Commit Graph

253 Commits

Author SHA1 Message Date
Aditya Atluri 6dcdf08e0d fixed multi-dim module kernel launch
Change-Id: Id1d81f2375d058979ab526433f905cf0ea3d23d6
2016-11-11 12:25:23 -06:00
Ben Sander d666fbaafe Add option to deny peer access.
Also fix test.

Change-Id: I1b247f6c4271442b008e560669bca4daf8eb94c7
2016-11-10 23:12:48 -06:00
Ben Sander 6e54a600b6 Use forceUnpinnedCopy to resolve P2p corner cases.
Change-Id: I2aebb419881246cebb696bec87798635bc71acc2
2016-11-10 23:12:48 -06:00
Ben Sander 0eeaa3bcd5 Enable async copy again.
Also add HIP_FORCE_SYNC_COPY chicken bit.

Change-Id: I76a385410494b99bf27305d3c08f55dd81987565
2016-11-10 23:12:48 -06:00
Ben Sander e767e0032e Refactor copy and P2P logic.
Prefer use of source-engine for DMA copies, even if user submits copy
in a stream attached to a different device.
The stream is now used only for synchronization, and HIP
makes the most optimal decision for which engine to perform the
copy - typically the source copy engine.

HIP now makes decision on which engine should perform the copy
and passes this to HCC using new apis.
HIP has additional information about peer
visibility and will make a decision which agent should perform
the copy .

Change-Id: I0cf4cfebeae256e6ca795f08a7ed7130f4857d1f
2016-11-10 23:12:48 -06:00
Ben Sander f3d38c2615 Improve Peer support and testing.
Change-Id: Icadc65988aaf145a265587ab0357c5bf4d26f3eb
2016-11-06 03:22:36 -06:00
Ben Sander 06ecfa3975 Set forceHostCopyEngine for other copy dirs. Support HIP_FORCE_P2P_HOST
Also: more debug for copy and P2p.

Change-Id: I87030c525410e041b2a00baaf6c68e6c0977ff42
2016-11-04 19:53:23 -05:00
Ben Sander 926e63c655 Refactor resolve-mem step1
Change-Id: I7b8b2bbb56d7b31a97b48ebd42002641cd07a460
2016-11-04 09:37:56 -05:00
Ben Sander 00276d141e Add debug for Peer APIs. Enable PeerMemcpy APIs by default.
Change-Id: I46e39a9e7b07686a78484c1f3b5495b08e052fbb
2016-11-04 08:51:16 -05:00
Aditya Atluri f097b6ef81 added inter thread data movement intrinsics
Change-Id: I2a8a8ed49429cb7f96439bd28c4b83b5142737df
2016-11-01 16:37:33 -05:00
Ben Sander 3d0fa30183 Print short hipLaunchKernel correctly.
Change-Id: I6ca03d7c707cd03d6982199830213953d5855f17
2016-10-27 23:09:32 -05:00
Ben Sander 18dbafe6e8 Add initial hipProfileStart/Stop
And modify sample to show how to use.
Still needs some work to understand interaction with CXL.

Change-Id: I2579824d2dd7863ea23874d34f0dabb3cb305d3e
2016-10-27 23:09:32 -05:00
Ben Sander 1d5034f583 Add two levels of HIP_PROFILE_API (1=short,2=long)
Change-Id: I7ef98589f8731fb879db109fd573c62b489f2b61
2016-10-27 23:09:31 -05:00
Ben Sander 90adf98a7e Fix scoped marker so begin/end ATP timestamps correct
Change-Id: Ic944d3fc00d7bc31b756c0e6c327b99eb489537e
2016-10-27 23:09:31 -05:00
Ben Sander f7c42efa3f Rename HIP_ATP_MARKER and profiling vars
HIP_PROFILE_API
HIP_DB_START_API
HIP_DB_STOP_API

Change-Id: I6c4da67212ff8217e6356a2622d4c6278a188c34
2016-10-27 23:09:31 -05:00
Ben Sander caf446fe53 Allow HIP_DB to be number or string flags (ie HIP_DB=api+mem+sync)
Add callbacks for processing env vars.

Change-Id: I4ddf50e2da56b1dae43f50657bc693b07b23c03d
2016-10-27 23:09:31 -05:00
Ben Sander 3132c36471 Add HIP_PROFILE_START_API, HIP_PROFILE_STOP_API
Refactor HIP_INIT_API to call recordApiTrace.

