AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

hipHostMalloc allocation are mapped to all devices by default.

Browse Source 
Support hipHostMallocPortable flag.
Default flags are hipHostMallocPortable | hipHostMallocMapped.

Also:
-refactor tests to move addCount and addCountReverse into HipTest
namespace.
-test multi-GPU host memory.
This commit is contained in:
Ben Sander
2017-05-10 17:32:25 -05:00
parent e0c3ea15b2
commit c3ccaa01e5
7 changed files with 181 additions and 96 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/src/runtimeApi/stream/hipStreamWaitEvent.cpp
+2 -38
View File
@@ -35,42 +35,6 @@ unsigned p_count = 100;
template <typename T>
__global__ void
addCount( const T *A_d,
T *C_d,
size_t NELEM,
int count)
{
size_t offset = (hipBlockIdx_x * hipBlockDim_x + hipThreadIdx_x);
size_t stride = hipBlockDim_x * hipGridDim_x ;
// Deliberately do this in an inefficient way to increase kernel runtime
for (int i=0; i<count; i++) {
for (size_t i=offset; i<NELEM; i+=stride) {
C_d[i] = A_d[i] + (T)count;
}
}
}
template <typename T>
__global__ void
addCountReverse( const T *A_d,
T *C_d,
int64_t NELEM,
int count)
{
size_t offset = (hipBlockIdx_x * hipBlockDim_x + hipThreadIdx_x);
size_t stride = hipBlockDim_x * hipGridDim_x ;
// Deliberately do this in an inefficient way to increase kernel runtime
for (int i=0; i<count; i++) {
for (int64_t i=NELEM-stride+offset; i>=0; i-=stride) {
C_d[i] = A_d[i] + (T)count;
}
}
}
//------
@@ -171,9 +135,9 @@ void Streamer<T>::runAsyncAfter(Streamer<T> *depStreamer, bool waitSameStream)
unsigned blocks = HipTest::setNumBlocks(blocksPerCU, threadsPerBlock, _numElements);
if (_reverse) {
hipLaunchKernelGGL(addCountReverse , dim3(blocks), dim3(threadsPerBlock), 0, _stream, _A_d, _C_d, _numElements, p_count);
hipLaunchKernelGGL(HipTest::addCountReverse , dim3(blocks), dim3(threadsPerBlock), 0, _stream, _A_d, _C_d, _numElements, p_count);
} else {
hipLaunchKernelGGL(addCount, dim3(blocks), dim3(threadsPerBlock), 0, _stream, _A_d, _C_d, _numElements, p_count);
hipLaunchKernelGGL(HipTest::addCount, dim3(blocks), dim3(threadsPerBlock), 0, _stream, _A_d, _C_d, _numElements, p_count);
}
HIPCHECK(hipEventRecord(_event, _stream));