交叉编译与目标机直接编译时GCC ARM选项差异及选型咨询

Great question! Let's break down the key differences between GCC ARM options for cross-compiling vs. native compiling on your Cubietruck (ARM Cortex-A7 dual-core, Debian Jessie 8.0 Armbian, armhf) platform, using the hardware info you provided.

When compiling directly on your Cubietruck, the local armhf GCC is already pre-configured to match your platform's hardware and OS, so you'll rarely need to manually specify most low-level ARM options:

• ABI & Base Architecture: The native armhf GCC defaults to the aapcs-linux ABI (required for armhf compatibility) and targets ARMv7-A with support for your CPU's features (swp, half, thumb, fastmult, vfp, edsp, neon, vfpv3, tls, vfpv4, idiva, idivt) out of the box. You don't need to add -mabi=aapcs-linux explicitly.
• CPU/Tune Options: While the default armhf GCC is optimized for general ARMv7-A cores, you can optionally add -mcpu=cortex-a7 and -mtune=cortex-a7 to enable microarchitecture-specific optimizations (like leveraging Cortex-A7's cache hierarchy and dual-core scheduling) for better performance. Without these, the compiler will generate code compatible with all ARMv7-A cores, but not optimized specifically for yours.
• Instruction Set Mode: The native GCC typically defaults to Thumb mode (-mthumb) for smaller code size, which is ideal for embedded systems. You only need to specify -marm if you have a specific reason to force ARM instruction set usage.

Cross-compilers are designed to support multiple ARM targets, so you must explicitly define your target platform's parameters to avoid generating incompatible code. Your initial choice of options is spot-on, and here's why they're necessary:

• -mabi=aapcs-linux: Ensures the compiler generates code compatible with the armhf ABI, matching what the native target uses. Omitting this could lead to ABI mismatches (e.g., incorrect function calling conventions) that break runtime execution.
• -mcpu=cortex-a7: Tells the compiler to generate instructions specific to the Cortex-A7 CPU, including support for its advanced features like VFPv4 and NEON. Without this, the cross-compiler might default to a lower ARM version (e.g., ARMv5) which won't utilize your hardware's capabilities.
• -mtune=cortex-a7: Optimizes instruction scheduling to align with Cortex-A7's microarchitecture, maximizing execution efficiency on your specific CPU.
• -marm/-mthumb: Lets you choose between ARM and Thumb instruction sets. Like native compiling, -mthumb is recommended for smaller binaries unless you have a use case requiring ARM instructions.
• Additional Cross-Compile Requirement: Unlike native compiling, you'll also need to set the --sysroot flag (or configure the cross-compiler with the target's root filesystem) to ensure the linker uses the correct libraries from your Cubietruck's Debian Jessie system, not the host's libraries.

内容的提问来源于stack exchange，提问作者dk13