Added explicit memory copy direction apis

- Fixed stale printf in context api
- Added 4 sync memcpy apis
  1. hipMemcpyHtoD
  2. hipMemcpyDtoH
  3. hipMemcpyDtoD
  4. hipMemcpyHtoH
- Added test for added apis

Change-Id: I4a9c382445b62631f8d0bcbb9a670322288b72b1


[ROCm/clr commit: f722a132a4]

projects/clr/hipamd/include/hcc_detail/hip_hcc.h

@@ -421,8 +421,9 @@ typedef uint64_t SeqNum_t ;
     ~ihipStream_t();
 
     // kind is hipMemcpyKind
-    void copySync (LockedAccessor_StreamCrit_t &crit, void* dst, const void* src, size_t sizeBytes, unsigned kind);
-    void locked_copySync (void* dst, const void* src, size_t sizeBytes, unsigned kind);
+    void copySync (LockedAccessor_StreamCrit_t &crit, void* dst, const void* src, size_t sizeBytes, unsigned kind, bool resolveOn = true);
+    void locked_copySync (void* dst, const void* src, size_t sizeBytes, unsigned kind, bool resolveOn = true);
+
 
     void copyAsync(void* dst, const void* src, size_t sizeBytes, unsigned kind);
 

projects/clr/hipamd/include/hcc_detail/hip_runtime_api.h

@@ -844,6 +844,14 @@ hipError_t hipHostFree(void* ptr);
  */
 hipError_t hipMemcpy(void* dst, const void* src, size_t sizeBytes, hipMemcpyKind kind);
 
+hipError_t hipMemcpyHtoD(hipDeviceptr dst, hipDeviceptr src, size_t sizeBytes);
+
+hipError_t hipMemcpyDtoH(hipDeviceptr dst, hipDeviceptr src, size_t sizeBytes);
+
+hipError_t hipMemcpyDtoD(hipDeviceptr dst, hipDeviceptr src, size_t sizeBytes);
+
+hipError_t hipMemcpyHtoH(hipDeviceptr dst, hipDeviceptr src, size_t sizeBytes);
+
 
 /**
  *  @brief Copies @p sizeBytes bytes from the memory area pointed to by @p src to the memory area pointed to by @p offset bytes from the start of symbol @p symbol.

projects/clr/hipamd/src/hip_context.cpp

@@ -32,7 +32,6 @@ thread_local std::stack<ihipCtx_t *>  tls_ctxStack;
 hipError_t ihipCtxStackUpdate()
 {
     //HIP_INIT_API();
-    printf("Inside ihipCtxStackUpdate\n");
     hipError_t e = hipSuccess;
 
     if(tls_ctxStack.empty()) {

projects/clr/hipamd/src/hip_hcc.cpp

@@ -1563,7 +1563,7 @@ void ihipStream_t::setAsyncCopyAgents(unsigned kind, ihipCommand_t *commandType,
 }
 
 
-void ihipStream_t::copySync(LockedAccessor_StreamCrit_t &crit, void* dst, const void* src, size_t sizeBytes, unsigned kind)
+void ihipStream_t::copySync(LockedAccessor_StreamCrit_t &crit, void* dst, const void* src, size_t sizeBytes, unsigned kind, bool resolveOn)
 {
     ihipCtx_t *ctx = this->getCtx();
     const ihipDevice_t *device = ctx->getDevice();
@@ -1582,7 +1582,7 @@ void ihipStream_t::copySync(LockedAccessor_StreamCrit_t &crit, void* dst, const
     bool dstInDeviceMem = dstPtrInfo._isInDeviceMem;
 
     // Resolve default to a specific Kind so we know which algorithm to use:
-    if (kind == hipMemcpyDefault) {
+    if (kind == hipMemcpyDefault && resolveOn) {
         kind = resolveMemcpyDirection(srcTracked, dstTracked, srcInDeviceMem, dstInDeviceMem);
     };
 
@@ -1753,14 +1753,12 @@ void ihipStream_t::copySync(LockedAccessor_StreamCrit_t &crit, void* dst, const
 
 
 // Sync copy that acquires lock:
-void ihipStream_t::locked_copySync(void* dst, const void* src, size_t sizeBytes, unsigned kind)
+void ihipStream_t::locked_copySync(void* dst, const void* src, size_t sizeBytes, unsigned kind, bool resolveOn)
 {
     LockedAccessor_StreamCrit_t crit (_criticalData);
-    copySync(crit, dst, src, sizeBytes, kind);
+    copySync(crit, dst, src, sizeBytes, kind, resolveOn);
 }
 
