Merge remote-tracking branch 'origin/main' into afuller/clarify-api

.github/workflows/on-pr-opt.yml → .github/workflows/build-and-run-all-tests.yml

@@ -1,10 +1,12 @@
-# Optional PR checks
-name: On PR - Optional
+# Build and then run all tests, on all supported archs.
+name: Build and run all tests
 
 on:
   workflow_dispatch:
   pull_request:
     branches: ["main"]
+  push:
+    branches: ["main"]
 
 jobs:
   build-tests:

.github/workflows/build-clients.yml

@@ -0,0 +1,56 @@
+name: Build clients on newest UMD
+
+on:
+  workflow_dispatch:
+    inputs:
+      timeout:
+        required: true
+        description: 'The timeout for the job in minutes'
+        type: number
+        default: 30
+  pull_request:
+    branches: ["main"]
+  push:
+    branches: ["main"]
+
+jobs:
+  build-tt-metal:
+    # Due to parsing bug, fromJSON is used to convert string to number.
+    # In pull_request or push events, the input context is not available, stating the default again here.
+    timeout-minutes: ${{ fromJSON(inputs.timeout || '30') }}
+    strategy:
+      fail-fast: false
+      matrix:
+        arch_name: [grayskull, wormhole_b0, blackhole]
+
+    name: Build tt-metal for ${{ matrix.arch_name }} with newest UMD
+    runs-on: ubuntu-20.04
+    container:
+      image: ghcr.io/tenstorrent/tt-metal/tt-metalium/ubuntu-20.04-amd64:latest
+      options: --user root
+
+    steps:
+      - name: Checkout client repo
+        uses: actions/checkout@v4
+        with:
+          # Clone under tt-metal directory
+          path: tt-metal
+          repository: tenstorrent/tt-metal
+          submodules: recursive
+          lfs: 'true'
+
+      - name: Checkout UMD
+        uses: actions/checkout@v4
+        with:
+          # Clone directly into tt-metal directory for umd
+          path: tt-metal/tt_metal/third_party/umd
+          submodules: recursive
+          lfs: 'true'
+
+      - name: Build tt-metal
+        run: |
+          cd tt-metal
+          export ARCH_NAME=${{ matrix.arch_name }}
+          export TT_METAL_HOME=$(pwd)
+          export PYTHONPATH=$(pwd)
+          ./build_metal.sh

.github/workflows/build-device.yml

@@ -1,19 +1,19 @@
 # Builds device.
 # Build is performed on all supported OS versions.
-name: Build Target
+name: Build Device
 
 on:
-  workflow_call:
-    inputs:
-      timeout:
-        required: true
-        type: number
   workflow_dispatch:
     inputs:
       timeout:
         required: true
-        description: 'The timeout for the build job in minutes'
+        description: 'The timeout for the job in minutes'
         type: number
+        default: 15
+  pull_request:
+    branches: ["main"]
+  push:
+    branches: ["main"]
 
 env:
   BUILD_TARGET: device
@@ -25,7 +25,9 @@ env:
 
 jobs:
   build:
-    timeout-minutes: ${{ inputs.timeout }}
+    # Due to parsing bug, fromJSON is used to convert string to number.
+    # In pull_request or push events, the input context is not available, stating the default again here.
+    timeout-minutes: ${{ fromJSON(inputs.timeout || '15') }}
     strategy:
       fail-fast: false
       matrix:

.github/workflows/build-image.yml

@@ -3,11 +3,17 @@ name: Build and Publish Docker Image
 
 on:
   workflow_dispatch:
-  workflow_call:
+    inputs:
+      timeout:
+        required: true
+        description: 'The timeout for the job in minutes'
+        type: number
+        default: 15
 
 jobs:
   build:
-    timeout-minutes: 15
+    # Due to parsing bug, fromJSON is used to convert string to number
+    timeout-minutes: ${{ fromJSON(inputs.timeout) }}
     strategy:
       fail-fast: false
       matrix:

.github/workflows/build-tests.yml

@@ -33,10 +33,12 @@ env:
   LIB_OUTPUT_DIR: ./build/lib
   DEPS_OUTPUT_DIR: ./build/_deps
   TEST_OUTPUT_DIR: ./build/test
+  CLUSTER_DESCRIPTORS_DIR: ./tests/api/cluster_descriptor_examples
 
 jobs:
   build:
-    timeout-minutes: ${{ inputs.timeout }}
+    # Due to parsing bug, fromJSON is used to convert string to number
+    timeout-minutes: ${{ fromJSON(inputs.timeout) }}
     strategy:
       fail-fast: false
       matrix:
@@ -74,7 +76,8 @@ jobs:
         run: |
           tar cvf artifact.tar ${{ env.TEST_OUTPUT_DIR }} \
           ${{ env.LIB_OUTPUT_DIR }} \
-          ${{ env.DEPS_OUTPUT_DIR }}
+          ${{ env.DEPS_OUTPUT_DIR }} \
+          ${{ env.CLUSTER_DESCRIPTORS_DIR }}
 
