Merge pull request #613 from InfiniTensor/issue/612

issue/612 - differentiate mempcy sync/async

include/infinicore/context/context.hpp

@@ -25,9 +25,9 @@ std::shared_ptr<Memory> allocateMemory(size_t size);
 std::shared_ptr<Memory> allocateHostMemory(size_t size);
 std::shared_ptr<Memory> allocatePinnedHostMemory(size_t size);
 
-void memcpyH2D(void *dst, const void *src, size_t size);
+void memcpyH2D(void *dst, const void *src, size_t size, bool async = true);
 void memcpyD2H(void *dst, const void *src, size_t size);
-void memcpyD2D(void *dst, const void *src, size_t size);
+void memcpyD2D(void *dst, const void *src, size_t size, bool async = true);
 void memcpyH2H(void *dst, const void *src, size_t size);
 
 // Timing APIs for performance measurement

src/infinicore/context/context_impl.cc

@@ -129,16 +129,16 @@ std::shared_ptr<Memory> allocatePinnedHostMemory(size_t size) {
     return ContextImpl::singleton().getCurrentRuntime()->allocatePinnedHostMemory(size);
 }
 
-void memcpyH2D(void *dst, const void *src, size_t size) {
-    return ContextImpl::singleton().getCurrentRuntime()->memcpyH2D(dst, src, size);
+void memcpyH2D(void *dst, const void *src, size_t size, bool async) {
+    return ContextImpl::singleton().getCurrentRuntime()->memcpyH2D(dst, src, size, async);
 }
 
 void memcpyD2H(void *dst, const void *src, size_t size) {
     return ContextImpl::singleton().getCurrentRuntime()->memcpyD2H(dst, src, size);
 }
 
-void memcpyD2D(void *dst, const void *src, size_t size) {
-    return ContextImpl::singleton().getCurrentRuntime()->memcpyD2D(dst, src, size);
+void memcpyD2D(void *dst, const void *src, size_t size, bool async) {
+    return ContextImpl::singleton().getCurrentRuntime()->memcpyD2D(dst, src, size, async);
 }
 
 void memcpyH2H(void *dst, const void *src, size_t size) {

src/infinicore/context/runtime/runtime.cc

@@ -76,16 +76,24 @@ std::shared_ptr<Memory> Runtime::allocatePinnedHostMemory(size_t size) {
         true);
 }
 
-void Runtime::memcpyH2D(void *dst, const void *src, size_t size) {
-    INFINICORE_CHECK_ERROR(infinirtMemcpyAsync(dst, src, size, INFINIRT_MEMCPY_H2D, stream_));
+void Runtime::memcpyH2D(void *dst, const void *src, size_t size, bool async) {
+    if (async) {
+        INFINICORE_CHECK_ERROR(infinirtMemcpyAsync(dst, src, size, INFINIRT_MEMCPY_H2D, stream_));
+    } else {
+        INFINICORE_CHECK_ERROR(infinirtMemcpy(dst, src, size, INFINIRT_MEMCPY_H2D));
+    }
 }
 
 void Runtime::memcpyD2H(void *dst, const void *src, size_t size) {
     INFINICORE_CHECK_ERROR(infinirtMemcpy(dst, src, size, INFINIRT_MEMCPY_D2H));
 }
 
-void Runtime::memcpyD2D(void *dst, const void *src, size_t size) {
-    INFINICORE_CHECK_ERROR(infinirtMemcpyAsync(dst, src, size, INFINIRT_MEMCPY_D2D, stream_));
+void Runtime::memcpyD2D(void *dst, const void *src, size_t size, bool async) {
+    if (async) {
+        INFINICORE_CHECK_ERROR(infinirtMemcpyAsync(dst, src, size, INFINIRT_MEMCPY_D2D, stream_));
+    } else {
+        INFINICORE_CHECK_ERROR(infinirtMemcpy(dst, src, size, INFINIRT_MEMCPY_D2D));
+    }
 }
 
 // Timing method implementations

src/infinicore/context/runtime/runtime.hpp

@@ -34,9 +34,9 @@ public:
     std::shared_ptr<Memory> allocateMemory(size_t size);
     std::shared_ptr<Memory> allocatePinnedHostMemory(size_t size);
 
-    void memcpyH2D(void *dst, const void *src, size_t size);
+    void memcpyH2D(void *dst, const void *src, size_t size, bool async = true);
     void memcpyD2H(void *dst, const void *src, size_t size);
-    void memcpyD2D(void *dst, const void *src, size_t size);
+    void memcpyD2D(void *dst, const void *src, size_t size, bool async = true);
 
     // Timing methods
     infinirtEvent_t createEvent();