feat: sync whisper.cpp (enable ggml-alloc) (#123)

* feat: test metal-and-alloc of whisper.cpp

* fix(ios): temp workaround for -fno-objc-arc

* fix(android): add ggml-alloc

* feat(ios): still use ARC for ggml-metal.m

* feat(example, ios): add extended virtual addressing entitlement

* feat: update whisper.cpp branch

* feat: update whisper.cpp branch

* feat: sync whisper.cpp

* feat(ios): temp disable metal

* feat(ios): use MLComputeUnitsAll by patch

android/src/main/CMakeLists.txt

@@ -8,6 +8,7 @@ set(RNWHISPER_LIB_DIR ${CMAKE_SOURCE_DIR}/../../../cpp)
 set(
     SOURCE_FILES
     ${RNWHISPER_LIB_DIR}/ggml.c
+    ${RNWHISPER_LIB_DIR}/ggml-alloc.c
     ${RNWHISPER_LIB_DIR}/whisper.cpp
     ${RNWHISPER_LIB_DIR}/rn-whisper.cpp
     ${CMAKE_SOURCE_DIR}/jni.cpp
@@ -21,7 +22,7 @@ function(build_library target_name)
         SHARED
         ${SOURCE_FILES}
     )
-    
+
     target_link_libraries(${target_name} ${LOG_LIB} android)
 
     if (${target_name} STREQUAL "whisper_v8fp16_va")

cpp/coreml/whisper-encoder.mm

@@ -22,7 +22,13 @@
 
     NSURL * url_model = [NSURL fileURLWithPath: path_model_str];
 
-    const void * data = CFBridgingRetain([[whisper_encoder_impl alloc] initWithContentsOfURL:url_model error:nil]);
+    // select which device to run the Core ML model on
+    MLModelConfiguration *config = [[MLModelConfiguration alloc] init];
+    //config.computeUnits = MLComputeUnitsCPUAndGPU;
+    //config.computeUnits = MLComputeUnitsCPUAndNeuralEngine;
+    config.computeUnits = MLComputeUnitsAll;
+
+    const void * data = CFBridgingRetain([[whisper_encoder_impl alloc] initWithContentsOfURL:url_model configuration:config error:nil]);
 
     if (data == NULL) {
         return NULL;