Test: run only the a5 jobs, on the new [self-hosted, a5, test] runner

.github/workflows/ci.yml

@@ -13,8 +13,15 @@ concurrency:
   cancel-in-progress: true
 
 jobs:
+  # TEMPORARY: only the a5 self-hosted jobs (ut-a5, st-onboard-a5) run, so the
+  # new [self-hosted, a5, test] runner can be validated in isolation. Every
+  # other job is gated off with `if: false` and the a5 jobs have their
+  # `needs:`/`detect-changes` gating dropped. Revert this whole block to
+  # restore the normal pipeline.
+
   # ---------- Pre-commit hooks (format, lint, clang-tidy) ----------
   pre-commit:
+    if: false
     runs-on: ubuntu-latest
     timeout-minutes: 30
     steps:
@@ -62,7 +69,7 @@ jobs:
   # See docs/python-packaging.md and tools/verify_packaging.sh.
   packaging-matrix:
     needs: [detect-changes, pre-commit]
-    if: needs.detect-changes.outputs.docs_only != 'true'
+    if: false
     runs-on: ${{ matrix.os }}
     timeout-minutes: 60
     strategy:
@@ -107,7 +114,7 @@ jobs:
   # ---------- Unit tests (no hardware, Python + C++) ----------
   ut:
     needs: [detect-changes, pre-commit]
-    if: needs.detect-changes.outputs.docs_only != 'true'
+    if: false
     runs-on: ${{ matrix.os }}
     timeout-minutes: 15
     strategy:
@@ -163,7 +170,7 @@ jobs:
   # ---------- Simulation scene tests ----------
   st-sim-a2a3:
     needs: [detect-changes, pre-commit]
-    if: needs.detect-changes.outputs.a2a3_changed == 'true'
+    if: false
     runs-on: ${{ matrix.os }}
     timeout-minutes: 30
     env:
@@ -257,7 +264,7 @@ jobs:
 
   st-sim-a5:
     needs: [detect-changes, pre-commit]
-    if: needs.detect-changes.outputs.a5_changed == 'true'
+    if: false
     runs-on: ${{ matrix.os }}
     timeout-minutes: 30
     env:
@@ -375,7 +382,7 @@ jobs:
   # corresponding profiling block — no cross-contamination from cmake cache.
   profiling-flags-smoke:
     needs: [detect-changes, pre-commit]
-    if: needs.detect-changes.outputs.a2a3_changed == 'true' || needs.detect-changes.outputs.a5_changed == 'true'
+    if: false
     runs-on: ubuntu-latest
     timeout-minutes: 30
 
@@ -473,7 +480,7 @@ jobs:
   # ---------- Unit tests (a2a3 hardware, Python + C++) ----------
   ut-a2a3:
     needs: [detect-changes, pre-commit]
-    if: needs.detect-changes.outputs.docs_only != 'true'
+    if: false
     runs-on: [self-hosted, a2a3]
     timeout-minutes: 30
     env:
@@ -559,7 +566,7 @@ jobs:
   # ---------- Scene tests (a2a3 hardware) ----------
   st-onboard-a2a3:
     needs: [detect-changes, pre-commit]
-    if: needs.detect-changes.outputs.a2a3_changed == 'true'
+    if: false
     runs-on: [self-hosted, a2a3]
     timeout-minutes: 60
     env:
@@ -728,6 +735,7 @@ jobs:
 
   # ---------- Detect platform-specific changes (runs on GitHub server) ----------
   detect-changes:
+    if: false
     runs-on: ubuntu-latest
     outputs:
       a2a3_changed: ${{ steps.check.outputs.a2a3_changed }}
@@ -784,9 +792,7 @@ jobs:
 