-
-
 void ihipStream_t::copyAsync(void* dst, const void* src, size_t sizeBytes, unsigned kind)
 {
     LockedAccessor_StreamCrit_t crit(_criticalData);

projects/clr/hipamd/src/hip_memory.cpp

@@ -446,12 +446,96 @@ hipError_t hipMemcpy(void* dst, const void* src, size_t sizeBytes, hipMemcpyKind
         e = ex._code;
     }
 
+    return ihipLogStatus(e);
+}
 
+hipError_t hipMemcpyHtoD(hipDeviceptr dst, hipDeviceptr src, size_t sizeBytes)
+{
+    HIP_INIT_API(dst, src, sizeBytes);
+
+    hipStream_t stream = ihipSyncAndResolveStream(hipStreamNull);
+
+    hc::completion_future marker;
+
+    hipError_t e = hipSuccess;
+
+    try {
+
+        stream->locked_copySync((void*)dst, (void*)src, sizeBytes, hipMemcpyHostToDevice, false);
+    }
+    catch (ihipException ex) {
+        e = ex._code;
+    }
 
     return ihipLogStatus(e);
 }
 
 
+hipError_t hipMemcpyDtoH(hipDeviceptr dst, hipDeviceptr src, size_t sizeBytes)
+{
+    HIP_INIT_API(dst, src, sizeBytes);
+
+    hipStream_t stream = ihipSyncAndResolveStream(hipStreamNull);
+
+    hc::completion_future marker;
+
+    hipError_t e = hipSuccess;
+
+    try {
+
+        stream->locked_copySync((void*)dst, (void*)src, sizeBytes, hipMemcpyDeviceToHost, false);
+    }
+    catch (ihipException ex) {
+        e = ex._code;
+    }
+
+    return ihipLogStatus(e);
+}
+
+hipError_t hipMemcpyDtoD(hipDeviceptr dst, hipDeviceptr src, size_t sizeBytes)
+{
+    HIP_INIT_API(dst, src, sizeBytes);
+
+    hipStream_t stream = ihipSyncAndResolveStream(hipStreamNull);
+
+    hc::completion_future marker;
+
+    hipError_t e = hipSuccess;
+
+    try {
+
+        stream->locked_copySync((void*)dst, (void*)src, sizeBytes, hipMemcpyDeviceToDevice, false);
+    }
+    catch (ihipException ex) {
+        e = ex._code;
+    }
+
+    return ihipLogStatus(e);
+}
+
+hipError_t hipMemcpyHtoH(hipDeviceptr dst, hipDeviceptr src, size_t sizeBytes)
+{
+    HIP_INIT_API(dst, src, sizeBytes);
+
+    hipStream_t stream = ihipSyncAndResolveStream(hipStreamNull);
+
+    hc::completion_future marker;
+
+    hipError_t e = hipSuccess;
+
+    try {
+
+        stream->locked_copySync((void*)dst, (void*)src, sizeBytes, hipMemcpyHostToHost, false);
+    }
+    catch (ihipException ex) {
+        e = ex._code;
+    }
+
+    return ihipLogStatus(e);
+}
+
+
+
 /**
  * @result #hipSuccess, #hipErrorInvalidDevice, #hipErrorInvalidMemcpyDirection,
  * @result #hipErrorInvalidValue : If dst==NULL or src==NULL, or other bad argument.

projects/clr/hipamd/tests/src/hipDrvMemcpy.cpp

@@ -0,0 +1,26 @@
+#include<iostream>
+#include<hip/hip_runtime.h>
+#include<hip/hip_runtime_api.h>
+
+#define LEN 1024
+#define SIZE LEN<<2
+
+int main(){
+  int *A, *B, *C;
+  hipDeviceptr Ad, Bd;
+  A = new int[LEN];
+  B = new int[LEN];
+  C = new int[LEN];
+  for(int i=0;i<LEN;i++){
+    A[i] = i;
+  }
+  hipMalloc((void**)&Ad, SIZE);
+  hipMalloc((void**)&Bd, SIZE);
+  hipMemcpyHtoD(Ad, A, SIZE);
+  hipMemcpyDtoD(Bd, Ad, SIZE);
+  hipMemcpyDtoH(B, Bd, SIZE);
+  hipMemcpyHtoH(C, B,SIZE);
+  for(int i=0;i<16;i++){
+    std::cout<<A[i]<<" "<<C[i]<<std::endl;
+  }
+}