Louis Verhaard | 0b8268a | 2020-08-05 16:11:29 +0200 | [diff] [blame] | 1 | # Copyright (C) 2020 Arm Limited or its affiliates. All rights reserved. |
| 2 | # |
| 3 | # SPDX-License-Identifier: Apache-2.0 |
| 4 | # |
| 5 | # Licensed under the Apache License, Version 2.0 (the License); you may |
| 6 | # not use this file except in compliance with the License. |
| 7 | # You may obtain a copy of the License at |
| 8 | # |
| 9 | # www.apache.org/licenses/LICENSE-2.0 |
| 10 | # |
| 11 | # Unless required by applicable law or agreed to in writing, software |
| 12 | # distributed under the License is distributed on an AS IS BASIS, WITHOUT |
| 13 | # WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 14 | # See the License for the specific language governing permissions and |
| 15 | # limitations under the License. |
| 16 | # Description: |
| 17 | # Unit tests for LUT support |
Jacob Bohlin | 1a66697 | 2020-09-11 10:04:15 +0200 | [diff] [blame] | 18 | import random |
| 19 | |
Louis Verhaard | 0b8268a | 2020-08-05 16:11:29 +0200 | [diff] [blame] | 20 | import numpy as np |
| 21 | |
Louis Verhaard | 0b8268a | 2020-08-05 16:11:29 +0200 | [diff] [blame] | 22 | from ethosu.vela import lut |
| 23 | from ethosu.vela import mark_tensors |
| 24 | from ethosu.vela import pass_packing |
| 25 | from ethosu.vela.data_type import DataType |
| 26 | from ethosu.vela.high_level_command_stream import DMA |
| 27 | from ethosu.vela.nn_graph import Graph |
Louis Verhaard | aee5d75 | 2020-09-30 09:01:52 +0200 | [diff] [blame] | 28 | from ethosu.vela.operation import Op |
Tim Hall | d8339a7 | 2021-05-27 18:49:40 +0100 | [diff] [blame] | 29 | from ethosu.vela.rewrite_graph import rewrite_graph_pre_order |
Louis Verhaard | 0b8268a | 2020-08-05 16:11:29 +0200 | [diff] [blame] | 30 | from ethosu.vela.rewrite_graph import verify_graph_health |
| 31 | from ethosu.vela.tensor import create_const_tensor |
| 32 | from ethosu.vela.tensor import TensorPurpose |
| 33 | from ethosu.vela.test import testutil |
| 34 | |
| 35 | |
Tim Hall | d8339a7 | 2021-05-27 18:49:40 +0100 | [diff] [blame] | 36 | def set_256_lut(op, key, arch): |
Jacob Bohlin | 1a66697 | 2020-09-11 10:04:15 +0200 | [diff] [blame] | 37 | random.seed(key) |
| 38 | values = random.choices(range(256), k=256) |
Louis Verhaard | 0b8268a | 2020-08-05 16:11:29 +0200 | [diff] [blame] | 39 | lut_tensor = create_const_tensor( |
| 40 | op.name + "_lut", [1, 1, 1, 256], DataType.int8, values, np.uint8, TensorPurpose.LUT |
| 41 | ) |
Tim Hall | d8339a7 | 2021-05-27 18:49:40 +0100 | [diff] [blame] | 42 | scratch_lut_tensor = lut_tensor.clone_into_fast_storage(arch) |
| 43 | op.set_activation_lut(scratch_lut_tensor) |
Louis Verhaard | 0b8268a | 2020-08-05 16:11:29 +0200 | [diff] [blame] | 44 | |
| 45 | |
Tim Hall | d8339a7 | 2021-05-27 18:49:40 +0100 | [diff] [blame] | 46 | def set_1K_lut(op, key, arch): |
Jacob Bohlin | 1a66697 | 2020-09-11 10:04:15 +0200 | [diff] [blame] | 47 | random.seed(key) |
| 48 | values = random.choices(range(256), k=256) |
Louis Verhaard | 0b8268a | 2020-08-05 16:11:29 +0200 | [diff] [blame] | 49 | lut_tensor = create_const_tensor( |
| 50 | op.name + "_lut", [1, 1, 1, 256], DataType.int32, values, np.uint32, TensorPurpose.LUT |