   # ---------- Unit tests (a5 hardware, Python + C++) ----------
   ut-a5:
-    needs: [detect-changes, pre-commit]
-    if: needs.detect-changes.outputs.docs_only != 'true'
-    runs-on: [self-hosted, a5]
+    runs-on: [self-hosted, test]
     timeout-minutes: 30
     env:
       SIMPLER_SCHEDULER_TIMEOUT_MS: "2000"
@@ -816,6 +822,10 @@ jobs:
           source .venv/bin/activate
           cmake -B tests/ut/cpp/build -S tests/ut/cpp -DSIMPLER_ENABLE_HARDWARE_TESTS=ON
           cmake --build tests/ut/cpp/build
+          # The no_hardware suite needs no device, so it runs outside the
+          # task-submit lock. --no-tests=error catches a label typo silently
+          # selecting nothing.
+          ctest --test-dir tests/ut/cpp/build -L no_hardware -j$(nproc) --no-tests=error --output-on-failure
           python3 -c "
           import json, os
           p = os.environ['DEVICE_RANGE'].split('-'); s, e = p[0], p[-1]  # tolerate a single id (no hyphen)
@@ -824,6 +834,10 @@ jobs:
                     open('tests/ut/cpp/build/resources.json', 'w'))
           "
           DEVICE_LIST=$(python -c "p='${DEVICE_RANGE}'.split('-'); s,e=p[0],p[-1]; print(','.join(str(i) for i in range(int(s),int(e)+1)))")
+          # No test carries requires_hardware_a5 yet — the only hardware UT,
+          # test_comm_lifecycle, is a2a3-bound. This selection is therefore
+          # empty and omits --no-tests=error until the first a5 hardware UT
+          # lands; the no_hardware run above is what gives this job coverage.
           task-submit --timeout 1800 --max-time 1800 --device "$DEVICE_LIST" --run "ctest --test-dir tests/ut/cpp/build -L '^requires_hardware(_a5)?\$' --resource-spec-file $PWD/tests/ut/cpp/build/resources.json -j$(nproc) --output-on-failure"
 
       - name: Build cann-examples/query (CANN host-API smoke)
@@ -854,9 +868,7 @@ jobs:
           cmake --build build
 
   st-onboard-a5:
-    needs: [detect-changes, pre-commit]
-    if: needs.detect-changes.outputs.a5_changed == 'true'
-    runs-on: [self-hosted, a5]
+    runs-on: [self-hosted, test]
     timeout-minutes: 60
     env:
       SIMPLER_SCHEDULER_TIMEOUT_MS: "2000"

src/a5/docs/hardware.md

@@ -150,6 +150,24 @@ For cross-generation portable code: **always go through ACL or CANN
 ini, never HAL**. HAL's CORE_NUM semantics shift between a3 and a5 in
 ways that have no public documentation.
 
+### CPU_TOPO compatibility on newer a5 drivers
+
+On `Ascend950PR_9579` with driver `25.7.rc1.6`, both host-side and
+device-side `AICPU + OCCUPY` report `0x3e`, so cpu_ids 1 through 5 are
+the complete user-schedulable pool. Launching five AICPU threads reaches
+each of those cpu_ids exactly once.
+
+The same driver returns `DRV_ERROR_NOT_SUPPORT` for both
+`halGetDeviceInfoByBuff(SYSTEM, CPU_TOPO)` and
+`dsmi_get_device_info(SOC_INFO, CPU_TOPO)`. Its public DSMI header only
+defines SOC_INFO subcommands 0 and 1.
+
+When CPU_TOPO is unavailable, the host runtime enumerates OCCUPY directly
+and treats each set bit as a distinct non-SMT physical CPU. This preserves
+the a5 cpu_id-to-cluster mapping while keeping the launch count equal to the
+actual user pool. Drivers that provide CPU_TOPO continue to use its detailed
+physical and hyperthread metadata.
+
 ## CANN AICPU thread dispatch under varying launch budgets
 
 How CANN distributes N AICPU threads across the user pool determines

src/a5/platform/onboard/host/aicpu_topology_probe.cpp

@@ -144,7 +144,13 @@ bool probe_aicpu_topology_uncached(uint32_t device_id, std::vector<AicpuLogicalC
     if (!query_occupy(device_id, occupy)) return false;
 
     DsmiCpuTopo topo{};
-    if (!query_cpu_topo(device_id, topo)) return false;
+    if (!query_cpu_topo(device_id, topo)) {
+        LOG_WARN(
+            "aicpu_topology_probe: CPU_TOPO unavailable; deriving topology from OCCUPY=0x%llx",
+            static_cast<unsigned long long>(occupy)
+        );
+        return derive_topology_from_occupy(occupy, out_user_cpus);
+    }
 
     for (uint32_t i = 0; i < topo.total_nums; ++i) {
         const DsmiSingleCpu &c = topo.cpus[i];
@@ -170,6 +176,21 @@ bool probe_aicpu_topology_uncached(uint32_t device_id, std::vector<AicpuLogicalC
 
