core::slice in kernel

[termoshtt]

This PR try to use core::slice in a kernel function

#[kernel]
#[depends("accel-core" = "0.2.0-alpha")]
pub unsafe fn add(a: &[f64], b: &[f64], c: &mut [f64]) {
    let i = accel_core::index() as usize;
    let n = c.len();
    if i < n {
        c[i] = a[i] + b[i]
    }
}

Problem

This kernel yields an error about PTX:

called `Result::unwrap()` on an `Err` value: cudaError(CUDA_ERROR_INVALID_PTX)

PTX-builder links rust code into LLVM byte-code (kernel.bc) while macro #[kernel] of accel-derive, and it will be loaded into accel::Module. "Invalid PTX" erro occurs on this loading. To get more information about how this PTX is invalid, I have tried to load PTX file manually:

llc -mcpu=sm_20 kernel.bc -o kernel.ptx  # compile LLVM byptecode to PTX
nvcc --cubin kernel.ptx  # Convert PTX to cubin

llc works, but nvcc raises an error:

ptxas fatal   : Unresolved extern function '_ZN4core9panicking18panic_bounds_check17h476c69b1512db11aE'

This function seems to be core::panicking::panic_bounds_check, which should be contained in libcore.
(#11 is opened here on Jan, and I still stay here)

[termoshtt]

Link error about panic_bounds_check has been resolved, but the result of add is still invalid (´・ω・｀)

a = [0.0, 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0, 12.0, 13.0, 14.0, 15.0]
b = [0.0, 2.0, 4.0, 6.0, 8.0, 10.0, 12.0, 14.0, 16.0, 18.0, 20.0, 22.0, 24.0, 26.0, 28.0, 30.0]
c = [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]

[sandeep-datta]

Can you try again with sm_30 as the target arch? Sm_20 has been deprecated.

[termoshtt]

moved https://gitlab.com/termoshtt/accel/issues/7