| 51 | ) |
Tim Hall | d8339a7 | 2021-05-27 18:49:40 +0100 | [diff] [blame] | 52 | scratch_lut_tensor = lut_tensor.clone_into_fast_storage(arch) |
| 53 | op.set_activation_lut(scratch_lut_tensor) |
Louis Verhaard | 0b8268a | 2020-08-05 16:11:29 +0200 | [diff] [blame] | 54 | |
| 55 | |
Tim Hall | d8339a7 | 2021-05-27 18:49:40 +0100 | [diff] [blame] | 56 | def set_2K_lut(op, key, arch): |
Jacob Bohlin | 1a66697 | 2020-09-11 10:04:15 +0200 | [diff] [blame] | 57 | random.seed(key) |
| 58 | values = random.choices(range(512), k=512) |
Louis Verhaard | 0b8268a | 2020-08-05 16:11:29 +0200 | [diff] [blame] | 59 | lut_tensor = create_const_tensor( |
| 60 | op.name + "_lut", [1, 1, 1, 512], DataType.int32, values, np.uint32, TensorPurpose.LUT |
| 61 | ) |
Tim Hall | d8339a7 | 2021-05-27 18:49:40 +0100 | [diff] [blame] | 62 | scratch_lut_tensor = lut_tensor.clone_into_fast_storage(arch) |
| 63 | op.set_activation_lut(scratch_lut_tensor) |
Louis Verhaard | 0b8268a | 2020-08-05 16:11:29 +0200 | [diff] [blame] | 64 | |
| 65 | |
| 66 | def process(arch, op_list): |
| 67 | # Returns subgraph with given operations |
| 68 | nng = Graph() |
| 69 | sg = testutil.create_subgraph(op_list) |
| 70 | nng.subgraphs.append(sg) |
| 71 | assert verify_graph_health(nng) |
| 72 | nng = mark_tensors.mark_tensor_purpose(nng, arch, False) |
| 73 | assert verify_graph_health(nng) |
Tim Hall | d8339a7 | 2021-05-27 18:49:40 +0100 | [diff] [blame] | 74 | rewrite_graph_pre_order(nng, sg, arch, [], []) |
Louis Verhaard | 0b8268a | 2020-08-05 16:11:29 +0200 | [diff] [blame] | 75 | pass_packing.pack_into_passes(nng, arch, False) |
| 76 | assert verify_graph_health(nng) |
| 77 | # Create a DMA instruction for every op |
| 78 | cmd_list = [] |
| 79 | for ps in sg.passes: |
Tim Hall | d8339a7 | 2021-05-27 18:49:40 +0100 | [diff] [blame] | 80 | for input_tens in ps.inputs: |
| 81 | if input_tens.src_tensor: |
| 82 | cmd_list.append(DMA(ps, input_tens.src_tensor, input_tens, None)) |
| 83 | |
Louis Verhaard | 0b8268a | 2020-08-05 16:11:29 +0200 | [diff] [blame] | 84 | sg.high_level_command_stream = cmd_list |
| 85 | return sg |
| 86 | |
| 87 | |
Patrik Gustavsson | e5cf95b | 2020-09-03 16:39:52 +0200 | [diff] [blame] | 88 | def filter_lut_cmds(cmd_list): |
| 89 | lut_cmd_list = [] |
| 90 | for cmd in cmd_list: |
| 91 | if "lut" in cmd.in_tensor.name: |
| 92 | lut_cmd_list.append(cmd) |
| 93 | return lut_cmd_list |
| 94 | |
| 95 | |
Louis Verhaard | 0b8268a | 2020-08-05 16:11:29 +0200 | [diff] [blame] | 96 | def test_optimize_high_level_cmd_stream_2K(): |
| 97 | # Tests lut.optimize_high_level_cmd_stream, blending 256 byte and 2K luts |
| 98 | arch = testutil.create_arch() |
| 99 | shape = [1, 1, 1, 1] |
| 100 | # u8 LUT op, should lead to DMA |
Louis Verhaard | aee5d75 | 2020-09-30 09:01:52 +0200 | [diff] [blame] | 101 | op0 = testutil.create_elemwise_op(Op.Add, "op0", shape, shape, shape) |
Tim Hall | d8339a7 | 2021-05-27 18:49:40 +0100 | [diff] [blame] | 102 | set_256_lut(op0, "lut0", arch) |
Louis Verhaard | 0b8268a | 2020-08-05 16:11:29 +0200 | [diff] [blame] | 103 | # u8 LUT op, should lead to DMA |
