Merge pull request #949 from gargrahul/single_stream_concurrent_kernels

Add extension for kernel concurrency on same stream

[ROCm/hip commit: 352b17346c]

projects/hip/src/hip_module.cpp

@@ -130,7 +130,7 @@ hipError_t ihipModuleLaunchKernel(hipFunction_t f, uint32_t globalWorkSizeX,
                                   uint32_t localWorkSizeX, uint32_t localWorkSizeY,
                                   uint32_t localWorkSizeZ, size_t sharedMemBytes,
                                   hipStream_t hStream, void** kernelParams, void** extra,
-                                  hipEvent_t startEvent, hipEvent_t stopEvent) {
+                                  hipEvent_t startEvent, hipEvent_t stopEvent, uint32_t flags) {
     auto ctx = ihipGetTlsDefaultCtx();
     hipError_t ret = hipSuccess;
 
@@ -202,8 +202,11 @@ hipError_t ihipModuleLaunchKernel(hipFunction_t f, uint32_t globalWorkSizeX,
         aql.kernel_object = f->_object;
         aql.setup = 3 << HSA_KERNEL_DISPATCH_PACKET_SETUP_DIMENSIONS;
         aql.header =
-            (HSA_PACKET_TYPE_KERNEL_DISPATCH << HSA_PACKET_HEADER_TYPE) |
-            (1 << HSA_PACKET_HEADER_BARRIER);  // TODO - honor queue setting for execute_in_order
+            (HSA_PACKET_TYPE_KERNEL_DISPATCH << HSA_PACKET_HEADER_TYPE);
+        if((flags & 0x1)== 0 ) {
+            //in_order
+            aql.header |= (1 << HSA_PACKET_HEADER_BARRIER);
+        }
 
         if (HCC_OPT_FLUSH) {
             aql.header |= (HSA_FENCE_SCOPE_AGENT << HSA_PACKET_HEADER_ACQUIRE_FENCE_SCOPE) |
@@ -250,9 +253,21 @@ hipError_t hipModuleLaunchKernel(hipFunction_t f, uint32_t gridDimX, uint32_t gr
                  hStream, kernelParams, extra);
     return ihipLogStatus(ihipModuleLaunchKernel(
         f, blockDimX * gridDimX, blockDimY * gridDimY, gridDimZ * blockDimZ, blockDimX, blockDimY,
-        blockDimZ, sharedMemBytes, hStream, kernelParams, extra, nullptr, nullptr));
+        blockDimZ, sharedMemBytes, hStream, kernelParams, extra, nullptr, nullptr, 0));
 }
 
+hipError_t hipExtModuleLaunchKernel(hipFunction_t f, uint32_t globalWorkSizeX,
+                                    uint32_t globalWorkSizeY, uint32_t globalWorkSizeZ,
+                                    uint32_t localWorkSizeX, uint32_t localWorkSizeY,
+                                    uint32_t localWorkSizeZ, size_t sharedMemBytes,
+                                    hipStream_t hStream, void** kernelParams, void** extra,
+                                    hipEvent_t startEvent, hipEvent_t stopEvent, uint32_t flags) {
+    HIP_INIT_API(hipHccModuleLaunchKernel, f, globalWorkSizeX, globalWorkSizeY, globalWorkSizeZ, localWorkSizeX,
+                 localWorkSizeY, localWorkSizeZ, sharedMemBytes, hStream, kernelParams, extra);
+    return ihipLogStatus(ihipModuleLaunchKernel(
+        f, globalWorkSizeX, globalWorkSizeY, globalWorkSizeZ, localWorkSizeX, localWorkSizeY,
+        localWorkSizeZ, sharedMemBytes, hStream, kernelParams, extra, startEvent, stopEvent, flags));
+}
 
 hipError_t hipHccModuleLaunchKernel(hipFunction_t f, uint32_t globalWorkSizeX,
                                     uint32_t globalWorkSizeY, uint32_t globalWorkSizeZ,
@@ -264,7 +279,7 @@ hipError_t hipHccModuleLaunchKernel(hipFunction_t f, uint32_t globalWorkSizeX,
                  localWorkSizeY, localWorkSizeZ, sharedMemBytes, hStream, kernelParams, extra);
     return ihipLogStatus(ihipModuleLaunchKernel(
         f, globalWorkSizeX, globalWorkSizeY, globalWorkSizeZ, localWorkSizeX, localWorkSizeY,
-        localWorkSizeZ, sharedMemBytes, hStream, kernelParams, extra, startEvent, stopEvent));
+        localWorkSizeZ, sharedMemBytes, hStream, kernelParams, extra, startEvent, stopEvent, 0));
 }
 
 namespace hip_impl {