step6, enhance error check, remove CUDA macro, enhance device id to fix none-zero id issue

CMakeLists.txt

@@ -448,11 +448,11 @@ endif()
 
 if (LLAMA_SYCL)
     set(ENABLE_AOT ats)
-    if (NOT ${CMAKE_C_COMPILER_ID} MATCHES "Clang")
-        message(WARNING "Only LLVM is supported for SYCL")
+    if (NOT ${CMAKE_C_COMPILER_ID} MATCHES "Intel")
+	    message(WARNING "${CMAKE_C_COMPILER_ID} Need IntelLLVM for SYCL")
     endif()
-    if (NOT ${CMAKE_CXX_COMPILER_ID} MATCHES "Clang")
-        message(WARNING "Only LLVM is supported for SYCL")
+    if (NOT ${CMAKE_CXX_COMPILER_ID} MATCHES "Intel")
+        message(WARNING "${CMAKE_CXX_COMPILER_ID} Need IntelLLVM for SYCL")
     endif()
 
     #find_package(SYCL REQUIRED)
@@ -476,6 +476,7 @@ if (LLAMA_SYCL)
 	    add_compile_definitions(GGML_USE_CUBLAS)
 	    #add_compile_definitions(GGML_SYCL_F16)
 	    #add_compile_options(-std=c++17 -O3 -fsycl)
+	add_compile_options(-I./)
         add_compile_options(-I/opt/intel/oneapi/compiler/2024.0/include)
         add_compile_options(-I/opt/intel/oneapi/compiler/2024.0/include/sycl)
         add_compile_options(-I/opt/intel/oneapi/dpcpp-ct/2024.0/include)

ggml-sycl.hpp

@@ -1,4 +1,17 @@
 #include <sycl/sycl.hpp>
 #include <dpct/dpct.hpp>
-typedef half ggml_fp16_t;
+// typedef sycl::half ggml_fp16_t;
 
+#define CHECK_TRY_ERROR(expr)                                                  \
+  [&]() {                                                                      \
+    try {                                                                      \
+      expr;                                                                    \
+      return dpct::success;                                                    \
+    } catch (std::exception const &e) {                                        \
+      std::cerr << e.what()<< "\nException caught at file:" << __FILE__        \
+        << ", line:" << __LINE__ <<", func:"<<__func__<< std::endl;            \
+      return dpct::default_error;                                              \
+    }                                                                          \
+  }()
+
+#define DEBUG_CUDA_MALLOC

run.sh

@@ -11,7 +11,7 @@ else
   export GGML_SYCL_DEVICE=0
 fi
 echo GGML_SYCL_DEVICE=$GGML_SYCL_DEVICE
-#export GGML_SYCL_DEBUG=1
+export GGML_SYCL_DEBUG=1
 #export GGML_SYCL_LIST_DEVICE=1
 #./build/bin/main -m models/llama-2-7b.Q4_0.gguf -p "${INPUT1}" -e -n 400 -ngl 33 -c 2048
 ./build/bin/main -m models/llama-2-7b.Q4_0.gguf -p "${INPUT2}" -n 400 -e -ngl 33

setup.sh

@@ -6,4 +6,4 @@ source /opt/intel/oneapi/setvars.sh
 
 #cmake .. -DLLAMA_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DGGML_SYCL_F16=ON
 cmake .. -DLLAMA_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx
-cmake --build . --config Release -v
+cmake --build . --config Release --target main