Louis Verhaard | aee5d75 | 2020-09-30 09:01:52 +0200 | [diff] [blame] | 104 | op1 = testutil.create_elemwise_op(Op.Add, "op1", shape, shape, shape) |
Tim Hall | d8339a7 | 2021-05-27 18:49:40 +0100 | [diff] [blame] | 105 | set_256_lut(op1, "lut1", arch) |
Louis Verhaard | 0b8268a | 2020-08-05 16:11:29 +0200 | [diff] [blame] | 106 | # u8 LUT op with different LUT, should lead to DMA |
Louis Verhaard | aee5d75 | 2020-09-30 09:01:52 +0200 | [diff] [blame] | 107 | op2 = testutil.create_elemwise_op(Op.Add, "op2", shape, shape, shape) |
Tim Hall | d8339a7 | 2021-05-27 18:49:40 +0100 | [diff] [blame] | 108 | set_256_lut(op2, "lut2", arch) |
Louis Verhaard | 0b8268a | 2020-08-05 16:11:29 +0200 | [diff] [blame] | 109 | # u8 LUT op with same LUT as in op1, should not lead to DMA |
Louis Verhaard | aee5d75 | 2020-09-30 09:01:52 +0200 | [diff] [blame] | 110 | op3 = testutil.create_elemwise_op(Op.Add, "op3", shape, shape, shape) |
Tim Hall | d8339a7 | 2021-05-27 18:49:40 +0100 | [diff] [blame] | 111 | set_256_lut(op3, "lut1", arch) |
Louis Verhaard | 0b8268a | 2020-08-05 16:11:29 +0200 | [diff] [blame] | 112 | # u8 LUT op with same LUT as in op2, should not lead to DMA |
Louis Verhaard | aee5d75 | 2020-09-30 09:01:52 +0200 | [diff] [blame] | 113 | op4 = testutil.create_elemwise_op(Op.Add, "op4", shape, shape, shape) |
Tim Hall | d8339a7 | 2021-05-27 18:49:40 +0100 | [diff] [blame] | 114 | set_256_lut(op4, "lut2", arch) |
Louis Verhaard | 0b8268a | 2020-08-05 16:11:29 +0200 | [diff] [blame] | 115 | # 2K LUT op, should lead to DMA, and will overwrite all previous LUTs in SHRAM |
Louis Verhaard | aee5d75 | 2020-09-30 09:01:52 +0200 | [diff] [blame] | 116 | op5_2K = testutil.create_elemwise_op(Op.Add, "op5", shape, shape, shape) |
Tim Hall | d8339a7 | 2021-05-27 18:49:40 +0100 | [diff] [blame] | 117 | set_2K_lut(op5_2K, "lut5", arch) |
Louis Verhaard | 0b8268a | 2020-08-05 16:11:29 +0200 | [diff] [blame] | 118 | # Another 2K LUT op, should lead to DMA, and will overwrite the previous LUT in SHRAM |
Louis Verhaard | aee5d75 | 2020-09-30 09:01:52 +0200 | [diff] [blame] | 119 | op6_2K = testutil.create_elemwise_op(Op.Add, "op6", shape, shape, shape) |
Tim Hall | d8339a7 | 2021-05-27 18:49:40 +0100 | [diff] [blame] | 120 | set_2K_lut(op6_2K, "lut6", arch) |
Louis Verhaard | 0b8268a | 2020-08-05 16:11:29 +0200 | [diff] [blame] | 121 | # u8 LUT op with same LUT as in op1, should lead to DMA |
Louis Verhaard | aee5d75 | 2020-09-30 09:01:52 +0200 | [diff] [blame] | 122 | op7 = testutil.create_elemwise_op(Op.Add, "op7", shape, shape, shape) |
Tim Hall | d8339a7 | 2021-05-27 18:49:40 +0100 | [diff] [blame] | 123 | set_256_lut(op7, "lut1", arch) |
Louis Verhaard | 0b8268a | 2020-08-05 16:11:29 +0200 | [diff] [blame] | 124 | |
| 125 | op_list = [op0, op1, op2, op3, op4, op5_2K, op6_2K, op7] |
| 126 | sg = process(arch, op_list) |
| 127 | orig_cmd_list = sg.high_level_command_stream |
| 128 | sg.high_level_command_stream = orig_cmd_list |
| 129 | lut.optimize_high_level_cmd_stream(sg, arch) |
| 130 | cmd_list = sg.high_level_command_stream |
| 131 | # Check that only the needed DMA commands are left |
| 132 | expected_dma_ops = [op0, op1, op2, op5_2K, op6_2K, op7] |
Patrik Gustavsson | e5cf95b | 2020-09-03 16:39:52 +0200 | [diff] [blame] | 133 | |
