xmos · Allan-xmos · Feb 24, 2026 · Feb 25, 2026 · Feb 25, 2026 · Feb 26, 2026
diff --git a/Jenkinsfile b/Jenkinsfile
diff --git a/examples/CMakeLists.txt b/examples/CMakeLists.txt
@@ -2,6 +2,12 @@ cmake_minimum_required(VERSION 3.21)
 include($ENV{XMOS_CMAKE_PATH}/xcommon.cmake)
 project(lib_voice_examples)
 
+if(DEFINED XCORE_TARGET)
+    set(XCORE_TARGET ${XCORE_TARGET})
+else()
+    set(XCORE_TARGET XK-EVK-XU316)
+endif()
+
 add_subdirectory(app_aec)
 add_subdirectory(app_vnr)
 

diff --git a/examples/app_aec/CMakeLists.txt b/examples/app_aec/CMakeLists.txt
@@ -2,7 +2,11 @@ cmake_minimum_required(VERSION 3.21)
 include($ENV{XMOS_CMAKE_PATH}/xcommon.cmake)
 project(app_aec)
 
-set(APP_HW_TARGET XK-EVK-XU316)
+if(DEFINED XCORE_TARGET)
+    set(APP_HW_TARGET ${XCORE_TARGET})
+else()
+    set(APP_HW_TARGET XK-EVK-XU316)
+endif()
 
 if(NOT BUILD_NATIVE)
     set(APP_COMPILER_FLAGS_1th -report -DAEC_THREADS=1)

diff --git a/examples/app_pipeline/CMakeLists.txt b/examples/app_pipeline/CMakeLists.txt
@@ -2,7 +2,12 @@ cmake_minimum_required(VERSION 3.21)
 include($ENV{XMOS_CMAKE_PATH}/xcommon.cmake)
 project(app_pipeline)
 
-set(APP_HW_TARGET XK-EVK-XU316)
+if(DEFINED XCORE_TARGET)
+    set(APP_HW_TARGET ${XCORE_TARGET})
+else()
+    set(APP_HW_TARGET XK-EVK-XU316)
+endif()
+
 set(APP_COMPILER_FLAGS_std_arch -report)
 set(APP_COMPILER_FLAGS_alt_arch -report -DALT_ARCH_MODE=1)
 

diff --git a/examples/app_pipeline/src/app.c b/examples/app_pipeline/src/app.c
@@ -6,17 +6,21 @@
 #include <xcore/channel.h>
 #include <xcore/chanend.h>
 #include <xcore/channel_transaction.h>
+#include <xcore/parallel.h>
 
-extern void pipeline_tile0_init(pipeline_state_tile0_t *state);
-extern void pipeline_tile1_init(pipeline_state_tile1_t *state);
+DECLARE_JOB(pipeline_wrapper_thread0, (chanend_t));
+DECLARE_JOB(pipeline_wrapper_thread1, (chanend_t));
 
-extern void pipeline_process_frame_tile0(pipeline_state_tile0_t *state,
+extern void pipeline_thread0_init(pipeline_state_thread0_t *state);
+extern void pipeline_thread1_init(pipeline_state_thread1_t *state);
+
+extern void pipeline_process_frame_thread0(pipeline_state_thread0_t *state,
         int32_t (*input_y_data)[AP_FRAME_ADVANCE],
         int32_t (*input_x_data)[AP_FRAME_ADVANCE],
         int32_t (*output_data)[AP_FRAME_ADVANCE],
         pipeline_metadata_t *md_output);
 
-extern void pipeline_process_frame_tile1(pipeline_state_tile1_t *state, pipeline_metadata_t *md_input,
+extern void pipeline_process_frame_thread1(pipeline_state_thread1_t *state, pipeline_metadata_t *md_input,
         int32_t (*input_data)[AP_FRAME_ADVANCE],
         int32_t output_data[AP_FRAME_ADVANCE]);
 
@@ -39,43 +43,61 @@ static inline void consumer(int32_t frame_y[AP_FRAME_ADVANCE]) {
     printf("frame done\n");
 }
 
