FASM编译报错：section is not aligned enough问题排查及方案验证

Hey there! Let's break down your problem and answer your compliance question clearly.

The Original Error Explained #

You tried to use align 32 inside the .text section to get your .table aligned for AVX's vmovaps instruction, but hit a "section is not aligned enough" error. Here's why that happened:

• FASM enforces that any align directive inside a section can't exceed the section's own alignment setting. By default, the .text section in ELF64 targets typically uses a 16-byte alignment (this depends on the target system, but it's common). When you tried to force a 32-byte alignment inside a section only set up for 16-byte alignment, FASM throws an error because it can't guarantee that alignment within the section's constraints.

Is Your Modified Solution Compliant? #

Absolutely—this is not just compliant, it's the recommended approach for this scenario. Here's why:

1. Separation of code and read-only data: The .rodata section is explicitly designed for read-only constant data in ELF binaries. Putting your lookup table here follows standard ELF best practices, keeping executable code (.text) and static data separate, which is good for security, memory protection, and maintainability.
2. Proper alignment guarantee: By declaring .rodata with align 32, you ensure the entire section starts at a 32-byte aligned address. Since your .table is placed at the start of this section, its address will automatically be 32-byte aligned—perfect for vmovaps which requires 32-byte aligned memory operands.
3. FASM and ELF specification compliance: This usage aligns with both FASM's section declaration rules and the ELF64 specification, which defines .rodata as a valid read-only data section with configurable alignment.

Alternative (Less Ideal) Fix #

If you really needed to keep the table in .text (which isn't recommended), you could adjust the .text section's alignment directly:

section '.text' executable public align 32
func:
vmovaps ymm0, YWORD [.table]
xor rax, rax
ret
align 32
.table:
DQ 1024
DQ 1024
DQ 1024
DQ 1024
DQ 2048
DQ 2048
DQ 2048
DQ 2048

But again, mixing constant data with executable code isn't a best practice, so your original modified solution is far better.

内容的提问来源于stack exchange，提问作者Jason