#9045: Merge 1 and 2 cq tests back into one file after enabling mixin…

…g of 1/2 cqs in the same process

models/demos/resnet/tests/test_metal_resnet50_performant.py

@@ -5,7 +5,13 @@
 import pytest
 import tt_lib
 
-from models.demos.resnet.tests.test_metal_resnet50 import run_resnet50_inference, run_model, run_trace_model
+from models.demos.resnet.tests.test_metal_resnet50 import (
+    run_resnet50_inference,
+    run_model,
+    run_trace_model,
+    run_2cq_model,
+    run_trace_2cq_model,
+)
 from models.utility_functions import skip_for_wormhole_b0
 
 
@@ -83,3 +89,81 @@ def test_run_resnet50_trace_inference(
     )
 
     device.enable_async(False)
+
+
+@skip_for_wormhole_b0("This test is not supported on WHB0, please use the TTNN version.")
+@pytest.mark.parametrize("device_params", [{"l1_small_size": 24576, "num_hw_cqs": 2}], indirect=True)
+@pytest.mark.parametrize("batch_size", [20], ids=["batch_20"])
+@pytest.mark.parametrize(
+    "weights_dtype",
+    [tt_lib.tensor.DataType.BFLOAT8_B],
+    ids=["weights_BFLOAT8_B"],
+)
+@pytest.mark.parametrize(
+    "activations_dtype",
+    [tt_lib.tensor.DataType.BFLOAT8_B],
+    ids=["activations_BFLOAT8_B"],
+)
+@pytest.mark.parametrize(
+    "math_fidelity",
+    [tt_lib.tensor.MathFidelity.LoFi],
+    ids=["LoFi"],
+)
+def test_run_resnet50_2cqs_inference(
+    device, use_program_cache, batch_size, weights_dtype, activations_dtype, math_fidelity, imagenet_sample_input
+):
+    run_resnet50_inference(
+        device,
+        batch_size,
+        weights_dtype,
+        activations_dtype,
+        math_fidelity,
+        imagenet_sample_input,
+        run_2cq_model,
+    )
+
+
+@skip_for_wormhole_b0("This test is not supported on WHB0, please use the TTNN version.")
+@pytest.mark.parametrize(
+    "device_params", [{"l1_small_size": 24576, "num_hw_cqs": 2, "trace_region_size": 1500000}], indirect=True
+)
+@pytest.mark.parametrize("batch_size", [20], ids=["batch_20"])
+@pytest.mark.parametrize(
+    "weights_dtype",
+    [tt_lib.tensor.DataType.BFLOAT8_B],
+    ids=["weights_BFLOAT8_B"],
+)
+@pytest.mark.parametrize(
+    "activations_dtype",
+    [tt_lib.tensor.DataType.BFLOAT8_B],
+    ids=["activations_BFLOAT8_B"],
+)
+@pytest.mark.parametrize(
+    "math_fidelity",
+    [tt_lib.tensor.MathFidelity.LoFi],
+    ids=["LoFi"],
+)
+@pytest.mark.parametrize("enable_async", [True, False])
+def test_run_resnet50_trace_2cqs_inference(
+    device,
+    use_program_cache,
+    batch_size,
+    weights_dtype,
+    activations_dtype,
+    math_fidelity,
+    imagenet_sample_input,
+    enable_async,
+):
+    device.enable_async(enable_async)
+
+    run_resnet50_inference(
+        device,
+        batch_size,
+        weights_dtype,
+        activations_dtype,
+        math_fidelity,
+        imagenet_sample_input,
+        run_trace_2cq_model,
+    )
+
+    device.enable_async(False)

models/demos/resnet/tests/test_perf_resnet.py

@@ -384,3 +384,50 @@ def test_perf_trace_bare_metal(
         f"resnet50_trace_{mode}",
     )
     device.enable_async(False)
+
+
+@skip_for_wormhole_b0(reason_str="Not tested on single WH")
+@pytest.mark.parametrize("device_params", [{"l1_small_size": 32768, "num_hw_cqs": 2}], indirect=True)
+@pytest.mark.models_performance_bare_metal
+@pytest.mark.parametrize(
+    "batch_size, expected_inference_time, expected_compile_time",
+    ((20, 0.0042, 16),),
+)
+def test_perf_2cqs_bare_metal(
+    device,
+    use_program_cache,
+    batch_size,
+    expected_inference_time,
+    expected_compile_time,
+    hf_cat_image_sample_input,
+):
+    run_perf_resnet(
+        batch_size, expected_inference_time, expected_compile_time, hf_cat_image_sample_input, device, "resnet50_2cqs"
+    )
+
+
+@skip_for_wormhole_b0(reason_str="Not tested on single WH")
+@pytest.mark.parametrize(
+    "device_params", [{"l1_small_size": 32768, "num_hw_cqs": 2, "trace_region_size": 1332224}], indirect=True
+)
+@pytest.mark.models_performance_bare_metal
+@pytest.mark.parametrize(
+    "batch_size, expected_inference_time, expected_compile_time",
+    ((20, 0.0042, 16),),
+)
+def test_perf_trace_2cqs_bare_metal(
+    device,
+    use_program_cache,
+    batch_size,
+    expected_inference_time,
+    expected_compile_time,
+    hf_cat_image_sample_input,
+):
+    run_perf_resnet(
+        batch_size,
+        expected_inference_time,
+        expected_compile_time,
+        hf_cat_image_sample_input,
+        device,
+        "resnet50_trace_2cqs",
+    )

tests/scripts/run_performance.sh

@@ -17,9 +17,7 @@ run_perf_models_other() {
 
     env pytest models/demos/ttnn_falcon7b/tests -m $test_marker
 
-    # Separate calls since we can't mix switching between number of cqs
     env pytest models/demos/resnet/tests/test_perf_resnet.py -m $test_marker
-    env pytest models/demos/resnet/tests/test_perf_resnet_2cqs.py -m $test_marker
 
     env pytest tests/ttnn/integration_tests/whisper/test_performance.py -m $test_marker
 

tests/scripts/single_card/nightly/run_gs_only.sh

@@ -10,5 +10,3 @@ fi
 echo "Running model nightly tests for GS only"
 
 env pytest models/demos/resnet/tests/test_metal_resnet50_performant.py
-
-env pytest models/demos/resnet/tests/test_metal_resnet50_2cqs_performant.py