fix multiple cards and signle card

Author: NeoZhangJianyu

ggml-sycl.cpp

@@ -5893,6 +5893,8 @@ GGML_CALL static void ggml_backend_sycl_set_tensor_async(ggml_backend_t backend,
 
     GGML_ASSERT(buf->buft == ggml_backend_sycl_buffer_type(sycl_ctx->device) && "unsupported buffer type");
     const queue_ptr stream = sycl_ctx->stream(sycl_ctx->device, 0);
+    printf("zjy ggml_backend_sycl_set_tensor_async sycl_ctx->device=%d stream=%p\n", sycl_ctx->device, stream);
+
     SYCL_CHECK(CHECK_TRY_ERROR((stream)->memcpy(
         (char *)tensor->data + offset, data, size).wait()));
 }
@@ -6156,22 +6158,17 @@ static ggml_guid_t ggml_backend_sycl_guid() {
 
 GGML_CALL ggml_backend_t ggml_backend_sycl_init(int device_id) {
     GGML_SYCL_DEBUG("[SYCL] call ggml_backend_sycl_init, device_id=%d\n", device_id);
-    printf("zjy ggml_backend_sycl_init 1\n");
     check_allow_device_id(device_id);
-    printf("zjy ggml_backend_sycl_init 2\n");
     ggml_backend_sycl_context * ctx = new ggml_backend_sycl_context(ggml_sycl_info(), device_id);
-    printf("zjy ggml_backend_sycl_init 3\n");
     if (ctx == nullptr) {
         fprintf(stderr, "%s: error: failed to allocate context\n", __func__);
         return nullptr;
     };
-    printf("zjy ggml_backend_sycl_init 3\n");
     ggml_backend_t sycl_backend = new ggml_backend {
         /* .guid      = */ ggml_backend_sycl_guid(),
         /* .interface = */ ggml_backend_sycl_interface,
         /* .context   = */ ctx
     };
-    printf("zjy ggml_backend_sycl_init 4\n");
     return sycl_backend;
 }
 

ggml-sycl/common.hpp

@@ -128,6 +128,7 @@ static void crash() {
           "Meet error in this line code!"); \
   } while (0)
 
+
 #if DPCT_COMPAT_RT_VERSION >= 11100
 #define GGML_SYCL_ASSUME(x) __builtin_assume(x)
 #else
@@ -241,22 +242,24 @@ class sycl_device_mgr {
 
     sycl::queue * create_queue_for_device(sycl::context &ctx, sycl::device &device) {
         dpct::select_device(dpct::dev_mgr::instance().get_device_id(device));
-        printf("zjy create_queue_for_device 1\n");
         auto res = dpct::get_current_device().create_queue(ctx, device);
-        printf("zjy create_queue_for_device 4\n");
         return res;
     }
 
     sycl::queue * create_queue_for_device_id(int device_id) {
-        printf("zjy create_queue_for_device_id 0 ctxs.size()=%d\n", ctxs.size());
-
-        sycl::context ctx=ctxs[device_id];
-        printf("zjy create_queue_for_device_id 1\n");
+        int i = get_device_index(device_id);
+        sycl::context ctx=ctxs[i];
         sycl::device device = dpct::dev_mgr::instance().get_device(device_id);;
-        printf("zjy create_queue_for_device_id 2\n");
         return create_queue_for_device(ctx, device);
     }
 
+    int get_device_index(int device_id) {
+        for (int i = 0; i < device_ids.size(); i++) {
+            if (device_ids[i] == device_id) return i;
+        }
+        return -1;
+    }
+
     void create_context_for_devices() {
         for (int i = 0; i < device_ids.size(); i++) {
             sycl::context ctx = sycl::context(devices[i]);
@@ -552,10 +555,9 @@ struct ggml_backend_sycl_context {
     explicit ggml_backend_sycl_context(struct ggml_sycl_device_info &sycl_device_info, int device_id) :
         device(device_id),
         name(GGML_SYCL_NAME + std::to_string(device)) {
-            printf("zjy ggml_backend_sycl_context 1\n");
             for (int i=0;i<GGML_SYCL_MAX_STREAMS; i++){
-                printf("zjy ggml_backend_sycl_context 2 i=%d id=%d\n", i, device_id );
                 qptrs[device_id][i] = sycl_device_info.device_mgr->create_queue_for_device_id(device_id);
+                printf("zjy ggml_backend_sycl_context sycl_ctx->device=%d i=%d stream=%p\n", device_id, i, qptrs[device_id][i]);
             }
     }
 

ggml-sycl/dpct/helper.hpp

@@ -891,9 +891,7 @@ namespace dpct
         void select_device(unsigned int id)
         {
             std::lock_guard<std::recursive_mutex> lock(m_mutex);
-             printf("zjy create_queue_for_device 2\n");
             check_id(id);
-             printf("zjy create_queue_for_device 3\n");
             _thread2dev_map[get_tid()] = id;
         }
         unsigned int device_count() { return _devs.size(); }