Disable IC3 IC4 Q8 XNNPACK (#871)

Summary:
Pull Request resolved: #871

We need a fix from quantizer; disable it for MVP but we can enable it for main branch later.

The issue is quantization scale difference due to calibration and XNNPACK won't execute the program; it only occurs when running it with both "--delegate" and "--quantize".

With portable kernel it executes.

Reviewed By: guangy10

Differential Revision: D50236323

fbshipit-source-id: b37cf0bb0294f9c7e65f945dc9c871e0b889b607

Author: kirklandsign

examples/xnnpack/__init__.py

@@ -18,8 +18,8 @@ class XNNPACKOptions(object):
     "add": XNNPACKOptions(True, True),
     "add_mul": XNNPACKOptions(True, True),
     "dl3": XNNPACKOptions(True, True),
-    "ic3": XNNPACKOptions(True, True),
-    "ic4": XNNPACKOptions(True, True),
+    "ic3": XNNPACKOptions(False, True),
+    "ic4": XNNPACKOptions(False, True),
     "mv2": XNNPACKOptions(True, True),
     "mv3": XNNPACKOptions(True, True),
     "resnet18": XNNPACKOptions(True, True),