Make the test harnesses link again: backend registry and engine lib path

src/test/CMakeLists.txt

@@ -95,7 +95,8 @@ function(add_npu_test TARGET_NAME OUTPUT_SUBDIR)
 
     # Basic link settings for test
     target_link_directories(${TARGET_NAME} PUBLIC
-        ${CMAKE_SOURCE_DIR}/../../lib
+        # the XRT-built engine libs; the HRX ones sit beside them in lib/hrx
+        ${CMAKE_SOURCE_DIR}/../../lib/xrt
         ${XRT_LIB_DIR}
         ${CMAKE_SOURCE_DIR}/../../build/tokenizer/lib
         C:/dev/boost_1_88_0/stage/lib

src/test/gemma4_12b/CMakeLists.txt

@@ -11,6 +11,7 @@ add_npu_test(
     USE_TOKENIZER
     USE_SAMPLER
     SOURCES
+        "${CMAKE_SOURCE_DIR}/../../common/AutoModel/model_backend.cpp"
         "${CMAKE_SOURCE_DIR}/../../common/AutoModel/modeling_gemma4_12b.cpp"
         "${CMAKE_SOURCE_DIR}/../../common/AutoModel/modeling_gemma4_12b_image.cpp"   
         "${CMAKE_SOURCE_DIR}/../../common/AutoModel/modeling_gemma4_12b_audio.cpp"             

src/test/gemma4_12b/Makefile

@@ -19,6 +19,7 @@
 TYPE ?= 3
 
 SOURCES += test.cpp
+SOURCES += ../../common/AutoModel/model_backend.cpp
 SOURCES += ../../common/AutoModel/automodel.cpp
 SOURCES += ../../common/AutoModel/modeling_gemma4_12b.cpp
 SOURCES += ../../common/AutoModel/modeling_gemma4_12b_image.cpp

src/test/gemma4_12b/test.cpp

@@ -3,8 +3,17 @@
 #include "utils/utils.hpp"
 #include "utils/vm_args.hpp"
 #include "AutoModel/modeling_gemma4_12b.hpp"
+#include "AutoModel/flm_backend.hpp"
 #include "model_list.hpp"
 
+/// \note builtin_backends.cpp names every engine and so links every prebuilt
+///       library; this harness registers just the ones it drives.
+namespace flm::backend {
+void register_builtin_backends(BackendRegistry& registry) {
+    registry.register_backend("gemma4-12b", kFlmBackendId, flm_factory<gemma4_12b_npu>());
+}
+}  // namespace flm::backend
+
 flm_rt::device npu_device_global;
 
 int main(int argc, char* argv[]) {

src/test/gemma4e_flash/CMakeLists.txt

@@ -20,6 +20,7 @@ add_npu_test(
     USE_SAMPLER
     USE_AVX512
     SOURCES
+        "${CMAKE_SOURCE_DIR}/../../common/AutoModel/model_backend.cpp"
         "${CMAKE_SOURCE_DIR}/../../common/AutoModel/modeling_gemma4e.cpp"
         "${CMAKE_SOURCE_DIR}/../../common/AutoModel/modeling_gemma4e_image.cpp"
         "${CMAKE_SOURCE_DIR}/../../common/AutoModel/modeling_gemma4e_audio.cpp"

src/test/gemma4e_flash/Makefile

@@ -17,6 +17,7 @@
 
 
 SOURCES += test.cpp
+SOURCES += ../../common/AutoModel/model_backend.cpp
 SOURCES += ../../common/AutoModel/automodel.cpp
 SOURCES += ../../common/AutoModel/modeling_gemma4e.cpp
 SOURCES += ../../common/AutoModel/modeling_gemma4e_image.cpp

src/test/gemma4e_flash/test.cpp

@@ -3,8 +3,18 @@
 #include "utils/utils.hpp"
 #include "utils/vm_args.hpp"
 #include "AutoModel/modeling_gemma4e.hpp"
+#include "AutoModel/flm_backend.hpp"
 #include "model_list.hpp"
 
