How to Seamlessly Port C/C++ Applications from x86 to ARM64 TaiShan Servers

After migrating two projects from x86 servers to aarch64 TaiShan servers, the author summarizes the experience and provides a detailed migration guide.

Programming Language Overview

Languages can be divided into compiled and interpreted categories. Compiled languages such as C/C++ produce machine instructions that are architecture‑specific, while interpreted languages like Java and Python generate platform‑independent bytecode executed by a virtual machine.

Compiled C/C++ programs built for x86 cannot run directly on TaiShan servers and require a porting compilation; interpreted Java/Python programs can run unchanged, though native libraries (e.g., .so files) must be recompiled.

Preparation

Install a recent GCC (≥7.3, minimum 4.8.5). Download links: http://ftp.gnu.org/gnu/gcc/gcc-7.3.0/ and installation guide https://gcc.gnu.org/install/ .

Common Porting Issues and Fixes

1.1 -m64 Compilation Option

Problem: gcc: error: unrecognized command line option ‘-m64’ on ARM64.

Cause: -m64 is an x86‑64 option.

Solution: Use -mabi=lp64 for ARM64.

1.2 char Signedness

Problem:

warning: comparison is always false due to limited range of data type

.

Cause: char is signed on x86 but unsigned on ARM64.

Solution: Add -fsigned-char to the compile flags.

2.1 Assembly Instruction Rewrite

ARM assembly differs from x86; all inline assembly must be rewritten for ARM64.

Example (x86):

Example (ARM64 using GCC built‑ins):

2.2 Replace x86 CRC32 Instructions

Problem: Unknown mnemonic crc32q on ARM64.

Solution: Use ARM64 CRC32 instructions ( crc32cb, crc32ch, crc32cw, crc32cx) and add -mcpu=generic+crc to the compile options.

Example (x86):

Example (ARM64):

2.3 Replace x86 bswap with ARM64 rev

Problem: Error: unknown mnemonic ‘bswap’.

Solution: Use rev on ARM64.

Illustration (x86):

Illustration (ARM64):

2.4 Replace x86 rep with ARM64 rept

Problem: unknown mnemonic ‘rep’.

Solution: Replace with rept as shown in the before/after images.

Before:

After:

2.5 Fast Porting of Inline SSE/SSE2

Use the open‑source sse2neon library ( https://github.com/open-estuary/sse2neon.git ) and follow the three‑step integration procedure described.

2.6 Weak Memory Ordering

ARM64’s weak memory model can cause unexpected results in lock‑free code; ensure proper memory barriers are used.

2.7 Atomic Operations on Misaligned Struct Members

ARM64 requires proper alignment for atomic instructions; remove or adjust #pragma pack usage that forces misalignment.

2.8 Hard‑coded Core Count

Search for hard‑coded CPU core numbers (e.g., in sched_setaffinity) and replace them with dynamic queries such as sysconf(_SC_NPROCESSORS_CONF).

2.9 Double‑to‑Integer Conversion Differences

Floating‑point to integer conversion behaves differently on x86 and TaiShan; follow the provided conversion tables to handle overflow and underflow consistently.

Compiler Optimization Options

4.1 GCC Floating‑Point Optimization

Using -O2 or higher enables fused multiply‑add ( fmadd) which can cause up to 16‑digit precision differences between x86 and ARM64.

Solution: Add -ffp-contract=off to disable this optimization.

4.2 Targeting Kunpeng Architecture

Add -march=armv8-a to generate code optimized for Kunpeng processors.

4.3 Tuning for Kunpeng Pipeline

For GCC ≥9.1, add -mtune=tsv110 to exploit the Kunpeng pipeline.