Add more unary backward ops sweeps (#13771)

* #1112: Copy sweeps commits from branch amalbasaTT/backward_ops-sweeps-3

* #11512: Refactoring

.github/workflows/ttnn-run-sweeps.yaml

@@ -120,6 +120,18 @@ on:
           - eltwise.unary_backward.sinh_bw.sinh_bw
           - eltwise.unary_backward.sin_bw.sin_bw
           - eltwise.unary_backward.square_bw.square_bw
+          - eltwise.unary_backward.rdiv_bw.rdiv_bw
+          - eltwise.unary_backward.bias_gelu_bw.bias_gelu_bw
+          - eltwise.unary_backward.pow_bw.pow_bw
+          - eltwise.unary_backward.exp_bw.exp_bw
+          - eltwise.unary_backward.tanh_bw.tanh_bw
+          - eltwise.unary_backward.sqrt_bw.sqrt_bw
+          - eltwise.unary_backward.add_bw.add_bw
+          - eltwise.unary_backward.assign_bw.assign_bw
+          - eltwise.unary_backward.fill_bw.fill_bw
+          - eltwise.unary_backward.hardsigmoid_bw.hardsigmoid_bw
+          - eltwise.unary_backward.lgamma_bw.lgamma_bw
+          - eltwise.unary_backward.multigammaln_bw.multigammaln_bw
           - eltwise.unary.lgamma
           - eltwise.unary.logit
           - eltwise.unary.mish

tests/sweep_framework/sweeps/eltwise/unary_backward/add_bw/add_bw.py

@@ -0,0 +1,108 @@
+# SPDX-FileCopyrightText: © 2024 Tenstorrent Inc.
+
+# SPDX-License-Identifier: Apache-2.0
+
+from typing import Optional, Tuple
+from functools import partial
+
+import torch
+import random
+import ttnn
+from tests.sweep_framework.utils import gen_shapes, sanitize_shape_rm
+from tests.tt_eager.python_api_testing.sweep_tests.generation_funcs import gen_func_with_cast_tt
+
+from tests.ttnn.utils_for_testing import check_with_pcc, start_measuring_time, stop_measuring_time
+from models.utility_functions import torch_random
+
+# Override the default timeout in seconds for hang detection.
+TIMEOUT = 30
+
+random.seed(0)
+
+
+# Parameters provided to the test vector generator are defined here.
+# They are defined as dict-type suites that contain the arguments to the run function as keys, and lists of possible inputs as values.
+# Each suite has a key name (in this case "suite_1" and "suite_2") which will associate the test vectors to this specific suite of inputs.
+# Developers can create their own generator functions and pass them to the parameters as inputs.
+parameters = {
+    "nightly": {
+        "input_shape": gen_shapes([1, 1, 32, 32], [6, 12, 256, 256], [1, 1, 1, 1], 4)
+        + gen_shapes([1, 32, 32], [12, 256, 256], [1, 1, 1], 4)
+        + gen_shapes([32, 32], [256, 256], [1, 1], 4),
+        "grad_dtype": [ttnn.bfloat16, ttnn.bfloat8_b],
+        "input_a_dtype": [ttnn.bfloat16, ttnn.bfloat8_b],
+        "input_layout": [ttnn.TILE_LAYOUT, ttnn.ROW_MAJOR_LAYOUT],
+        "grad_memory_config": [ttnn.DRAM_MEMORY_CONFIG, ttnn.L1_MEMORY_CONFIG],
+        "input_a_memory_config": [ttnn.DRAM_MEMORY_CONFIG, ttnn.L1_MEMORY_CONFIG],
+        "output_memory_config": [ttnn.DRAM_MEMORY_CONFIG, ttnn.L1_MEMORY_CONFIG],
+    },
+}
+
+
+# Invalidate vector is called during the generation phase where each vector will be passed in.
+# If invalidated, the vector will still be stored but will be skipped.
+# Returns False, None if the vector is valid, and True, str with a reason for invalidation if it is invalid.
+def invalidate_vector(test_vector) -> Tuple[bool, Optional[str]]:
+    if test_vector["input_layout"] == ttnn.ROW_MAJOR_LAYOUT and (
+        test_vector["input_a_dtype"] == ttnn.bfloat8_b or test_vector["grad_dtype"] == ttnn.bfloat8_b
+    ):
+        return True, "bfloat8_b is not supported on row major layout"
+    return False, None
+
+
+# This is the run instructions for the test, defined by the developer.
+# The run function must take the above-defined parameters as inputs.
+# The runner will call this run function with each test vector, and the returned results from this function will be stored.
+# If you defined a mesh_device_fixture above, the object you yielded will be passed into this function as 'device'. Otherwise, it will be the default ttnn device opened by the infra.
+def run(
+    input_shape,
+    grad_dtype,
+    input_a_dtype,
+    input_layout,
+    grad_memory_config,
+    input_a_memory_config,
+    output_memory_config,
+    *,
+    device,
+) -> list:
+    data_seed = random.randint(0, 20000000)
+    torch.manual_seed(data_seed)
+
+    if input_layout == ttnn.ROW_MAJOR_LAYOUT:
+        input_shape = sanitize_shape_rm(input_shape)
+
+    torch_grad_tensor = gen_func_with_cast_tt(
+        partial(torch_random, low=-100, high=100, dtype=torch.float32), grad_dtype
+    )(input_shape)
+    torch_input_tensor_a = gen_func_with_cast_tt(
+        partial(torch_random, low=-100, high=100, dtype=torch.float32), input_a_dtype
+    )(input_shape)
+    torch_input_tensor_a.requires_grad = True
+
+    scalar = torch.tensor(1, dtype=torch.bfloat16).uniform_(-100, 100).item()
+
+    golden_function = ttnn.get_golden_function(ttnn.add_bw)
+    torch_output_tensor = golden_function(torch_grad_tensor, torch_input_tensor_a, scalar)[0]
+
+    grad_tensor = ttnn.from_torch(
+        torch_grad_tensor,
+        dtype=grad_dtype,
+        layout=input_layout,
+        device=device,
+        memory_config=grad_memory_config,
+    )
+
+    input_tensor_a = ttnn.from_torch(
+        torch_input_tensor_a.detach().clone(),
+        dtype=input_a_dtype,
+        layout=input_layout,
+        device=device,
+        memory_config=input_a_memory_config,
+    )
+
+    start_time = start_measuring_time()
+    output_tensor = ttnn.add_bw(grad_tensor, input_tensor_a, scalar, memory_config=output_memory_config)[0]
+    output_tensor = ttnn.to_torch(output_tensor)
+    e2e_perf = stop_measuring_time(start_time)
+
+    return [check_with_pcc(torch_output_tensor, output_tensor, 0.999), e2e_perf]