+/// \note builtin_backends.cpp names every engine and so links every prebuilt
+///       library; this harness registers just the ones it drives.
+namespace flm::backend {
+void register_builtin_backends(BackendRegistry& registry) {
+    registry.register_backend("gemma4e", kFlmBackendId, flm_factory<gemma4e_npu>());
+    registry.register_backend("gemma4e-flash", kFlmBackendId, flm_factory<gemma4e_flash>());
+}
+}  // namespace flm::backend
+
 flm_rt::device npu_device_global;
 
 int main(int argc, char* argv[]) {

src/test/gemma4e_npu/CMakeLists.txt

@@ -20,6 +20,7 @@ add_npu_test(
     USE_SAMPLER
     USE_AVX512
     SOURCES
+        "${CMAKE_SOURCE_DIR}/../../common/AutoModel/model_backend.cpp"
         "${CMAKE_SOURCE_DIR}/../../common/AutoModel/modeling_gemma4e.cpp"
         "${CMAKE_SOURCE_DIR}/../../common/AutoModel/modeling_gemma4e_image.cpp"
         "${CMAKE_SOURCE_DIR}/../../common/AutoModel/modeling_gemma4e_audio.cpp"

src/test/gemma4e_npu/Makefile

@@ -17,6 +17,7 @@
 
 
 SOURCES += test.cpp
+SOURCES += ../../common/AutoModel/model_backend.cpp
 SOURCES += ../../common/AutoModel/automodel.cpp
 SOURCES += ../../common/AutoModel/modeling_gemma4e.cpp
 SOURCES += ../../common/AutoModel/modeling_gemma4e_image.cpp

src/test/gemma4e_npu/test.cpp

@@ -3,8 +3,18 @@
 #include "utils/utils.hpp"
 #include "utils/vm_args.hpp"
 #include "AutoModel/modeling_gemma4e.hpp"
+#include "AutoModel/flm_backend.hpp"
 #include "model_list.hpp"
 
+/// \note builtin_backends.cpp names every engine and so links every prebuilt
+///       library; this harness registers just the ones it drives.
+namespace flm::backend {
+void register_builtin_backends(BackendRegistry& registry) {
+    registry.register_backend("gemma4e", kFlmBackendId, flm_factory<gemma4e_npu>());
+    registry.register_backend("gemma4e-flash", kFlmBackendId, flm_factory<gemma4e_flash>());
+}
+}  // namespace flm::backend
+
 flm_rt::device npu_device_global;
 
 int main(int argc, char* argv[]) {

src/test/gemma_npu/CMakeLists.txt

@@ -11,6 +11,7 @@ add_npu_test(
     USE_TOKENIZER
     USE_SAMPLER
     SOURCES
+        "${CMAKE_SOURCE_DIR}/../../common/AutoModel/model_backend.cpp"
         "${CMAKE_SOURCE_DIR}/../../common/AutoModel/modeling_gemma3.cpp"
         "${CMAKE_SOURCE_DIR}/../../common/AutoModel/modeling_gemma3_image.cpp"
         "${CMAKE_SOURCE_DIR}/../../common/image/image_reader.cpp"

src/test/gemma_npu/Makefile

@@ -18,6 +18,7 @@
 -include ../common.mk
 
 SOURCES += test.cpp
+SOURCES += ../../common/AutoModel/model_backend.cpp
 SOURCES += ../../common/AutoModel/automodel.cpp
 SOURCES += ../../common/AutoModel/modeling_gemma3.cpp
 SOURCES += ../../common/AutoModel/modeling_gemma3_image.cpp

src/test/gemma_npu/test.cpp

@@ -3,8 +3,17 @@
 #include "utils/utils.hpp"
 #include "utils/vm_args.hpp"
 #include "AutoModel/modeling_gemma3.hpp"
+#include "AutoModel/flm_backend.hpp"
 #include "model_list.hpp"
 
