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

Changed StagingBuffer class to UnpinnedCopyEngine

Browse Source 
Change-Id: I1e212bfc8030dcf225ecf78fd7b23fda9b1de92f
This commit is contained in:
Rahul Garg
2016-08-09 21:29:42 +05:30
parent 023b1ecf33
commit 2ac93c340d
4 changed files with 51 additions and 51 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/hip_hcc.cpp
+6 -6
View File
@@ -85,7 +85,7 @@ int HIP_DISABLE_HW_COPY_DEP = 0;
std::once_flag hip_initialized;
// Array of pointers to devices.
// Array of pointers to devices.
ihipDevice_t **g_deviceArray;
@@ -106,7 +106,7 @@ thread_local int tls_defaultDeviceId = 0;
// This is the implicit context used by all HIP commands.
// It can be set by hipSetDevice or by the CTX manipulation commands:
thread_local ihipCtx_t *tls_defaultCtx;
thread_local ihipCtx_t *tls_defaultCtx;
thread_local hipError_t tls_lastHipError = hipSuccess;
@@ -550,7 +550,7 @@ void ihipCtxCriticalBase_t<CtxMutex>::addStream(ihipStream_t *stream)
// ihipDevice_t
//=================================================================================================
ihipDevice_t::ihipDevice_t(unsigned deviceId, unsigned deviceCnt, hc::accelerator &acc) :
_deviceId(deviceId),
_deviceId(deviceId),
_acc(acc)
{
hsa_agent_t *agent = static_cast<hsa_agent_t*> (acc.get_hsa_agent());
@@ -567,8 +567,8 @@ ihipDevice_t::ihipDevice_t(unsigned deviceId, unsigned deviceCnt, hc::accelerato
initProperties(&_props);
_stagingBuffer[0] = new StagingBuffer(_hsaAgent,g_cpu_agent, HIP_STAGING_SIZE*1024, HIP_STAGING_BUFFERS,HIP_H2D_MEM_TRANSFER_THRESHOLD_DIRECT_OR_STAGING,HIP_H2D_MEM_TRANSFER_THRESHOLD_STAGING_OR_PININPLACE,HIP_D2H_MEM_TRANSFER_THRESHOLD);
_stagingBuffer[1] = new StagingBuffer(_hsaAgent,g_cpu_agent, HIP_STAGING_SIZE*1024, HIP_STAGING_BUFFERS,HIP_H2D_MEM_TRANSFER_THRESHOLD_DIRECT_OR_STAGING,HIP_H2D_MEM_TRANSFER_THRESHOLD_STAGING_OR_PININPLACE,HIP_D2H_MEM_TRANSFER_THRESHOLD);
_stagingBuffer[0] = new UnpinnedCopyEngine(_hsaAgent,g_cpu_agent, HIP_STAGING_SIZE*1024, HIP_STAGING_BUFFERS,HIP_H2D_MEM_TRANSFER_THRESHOLD_DIRECT_OR_STAGING,HIP_H2D_MEM_TRANSFER_THRESHOLD_STAGING_OR_PININPLACE,HIP_D2H_MEM_TRANSFER_THRESHOLD);
_stagingBuffer[1] = new UnpinnedCopyEngine(_hsaAgent,g_cpu_agent, HIP_STAGING_SIZE*1024, HIP_STAGING_BUFFERS,HIP_H2D_MEM_TRANSFER_THRESHOLD_DIRECT_OR_STAGING,HIP_H2D_MEM_TRANSFER_THRESHOLD_STAGING_OR_PININPLACE,HIP_D2H_MEM_TRANSFER_THRESHOLD);
_primaryCtx = new ihipCtx_t(this, deviceCnt, hipDeviceMapHost);
}
@@ -1832,5 +1832,5 @@ hipError_t hipHccGetAcceleratorView(hipStream_t stream, hc::accelerator_view **a
//// TODO - add identifier numbers for streams and devices to help with debugging.
#if ONE_OBJECT_FILE
#include "staging_buffer.cpp"
#include "unpinned_copy_engine.cpp"
#endif
src/staging_buffer.cpp → src/unpinned_copy_engine.cpp
+8 -8
View File
@@ -21,7 +21,7 @@ THE SOFTWARE.