tests/sweep_framework/sweeps/eltwise/unary_backward/assign_bw/assign_bw.py

@@ -0,0 +1,106 @@
+# SPDX-FileCopyrightText: © 2024 Tenstorrent Inc.
+
+# SPDX-License-Identifier: Apache-2.0
+
+from typing import Optional, Tuple
+from functools import partial
+
+import torch
+import random
+import ttnn
+from tests.sweep_framework.utils import gen_shapes, sanitize_shape_rm
+from tests.tt_eager.python_api_testing.sweep_tests.generation_funcs import gen_func_with_cast_tt
+
+from tests.ttnn.utils_for_testing import check_with_pcc, start_measuring_time, stop_measuring_time
+from models.utility_functions import torch_random
+
+# Override the default timeout in seconds for hang detection.
+TIMEOUT = 30
+
+random.seed(0)
+
+
+# Parameters provided to the test vector generator are defined here.
+# They are defined as dict-type suites that contain the arguments to the run function as keys, and lists of possible inputs as values.
+# Each suite has a key name (in this case "suite_1" and "suite_2") which will associate the test vectors to this specific suite of inputs.
+# Developers can create their own generator functions and pass them to the parameters as inputs.
+parameters = {
+    "nightly": {
+        "input_shape": gen_shapes([1, 1, 1, 1], [6, 12, 256, 256], [1, 1, 1, 1], 16)
+        + gen_shapes([1, 1, 1], [12, 256, 256], [1, 1, 1], 16)
+        + gen_shapes([1, 1], [256, 256], [1, 1], 16),
+        "grad_dtype": [ttnn.bfloat16],
+        "input_a_dtype": [ttnn.bfloat16],
+        "input_layout": [ttnn.TILE_LAYOUT, ttnn.ROW_MAJOR_LAYOUT],
+        "grad_memory_config": [ttnn.DRAM_MEMORY_CONFIG, ttnn.L1_MEMORY_CONFIG],
+        "input_a_memory_config": [ttnn.DRAM_MEMORY_CONFIG, ttnn.L1_MEMORY_CONFIG],
+        "output_memory_config": [ttnn.DRAM_MEMORY_CONFIG, ttnn.L1_MEMORY_CONFIG],
+    },
+}
+
+
+# Invalidate vector is called during the generation phase where each vector will be passed in.
+# If invalidated, the vector will still be stored but will be skipped.
+# Returns False, None if the vector is valid, and True, str with a reason for invalidation if it is invalid.
+def invalidate_vector(test_vector) -> Tuple[bool, Optional[str]]:
+    if test_vector["input_layout"] == ttnn.ROW_MAJOR_LAYOUT and (
+        test_vector["grad_dtype"] == ttnn.bfloat8_b or test_vector["input_a_dtype"] == ttnn.bfloat8_b
+    ):
+        return True, "bfloat8_b is only supported on tiled layout"
+    return False, None
+
+
+# This is the run instructions for the test, defined by the developer.
+# The run function must take the above-defined parameters as inputs.
+# The runner will call this run function with each test vector, and the returned results from this function will be stored.
+# If you defined a mesh_device_fixture above, the object you yielded will be passed into this function as 'device'. pOtherwise, it will be the default ttnn device opened by the infra.
+def run(
+    input_shape,
+    grad_dtype,
+    input_a_dtype,
+    input_layout,
+    grad_memory_config,
+    input_a_memory_config,
+    output_memory_config,
+    *,
+    device,
+) -> list:
+    data_seed = random.randint(0, 20000000)
+    torch.manual_seed(data_seed)
+
+    if input_layout == ttnn.ROW_MAJOR_LAYOUT:
+        input_shape = sanitize_shape_rm(input_shape)
+
+    torch_grad_tensor = gen_func_with_cast_tt(
+        partial(torch_random, low=-100, high=100, dtype=torch.float32), grad_dtype
+    )(input_shape)
+    torch_input_tensor_a = gen_func_with_cast_tt(
+        partial(torch_random, low=-100, high=100, dtype=torch.float32), input_a_dtype
+    )(input_shape)
+    torch_input_tensor_a.requires_grad = True
+
+    golden_function = ttnn.get_golden_function(ttnn.assign_bw)
+    torch_output_tensor = golden_function(torch_grad_tensor, torch_input_tensor_a)[0]
+
+    grad_tensor = ttnn.from_torch(
+        torch_grad_tensor,
+        dtype=grad_dtype,
+        layout=input_layout,
+        device=device,
+        memory_config=grad_memory_config,
+    )
+
+    input_tensor_a = ttnn.from_torch(
+        torch_input_tensor_a,
+        dtype=input_a_dtype,
+        layout=input_layout,
+        device=device,
+        memory_config=input_a_memory_config,
+    )
+
+    start_time = start_measuring_time()
+    output_tensor = ttnn.assign_bw(grad_tensor, input_tensor_a, memory_config=output_memory_config)[0]
+    output_tensor = ttnn.to_torch(output_tensor)
+    e2e_perf = stop_measuring_time(start_time)
+
+    return [check_with_pcc(torch_output_tensor, output_tensor, 0.999), e2e_perf]