| 134 | cmd_list = filter_lut_cmds(cmd_list) |
| 135 | orig_cmd_list = filter_lut_cmds(orig_cmd_list) |
| 136 | |
Louis Verhaard | 0b8268a | 2020-08-05 16:11:29 +0200 | [diff] [blame] | 137 | for (cmd, op) in zip(cmd_list, expected_dma_ops): |
Tim Hall | d8339a7 | 2021-05-27 18:49:40 +0100 | [diff] [blame] | 138 | assert cmd.in_tensor == op.activation_lut.src_tensor |
Louis Verhaard | 0b8268a | 2020-08-05 16:11:29 +0200 | [diff] [blame] | 139 | # Check that lut0, lut1 and lut2 in op0, op1, op2 are stored on different addresses |
| 140 | assert orig_cmd_list[0].out_tensor.address != orig_cmd_list[1].out_tensor.address |
| 141 | assert orig_cmd_list[0].out_tensor.address != orig_cmd_list[2].out_tensor.address |
| 142 | assert orig_cmd_list[1].out_tensor.address != orig_cmd_list[2].out_tensor.address |
| 143 | # Check that lut1 in op1 and op3 have same address |
| 144 | assert orig_cmd_list[1].out_tensor.address == orig_cmd_list[3].out_tensor.address |
| 145 | # Check that lut2 in op2 and op4 have same address |
| 146 | assert orig_cmd_list[2].out_tensor.address == orig_cmd_list[4].out_tensor.address |
| 147 | # Check that lut-s for 16 bit (op5 and op6) are stored on same address |
| 148 | assert orig_cmd_list[5].out_tensor.address == orig_cmd_list[6].out_tensor.address |
| 149 | |
| 150 | |
| 151 | def test_optimize_high_level_cmd_stream_1K(): |
| 152 | # Tests lut.optimize_high_level_cmd_stream, blending 256 and 1K luts |
| 153 | arch = testutil.create_arch() |
| 154 | shape = [1, 1, 1, 1] |
| 155 | # u8 LUT op, should lead to DMA |
Louis Verhaard | aee5d75 | 2020-09-30 09:01:52 +0200 | [diff] [blame] | 156 | op0 = testutil.create_elemwise_op(Op.Add, "op0", shape, shape, shape) |
Tim Hall | d8339a7 | 2021-05-27 18:49:40 +0100 | [diff] [blame] | 157 | set_256_lut(op0, "lut0", arch) |
Louis Verhaard | 0b8268a | 2020-08-05 16:11:29 +0200 | [diff] [blame] | 158 | # u8 LUT op, should lead to DMA |
Louis Verhaard | aee5d75 | 2020-09-30 09:01:52 +0200 | [diff] [blame] | 159 | op1 = testutil.create_elemwise_op(Op.Add, "op1", shape, shape, shape) |
Tim Hall | d8339a7 | 2021-05-27 18:49:40 +0100 | [diff] [blame] | 160 | set_256_lut(op1, "lut1", arch) |
Louis Verhaard | 0b8268a | 2020-08-05 16:11:29 +0200 | [diff] [blame] | 161 | # 1K LUT op with different LUT, should lead to DMA |
Louis Verhaard | aee5d75 | 2020-09-30 09:01:52 +0200 | [diff] [blame] | 162 | op2_1K = testutil.create_elemwise_op(Op.Add, "op2", shape, shape, shape) |
Tim Hall | d8339a7 | 2021-05-27 18:49:40 +0100 | [diff] [blame] | 163 | set_1K_lut(op2_1K, "lut2", arch) |
Louis Verhaard | 0b8268a | 2020-08-05 16:11:29 +0200 | [diff] [blame] | 164 | # u8 LUT op with same LUT as in op1, should not lead to DMA |
Louis Verhaard | aee5d75 | 2020-09-30 09:01:52 +0200 | [diff] [blame] | 165 | op3 = testutil.create_elemwise_op(Op.Add, "op3", shape, shape, shape) |
Tim Hall | d8339a7 | 2021-05-27 18:49:40 +0100 | [diff] [blame] | 166 | set_256_lut(op3, "lut1", arch) |
Louis Verhaard | 0b8268a | 2020-08-05 16:11:29 +0200 | [diff] [blame] | 167 | # 1K LUT op with same LUT as in op2, should not lead to DMA |