Change-Id: Ieff4b5018236f59e49e1b9841474440a34f821df
2016-10-27 23:09:31 -05:00
Ben Sander 45e2dd4d69 Add per-thread API seqnum to debug
Change-Id: Ib13733a3e84cd56bae13a32bae40f936c20b7543
2016-10-27 23:09:31 -05:00
Ben Sander 7d69e858a5 Improve HIP TID printing in debug mode.
Map long thread-id to a short one that is printed with each message.
Remove clunky stirng creation code for tid_tr.
Print TID on every message.

Change-Id: I780a91d8ce789cb4957789036b478bf5cde8c4e4
2016-10-27 23:09:31 -05:00
Aditya Atluri 8145e1b129 reverted change for cache size query
Change-Id: I44a1f43818cd287a2a3b6265f43d183f9bd5b71c
2016-10-25 11:03:35 -05:00
Aditya Atluri e28f65604b correct cachesize to output correct value
Change-Id: I5db031591eb718b0c12e78a35e4b19349de9526d
2016-10-25 09:33:45 -05:00
Ben Sander 714968cdc6 Fix P2P for async
Also improve HIP debug message: Add more DB_COPY1 messages. memcpyStr,
expand HIP_DB bitmask.
2016-10-20 10:02:23 -05:00
Ben Sander d4b23da91f Add HIP_WAIT_MODE env var.
Also weaken cases where hipSetDeviceFlags returns hipErrorInvalidValue.

Change-Id: I7f113338be6fe498eaf1ab40fd0fd6b23849bb5e
2016-10-18 22:27:16 -05:00
Ben Sander c7303d5924 Add hipDeviceSchedule* support to queue wait
Change-Id: Iffa7a356500b026f3737c3f5719ca9f62b10d855
2016-10-18 22:27:16 -05:00
Ben Sander 0787c41ac0 Remove some TODO items
Change-Id: I7e9de2e43a8584f8dc9ee6d45c8ed00ca465f591
2016-10-18 22:27:16 -05:00
Ben Sander 403cdf93c3 Move some internal headers from "include/hip/" to src.
Change-Id: I7041bd5c803d9318979f4a7c1d658445c614691e
2016-10-18 22:27:16 -05:00
Maneesh Gupta 88d6cad3fb src/*: Update copyright header
Change-Id: I455f5d0d12fe9cb39a3ba873bd22b4c25ed07cbf
2016-10-15 22:55:22 +05:30
Ben Sander aa3e04cf9a Cleanup files from code review.
- Remove some stale code
- Update docs
- Correct define for __HIP_ARCH_HAS_GLOBAL_INT64_ATOMICS__

Change-Id: Ic5e3cdb8269b1c18f6d2693700b55e08c4d0080e
2016-10-15 11:51:20 -05:00
Ben Sander dcadaadb24 Add code to use new HCC API accelerator_view::dispatch_hsa_kernel.
Disabed by default, can enable with USE_DISPATCH_HSA_KERNEL=1

Change-Id: I7a6ba76f2bada34952ed47f5335ce695fa2faea5
2016-10-14 23:46:29 -05:00
Ben Sander 45bc107ad0 Fix HIP_USE_PRODUCT_NAME detection.
Change-Id: I6879ec3a11845bea66a18a9328bd4eaf54713420
2016-10-13 11:51:53 -05:00
pensun 0ac70e7184 Add ifdef guard for the feature requires ROCm1.3
Change-Id: I7154517c47000c37fe5eb09a3c1cf2a9aacbe27c
2016-10-13 10:57:31 -05:00
Aditya Atluri e25a1e50b3 added constant memory property to 16KB
Change-Id: If067b4057c2e3fc0c26cf4604a1d4fac7f139b12
2016-10-13 10:47:40 -05:00
pensun 44cc44011b Change to query device name using HSA_AMD_AGENT_INFO_PRODUCT_NAME;
Note: this commit depends on ROCR runtime in ROCm 1.3 release.

