From d8e323d4b5b27d1bc2714b611e88ecb020d9ad5d Mon Sep 17 00:00:00 2001 From: Alex Voicu Date: Wed, 8 Nov 2017 00:08:55 +0000 Subject: [PATCH] Clean up trailing whitespace so as to reduce noise in #246. --- include/hip/hcc_detail/hip_runtime.h | 4 +- src/hip_memory.cpp | 18 ++--- tests/src/runtimeApi/event/record_event.cpp | 20 +++--- tests/src/runtimeApi/memory/hipMemcpy.cpp | 66 +++++++++---------- .../multiThread/hipMultiThreadStreams1.cpp | 4 +- tests/src/runtimeApi/stream/hipNullStream.cpp | 10 +-- .../src/runtimeApi/stream/hipStreamSync2.cpp | 28 ++++---- .../runtimeApi/stream/hipStreamWaitEvent.cpp | 24 +++---- 8 files changed, 87 insertions(+), 87 deletions(-) diff --git a/include/hip/hcc_detail/hip_runtime.h b/include/hip/hcc_detail/hip_runtime.h index 370ac2abbb..d3211ed3f5 100644 --- a/include/hip/hcc_detail/hip_runtime.h +++ b/include/hip/hcc_detail/hip_runtime.h @@ -53,7 +53,7 @@ THE SOFTWARE. // define HIP_ENABLE_PRINTF to enable printf #ifdef HIP_ENABLE_PRINTF #define HCC_ENABLE_ACCELERATOR_PRINTF 1 -#endif +#endif //--- // Remainder of this file only compiles with HCC @@ -481,7 +481,7 @@ do {\ type* var = \ (type*)__get_dynamicgroupbaseptr(); \ -#define HIP_DYNAMIC_SHARED_ATTRIBUTE +#define HIP_DYNAMIC_SHARED_ATTRIBUTE diff --git a/src/hip_memory.cpp b/src/hip_memory.cpp index 5a4b5f4b4e..8a5225d499 100644 --- a/src/hip_memory.cpp +++ b/src/hip_memory.cpp @@ -69,7 +69,7 @@ int sharePtr(void *ptr, ihipCtx_t *ctx, bool shareWithAll, unsigned hipFlags) if (shareWithAll) { hsa_status_t s = hsa_amd_agents_allow_access(g_deviceCnt+1, g_allAgents, NULL, ptr); - tprintf (DB_MEM, " allow access to CPU + all %d GPUs (shareWithAll)\n", g_deviceCnt); + tprintf (DB_MEM, " allow access to CPU + all %d GPUs (shareWithAll)\n", g_deviceCnt); if (s != HSA_STATUS_SUCCESS) { ret = -1; } @@ -126,7 +126,7 @@ void * allocAndSharePtr(const char *msg, size_t sizeBytes, ihipCtx_t *ctx, bool if (HIP_INIT_ALLOC != -1) { // TODO , dont' call HIP API directly here: hipMemset(ptr, HIP_INIT_ALLOC, sizeBytes); - } + } if (ptr != nullptr) { int r = sharePtr(ptr, ctx, shareWithAll, hipFlags); @@ -255,7 +255,7 @@ hipError_t hipMalloc(void** ptr, size_t sizeBytes) hip_status = hipErrorMemoryAllocation; } - } + } return ihipLogStatus(hip_status); @@ -288,10 +288,10 @@ hipError_t hipHostMalloc(void** ptr, size_t sizeBytes, unsigned int flags) } - const unsigned supportedFlags = hipHostMallocPortable - | hipHostMallocMapped - | hipHostMallocWriteCombined - | hipHostMallocCoherent + const unsigned supportedFlags = hipHostMallocPortable + | hipHostMallocMapped + | hipHostMallocWriteCombined + | hipHostMallocCoherent | hipHostMallocNonCoherent; @@ -304,7 +304,7 @@ hipError_t hipHostMalloc(void** ptr, size_t sizeBytes, unsigned int flags) hip_status = hipErrorInvalidValue; } else { auto device = ctx->getWriteableDevice(); - + unsigned amFlags = 0; if (flags & hipHostMallocCoherent) { amFlags = amHostCoherent; @@ -585,7 +585,7 @@ hipError_t hipMalloc3DArray(hipArray_t *array, hsa_ext_image_data_info_t imageInfo; hsa_status_t status = hsa_ext_image_data_get_info(*agent, &imageDescriptor, permission, &imageInfo); size_t alignment = imageInfo.alignment <= allocGranularity ? 