In WebAssembly, the align attribute is used to specify the alignment of data when accessing linear memory. Alignment is a concept in computer memory access, referring to how the starting address of data is positioned relative to a value (typically a power of two). Correct alignment can improve memory access efficiency, as many processors are optimized for aligned memory accesses.
On certain processor architectures, misalignment (where the starting address is not an integer multiple of the data size) can lead to performance penalties or even hardware exceptions. Therefore, specifying the correct alignment is crucial for ensuring code performance and correctness.
For example, if you are reading a 32-bit integer (4 bytes), on many processors, the most efficient approach is to read from an address that is a multiple of 4 (i.e., its alignment is 4). In WebAssembly's text format, this can be specified using the align attribute:
(i32.load align=4 (i32.const 0))
This code indicates loading a 32-bit integer from address 0, where the address should be a multiple of 4. If the actual starting address of the integer is not a multiple of 4, the align attribute may instruct the compiler or virtual machine to make necessary adjustments to satisfy this requirement.
However, in practical use of WebAssembly, the align attribute is typically a suggestion, and the WebAssembly runtime ensures that accessing data at misaligned addresses does not cause hardware exceptions. This means that even when align=4 is specified, the WebAssembly virtual machine can still handle reading a 4-byte integer from a non-multiple-of-4 address, though this may incur performance penalties. In WebAssembly's binary format, align is actually encoded as the logarithm base 2 of the alignment, so align=4 is encoded as 2 (since 2^2 = 4).