EPR #410560 - Forum: [169727] Bug? CL_MEM_COPY_HOST_PTR is not thread safe
- Ignore the transfer queue on memory release, since it releases resources after every operation
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuresource.cpp#201 edit
ECR #304775 - Don't disable second SDMA if configuration has just 1 compute ring.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuvirtual.cpp#343 edit
EPR #410797 - Specific OCL kernel is 5x slower on Hawaii than on Nvidia K40 GPU when tested under Linux.
- The logic for local workgroup size search was prioritizing ALU utilization, but with multidemensional launches X dimension could affect address calculation and cacheline utlization more than others. Add cacheline size into the consideration.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpukernel.cpp#270 edit
EPR #411058 - [CQE OCL][Lnx][QR][CZ]MultiDevice_Context fails in 2.0 conformance wimpyfull due to CL# 1101352
- The detection of different map types is overcomplicated with possibility of multiple maps and multithreading environment. Thus keep USWC indirect map optimization based on the allocation flags.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpumemory.cpp#114 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpumemory.hpp#46 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuvirtual.cpp#342 edit
EPR #410824 - [CQE OCL][CZ][S/G][QR] Two Bolt sample failing on CPU; Faulty CL: 1101352
- The test performs double maps with different map flags. Optimization could choose different map schemes for each call and memory coherency could be broken. Add extra conditions to detect multiple maps and use the same path as the first map.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpumemory.hpp#45 edit
EPR #410736 - [CQE OCL][ISV][QR][G] FFMPEG app generating corrupted video output; Faulty CL:1101352
- Add detection for AHP allocation.
FFmpeg uses AHP allocations with CL_MAP_READ flag, but actually performs CPU write into the buffer. With indirect map runtime executes useless transfer on map and doesn't write updated memory on unmap, because a wrong flag sent by the app.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpumemory.cpp#113 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpumemory.hpp#44 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuvirtual.cpp#341 edit
... //depot/stg/opencl/drivers/opencl/tests/ocltst/module/perf/TestList.cpp#40 edit
EPR #397491 - Replace "switch" with "if" so that new ASIC id doesn't need to be added.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gslbe/src/rt/GSLDevice.cpp#103 edit
EPR #403782 - IOMMU2/SVM
- For finegrainsystem, the app can pass a malloced pointer directly to the kernel. Copy pointer directly to the aqlArgBuf without exiting.
ReviewBoardURL = http://ocltc.amd.com/reviews/r/6378/diff/
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpukernel.cpp#269 edit
EPR #409950 - [IV][OCL] Multiple OCL samples crashed on multiple machines for 32-bit OS.
There are two issues:
1. the SC dll should be dynamically loaded only when it is available. This is to allow apps to run on CPU device without the SC dll. This CL fixes it. It also allows user to use env var AMD_OCL_SC_LIB to provide the name or complete path of SC dll to load.
2. The test fails because amdhsasc.dll is not included in base driver for 32 bit OS. The proper solution should be ask package team to include amdhsasc.dll in the base driver. Also amdhsasc.dll should be renamed amdoclsc.dll since it is not only used for HSAIL but also used by AMDIL. The benefit of separate SC component as a shared library is decreased build time since changes in SC does not require rebuild of amdocl.dll, and ease of debugging and regression analysis by allowing swapping SC comopnent.
However since 15.10 branch is close, there is not enough time to make changes to package. Therefore this CL implements a workaround for this issue without change to the package. We will implement the proper fix in the next relase.
The workaround implemented by this CL embeds SC statically in amdocl.dll. The runtime loads SC dll specified by env var AMD_OCL_SC_LIB only if it is available. If the SC dll is not available, it will use the embeded SC.
Affected files ...
... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/build/Makefile.api#96 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/api/v0_8/acl.cpp#22 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/api/v0_8/aclLoaders.cpp#9 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/gpu/Makefile#44 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/gpu/sclibdefs.opencl#20 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/include/v0_8/aclStructs.h#13 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/include/v0_8/aclTypes.h#4 edit
... //depot/stg/opencl/drivers/opencl/compiler/tools/aoc2/build/Makefile.aoc2#21 edit
... //depot/stg/opencl/drivers/opencl/opencldefs#148 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudevice.cpp#485 edit
... //depot/stg/opencl/drivers/opencl/runtime/utils/flags.hpp#220 edit
ECR #333753 - Compiler Lib/aoc2/devloader: move devloader functionality into aoc2
[Purpose] To get rid of obsolete runtimenew dependency in compiler
1. Devloader functionality moved into aoc2;
2. Devloader is removed from the tree & make system;
3. Related changes in test_driver.pl;
4. Functions alignedMalloc & alignedFree are moved to libUtils.h;
5. Function aclHsaLoader is renamed to _aclHsaLoader to indicate that it is not a Compiler Lib API's function.
[Testing] make smoke, pre check-in
[Reviewers] Nikolay Haustov, Brian Sumner
Affected files ...