tests/sweep_framework/sweeps/eltwise/unary_backward/bias_gelu_bw/bias_gelu_bw.py

@@ -0,0 +1,128 @@
+# SPDX-FileCopyrightText: © 2024 Tenstorrent Inc.
+
+# SPDX-License-Identifier: Apache-2.0
+
+from typing import Optional, Tuple
+from functools import partial
+
+import torch
+import random
+import ttnn
+from tests.sweep_framework.utils import gen_shapes, gen_rand_exclude_range
+from tests.tt_eager.python_api_testing.sweep_tests.generation_funcs import gen_func_with_cast_tt
+
+from tests.ttnn.utils_for_testing import check_with_pcc, start_measuring_time, stop_measuring_time
+from models.utility_functions import torch_random
+
+# Override the default timeout in seconds for hang detection.
+TIMEOUT = 30
+
+random.seed(0)
+
+
+# Parameters provided to the test vector generator are defined here.
+# They are defined as dict-type suites that contain the arguments to the run function as keys, and lists of possible inputs as values.
+# Each suite has a key name (in this case "suite_1" and "suite_2") which will associate the test vectors to this specific suite of inputs.
+# Developers can create their own generator functions and pass them to the parameters as inputs.
+parameters = {
+    "nightly": {
+        "input_shape": gen_shapes([1, 1, 1, 1], [6, 12, 256, 256], [1, 1, 1, 1], 8)
+        + gen_shapes([1, 1, 1], [12, 256, 256], [1, 1, 1], 8)
+        + gen_shapes([1, 1], [256, 256], [1, 1], 8),
+        "approximate": ["none"],
+        "grad_dtype": [ttnn.bfloat16, ttnn.bfloat8_b],
+        "input_a_dtype": [ttnn.bfloat16],
+        "input_layout": [ttnn.TILE_LAYOUT, ttnn.ROW_MAJOR_LAYOUT],
+        "grad_memory_config": [ttnn.DRAM_MEMORY_CONFIG, ttnn.L1_MEMORY_CONFIG],
+        "input_a_memory_config": [ttnn.DRAM_MEMORY_CONFIG, ttnn.L1_MEMORY_CONFIG],
+        "output_memory_config": [ttnn.DRAM_MEMORY_CONFIG, ttnn.L1_MEMORY_CONFIG],
+    },
+    "xfail": {
+        "input_shape": gen_shapes([1, 1, 32, 32], [6, 12, 256, 256], [1, 1, 32, 32], 8)
+        + gen_shapes([1, 1, 1], [12, 256, 256], [1, 32, 32], 8)
+        + gen_shapes([1, 1], [256, 256], [32, 32], 8),
+        "approximate": ["none", "tanh"],
+        "grad_dtype": [ttnn.bfloat16, ttnn.bfloat8_b],
+        "input_a_dtype": [ttnn.bfloat16, ttnn.bfloat8_b],
+        "input_layout": [ttnn.TILE_LAYOUT, ttnn.ROW_MAJOR_LAYOUT],
+        "grad_memory_config": [ttnn.DRAM_MEMORY_CONFIG, ttnn.L1_MEMORY_CONFIG],
+        "input_a_memory_config": [ttnn.DRAM_MEMORY_CONFIG, ttnn.L1_MEMORY_CONFIG],
+        "output_memory_config": [ttnn.DRAM_MEMORY_CONFIG, ttnn.L1_MEMORY_CONFIG],
+    },
+}
+
+
+# Invalidate vector is called during the generation phase where each vector will be passed in.
+# If invalidated, the vector will still be stored but will be skipped.
+# Returns False, None if the vector is valid, and True, str with a reason for invalidation if it is invalid.
+def invalidate_vector(test_vector) -> Tuple[bool, Optional[str]]:
+    if test_vector["input_layout"] == ttnn.ROW_MAJOR_LAYOUT:
+        return True, "Inputs to eltwise binary must be tilized"
+    if test_vector["input_layout"] == ttnn.ROW_MAJOR_LAYOUT and (
+        test_vector["grad_dtype"] == ttnn.bfloat8_b or test_vector["input_a_dtype"] == ttnn.bfloat8_b
+    ):
+        return True, "bfloat8_b is only supported on tiled layout"
+    return False, None
+
+
+# This is the run instructions for the test, defined by the developer.
+# The run function must take the above-defined parameters as inputs.
+# The runner will call this run function with each test vector, and the returned results from this function will be stored.
+# If you defined a mesh_device_fixture above, the object you yielded will be passed into this function as 'device'. Otherwise, it will be the default ttnn device opened by the infra.
+def run(
+    input_shape,
+    approximate,
+    grad_dtype,
+    input_a_dtype,
+    input_layout,
+    grad_memory_config,
+    input_a_memory_config,
+    output_memory_config,
+    *,
+    device,
+) -> list:
+    data_seed = random.randint(0, 20000000)
+    torch.manual_seed(data_seed)
+
+    torch_grad_tensor = gen_func_with_cast_tt(
+        partial(torch_random, low=-100, high=100, dtype=torch.float32), grad_dtype
+    )(input_shape)
+    torch_input_tensor_a = gen_func_with_cast_tt(
+        partial(torch_random, low=-100, high=100, dtype=torch.float32), input_a_dtype
+    )(input_shape)
+    torch_input_tensor_a.requires_grad = True
+
+    scalar = torch.tensor(1, dtype=torch.bfloat16).uniform_(-100, 100).item()
+
+    golden_function = ttnn.get_golden_function(ttnn.bias_gelu_bw)
+    torch_output_tensor = golden_function(torch_grad_tensor, torch_input_tensor_a, scalar, value=approximate)[0]
+    # intermediate_result = torch.nn.functional.gelu(torch.add(torch_input_tensor_a, scalar), approximate=approximate)
+    # intermediate_result.backward(gradient=torch_grad_tensor)
+    # torch_output_tensor = torch_input_tensor_a.grad
+
+    grad_tensor = ttnn.from_torch(
+        torch_grad_tensor,
+        dtype=grad_dtype,
+        layout=input_layout,
+        device=device,
+        memory_config=grad_memory_config,
+    )
+
+    input_tensor_a = ttnn.from_torch(
+        torch_input_tensor_a,
+        dtype=input_a_dtype,
+        layout=input_layout,
+        device=device,
+        memory_config=input_a_memory_config,
+    )
+
+    start_time = start_measuring_time()
+    output_tensor = ttnn.bias_gelu_bw(
+        grad_tensor, input_tensor_a, scalar, approximate=approximate, memory_config=output_memory_config
+    )[0]
+    output_tensor = ttnn.to_torch(output_tensor)
+    e2e_perf = stop_measuring_time(start_time)
+
+    info_string = f"Dtypes - grad:{grad_dtype}, input:{input_a_dtype}. Approximation:{approximate}"
+
+    return [check_with_pcc(torch_output_tensor, output_tensor, 0.999), e2e_perf, info_string]