forked from filecoin-project/ec-gpu
-
Notifications
You must be signed in to change notification settings - Fork 0
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
fix: generate OpenCL source at compile time (filecoin-project#28)
The OpenCL kernel source could be generated at run-time, but with this change it makes it more similar to the CUDA version, where we need to compile the kernel at compile-time. This is to improve the symmetry between those features and should also simplify the code for future changes.
- Loading branch information
Showing
4 changed files
with
139 additions
and
113 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -1,106 +1,152 @@ | ||
| /// The build script is needed to compile the CUDA kernel. | ||
| /// | ||
| /// It will compile the kernel at compile time if the `fft` and/or the `multiexp` features are | ||
| /// enabled. | ||
| /// enabled. It also generates the OpenCL source code at compile time. | ||
| #[cfg(all( | ||
| any(feature = "fft", feature = "multiexp"), | ||
| not(feature = "cargo-clippy") | ||
| ))] | ||
| #[path = "src/source.rs"] | ||
| mod source; | ||
|
|
||
| /// The build script is use to generate the CUDA kernel and OpenCL source at compile-time, if the | ||
| /// if the `fft` and/or the `multiexp` features are enabled. | ||
| #[cfg(all( | ||
| feature = "cuda", | ||
| any(feature = "fft", feature = "multiexp"), | ||
| not(feature = "cargo-clippy") | ||
| ))] | ||
| fn main() { | ||
| #[cfg(feature = "cuda")] | ||
| kernel::generate_cuda(); | ||
| #[cfg(feature = "opencl")] | ||
| kernel::generate_opencl(); | ||
| } | ||
|
|
||
| // This is a hack for the case when we run Clippy while we don't generate any GPU kernel. For | ||
| // Clippy we don't need a proper source or properly compiled kernel, but just some arbitrary bytes. | ||
| #[cfg(not(all( | ||
| any(feature = "fft", feature = "multiexp"), | ||
| not(feature = "cargo-clippy") | ||
| )))] | ||
| fn main() { | ||
| println!("cargo:rustc-env=CUDA_KERNEL_FATBIN=../build.rs"); | ||
| println!("cargo:rustc-env=OPENCL_KERNEL_SOURCE=../build.rs"); | ||
| } | ||
|
|
||
| // Put the code into a module, so that we need to repeat the feature flags less often. | ||
| #[cfg(all( | ||
| any(feature = "fft", feature = "multiexp"), | ||
| not(feature = "cargo-clippy") | ||
| ))] | ||
| mod kernel { | ||
| use std::path::PathBuf; | ||
| use std::process::Command; | ||
| use std::{env, fs}; | ||
|
|
||
| use blstrs::Bls12; | ||
| use sha2::{Digest, Sha256}; | ||
|
|
||
| #[path = "src/source.rs"] | ||
| mod source; | ||
|
|
||
| // This is a hack for the case when the documentation is built on docs.rs. For the | ||
| // documentation we don't need a properly compiled kernel, but just some arbitrary bytes. | ||
| if env::var("DOCS_RS").is_ok() { | ||
| println!("cargo:rustc-env=CUDA_KERNEL_FATBIN=../build.rs"); | ||
| return; | ||
| } | ||
| #[cfg(feature = "cuda")] | ||
| pub(crate) fn generate_cuda() { | ||
| use sha2::{Digest, Sha256}; | ||
|
|
||
| let kernel_source = source::gen_source::<Bls12, source::Limb32>(); | ||
|
|
||
| let out_dir = env::var("OUT_DIR").expect("OUT_DIR was not set."); | ||
|
|
||
| // Make it possible to override the default options. Though the source and output file is | ||
