Added new Memory API's (#1399)

Added new memory API's hipMemAllocPitch, hipMemAllocHost, hipMemsetD16, hipMemsetD16Async, hipMemsetD8Async
Modified to support all scenarios hipMemcpyParam2DAsync, hipMemcpyParam2D.


[ROCm/clr commit: 96890792b0]
Este commit está contenido en:
ansurya
2019-10-04 13:36:31 +05:30
cometido por Maneesh Gupta
padre e5a2ba9602
commit efc64a4f8a
Se han modificado 12 ficheros con 404 adiciones y 48 borrados
@@ -21,7 +21,7 @@ THE SOFTWARE.
*/
/* HIT_START
* BUILD: %t %s ../../test_common.cpp EXCLUDE_HIP_PLATFORM all
* BUILD: %t %s ../../test_common.cpp EXCLUDE_HIP_PLATFORM nvcc
* TEST: %t
* HIT_END
*/
@@ -34,6 +34,21 @@ void printSep() {
"======================================================================================\n");
}
inline void initMemCpyParam2D(hip_Memcpy2D &ins, const size_t dpitch,
const size_t spitch, const size_t width,
const size_t height, hipMemoryType dstType,
enum hipMemoryType srcType) {
ins.srcXInBytes=0;
ins.srcY=0;
ins.srcPitch=spitch;
ins.dstXInBytes=0;
ins.dstY=0;
ins.dstPitch=dpitch;
ins.WidthInBytes=width;
ins.Height=height;
ins.dstMemoryType= dstType;
ins.srcMemoryType= srcType;
}
//---
// Test copies of a matrix numW by numH
@@ -65,7 +80,11 @@ void memcpy2Dtest(size_t numW, size_t numH, bool usePinnedHost) {
unsigned blocks = HipTest::setNumBlocks(blocksPerCU, threadsPerBlock, numW * numH);
HIPCHECK(hipMemcpy2D(A_d, pitch_A, A_h, width, width, numH, hipMemcpyHostToDevice));
HIPCHECK(hipMemcpy2D(B_d, pitch_B, B_h, width, width, numH, hipMemcpyHostToDevice));
hip_Memcpy2D ins;
initMemCpyParam2D(ins,pitch_B,width,width,numH,hipMemoryTypeDevice,hipMemoryTypeHost);
ins.dstDevice = (hipDeviceptr_t)B_d;
ins.srcHost = B_h;
HIPCHECK(hipMemcpyParam2D(&ins));
hipLaunchKernelGGL(HipTest::vectorADD, dim3(blocks), dim3(threadsPerBlock), 0, 0, A_d, B_d, C_d,
(pitch_C / sizeof(T)) * numH);
@@ -115,7 +134,11 @@ void memcpyArraytest(size_t numW, size_t numH, bool usePinnedHost, bool usePitch
unsigned blocks = HipTest::setNumBlocks(blocksPerCU, threadsPerBlock, numW * numH);
HIPCHECK(hipMemcpyToArray(A_d, 0, 0, (void*)A_h, width, hipMemcpyHostToDevice));
HIPCHECK(hipMemcpyToArray(B_d, 0, 0, (void*)B_h, width, hipMemcpyHostToDevice));
hip_Memcpy2D ins;
initMemCpyParam2D(ins,width,width,width,numH,hipMemoryTypeArray,hipMemoryTypeHost);
ins.dstArray = B_d;
ins.srcHost = B_h;
HIPCHECK(hipMemcpyParam2D(&ins));
hipLaunchKernelGGL(HipTest::vectorADD, dim3(blocks), dim3(threadsPerBlock), 0, 0,
(T*)A_d->data, (T*)B_d->data, (T*)C_d->data, numW);
@@ -152,15 +175,21 @@ void memcpyArraytest(size_t numW, size_t numH, bool usePinnedHost, bool usePitch
} else {
HIPCHECK(hipMemcpy2DToArray(A_d, 0, 0, (void*)A_h, width, width, numH,
hipMemcpyHostToDevice));
HIPCHECK(hipMemcpy2DToArray(B_d, 0, 0, (void*)B_h, width, width, numH,
hipMemcpyHostToDevice));
hip_Memcpy2D ins;
initMemCpyParam2D(ins,width,width,width,numH,hipMemoryTypeArray,hipMemoryTypeHost);
ins.dstArray = B_d;
ins.srcHost = B_h;
HIPCHECK(hipMemcpyParam2D(&ins));
}
hipLaunchKernelGGL(HipTest::vectorADD, dim3(blocks), dim3(threadsPerBlock), 0, 0,
(T*)A_d->data, (T*)B_d->data, (T*)C_d->data, numW * numH);
HIPCHECK(hipMemcpy2D((void*)C_h, width, (void*)C_d->data, width, width, numH,
hipMemcpyDeviceToHost));
printf("memcpy srcArray to dstHost\n");
hip_Memcpy2D ins;
initMemCpyParam2D(ins,width,width,width,numH,hipMemoryTypeHost,hipMemoryTypeArray);
ins.srcArray = C_d;
ins.dstHost = C_h;
HIPCHECK(hipMemcpyParam2D(&ins));
HIPCHECK(hipDeviceSynchronize());
HipTest::checkVectorADD(A_h, B_h, C_h, numW * numH);
@@ -227,8 +256,6 @@ int main(int argc, char* argv[]) {
printf("\n\n=== tests&1 (types)\n");
printSep();
HIPCHECK(hipDeviceReset());
size_t width = N / 6;
size_t height = N / 6;
memcpy2Dtest<float>(321, 211, 0);
memcpy2Dtest<double>(322, 211, 0);
memcpy2Dtest<char>(320, 211, 0);