ORT 1.24.1 release cherry pick round 2 (#27233)

#27021: [Disable matmul 1d
tests on
DML](1afc8bc)


#27206: [[CUDA] Run
FlashAttention regression test only when FlashAttention
is…](4d95d97)

#27120: [POWER : Fix build
failure due to unsupported cpuinfo on
ppc64le](2843ec0)

---------

Co-authored-by: Ti-Tai Wang <titaiwang@microsoft.com>
Co-authored-by: Hariharan Seshadri <shariharan91@gmail.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: BODAPATIMAHESH <148746454+BODAPATIMAHESH@users.noreply.github.com>

Author: 5 people

cmake/onnxruntime_unittests.cmake

@@ -1553,8 +1553,13 @@ endif()
                               onnxruntime_common ${CMAKE_DL_LIBS})
         set_target_properties(onnxruntime_runtime_path_test_shared_library PROPERTIES AIX_SHARED_LIBRARY_ARCHIVE OFF)
       else()
-        target_link_libraries(onnxruntime_runtime_path_test_shared_library PRIVATE
-                              onnxruntime_common cpuinfo ${CMAKE_DL_LIBS})
+        if (CPUINFO_SUPPORTED)
+          target_link_libraries(onnxruntime_runtime_path_test_shared_library PRIVATE
+                                onnxruntime_common cpuinfo ${CMAKE_DL_LIBS})
+        else()
+          target_link_libraries(onnxruntime_runtime_path_test_shared_library PRIVATE
+                                onnxruntime_common ${CMAKE_DL_LIBS})
+        endif()
       endif()
       target_include_directories(onnxruntime_runtime_path_test_shared_library PRIVATE ${ONNXRUNTIME_ROOT})
 

onnxruntime/test/python/transformers/test_gqa.py

@@ -1775,6 +1775,7 @@ def test_flash_decode_parity(self):
         del os.environ["ORT_DISABLE_FLASH_DECODE"]
 
 
+@unittest.skipIf(not has_flash_attention(), "Flash Attention is not available, skipping tests.")
 class TestGQARegressions(unittest.TestCase):
     """Specific regression tests for historical bugs."""
 

onnxruntime/test/testdata/onnx_backend_test_series_filters.jsonc

@@ -785,7 +785,10 @@
         "^test_reduce_max_empty_set_cpu",   // DNNL result in "(shapes (2, 1, 4), (1, 0, 1) mismatch)". this is the same for test_reduce_min_empty_set which is already in the list
         "^test_reduce_min_empty_set_cpu",
         "^test_resize_upsample_sizes_nearest_not_smaller_cpu",
-        "^test_clip_min_greater_than_max_cpu"
+        "^test_clip_min_greater_than_max_cpu",
+        // Fail since v1.20.1 (new matmul 1D tests)
+        "^test_matmul_1d_1d_cpu",
+        "^test_matmul_4d_1d_cpu"
     ],
     // ORT first supported opset 7, so models with nodes that require versions prior to opset 7 are not supported
     "tests_with_pre_opset7_dependencies": [