+/// \note builtin_backends.cpp names every engine and so links every prebuilt
+///       library; this harness registers just the ones it drives.
+namespace flm::backend {
+void register_builtin_backends(BackendRegistry& registry) {
+    registry.register_backend("gemma3", kFlmBackendId, flm_factory<gemma_npu>());
+}
+}  // namespace flm::backend
+
 flm_rt::device npu_device_global;
 
 int main(int argc, char* argv[]) {

src/test/gemma_text_npu/CMakeLists.txt

@@ -11,6 +11,7 @@ add_npu_test(
     USE_TOKENIZER
     USE_SAMPLER
     SOURCES
+        "${CMAKE_SOURCE_DIR}/../../common/AutoModel/model_backend.cpp"
         "${CMAKE_SOURCE_DIR}/../../common/AutoModel/modeling_gemma3_text.cpp"
 )
 

src/test/gemma_text_npu/Makefile

@@ -15,6 +15,7 @@
 # =============================================================================
 -include ../common.mk
 
+SOURCES += ../../common/AutoModel/model_backend.cpp
 SOURCES += $(wildcard ../../common/AutoModel/automodel.cpp)
 SOURCES += $(wildcard ../../common/AutoModel/modeling_gemma3_text.cpp)
 SOURCES += $(wildcard ../../common/tokenizer/tokenizer.cpp)

src/test/gemma_text_npu/test.cpp

@@ -3,8 +3,17 @@
 #include "utils/utils.hpp"
 #include "utils/vm_args.hpp"
 #include "AutoModel/modeling_gemma3_text.hpp"
+#include "AutoModel/flm_backend.hpp"
 #include "model_list.hpp"
 
+/// \note builtin_backends.cpp names every engine and so links every prebuilt
+///       library; this harness registers just the ones it drives.
+namespace flm::backend {
+void register_builtin_backends(BackendRegistry& registry) {
+    registry.register_backend("gemma3-text", kFlmBackendId, flm_factory<gemma_text_npu>());
+}
+}  // namespace flm::backend
+
 flm_rt::device npu_device_global;
 // Model-specific factory function for Gemma3 Text family only
 inline std::pair<std::string, std::unique_ptr<AutoModel>> get_gemma_text_model(const std::string& model_tag, flm_rt::device* npu_device_inst) {

src/test/gpt_oss_npu/CMakeLists.txt

@@ -11,6 +11,7 @@ add_npu_test(
     USE_TOKENIZER
     USE_SAMPLER
     SOURCES
+        "${CMAKE_SOURCE_DIR}/../../common/AutoModel/model_backend.cpp"
         "${CMAKE_SOURCE_DIR}/../../common/AutoModel/modeling_gpt_oss.cpp"
 )
 

src/test/gpt_oss_npu/Makefile

@@ -16,6 +16,7 @@
 -include ../common.mk
 
 
+SOURCES += ../../common/AutoModel/model_backend.cpp
 SOURCES += $(wildcard ../../common/AutoModel/automodel.cpp)
 SOURCES += $(wildcard ../../common/AutoModel/modeling_gpt_oss.cpp)
 SOURCES += $(wildcard ../../include/AutoModel/automodel.hpp)

src/test/gpt_oss_npu/test.cpp

@@ -7,9 +7,18 @@
 #include "utils/utils.hpp"
 #include "utils/vm_args.hpp"
 #include "AutoModel/modeling_gpt_oss.hpp"
+#include "AutoModel/flm_backend.hpp"
 #include "model_list.hpp"
 #include "utils/vm_args.hpp"
 
+/// \note builtin_backends.cpp names every engine and so links every prebuilt
+///       library; this harness registers just the ones it drives.
+namespace flm::backend {
+void register_builtin_backends(BackendRegistry& registry) {
+    registry.register_backend("gpt-oss", kFlmBackendId, flm_factory<gpt_oss_npu>());
+}
+}  // namespace flm::backend
+
 flm_rt::device npu_device_global;
 
 

src/test/hunyuan_npu/CMakeLists.txt

@@ -11,6 +11,7 @@ add_npu_test(
     USE_TOKENIZER
     USE_SAMPLER
     SOURCES
+        "${CMAKE_SOURCE_DIR}/../../common/AutoModel/model_backend.cpp"
         "${CMAKE_SOURCE_DIR}/../../common/AutoModel/modeling_hunyuan.cpp"
 )
 

src/test/hunyuan_npu/Makefile

@@ -36,6 +36,7 @@ endif
 
 SOURCES += test.cpp
 SOURCES += bench_prefill.cpp
+SOURCES += ../../common/AutoModel/model_backend.cpp
 SOURCES += ../../common/AutoModel/automodel.cpp
 SOURCES += ../../common/AutoModel/modeling_hunyuan.cpp
 SOURCES += ../../common/tokenizer/tokenizer.cpp

src/test/hunyuan_npu/test.cpp

@@ -25,9 +25,18 @@
 #include "utils/utils.hpp"
 #include "utils/vm_args.hpp"
 #include "AutoModel/modeling_hunyuan.hpp"
+#include "AutoModel/flm_backend.hpp"
 #include "model_list.hpp"
 #include "bench_samples.hpp"
 
+/// \note builtin_backends.cpp names every engine and so links every prebuilt
+///       library; this harness registers just the ones it drives.
+namespace flm::backend {
+void register_builtin_backends(BackendRegistry& registry) {
+    registry.register_backend("hunyuan", kFlmBackendId, flm_factory<hunyuan_npu>());
+}
+}  // namespace flm::backend
+
 flm_rt::device npu_device_global;
 
 /// \brief the fixed translator instruction, installed as the system prompt

src/test/lfm2_npu/CMakeLists.txt

@@ -11,6 +11,7 @@ add_npu_test(
     USE_TOKENIZER
     USE_SAMPLER
     SOURCES
+        "${CMAKE_SOURCE_DIR}/../../common/AutoModel/model_backend.cpp"
         "${CMAKE_SOURCE_DIR}/../../common/AutoModel/modeling_lfm2.cpp"
 )
 

src/test/lfm2_npu/Makefile

@@ -16,6 +16,7 @@
 -include ../common.mk
 
 
+SOURCES += ../../common/AutoModel/model_backend.cpp
 SOURCES += $(wildcard ../../common/AutoModel/automodel.cpp)
 SOURCES += $(wildcard ../../common/AutoModel/modeling_lfm2.cpp)
 SOURCES += $(wildcard ../../include/AutoModel/automodel.hpp)

src/test/lfm2_npu/test.cpp

@@ -4,10 +4,20 @@
 #include "utils/utils.hpp"
 #include "utils/vm_args.hpp"
 #include "AutoModel/modeling_lfm2.hpp"
+#include "AutoModel/flm_backend.hpp"
 #include "model_list.hpp"
 #include "utils/vm_args.hpp"
 #include "metrices.hpp"
 
+/// \note builtin_backends.cpp names every engine and so links every prebuilt
+///       library; this harness registers just the ones it drives.
+namespace flm::backend {
+void register_builtin_backends(BackendRegistry& registry) {
+    registry.register_backend("lfm2", kFlmBackendId, flm_factory<lfm2_npu>());
+    registry.register_backend("lfm2.5-tk", kFlmBackendId, flm_factory<lfm2_npu>());
+}
+}  // namespace flm::backend
+
 flm_rt::device npu_device_global;
 
 // Model-specific factory function for Llama family and DeepSeek_r1_8b

src/test/llama_npu/CMakeLists.txt

@@ -11,6 +11,7 @@ add_npu_test(
     USE_TOKENIZER
     USE_SAMPLER
     SOURCES
+        "${CMAKE_SOURCE_DIR}/../../common/AutoModel/model_backend.cpp"
         "${CMAKE_SOURCE_DIR}/../../common/AutoModel/modeling_llama3.cpp"
 )
 

src/test/llama_npu/Makefile

@@ -16,6 +16,7 @@
 -include ../common.mk
 
 
+SOURCES += ../../common/AutoModel/model_backend.cpp
 SOURCES += $(wildcard ../../common/AutoModel/automodel.cpp)
 SOURCES += $(wildcard ../../common/AutoModel/modeling_llama3.cpp)
 SOURCES += $(wildcard ../../common/tokenizer/tokenizer.cpp)

src/test/llama_npu/test.cpp

@@ -7,9 +7,19 @@
 #include "utils/utils.hpp"
 #include "utils/vm_args.hpp"
 #include "AutoModel/modeling_llama3.hpp"
+#include "AutoModel/flm_backend.hpp"
 #include "model_list.hpp"
 #include "utils/vm_args.hpp"
 
