Add RISC-V port

[tingsi5]

The port is based on pattonkan/sse2rvv@ca0e0b4, which for now supports SSE2 and SSE4.2 that embree uses. It is verified using Ubuntu 24.04.1 RISC-V preinstalled server image from https://cdimage.ubuntu.com/releases/24.04/release/ with QEMU on Ubuntu 24.04.

Note this is just the initial port of RVV and has a limitation for VLEN=128 only. We expect further performance improvement from porting the other extensions.

Test steps:

1. Follow https://wiki.ubuntu.com/RISC-V/QEMU to install required packages and run the QEMU using the following command with vector extension enabled:

   # qemu-system-riscv64 \
   -machine virt -nographic -m 4096 -smp 4 \
   -bios /usr/lib/riscv64-linux-gnu/opensbi/generic/fw_jump.bin \
   -kernel /usr/lib/u-boot/qemu-riscv64_smode/uboot.elf \
   -device virtio-net-device,netdev=eth0 -netdev user,id=eth0 \
   -device virtio-rng-pci \
   -drive file=ubuntu-24.04.1-preinstalled-server-riscv64.img,format=raw,if=virtio \
   -cpu rv64,v=true,vlen=128,elen=64,vext_spec=v1.0,zba=true,zbb=true,zbs=true
   

2. Use clang-18 to build embree in QEMU guest
   sudo apt update; sudo apt install cmake libtbb-dev libglfw3-dev clang-18
cd embree; mkdir build; cd build; cmake -DCMAKE_CXX_COMPILER=clang++-18 -DCMAKE_C_COMPILER=clang-18 ..; make -j4; sudo make install

3. Untar the archive https://github.com/RenderKit/embree/releases/download/v4.3.3/embree-4.3.3-testing.tar.gz to /usr/local and run tests
   cd /usr/local/testing; sudo cmake -B build -DCMAKE_CXX_COMPILER=clang++-18 -DCMAKE_C_COMPILER=clang-18 -DEMBREE_TESTING_INTENSITY=1; sudo cmake --build build --target test
   The results are:

   99% tests passed, 2 tests failed out of 269
   
   Total Test time (real) = 13564.67 sec
   
   The following tests FAILED:
           246 - embree_verify (Timeout)
           258 - bvh_builder (Timeout)
   

   Running embree_bvh_builder alone is passed, and running embree_verify will have a segmentation fault in SSE4.2.regression_static_memory_monitor from libtbb for an invalid register value. Use regex to select only the test is passed though. This doesn't seem relate to the port and can be listed as a known issue.

   Thread 16 "embree_verify" received signal SIGSEGV, Segmentation fault.
   [Switching to Thread 0x7fffee600140 (LWP 24732)]
   0x00007ffff77d2a50 in tbb::detail::r1::cancel_group_execution(tbb::detail::d1::task_group_context&) () from /lib/riscv64-linux-gnu/libtbb.so.12
   (gdb) bt
   #0  0x00007ffff77d2a50 in tbb::detail::r1::cancel_group_execution(tbb::detail::d1::task_group_context&) () from /lib/riscv64-linux-gnu/libtbb.so.12
   #1  0x00007ffff77d5bee in ?? () from /lib/riscv64-linux-gnu/libtbb.so.12
   #2  0x00007ffff77d666e in ?? () from /lib/riscv64-linux-gnu/libtbb.so.12
   #3  0x00007ffff72f10f4 in start_thread (arg=<optimized out>)
       at ./nptl/pthread_create.c:447
   #4  0x00007ffff7343908 in __thread_start_clone3 ()
       at ../sysdeps/unix/sysv/linux/riscv/clone3.S:71
   (gdb) x/5i $pc-10
      0x7ffff77d2a46 <_ZN3tbb6detail2r122cancel_group_executionERNS0_2d118task_group_contextE+284>:        ld       a5,-16(a4)
      0x7ffff77d2a4a <_ZN3tbb6detail2r122cancel_group_executionERNS0_2d118task_group_contextE+288>:
       beqz        a5,0x7ffff77d2a54 <_ZN3tbb6detail2r122cancel_group_executionERNS0_2d118task_group_contextE+298>
      0x7ffff77d2a4c <_ZN3tbb6detail2r122cancel_group_executionERNS0_2d118task_group_contextE+290>:
       beq s1,a5,0x7ffff77d2ab6 <_ZN3tbb6detail2r122cancel_group_executionERNS0_2d118task_group_contextE+396>
   => 0x7ffff77d2a50 <_ZN3tbb6detail2r122cancel_group_executionERNS0_2d118task_group_contextE+294>:        ld       a5,16(a5)
      0x7ffff77d2a52 <_ZN3tbb6detail2r122cancel_group_executionERNS0_2d118task_group_contextE+296>:
       bnez        a5,0x7ffff77d2a4c <_ZN3tbb6detail2r122cancel_group_executionERNS0_2d118task_group_contextE+290>
   (gdb) i r a5
   a5             0x4445535341505b20       4919429785015311136
   

   ubuntu@ubuntu:~$ embree_verify --no-colors --run ".*regression_static_memory_monitor.*"
                                                   SSE2.regression_static_memory_monitor ............................................................... [PASSED]
                                                 SSE4.2.regression_static_memory_monitor ............................................................... [PASSED]
   
                                                                            Tests passed: 2
                                                                            Tests failed: 0
                                                                Tests failed and ignored: 0
   

[topperc]

I think _mm_setcsr is for floating point rounding modes. So this should be a write to frm not vxrm.

[tingsi5]

Update SSE2RVV to include the proper implementation.