E6042

Error · Topology/Hardware

E6042 — Code on device 1 or a later device of a declared GPU (arch: nvptx64 with count:). The CUDA runtime gets only device 0 of a declared GPU, and the memory of every device is on GPU 0, so the code has no GPU to run on yet. Run it on device 0.

What the compiler reports

Error[E6042] at {{.*}}.vx:[[@LINE+1]]:{{[0-9]+}}: `on Topology::Gpus[1]` of function 'runs': device 1 of 'Gpus' has no GPU to run on yet, because the CUDA runtime gets only device 0 of a declared GPU. Run the code on `Topology::Gpus[0]`
Error[E6042] at {{.*}}.vx:[[@LINE+1]]:3: `spawn on(Topology::Gpus[1])`: device 1 of 'Gpus' has no GPU to run on yet
2 error(s) emitted

The fragments the test suite holds the compiler to. An ellipsis marks text the test does not constrain; the emitted message also carries a source location.

A program that triggers it

//
// Part of the Vx Project, under the Apache License v2.0 with LLVM Exceptions.
// See LICENSE for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
// `Gpus` is a GPU with two devices. For now the CUDA runtime gets only device 0,
// so code on device 1 is error E6042, in an `on` clause and in a `spawn`. Code
// on device 0 and a tensor placed on device 1 are fine.
//

Memory GPU_RAM {
  scope: device, capacity: 1 GB, bandwidth: 1 TB/s
}

Topology Gpus {
  count: 2,
  arch: nvptx64,
  memory: Memory::GPU_RAM,
  visible: [Memory::GPU_RAM],
  transfer Memory::CPU_DRAM -> Memory::GPU_RAM : 100,
}

fn placed(_t : Tensor<f32, [4], Topology::Gpus[1]>) -> i32 {
  return 0;
}

fn runs() on Topology::Gpus[1] -> i32 {
  return 0;
}

fn main() -> i32 {
  spawn on(Topology::Gpus[0]) {
    let _a = 1;
  }
  spawn on(Topology::Gpus[1]) {
    let _a = 2;
  }
  return 0;
}

From tests/frontend/fail/code_on_device_1_of_a_declared_gpu.vx, which asserts this diagnostic on every commit.

Related