|
| 1 | +# Usage |
| 2 | + |
| 3 | +This feature is work-in-progress, and not ready for usage. The instructions here are for contributors, or people interested in following the latest progress. |
| 4 | +We currently work on launching the following Rust kernel on the GPU. To follow along, copy it to a `src/lib.rs` file. |
| 5 | + |
| 6 | +```rust |
| 7 | +#![feature(abi_gpu_kernel)] |
| 8 | +#![no_std] |
| 9 | + |
| 10 | +#[cfg(target_os = "linux")] |
| 11 | +extern crate libc; |
| 12 | +#[cfg(target_os = "linux")] |
| 13 | +use libc::c_char; |
| 14 | + |
| 15 | +use core::mem; |
| 16 | + |
| 17 | +#[panic_handler] |
| 18 | +fn panic(_: &core::panic::PanicInfo) -> ! { |
| 19 | + loop {} |
| 20 | +} |
| 21 | + |
| 22 | +#[cfg(target_os = "linux")] |
| 23 | +#[unsafe(no_mangle)] |
| 24 | +#[inline(never)] |
| 25 | +fn main() { |
| 26 | + let array_c: *mut [f64; 256] = |
| 27 | + unsafe { libc::calloc(256, (mem::size_of::<f64>()) as libc::size_t) as *mut [f64; 256] }; |
| 28 | + let output = c"The first element is zero %f\n"; |
| 29 | + let output2 = c"The first element is NOT zero %f\n"; |
| 30 | + let output3 = c"The second element is %f\n"; |
| 31 | + unsafe { |
| 32 | + let val: *const c_char = if (*array_c)[0] < 0.1 { |
| 33 | + output.as_ptr() |
| 34 | + } else { |
| 35 | + output2.as_ptr() |
| 36 | + }; |
| 37 | + libc::printf(val, (*array_c)[0]); |
| 38 | + } |
| 39 | + |
| 40 | + unsafe { |
| 41 | + kernel_1(array_c); |
| 42 | + } |
| 43 | + core::hint::black_box(&array_c); |
| 44 | + unsafe { |
| 45 | + let val: *const c_char = if (*array_c)[0] < 0.1 { |
| 46 | + output.as_ptr() |
| 47 | + } else { |
| 48 | + output2.as_ptr() |
| 49 | + }; |
| 50 | + libc::printf(val, (*array_c)[0]); |
| 51 | + libc::printf(output3.as_ptr(), (*array_c)[1]); |
| 52 | + } |
| 53 | +} |
| 54 | + |
| 55 | +#[cfg(target_os = "linux")] |
| 56 | +unsafe extern "C" { |
| 57 | + pub fn kernel_1(array_b: *mut [f64; 256]); |
| 58 | +} |
| 59 | +``` |
| 60 | + |
| 61 | +## Compile instructions |
| 62 | +It is important to use a clang compiler build on the same llvm as rustc. Just calling clang without the full path will likely use your system clang, which probably will be incompatible. So either substitute clang/lld invocations below with absolute path, or set your `PATH` accordingly. |
| 63 | + |
| 64 | +First we generate the host (cpu) code. The first build is just to compile libc, take note of the hashed path. Then we call rustc directly to build our host code, while providing the libc artifact to rustc. |
| 65 | +``` |
| 66 | +cargo +offload build -r -v |
| 67 | +rustc +offload --edition 2024 src/lib.rs -g --crate-type cdylib -C opt-level=3 -C panic=abort -C lto=fat -L dependency=/absolute_path_to/target/release/deps --extern libc=/absolute_path_to/target/release/deps/liblibc-<HASH>.rlib --emit=llvm-bc,llvm-ir -Zoffload=Enable -Zunstable-options |
| 68 | +``` |
| 69 | + |
| 70 | +Now we generate the device code. Replace the target-cpu with the right code for your gpu. |
| 71 | +``` |
| 72 | +RUSTFLAGS="-Ctarget-cpu=gfx90a --emit=llvm-bc,llvm-ir" cargo +offload build -Zunstable-options -r -v --target amdgcn-amd-amdhsa -Zbuild-std=core |
| 73 | +``` |
| 74 | + |
| 75 | +Now find the <libname>.ll under target/amdgcn-amd-amdhsa folder and copy it to a device.ll file (or adjust the file names below). |
| 76 | +If you work on an NVIDIA or Intel gpu, please adjust the names acordingly and open an issue to share your results (either if you succeed or fail). |
