#12349: Enabling pytests on BH that are now working

tests/tt_eager/python_api_testing/sweep_tests/pytests/tt_dnn/test_div.py

@@ -16,15 +16,14 @@
 from tests.tt_eager.python_api_testing.sweep_tests.run_pytorch_ci_tests import (
     run_single_pytorch_test,
 )
-from models.utility_functions import is_grayskull, skip_for_blackhole
+from models.utility_functions import is_grayskull
 
 mem_configs = [
     ttnn.MemoryConfig(ttnn.TensorMemoryLayout.INTERLEAVED, ttnn.BufferType.DRAM),
     ttnn.MemoryConfig(ttnn.TensorMemoryLayout.INTERLEAVED, ttnn.BufferType.L1),
 ]
 
 
-@skip_for_blackhole("Only supported for WH, see #12349")
 @pytest.mark.parametrize("accurate_mode", [False, True])
 @pytest.mark.parametrize("round_mode", ["None", "trunc", "floor"])
 @pytest.mark.parametrize(

tests/tt_eager/python_api_testing/sweep_tests/pytests/tt_dnn/test_div_unary.py

@@ -16,15 +16,14 @@
 from tests.tt_eager.python_api_testing.sweep_tests.run_pytorch_ci_tests import (
     run_single_pytorch_test,
 )
-from models.utility_functions import is_grayskull, skip_for_blackhole
+from models.utility_functions import is_grayskull
 
 mem_configs = [
     ttnn.MemoryConfig(ttnn.TensorMemoryLayout.INTERLEAVED, ttnn.BufferType.DRAM),
     ttnn.MemoryConfig(ttnn.TensorMemoryLayout.INTERLEAVED, ttnn.BufferType.L1),
 ]
 
 
-@skip_for_blackhole("Only supported on WH, see #12349")
 @pytest.mark.parametrize("accurate_mode", [True])
 @pytest.mark.parametrize("round_mode", ["None", "trunc", "floor"])
 @pytest.mark.parametrize(

tests/tt_eager/python_api_testing/sweep_tests/pytests/tt_dnn/test_fmod.py

@@ -16,15 +16,14 @@
 from tests.tt_eager.python_api_testing.sweep_tests.run_pytorch_ci_tests import (
     run_single_pytorch_test,
 )
-from models.utility_functions import skip_for_grayskull, skip_for_blackhole
+from models.utility_functions import skip_for_grayskull
 
 mem_configs = [
     ttnn.MemoryConfig(ttnn.TensorMemoryLayout.INTERLEAVED, ttnn.BufferType.DRAM),
     ttnn.MemoryConfig(ttnn.TensorMemoryLayout.INTERLEAVED, ttnn.BufferType.L1),
 ]
 
 
-@skip_for_blackhole("Only supported on WH, see #12349")
 @pytest.mark.parametrize(
     "input_shapes",
     [

tests/tt_eager/python_api_testing/sweep_tests/pytests/tt_dnn/test_frac.py

@@ -7,7 +7,7 @@
 import random
 from functools import partial
 import ttnn
-from models.utility_functions import skip_for_grayskull, skip_for_blackhole
+from models.utility_functions import skip_for_grayskull
 
 from tests.tt_eager.python_api_testing.sweep_tests import (
     comparison_funcs,
@@ -23,7 +23,6 @@
 ]
 
 
-@skip_for_blackhole("Unsupported on BH, see #12349")
 @pytest.mark.parametrize(
     "input_shapes",
     [

tests/tt_eager/python_api_testing/unit_testing/misc/test_scaled_dot_product_attention_decode.py

@@ -469,6 +469,7 @@ def run_test_sdpa_decode_single_iter(
     assert out_pass
 
 
+@skip_for_blackhole("Unsupported on BH, see #12349")
 @skip_for_grayskull("Unsupported in GS since L1 runs OOM with most configs")
 @pytest.mark.parametrize(
     "dtype, q_dtype",
@@ -518,6 +519,7 @@ def test_sdpa_decode(
         )
 
 
+@skip_for_blackhole("Unsupported on BH, see #12349")
 @skip_for_grayskull("Unsupported in GS since L1 runs OOM with most configs")
 @pytest.mark.parametrize(
     "dtype, q_dtype",
@@ -550,6 +552,7 @@ def test_sdpa_decode_non_causal(device, b, nh, nkv, s, d, dtype, grid_size, q_dt
     assert device.num_program_cache_entries() == 1
 
 
+@skip_for_blackhole("Unsupported on BH, see #12349")
 @skip_for_grayskull("Unsupported in GS since L1 runs OOM with most configs")
 @pytest.mark.parametrize(
     "dtype, q_dtype",
@@ -869,6 +872,7 @@ def test_sdpa_decode_paged_attention(
     assert device.num_program_cache_entries() == 4
 
 
+@skip_for_blackhole("Unsupported on BH, see #12349")
 @skip_for_grayskull("Unsupported in GS since L1 runs OOM with most configs")
 @pytest.mark.parametrize(
     "dtype, q_dtype",
@@ -898,6 +902,7 @@ def test_sdpa_decode_sharded(device, b, nh, nkv, s, d, dtype, grid_size, q_dtype
     )
 
