Fixes for GPU tests and GPU workflow

.github/workflows/selfhosted.yml

@@ -0,0 +1,23 @@
+# CI workflow that runs on selfhosted GPU
+name: Tests with gpu
+
+on:
+  pull_request:
+    types: [labeled]
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.head_ref }}
+  cancel-in-progress: true
+
+jobs:
+  build:
+    if: contains(join(github.event.pull_request.labels.*.name), 'run-on')
+    uses: qiboteam/workflows/.github/workflows/selfhosted.yml@v1
+    with:
+      used-labels: ${{ toJSON(github.event.pull_request.labels.*.name) }}
+      python-version: "3.10"
+      artifact-url: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
+      poetry-extras: "--with cuda11,tests"
+
+    secrets:
+      repo_token: ${{ secrets.GITHUB_TOKEN }}

src/qibojit/backends/clifford_operations_cpu.py

@@ -3,7 +3,7 @@
 import numpy as np
 from numba import njit, prange, uint64
 
-PARALLEL = False
+PARALLEL = True
 
 
 @njit("Tuple((u1[:], u1[:,:], u1[:,:]))(u1[:,:], u8)", parallel=PARALLEL, cache=True)

src/qibojit/backends/cpu.py

@@ -64,6 +64,7 @@ def __init__(self):
             4: self.gates.apply_four_qubit_gate_kernel,
             5: self.gates.apply_five_qubit_gate_kernel,
         }
+
         if sys.platform == "darwin":  # pragma: no cover
             self.set_threads(psutil.cpu_count(logical=False))
         else:

src/qibojit/tests/test_backends.py

@@ -20,6 +20,7 @@ def test_thread_setter(backend):
     original_threads = numba.get_num_threads()
     backend.set_threads(1)
     assert numba.get_num_threads() == 1
+    backend.set_threads(original_threads)
 
 
 @pytest.mark.parametrize("array_type", [None, "float32", "float64"])