Fix parsing cluster descriptor for Blackhole and Grayskull(#280)

---------

Co-authored-by: Almeet Bhullar <abhullar@tenstorrent.com>

device/hugepage.h

@@ -27,4 +27,4 @@ std::string find_hugepage_dir(std::size_t pagesize);
 // Today we assume there's only one pipeline running within the system.
 // One hugepage per device such that each device gets unique memory.
 int open_hugepage_file(const std::string &dir, chip_id_t physical_device_id, uint16_t channel);
-}
+}

device/libs/create_ethernet_map.h

@@ -1,3 +1,9 @@
+/*
+ * SPDX-FileCopyrightText: (c) 2024 Tenstorrent Inc.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
 // External function to create the cluster descriptor yaml file.
 extern "C" {
 int create_ethernet_map(char *file);

device/tt_cluster_descriptor.cpp

@@ -523,16 +523,21 @@ void tt_ClusterDescriptor::load_ethernet_connections_from_connectivity_descripto
 }
 
 void tt_ClusterDescriptor::load_chips_from_connectivity_descriptor(YAML::Node &yaml, tt_ClusterDescriptor &desc) {
+
+    for (YAML::const_iterator node = yaml["arch"].begin(); node != yaml["arch"].end(); ++node) {
+        chip_id_t chip_id = node->first.as<int>();
+        desc.all_chips.insert(chip_id);
+    }
+
     for (YAML::const_iterator node = yaml["chips"].begin(); node != yaml["chips"].end(); ++node) {
         chip_id_t chip_id = node->first.as<int>();
         std::vector<int> chip_rack_coords = node->second.as<std::vector<int>>();
         log_assert(chip_rack_coords.size() == 4, "Galaxy (x, y, rack, shelf) coords must be size 4");
         eth_coord_t chip_location{
             chip_rack_coords.at(0), chip_rack_coords.at(1), chip_rack_coords.at(2), chip_rack_coords.at(3)};
-        
+
         desc.chip_locations.insert({chip_id, chip_location});
         desc.coords_to_chip_ids[std::get<2>(chip_location)][std::get<3>(chip_location)][std::get<1>(chip_location)][std::get<0>(chip_location)] = chip_id;
-        desc.all_chips.insert(chip_id);
     }
 
     for(const auto& chip : yaml["chips_with_mmio"]) {
@@ -558,7 +563,7 @@ void tt_ClusterDescriptor::load_chips_from_connectivity_descriptor(YAML::Node &y
             std::get<3>(chip_location));
     }
 
-		if (yaml["boardtype"]) {
+	if (yaml["boardtype"]) {
         for (const auto& chip_board_type : yaml["boardtype"].as<std::map<int, std::string>>()) {
             auto &chip = chip_board_type.first;
             BoardType board_type;
@@ -568,7 +573,13 @@ void tt_ClusterDescriptor::load_chips_from_connectivity_descriptor(YAML::Node &y
                 board_type = BoardType::N300;
             } else if (chip_board_type.second == "GALAXY") {
                 board_type = BoardType::GALAXY;
+            } else if (chip_board_type.second == "e150") {
+                board_type = BoardType::E150;
+            }
+            else if (chip_board_type.second == "p150A") {
+                board_type = BoardType::P150A;
             } else {
+                log_warning(LogSiliconDriver, "Unknown board type for chip {}. This might happen because chip is running old firmware. Defaulting to DEFAULT", chip);
                 board_type = BoardType::DEFAULT;
             }
             desc.chip_board_type.insert({chip, board_type});
@@ -622,7 +633,7 @@ const std::unordered_map<chip_id_t, std::unordered_map<ethernet_channel_t, std::
 
 const std::unordered_map<chip_id_t, eth_coord_t>& tt_ClusterDescriptor::get_chip_locations() const {
     static auto locations = std::unordered_map<chip_id_t, eth_coord_t>();
-    if (locations.empty()) {
+    if (locations.empty() and !this->chip_locations.empty()) {
         for (auto chip_id : this->enabled_active_chips) {
             locations[chip_id] = chip_locations.at(chip_id);
         }
@@ -632,6 +643,7 @@ const std::unordered_map<chip_id_t, eth_coord_t>& tt_ClusterDescriptor::get_chip
 }
 
