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

Update hipHostRegister debug and pointerTracker debug and notes

Browse Source
This commit is contained in:
Ben Sander
2017-02-17 15:43:09 -06:00
parent 94c85fd4fc
commit 0a554f4dc1
3 changed files with 87 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/hip_hcc.cpp
+28 -12
View File
@@ -1803,6 +1803,20 @@ void ihipStream_t::resolveHcMemcpyDirection(unsigned hipMemKind,
}
void printPointerInfo(unsigned dbFlag, const char *tag, const void *ptr, const hc::AmPointerInfo &ptrInfo)
{
tprintf (dbFlag, " %s=%p baseHost=%p baseDev=%p sz=%zu home_dev=%d tracked=%d isDevMem=%d registered=%d\n",
tag, ptr,
ptrInfo._hostPointer, ptrInfo._devicePointer, ptrInfo._sizeBytes,
ptrInfo._appId, ptrInfo._sizeBytes != 0, ptrInfo._isInDeviceMem, !ptrInfo._isAmManaged);
}
// TODO : For registered and host memory, if the portable flag is set, we need to recognize that and perform appropriate copy operation.
// What can happen now is that Portable memory is mapped into multiple devices but Peer access is not enabled. i
// The peer detection logic doesn't see that the memory is already mapped and so tries to use an unpinned copy algorithm. If this is PinInPlace, then an error can occur.
// Need to track Portable flag correctly or use new RT functionality to query the peer status for the pointer.
//
// TODO - remove kind parm from here or use it below?
void ihipStream_t::locked_copySync(void* dst, const void* src, size_t sizeBytes, unsigned kind, bool resolveOn)
{
@@ -1819,6 +1833,16 @@ void ihipStream_t::locked_copySync(void* dst, const void* src, size_t sizeBytes,
bool dstTracked = (hc::am_memtracker_getinfo(&dstPtrInfo, dst) == AM_SUCCESS);
bool srcTracked = (hc::am_memtracker_getinfo(&srcPtrInfo, src) == AM_SUCCESS);
// Some code in HCC and in printPointerInfo uses _sizeBytes==0 as an indication ptr is not valid, so check it here:
if (!dstTracked) {
assert (dstPtrInfo._sizeBytes == 0);
}
if (!srcTracked) {
assert (srcPtrInfo._sizeBytes == 0);
}
hc::hcCommandKind hcCopyDir;
ihipCtx_t *copyDevice;
bool forceUnpinnedCopy;
@@ -1831,12 +1855,8 @@ void ihipStream_t::locked_copySync(void* dst, const void* src, size_t sizeBytes,
dst, dstPtrInfo._appId, dstPtrInfo._isInDeviceMem,
src, srcPtrInfo._appId, srcPtrInfo._isInDeviceMem,
sizeBytes, hcMemcpyStr(hcCopyDir), forceUnpinnedCopy);
tprintf (DB_COPY, " dst=%p baseHost=%p baseDev=%p sz=%zu home_dev=%d tracked=%d isDevMem=%d\n",
dst, dstPtrInfo._hostPointer, dstPtrInfo._devicePointer, dstPtrInfo._sizeBytes,
dstPtrInfo._appId, dstTracked, dstPtrInfo._isInDeviceMem);
tprintf (DB_COPY, " src=%p baseHost=%p baseDev=%p sz=%zu home_dev=%d tracked=%d isDevMem=%d\n",
src, srcPtrInfo._hostPointer, srcPtrInfo._devicePointer, srcPtrInfo._sizeBytes,
srcPtrInfo._appId, srcTracked, srcPtrInfo._isInDeviceMem);
printPointerInfo(DB_COPY, " dst", dst, dstPtrInfo);
printPointerInfo(DB_COPY, " src", src, srcPtrInfo);
this->ensureHaveQueue(crit);
@@ -1921,12 +1941,8 @@ void ihipStream_t::locked_copyAsync(void* dst, const void* src, size_t sizeBytes
dst, dstPtrInfo._appId, dstPtrInfo._isInDeviceMem,
src, srcPtrInfo._appId, srcPtrInfo._isInDeviceMem,
sizeBytes, hcMemcpyStr(hcCopyDir), forceUnpinnedCopy);
tprintf (DB_COPY, " dst=%p baseHost=%p baseDev=%p sz=%zu home_dev=%d tracked=%d isDevMem=%d\n",
dst, dstPtrInfo._hostPointer, dstPtrInfo._devicePointer, dstPtrInfo._sizeBytes,
dstPtrInfo._appId, dstTracked, dstPtrInfo._isInDeviceMem);
tprintf (DB_COPY, " src=%p baseHost=%p baseDev=%p sz=%zu home_dev=%d tracked=%d isDevMem=%d\n",
src, srcPtrInfo._hostPointer, srcPtrInfo._devicePointer, srcPtrInfo._sizeBytes,
srcPtrInfo._appId, srcTracked, srcPtrInfo._isInDeviceMem);
printPointerInfo(DB_COPY, " dst", dst, dstPtrInfo);
printPointerInfo(DB_COPY, " src", src, srcPtrInfo);
// "tracked" really indicates if the pointer's virtual address is available in the GPU address space.
// If both pointers are not tracked, we need to fall back to a sync copy.