| 77 | +First we compile our .ll files (good for manual inspections) to .bc files and clean up leftover artifacts. The cleanup is important, otherwise caching might interfere on following runs. |
| 78 | +``` |
| 79 | +opt lib.ll -o lib.bc |
| 80 | +opt device.ll -o device.bc |
| 81 | +rm *.o |
| 82 | +rm bare.amdgcn.gfx90a.img* |
| 83 | +``` |
| 84 | + |
| 85 | +``` |
| 86 | +clang-offload-packager" "-o" "host.out" "--image=file=device.bc,triple=amdgcn-amd-amdhsa,arch=gfx90a,kind=openmp" |
| 87 | +
|
| 88 | +clang-21" "-cc1" "-triple" "x86_64-unknown-linux-gnu" "-S" "-save-temps=cwd" "-disable-free" "-clear-ast-before-backend" "-main-file-name" "lib.rs" "-mrelocation-model" "pic" "-pic-level" "2" "-pic-is-pie" "-mframe-pointer=all" "-fmath-errno" "-ffp-contract=on" "-fno-rounding-math" "-mconstructor-aliases" "-funwind-tables=2" "-target-cpu" "x86-64" "-tune-cpu" "generic" "-resource-dir" "/<ABSOLUTE_PATH_TO>/rust/build/x86_64-unknown-linux-gnu/llvm/lib/clang/21" "-ferror-limit" "19" "-fopenmp" "-fopenmp-offload-mandatory" "-fgnuc-version=4.2.1" "-fskip-odr-check-in-gmf" "-fembed-offload-object=host.out" "-fopenmp-targets=amdgcn-amd-amdhsa" "-faddrsig" "-D__GCC_HAVE_DWARF2_CFI_ASM=1" "-o" "host.s" "-x" "ir" "lib.bc" |
| 89 | +
|
| 90 | +clang-21" "-cc1as" "-triple" "x86_64-unknown-linux-gnu" "-filetype" "obj" "-main-file-name" "lib.rs" "-target-cpu" "x86-64" "-mrelocation-model" "pic" "-o" "host.o" "host.s" |
| 91 | +
|
| 92 | +clang-linker-wrapper" "--should-extract=gfx90a" "--device-compiler=amdgcn-amd-amdhsa=-g" "--device-compiler=amdgcn-amd-amdhsa=-save-temps=cwd" "--device-linker=amdgcn-amd-amdhsa=-lompdevice" "--host-triple=x86_64-unknown-linux-gnu" "--save-temps" "--linker-path=/ABSOlUTE_PATH_TO/rust/build/x86_64-unknown-linux-gnu/lld/bin/ld.lld" "--hash-style=gnu" "--eh-frame-hdr" "-m" "elf_x86_64" "-pie" "-dynamic-linker" "/lib64/ld-linux-x86-64.so.2" "-o" "bare" "/lib/../lib64/Scrt1.o" "/lib/../lib64/crti.o" "/ABSOLUTE_PATH_TO/crtbeginS.o" "-L/ABSOLUTE_PATH_TO/rust/build/x86_64-unknown-linux-gnu/llvm/bin/../lib/x86_64-unknown-linux-gnu" "-L/ABSOLUTE_PATH_TO/rust/build/x86_64-unknown-linux-gnu/llvm/lib/clang/21/lib/x86_64-unknown-linux-gnu" "-L/lib/../lib64" "-L/usr/lib64" "-L/lib" "-L/usr/lib" "host.o" "-lstdc++" "-lm" "-lomp" "-lomptarget" "-L/ABSOLUTE_PATH_TO/rust/build/x86_64-unknown-linux-gnu/llvm/lib" "-lgcc_s" "-lgcc" "-lpthread" "-lc" "-lgcc_s" "-lgcc" "/ABSOLUTE_PATH_TO/crtendS.o" "/lib/../lib64/crtn.o" |
| 93 | +``` |
| 94 | + |
| 95 | +Especially for the last command I recommend to not fix the paths, but rather just re-generate them by copying a bare-mode openmp example and compiling it with your clang. By adding `-###` to your clang invocation, you can see the invidual steps. |
| 96 | +``` |
| 97 | +myclang++ -fuse-ld=lld -O3 -fopenmp -fopenmp-offload-mandatory --offload-arch=gfx90a omp_bare.cpp -o main -### |
| 98 | +``` |
| 99 | + |
| 100 | +In the final step, you can now run your binary |
| 101 | + |
| 102 | +``` |
| 103 | +./main |
| 104 | +The first element is zero 0.000000 |
| 105 | +The first element is NOT zero 21.000000 |
| 106 | +The second element is 0.000000 |
| 107 | +``` |
| 108 | + |
| 109 | +To receive more information about the memory transfer, you can enable info printing with |
| 110 | +``` |
| 111 | +LIBOMPTARGET_INFO=-1 ./main |
| 112 | +``` |
0 commit comments