commit dd14c1b2ba22f4b37be9ec1a9a7d61741d36506e
author Won Jeon <won.jeon@arm.com> Thu Jun 29 23:20:00 2023 +0000
committer Eric Kunze <eric.kunze@arm.com> Thu Jul 06 21:36:53 2023 +0000
tree 74d772c12768e253d3d79c619cdaf7cb7922af14
parent d69e2838b2e2f46401c5da19f662c9e0d5a5df06

Add reference model framework test for INT8 squared difference

Signed-off-by: Won Jeon <won.jeon@arm.com>
Change-Id: I6fba6907cef0616c18dc461dbb92d2aceb582f6c

verif/frameworks/tosa_verif_framework_generator.py

diff --git a/verif/frameworks/tosa_verif_framework_generator.py b/verif/frameworks/tosa_verif_framework_generator.py
index 94a1a15..09a06b4 100755
--- a/verif/frameworks/tosa_verif_framework_generator.py
+++ b/verif/frameworks/tosa_verif_framework_generator.py
@@ -369,7 +369,10 @@
     "squared_difference": {
         "operands": (2, 0),
         "build_fcn": (TBuilder.SquaredDifference, TGen.tgBFuzz, ArgGen.agNone),
-        "types": TYPE_F,
+        "types": {
+            "tf": TYPE_F,
+            "tflite": list(TYPE_FI + [QuantType.ALL_I8]),
+        },
     },
     "equal": {
         "operands": (2, 0),
@@ -1119,7 +1122,12 @@
                     max_val = float(qmax - qzero[idx]) * scale
                 else:
                     scale = (max_val - min_val) / float(qmax - qmin)
-                    zeropoint = int(round((-min_val) / scale)) + qmin
+                    zeropoint = -int(round((-min_val) / scale)) + qmin
+
+                # Exit if min_val <= 0.0, in order to avoid assertion error
+                # from tf.quantization.fake_quant_with_min_max_args
+                if min_val > 0.0:
+                    return True
 
                 # run through tf.fakequant first to assure quantization error aligned
                 fakequant_val = tf.quantization.fake_quant_with_min_max_args(