... //depot/stg/opencl/drivers/opencl/Makefile#48 edit
... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/amdocl.def.in#10 edit
... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/amdocl.map.in#11 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/amdoclcl.def.in#8 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/amdoclcl.map.in#7 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/gpu/brig_loader.cpp#15 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/gpu/scwrapper/scClientAPI.cpp#20 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/utils/v0_8/libUtils.h#13 edit
... //depot/stg/opencl/drivers/opencl/compiler/loader/devloader/Makefile#8 delete
... //depot/stg/opencl/drivers/opencl/compiler/loader/devloader/build/Makefile#3 delete
... //depot/stg/opencl/drivers/opencl/compiler/loader/devloader/build/Makefile.devloader#11 delete
... //depot/stg/opencl/drivers/opencl/compiler/loader/devloader/devloader.cpp#6 delete
... //depot/stg/opencl/drivers/opencl/compiler/tools/aoc2/aoc2.cpp#61 edit
... //depot/stg/opencl/drivers/opencl/compiler/tools/aoc2/build/Makefile.aoc2#20 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuprogram.cpp#185 edit
... //depot/stg/opencl/drivers/opencl/tests/hsa/bin/test_driver.pl#5 edit
EPR #409798 - clCompileProgram and clLinkProgram regression for SPIR - set the correct IR type while extracting from binary (aclSPIR, aclLLVMIR) for single SPIR module for CPU.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/cpu/cpuprogram.cpp#62 edit
EPR #409840 - [CQE OCL][LNX][QR] OpenCL SPIR Conf test "Compile_and_link" failed in all Asics due to CL#1098110 - Set the IR type to SPIR only for single SPIR modules.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuprogram.cpp#184 edit
EPR #409798 - clCompileProgram and clLinkProgram regression for SPIR - set the correct IR type while extracting from binary (aclSPIR, aclLLVMIR)
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuprogram.cpp#183 edit
ECR #304775 - Optimize oclBandwidthTest from nVidia SDK
- Cache pinned memory, since the benchmark sends the same transfer in a single batch. Thus we could avoid pin/unpin
- Swap SDMA engine allocation order. Blit manager allocates a queue on device, thus the first app queue was getting the paging second SDMA.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpublit.cpp#112 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpublit.hpp#37 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuvirtual.cpp#339 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuvirtual.hpp#121 edit
EPR #407215 - reset host memory pointer of a image view based on original image
ReviewBoard #6301
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudevice.cpp#480 edit
ECR #333755 - Part 2- Update to foundation spec 1.0 20141019:
- hsa_dispatch_packet_t now becomes hsa_kernel_dispatch_packet_t
- all bit mask in a struct are removed and replaced by enums that indicates the bit position and width.
Test: TC precheckin
Review: Hari, Fan, Shucai, German, Yunjun.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpukernel.cpp#268 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpukernel.hpp#103 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpusched.hpp#15 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuvirtual.cpp#338 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/hsa_foundation/hsavirtual.cpp#25 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/hsa_foundation/hsavirtual.hpp#12 edit
EPR #408459 - changed the implementation of svmAlloc, so that the first device can create amd::Memory object, and the rest of devices only added gpu memory to it. This is part of changes for mgpu support for svmalloc
code review:
http://ocltc.amd.com/reviews/r/6245/
precheckin testing results:
http://ocltc.amd.com:8111/viewModification.html?modId=43136&personal=true&buildTypeId=&tab=vcsModificationTests
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/cpu/cpudevice.hpp#88 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/device.hpp#233 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudevice.cpp#479 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudevice.hpp#133 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/hsa/hsadevice.cpp#87 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/hsa/hsadevice.hpp#44 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/hsa_foundation/hsadevice.cpp#19 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/hsa_foundation/hsadevice.hpp#6 edit
... //depot/stg/opencl/drivers/opencl/runtime/platform/context.cpp#34 edit
EPR #408185 - Use pinned memory if directaccess is true and remoteAlloc is used.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudevice.cpp#478 edit
EPR #408459 - added an env variable OCL_FORCE_CPU_SVM in the runtime, so that the svm feature for CPU can be enabled manually even for non OpenCL 2.0 support for CPU device.
code review:
http://ocltc.amd.com/reviews/r/6190/
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/cpu/cpudevice.cpp#268 edit
... //depot/stg/opencl/drivers/opencl/runtime/utils/flags.hpp#218 edit
EPR #406328 - removed the customSVMallocator from runtime, and also changed the name of customSvmAllocDevice to svmAllocDevice, because we don't use custom svm allocator for devices.
precheckin testing:
http://ocltc.amd.com:8111/viewModification.html?modId=43040&personal=true&buildTypeId=&tab=vcsModificationBuilds&show_all_builds=true
code review:
http://ocltc.amd.com/reviews/r/6222/
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/cpu/cpusettings.cpp#29 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/device.cpp#171 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/device.hpp#232 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpusettings.cpp#293 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/hsa/hsasettings.cpp#36 edit
... //depot/stg/opencl/drivers/opencl/runtime/platform/context.cpp#33 edit
... //depot/stg/opencl/drivers/opencl/runtime/platform/context.hpp#23 edit
EPR #408506 - Extended the reported global memory size(CL_DEVICE_GLOBAL_FREE_MEMORY_AMD) to include a portion of remote memory for APU
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudevice.cpp#476 edit
ECR #304775 - add flag to force CL_FP_DENORM on gpu
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudevice.cpp#475 edit
... //depot/stg/opencl/drivers/opencl/runtime/utils/flags.hpp#216 edit
ECR #304775 - Align the queue size to match the multidispatch scheduler requirements
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuvirtual.cpp#337 edit
EPR #397491 - changed the CPU SVM capability availablility only for OpenCL 2.0, not for 1.2.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/cpu/cpudevice.cpp#267 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/cpu/cpusettings.cpp#28 edit
foreman