0 : imageInfo.alignment; - + *ptr = hip_internal::allocAndSharePtr("device_array", allocSize, ctx, false, am_flags, 0, alignment); if (size && (*ptr == NULL)) { diff --git a/tests/src/runtimeApi/event/record_event.cpp b/tests/src/runtimeApi/event/record_event.cpp index bd8a3ada8e..b9653bf522 100644 --- a/tests/src/runtimeApi/event/record_event.cpp +++ b/tests/src/runtimeApi/event/record_event.cpp @@ -52,7 +52,7 @@ void test(unsigned testMask, int *C_d, int *C_h, int64_t numElements, hipStream_ if (!(testMask & p_tests)) { return; } - printf ("\ntest 0x%3x: stream=%p waitStart=%d syncMode=%s\n", + printf ("\ntest 0x%3x: stream=%p waitStart=%d syncMode=%s\n", testMask, stream, waitStart, syncModeString(syncMode)); size_t sizeBytes = numElements * sizeof(int); @@ -85,8 +85,8 @@ void test(unsigned testMask, int *C_d, int *C_h, int64_t numElements, hipStream_ HIPCHECK(hipEventSynchronize(start)); } - - hipError_t expectedStopError = hipSuccess; + + hipError_t expectedStopError = hipSuccess; // How to wait for the events to finish: switch (syncMode) { @@ -97,12 +97,12 @@ void test(unsigned testMask, int *C_d, int *C_h, int64_t numElements, hipStream_ HIPCHECK(hipStreamSynchronize(stream)); // wait for recording to finish... break; case syncStopEvent: - HIPCHECK(hipEventSynchronize(stop)); + HIPCHECK(hipEventSynchronize(stop)); break; default: assert(0); }; - + float t; @@ -111,25 +111,25 @@ void test(unsigned testMask, int *C_d, int *C_h, int64_t numElements, hipStream_ failed ("start event not in expected state, was %d=%s\n", e, hipGetErrorName(e)); } - if (e == hipSuccess) + if (e == hipSuccess) assert (t==0.0f); - + // stop usually ready unless we skipped the synchronization (syncNone) HIPCHECK_API(hipEventElapsedTime(&t, stop, stop), expectedStopError); - if (e == hipSuccess) + if (e == hipSuccess) assert (t==0.0f); e = hipEventElapsedTime(&t, start, stop); HIPCHECK_API(e, expectedStopError); - if (expectedStopError == hipSuccess) + if (expectedStopError == hipSuccess) assert (t>0.0f); printf ("time=%6.2f error=%s\n", t, hipGetErrorName(e)); e = hipEventElapsedTime(&t, stop, start); HIPCHECK_API(e, expectedStopError); - if (expectedStopError == hipSuccess) + if (expectedStopError == hipSuccess) assert (t<0.0f); printf ("negtime=%6.2f error=%s\n", t, hipGetErrorName(e)); diff --git a/tests/src/runtimeApi/memory/hipMemcpy.cpp b/tests/src/runtimeApi/memory/hipMemcpy.cpp index e8e803e44c..d8438fa848 100644 --- a/tests/src/runtimeApi/memory/hipMemcpy.cpp +++ b/tests/src/runtimeApi/memory/hipMemcpy.cpp @@ -58,7 +58,7 @@ public: void offset(int offset) { _offset = offset; }; int offset() const { return _offset; }; - + private: T * _A_d; T* _B_d; @@ -72,7 +72,7 @@ private: template DeviceMemory::DeviceMemory(size_t numElements) - : _maxNumElements(numElements), + : _maxNumElements(numElements), _offset(0) { T ** np = nullptr; @@ -93,7 +93,7 @@ DeviceMemory::~DeviceMemory () HipTest::freeArrays (_A_d, _B_d, _C_d, np, np, np, 0); HIPCHECK (hipFree(_C_dd)); - + _C_dd = NULL; }; @@ -125,7 +125,7 @@ public: T * A_hh; T* B_hh; - bool _usePinnedHost; + bool _usePinnedHost; private: size_t _maxNumElements; @@ -165,11 +165,11 @@ HostMemory::HostMemory(size_t numElements, bool usePinnedHost) template void -HostMemory::reset(size_t numElements, bool full) +HostMemory::reset(size_t numElements, bool full) { // Initialize the host