-
Notifications
You must be signed in to change notification settings - Fork 74
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
- Loading branch information
1 parent
5907587
commit b81a355
Showing
4 changed files
with
112 additions
and
91 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
1 change: 1 addition & 0 deletions
1
tests/tt_metal/tt_metal/dispatch/dispatch_buffer/CMakeLists.txt
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
110 changes: 110 additions & 0 deletions
110
tests/tt_metal/tt_metal/dispatch/dispatch_buffer/test_sub_device.cpp
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,110 @@ | ||
| // SPDX-FileCopyrightText: © 2024 Tenstorrent Inc. | ||
| // | ||
| // SPDX-License-Identifier: Apache-2.0 | ||
|
|
||
| #include <cstddef> | ||
| #include <cstdint> | ||
| #include <array> | ||
| #include <tuple> | ||
| #include <vector> | ||
|
|
||
| #include "gtest/gtest.h" | ||
| #include "tt_metal/common/core_coord.hpp" | ||
| #include "tt_metal/impl/buffers/global_semaphore.hpp" | ||
| #include "tt_metal/impl/device/device.hpp" | ||
| #include "tt_metal/impl/event/event.hpp" | ||
| #include "tt_metal/impl/sub_device/sub_device.hpp" | ||
| #include "test_utils/stimulus.hpp" | ||
| #include "command_queue_fixture.hpp" | ||
| #include "command_queue_test_utils.hpp" | ||
| #include "sub_device_test_utils.hpp" | ||
|
|
||
| TEST_F(CommandQueueSingleCardFixture, TensixTestSubDeviceAllocations) { | ||
| uint32_t local_l1_size = 3200; | ||
| SubDevice sub_device_1(std::array{CoreRangeSet(CoreRange({0, 0}, {2, 2}))}); | ||
| SubDevice sub_device_2(std::array{CoreRangeSet(std::vector{CoreRange({3, 3}, {3, 3}), CoreRange({4, 4}, {4, 4})})}); | ||
| CoreRangeSet sharded_cores_1 = CoreRange({0, 0}, {2, 2}); | ||
| CoreRangeSet sharded_cores_2 = CoreRange({4, 4}, {4, 4}); | ||
|
|
||
| auto sharded_cores_1_vec = corerange_to_cores(sharded_cores_1, std::nullopt, true); | ||
| auto sharded_cores_2_vec = corerange_to_cores(sharded_cores_2, std::nullopt, true); | ||
|
|
||
| ShardSpecBuffer shard_spec_buffer_1 = ShardSpecBuffer(sharded_cores_1, {1, 1}, ShardOrientation::ROW_MAJOR, false, {1, 1}, {sharded_cores_1.num_cores(), 1}); | ||
| uint32_t page_size_1 = 32; | ||
| ShardedBufferConfig shard_config_1 = {nullptr, sharded_cores_1.num_cores() * page_size_1, page_size_1, BufferType::L1, TensorMemoryLayout::HEIGHT_SHARDED, shard_spec_buffer_1}; | ||
| auto input_1 = tt::test_utils::generate_uniform_random_vector<uint32_t>(0, 100, shard_config_1.size / sizeof(uint32_t)); | ||
|
|
||
| ShardSpecBuffer shard_spec_buffer_2 = ShardSpecBuffer(sharded_cores_2, {1, 1}, ShardOrientation::ROW_MAJOR, false, {1, 1}, {sharded_cores_2.num_cores(), 1}); | ||
| uint32_t page_size_2 = 64; | ||
| ShardedBufferConfig shard_config_2 = {nullptr, sharded_cores_2.num_cores() * page_size_2, page_size_2, BufferType::L1, TensorMemoryLayout::HEIGHT_SHARDED, shard_spec_buffer_2}; | ||
| auto input_2 = tt::test_utils::generate_uniform_random_vector<uint32_t>(0, 100, shard_config_2.size / sizeof(uint32_t)); | ||
|
|
||
| uint32_t page_size_3 = 1024; | ||
| InterleavedBufferConfig interleaved_config = {nullptr, page_size_3, page_size_3, BufferType::L1, TensorMemoryLayout::INTERLEAVED}; | ||
| auto input_3 = tt::test_utils::generate_uniform_random_vector<uint32_t>(0, 100, interleaved_config.size / sizeof(uint32_t)); | ||
|
|
||
| for (Device *device : devices_) { | ||