| // always set automatically. | ||
| let mut nvcc = match env::var("EC_GPU_CUDA_NVCC_ARGS") { | ||
| Ok(args) => execute::command(format!("nvcc {}", args)), | ||
| Err(_) => { | ||
| let mut command = Command::new("nvcc"); | ||
| command | ||
| .arg("--optimize=6") | ||
| // Compile with as many threads as CPUs are available. | ||
| .arg("--threads=0") | ||
| .arg("--fatbin") | ||
| .arg("--gpu-architecture=sm_86") | ||
| .arg("--generate-code=arch=compute_86,code=sm_86") | ||
| .arg("--generate-code=arch=compute_80,code=sm_80") | ||
| .arg("--generate-code=arch=compute_75,code=sm_75"); | ||
| command | ||
| // This is a hack for the case when the documentation is built on docs.rs. For the | ||
| // documentation we don't need a properly compiled kernel, but just some arbitrary bytes. | ||
| if env::var("DOCS_RS").is_ok() { | ||
| println!("cargo:rustc-env=CUDA_KERNEL_FATBIN=../build.rs"); | ||
| return; | ||
| } | ||
| }; | ||
|
|
||
| // Hash the source and and the compile flags. Use that as the filename, so that the kernel is | ||
| // only rebuilt if any of them change. | ||
| let mut hasher = Sha256::new(); | ||
| hasher.update(kernel_source.as_bytes()); | ||
| hasher.update(&format!("{:?}", &nvcc)); | ||
| let kernel_digest = hex::encode(hasher.finalize()); | ||
|
|
||
| let source_path: PathBuf = [&out_dir, &format!("{}.cu", &kernel_digest)] | ||
| .iter() | ||
| .collect(); | ||
| let fatbin_path: PathBuf = [&out_dir, &format!("{}.fatbin", &kernel_digest)] | ||
| .iter() | ||
| .collect(); | ||
|
|
||
| fs::write(&source_path, &kernel_source).unwrap_or_else(|_| { | ||
| panic!( | ||
| "Cannot write kernel source at {}.", | ||
| source_path.to_str().unwrap() | ||
| ) | ||
| }); | ||
|
|
||
| // Only compile if the output doesn't exist yet. | ||
| if !fatbin_path.as_path().exists() { | ||
| let status = nvcc | ||
| .arg("--output-file") | ||
| .arg(&fatbin_path) | ||
| .arg(&source_path) | ||
| .status() | ||
| .expect("Cannot run nvcc. Install the NVIDIA toolkit or disable the `cuda` feature."); | ||
|
|
||
| if !status.success() { | ||
|
|
||
| let kernel_source = crate::source::gen_source::<Bls12, crate::source::Limb32>(); | ||
| let out_dir = env::var("OUT_DIR").expect("OUT_DIR was not set."); | ||
|
|
||
| // Make it possible to override the default options. Though the source and output file is | ||
| // always set automatically. | ||
| let mut nvcc = match env::var("EC_GPU_CUDA_NVCC_ARGS") { | ||
| Ok(args) => execute::command(format!("nvcc {}", args)), | ||
| Err(_) => { | ||
| let mut command = std::process::Command::new("nvcc"); | ||
| command | ||
| .arg("--optimize=6") | ||
| // Compile with as many threads as CPUs are available. | ||
| .arg("--threads=0") | ||
| .arg("--fatbin") | ||
| .arg("--gpu-architecture=sm_86") | ||
| .arg("--generate-code=arch=compute_86,code=sm_86") | ||
| .arg("--generate-code=arch=compute_80,code=sm_80") | ||
| .arg("--generate-code=arch=compute_75,code=sm_75"); | ||
| command | ||
| } | ||
| }; | ||
|
|
||
| // Hash the source and the compile flags. Use that as the filename, so that the kernel is only | ||
| // rebuilt if any of them change. | ||
