Hi,
I ran the following with environment variables NUMBA_JIT_COVERAGE=1 and NUMBA_COVERAGE=1.
import numpy as np
from numba import cuda
import numba as nb
@nb.jit
def fn_w_branch(x):
x[1] = 1
x[2] = x[0] + x[1]
if x[0] == 0:
x[0] = 5
else:
x[0] = 7
@cuda.jit
def fn_1(x):
x[0] = 5
@cuda.jit
def fn(x):
x[0] = dev_fn(5)
@cuda.jit(device=True)
def dev_fn(x):
if x == 5:
return x + 5
else:
return x + 7
items = np.zeros(12)
fn_1[1,1](items)
print(items)
fn[1,1](items)
print(items)
fn_w_branch(items)
print(items)
The output I get after coverage run file.py coverage report -m is as follows:
Name Stmts Miss Cover Missing
---------------------------------------------
reproducer.py 28 5 82% 16, 20, 24-27
---------------------------------------------
TOTAL 28 5 82%
It seems the regular numba jit compiled code was measured but not the cuda jit code. Is there a way to make cuda jit compiled code emit coverage data?
Thanks.
Hi,
I ran the following with environment variables
NUMBA_JIT_COVERAGE=1andNUMBA_COVERAGE=1.The output I get after
coverage run file.pycoverage report -mis as follows:It seems the regular numba jit compiled code was measured but not the cuda jit code. Is there a way to make cuda jit compiled code emit coverage data?
Thanks.