Breakpoints

A breakpoint makes your program stop whenever a certain point in the program is reached.

You can set breakpoints with the break command to specify the place where your program should stop in the kernel. Define breakpoints by line numbers or function names.

You have several ways to set breakpoints:

• break <function>

• break <filename>:<linenumber>

• break <filename>:<function>

• break … if <condition>

See also:

• Breakpoints in GDB*.

Consider the following numba-dpex kernel code (refer numba_dpex/examples/debug/simple_sum.py for full example):

import dpnp as np

import numba_dpex as ndpx


@ndpx.kernel(debug=True)
def data_parallel_sum(item, a, b, c):
    i = item.get_id(0)
    c[i] = a[i] + b[i]  # Condition breakpoint location


global_size = 10
N = global_size

a = np.array(np.random.random(N), dtype=np.float32)
b = np.array(np.random.random(N), dtype=np.float32)
c = np.ones_like(a)

ndpx.call_kernel(data_parallel_sum, ndpx.Range(global_size), a, b, c)

print("Done...")

break function

The debugger output:

$ NUMBA_OPT=0 gdb-oneapi -q python
(gdb) set breakpoint pending on
(gdb) break data_parallel_sum
(gdb) run simple_sum.py
...
Thread 2.2 hit Breakpoint 1, with SIMD lanes [0-7], __main__::data_parallel_sum () at simple_sum.py:20
20          @dpex.kernel(debug=True)
(gdb) continue
Thread 2.3 hit Breakpoint 1, with SIMD lanes [0-1], __main__::data_parallel_sum () at simple_sum.py:20
20          @dpex.kernel(debug=True)
(gdb) continue
...
Done...

break filename:linenumber

The debugger output:

$ NUMBA_OPT=0 gdb-oneapi -q python
(gdb) set breakpoint pending on
(gdb) break simple_sum.py:20
(gdb) run simple_sum.py
...
Thread 2.2 hit Breakpoint 1, with SIMD lanes [0-7], __main__::data_parallel_sum () at simple_sum.py:20
20          @dpex.kernel(debug=True)
(gdb) continue
Thread 2.3 hit Breakpoint 1, with SIMD lanes [0-1], __main__::data_parallel_sum () at simple_sum.py:20
20          @dpex.kernel(debug=True)
(gdb) continue
...
Done...

break filename:function

The debugger output:

$ NUMBA_OPT=0 gdb-oneapi -q python
(gdb) set breakpoint pending on
(gdb) break simple_sum.py:data_parallel_sum
(gdb) run simple_sum.py
...
Thread 2.2 hit Breakpoint 1, with SIMD lanes [0-7], __main__::data_parallel_sum () at simple_sum.py:20
20          @dpex.kernel(debug=True)
(gdb) continue
Thread 2.3 hit Breakpoint 1, with SIMD lanes [0-1], __main__::data_parallel_sum () at simple_sum.py:20
20          @dpex.kernel(debug=True)
(gdb) continue
...
Done...

break … if cond

The debugger output:

$ NUMBA_OPT=0 gdb-oneapi -q python
(gdb) set breakpoint pending on
(gdb) break simple_sum.py:23 if i == 1
(gdb) run simple_sum.py
...
Thread 2.2 hit Breakpoint 1, with SIMD lane 1, __main__::data_parallel_sum () at simple_sum.py:23
23          c[i] = a[i] + b[i]
(gdb) print i
$1 = 1
(gdb) continue
...
Done...

Breakpoints with nested functions

Consider numba-dpex kernel code. See the source file numba_dpex/examples/debug/simple_dpex_func.py:

import dpnp as np

import numba_dpex as ndpx


@ndpx.device_func(debug=True)
def func_sum(a_in_func, b_in_func):
    result = a_in_func + b_in_func  # breakpoint location
    return result


@ndpx.kernel(debug=True)
def kernel_sum(item, a_in_kernel, b_in_kernel, c_in_kernel):
    i = item.get_id(0)
    c_in_kernel[i] = func_sum(a_in_kernel[i], b_in_kernel[i])


global_size = 10
a = np.arange(global_size, dtype=np.float32)
b = np.arange(global_size, dtype=np.float32)
c = np.empty_like(a)

ndpx.call_kernel(kernel_sum, ndpx.Range(global_size), a, b, c)

print("Done...")

The debugger output:

$ NUMBA_OPT=0 gdb-oneapi -q python
(gdb) set breakpoint pending on
(gdb) break simple_dpex_func.py:func_sum
(gdb) run simple_dpex_func.py
...
Thread 2.2 hit Breakpoint 1, with SIMD lanes [0-7], __main__::func_sum () at simple_dpex_func.py:22
22          result = a_in_func + b_in_func
(gdb) continue
Thread 2.3 hit Breakpoint 1, with SIMD lanes [0-1], __main__::func_sum () at simple_dpex_func.py:22
22          result = a_in_func + b_in_func
(gdb) continue
...
Done...