data: for (size_t i=0; i void memcpytest2(DeviceMemory *dmem, HostMemory *hmem, size_t numElements, bool useHostToHost, bool useDeviceToDevice, bool useMemkindDefault) { size_t sizeElements = numElements * sizeof(T); - printf ("test: %s<%s> size=%lu (%6.2fMB) usePinnedHost:%d, useHostToHost:%d, useDeviceToDevice:%d, useMemkindDefault:%d, offsets:dev:%+d host:+%d\n", - __func__, + printf ("test: %s<%s> size=%lu (%6.2fMB) usePinnedHost:%d, useHostToHost:%d, useDeviceToDevice:%d, useMemkindDefault:%d, offsets:dev:%+d host:+%d\n", + __func__, TYPENAME(T), sizeElements, sizeElements/1024.0/1024.0, hmem->_usePinnedHost, useHostToHost, useDeviceToDevice, useMemkindDefault, @@ -273,8 +273,8 @@ void memcpytest2_for_type(size_t numElements) { printSep(); - DeviceMemory memD(numElements); - HostMemory memU(numElements, 0/*usePinnedHost*/); + DeviceMemory memD(numElements); + HostMemory memU(numElements, 0/*usePinnedHost*/); HostMemory memP(numElements, 1/*usePinnedHost*/); for (int usePinnedHost =0; usePinnedHost<=1; usePinnedHost++) { @@ -307,11 +307,11 @@ void memcpytest2_sizes(size_t maxElem=0) maxElem = free/sizeof(T)/8; } - printf (" device#%d: hipMemGetInfo: free=%zu (%4.2fMB) total=%zu (%4.2fMB) maxSize=%6.1fMB\n", + printf (" device#%d: hipMemGetInfo: free=%zu (%4.2fMB) total=%zu (%4.2fMB) maxSize=%6.1fMB\n", deviceId, free, (float)(free/1024.0/1024.0), total, (float)(total/1024.0/1024.0), maxElem*sizeof(T)/1024.0/1024.0); HIPCHECK ( hipDeviceReset() ); - DeviceMemory memD(maxElem); - HostMemory memU(maxElem, 0/*usePinnedHost*/); + DeviceMemory memD(maxElem); + HostMemory memU(maxElem, 0/*usePinnedHost*/); HostMemory memP(maxElem, 1/*usePinnedHost*/); for (size_t elem=1; elem<=maxElem; elem*=2) { @@ -336,11 +336,11 @@ void memcpytest2_offsets(size_t maxElem, bool devOffsets, bool hostOffsets) HIPCHECK(hipMemGetInfo(&free, &total)); - printf (" device#%d: hipMemGetInfo: free=%zu (%4.2fMB) total=%zu (%4.2fMB) maxSize=%6.1fMB\n", + printf (" device#%d: hipMemGetInfo: free=%zu (%4.2fMB) total=%zu (%4.2fMB) maxSize=%6.1fMB\n", deviceId, free, (float)(free/1024.0/1024.0), total, (float)(total/1024.0/1024.0), maxElem*sizeof(T)/1024.0/1024.0); HIPCHECK ( hipDeviceReset() ); - DeviceMemory memD(maxElem); - HostMemory memU(maxElem, 0/*usePinnedHost*/); + DeviceMemory memD(maxElem); + HostMemory memU(maxElem, 0/*usePinnedHost*/); HostMemory memP(maxElem, 1/*usePinnedHost*/); size_t elem = maxElem / 2; @@ -380,16 +380,16 @@ void multiThread_1(bool serialize, bool usePinnedHost) { printSep(); printf ("test: %s<%s> serialize=%d usePinnedHost=%d\n", __func__, TYPENAME(T), serialize, usePinnedHost); - DeviceMemory memD(N); - HostMemory mem1(N, usePinnedHost); - HostMemory mem2(N, usePinnedHost); + DeviceMemory memD(N); + HostMemory mem1(N, usePinnedHost); + HostMemory mem2(N, usePinnedHost); std::thread t1 (memcpytest2, &memD, &mem1, N, 0,0,0); if (serialize) { t1.join(); } - + std::thread t2 (memcpytest2,&memD, &mem2, N, 0,0,0); if (serialize) { t2.join(); @@ -427,21 +427,21 @@ int main(int argc, char *argv[]) // Some tests around the 64KB boundary which have historically shown issues: printf ("\n\n=== tests&0x2 (64KB boundary)\n"); size_t maxElem = 32*1024*1024; - DeviceMemory memD(maxElem); - HostMemory memU(maxElem, 0/*usePinnedHost*/); - HostMemory