AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
1,396 Commits 3 Branches 0 Tags
31dfd8bb1fdfd66a56119feeb79ba75b077349ff
Commit Graph

470 Commits

Author SHA1 Message Date
Aditya Atluri e39de0f560 reorg make_datatype feature 
[ROCm/hip commit: 9ca8584ec0]
 2016-04-21 14:51:19 -05:00
Aditya Atluri 552acd5ed5 added full data type support for __ldg 
[ROCm/hip commit: ba59ea87ab]
 2016-04-21 11:17:26 -05:00
bwicakso d24ba1ecbb Fix for kernel synchronization 
The completion future of a particular kernel is lost if there are
multiple kernels in the stream. This can cause a racing condition where
the signal associated with the unreferenced completion_future might get
released by hcc runtime.


[ROCm/hip commit: df98fd8531]
 2016-04-20 15:51:39 -05:00
Aditya Atluri 182ce03846 added support pinned dma memcpy between host and device 
[ROCm/hip commit: f74b7a3636]
 2016-04-20 14:21:22 -05:00
Aditya Atluri 7feb990d97 added copyright for ldg 
[ROCm/hip commit: de7952cd06]
 2016-04-20 12:28:02 -05:00
Aditya Atluri 15fc041fd6 added support for __ldg 
[ROCm/hip commit: 805b268ad4]
 2016-04-20 12:25:40 -05:00
Ben Sander 5858b85e8d Fix hipDeviceReset synchronization 
[ROCm/hip commit: 453615ed57]
 2016-04-19 11:56:12 -05:00
Ben Sander edda485753 Set chicken bits to 0. 
[ROCm/hip commit: 8c97a258de]
 2016-04-19 11:56:12 -05:00
Ben Sander bd8388bfc3 Add USE_HCC_LOCK_API. 
Default off, if set will compile assuming HCC lock APIs exist.


[ROCm/hip commit: 3bcefdaa2d]
 2016-04-18 23:50:52 -05:00
Ben Sander f601813057 Merge branch 'privatestaging' of https://github.com/AMDComputeLibraries/HIP-privatestaging into privatestaging 
[ROCm/hip commit: 22d5738f82]
 2016-04-18 21:51:13 -05:00
Ben Sander 32f9d67741 Fixes for P2P and hipDeviceReset 
- devicereset would lose track of default stream and thus subsequent
  synchronization calls might not actually sychronize.
- Also deviceReset now correctly frees streams.
- fix waits in P2P staging copy - first phase (Device0-to-Staging) must
  wait for second phase (Staging to Device1) to finish draining the
  buffer.


[ROCm/hip commit: e020d68309]
 2016-04-18 20:49:33 -05:00
Aditya Atluri 348e5be858 Update hip_hcc.cpp 
[ROCm/hip commit: ea647727df]
 2016-04-18 11:36:51 -05:00
Ben Sander 698ea7f30f Move HIP_HCC define to CMake 
[ROCm/hip commit: 65abde6626]
 2016-04-17 07:40:04 -05:00
Ben Sander 95e4be12de fix direction of peer enable/disable APIs 
- Enable(peerId) makes memory physically located on peer visible to
current device.


[ROCm/hip commit: 74a7dee58f]
 2016-04-17 07:38:05 -05:00
Ben Sander 262b69f387 Merge branch 'privatestaging' into p2p 
Conflicts:
	include/hcc_detail/hip_hcc.h
	src/hip_hcc.cpp


[ROCm/hip commit: 49cc5aec91]
 2016-04-17 06:46:52 -05:00
Aditya Atluri 0b2dfa28a5 Added copyright 
[ROCm/hip commit: 8412e9cbba]
 2016-04-16 17:21:53 -05:00
Aditya Atluri 6cd86531b3 Corrected Memcpydefault 
[ROCm/hip commit: 8dc1bdcbe6]
 2016-04-16 17:10:13 -05:00
Ben Sander d4f449a95e P2P Update. 
- add P2P staging buffer copy.
- If copy device does not have sufficient access permissions, fall back
  to staging buffer.
- improve docs for which copy device is used.