-void pipeline_wrapper_tile0(chanend_t c_pcm_out)
+void pipeline_wrapper_thread0(chanend_t c_pcm_out)
 {
     int32_t DWORD_ALIGNED frame_y[AP_MAX_Y_CHANNELS][AP_FRAME_ADVANCE];
     int32_t DWORD_ALIGNED frame_x[AP_MAX_X_CHANNELS][AP_FRAME_ADVANCE];
 
     // Initialise pipeline
-    pipeline_state_tile0_t DWORD_ALIGNED pipeline_tile0_state;
-    pipeline_tile0_init(&pipeline_tile0_state);
+    pipeline_state_thread0_t DWORD_ALIGNED pipeline_thread0_state;
+    pipeline_thread0_init(&pipeline_thread0_state);
 
     for(unsigned b = 0; b < 5; b++){
         producer(frame_y, frame_x);
 
         pipeline_metadata_t md;
-        int32_t DWORD_ALIGNED tile0_output[AP_MAX_Y_CHANNELS][AP_FRAME_ADVANCE];
-        pipeline_process_frame_tile0(&pipeline_tile0_state, frame_y, frame_x, tile0_output, &md);
+        int32_t DWORD_ALIGNED thread0_output[AP_MAX_Y_CHANNELS][AP_FRAME_ADVANCE];
+        pipeline_process_frame_thread0(&pipeline_thread0_state, frame_y, frame_x, thread0_output, &md);
 
-        // Send data to process to the other tile and receive processed output back
-        //Transfer to other tile
+        // Send data to process to the other thread and receive processed output back
+        //Transfer to other thread
+        chan_out_byte(c_pcm_out, 1);
         chan_out_buf_byte(c_pcm_out, (uint8_t*)&md, sizeof(pipeline_metadata_t));
-        chan_out_buf_word(c_pcm_out, (uint32_t*)&tile0_output[0][0], (AP_MAX_Y_CHANNELS * AP_FRAME_ADVANCE));
+        chan_out_buf_word(c_pcm_out, (uint32_t*)&thread0_output[0][0], (AP_MAX_Y_CHANNELS * AP_FRAME_ADVANCE));
     }
+    chan_out_byte(c_pcm_out, 0);
 }
 
-void pipeline_wrapper_tile1(chanend_t c_pcm_in)
+void pipeline_wrapper_thread1(chanend_t c_pcm_in)
 {
-    pipeline_state_tile1_t DWORD_ALIGNED pipeline_tile1_state;
+    pipeline_state_thread1_t DWORD_ALIGNED pipeline_thread1_state;
     pipeline_metadata_t md;
-    int32_t DWORD_ALIGNED tile0_output[AP_MAX_Y_CHANNELS][AP_FRAME_ADVANCE];
+    int32_t DWORD_ALIGNED thread0_output[AP_MAX_Y_CHANNELS][AP_FRAME_ADVANCE];
     int32_t DWORD_ALIGNED pipeline_output[AP_FRAME_ADVANCE];
 
-    pipeline_tile1_init(&pipeline_tile1_state);
+    pipeline_thread1_init(&pipeline_thread1_state);
     while(1) {
+        uint8_t status = chan_in_byte(c_pcm_in);
+        if (!status) break;
         chan_in_buf_byte(c_pcm_in, (uint8_t*)&md, sizeof(pipeline_metadata_t));
-        chan_in_buf_word(c_pcm_in, (uint32_t*)&tile0_output[0][0], (AP_MAX_Y_CHANNELS * AP_FRAME_ADVANCE));
+        chan_in_buf_word(c_pcm_in, (uint32_t*)&thread0_output[0][0], (AP_MAX_Y_CHANNELS * AP_FRAME_ADVANCE));
 