#include "hsa_ext_amd.h"
#include "hcc_detail/staging_buffer.h"
#include "hcc_detail/unpinned_copy_engine.h"
#ifdef HIP_HCC
#include "hcc_detail/hip_runtime.h"
@@ -62,7 +62,7 @@ hsa_status_t findGlobalPool(hsa_amd_memory_pool_t pool, void* data) {
}
//-------------------------------------------------------------------------------------------------
StagingBuffer::StagingBuffer(hsa_agent_t hsaAgent, hsa_agent_t cpuAgent, size_t bufferSize, int numBuffers, int thresholdH2DDirectStaging,int thresholdH2DStagingPinInPlace,int thresholdD2H) :
UnpinnedCopyEngine::UnpinnedCopyEngine(hsa_agent_t hsaAgent, hsa_agent_t cpuAgent, size_t bufferSize, int numBuffers, int thresholdH2DDirectStaging,int thresholdH2DStagingPinInPlace,int thresholdD2H) :
_hsaAgent(hsaAgent),
_cpuAgent(cpuAgent),
_bufferSize(bufferSize),
@@ -93,7 +93,7 @@ StagingBuffer::StagingBuffer(hsa_agent_t hsaAgent, hsa_agent_t cpuAgent, size_t
//---
StagingBuffer::~StagingBuffer()
UnpinnedCopyEngine::~UnpinnedCopyEngine()
{
for (int i=0; i<_numBuffers; i++) {
if (_pinnedStagingBuffer[i]) {
@@ -112,7 +112,7 @@ StagingBuffer::~StagingBuffer()
//IN: dst - dest pointer - must be accessible from host CPU.
//IN: src - src pointer for copy. Must be accessible from agent this buffer is associated with (via _hsaAgent)
//IN: waitFor - hsaSignal to wait for - the copy will begin only when the specified dependency is resolved. May be NULL indicating no dependency.
void StagingBuffer::CopyHostToDevicePinInPlace(void* dst, const void* src, size_t sizeBytes, hsa_signal_t *waitFor)
void UnpinnedCopyEngine::CopyHostToDevicePinInPlace(void* dst, const void* src, size_t sizeBytes, hsa_signal_t *waitFor)
{
std::lock_guard<std::mutex> l (_copy_lock);
@@ -166,7 +166,7 @@ void StagingBuffer::CopyHostToDevicePinInPlace(void* dst, const void* src, size_
//IN: dst - dest pointer - must be accessible from host CPU.
//IN: src - src pointer for copy. Must be accessible from agent this buffer is associated with (via _hsaAgent)
//IN: waitFor - hsaSignal to wait for - the copy will begin only when the specified dependency is resolved. May be NULL indicating no dependency.
void StagingBuffer::CopyHostToDevice(int tempIndex,int isLargeBar,void* dst, const void* src, size_t sizeBytes, hsa_signal_t *waitFor)
void UnpinnedCopyEngine::CopyHostToDevice(int tempIndex,int isLargeBar,void* dst, const void* src, size_t sizeBytes, hsa_signal_t *waitFor)
{
if((tempIndex==1)&&(isLargeBar)&&(sizeBytes < _hipH2DTransferThresholdDirectOrStaging)){
memcpy(dst,src,sizeBytes);
@@ -227,7 +227,7 @@ void StagingBuffer::CopyHostToDevice(int tempIndex,int isLargeBar,void* dst, con
}
void StagingBuffer::CopyDeviceToHostPinInPlace(void* dst, const void* src, size_t sizeBytes, hsa_signal_t *waitFor)
void UnpinnedCopyEngine::CopyDeviceToHostPinInPlace(void* dst, const void* src, size_t sizeBytes, hsa_signal_t *waitFor)
{
std::lock_guard<std::mutex> l (_copy_lock);
@@ -272,7 +272,7 @@ void StagingBuffer::CopyDeviceToHostPinInPlace(void* dst, const void* src, size_
//IN: dst - dest pointer - must be accessible from agent this buffer is associated with (via _hsaAgent).
//IN: src - src pointer for copy. Must be accessible from host CPU.
//IN: waitFor - hsaSignal to wait for - the copy will begin only when the specified dependency is resolved. May be NULL indicating no dependency.
void StagingBuffer::CopyDeviceToHost(int tempIndex,void* dst, const void* src, size_t sizeBytes, hsa_signal_t *waitFor)
void UnpinnedCopyEngine::CopyDeviceToHost(int tempIndex,void* dst, const void* src, size_t sizeBytes, hsa_signal_t *waitFor)
{
if((tempIndex==1) && (sizeBytes> _hipD2HTransferThreshold)){
CopyDeviceToHostPinInPlace(dst, src, sizeBytes, waitFor);
@@ -339,7 +339,7 @@ void StagingBuffer::CopyDeviceToHost(int tempIndex,void* dst, const void* src, s
//IN: dst - dest pointer - must be accessible from agent this buffer is associated with (via _hsaAgent).
//IN: src - src pointer for copy. Must be accessible from host CPU.
//IN: waitFor - hsaSignal to wait for - the copy will begin only when the specified dependency is resolved. May be NULL indicating no dependency.
void StagingBuffer::CopyPeerToPeer(void* dst, hsa_agent_t dstAgent, const void* src, hsa_agent_t srcAgent, size_t sizeBytes, hsa_signal_t *waitFor)
void UnpinnedCopyEngine::CopyPeerToPeer(void* dst, hsa_agent_t dstAgent, const void* src, hsa_agent_t srcAgent, size_t sizeBytes, hsa_signal_t *waitFor)
{
std::lock_guard<std::mutex> l (_copy_lock);