 
+@skip_for_blackhole("Unsupported on BH, see #12349")
 @skip_for_grayskull("Unsupported in GS since L1 runs OOM with most configs")
 @pytest.mark.skip("Skipping Perf Test in CI")
 def test_sdpa_decode_perf(device, use_program_cache):
@@ -952,6 +957,7 @@ def test_sdpa_decode_perf(device, use_program_cache):
         )
 
 
+@skip_for_blackhole("Unsupported on BH, see #12349")
 @skip_for_grayskull("Unsupported in GS since L1 runs OOM with most configs")
 @pytest.mark.parametrize(
     "dtype",
@@ -1171,6 +1177,7 @@ def run_test_sdpa_decode_ndpcc(device, b, nh, nkv, s, d, dtype, grid_size, q_dty
     logger.info(f"PCC failed Start Pos: {failed_start_pos}")
 
 
+@skip_for_blackhole("Unsupported on BH, see #12349")
 @pytest.mark.timeout(600)
 @pytest.mark.skip("Skipping due to causing 45 minutes timeout on tt eager unit tests")
 @skip_for_grayskull("Unsupported in GS since L1 runs OOM with most configs")

tests/tt_eager/python_api_testing/unit_testing/misc/test_sum.py

@@ -9,7 +9,6 @@
 from models.utility_functions import skip_for_blackhole
 
 
-@skip_for_blackhole("Mismatching on BH, see #12349")
 @pytest.mark.parametrize(
     "shape_dim",
     (

tests/tt_eager/python_api_testing/unit_testing/misc/test_transpose.py

@@ -631,6 +631,7 @@ def test_transpose_bfloat8_b(device, shape, swap_dims):
     assert_with_pcc(torch_output, tt_output, 0.9999)
 
 
+@skip_for_blackhole("Mismatching on BH, see #12349")
 @pytest.mark.parametrize(
     "dtype",
     (ttnn.bfloat16, ttnn.float32),
@@ -649,6 +650,7 @@ def test_transpose_hc(dtype, shape, device):
     transpose(shape, device, dim0=1, dim1=-2, input_dtype=dtype)
 
 
+@skip_for_blackhole("Mismatching on BH, see #12349")
 @pytest.mark.parametrize(
     "dtype",
     (ttnn.bfloat16, ttnn.float32),
@@ -677,6 +679,7 @@ def test_transpose_2D(dtype, shape, layout, device):
     assert_with_pcc(torch_output, tt_output, 0.9999)
 
 
+@skip_for_blackhole("Mismatching on BH, see #12349")
 @pytest.mark.parametrize(
     "dtype",
     (ttnn.bfloat16, ttnn.float32),
@@ -758,6 +761,7 @@ def test_transpose_failures(config, device):
     assert_with_pcc(torch_output, tt_output, 0.9999)
 
 
+@skip_for_blackhole("Mismatching on BH, see #12349")
 @pytest.mark.parametrize(
     "config",
     [

tests/tt_eager/python_api_testing/unit_testing/test_prod_nc.py

@@ -28,7 +28,6 @@ def get_tensors(input_shape, output_shape, device):
     return tt_input, tt_output, torch_input
 
 
-@skip_for_blackhole("Mismatching on BH, see #12349")
 @pytest.mark.parametrize(
     "input_shape",
     (

ttnn/cpp/ttnn/operations/eltwise/binary/device/binary_composite_op.cpp

@@ -280,7 +280,7 @@ Tensor ExecuteBinaryRemainder::invoke(const Tensor& input, float scalar, const s
 // Binary FMOD will be overloaded by unary FMOD in another PR
 Tensor ExecuteBinaryFmod::invoke(const Tensor& input_a, const Tensor& input_b, const std::optional<MemoryConfig>& output_mem_config) {
     auto arch = input_a.device()->arch();
-    TT_FATAL(arch == tt::ARCH::WORMHOLE_B0, "Op is only supported on Wormhole");
+    TT_FATAL(arch == tt::ARCH::WORMHOLE_B0 or arch == tt::ARCH::BLACKHOLE, "Op is only supported on Wormhole or Blackhole");
     DataType input_dtype = input_a.get_dtype();
     Tensor a = typecast(input_a, DataType::FLOAT32);
     Tensor b = typecast(input_b, DataType::FLOAT32);

ttnn/cpp/ttnn/operations/eltwise/unary/device/unary_composite_op.cpp

@@ -817,7 +817,7 @@ Tensor _normalize_global(const Tensor& y,  const std::optional<MemoryConfig>& ou
 
 Tensor _frac(const Tensor& input, const std::optional<MemoryConfig>& output_mem_config) {
     auto arch = input.device()->arch();
-    TT_FATAL(arch == tt::ARCH::WORMHOLE_B0, "Op is only supported on Wormhole");
+    TT_FATAL(arch == tt::ARCH::WORMHOLE_B0 or arch == tt::ARCH::BLACKHOLE, "Op is only supported on Wormhole or Blackhole");
     Tensor trunc_res = ttnn::trunc(input);
     Tensor result = ttnn::subtract(input, trunc_res, std::nullopt, output_mem_config);
     return result;