[ROCm/hip commit: dcabc9dbf7]
 2016-04-16 10:18:56 -05:00
Aditya Atluri c53211e700 improved error returns for hipHostRegister/UnRegister functions 
[ROCm/hip commit: 2e80de6dcb]
 2016-04-15 10:42:31 -05:00
Aditya Atluri 4dd3424b15 added more features to hipHostRegister 
[ROCm/hip commit: 8a43f18333]
 2016-04-15 10:32:01 -05:00
Aditya Atluri 12cb3de301 added hipHostRegister/UnRegister api 
[ROCm/hip commit: fa7a3822f6]
 2016-04-15 10:08:10 -05:00
Ben Sander 5060403b2a Merge branch 'p2p' of https://github.com/AMDComputeLibraries/HIP-privatestaging into p2p 
Conflicts:
	RELEASE.md
	include/hcc_detail/hip_hcc.h
	samples/1_Utils/hipInfo/hipInfo.cpp
	src/hip_hcc.cpp
	src/hip_peer.cpp


[ROCm/hip commit: 31dc13d2ec]
 2016-04-11 09:17:27 -05:00
Ben Sander 7c92c9cc22 P2p checkpoint. 
- set USE_PEER_TO_PEER=3 (requires HCC "am_memtracker_update_peers")
- when enabling peer, turn it on for previously allocated memory.
- hipDeviceCanAccessPeer is no longer self-ware (self does not qualify
  as a peer)
- device peerlist always includes self, so when we call allow_access
  we never remove self access.
- hipDeviceReset() removes old peer mappings.


[ROCm/hip commit: 1f53c55d3e]
 2016-04-11 12:52:18 -05:00
Ben Sander b174e8bba7 Clean up disable. 
Add USE_HCC_LOCK (disabled)
Disable USE_PEER_TO_PEER.


[ROCm/hip commit: b0529e04f1]
 2016-04-11 09:09:36 -05:00
Ben Sander c964e3c75a P2p checkpoint. 
- set USE_PEER_TO_PEER=3 (requires HCC "am_memtracker_update_peers")
- when enabling peer, turn it on for previously allocated memory.
- hipDeviceCanAccessPeer is no longer self-ware (self does not qualify
  as a peer)
- device peerlist always includes self, so when we call allow_access
  we never remove self access.
- hipDeviceReset() removes old peer mappings.


[ROCm/hip commit: 83f0de7806]
 2016-04-11 07:58:59 -05:00
Ben Sander 80e2e37c51 Remove stray debug msgs, hipInfo don't print self as peer. 
[ROCm/hip commit: d89539d40f]
 2016-04-11 07:58:58 -05:00
Ben Sander 69a9feb1fb fix bugs in P2P implementation 
- addPeers polarity reversed, would never add.
- check allow_access return value, pipe error to hipMalloc.


[ROCm/hip commit: fb31eaf07b]
 2016-04-11 07:58:58 -05:00
Ben Sander 216c9a0c16 Use HCC get_is_pool, add USE_PEER_TO_PEER level 2, 3. 
[ROCm/hip commit: 7a11a2178d]
 2016-04-11 07:58:58 -05:00
Ben Sander 06e2dbd128 For P2P, use the peer list when allocating Device memory or pinned host. 
Each new allocation is automatically mapped into the address space of
all enabled peers.


[ROCm/hip commit: 813b063888]
 2016-04-11 07:58:58 -05:00
Ben Sander ac11c83d76 P2P checkpoint. 
Maintain enabled peer tables for each device.


[ROCm/hip commit: f2aa470f7f]
 2016-04-11 07:58:58 -05:00
Ben Sander 56d684e2fe Checkpoint initial peer2peer implementation. 
[ROCm/hip commit: 69f2469cbb]
 2016-04-11 07:58:58 -05:00
Aditya Atluri ed3013cb6e added hipHostRegister support no multi-gpu 
[ROCm/hip commit: 07d570468e]
 2016-04-11 10:28:16 -05:00
Ben Sander 772983e547 Remove stray debug msgs, hipInfo don't print self as peer. 
[ROCm/hip commit: 816de09842]
 2016-04-09 04:35:06 -05:00
Ben Sander 70108729ba fix bugs in P2P implementation 
- addPeers polarity reversed, would never add.
- check allow_access return value, pipe error to hipMalloc.