 }  // namespace
 
+bool derive_topology_from_occupy(uint64_t occupy, std::vector<AicpuLogicalCpu> &out_user_cpus) {
+    out_user_cpus.clear();
+    for (int32_t cpu_id = 0; cpu_id < 64; ++cpu_id) {
+        if (((occupy >> cpu_id) & 1ULL) == 0) continue;
+        AicpuLogicalCpu cpu{};
+        cpu.cpu_id = cpu_id;
+        cpu.phy_cpu_id = cpu_id;
+        cpu.hyperthread_id = 0;
+        cpu.cluster_id = cpu.phy_cpu_id / 2;
+        cpu.die_id = cpu.phy_cpu_id / 4;
+        out_user_cpus.push_back(cpu);
+    }
+    return !out_user_cpus.empty();
+}
+
 bool probe_aicpu_topology(uint32_t device_id, std::vector<AicpuLogicalCpu> &out_user_cpus) {
     {
         std::lock_guard<std::mutex> lk(s_topo_cache_mu);

src/a5/platform/onboard/host/aicpu_topology_probe.h

@@ -17,9 +17,10 @@
 
 namespace pto::a5 {
 
-// Per-cpu_id metadata used by the packing algorithm. Filled from DSMI
-// CPU_TOPO + halGetDeviceInfo(AICPU, OCCUPY). cluster/die ids derive from
-// phy_cpu_id via the a5 mapping (cluster = phy/2, die = phy/4).
+// Per-cpu_id metadata used by the packing algorithm. Filled from the driver
+// CPU_TOPO data when available, otherwise derived from the AICPU OCCUPY
+// bitmap. cluster/die ids follow the a5 mapping (cluster = phy/2,
+// die = phy/4).
 struct AicpuLogicalCpu {
     int32_t cpu_id;
     int32_t phy_cpu_id;
@@ -33,15 +34,21 @@ struct AicpuLogicalCpu {
 // only contains cpu_ids that are in the device-side OCCUPY bitmap (i.e.
 // user-schedulable), sorted by cpu_id ascending.
 //
-// This function performs three driver calls:
+// This function uses these driver calls:
 //   * halGetDeviceInfo(AICPU, OCCUPY) — user-schedulable bitmap
 //   * halGetDeviceInfoByBuff(SYSTEM, CPU_TOPO)  (primary)
 //   * dsmi_get_device_info(SOC_INFO, CPU_TOPO)  (fallback)
 //
+// Drivers that do not expose CPU_TOPO use the OCCUPY-only topology below.
 // All driver entry points are dlsym'd from the host process (CANN is
 // expected to be already loaded by the surrounding `aclInit` path).
 bool probe_aicpu_topology(uint32_t device_id, std::vector<AicpuLogicalCpu> &out_user_cpus);
 
+// Build conservative a5 topology metadata from a user-schedulable OCCUPY
+// bitmap. Every set bit is a distinct non-SMT physical CPU; AICPU cpu_ids
+// retain the a5 two-physical-CPUs-per-cluster, two-clusters-per-die layout.
+bool derive_topology_from_occupy(uint64_t occupy, std::vector<AicpuLogicalCpu> &out_user_cpus);
+
 // Compute the `ALLOWED_CPUS` selection for the surviving threads.
 //
 // Inputs:

tests/ut/cpp/CMakeLists.txt

@@ -732,6 +732,27 @@ target_link_libraries(test_a2a3_aicpu_affinity_select PRIVATE
 add_test(NAME test_a2a3_aicpu_affinity_select COMMAND test_a2a3_aicpu_affinity_select)
 set_tests_properties(test_a2a3_aicpu_affinity_select PROPERTIES LABELS "no_hardware")
 
