Merge pull request #384 from es-ude/365-implement-inference-execution…

…-with-pytest

365 implement inference execution with pytest

elasticai/creator/nn/fixed_point/conv1d/_testbench_test.py

@@ -0,0 +1,141 @@
+from typing import Any, Callable
+
+import pytest
+import torch
+
+from elasticai.creator.vhdl.auto_wire_protocols.port_definitions import create_port
+from elasticai.creator.vhdl.design.ports import Port
+
+from ..number_converter import FXPParams, NumberConverter
+from .testbench import Conv1dTestbench
+from .design import Conv1dDesign
+
+
+class DummyConv1d:
+    def __init__(self, fxp_params: FXPParams, in_channels: int, out_channels: int):
+        self.name: str = "conv1d"
+        self.kernel_size: int = 1
+        self.input_signal_length = 1
+        self.in_channels = in_channels
+        self.out_channels = out_channels
+        self.port: Port = create_port(
+            y_width=fxp_params.total_bits,
+            x_width=fxp_params.total_bits,
+            x_count=1,
+            y_count=2,
+        )
+
+
+def parameters_for_reported_content_parsing(fxp_params, input_expected_pairs):
+    def add_expected_prefix_to_pairs(pairs):
+        _converter_for_batch = NumberConverter(
+            FXPParams(8, 0)
+        )  # max for 255 lines of inputs
+        pairs_with_prefix = list()
+        for i, (pairs_text, pairs_number) in enumerate(pairs):
+            pairs_with_prefix.append(list())
+            pairs_with_prefix[i].append(list())
+            pairs_with_prefix[i].append(pairs_number)
+            for batch_number, batch_channel_text in enumerate(pairs_text):
+                for out_channel_text in batch_channel_text:
+                    for value_text in out_channel_text:
+                        pairs_with_prefix[i][0].append(
+                            f"result: {_converter_for_batch.integer_to_bits(batch_number)},"
+                            f" {value_text}"
+                        )
+        return pairs_with_prefix
+
+    pairs_with_prefix = [
+        (fxp_params, a, b)
+        for a, b in add_expected_prefix_to_pairs(input_expected_pairs)
+    ]
+    return pairs_with_prefix
+
+
+@pytest.fixture
+def create_uut() -> Callable[[FXPParams, int, int], Conv1dDesign]:
+    def create(fxp_params, in_channels: int, out_channels: int) -> Conv1dDesign:
+        return DummyConv1d(fxp_params, in_channels=in_channels, out_channels=out_channels)
+
+    return create
+
+
+@pytest.mark.parametrize(
+    "fxp_params, reported, y", (
+        parameters_for_reported_content_parsing(
+            fxp_params=FXPParams(total_bits=3, frac_bits=0),
+            input_expected_pairs=[
+                ([[["010"]]], [[[2.0]]]),
+                ([[["001", "010"]]], [[[1.0, 2.0]]]),
+                ([[["111", "001"]]], [[[-1.0, 1.0]]]),
+            ]
+        ) +
+        parameters_for_reported_content_parsing(
+            fxp_params=FXPParams(total_bits=4, frac_bits=1),
+            input_expected_pairs=[
+                ([[["0001", "1111"]]], [[[0.5, -0.5]]]),
+                ([[["0001", "0011"]], [["1000", "1111"]]], [[[0.5, 1.5]], [[-4.0, -0.5]]]),
+            ]
+        )
+    )
+)
+def test_parse_reported_content_one_out_channel(fxp_params, reported, y, create_uut):
+    in_channels = None
+    out_channels = 1
+    bench = Conv1dTestbench(
+        name="conv1d_testbench", fxp_params=fxp_params, uut=create_uut(fxp_params, in_channels, out_channels)
+    )
+    print(reported)
+    assert y == bench.parse_reported_content(reported)
+
+
+@pytest.mark.parametrize(
+    "fxp_params, reported, y", (
+        parameters_for_reported_content_parsing(
+            fxp_params=FXPParams(total_bits=3, frac_bits=0),
+            input_expected_pairs=[
+                ([[["010"],["010"]]], [[[2.0],[2.0]]]),
+                ([[["001", "010"], ["001", "010"]]], [[[1.0, 2.0], [1.0, 2.0]]]),
+                ([[["111", "001"], ["111", "001"]]], [[[-1.0, 1.0], [-1.0, 1.0]]]),
+            ]
+        ) +
+        parameters_for_reported_content_parsing(
+            fxp_params=FXPParams(total_bits=4, frac_bits=1),
+            input_expected_pairs=[
+                ([[["0001", "1111"], ["0001", "1111"]]], [[[0.5, -0.5], [0.5, -0.5]]]),
+                ([[["0001", "0011"], ["0001", "0011"]], [["1000", "1111"], ["1000", "1111"]]],
+                 [[[0.5, 1.5], [0.5, 1.5]], [[-4.0, -0.5], [-4.0, -0.5]]]),
+            ]
+        )
+    )
+)
+def test_parse_reported_content_two_out_channel(fxp_params, reported, y, create_uut):
+    in_channels = None
+    out_channels = 2
+    bench = Conv1dTestbench(
+        name="conv1d_testbench", fxp_params=fxp_params, uut=create_uut(fxp_params, in_channels, out_channels)
+    )
+    print(reported)
+    assert y == bench.parse_reported_content(reported)
+
+def test_input_preparation_with_one_in_channel(create_uut):
+    fxp_params = FXPParams(total_bits=3, frac_bits=0)
+    in_channels = 1
+    out_channels = None
+    bench = Conv1dTestbench(
+        name="bench_name", fxp_params=fxp_params, uut=create_uut(fxp_params, in_channels, out_channels),
+    )
+    input = torch.Tensor([[[1.0, 1.0]]])
+    expected = [{"x_0_0": "001", "x_0_1": "001"}]
+    assert expected == bench.prepare_inputs(input.tolist())
+
+def test_input_preparation_with_two_in_channel(create_uut):
+    fxp_params = FXPParams(total_bits=3, frac_bits=0)
+    in_channels = 1
+    out_channels = None
+    bench = Conv1dTestbench(
+        name="bench_name", fxp_params=fxp_params, uut=create_uut(fxp_params, in_channels, out_channels),
+    )
+    input = torch.Tensor([[[1.0, 1.0], [1.0, 2.0]]])
+    expected = [{"x_0_0": "001", "x_0_1": "001", "x_1_0": "001", "x_1_1": "010"}]
+    assert expected == bench.prepare_inputs(input.tolist())