[ROCm/hip commit: 7886c9e3d9]
 2016-04-09 04:11:31 -05:00
Ben Sander bc45411c71 Merge branch 'privatestaging' of https://github.com/AMDComputeLibraries/HIP-privatestaging into privatestaging 
[ROCm/hip commit: 11c22b24b7]
 2016-04-07 15:55:35 -05:00
Ben Sander aeac589552 use hsa_free directly for staging buffer free rather than am_free 
[ROCm/hip commit: 56c801f9e4]
 2016-04-07 15:54:28 -05:00
Ben Sander c0d88c2f6d Use HCC get_is_pool, add USE_PEER_TO_PEER level 2, 3. 
[ROCm/hip commit: 15b4b1f81f]
 2016-04-07 14:28:42 -05:00
pensun bdef1bb4e8 clean up unused comments 
[ROCm/hip commit: 4b2c5976ce]
 2016-04-07 09:46:00 -05:00
Ben Sander 84b5016a79 For P2P, use the peer list when allocating Device memory or pinned host. 
Each new allocation is automatically mapped into the address space of
all enabled peers.


[ROCm/hip commit: 41f7317fb5]
 2016-04-06 16:44:31 -05:00
Ben Sander c5240bd079 P2P checkpoint. 
Maintain enabled peer tables for each device.


[ROCm/hip commit: 36926e6233]
 2016-04-06 15:50:47 -05:00
Ben Sander 8db4a6d46e Checkpoint initial peer2peer implementation. 
[ROCm/hip commit: b02e9163ab]
 2016-04-06 15:50:47 -05:00
Aditya Atluri 70c1ac9285 added device math functions 
[ROCm/hip commit: 26985b746b]
 2016-03-31 12:29:19 -05:00
Aditya Atluri fe0f4e8bc2 moved cpp specific device code from headers to source 
[ROCm/hip commit: 72c72d8748]
 2016-03-31 04:45:00 -05:00
Aditya Atluri f052a2752a Merge branch 'privatestaging' of https://github.com/AMDComputeLibraries/HIP-privatestaging into privatestaging 
[ROCm/hip commit: 39af5d9caf]
 2016-03-29 11:18:09 -05:00
Ben Sander 24f82925bb Add runtime switch to control HIP_ATP_MARKER 
Only generate the function strings if requested at
compile-time && runtime.


[ROCm/hip commit: 740b730cac]
 2016-03-29 17:27:30 -05:00
Aditya Atluri ffe185d731 Calling intrinsics from global kernel 
[ROCm/hip commit: 61cd2bb399]
 2016-03-29 11:17:55 -05:00
Ben Sander 630ef59d7b Tweak thread-safe implementation. 
introduce LockedAccessor option so destructor does not unlock.
Allows locks to exist across function boundaries, required
for hipLaunchKernel macro which has several unusual requirements.
(including C comppatibility, must use variadic macro, more).


[ROCm/hip commit: 8635863724]
 2016-03-28 21:41:47 -05:00
Ben Sander 9e3ac64c54 Stream thread-safe checkpoint. 
Moving data structures to critical / protected section.


[ROCm/hip commit: 3f18bab2c7]
 2016-03-28 09:46:40 -05:00
Ben Sander c9b5b99078 Stream thread-safe checkpoint. 
[ROCm/hip commit: ec049e7f0b]
 2016-03-28 04:22:20 -05:00
Ben Sander 48773c6d4f Improve debug print messages. 
- Remove "call-to-call" for hipStreamCreate and hipEventCreate.
  These now call an internal functions rather than calling through
  hipStreamCreateWithFalgs and hipEventCreateWithFlags.

- Add HIP_INIT_API for more functions so they trace correctly.

- Use stream#DEVICE.STREAMID in debug messages via new specialization in
  tace_helper.


[ROCm/hip commit: 7934cf620d]
 2016-03-26 12:35:04 -05:00