Inline assembler
C and C++ allows adding assembler inside functions, so that some very low-level operations can be done. This is sometimes used for optimizations, but sometimes also to do things that C and C++ can not do otherwise. Storm also provides this functionality, but instead of being a core part of the compiler, it is simply implemented as a library.
The inline assembler is located in lang:asm, and is therefore implemented as any other language in
the compiler. This means that .asm-files can be created and used directly with the assembler (not
implemented yet). The assembler language also works as a language extension and can therefore be
embedded inside other languages.
Separate files
This is not supported yet. Creating .asm-files will result in the compiler issuing a warning about
a missing reader class for asm.
Inline
To use the assembler inside of Basic Storm, simply use lang:asm, and you can use the asm{} block
to write inline assembly code. The asm block does not return anything, and when evaluated it will
simply execute the instructions written inside the block. Any local variables within the current
scope are visible inside the asm block as well, and can be accessed as if they were regular
registers.
The following function uses inline assembler to add two integer variables:
use lang:asm;
Int asmAdd(Int a, Int b) {
Int r;
asm {
mov eax, a;
add eax, b;
mov r, eax;
}
r;
}
Semantics
The assembler does not generate machine code directly, instead it generates an intermediate representation that is similar to bytecode in other languages. This bytecode is heavily inspired from the X86 instruction set, but has some differences which aims to make the bytecode portable. Bytecode is never interpreted, it is always translated into machine code before it is executed.
In the bytecode, there are four different data sizes, 1 byte (Byte), 4 bytes (Int), 8 bytes
(Long) and pointer-size. Trying to mix different data types in the same instruction will fail,
except for some instructions (for example when taking the address of something or casting). Function
calls also have a special form.
It is worth to notice that some of the limitations in X86 assembly does not exist in the bytecode. For example, it is not possible to have both a source and a destination operand in memory in raw X86 assembly, but the code generation solves this by emitting extra instructions as necessary. The code generation always takes care of preserving any registers when doing these kind of things, which means that it is entirely transparent.