Skip to content
KernelIndex
Search⌘K

submission 623790

0xdev_fiver · python · License unknown

Use it

Vendorable · source mirrored · license unknownView source →

No package. Vendor the mirrored source: 46 lines, June 9 Researcher Reciprocity License v1.0.

submission.py
curl "https://kernelindex.com/api/v1/implementations/kernelbot-vectoradd-v2-623790?include=source"
interfacepython
Compatibility
measured onNVIDIA A100
declared hardwareNVIDIA A100
architecturessm_80
dtypesfp16

Benchmark evidence

1 measurement across 1 GPU, fastest first.

Operation / workload
Hardware
Latency
Rank
Observed
FP16 vector additionsuite of 5 cases
NVIDIA A100
913.7µs
#18 of 87
2026-03-24

Reported · How evidence levels are derived →

Source and license

sourceavailable
revision digestsha256:e3804683a649f3a5b39e94cdbf67980a6fe167d027fd4c13ea25432d1ba83c53
license declaredunknown
license concludedunknown
authors0xdev_fiver
imported2026-08-15

Kernel source

submission.py46 lines
from utils import make_match_reference, DeterministicContext
import torch
from task import input_t, output_t

@torch.compile
def custom_kernel(data: input_t) -> output_t:
    A, B, output = data
    output[...] = A + B
    return output
    
    
def ref_kernel(data: input_t) -> output_t:
    """
    Reference implementation of vector addition using PyTorch.
    Args:
        data: Tuple of tensors [A, B] to be added.
    Returns:
        Tensor containing element-wise sums.
    """
    with DeterministicContext():
        A, B, output = data
        output[...] = A + B
        return output


def generate_input(size: int, seed: int) -> input_t:
    """
    Generates random input tensors of specified shapes.
    Returns:
        Tuple of tensors [A, B] to be added.
    """
    gen = torch.Generator(device="cuda")
    gen.manual_seed(seed)
    A = torch.randn(
        size, size, device="cuda", dtype=torch.float16, generator=gen
    ).contiguous()
    B = torch.randn(
        size, size, device="cuda", dtype=torch.float16, generator=gen
    ).contiguous()
    C = torch.empty(size, size, device="cuda", dtype=torch.float16).contiguous()
    return A, B, C


check_implementation = make_match_reference(ref_kernel)

scrolls · 46 lines total

Source code from GPU Mode and the KernelBot dataset · June 9 Researcher Reciprocity License v1.0

Best evidence level for this revision: reported

JSON