On 04/08/2019 09:09, Bu Le wrote:
Hi, May I ask why aarch64 do not implement fPIC on large/medium code model? Comparing with x86, where large code model PIC is implemented by "lea + movabs + add" instructions, aarch64 seems also has these instructions in its ABI (adr + add). However, under the large code model, aarch64 use LDR instruction to do the absolute relocation, which makes it not able to be dynamic linked. I hit this problem when I am recently working on a open source project CESM, which can be compiled successfully on x86 with gcc -mcmodel=medium. When I recompile it on a aarch64 machine, I hit following error: gptl.c:(.text+0x14): relocation truncated to fit: R_AARCH64_ADR_PREL_PG_HI21 against `.bss' gptl.c:(.text+0x5c): relocation truncated to fit: R_AARCH64_ADR_PREL_PG_HI21 against `.bss' gptl.c:(.text+0x1ac): relocation truncated to fit: R_AARCH64_ADR_PREL_PG_HI21 against `.bss' ... The program can be compiled successfully with mcmodel=large. But the compiled executables cannot run because all the other depends in the open source project are PIC executables. Thanks, Bruce
I don't think there are any technical reasons why this hasn't been done yet. It's just a matter of priorities.
R.