memP(maxElem, 0/*usePinnedHost*/); + DeviceMemory memD(maxElem); + HostMemory memU(maxElem, 0/*usePinnedHost*/); + HostMemory memP(maxElem, 0/*usePinnedHost*/); // These all pass: - memcpytest2(&memD, &memP, 15*1024*1024, 0, 0, 0); - memcpytest2(&memD, &memP, 16*1024*1024, 0, 0, 0); - memcpytest2(&memD, &memP, 16*1024*1024+16*1024, 0, 0, 0); + memcpytest2(&memD, &memP, 15*1024*1024, 0, 0, 0); + memcpytest2(&memD, &memP, 16*1024*1024, 0, 0, 0); + memcpytest2(&memD, &memP, 16*1024*1024+16*1024, 0, 0, 0); // Just over 64MB: - memcpytest2(&memD, &memP, 16*1024*1024+512*1024, 0, 0, 0); - memcpytest2(&memD, &memP, 17*1024*1024+1024, 0, 0, 0); - memcpytest2(&memD, &memP, 32*1024*1024, 0, 0, 0); - memcpytest2(&memD, &memU, 32*1024*1024, 0, 0, 0); - memcpytest2(&memD, &memP, 32*1024*1024, 1, 1, 0); - memcpytest2(&memD, &memP, 32*1024*1024, 1, 1, 0); + memcpytest2(&memD, &memP, 16*1024*1024+512*1024, 0, 0, 0); + memcpytest2(&memD, &memP, 17*1024*1024+1024, 0, 0, 0); + memcpytest2(&memD, &memP, 32*1024*1024, 0, 0, 0); + memcpytest2(&memD, &memU, 32*1024*1024, 0, 0, 0); + memcpytest2(&memD, &memP, 32*1024*1024, 1, 1, 0); + memcpytest2(&memD, &memP, 32*1024*1024, 1, 1, 0); } @@ -464,7 +464,7 @@ int main(int argc, char *argv[]) // Simplest cases: serialize the threads, and also used pinned memory: // This verifies that the sub-calls to memcpytest2 are correct. - multiThread_1(true, true); + multiThread_1(true, true); // Serialize, but use unpinned memory to stress the unpinned memory xfer path. multiThread_1(true, false); diff --git a/tests/src/runtimeApi/multiThread/hipMultiThreadStreams1.cpp b/tests/src/runtimeApi/multiThread/hipMultiThreadStreams1.cpp index 4f73b67ad7..9b2e749cf9 100644 --- a/tests/src/runtimeApi/multiThread/hipMultiThreadStreams1.cpp +++ b/tests/src/runtimeApi/multiThread/hipMultiThreadStreams1.cpp @@ -41,8 +41,8 @@ void printSep() // Designed to stress a small number of simple smoke tests template< - typename T=float, - class P=HipTest::Unpinned, + typename T=float, + class P=HipTest::Unpinned, class C=HipTest::Memcpy > void simpleVectorAdd(size_t numElements, int iters, hipStream_t stream) diff --git a/tests/src/runtimeApi/stream/hipNullStream.cpp b/tests/src/runtimeApi/stream/hipNullStream.cpp index b610315608..af5547e3fb 100644 --- a/tests/src/runtimeApi/stream/hipNullStream.cpp +++ b/tests/src/runtimeApi/stream/hipNullStream.cpp @@ -119,7 +119,7 @@ void Streamer::reset() { HipTest::setDefaultData(_numElements, _A_h, _B_h, _C_h); H2D(); - + } @@ -238,7 +238,7 @@ int main(int argc, char *argv[]) nullStreamer->D2H(); HIPCHECK(hipDeviceSynchronize()); - HipTest::checkTest(expected_H, nullStreamer->_C_h, numElements); + HipTest::checkTest(expected_H, nullStreamer->_C_h, numElements); } } @@ -263,7 +263,7 @@ int main(int argc, char *argv[]) HIPCHECK(hipDeviceSynchronize()); - HipTest::checkTest(expected_H, nullStreamer->_C_h, numElements); + HipTest::checkTest(expected_H, nullStreamer->_C_h, numElements); } } @@ -289,10 +289,10 @@ int main(int argc, char *argv[]) // Copy with stream1, this could go async if the streamSync doesn't synchronize ALL the