-        pipeline_process_frame_tile1(&pipeline_tile1_state, &md, tile0_output, pipeline_output);
+        pipeline_process_frame_thread1(&pipeline_thread1_state, &md, thread0_output, pipeline_output);
 
         consumer(pipeline_output);
     }
 }
+
+// producer -> stage1 -> (thread0_to_thread1) -> stage2 -> stage3 -> stage4 -> consumer
+// producer and stage1 run on thread0
+// stage2, stage3, stage4 and consumer run on thread1
+
+int main() {
+    channel_t ch = chan_alloc();
+    PAR_JOBS(
+        PJOB(pipeline_wrapper_thread0, (ch.end_a)),
+        PJOB(pipeline_wrapper_thread1, (ch.end_b))
+    );
+    chan_free(ch);
+    return 0;
+}
diff --git a/examples/app_pipeline/src/main.xc b/examples/app_pipeline/src/main.xc
diff --git a/examples/app_pipeline/src/pipeline.c b/examples/app_pipeline/src/pipeline.c
@@ -5,8 +5,8 @@
 
 #include "pipeline_state.h"
 
-void pipeline_tile0_init(pipeline_state_tile0_t *state) {
-    memset(state, 0, sizeof(pipeline_state_tile0_t));
+void pipeline_thread0_init(pipeline_state_thread0_t *state) {
+    memset(state, 0, sizeof(pipeline_state_thread0_t));
 
     // Initialise AEC, DE, ADEC stages
     aec_conf_t aec_de_mode_conf, aec_non_de_mode_conf;
@@ -44,8 +44,8 @@ void pipeline_tile0_init(pipeline_state_tile0_t *state) {
     stage1_init(&state->stage_1_state, &aec_de_mode_conf, &aec_non_de_mode_conf, &adec_conf);
 }
 
-void pipeline_tile1_init(pipeline_state_tile1_t *state) {
-    memset(state, 0, sizeof(pipeline_state_tile1_t));
+void pipeline_thread1_init(pipeline_state_thread1_t *state) {
+    memset(state, 0, sizeof(pipeline_state_thread1_t));
 
     // Initialise IC, VNR
     ic_init(&state->ic_state);
@@ -58,7 +58,7 @@ void pipeline_tile1_init(pipeline_state_tile1_t *state) {
     agc_init(&state->agc_state, &agc_conf_asr);
 }
 
-void pipeline_process_frame_tile0(pipeline_state_tile0_t *state,
+void pipeline_process_frame_thread0(pipeline_state_thread0_t *state,
         int32_t (*input_y_data)[AP_FRAME_ADVANCE],
         int32_t (*input_x_data)[AP_FRAME_ADVANCE],
         int32_t (*output_data)[AP_FRAME_ADVANCE],
@@ -79,7 +79,7 @@ void pipeline_process_frame_tile0(pipeline_state_tile0_t *state,
     memcpy(md_output, &md, sizeof(pipeline_metadata_t));
 }
 
-void pipeline_process_frame_tile1(pipeline_state_tile1_t *state, pipeline_metadata_t *md_input,
+void pipeline_process_frame_thread1(pipeline_state_thread1_t *state, pipeline_metadata_t *md_input,
         int32_t (*input_data)[AP_FRAME_ADVANCE],
         int32_t output_data[AP_FRAME_ADVANCE])
 {

diff --git a/examples/app_pipeline/src/pipeline_state.h b/examples/app_pipeline/src/pipeline_state.h
@@ -16,7 +16,7 @@ typedef struct {
 typedef struct {
     // Stage1 - AEC, DE, ADEC
     stage1_t DWORD_ALIGNED stage_1_state;
-} pipeline_state_tile0_t;
+} pipeline_state_thread0_t;
 
 typedef struct {
     // IC, VNR
@@ -26,6 +26,6 @@ typedef struct {
     ns_state_t DWORD_ALIGNED ns_state;
     // AGC
     agc_state_t agc_state;
-} pipeline_state_tile1_t;
+} pipeline_state_thread1_t;
 
 #endif
diff --git a/examples/app_vnr/CMakeLists.txt b/examples/app_vnr/CMakeLists.txt
@@ -2,7 +2,11 @@ cmake_minimum_required(VERSION 3.21)
 include($ENV{XMOS_CMAKE_PATH}/xcommon.cmake)
 project(app_vnr)
 
-set(APP_HW_TARGET XK-EVK-XU316)
+if(DEFINED XCORE_TARGET)
+    set(APP_HW_TARGET ${XCORE_TARGET})
+else()
+    set(APP_HW_TARGET XK-EVK-XU316)
+endif()
 
 if(NOT BUILD_NATIVE)
     set(APP_COMPILER_FLAGS -report)

diff --git a/lib_voice/lib_build_info.cmake b/lib_voice/lib_build_info.cmake
@@ -1,20 +1,21 @@
 set(LIB_NAME lib_voice)
 set(LIB_VERSION 1.0.0)
-set(LIB_DEPENDENT_MODULES "lib_xcore_math(2.4.1)")
+set(LIB_DEPENDENT_MODULES "lib_xcore_math(lib_voice_fixes)")
 
 set(LIB_COMPILER_FLAGS
             -g
             -Os
             -DHEADROOM_CHECK=0)
 
-if(BUILD_NATIVE)
-    list(APPEND LIB_COMPILER_FLAGS
-        -D__xtflm_conf_h_exists__
-        -DNN_USE_REF
-    )
-endif()
+# if(BUILD_NATIVE)
+#     list(APPEND LIB_COMPILER_FLAGS
+#         -D__xtflm_conf_h_exists__
+#         -DNN_USE_REF
+#     )
+# endif()
 
 set(LIB_CXX_SRCS "")
+set(lib_ASM_SRCS "")
 include(${CMAKE_CURRENT_LIST_DIR}/vnr_model.cmake)
 file(RELATIVE_PATH MODEL_OUT_DIR_REL ${CMAKE_CURRENT_LIST_DIR} ${MODEL_OUT_DIR})
 
@@ -70,7 +71,7 @@ foreach(target ${APP_BUILD_TARGETS})
 
     # Link aitools with the targets
     target_link_libraries(${target} PRIVATE tflite_micro)
-if(BUILD_NATIVE)
-    target_compile_features(${target} PRIVATE cxx_std_11)
-endif()
+# if(BUILD_NATIVE)
+#     target_compile_features(${target} PRIVATE cxx_std_11)
+# endif()
 endforeach()
diff --git a/lib_voice/src/aec/aec_process_frame.c b/lib_voice/src/aec/aec_process_frame.c
@@ -10,7 +10,7 @@ enum e_fft {Y_FFT, X_FFT, ERROR_FFT};
 #define REF_ACTIVE_THRESHOLD_dB (-60) // Reference input level above which it is considered active
 #define REF_ACTIVE_THRESHOLD f64_to_float_s32(pow(10, REF_ACTIVE_THRESHOLD_dB/20.0))
 
-#ifdef __XS3A__
+#if defined(__XS3A__) || defined(__VX4B__) 
 #include <xcore/parallel.h>
 DECLARE_JOB(calc_time_domain_ema_energy_task, (const aec_par_tasks_and_channels_t*, aec_filter_state_t *, int32_t*, int, int, enum e_td_ema));
 DECLARE_JOB(fft_task, (const aec_par_tasks_and_channels_t*, aec_filter_state_t*, aec_filter_state_t*, int, int, enum e_fft));
@@ -273,7 +273,7 @@ void filter_adapt_task(const aec_par_tasks_t *s, aec_filter_state_t *main_state,
     }
 }
 
-#ifdef __XS3A__
+#if defined(__XS3A__) || defined(__VX4B__) 
 
 #define PAR_THREADS_PJOBS(FUNC, ARR, NUM_THREADS, ...)        \
 do {                                                            \

diff --git a/lib_voice/vnr_model.cmake b/lib_voice/vnr_model.cmake
@@ -13,7 +13,15 @@ execute_process(
 )
 
 # Add tflite_micro
-set(XMOS_AITOOLSLIB_PATH_CMAKE "${XMOS_AITOOLSLIB_PATH}/buildfiles/aitoolslib.cmake")
+if (APP_BUILD_ARCH STREQUAL "xs3a")
+    set(XMOS_AITOOLSLIB_PATH_CMAKE "${XMOS_AITOOLSLIB_PATH}/buildfiles/aitoolslib.cmake")
+    set(MODEL_TH 0.50)
+    set(ARCH_STR "XS3A")
+elseif (APP_BUILD_ARCH STREQUAL "vx4b")
+    set(XMOS_AITOOLSLIB_PATH_CMAKE "${CMAKE_CURRENT_LIST_DIR}/../new_ai_tools/libxtflitemicro.cmake")
+    set(MODEL_TH 2)
+    set(ARCH_STR "VX4A")
+endif()
 
 if(XMOS_AITOOLSLIB_PATH STREQUAL "")
     message(FATAL_ERROR "Path to XMOS AI tools NOT found")
@@ -38,12 +46,11 @@ set(MODEL_OUT_DIR ${CMAKE_CURRENT_BINARY_DIR}/src.autogen/vnr_model/)
 set(MODEL_IN_PATH ${CMAKE_CURRENT_LIST_DIR}/src/vnr/model/trained_model.tflite)
 set(MODEL_OUT_PATH ${MODEL_OUT_DIR}/trained_model_xcore.tflite)
 set(MODEL_N_CORES 1)
-set(MODEL_TH 0.50)
 
 file(MAKE_DIRECTORY ${MODEL_OUT_DIR})
 
 add_custom_command(
     OUTPUT ${MODEL_OUT_PATH}.cpp ${MODEL_OUT_PATH}.h ${MODEL_OUT_PATH}
-    COMMAND xcore-opt ${MODEL_IN_PATH} -tc ${MODEL_N_CORES} -o ${MODEL_OUT_PATH} --xcore-conv-err-threshold ${MODEL_TH}
+    COMMAND xcore-opt ${MODEL_IN_PATH} -tc ${MODEL_N_CORES} -o ${MODEL_OUT_PATH} --xcore-conv-err-threshold ${MODEL_TH} --xcore-target-arch=${ARCH_STR}
     DEPENDS ${MODEL_IN_PATH}
 )
diff --git a/new_ai_tools/libxtflitemicro.cmake b/new_ai_tools/libxtflitemicro.cmake
@@ -0,0 +1,18 @@
+set(XMOS_AITOOLSLIB_DEFINITIONS
+    "TF_LITE_STATIC_MEMORY"
+    "TF_LITE_STRIP_ERROR_STRINGS"
+    "XCORE"
+    "NO_INTERPRETER"
+)
+
+set(XMOS_AITOOLSLIB_LIBRARIES "${CMAKE_CURRENT_LIST_DIR}/libxtflitemicro_vx4a.a")
+set(XMOS_AITOOLSLIB_INCLUDES "${XMOS_AITOOLSLIB_PATH}/include")
+
+
+# list(APPEND APP_COMPILER_FLAGS
+    # -Wfptrgroup
+    # -ffunction-sections
+    # -fdata-sections
+    # -Wl,--gc-sections
+    # -D__VX4A__=1
+# )
diff --git a/new_ai_tools/libxtflitemicro_vx4a.a b/new_ai_tools/libxtflitemicro_vx4a.a
diff --git a/requirements.txt b/requirements.txt
@@ -1,4 +1,4 @@
 # python_version 3.11
 # pip_version 25.*
 
-xmos-ai-tools==1.4.2
+xmos-ai-tools==1.4.3.dev18
diff --git a/tests/lib_aec/aec_unit_tests/CMakeLists.txt b/tests/lib_aec/aec_unit_tests/CMakeLists.txt
@@ -2,7 +2,12 @@ cmake_minimum_required(VERSION 3.21)
 include($ENV{XMOS_CMAKE_PATH}/xcommon.cmake)
 project(aec_unit_tests)
 
-set(APP_HW_TARGET XK-EVK-XU316)
+if(DEFINED XCORE_TARGET)
+    set(APP_HW_TARGET ${XCORE_TARGET})
+else()
+    set(APP_HW_TARGET XK-EVK-XU316)
+endif()
+
 set(XMOS_SANDBOX_DIR ${CMAKE_CURRENT_LIST_DIR}/../../../../)
 include(${XMOS_SANDBOX_DIR}/lib_voice/tests/etc/build_options.cmake)