Arm backend: Enable linear 16a8w tests

Enable tests of int16 activations and int8 weight quantization.
Test for large_rand is disabled to sort out why the test is flaky.

Signed-off-by: Per Åstrand <[email protected]>
Change-Id: I9de5d472f8862edebcf82c140399985db930c069

Author: per

backends/arm/scripts/parse_test_names.py

@@ -95,6 +95,9 @@ def parse_test_name(
     op = op.removesuffix("_1d")
     op = op.removesuffix("_2d")
 
+    # Remove suffix for 16 bit activation and 8 bit weight test cases
+    op = op.removesuffix("_16a8w")
+
     assert target != "None", f"{test_name} does not contain one of {TARGETS}"
     assert (
         op in op_name_map.keys()

backends/arm/test/ops/test_linear.py

@@ -8,7 +8,6 @@
 
 from typing import Tuple
 
-import pytest
 import torch
 from executorch.backends.arm.quantizer.arm_quantizer import (
     get_symmetric_a16w8_quantization_config,
@@ -276,10 +275,14 @@ def get_symmetric_a16w8_linear_quantizer(
     )
 
 
-@common.parametrize("test_data", test_data_rank1_INT | test_data_rank4_INT)
-@pytest.mark.xfail(
-    reason="missing int16 linear ops support; fails at TOSA reference model run with Invalid TOSA graph"
-)
+test_data_all_16a8w = test_data_rank1_INT | test_data_rank4_INT
+# TODO: Remove large rand test as they are flaky until sorted out why: MLETORCH-1377
+for k in list(test_data_all_16a8w.keys()):
+    if "large_rand" in k:
+        test_data_all_16a8w.pop(k)
+
+
+@common.parametrize("test_data", test_data_all_16a8w)
 def test_linear_16a8w_tosa_INT(test_data: torch.Tensor):
     """Test linear operation with 16A8W quantization (16-bit activations, 8-bit weights)"""
     test_data, out_features, has_bias, per_channel_quantization = test_data()