       - name: Upload build artifacts archive
         uses: actions/upload-artifact@v4

.github/workflows/pre-commit.yml

@@ -2,8 +2,11 @@
 name: Run Pre-commit Hooks
 
 on:
-  workflow_call:
   workflow_dispatch:
+  pull_request:
+    branches: ["main"]
+  push:
+    branches: ["main"]
 
 jobs:
   pre-commit:

.github/workflows/run-tests.yml

@@ -45,7 +45,8 @@ env:
 
 jobs:
   test:
-    timeout-minutes: ${{ inputs.timeout }}
+    # Due to parsing bug, fromJSON is used to convert string to number
+    timeout-minutes: ${{ fromJSON(inputs.timeout) }}
     strategy:
       fail-fast: false
       matrix:

.github/workflows/test-runner.yaml

@@ -2,10 +2,17 @@ name: Check runner
 
 on:
   workflow_dispatch:
+    inputs:
+      timeout:
+        required: true
+        description: 'The timeout for the job in minutes'
+        type: number
+        default: 10
 
 jobs:
   check-runners-host:
-    timeout-minutes: 10
+    # Due to parsing bug, fromJSON is used to convert string to number
+    timeout-minutes: ${{ fromJSON(inputs.timeout) }}
     strategy:
       fail-fast: false
       matrix:
@@ -52,7 +59,8 @@ jobs:
           du -h --max-depth=1 | sort -rh
 
   check-runners-docker:
-    timeout-minutes: 10
+    # Due to parsing bug, fromJSON is used to convert string to number
+    timeout-minutes: ${{ fromJSON(inputs.timeout) }}
     strategy:
       fail-fast: false
       matrix:

common/disjoint_set.hpp

@@ -0,0 +1,42 @@
+/*
+ * SPDX-FileCopyrightText: (c) 2024 Tenstorrent Inc.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+#pragma once
+
+#include <unordered_map>
+#include <unordered_set>
+
+// A standard disjoint set data structure to track connected components.
+template <typename T>
+class DisjointSet {
+public:
+    void add_item(T item) { parent[item] = item; }
+
+    int get_set(T item) {
+        while (parent[item] != item) {
+            item = parent[item];
+        }
+        return item;
+    }
+
+    void merge(T item1, T item2) {
+        T set1 = get_set(item1);
+        T set2 = get_set(item2);
+        parent[set1] = set2;
+    }
+
+    bool are_same_set(T item1, T item2) { return get_set(item1) == get_set(item2); }
+
+    int get_num_sets() {
+        std::unordered_set<T> sets;
+        for (auto [item, _] : parent) {
+            sets.insert(get_set(item));
+        }
+        return sets.size();
+    }
+
+private:
+    std::unordered_map<T, T> parent;
+};

common/utils.hpp

@@ -0,0 +1,31 @@
+/*
+ * SPDX-FileCopyrightText: (c) 2024 Tenstorrent Inc.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+#pragma once
+
+#include <filesystem>
+#include <iostream>
+#include <string>
+
+namespace tt::umd::utils {
+
+std::string get_abs_path(std::string path) {
+    // Note that __FILE__ might be resolved at compile time to an absolute or relative address, depending on the
+    // compiler.
+    std::filesystem::path current_file_path = std::filesystem::path(__FILE__);
+    std::filesystem::path umd_root;
+    if (current_file_path.is_absolute()) {
+        umd_root = current_file_path.parent_path().parent_path();
+    } else {
+        std::filesystem::path umd_root_relative =
+            std::filesystem::relative(std::filesystem::path(__FILE__).parent_path().parent_path(), "../");
+        umd_root = std::filesystem::canonical(umd_root_relative);
+    }
+    std::filesystem::path abs_path = umd_root / path;
+    return abs_path.string();
+}
+
+}  // namespace tt::umd::utils

device/api/umd/device/architecture_implementation.h

@@ -12,8 +12,8 @@
 #include <vector>
 
 #include "umd/device/tlb.h"
-#include "umd/device/xy_pair.h"
 #include "umd/device/tt_arch_types.h"
+#include "umd/device/xy_pair.h"
 
 struct tt_driver_host_address_params;
 struct tt_driver_eth_interface_params;
@@ -22,7 +22,7 @@ struct tt_driver_noc_params;
 namespace tt::umd {
 
 class architecture_implementation {
-   public:
+public:
     virtual ~architecture_implementation() = default;
 
     virtual tt::ARCH get_architecture() const = 0;
@@ -65,7 +65,8 @@ class architecture_implementation {
     virtual std::tuple<xy_pair, xy_pair> multicast_workaround(xy_pair start, xy_pair end) const = 0;
     virtual tlb_configuration get_tlb_configuration(uint32_t tlb_index) const = 0;
     virtual std::optional<std::tuple<std::uint64_t, std::uint64_t>> describe_tlb(std::int32_t tlb_index) const = 0;
-    virtual std::pair<std::uint64_t, std::uint64_t> get_tlb_data(std::uint32_t tlb_index, const tlb_data& data) const = 0;
+    virtual std::pair<std::uint64_t, std::uint64_t> get_tlb_data(
+        std::uint32_t tlb_index, const tlb_data& data) const = 0;
 
     virtual tt_driver_host_address_params get_host_address_params() const = 0;
     virtual tt_driver_eth_interface_params get_eth_interface_params() const = 0;