| auto sub_device_manager_1 = device->create_sub_device_manager({sub_device_1}, local_l1_size); | ||
| auto sub_device_manager_2 = device->create_sub_device_manager({sub_device_1, sub_device_2}, local_l1_size); | ||
| DeviceAddr l1_unreserved_base = device->get_base_allocator_addr(HalMemType::L1); | ||
| DeviceAddr max_addr = l1_unreserved_base + local_l1_size; | ||
|
|
||
| shard_config_1.device = device; | ||
| shard_config_2.device = device; | ||
| interleaved_config.device = device; | ||
|
|
||
| std::vector<CoreCoord> physical_cores_1; | ||
| physical_cores_1.reserve(sharded_cores_1_vec.size()); | ||
| for (const auto& core : sharded_cores_1_vec) { | ||
| physical_cores_1.push_back(device->worker_core_from_logical_core(core)); | ||
| } | ||
|
|
||
| std::vector<CoreCoord> physical_cores_2; | ||
| physical_cores_2.reserve(sharded_cores_2_vec.size()); | ||
| for (const auto& core : sharded_cores_2_vec) { | ||
| physical_cores_2.push_back(device->worker_core_from_logical_core(core)); | ||
| } | ||
|
|
||
| device->load_sub_device_manager(sub_device_manager_1); | ||
|
|
||
| auto buffer_1 = CreateBuffer(shard_config_1, SubDeviceId{0}); | ||
| EXPECT_EQ(buffer_1->address(), max_addr - page_size_1); | ||
| EnqueueWriteBuffer(device->command_queue(), buffer_1, input_1, false); | ||
| std::vector<uint32_t> output_1; | ||
| EnqueueReadBuffer(device->command_queue(), buffer_1, output_1, true); | ||
| EXPECT_EQ(input_1, output_1); | ||
| auto input_1_it = input_1.begin(); | ||
| for (const auto& physical_core : physical_cores_1) { | ||
| auto readback = tt::llrt::read_hex_vec_from_core( | ||
| device->id(), physical_core, buffer_1->address(), page_size_1); | ||
| EXPECT_TRUE(std::equal(input_1_it, input_1_it + page_size_1 / sizeof(uint32_t), readback.begin())); | ||
| input_1_it += page_size_1 / sizeof(uint32_t); | ||
| } | ||
|
|
||
| auto buffer_2 = CreateBuffer(interleaved_config); | ||
| EXPECT_THROW(CreateBuffer(shard_config_1, SubDeviceId{1}), std::exception); | ||
| EXPECT_THROW(device->clear_loaded_sub_device_manager(), std::exception); | ||
| EXPECT_THROW(device->load_sub_device_manager(sub_device_manager_2), std::exception); | ||
| DeallocateBuffer(*buffer_1); | ||
| device->clear_loaded_sub_device_manager(); | ||
| device->load_sub_device_manager(sub_device_manager_2); | ||
|
|
||
| auto buffer_3 = CreateBuffer(shard_config_2, SubDeviceId{1}); | ||
| EXPECT_EQ(buffer_3->address(), max_addr - page_size_2); | ||
| EnqueueWriteBuffer(device->command_queue(), buffer_3, input_2, false); | ||
| std::vector<uint32_t> output_2; | ||
| EnqueueReadBuffer(device->command_queue(), buffer_3, output_2, true); | ||
| EXPECT_EQ(input_2, output_2); | ||
| auto input_2_it = input_2.begin(); | ||
| for (const auto& physical_core : physical_cores_2) { | ||
| auto readback = tt::llrt::read_hex_vec_from_core( | ||
| device->id(), physical_core, buffer_3->address(), page_size_2); | ||
| EXPECT_TRUE(std::equal(input_2_it, input_2_it + page_size_2 / sizeof(uint32_t), readback.begin())); | ||
| input_2_it += page_size_2 / sizeof(uint32_t); | ||
| } | ||
|
|
||
| auto buffer_4 = CreateBuffer(shard_config_1, SubDeviceId{0}); | ||
| EXPECT_EQ(buffer_4->address(), max_addr - page_size_1); | ||
| EXPECT_THROW(CreateBuffer(interleaved_config, SubDeviceId{0}), std::exception); | ||
| } | ||
| } |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters