Wykres commitów

952 Commity

Autor SHA1 Wiadomość Data
Aryan Salmanpour 7e48231252 [hip] Move _criticalData of ihipStream_t class to private section and use criticalData() to access it (#1177) 2019-07-04 00:42:19 +00:00
Maneesh Gupta a220a8e8e9 Added missing NULL checks and corrected API return values (#1188)
* Added missing NULL checks and corrected API return values as per validation

* Added missing NULL checks
2019-07-03 08:51:39 +00:00
Anusha Godavarthy Surya 1a7c7e3b06 Added missing NULL checks 2019-06-27 20:19:30 +05:30
Anusha Godavarthy Surya 4989452413 Added missing NULL checks and corrected API return values as per validation 2019-06-27 00:19:05 +05:30
wkwchau 3742f24477 Fixed bug in hipOccupancyMaxPotentialBlockSize for the SGPRs limitation of gfx8 devices (#1176) 2019-06-26 15:18:00 +05:30
Aaron Enye Shi f87b900f96 Fix dlpi_name info empty when using GCC on ub18 (#1181)
This fixes a bug where GCC++ on Ubuntu 18.04 creates failing executables compared to GCC++ on 16.04 and clang++. While creating function names on Ubuntu 18.04, dl_phdr_info seems to provide a non-zero value for dlpi_addr on initial iteration, and an empty string in dlpi_name. This is causing failure when linking with g++, since the empty string prevents the kernel function from being loaded. Clang++ and GCC on UB16 provide a zero value for dlpi_addr. To fix this, we need to verify both addr and name exists, so that /proc/self/exe can be properly loaded.
2019-06-25 06:32:29 +05:30
Aryan Salmanpour d6ad690cb6 [hip] implement the hipExtLaunchMultiKernelMultiDevice API (#1165)
* [hip] implement the hipExtLaunchMultiKernelMultiDevice API

* add a guard to check the HCC version for acquire_locked_hsa_queue() API which was introdued in HCC for ROCm 2.5

* modified code based on the requested changes

* changes to lock all streams before launching kernels for each device and unlock them after the dispatches

* check each stream to be valid before starting to lock all the streams
2019-06-20 05:59:05 +05:30
wkwchau 28c34ead70 Implement the hipOccupancyMaxPotentialBlockSize function (#1162)
* Implement the hipOccupancyMaxPotentialBlockSize function

* Replaced hipGetDeviceProperties() call by ihipGetDeviceProperties() in ihipOccupancyMaxPotentialBlockSize()

* Add test for hipOccupancyMaxPotentialBlockSize in Module API

* Added extern declaration for ihipGetDeviceProperties() to be accessed inside ihipOccupancyMaxPotentialBlockSize()

* fixed hipOccupancyMaxPotentialBlockSize test build issue

* Fix hipOccupancyMaxPotentialBlockSize dtest

* Add BUILD_CMD in hipOccupancyMaxPotentialBlockSize dtest

* Revert "Add BUILD_CMD in hipOccupancyMaxPotentialBlockSize dtest"

This reverts commit 0480ff56f1441fc515d2c26ce33783e303423938.

* Disable hipOccupancyMaxPotentialBlockSize dtest on NVCC

* move extern declaration of ihipGetDeviceProperties to hip_module.cpp

* Update the limiation of 32 wavefronts per CU and 800/512 SGPRs for VI/pre-VI chips to calculate the occupancy
2019-06-20 05:58:29 +05:30
Maneesh Gupta 3b3118d459 Merge pull request #1167 from eshcherb/hip_prof_refactoring_190611
prof layer includes refactoring
2019-06-19 13:36:33 +05:30
Rahul Garg 107734f7ad HACK for SWDEV-173477/SWDEV-190701 2019-06-13 18:15:31 -07:00
Evgeny c6600ba26b prof layer includes refactoring 2019-06-11 20:13:29 -05:00
Maneesh Gupta 1d5d923d36 Merge pull request #1140 from scchan/program_state_stage_2-rebase-20190524
migrate more program_state logic from header into shared library (phase II)
2019-06-05 16:09:01 +05:30
Maneesh Gupta d4fa74ff09 Merge branch 'master' into implicit-kernarg 2019-06-04 13:24:19 +05:30
Maneesh Gupta 40a09318e4 Merge pull request #1155 from gargrahul/fix_kernel_lp_dim_trace
Fix wrong grid dim shown in trace
2019-06-04 13:21:39 +05:30
Maneesh Gupta 4b3d59a93e Merge pull request #1130 from lmoriche/master
Add support for code object v3
2019-06-04 13:20:52 +05:30
cdevadas 214ec53da3 Runtime changes to append implicit kernel arguments.
Appended 48 empty bytes to the kernarg area at runtime. The implicit arguments are enabled primarily for the hostcall services
and it is completely abstracted from the user code. Enabled it for both hip-clang and hip-hcc.
2019-06-04 10:45:49 +05:30
Rahul Garg 7a2e3b6a1c Fix wrong grid dim shown in trace 2019-05-31 22:30:24 +05:30
Siu Chi Chan 1fb9ab2d44 replace std::vector for kernarg 2019-05-24 17:27:43 -04:00
Siu Chi Chan 1a2d332e76 move executable_cache into program_state.cpp 2019-05-24 17:27:25 -04:00
Siu Chi Chan e2c0122892 remove executables() from program_state 2019-05-24 17:27:01 -04:00
Siu Chi Chan 6852be819f moving agent_globals_impl into hip_module 2019-05-24 16:43:38 -04:00
Laurent Morichetti de89102528 Add support for code object v3
Use the code object manager library to parse the code object metadata. Both
code object v2 and v3 formats are now supported for HCC generated binaries.
2019-05-23 18:03:32 -07:00
Evgeny Mankov 9cb3e9aa5e [HIP][HIPIFY] Make hipMemcpyParam2D coherent with cuMemcpy2D
+ Makes hip_Memcpy2D struct compatible with CUDA_MEMCPY2D struct
+ Add hipMemcpyParam2D support in nvcc fallback path
+ Update hipify-clang, tests and docs accordingly
2019-05-22 18:31:39 +03:00
Alex Voicu a538eb705a Add HIPRTC, glorious ersatz for NVRTC (#1097)
* Add ersatz for NVRTC.

* Fix extraneous paren and use correct namespace.

* Use lowerCamelCase (yuck, yuck) consistently.

* Link against FS when building hiprtc lib.

* Correctly mark Manipulators. Fix dual compile.

* Add unit tests. Extend HIT to accept linker options.

* Make sure the HIPRTC library is installed.

* Better logging. Try to auto-detect the target.

* Stop specifying the target explicitly.

* Add missing flavour of `hipModuleLaunchKernel`.

* Program was already destroyed.

* Don't use `--genco`. Fix mangled name trimming.

* Fix HIPRTC breakage due to upstream noise.

* [dtests] Replace RUN -> TEST in hiprtc tests

Change-Id: Ie499e92dfe4e5c94634b1c2b76cf52d241bcfea3

* [hit] Set HIP_PATH to HIP_ROOT_DIR for all tests

Change-Id: Ib0ad1f99bc71c03e363e055dd508a7a4a210680a
2019-05-16 18:28:54 +05:30
Wenkai Du 56d2dc0022 Use NUMA distance for hop count calculation 2019-05-15 21:50:35 +00:00
Maneesh Gupta c9fdb42b91 Merge pull request #1083 from gargrahul/fix_hip_impl_visible_agents
Maintain HIP_VISIBLE_DEVICES for kernel launch
2019-05-13 14:20:18 +05:30
Rahul Garg e1f3dc0c80 Add fine grained host memory lock support (#1095)
* Add fine grained host memory lock support

* Fix default flag check
2019-05-13 11:48:26 +05:30
Siu Chi Chan 05a1b696da migrate program_state logic from header into shared library (phase I) (#1077)
* Revert "Revert "Use COMgr to read Kernel Args Metadata (#1006)""

This reverts commit 235c6877c8.

* Revert "Use COMgr to read Kernel Args Metadata (#1006)"

This reverts commit 2c80975e9c.

* Revert "improve program state commentary"

This reverts commit f5e4fff6cc.

* Revert "load program state once per agent"

This reverts commit 2845b4c4b8.

* start moving function_names() into the hip shared lib

* start moving code_object_blobs to a new "state" object

* Consolidate various program state related static objects into a
single program_state object

* minor clean up

* move more stuffs from functional_grid_launch into program_state

* debug make_kernarg

* moving lookup for kernargs size_align into program_state

* clean up old code for kernarg size and alignment

* update hip_module to use newer api in program_state

* Create public member functions for program_state

* move most program state functions into shared library

* Pass the data buffer size to load_executable
Otherwise, it can't figure what the data size is
just from the char* (since the data is not really a string)

* turning free functions in program state into members of program_state_impl

* change the free function globals() into a member of program_state_impl

* replace the static mutex used for populating globals

* moving associate_code_object_symbols_with_host_allocation into
program_state_impl

* move load_code_object_and_freeze_executable into program_state_impl

* moving executables and functions_names into program_state_impl

* moving kernels() into program_state_impl

* moving functions() into program_state_impl

* move get_kernargs into program_state_impl

* moving kernel_descriptor into program_state_impl

* moving kernargs_size_align calculation into program_state_impl

* Changing the handle to program_state_impl to a pointer

* moving program_state_impl into a separate inline source file

* fixing/cleaning up some header file includes

* moving member function for kernargs_size_align into program_state.cpp

* moving Kernel_descriptor into program_state.inl

* add a new class to manage agent globals

* moving all agent globals processing functions into agent_globals_impl

* load program state once per agent

re-merging PR991 against other program state changes

* fix per-agent program state member initialization

* cache executables based on elf name, isa, and agent.

This avoids program state reloading executables after a shared library is dlopened.

re-merging PR1057 against other program state changes

* protect executables cache by a global mutex

* return ref to executables cache

* adapt PR#981 Make hipModuleGetGlobal be in HIP runtime
2019-05-12 19:24:03 +05:30
Maneesh Gupta c6c5e4cee8 Merge pull request #1081 from mangupta/swdev-181624
Implement hipExtGetLinkTypeAndHopCount for ROCm devices
2019-05-07 16:15:41 +05:30
wkwchau ebf986dcee Return hipErrorInsufficientDriver status when CPU device not found (#1064)
* Return hipErrorInsufficientDriver status when CPU device not found - no exception thrown

* Return hipErrorInsufficientDriver status when CPU device not found
2019-05-07 15:58:25 +05:30
Rahul Garg 3be54a903c Maintain HIP_VISIBLE_DEVICES for kernel launch 2019-05-07 05:09:02 +05:30
Maneesh Gupta 2f43f110d9 Implement hipExtGetLinkTypeAndHopCount for ROCm devices
Change-Id: Ie5bb4f640ac6d189c7fceeab22627a7494fd10bd
2019-05-06 15:54:31 +05:30
Sameer Sahasrabuddhe c74a97f756 minor cleanup: eliminate repetition 2019-04-25 20:41:16 +05:30
Rahul Garg 94769fc8dd Add hipMallocManaged default functional support (#1036)
* Add hipMallocManaged default functional support

* Fix build error

* Add dtest
2019-04-24 16:50:03 +05:30
Yaxun (Sam) Liu 710e633bdd Fix missing arg in HIP_INIT_API 2019-04-18 16:18:31 -04:00
Maneesh Gupta 22660bed74 Merge pull request #1019 from scchan/lazy_binding
minor workaround for lazy binding
2019-04-16 08:36:10 +05:30
Jeff Daily cf8fb43e6b In hipFree, synchronize owner of memory (#1018)
* In hipFree, if memory is associated with a device, synchronize that device's streams.