Louis Verhaard | aee5d75 | 2020-09-30 09:01:52 +0200 | [diff] [blame] | 168 | op4_1K = testutil.create_elemwise_op(Op.Add, "op4", shape, shape, shape) |
Tim Hall | d8339a7 | 2021-05-27 18:49:40 +0100 | [diff] [blame] | 169 | set_1K_lut(op4_1K, "lut2", arch) |
Louis Verhaard | 0b8268a | 2020-08-05 16:11:29 +0200 | [diff] [blame] | 170 | # 1K LUT op, should lead to DMA, and will overwrite lut2 |
Louis Verhaard | aee5d75 | 2020-09-30 09:01:52 +0200 | [diff] [blame] | 171 | op5_2K = testutil.create_elemwise_op(Op.Add, "op5", shape, shape, shape) |
Tim Hall | d8339a7 | 2021-05-27 18:49:40 +0100 | [diff] [blame] | 172 | set_1K_lut(op5_2K, "lut5", arch) |
Louis Verhaard | 0b8268a | 2020-08-05 16:11:29 +0200 | [diff] [blame] | 173 | # u8 LUT op, lut0 should still be present, should not lead to DMA |
Louis Verhaard | aee5d75 | 2020-09-30 09:01:52 +0200 | [diff] [blame] | 174 | op6 = testutil.create_elemwise_op(Op.Add, "op6", shape, shape, shape) |
Tim Hall | d8339a7 | 2021-05-27 18:49:40 +0100 | [diff] [blame] | 175 | set_256_lut(op6, "lut0", arch) |
Louis Verhaard | 0b8268a | 2020-08-05 16:11:29 +0200 | [diff] [blame] | 176 | # 1K LUT op with same LUT as in op2, should lead to DMA |
Louis Verhaard | aee5d75 | 2020-09-30 09:01:52 +0200 | [diff] [blame] | 177 | op7 = testutil.create_elemwise_op(Op.Add, "op7", shape, shape, shape) |
Tim Hall | d8339a7 | 2021-05-27 18:49:40 +0100 | [diff] [blame] | 178 | set_1K_lut(op7, "lut2", arch) |
Louis Verhaard | 0b8268a | 2020-08-05 16:11:29 +0200 | [diff] [blame] | 179 | |
| 180 | op_list = [op0, op1, op2_1K, op3, op4_1K, op5_2K, op6, op7] |
| 181 | sg = process(arch, op_list) |
| 182 | orig_cmd_list = sg.high_level_command_stream |
| 183 | sg.high_level_command_stream = orig_cmd_list |
| 184 | lut.optimize_high_level_cmd_stream(sg, arch) |
| 185 | cmd_list = sg.high_level_command_stream |
Patrik Gustavsson | e5cf95b | 2020-09-03 16:39:52 +0200 | [diff] [blame] | 186 | |
| 187 | cmd_list = filter_lut_cmds(cmd_list) |
| 188 | orig_cmd_list = filter_lut_cmds(orig_cmd_list) |
| 189 | |
Louis Verhaard | 0b8268a | 2020-08-05 16:11:29 +0200 | [diff] [blame] | 190 | # Check that only the needed DMA commands are left |
| 191 | expected_dma_ops = [op0, op1, op2_1K, op5_2K, op7] |
| 192 | for (cmd, op) in zip(cmd_list, expected_dma_ops): |
Tim Hall | d8339a7 | 2021-05-27 18:49:40 +0100 | [diff] [blame] | 193 | assert cmd.in_tensor == op.activation_lut.src_tensor |
Louis Verhaard | 0b8268a | 2020-08-05 16:11:29 +0200 | [diff] [blame] | 194 | # Check that lut0, lut1 and lut2 in op0, op1, op2 are stored on different addresses |
| 195 | assert orig_cmd_list[0].out_tensor.address != orig_cmd_list[1].out_tensor.address |
| 196 | assert orig_cmd_list[0].out_tensor.address != orig_cmd_list[2].out_tensor.address |
| 197 | assert orig_cmd_list[1].out_tensor.address != orig_cmd_list[2].out_tensor.address |
| 198 | # Check that lut1 in op1 and op3 have same address |
| 199 | assert orig_cmd_list[1].out_tensor.address == orig_cmd_list[3].out_tensor.address |
| 200 | # Check that lut2 in op2 and op4 and op7 have same address |
| 201 | assert orig_cmd_list[2].out_tensor.address == orig_cmd_list[4].out_tensor.address |
| 202 | assert orig_cmd_list[2].out_tensor.address == orig_cmd_list[7].out_tensor.address |