elasticai/creator/nn/fixed_point/conv1d/conv1d.tpl.vhd

@@ -1,6 +1,55 @@
--- Dummy File for testing implementation of conv1d Design
-${total_bits}
-${frac_bits}
-${in_channels}
-${out_channels}
-${kernel_size}
+library ieee;
+use ieee.std_logic_1164.all;
+
+entity ${name} is
+    port (
+        enable : in std_logic;
+        clock  : in std_logic;
+        x_address : out std_logic_vector(${x_address_width}-1 downto 0);
+        y_address : in std_logic_vector(${y_address_width}-1 downto 0);
+
+        x   : in std_logic_vector(${x_width}-1 downto 0);
+        y  : out std_logic_vector(${y_width}-1 downto 0);
+
+        done   : out std_logic
+    );
+end;
+
+architecture rtl of ${name} is
+    constant TOTAL_WIDTH : natural := ${x_width};
+    constant FRAC_WIDTH : natural := ${frac_width};
+    constant VECTOR_WIDTH : natural := ${vector_width};
+    constant KERNEL_SIZE : natural := ${kernel_size};
+    constant IN_CHANNELS : natural := ${in_channels};
+    constant OUT_CHANNELS : natural := ${out_channels};
+    constant X_ADDRESS_WIDTH : natural := ${x_address_width};
+    constant Y_ADDRESS_WIDTH : natural := ${y_address_width};
+
+    signal reset : std_logic;
+
+begin
+
+    reset <= not enable;
+
+    ${name}_conv1d : entity work.conv1d_fxp_MAC_RoundToZero
+        generic map(
+            TOTAL_WIDTH => TOTAL_WIDTH,
+            FRAC_WIDTH => FRAC_WIDTH,
+            VECTOR_WIDTH => VECTOR_WIDTH,
+            KERNEL_SIZE => KERNEL_SIZE,
+            IN_CHANNELS => IN_CHANNELS,
+            OUT_CHANNELS => OUT_CHANNELS,
+            X_ADDRESS_WIDTH => X_ADDRESS_WIDTH,
+            Y_ADDRESS_WIDTH => Y_ADDRESS_WIDTH
+        )
+        port map (
+            clock => clock,
+            enable => enable,
+            reset => reset,
+            x => x,
+            x_address => x_address,
+            y => y,
+            y_address => y_address,
+            done => done
+        );
+end rtl;