streams. HIPCHECK(hipMemcpyAsync(streamers[0]->_C_h, streamers[0]->_C_d, streamers[0]->_numElements*sizeof(int), hipMemcpyDeviceToHost, streamers[1]->_stream)); - + HIPCHECK(hipDeviceSynchronize()); - HipTest::checkTest(expected_H, streamers[0]->_C_h, numElements); + HipTest::checkTest(expected_H, streamers[0]->_C_h, numElements); } diff --git a/tests/src/runtimeApi/stream/hipStreamSync2.cpp b/tests/src/runtimeApi/stream/hipStreamSync2.cpp index c6a58ce7d4..4c49d80c05 100644 --- a/tests/src/runtimeApi/stream/hipStreamSync2.cpp +++ b/tests/src/runtimeApi/stream/hipStreamSync2.cpp @@ -59,23 +59,23 @@ const char *syncModeString(int syncMode) { void test(unsigned testMask, int *C_d, int *C_h, int64_t numElements, SyncMode syncMode, bool expectMismatch) { - // This test sends a long-running kernel to the null stream, then tests to see if the + // This test sends a long-running kernel to the null stream, then tests to see if the // specified synchronization technique is effective. // - // Some syncMode are not expected to correctly sync (for example "syncNone"). in these + // Some syncMode are not expected to correctly sync (for example "syncNone"). in these // cases the test sets expectMismatch and the check logic below will attempt to ensure that // the undesired synchronization did not occur - ie ensure the kernel is still running and did // not yet update the stop event. This can be tricky since if the kernel runs fast enough it - // may complete before the check. To prevent this, the addCountReverse has a count parameter - // which causes it to loop repeatedly, and the results are checked in reverse order. + // may complete before the check. To prevent this, the addCountReverse has a count parameter + // which causes it to loop repeatedly, and the results are checked in reverse order. // // Tests with expectMismatch=true should ensure the kernel finishes correctly. This results // are checked and we test to make sure stop event has completed. - + if (!(testMask & p_tests)) { return; } - printf ("\ntest 0x%02x: syncMode=%s expectMismatch=%d\n", + printf ("\ntest 0x%02x: syncMode=%s expectMismatch=%d\n", testMask, syncModeString(syncMode), expectMismatch); size_t sizeBytes = numElements * sizeof(int); @@ -98,7 +98,7 @@ void test(unsigned testMask, int *C_d, int *C_h, int64_t numElements, SyncMode s unsigned blocks = HipTest::setNumBlocks(blocksPerCU, threadsPerBlock, numElements); // Launch kernel into null stream, should result in C_h == count. hipLaunchKernelGGL(HipTest::addCountReverse , dim3(blocks), dim3(threadsPerBlock), 0, 0 /*stream*/, C_d, C_h, numElements, count); - HIPCHECK(hipEventRecord(stop, 0/*default*/)); + HIPCHECK(hipEventRecord(stop, 0/*default*/)); switch (syncMode) { case syncNone: @@ -108,18 +108,18 @@ void test(unsigned testMask, int *C_d, int *C_h, int64_t numElements, SyncMode s break; case syncOtherStream: // Does this synchronize with the null stream? - HIPCHECK(hipStreamSynchronize(otherStream)); + HIPCHECK(hipStreamSynchronize(otherStream)); break; case syncMarkerThenOtherStream: case syncMarkerThenOtherNonBlockingStream: - - // this may wait for NULL stream depending hipStreamNonBlocking flag above - HIPCHECK(hipEventRecord(otherStreamEvent, otherStream)); - HIPCHECK(hipStreamSynchronize(otherStream)); + // this may wait for NULL stream depending hipStreamNonBlocking flag above + HIPCHECK(hipEventRecord(otherStreamEvent, otherStream)); + + HIPCHECK(hipStreamSynchronize(otherStream)); break; case