Mingw下malloc与calloc性能相当？是否底层调用calloc？

Great question! The stark performance difference you're observing between Linux and Mingw boils down to fundamental gaps in how each platform's C runtime handles memory allocation—not necessarily that Mingw's malloc directly calls calloc under the hood. Let’s break this down clearly:

1. Linux Glibc: malloc vs calloc = No Zeroing vs Wasted Zeroing #

On Linux, the GNU C Library (glibc) draws a sharp line between these two functions:

• malloc grabs memory (either from the process heap or via mmap for large blocks) but skips zeroing it entirely. For heap allocations, you’ll get leftover garbage data from previous uses; even for mmap-backed blocks (which the kernel zeros), malloc avoids any user-space zeroing overhead.
• calloc does double duty: allocates memory and zeros every single byte of it. When you only use 0.1% of a huge allocation, calloc wastes massive cycles zeroing out 99.9% of memory you’ll never touch—this is exactly why you saw a 100x slowdown compared to malloc.

2. Windows CRT: malloc Often Delivers Zeroed Memory by Default #

Mingw relies on Windows’ native C Runtime (CRT) like UCRT or msvcrt, and Windows’ memory model works very differently:

• For large allocations, the CRT’s malloc uses VirtualAlloc under the hood. Windows kernel policy mandates that all newly mapped virtual memory pages are zeroed (to prevent sensitive data leaks between processes). This means malloc already gives you memory identical to what calloc would provide—no extra zeroing step needed.
• The CRT’s calloc implementation leverages this: instead of running a separate zeroing routine, it may just call malloc directly (with minor size validation) since the memory is already zeroed. Even for smaller heap allocations (via HeapAlloc), Windows’ heap manager often returns zeroed memory, or the zeroing overhead is negligible for small blocks.
• End result: malloc and calloc do nearly identical work in Mingw, so you see no meaningful performance gap.

3. Quick Ways to Verify This (If You’re Curious) #

If you want to confirm this behavior for yourself:

• Write a small test program that allocates a large block with malloc, then checks if the first few thousand bytes are zero. On Mingw, they’ll almost certainly be all zeros.
• Use a debugger like GDB to step into malloc and calloc calls in your Mingw-compiled program—you’ll likely see both functions eventually route to the same Windows kernel allocation functions.

内容的提问来源于stack exchange，提问作者Luke