+# a5 CPU_TOPO fallback and affinity selection. Pure logic; the fallback
+# enumerates the driver OCCUPY bitmap when detailed topology is unavailable.
+set(A5_ONBOARD_HOST_DIR ${CMAKE_SOURCE_DIR}/../../../src/a5/platform/onboard/host)
+add_executable(test_a5_aicpu_topology_fallback
+    a5/test_aicpu_topology_fallback.cpp
+    ${A5_ONBOARD_HOST_DIR}/aicpu_topology_probe.cpp
+)
+target_include_directories(test_a5_aicpu_topology_fallback PRIVATE
+    ${GTEST_INCLUDE_DIRS}
+    ${A5_ONBOARD_HOST_DIR}
+    ${SIMPLER_LOG_DIR}/include
+)
+target_link_libraries(test_a5_aicpu_topology_fallback PRIVATE
+    ${GTEST_MAIN_LIB}
+    ${GTEST_LIB}
+    ${CMAKE_DL_LIBS}
+    pthread
+)
+add_test(NAME test_a5_aicpu_topology_fallback COMMAND test_a5_aicpu_topology_fallback)
+set_tests_properties(test_a5_aicpu_topology_fallback PROPERTIES LABELS "no_hardware")
+
 # Hardware-gated tests.  Block is only entered when the project is configured
 # with -DSIMPLER_ENABLE_HARDWARE_TESTS=ON.  CI's no-hw `ut` job does not pass
 # this flag, so nothing here is compiled there — this is the structural

tests/ut/cpp/a5/test_aicpu_topology_fallback.cpp

@@ -0,0 +1,62 @@
+/*
+ * Copyright (c) PyPTO Contributors.
+ * This program is free software, you can redistribute it and/or modify it under the terms and conditions of
+ * CANN Open Software License Agreement Version 2.0 (the "License").
+ * Please refer to the License for details. You may not use this file except in compliance with the License.
+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
+ * INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
+ * See LICENSE in the root of the software repository for the full text of the License.
+ * -----------------------------------------------------------------------------------------------------------
+ */
+
+#include <gtest/gtest.h>
+
+#include <cstdint>
+#include <vector>
+
+#include "aicpu_topology_probe.h"
+
+extern "C" {
+void unified_log_error(const char *, const char *, ...) {}
+void unified_log_warn(const char *, const char *, ...) {}
+void unified_log_info_v(const char *, int, const char *, ...) {}
+}
+
+namespace {
+
+using pto::a5::AicpuLogicalCpu;
+using pto::a5::compute_allowed_cpus;
+using pto::a5::derive_topology_from_occupy;
+
+TEST(A5AicpuTopologyFallback, EnumeratesEveryOccupiedCpu) {
+    std::vector<AicpuLogicalCpu> cpus;
+
+    ASSERT_TRUE(derive_topology_from_occupy(0x3e, cpus));
+    ASSERT_EQ(cpus.size(), 5U);
+
+    for (int32_t i = 0; i < 5; ++i) {
+        EXPECT_EQ(cpus[i].cpu_id, i + 1);
+        EXPECT_EQ(cpus[i].phy_cpu_id, i + 1);
+        EXPECT_EQ(cpus[i].hyperthread_id, 0);
+        EXPECT_EQ(cpus[i].cluster_id, (i + 1) / 2);
+        EXPECT_EQ(cpus[i].die_id, (i + 1) / 4);
+    }
+}
+
+TEST(A5AicpuTopologyFallback, PreservesAffinitySelection) {
+    std::vector<AicpuLogicalCpu> cpus;
+    ASSERT_TRUE(derive_topology_from_occupy(0x3e, cpus));
+
+    std::vector<int32_t> allowed;
+    ASSERT_TRUE(compute_allowed_cpus(cpus, /*n_sched=*/2, /*n_orch=*/1, allowed));
+    EXPECT_EQ(allowed, (std::vector<int32_t>{4, 5, 1}));
+}
+
+TEST(A5AicpuTopologyFallback, RejectsEmptyOccupyMask) {
+    std::vector<AicpuLogicalCpu> cpus = {{1, 1, 0, 0, 0}};
+
+    EXPECT_FALSE(derive_topology_from_occupy(0, cpus));
+    EXPECT_TRUE(cpus.empty());
+}
+
+}  // namespace