Created attachment 862766 [details] updated ebuild The attached ebuild enables dev-libs/pocl to build successfully on a riscv platform.
Created attachment 862767 [details] Build log
I'd suggest providing a patch for the change, also the logical is not correct AFAIC, please don't keyword it before the patch is accepted As comment in bug #907331, it would great to have a short summary about why request this keywording (reasons behind), and how it works well (have run-time test passed)? I would guess it's used by opencl related apps
Created attachment 862772 [details, diff] add riscv64 support
Created attachment 862847 [details] output of clinfo I want to enable the CPU component of OpenCL on my RISCV SBC. I have just started testing and I see the following error when i run clinfo which I am not sure what to do to fix. Kernel library file /usr/lib64/OpenCL/vendors/pocl/../../../../share/pocl/kernel-riscv64-unknown-linux-gnu-sifive-u74.bc doesn't exist. Aborted (core dumped) tarFive ~ # ls /usr/share/pocl/kernel-riscv64-unknown-linux-gnu-generic-rv64.bc /usr/share/pocl/kernel-riscv64-unknown-linux-gnu-generic-rv64.bc StarFive ~ # file /usr/share/pocl/kernel-riscv64-unknown-linux-gnu-generic-rv64.bc /usr/share/pocl/kernel-riscv64-unknown-linux-gnu-generic-rv64.bc: LLVM IR bitcode
It looks like for riscv you need to figure out the appropriate -mcpu to use Once I changed it to if use riscv ; then host_cpu_variants="sifive-u74" then it worked better but there are still one or two issues. StarFive ~ # poclcc -l LIST OF DEVICES: 0: Vendor: Imagination Technologies Name: PowerVR B-Series BXE-4-32 Version: OpenCL 3.0 /hashcat -I hashcat (v6.2.6-460-gca4dc7acf) starting in backend information mode * Device #1: This device's local mem size is too small. OpenCL Info: ============ OpenCL Platform ID #1 Vendor..: Imagination Technologies Name....: PowerVR Version.: OpenCL 3.0 Backend Device ID #1 Type...........: GPU Vendor.ID......: 2147483648 Vendor.........: Imagination Technologies Name...........: PowerVR B-Series BXE-4-32 Version........: OpenCL 3.0 Processor(s)...: 1 Clock..........: 594 Memory.Total...: 7927 MB (limited to 990 MB allocatable in one block) Memory.Free....: 1152 MB Local.Memory...: 4 KB OpenCL.Version.: OpenCL C 1.2 Driver.Version.: 1.17@6210866 OpenCL Platform ID #2 Vendor..: Mesa Name....: Clover Version.: OpenCL 1.1 Mesa 23.0.0-devel (git-a89acaf05d) OpenCL Platform ID #3 Vendor..: The pocl project Name....: Portable Computing Language Version.: OpenCL 3.0 PoCL 3.1 Linux, RelWithDebInfo+Asserts, RELOC, SPIR, SPIR-V, LLVM 15.0.7, SLEEF Backend Device ID #2 Type...........: CPU Vendor.ID......: 2147483648 Vendor.........: Unknown x86 Name...........: pthread Version........: OpenCL 3.0 PoCL HSTR: pthread-riscv64-unknown-linux-gnu-sifive-u74 Processor(s)...: 4 Clock..........: 1500 Memory.Total...: 5879 MB (limited to 1024 MB allocatable in one block) Memory.Free....: 2907 MB Local.Memory...: 512 KB OpenCL.Version.: OpenCL C 1.2 PoCL Driver.Version.: 3.1 -- FAILED TEST --- StarFive ~ # python3 test.py Execution time of test without OpenCL: 30.688774824142456 s =============================================================== Platform name: PowerVR Platform profile: EMBEDDED_PROFILE Platform vendor: Imagination Technologies Platform version: OpenCL 3.0 --------------------------------------------------------------- Device name: PowerVR B-Series BXE-4-32 Device type: ALL | GPU Device memory: 7927 MB Device max clock speed: 594 MHz Device compute units: 1 Execution time of test: 0.014451 s Results OK =============================================================== Platform name: Portable Computing Language Platform profile: FULL_PROFILE Platform vendor: The pocl project Platform version: OpenCL 3.0 PoCL 3.1 Linux, RelWithDebInfo+Asserts, RELOC, SPIR, SPIR-V, LLVM 15.0.7, SLEEF --------------------------------------------------------------- Device name: pthread Device type: ALL | CPU Device memory: 5879 MB Device max clock speed: 1500 MHz Device compute units: 4 Segmentation fault (core dumped)
(In reply to Andrew Cameron from comment #4) > Created attachment 862847 [details] > output of clinfo > > I want to enable the CPU component of OpenCL on my RISCV SBC. > > I have just started testing and I see the following error when i run clinfo > which I am not sure what to do to fix. > > Kernel library file > /usr/lib64/OpenCL/vendors/pocl/../../../../share/pocl/kernel-riscv64-unknown- > linux-gnu-sifive-u74.bc doesn't exist. > Aborted (core dumped) I reproduced this coredump with host_cpu_variants="generic" (In reply to Andrew Cameron from comment #5) > It looks like for riscv you need to figure out the appropriate -mcpu to use > Once I changed it to > if use riscv ; then host_cpu_variants="sifive-u74" > I'm not quite sure what guidelines to use to add host_cpu_variants, I tried changing to host_cpu_variants="generic-rv64;rocket-rv64;sifive-s21;sifive-s51;sifive-s54;sifive-s76;sifive-u54;sifive-u74;sifive-x280" (Source: https://github.com/llvm/llvm-project/blob/071e9d7bac7a5c879b1c67e1c4e847814f6d8254/clang/test/Misc/target-invalid-cpu-note.c#L88) and it compile fails on sifive-u74 (hifive unmatched board), obviously we should pick some more generic variants. For now, I think host_cpu_variants="generic-rv64;sifive-u74" is ok on my unmatched board. (In reply to Andrew Cameron from comment #5) > pthread-riscv64-unknown-linux-gnu-sifive-u74 > Processor(s)...: 4 > Clock..........: 1500 > Memory.Total...: 5879 MB (limited to 1024 MB allocatable in one block) > Memory.Free....: 2907 MB > Local.Memory...: 512 KB > OpenCL.Version.: OpenCL C 1.2 PoCL > Driver.Version.: 3.1 > > -- FAILED TEST --- > StarFive ~ # python3 test.py > Execution time of test without OpenCL: 30.688774824142456 s > =============================================================== > Platform name: PowerVR > Platform profile: EMBEDDED_PROFILE > Platform vendor: Imagination Technologies > Platform version: OpenCL 3.0 > --------------------------------------------------------------- > Device name: pthread > Device type: ALL | CPU > Device memory: 5879 MB > Device max clock speed: 1500 MHz > Device compute units: 4 > Segmentation fault (core dumped) How to reproduce this error?
I can confirm that host_cpu_variants="generic-rv64;sifive-u74" also seems to work on my device. To test and see if you can reproduce the error I got you need to emerge pyopencl and all its dependencies. Then you can execute my test program that I will add as an attachment. python test.py Please post the output you get so we can compare results.
Created attachment 862915 [details] Test program I am using to test POCL
Any Updates on this?