syncDevice: - HIPCHECK(hipDeviceSynchronize()); + HIPCHECK(hipDeviceSynchronize()); break; default: assert(0); @@ -197,7 +197,7 @@ void runTests(int64_t numElements) int main(int argc, char *argv[]) { // Can' destroy the default stream:// TODO - move to another test - HIPCHECK_API(hipStreamDestroy(0), hipErrorInvalidResourceHandle); + HIPCHECK_API(hipStreamDestroy(0), hipErrorInvalidResourceHandle); HipTest::parseStandardArguments(argc, argv, true /*failOnUndefinedArg*/); diff --git a/tests/src/runtimeApi/stream/hipStreamWaitEvent.cpp b/tests/src/runtimeApi/stream/hipStreamWaitEvent.cpp index f5b1b79550..cf463be76a 100644 --- a/tests/src/runtimeApi/stream/hipStreamWaitEvent.cpp +++ b/tests/src/runtimeApi/stream/hipStreamWaitEvent.cpp @@ -88,7 +88,7 @@ private: template Streamer::Streamer(int deviceId, T * A_d, size_t numElements, int commandType) : - _preA_d(NULL), + _preA_d(NULL), _A_d(A_d), _deviceId(deviceId), _numElements(numElements), @@ -239,7 +239,7 @@ size_t Streamer::check(int streamerNum, T initValue, T expectedOffset, bool e return _mismatchCount; } - + //--- //Parse arguments specific to this test. @@ -300,7 +300,7 @@ void checkAll(int initValue, std::vector &streamers, std::vector< for (int i=0; iexpectedAdd(); - + mismatchCount += streamers[i]->check(i+1, initValue, expected, expectPass); } @@ -330,7 +330,7 @@ void checkAll(int initValue, std::vector &streamers, std::vector< void sync_none(void) {}; -void sync_allDevices(int numDevices) +void sync_allDevices(int numDevices) { for (int d=0; d streamers) +void sync_queryAllUntilComplete(std::vector streamers) { for (int i=streamers.size()-1; i>=0; i--) { streamers[i]->queryUntilComplete(); @@ -347,7 +347,7 @@ void sync_queryAllUntilComplete(std::vector streamers) } -void sync_streamWaitEvent(hipEvent_t lastEvent, int sideDeviceId, hipStream_t sideStream, bool waitHere) +void sync_streamWaitEvent(hipEvent_t lastEvent, int sideDeviceId, hipStream_t sideStream, bool waitHere) { HIPCHECK(hipSetDevice(sideDeviceId)); @@ -389,7 +389,7 @@ int main(int argc, char *argv[]) initArray_h[i] = initValue; } HIPCHECK(hipMemcpy(initArray_d, initArray_h, sizeElements, hipMemcpyHostToDevice)); - + int numDevices; HIPCHECK(hipGetDeviceCount(&numDevices)); @@ -414,7 +414,7 @@ int main(int argc, char *argv[]) // A sideband stream channel that is independent from above. - // Used to check to ensure the WaitEvent or other synchronization is working correctly since by default sideStream is + // Used to check to ensure the WaitEvent or other synchronization is working correctly since by default sideStream is // asynchronous wrt the other streams. std::vector sideStreams; for (int d=0; d Test 0x1000 simple null stream tests\n"); + printf ("==> Test 0x1000 simple null stream tests\n"); // try some null stream: hipStreamQuery(0); @@ -463,7 +463,7 @@ int main(int argc, char *argv[]) HIPCHECK(hipEventRecord(e1, s1)) HIPCHECK(hipStreamWaitEvent(hipStream_t(0), e1, 0/*flags*/)); - + HIPCHECK(hipStreamDestroy(s1)); HIPCHECK(hipEventDestroy(e1)); } @@ -476,11 +476,11 @@ int main(int argc, char *argv[]) HIPCHECK(hipEventRecord(e1, hipStream_t(0))) HIPCHECK(hipStreamWaitEvent(s1, e1, 0/*flags*/)); - + HIPCHECK(hipStreamDestroy(s1)); HIPCHECK(hipEventDestroy(e1)); } - + }