZp):
     return PoolAttributeAddOutputZp(builder, outputZp)
+def PoolAttributeAddAccumDtype(builder, accumDtype): builder.PrependUint32Slot(5, accumDtype, 0)
+def AddAccumDtype(builder, accumDtype):
+    return PoolAttributeAddAccumDtype(builder, accumDtype)
 def PoolAttributeEnd(builder): return builder.EndObject()
 def End(builder):
     return PoolAttributeEnd(builder)
\ No newline at end of file
diff --git a/python/tosa/TransposeConvAttribute.py b/python/tosa/TransposeConvAttribute.py
index a2824e2..1a6bbde 100644
--- a/python/tosa/TransposeConvAttribute.py
+++ b/python/tosa/TransposeConvAttribute.py
@@ -123,7 +123,14 @@ class TransposeConvAttribute(object):
             return self._tab.Get(flatbuffers.number_types.Int32Flags, o + self._tab.Pos)
         return 0
 
-def TransposeConvAttributeStart(builder): builder.StartObject(5)
+    # TransposeConvAttribute
+    def AccumDtype(self):
+        o = flatbuffers.number_types.UOffsetTFlags.py_type(self._tab.Offset(14))
+        if o != 0:
+            return self._tab.Get(flatbuffers.number_types.Uint32Flags, o + self._tab.Pos)
+        return 0
+
+def TransposeConvAttributeStart(builder): builder.StartObject(6)
 def Start(builder):
     return TransposeConvAttributeStart(builder)
 def TransposeConvAttributeAddOutPad(builder, outPad): builder.PrependUOffsetTRelativeSlot(0, flatbuffers.number_types.UOffsetTFlags.py_type(outPad), 0)
@@ -150,6 +157,9 @@ def AddInputZp(builder, inputZp):
 def TransposeConvAttributeAddWeightZp(builder, weightZp): builder.PrependInt32Slot(4, weightZp, 0)
 def AddWeightZp(builder, weightZp):
     return TransposeConvAttributeAddWeightZp(builder, weightZp)
+def TransposeConvAttributeAddAccumDtype(builder, accumDtype): builder.PrependUint32Slot(5, accumDtype, 0)
+def AddAccumDtype(builder, accumDtype):
+    return TransposeConvAttributeAddAccumDtype(builder, accumDtype)
 def TransposeConvAttributeEnd(builder): return builder.EndObject()
 def End(builder):
     return TransposeConvAttributeEnd(builder)
\ No newline at end of file