| let mut hasher = Sha256::new(); | ||
| hasher.update(kernel_source.as_bytes()); | ||
| hasher.update(&format!("{:?}", &nvcc)); | ||
| let kernel_digest = hex::encode(hasher.finalize()); | ||
|
|
||
| let source_path: PathBuf = [&out_dir, &format!("{}.cu", &kernel_digest)] | ||
| .iter() | ||
| .collect(); | ||
| let fatbin_path: PathBuf = [&out_dir, &format!("{}.fatbin", &kernel_digest)] | ||
| .iter() | ||
| .collect(); | ||
|
|
||
| fs::write(&source_path, &kernel_source).unwrap_or_else(|_| { | ||
| panic!( | ||
| "nvcc failed. See the kernel source at {}", | ||
| "Cannot write kernel source at {}.", | ||
| source_path.to_str().unwrap() | ||
| ); | ||
| ) | ||
| }); | ||
|
|
||
| // Only compile if the output doesn't exist yet. | ||
| if !fatbin_path.as_path().exists() { | ||
| let status = nvcc | ||
| .arg("--output-file") | ||
| .arg(&fatbin_path) | ||
| .arg(&source_path) | ||
| .status() | ||
| .expect( | ||
| "Cannot run nvcc. Install the NVIDIA toolkit or disable the `cuda` feature.", | ||
| ); | ||
|
|
||
| if !status.success() { | ||
| panic!( | ||
| "nvcc failed. See the kernel source at {}", | ||
| source_path.to_str().unwrap() | ||
| ); | ||
| } | ||
| } | ||
|
|
||
| // The idea to put the path to the farbin into a compile-time env variable is from | ||
| // https://github.com/LutzCle/fast-interconnects-demo/blob/b80ea8e04825167f486ab8ac1b5d67cf7dd51d2c/rust-demo/build.rs | ||
| println!( | ||
| "cargo:rustc-env=CUDA_KERNEL_FATBIN={}", | ||
| fatbin_path.to_str().unwrap() | ||
| ); | ||
| } | ||
|
|
||
| // The idea to put the path to the farbin into a compile-time env variable is from | ||
| // https://github.com/LutzCle/fast-interconnects-demo/blob/b80ea8e04825167f486ab8ac1b5d67cf7dd51d2c/rust-demo/build.rs | ||
| println!( | ||
| "cargo:rustc-env=CUDA_KERNEL_FATBIN={}", | ||
| fatbin_path.to_str().unwrap() | ||
| ); | ||
| } | ||
| #[cfg(feature = "opencl")] | ||
| pub(crate) fn generate_opencl() { | ||
| let kernel_source = crate::source::gen_source::<Bls12, crate::source::Limb64>(); | ||
| let out_dir = env::var("OUT_DIR").expect("OUT_DIR was not set."); | ||
|
|
||
| #[cfg(not(all( | ||
| feature = "cuda", | ||
| any(feature = "fft", feature = "multiexp"), | ||
| not(feature = "cargo-clippy") | ||
| )))] | ||
| fn main() { | ||
| // This is a hack for the case when the `cuda` and `cargo-clippy` features are enabled. For | ||
| // Clippy we don't need a properly compiled kernel, but just some arbitrary bytes. | ||
| println!("cargo:rustc-env=CUDA_KERNEL_FATBIN=../build.rs"); | ||
| // Generating the kernel source is cheap, hence use a fixed name and override it on every | ||
| // build. | ||
| let source_path: PathBuf = [&out_dir, "kernel.cl"].iter().collect(); | ||
|
|
||
| fs::write(&source_path, &kernel_source).unwrap_or_else(|_| { | ||
| panic!( | ||
| "Cannot write kernel source at {}.", | ||
| source_path.to_str().unwrap() | ||
| ) | ||
| }); | ||
|
|
||
| // For OpenCL we only need the kernel source, it is compiled at runtime. | ||
| #[cfg(feature = "opencl")] | ||
| println!( | ||
| "cargo:rustc-env=OPENCL_KERNEL_SOURCE={}", | ||
| source_path.to_str().unwrap() | ||
| ); | ||
| } | ||
| } |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters