Logging

1
2
3
4
5

set logging on
set logging file gdb_output.txt
set logging overwrite on      # optional, overwrites existing file
set logging redirect on       # No echoing on terminal
set logging off

Breaking Execution

Function Call

1

break function_name

1

break file.c:function_name

1

break 'Class::Method(Type)'

1
2

set breakpoint pending on
break function_name

Specific Line

1

break file.c:42

1

break 42

When Symbol Comes into Scope

1

rbreak symbol_name

When Symbol Changes Value

1
2
3

watch variable        # on write
rwatch variable       # on read
awatch variable       # on read/write

Requires variable to have a stable memory address (e.g., global or in-scope local).

I/O

Set Pretty Printing

1

set print pretty on

1

set print elements N

Limits number of elements printed (default: 200).

Dump Binary Memory

1

dump binary memory filename start_addr end_addr

Example:

1

dump binary memory dump.bin 0x600000 0x601000

1
2
3

set $start = &v[0]
set $end = $start + v.size()
dump binary memory <FILE_NAME>.bin $start $end

TL;DR (First Principles):

When GDB executes a program, it maps your variables—including arrays and matrices—into memory. You can tell GDB to dump raw memory of any variable to a binary file. This lets you analyze the data outside GDB (e.g., with Python + NumPy) without GDB choking on printing large data inline.

1. Understand What You're Dumping

You need:
- The starting memory address (&my_matrix)
- The exact byte size to dump (sizeof(my_matrix))
-
Example: for a float my_matrix[100][100]:

1

sizeof(my_matrix) = 100 * 100 * sizeof(float) = 40000 bytes

2. Basic Syntax

1

dump binary memory <filename> <start_address> <end_address>

Or if you know the size:

1

dump binary memory matrix.bin &my_matrix &my_matrix + sizeof(my_matrix)

But GDB doesn't support pointer arithmetic like that directly on sizeof. Instead:

1

dump binary memory matrix.bin &my_matrix (&my_matrix)+40000

(You can replace 40000 with any exact byte count)

Or calculate in GDB:

1

gdb p sizeof(my_matrix)

Then manually use the value:

1

dump binary memory matrix.bin &my_matrix (&my_matrix)+40000

What Happens Internally

• GDB interprets the address &my_matrix as a raw memory location.
• It writes the raw byte sequence (just 1s and 0s) to matrix.bin.

Load It in Python (Post-GDB)

1
2
3
4
5

import numpy as np

data = np.fromfile("matrix.bin", dtype=np.float32)
matrix = data.reshape((100, 100))
print(matrix)

Make sure dtype and shape match your C++ matrix layout.

🧪 Sanity Check in GDB

Want to confirm it’s dumping the correct values?

1

x/10f &my_matrix[0][0]   # examine 10 floats

Then check if those match the first line of matrix[0][:10] in NumPy.