 chip_id_t tt_ClusterDescriptor::get_shelf_local_physical_chip_coords(chip_id_t virtual_coord) {
+    log_assert(!this->chip_locations.empty(), "Getting physical chip coordinates is only valid for systems where chips have coordinates");
     // Physical cooridnates of chip inside a single rack. Calculated based on Galaxy topology.
     // See: https://yyz-gitlab.local.tenstorrent.com/tenstorrent/budabackend/-/wikis/uploads/23e7a5168f38dfb706f9887fde78cb03/image.png
     int x = std::get<0>(get_chip_locations().at(virtual_coord));
@@ -667,6 +679,7 @@ const std::unordered_map<chip_id_t, bool>& tt_ClusterDescriptor::get_noc_transla
 std::size_t tt_ClusterDescriptor::get_number_of_chips() const { return this->enabled_active_chips.size(); }
 
 int tt_ClusterDescriptor::get_ethernet_link_distance(chip_id_t chip_a, chip_id_t chip_b) const {
+    log_assert(!this->chip_locations.empty(), "Getting physical chip coordinates is only valid for systems where chips have coordinates");
     return this->get_ethernet_link_coord_distance(chip_locations.at(chip_a), chip_locations.at(chip_b));
 }
 

device/tt_cluster_descriptor.h

@@ -25,8 +25,10 @@ namespace YAML { class Node; }
 enum BoardType : uint32_t {
     N150 = 0,
     N300 = 1,
-    GALAXY = 2,
-    DEFAULT = 3,
+    E150 = 2,
+    P150A = 3,
+    GALAXY = 4,
+    DEFAULT = 5,
 };
 
 class tt_ClusterDescriptor {
@@ -88,9 +90,13 @@ class tt_ClusterDescriptor {
   // get_cluster_descriptor_file_path will create ethernet map in the background.
   static std::string get_cluster_descriptor_file_path();
   static std::unique_ptr<tt_ClusterDescriptor> create_from_yaml(const std::string &cluster_descriptor_file_path);
+
+  // TODO: This function is used to create mock cluster descriptor yaml files, for example for simulation.
+  // The name of the function is kept to not gate the changes regarding create-ethernet-map.
+  // It should be renamed to something like create_mock_cluster_descriptor and changed in tt-metal/tt-debuda.
   static std::unique_ptr<tt_ClusterDescriptor> create_for_grayskull_cluster(
-      const std::set<chip_id_t> &logical_mmio_device_ids,
-      const std::vector<chip_id_t> &physical_mmio_device_ids);
+    const std::set<chip_id_t> &logical_mmio_device_ids,
+    const std::vector<chip_id_t> &physical_mmio_device_ids);
 
   const std::unordered_map<chip_id_t, std::uint32_t>& get_harvesting_info() const;
   const std::unordered_map<chip_id_t, bool>& get_noc_translation_table_en() const;

device/tt_silicon_driver.cpp

@@ -302,14 +302,14 @@ tt_SiliconDevice::tt_SiliconDevice(const std::string &sdesc_path, const std::str
         log_info(LogSiliconDriver, "Detected {} PCI device{} : {}", m_num_pci_devices, (m_num_pci_devices > 1) ? "s":"", available_device_ids);
         log_debug(LogSiliconDriver, "Passed target devices: {}", target_devices);
     }
-
-    if (ndesc_path == "") {
-        ndesc = tt_ClusterDescriptor::create_for_grayskull_cluster(target_devices, available_device_ids);
-    }
-    else {
-        ndesc = tt_ClusterDescriptor::create_from_yaml(ndesc_path);
+
+    std::string cluster_descriptor_path = ndesc_path;
+    if (cluster_descriptor_path == "") {
+        cluster_descriptor_path = tt_ClusterDescriptor::get_cluster_descriptor_file_path();
     }
 
+    ndesc = tt_ClusterDescriptor::create_from_yaml(cluster_descriptor_path);
+
     for (auto &d: target_devices){
         if (ndesc->is_chip_mmio_capable(d)){
             target_mmio_device_ids.insert(d);

tests/api/test_chip.cpp

@@ -47,16 +47,9 @@ inline std::unique_ptr<tt_ClusterDescriptor> get_cluster_desc() {
 
     // TODO: Remove different branch for different archs
     std::unique_ptr<tt_ClusterDescriptor> cluster_desc;
-    if (device_arch == tt::ARCH::GRAYSKULL) {
-        cluster_desc = tt_ClusterDescriptor::create_for_grayskull_cluster(pci_device_ids_set, pci_device_ids);
-    } else if (device_arch == tt::ARCH::BLACKHOLE) {
-        std::string yaml_path = test_utils::GetAbsPath("blackhole_1chip_cluster.yaml");
-        cluster_desc = tt_ClusterDescriptor::create_from_yaml(yaml_path);
-    } else {
-        // TODO: remove getting manually cluster descriptor from yaml.
-        std::string yaml_path = tt_ClusterDescriptor::get_cluster_descriptor_file_path();
-        cluster_desc = tt_ClusterDescriptor::create_from_yaml(yaml_path);
-    }
+    // TODO: remove getting manually cluster descriptor from yaml.
+    std::string yaml_path = tt_ClusterDescriptor::get_cluster_descriptor_file_path();
+    cluster_desc = tt_ClusterDescriptor::create_from_yaml(yaml_path);
 
     return cluster_desc;
 }
@@ -116,7 +109,7 @@ inline std::unique_ptr<Cluster> get_cluster() {
 
 
     // TODO: Don't pass each of these arguments.
-    return std::unique_ptr<Cluster>(new Cluster(soc_path, device_arch == tt::ARCH::GRAYSKULL ? "" : yaml_path, detected_num_chips_set));
+    return std::unique_ptr<Cluster>(new Cluster(soc_path, tt_ClusterDescriptor::get_cluster_descriptor_file_path(), detected_num_chips_set));
 }
 
 // TODO: Once default auto TLB setup is in, check it is setup properly.

tests/api/test_cluster.cpp

@@ -56,18 +56,10 @@ inline std::unique_ptr<tt_ClusterDescriptor> get_cluster_desc() {
         return nullptr;
     }
 
-    // TODO: Remove different branch for different archs
     std::unique_ptr<tt_ClusterDescriptor> cluster_desc;
-    if (device_arch == tt::ARCH::GRAYSKULL) {
-        cluster_desc = tt_ClusterDescriptor::create_for_grayskull_cluster(pci_device_ids_set, pci_device_ids);
-    } else if (device_arch == tt::ARCH::BLACKHOLE) {
-        std::string yaml_path = test_utils::GetAbsPath("blackhole_1chip_cluster.yaml");
-        cluster_desc = tt_ClusterDescriptor::create_from_yaml(yaml_path);
-    } else {
-        // TODO: remove getting manually cluster descriptor from yaml.
-        std::string yaml_path = tt_ClusterDescriptor::get_cluster_descriptor_file_path();
-        cluster_desc = tt_ClusterDescriptor::create_from_yaml(yaml_path);
-    }
+    // TODO: remove getting manually cluster descriptor from yaml.
+    std::string yaml_path = tt_ClusterDescriptor::get_cluster_descriptor_file_path();
+    cluster_desc = tt_ClusterDescriptor::create_from_yaml(yaml_path);
 
     return cluster_desc;
 }
@@ -126,7 +118,7 @@ inline std::unique_ptr<Cluster> get_cluster() {
 
     // TODO: Don't pass each of these arguments.
     return std::unique_ptr<Cluster>(
-        new Cluster(soc_path, device_arch == tt::ARCH::GRAYSKULL ? "" : yaml_path, detected_num_chips_set));
+        new Cluster(soc_path, tt_ClusterDescriptor::get_cluster_descriptor_file_path(), detected_num_chips_set));
 }
 
 // TODO: Should not be wormhole specific.

tests/api/test_cluster_descriptor.cpp

@@ -38,16 +38,9 @@ inline std::unique_ptr<tt_ClusterDescriptor> get_cluster_desc() {
 
     // TODO: Remove different branch for different archs
     std::unique_ptr<tt_ClusterDescriptor> cluster_desc;
-    if (device_arch == tt::ARCH::GRAYSKULL) {
-        cluster_desc = tt_ClusterDescriptor::create_for_grayskull_cluster(pci_device_ids_set, pci_device_ids);
-    } else if (device_arch == tt::ARCH::BLACKHOLE) {
-        std::string yaml_path = test_utils::GetAbsPath("blackhole_1chip_cluster.yaml");
-        cluster_desc = tt_ClusterDescriptor::create_from_yaml(yaml_path);
-    } else {
-        // TODO: remove getting manually cluster descriptor from yaml.
-        std::string yaml_path = tt_ClusterDescriptor::get_cluster_descriptor_file_path();
-        cluster_desc = tt_ClusterDescriptor::create_from_yaml(yaml_path);
-    }
+    // TODO: remove getting manually cluster descriptor from yaml.
+    std::string yaml_path = tt_ClusterDescriptor::get_cluster_descriptor_file_path();
+    cluster_desc = tt_ClusterDescriptor::create_from_yaml(yaml_path);
 
     return cluster_desc;
 }

tests/api/test_mockup_device.cpp

@@ -55,4 +55,4 @@ TEST(ApiMockupTest, CreateDevice) {
     auto device_driver = std::make_unique<tt_MockupDevice>(get_soc_descriptor_file(arch));
 }
 
-}  // namespace test::mockup_device
+}  // namespace test::mockup_device

tests/blackhole/test_silicon_driver_bh.cpp

@@ -59,7 +59,7 @@ std::int32_t get_static_tlb_index(tt_xy_pair target) {
 
 std::set<chip_id_t> get_target_devices() {
     std::set<chip_id_t> target_devices;
-    std::unique_ptr<tt_ClusterDescriptor> cluster_desc_uniq = tt_ClusterDescriptor::create_from_yaml(test_utils::GetAbsPath("blackhole_1chip_cluster.yaml"));
+    std::unique_ptr<tt_ClusterDescriptor> cluster_desc_uniq = tt_ClusterDescriptor::create_from_yaml(tt_ClusterDescriptor::get_cluster_descriptor_file_path());
     for (int i = 0; i < cluster_desc_uniq->get_number_of_chips(); i++) {
         target_devices.insert(i);
     }
@@ -72,7 +72,7 @@ TEST(SiliconDriverBH, CreateDestroy) {
     tt_device_params default_params;
     // Initialize the driver with a 1x1 descriptor and explictly do not perform harvesting
     for(int i = 0; i < 50; i++) {
-        tt_SiliconDevice device = tt_SiliconDevice(test_utils::GetAbsPath("tests/soc_descs/blackhole_140_arch_no_eth.yaml"), test_utils::GetAbsPath("blackhole_1chip_cluster.yaml"), target_devices, num_host_mem_ch_per_mmio_device, false, true, false);
+        tt_SiliconDevice device = tt_SiliconDevice(test_utils::GetAbsPath("tests/soc_descs/blackhole_140_arch_no_eth.yaml"), tt_ClusterDescriptor::get_cluster_descriptor_file_path(), target_devices, num_host_mem_ch_per_mmio_device, false, true, false);
         set_params_for_remote_txn(device);
         device.start_device(default_params);
         device.deassert_risc_reset();
@@ -205,7 +205,7 @@ TEST(SiliconDriverBH, UnalignedStaticTLB_RW) {
 
     uint32_t num_host_mem_ch_per_mmio_device = 1;
 
-    tt_SiliconDevice device = tt_SiliconDevice(test_utils::GetAbsPath("tests/soc_descs/blackhole_140_arch_no_eth.yaml"), test_utils::GetAbsPath("blackhole_1chip_cluster.yaml"), target_devices, num_host_mem_ch_per_mmio_device, false, true, true);
+    tt_SiliconDevice device = tt_SiliconDevice(test_utils::GetAbsPath("tests/soc_descs/blackhole_140_arch_no_eth.yaml"), tt_ClusterDescriptor::get_cluster_descriptor_file_path(), target_devices, num_host_mem_ch_per_mmio_device, false, true, true);
     set_params_for_remote_txn(device);
     auto mmio_devices = device.get_target_mmio_device_ids();
 
@@ -264,7 +264,7 @@ TEST(SiliconDriverBH, StaticTLB_RW) {
 
     uint32_t num_host_mem_ch_per_mmio_device = 1;
 
-    tt_SiliconDevice device = tt_SiliconDevice(test_utils::GetAbsPath("tests/soc_descs/blackhole_140_arch_no_eth.yaml"), test_utils::GetAbsPath("blackhole_1chip_cluster.yaml"), target_devices, num_host_mem_ch_per_mmio_device, false, true, true);
+    tt_SiliconDevice device = tt_SiliconDevice(test_utils::GetAbsPath("tests/soc_descs/blackhole_140_arch_no_eth.yaml"), tt_ClusterDescriptor::get_cluster_descriptor_file_path(), target_devices, num_host_mem_ch_per_mmio_device, false, true, true);
     set_params_for_remote_txn(device);
     auto mmio_devices = device.get_target_mmio_device_ids();
 
@@ -314,7 +314,7 @@ TEST(SiliconDriverBH, DynamicTLB_RW) {
     std::set<chip_id_t> target_devices = get_target_devices();
 
     uint32_t num_host_mem_ch_per_mmio_device = 1;
-    tt_SiliconDevice device = tt_SiliconDevice(test_utils::GetAbsPath("tests/soc_descs/blackhole_140_arch_no_eth.yaml"), test_utils::GetAbsPath("blackhole_1chip_cluster.yaml"), target_devices, num_host_mem_ch_per_mmio_device, false, true, true);
+    tt_SiliconDevice device = tt_SiliconDevice(test_utils::GetAbsPath("tests/soc_descs/blackhole_140_arch_no_eth.yaml"), tt_ClusterDescriptor::get_cluster_descriptor_file_path(), target_devices, num_host_mem_ch_per_mmio_device, false, true, true);
 
     set_params_for_remote_txn(device);
 
@@ -378,7 +378,7 @@ TEST(SiliconDriverBH, MultiThreadedDevice) {
     std::set<chip_id_t> target_devices = get_target_devices();
 
     uint32_t num_host_mem_ch_per_mmio_device = 1;
-    tt_SiliconDevice device = tt_SiliconDevice(test_utils::GetAbsPath("tests/soc_descs/blackhole_140_arch_no_eth.yaml"), test_utils::GetAbsPath("blackhole_1chip_cluster.yaml"), target_devices, num_host_mem_ch_per_mmio_device, false, true, true);
+    tt_SiliconDevice device = tt_SiliconDevice(test_utils::GetAbsPath("tests/soc_descs/blackhole_140_arch_no_eth.yaml"), tt_ClusterDescriptor::get_cluster_descriptor_file_path(), target_devices, num_host_mem_ch_per_mmio_device, false, true, true);
 
     set_params_for_remote_txn(device);
 
@@ -437,7 +437,7 @@ TEST(SiliconDriverBH, MultiThreadedMemBar) {
     uint32_t base_addr = l1_mem::address_map::DATA_BUFFER_SPACE_BASE;
     uint32_t num_host_mem_ch_per_mmio_device = 1;
 
-    tt_SiliconDevice device = tt_SiliconDevice(test_utils::GetAbsPath("tests/soc_descs/blackhole_140_arch_no_eth.yaml"), test_utils::GetAbsPath("blackhole_1chip_cluster.yaml"), target_devices, num_host_mem_ch_per_mmio_device, false, true, true);
+    tt_SiliconDevice device = tt_SiliconDevice(test_utils::GetAbsPath("tests/soc_descs/blackhole_140_arch_no_eth.yaml"), tt_ClusterDescriptor::get_cluster_descriptor_file_path(), target_devices, num_host_mem_ch_per_mmio_device, false, true, true);
     set_params_for_remote_txn(device);
     for(int i = 0; i < target_devices.size(); i++) {
         // Iterate over devices and only setup static TLBs for functional worker cores
@@ -539,7 +539,7 @@ TEST(SiliconDriverBH, DISABLED_BroadcastWrite) { // Cannot broadcast to tensix/e
 
     uint32_t num_host_mem_ch_per_mmio_device = 1;
 
-    tt_SiliconDevice device = tt_SiliconDevice(test_utils::GetAbsPath("tests/soc_descs/blackhole_140_arch_no_eth.yaml"), test_utils::GetAbsPath("blackhole_1chip_cluster.yaml"), target_devices, num_host_mem_ch_per_mmio_device, false, true, true);
+    tt_SiliconDevice device = tt_SiliconDevice(test_utils::GetAbsPath("tests/soc_descs/blackhole_140_arch_no_eth.yaml"), tt_ClusterDescriptor::get_cluster_descriptor_file_path(), target_devices, num_host_mem_ch_per_mmio_device, false, true, true);
     set_params_for_remote_txn(device);
     auto mmio_devices = device.get_target_mmio_device_ids();
 
@@ -596,7 +596,7 @@ TEST(SiliconDriverBH, DISABLED_VirtualCoordinateBroadcast) { // same problem as
 
     uint32_t num_host_mem_ch_per_mmio_device = 1;
 
-    tt_SiliconDevice device = tt_SiliconDevice(test_utils::GetAbsPath("tests/soc_descs/blackhole_140_arch_no_eth.yaml"), test_utils::GetAbsPath("blackhole_1chip_cluster.yaml"), target_devices, num_host_mem_ch_per_mmio_device, false, true, true);
+    tt_SiliconDevice device = tt_SiliconDevice(test_utils::GetAbsPath("tests/soc_descs/blackhole_140_arch_no_eth.yaml"), tt_ClusterDescriptor::get_cluster_descriptor_file_path(), target_devices, num_host_mem_ch_per_mmio_device, false, true, true);
     set_params_for_remote_txn(device);
     auto mmio_devices = device.get_target_mmio_device_ids();