Arm backend: Update vgf ops tests [Part 2]

backends/arm/test/models/test_nss.py

@@ -110,7 +110,7 @@ def test_nss_u85_INT():
     reason="[MLETORCH-1430]: Double types are not supported in buffers in MSL"
 )
 @common.SkipIfNoModelConverter
-def test_nss_vgf_no_quant():
+def test_nss_vgf_FP():
     pipeline = VgfPipeline[input_t](
         nss().eval(),
         example_inputs(),
@@ -119,12 +119,14 @@ def test_nss_vgf_no_quant():
         use_to_edge_transform_and_lower=True,
         run_on_vulkan_runtime=True,
         quantize=False,
+        # Override tosa version to test FP-only path
+        tosa_version="TOSA-1.0+FP",
     )
     pipeline.run()
 
 
 @common.SkipIfNoModelConverter
-def test_nss_vgf_quant():
+def test_nss_vgf_INT():
     pipeline = VgfPipeline[input_t](
         nss().eval(),
         example_inputs(),
@@ -134,6 +136,8 @@ def test_nss_vgf_quant():
         use_to_edge_transform_and_lower=True,
         run_on_vulkan_runtime=True,
         quantize=True,
+        # Override tosa version to test INT-only path
+        tosa_version="TOSA-1.0+INT",
     )
     pipeline.run()
 

backends/arm/test/ops/test_hardsigmoid.py

@@ -90,21 +90,25 @@ def test_hardsigmoid_u85_INT(test_data: torch.Tensor):
 
 @common.parametrize("test_data", test_data_suite)
 @common.SkipIfNoModelConverter
-def test_hardsigmoid_vgf_FP(test_data: torch.Tensor):
+def test_hardsigmoid_vgf_no_quant(test_data: torch.Tensor):
     pipeline = VgfPipeline[input_t1](
-        Hardsigmoid(), (test_data(),), aten_op, exir_op=[], tosa_version="TOSA-1.0+FP"
+        Hardsigmoid(),
+        (test_data(),),
+        aten_op,
+        exir_op=[],
+        quantize=False,
     )
     pipeline.run()
 
 
 @common.parametrize("test_data", test_data_suite)
 @common.SkipIfNoModelConverter
-def test_hardsigmoid_vgf_INT(test_data: torch.Tensor):
+def test_hardsigmoid_vgf_quant(test_data: torch.Tensor):
     pipeline = VgfPipeline[input_t1](
         Hardsigmoid(),
         (test_data(),),
         aten_op,
         exir_op=[],
-        tosa_version="TOSA-1.0+INT",
+        quantize=True,
     )
     pipeline.run()

backends/arm/test/ops/test_hardswish.py

@@ -80,21 +80,25 @@ def test_hardswish_u85_INT(test_data):
 
 @common.parametrize("test_data", test_data_suite)
 @common.SkipIfNoModelConverter
-def test_hardswish_vgf_FP(test_data):
+def test_hardswish_vgf_no_quant(test_data):
     pipeline = VgfPipeline[input_t1](
-        Hardswish(), (test_data(),), aten_op, exir_op, tosa_version="TOSA-1.0+FP"
+        Hardswish(),
+        (test_data(),),
+        aten_op,
+        exir_op,
+        quantize=False,
     )
     pipeline.run()
 
 
 @common.parametrize("test_data", test_data_suite)
 @common.SkipIfNoModelConverter
-def test_hardswish_vgf_INT(test_data):
+def test_hardswish_vgf_quant(test_data):
     pipeline = VgfPipeline[input_t1](
         Hardswish(),
         (test_data(),),
         aten_op,
         exir_op,
-        tosa_version="TOSA-1.0+INT",
+        quantize=True,
     )
     pipeline.run()

backends/arm/test/ops/test_hardtanh.py

@@ -89,21 +89,25 @@ def test_hardtanh_u85_INT(test_data: torch.Tensor):
 
 @common.parametrize("test_data", test_data_suite)
 @common.SkipIfNoModelConverter
-def test_hardtanh_vgf_FP(test_data: torch.Tensor):
+def test_hardtanh_vgf_no_quant(test_data: torch.Tensor):
     pipeline = VgfPipeline[input_t](
-        HardTanh(), (test_data(),), aten_op, exir_op, tosa_version="TOSA-1.0+FP"
+        HardTanh(),
+        (test_data(),),
+        aten_op,
+        exir_op,
+        quantize=False,
     )
     pipeline.run()
 
 
 @common.parametrize("test_data", test_data_suite)
 @common.SkipIfNoModelConverter
-def test_hardtanh_vgf_INT(test_data: torch.Tensor):
+def test_hardtanh_vgf_quant(test_data: torch.Tensor):
     pipeline = VgfPipeline[input_t](
         HardTanh(),
         (test_data(),),
         aten_op,
         exir_op,
-        tosa_version="TOSA-1.0+INT",
+        quantize=True,
     )
     pipeline.run()

backends/arm/test/ops/test_index_select.py

@@ -137,41 +137,41 @@ def test_index_select_u55_INT_not_delegated(test_data: input_params):
 
 @pytest.mark.parametrize("test_data", list(test_data.values()))
 @common.SkipIfNoModelConverter
-def test_index_select_vgf_FP(test_data: input_params):
+def test_index_select_vgf_no_quant(test_data: input_params):
     op, inp = test_data
     pipeline = VgfPipeline[input_params](
         op,
         inp,
         op.aten_op,
         op.exir_op,
-        tosa_version="TOSA-1.0+FP",
+        quantize=False,
     )
     pipeline.run()
 
 
 @pytest.mark.parametrize("test_data", list(test_data.values())[:-1])
 @common.SkipIfNoModelConverter
-def test_index_select_vgf_INT(test_data: input_params):
+def test_index_select_vgf_quant(test_data: input_params):
     op, inp = test_data
     pipeline = VgfPipeline[input_params](
         op,
         inp,
         op.aten_op,
         op.exir_op,
-        tosa_version="TOSA-1.0+INT",
+        quantize=True,
     )
     pipeline.run()
 
 
 @pytest.mark.parametrize("test_data", list(test_data.values())[-1:])
 @common.SkipIfNoModelConverter
-def test_index_select_vgf_INT_rand(test_data: input_params):
+def test_index_select_vgf_quant_rand(test_data: input_params):
     op, inp = test_data
     pipeline = VgfPipeline[input_params](
         op,
         inp,
         op.aten_op,
         op.exir_op,
-        tosa_version="TOSA-1.0+INT",
+        quantize=True,
     )
     pipeline.run()

backends/arm/test/ops/test_layer_norm.py

@@ -115,26 +115,26 @@ def test_native_layer_norm_u85_INT(test_data):
 
 @common.parametrize("test_data", test_data_suite)
 @common.SkipIfNoModelConverter
-def test_native_layer_norm_vgf_FP(test_data):
+def test_native_layer_norm_vgf_no_quant(test_data):
     test_input, model = test_data()
     pipeline = VgfPipeline[input_t](
         model,
         test_input,
         "torch.ops.aten.layer_norm.default",
-        tosa_version="TOSA-1.0+FP",
+        quantize=False,
     )
     pipeline.run()
 
 
 @common.parametrize("test_data", test_data_suite)
 @common.SkipIfNoModelConverter
-def test_native_layer_norm_vgf_INT(test_data):
+def test_native_layer_norm_vgf_quant(test_data):
     test_input, model = test_data()
     pipeline = VgfPipeline[input_t](
         model,
         test_input,
         "torch.ops.aten.sub.Tensor",
-        tosa_version="TOSA-1.0+INT",
+        quantize=True,
     )
     pipeline.run()
 

backends/arm/test/ops/test_le.py

@@ -246,51 +246,51 @@ def test_le_scalar_16a8w_u85_INT16(test_module):
 
 @common.parametrize("test_module", test_data_tensor)
 @common.SkipIfNoModelConverter
-def test_le_tensor_vgf_FP(test_module):
+def test_le_tensor_vgf_no_quant(test_module):
     pipeline = VgfPipeline[input_t](
         test_module(),
         test_module().get_inputs(),
         LessEqual.aten_op_tensor,
         LessEqual.exir_op,
-        tosa_version="TOSA-1.0+FP",
+        quantize=False,
     )
     pipeline.run()
 
 
 @common.parametrize("test_module", test_data_tensor)
 @common.SkipIfNoModelConverter
-def test_le_tensor_vgf_INT(test_module):
+def test_le_tensor_vgf_quant(test_module):
     pipeline = VgfPipeline[input_t](
         test_module(),
         test_module().get_inputs(),
         LessEqual.aten_op_tensor,
         LessEqual.exir_op,
-        tosa_version="TOSA-1.0+INT",
+        quantize=True,
     )
     pipeline.run()
 
 
 @common.parametrize("test_module", test_data_scalar)
 @common.SkipIfNoModelConverter
-def test_le_scalar_vgf_FP(test_module):
+def test_le_scalar_vgf_no_quant(test_module):
     pipeline = VgfPipeline[input_t](
         test_module(),
         test_module().get_inputs(),
         LessEqual.aten_op_scalar,
         LessEqual.exir_op,
-        tosa_version="TOSA-1.0+FP",
+        quantize=False,
     )
     pipeline.run()
 
 
 @common.parametrize("test_module", test_data_scalar)
 @common.SkipIfNoModelConverter
-def test_le_scalar_vgf_INT(test_module):
+def test_le_scalar_vgf_quant(test_module):
     pipeline = VgfPipeline[input_t](
         test_module(),
         test_module().get_inputs(),
         LessEqual.aten_op_tensor,
         LessEqual.exir_op,
-        tosa_version="TOSA-1.0+INT",
+        quantize=True,
     )
     pipeline.run()

backends/arm/test/ops/test_leaky_relu.py

@@ -95,14 +95,14 @@ def test_leaky_relu_u85_INT(test_data):
 
 @common.parametrize("test_data", LeakyReLU.test_data)
 @common.SkipIfNoModelConverter
-def test_leaky_relu_vgf_FP(test_data):
+def test_leaky_relu_vgf_no_quant(test_data):
     data, slope = test_data()
     pipeline = VgfPipeline[input_t1](
         LeakyReLU(slope),
         data,
         [],
         use_to_edge_transform_and_lower=True,
-        tosa_version="TOSA-1.0+FP",
+        quantize=False,
     )
     pipeline.add_stage_after(
         "to_edge_transform_and_lower", pipeline.tester.check_not, [aten_op]
@@ -112,14 +112,14 @@ def test_leaky_relu_vgf_FP(test_data):
 
 @common.parametrize("test_data", LeakyReLU.test_data)
 @common.SkipIfNoModelConverter
-def test_leaky_relu_vgf_INT(test_data):
+def test_leaky_relu_vgf_quant(test_data):
     data, slope = test_data()
     pipeline = VgfPipeline[input_t1](
         LeakyReLU(slope),
         data,
         [],
         use_to_edge_transform_and_lower=True,
-        tosa_version="TOSA-1.0+INT",
+        quantize=True,
     )
     pipeline.add_stage_after("quantize", pipeline.tester.check_not, [aten_op])
     pipeline.run()

backends/arm/test/ops/test_linalg_vector_norm.py

@@ -128,7 +128,7 @@ def test_vector_norm_u85_INT_fvp(test_module):
 
 @common.parametrize("test_module", test_modules)
 @common.SkipIfNoModelConverter
-def test_vector_norm_vgf_FP(test_module):
+def test_vector_norm_vgf_no_quant(test_module):
     model, input_tensor = test_module
     # FP VGF
     aten_op = "torch.ops.aten.linalg_vector_norm.default"
@@ -138,14 +138,14 @@ def test_vector_norm_vgf_FP(test_module):
         input_tensor,
         aten_op,
         exir_op,
-        tosa_version="TOSA-1.0+FP",
+        quantize=False,
     )
     pipeline.run()
 
 
 @common.parametrize("test_module", test_modules)
 @common.SkipIfNoModelConverter
-def test_vector_norm_vgf_INT(test_module):
+def test_vector_norm_vgf_quant(test_module):
     model, input_tensor = test_module
     # Should not found this op
     exir_op = "executorch_exir_dialects_edge__ops_aten_linalg_vector_norm_default"
@@ -155,6 +155,6 @@ def test_vector_norm_vgf_INT(test_module):
         input_tensor,
         aten_op_q_decomposed_q,
         exir_op,
-        tosa_version="TOSA-1.0+INT",
+        quantize=True,
     )
     pipeline.run()

backends/arm/test/ops/test_linear.py

@@ -209,39 +209,31 @@ def test_linear_u85_INT(test_data: torch.Tensor):
 
 @common.parametrize("test_data", test_data_rank1_FP | test_data_rank4_FP)
 @common.SkipIfNoModelConverter
-def test_linear_vgf_FP(test_data: torch.Tensor):
+def test_linear_vgf_no_quant(test_data: torch.Tensor):
     test_data, out_features, has_bias = test_data()
     in_features = test_data.shape[-1]
     pipeline = VgfPipeline[input_t1](
-        Linear(
-            in_features=in_features,
-            out_features=out_features,
-            bias=has_bias,
-        ),
+        Linear(in_features=in_features, out_features=out_features, bias=has_bias),
         (test_data,),
         aten_op=aten_op,
         exir_op=[],
-        tosa_version="TOSA-1.0+FP",
+        quantize=False,
     )
     pipeline.run()
 
 
 @common.parametrize("test_data", test_data_rank1_INT | test_data_rank4_INT)
 @common.SkipIfNoModelConverter
-def test_linear_vgf_INT(test_data: torch.Tensor):
+def test_linear_vgf_quant(test_data: torch.Tensor):
     test_data, out_features, has_bias, per_channel_quantization = test_data()
     in_features = test_data.shape[-1]
     pipeline = VgfPipeline[input_t1](
-        Linear(
-            in_features=in_features,
-            out_features=out_features,
-            bias=has_bias,
-        ),
+        Linear(in_features=in_features, out_features=out_features, bias=has_bias),
         (test_data,),
         aten_op=aten_op,
         exir_op=[],
-        tosa_version="TOSA-1.0+INT",
         per_channel_quantization=per_channel_quantization,
+        quantize=True,
     )
     pipeline.run()