Change-Id: I90385ef6d11ee8a1e8adae1d3fdf21747347544c
2016-10-12 20:01:30 -05:00
Aditya Atluri e5325a1ab4 Added hipDeviceGetLimit api
1. hipDeviceGetLimit API for HCC path is added
2. Test for hipDeviceGetLimit API is added
3. The feature added only supports querying heap size
4. Corrected indents for malloc and free device functions
5. Removed redundant data structures
6. Added g_heap_malloc_size to store the heap size

Change-Id: If48d1b0ce9270e994f1c542cc283ddbb14746bbb
2016-10-12 19:58:48 -05:00
Aditya Atluri 2918ff273d added malloc and free device functions
1. Added malloc and free device functions
2. Added test which check malloc and free functions
TODO: Need to add support for multiple device. Works only on one device (multi device support id NOT available).

Change-Id: Id11fc36463915d6ad46c264d5a20c8feb2d2c17c
2016-10-12 19:08:34 -05:00
Ben Sander 250f711ed0 Add DISABLE_COPY_EXT option. 2016-10-05 12:18:42 -05:00
Ben Sander 6950d3f9e9 Add HIP_BLOCKING_SYNC environment var to control stream sync behavior. 2016-10-05 12:18:42 -05:00
Maneesh Gupta bbfc08f419 Move include/* to include/hip/*
Change-Id: I7a7b2839b4df59c7a4c503550f99fdc9e45c0f54
2016-10-04 22:17:18 +05:30
Ben Sander 806e58ca63 Refactor asyncCopy and syncCopy to fix deadlock case.
- Minimize time that locks are held.
- Eliminate copy code that locked stream and ctx at same time.
    - Stream was locked to ensure thread-safe enqueue to the queue.
    - Devices were locked to query peer-lists.

Change-Id: Ibe8880bb7fb995a3da8f90ff911f212d81525018
2016-09-27 15:45:40 -05:00
Ben Sander 4fd4168796 Add debug option to print ThreadID with each message.
Also print messages with single fprintf to prevents threads from
interleaving.

Change-Id: Ib3999fe6b1e67b4a16cd7dcde82f3dfc99dd48ff
2016-09-27 15:45:40 -05:00
Ben Sander f6717e326e Fix signal resource issue.
Remove memory leak with new hc::completion_future.
Implement HIP_LAUNCH_BLOCKING with queue-level wait.

Change-Id: I45975f81c4d239fdeed7776970988d28449865dc
2016-09-26 16:47:32 -05:00
Ben Sander 944d4eaa18 Peer-to-Peer improvements.
- Bug fix for peer visibility.  Now contexts correctly detect when they can use SDMA for P2P vs staging buffers.
- Interface to new HCC copy_ext function.
- Improve context and peer print /debug options.
- Add comments and usage to hipPeerToPeer_simple test.
2016-09-22 14:21:19 -05:00
Ben Sander 926aefa930 Remove unpinned_copy code. Other cleanup.
Change-Id: Ie3f71439cf1ba729ef223d078917c403d3de879a
2016-09-22 14:21:19 -05:00
Ben Sander 5bc18c8603 Cleanup. Remove cfs, ihipSignal_t, staging buffer calls.
Change-Id: I8bb67c484e3a65be06a03665f059217930da2bed
2016-09-22 14:21:19 -05:00
Ben Sander 17e21309f1 Cleanup: Remove HIP signal pool.
Change-Id: Icebfd0509d12396cc5933d5556d68b53e1be36e0
2016-09-22 14:21:19 -05:00
Ben Sander c365785c5c Remove HIP command dependency tracking.
Change-Id: I991c13bc5108193959ba70f9f6f9c692c9ad3a5b
2016-09-22 14:21:19 -05:00
Ben Sander b0509c5a58 Cleanup, remove preCopyCommand.
Change-Id: I3768d3789a99be8136b43179d4152fa1875665cb
2016-09-22 14:21:19 -05:00
Ben Sander 3596901c3d Change HIP async copy to call av::copy_async.
Change-Id: I4274b63ced3940d5249c32bd9d156296529c70e8
2016-09-22 14:21:19 -05:00
Ben Sander 8598d8c6d3 Use HCC's synchronous accelerator_view::copy
Replace large block of HIP code with a call to HCC av::copy().

Change-Id: Ic32e1801cf8d4cd116ac02b72c41b1a1e4b6065c
2016-09-22 14:21:19 -05:00
Ben Sander 60f797cc5a Remove USE_AV_COPY, USE_PEER_TO_PEER fallback paths.
Change-Id: I9c20173e62029c4caebabc98784c6d7697758e4f
2016-09-22 14:21:19 -05:00