This changes the behavior from synchronizing the currently set TLS device.

* All devices sync in hipFree for _appId=-1 case.

* Revert "All devices sync in hipFree for _appId=-1 case."

This reverts commit 1efb34d6a8426661e45bc5f763422a1147aeac10.

* add HIP_SYNC_FREE env var
2019-04-16 08:35:55 +05:30
Yaxun (Sam) Liu 5c67ee11f4 Fix regression on multi-gpu due to PR#997 2019-04-05 22:54:41 -04:00
Siu Chi Chan b5045af7e9 minor workaround for lazy binding 2019-04-02 17:28:06 -04:00
Yaxun Sam Liu 8f5c812a68 hip-clang: fix kernel not found on multi-gpu
__hipRegisterFunction is called during by .init functions during program initialization.
It calls hipModuleGetFunction to locate kernel symbol in code objects. hipModuleGetFunction
assumes current device when locating kernel symbols. This works for HCC but not for hip-clang,
since hip-clang needs to locate kernel symbols for different devices without switching
between devices.

This patch introduces a new hsa agent parameter to ihipModuleGetFunction, which allows
__hipRegisterFunction to choose the correct hsa agent when locating kernel symbols. By
default it uses this_agent(), therefore this patch has no impact on HCC.
2019-03-31 10:08:20 -04:00
Wen-Heng (Jack) Chung 04915cea2f Make hipModuleGetGlobal be in HIP runtime so it can be discovered at runtime (#981)
* Make hipModuleGetGlobal be in HIP runtime so it can be discovered at runtime

In HIP PR #929, quite a few HIP public APIs were made as inline functions with
hidden visibility. It was necessary to support applications with shared
libraries with GPU kernels launched via hipLaunchKernelGGL(), after HIP runtime
is initialized.

In empirical tests, the implementation has been proved to be a bit too
excessive, especially for hipModuleGetGlobal(). The function is used by another
type of client applications which relies on the existence of this function
within HIP runtime so global symbols from HSA code objects loaded dynamically
at runtime can be retrieved programmtically.

This commit moves hipModuleGetGlobal() back to src/hip_module.cpp, and makes it
visible and not inline, to fulfill requirements for applications
aforementioned. It does not change the behavior of applications depending on
hipLaunchKernelGGL().

* Add HIP_INIT_API into the implementation of hipModuleGetGlobal

Address review comments.

* Fix failing HIP unit tests
2019-03-29 03:45:04 +00:00
Maneesh Gupta f9f4cee347 Merge pull request #992 from gargrahul/handle_d2d_memcpy2d
Handle D2D in memcpy2D
2019-03-28 04:41:36 +00:00
Rahul Garg 50d623981e Handle D2D in memcpy2D 2019-03-28 02:21:45 +05:30
Jeff Daily 2845b4c4b8 load program state once per agent 2019-03-27 18:19:10 +00:00
Rahul Garg 9b38380c03 Let hipHostMalloc always share/map pinned host ptr 2019-03-26 10:19:13 +05:30
Rahul Garg ad11972f47 Avoid double mapping of devices to hostMalloc buffer 2019-03-25 23:07:05 +05:30
Maneesh Gupta c20d233585 Merge pull request #970 from mangupta/swdev-172995
hipExtMallocWithFlags implementation
2019-03-25 07:46:53 +00:00
Maneesh Gupta 9de28dfa5a Merge pull request #962 from gargrahul/add_2d_copy_fallback
Add 2D fallback to use copy kernel
2019-03-25 07:46:43 +00:00
Rahul Garg bec3995700 2D Fallback needs hcc workweek 19101 or higher 2019-03-25 12:07:28 +05:30
Maneesh Gupta 45255ab492 hipExtMallocWithFlags needs hcc workweek 19115 or higher 2019-03-25 11:41:20 +05:30