+/// \note builtin_backends.cpp names every engine and so links every prebuilt
+///       library; this harness registers just the ones it drives.
+namespace flm::backend {
+void register_builtin_backends(BackendRegistry& registry) {
+    registry.register_backend("llama3", kFlmBackendId, flm_factory<llama_npu>());
+    registry.register_backend("deepseek-r1", kFlmBackendId, flm_factory<llama_npu>());
+}
+}  // namespace flm::backend
+
 flm_rt::device npu_device_global;
 
 int main(int argc, char* argv[]) {

src/test/minicpm_v_4_7_npu/CMakeLists.txt

@@ -4,22 +4,59 @@ project(minicpm_v_4_7_npu VERSION 1.0.0 LANGUAGES CXX)
 include(${CMAKE_CURRENT_LIST_DIR}/../CMakeLists.txt)
 npu_test_setup()
 
+# Find and enable OpenMP for multi-threading
+find_package(OpenMP)
+if(OpenMP_CXX_FOUND)
+    message(STATUS "OpenMP found - enabling multi-threading support for test")
+else()
+    message(WARNING "OpenMP not found - some optimizations will run single-threaded")
+endif()
+
+# Same image sources as the Makefile: see the note there on why imageproc.cpp is
+# linked although MiniCPM resizes with its own PIL-exact code.
 add_npu_test(
     test_minicpm_v_4_7_npu
     test/minicpm_v_4_7_npu
     USE_AUTOMODEL
     USE_TOKENIZER
     USE_SAMPLER
+    USE_AVX512
     SOURCES
         "${CMAKE_SOURCE_DIR}/../../common/AutoModel/model_backend.cpp"
         "${CMAKE_SOURCE_DIR}/../../common/AutoModel/modeling_minicpm_v.cpp"
+        "${CMAKE_SOURCE_DIR}/../../common/AutoModel/modeling_minicpm_v_image.cpp"
+        "${CMAKE_SOURCE_DIR}/../../common/image_process_utils/imageproc.cpp"
+        "${CMAKE_SOURCE_DIR}/../../common/image_process_utils/imageprocAVX512.cpp"
+        "${CMAKE_SOURCE_DIR}/../../common/image/image_reader.cpp"
+)
+
+target_compile_definitions(test_minicpm_v_4_7_npu PUBLIC
+    USEAVX512=1
 )
 
+# Add OpenMP compiler flags if available
+if(OpenMP_CXX_FOUND)
+    target_compile_options(test_minicpm_v_4_7_npu PUBLIC
+        $<$<CXX_COMPILER_ID:MSVC>:/openmp>
+        $<$<NOT:$<CXX_COMPILER_ID:MSVC>>:-fopenmp>
+    )
+endif()
+
 target_link_libraries(test_minicpm_v_4_7_npu PUBLIC
     minicpm_v_4_7_npu
     xrt_coreutil
+    avformat
+    avcodec
+    avutil
+    swscale
+    zlib
 )
 
+# Link OpenMP if available
+if(OpenMP_CXX_FOUND)
+    target_link_libraries(test_minicpm_v_4_7_npu PUBLIC OpenMP::OpenMP_CXX)
+endif()
+
 # Add test target
 add_custom_target(test_minicpm_v_4_7_npu_target
     DEPENDS test_minicpm_v_4_7_npu

src/test/nanbeige_npu/CMakeLists.txt

@@ -11,6 +11,7 @@ add_npu_test(
     USE_TOKENIZER
     USE_SAMPLER
     SOURCES
+        "${CMAKE_SOURCE_DIR}/../../common/AutoModel/model_backend.cpp"
         "${CMAKE_SOURCE_DIR}/../../common/AutoModel/modeling_nanbeige.cpp"
 )