Fix OpenVINO tests in CI

backends/openvino/runtime/OpenvinoApi.h

@@ -76,6 +76,8 @@ using ov_core_free_fn = void (*)(ov_core_t*);
 using ov_core_get_available_devices_fn =
     ov_status_e (*)(const ov_core_t*, ov_available_devices_t*);
 using ov_available_devices_free_fn = void (*)(ov_available_devices_t*);
+using ov_core_set_property_fn =
+    ov_status_e (*)(const ov_core_t*, const char*, ...);
 using ov_core_import_model_fn = ov_status_e (*)(
     const ov_core_t*,
     const char*,
@@ -115,6 +117,7 @@ struct OpenvinoFunctions {
   ov_core_free_fn core_free = nullptr;
   ov_core_get_available_devices_fn core_get_available_devices = nullptr;
   ov_available_devices_free_fn available_devices_free = nullptr;
+  ov_core_set_property_fn core_set_property = nullptr;
   ov_core_import_model_fn core_import_model = nullptr;
   ov_compiled_model_create_infer_request_fn
       compiled_model_create_infer_request = nullptr;

backends/openvino/runtime/OpenvinoBackend.cpp

@@ -236,6 +236,7 @@ bool OpenvinoBackend::ensure_loaded() const {
     LOAD_SYM(core_free, ov_core_free)
     LOAD_SYM(core_get_available_devices, ov_core_get_available_devices)
     LOAD_SYM(available_devices_free, ov_available_devices_free)
+    LOAD_SYM(core_set_property, ov_core_set_property)
     LOAD_SYM(core_import_model, ov_core_import_model)
     LOAD_SYM(
         compiled_model_create_infer_request,
@@ -321,17 +322,44 @@ exr::Result<exr::DelegateHandle*> OpenvinoBackend::init(
   size_t data_size = processed->size();
 
   std::string device = "CPU";
+  std::string inference_precision;
   for (auto& compile_spec : compile_specs) {
+    std::string* target = nullptr;
     if (std::strcmp(compile_spec.key, "device") == 0) {
-      const char* buf = static_cast<const char*>(compile_spec.value.buffer);
-      size_t len = compile_spec.value.nbytes;
-      // Strip trailing null bytes that may be included in nbytes.
-      while (len > 0 && buf[len - 1] == '\0') {
-        --len;
-      }
-      if (len > 0) {
-        device.assign(buf, len);
-      }
+      target = &device;
+    } else if (std::strcmp(compile_spec.key, "inference_precision") == 0) {
+      target = &inference_precision;
+    } else {
+      continue;
+    }
+    const char* buf = static_cast<const char*>(compile_spec.value.buffer);
+    size_t len = compile_spec.value.nbytes;
+    // Strip trailing null bytes that may be included in nbytes.
+    while (len > 0 && buf[len - 1] == '\0') {
+      --len;
+    }
+    if (len > 0) {
+      target->assign(buf, len);
+    }
+  }
+
+  // The device picks its default precision when the model is imported, e.g.
+  // bf16 for f32 models on CPUs with native bf16 support.
+  if (!inference_precision.empty()) {
+    status = ov_.core_set_property(
+        core,
+        device.c_str(),
+        "INFERENCE_PRECISION_HINT",
+        inference_precision.c_str());
+    if (status != OV_STATUS_OK) {
+      ET_LOG(
+          Error,
+          "OpenVINO: failed to set inference precision '%s' for device '%s' (status=%d)",
+          inference_precision.c_str(),
+          device.c_str(),
+          status);
+      ov_.core_free(core);
+      return exr::Error::InvalidArgument;
     }
   }
 
@@ -343,8 +371,10 @@ exr::Result<exr::DelegateHandle*> OpenvinoBackend::init(
   if (status != OV_STATUS_OK || !compiled_model) {
     ET_LOG(
         Error,
-        "OpenVINO: failed to import model for device '%s' (status=%d)",
+        "OpenVINO: failed to import model for device '%s'%s%s (status=%d)",
         device.c_str(),
+        inference_precision.empty() ? "" : " with inference_precision ",
+        inference_precision.c_str(),
         status);
     return exr::Error::Internal;
   }
@@ -541,8 +571,12 @@ exr::Result<ov_tensor_t*> OpenvinoBackend::create_ov_tensor(
   }
   // shape is zero-initialized; shape_free is only needed after a successful
   // shape_create (the zero state is safe to skip).
+  // ov_shape_create rejects rank 0, so 0-d tensors use the zero shape as-is.
   ov_shape_t shape = {};
-  ov_status_e status = ov_.shape_create(rank, dims, &shape);
+  ov_status_e status = OV_STATUS_OK;
+  if (rank > 0) {
+    status = ov_.shape_create(rank, dims, &shape);
+  }
   dims_heap.reset(); // Release heap dims (no-op if stack was used).
   if (status != OV_STATUS_OK) {
     return exr::Error::Internal;
@@ -551,7 +585,9 @@ exr::Result<ov_tensor_t*> OpenvinoBackend::create_ov_tensor(
   ov_tensor_t* ov_tensor = nullptr;
   status = ov_.tensor_create_from_host_ptr(
       ov_type, shape, tensor.mutable_data_ptr(), &ov_tensor);
-  ov_.shape_free(&shape);
+  if (rank > 0) {
+    ov_.shape_free(&shape);
+  }
   if (status != OV_STATUS_OK || !ov_tensor) {
     return exr::Error::Internal;
   }

backends/openvino/tests/ops/base_openvino_op_test.py

@@ -33,6 +33,9 @@ def execute_layer_test(
 
         # Convert to edge dialect and lower the module to the backend with a custom partitioner
         compile_spec = [CompileSpec("device", self.device.encode())]
+        # CPUs with native bf16 otherwise run f32 models in bf16, which misses atol.
+        if self.device == "CPU":
+            compile_spec.append(CompileSpec("inference_precision", b"f32"))
         lowered_module: EdgeProgramManager = to_edge_transform_and_lower(
             aten_dialect,
             partitioner=[

backends/openvino/tests/test_runner.py

@@ -1,4 +1,5 @@
 import argparse
+import sys
 import unittest
 
 
@@ -71,3 +72,4 @@ def parse_arguments():
         print("OpenVINO backend tests completed successfully")
     else:
         print("OpenVINO backend tests completed with failures")
+        sys.exit(1)