Code block that allows the use of architecture-specific instructions.
The
Asm block is used to insert specific machine-code instructions in a program in order to perform operations that cannot be carried out using the features of the language or to hand-optimize performance-sensitive sections of code.
The current FreeBASIC compiler currently only produces code for Intel
80x86-based machines; however, in the future, the compiler might be ported to a platform which does not support the same instruction set. Therefore,
Asm blocks should only be used when necessary, and a FreeBASIC-only alternative should be provided if possible.
The return value of a function may be set by using the
Function keyword within brackets as shown in the example below.
Asm block comments have the same syntax as usual FreeBASIC
Comments - use FreeBASIC-like "
' " comments, not "
; " as usual in ASM.
x86 Specific:
Syntax
The syntax of the inline assembler is a simplified form of Intel syntax. Intel syntax is used by the majority of x86 assemblers, such as MASM, TASM, NASM, YASM and FASM. In general, the destination of an instruction is placed first, followed by the source. Variables and functions defined by a program may be referenced in an Asm block. The assembler used by FreeBASIC is GAS, using the
.intel_syntax noprefix directive, and ASM blocks are passed through unmodified, except for the substitution of local variable names for stack frame references, and commenting removal.
Instruction syntax is mostly the same as FASM uses, one important difference is that GAS requires size settings to be followed by the word "ptr".
' Assuming "blah" is a FB global or local UINTEGER variable
mov eax, [blah] ' OK: size is apparent from eax
inc [blah] ' Not OK: size is not given
inc dword [blah] ' Not OK: size given, but still not accepted by GAS
inc dword Ptr [blah] ' OK: "ptr" is needed by GAS here
Register Preservation
When an ASM block is opened, the registers ebx, esi, and edi are pushed to the stack, when the block is closed, these registers are popped back from the stack. This is because these registers are required to be preserved by most or all OS's using the x86 CPU. You can therefore use these registers without explicitly preserving them yourself. You should not change esp and ebp, since they are usually used to address local variables.
Register Names
The names of the registers for the x86 architecture are written as follows in an Asm block:
- 4-byte integer registers: eax, ebx, ecx, edx, ebp, esp, edi, esi
- 2-byte integer registers: ax, bx, cx, dx, bp, sp, di, si (low words of 4-byte e- registers)
- 1-byte integer registers: al, ah, bl, bh, cl, ch, dl, dh (low and high bytes of 2-byte -x registers)
- Floating-point registers: st(0), st(1), st(2), st(3), st(4), st(5), st(6), st(7)
- MMX registers (aliased onto floating-point registers): mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7
- SSE registers: xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7
Instruction Set
See these external references:
Unsafe instructions
Note that the FreeBASIC compiler produces 32-bit protected-mode code for the x86 which usually runs in an unprivileged user level; therefore, privileged and sensitive instructions will assemble fine, but possibly won't work correctly or cause a runtime "General Protection Fault", "Illegal instruction", or SIGILL error. The following are the privileged and sensitive instructions as of the Intel Pentium 4 and Xeon:
- cli *1
- clts
- hlt
- in *1
- ins *1
- int *1
- into *1
- invd
- invlpg
- lgdt
- lidt
- lldt
- lmsw
- ltr
- mov to/from CRn, DRn, TRn
- out *1
- outs *1
- rdmsr
- rdpmc *2
- rdtsc *2
- sti *1
- str
- wbinvd
- wrmsr
- all SSE2 and higher instructions *2
*1: sensitive to IOPL, fine in DOS
*2: sensitive to permission bits in CR4, see below
The privileged instructions will work "correctly" in DOS when running on a Ring 0 DPMI kernel, like the (non-default) Ring 0 version of CWSDPMI, WDOSX or D3X, nevertheless most of them are not really useful and dangerous when executed from DPMI code. RDTSC (Read Time Stamp Counter) has been shown to be allowed by most, or all OS'es.
However the usefulness of RDTSC has been diminished with the advent of multi-core and hibernating CPUs. SSE2 and higher instructions are disabled "by default" after CPU initialization, Windows and Linux usually do enable them, in DOS it is business of the DPMI host: HDPMI32 will enable them, CWSDPMI won't. The INT instruction is usable in the DOS version/target only, note that it works slightly differently from real mode DOS, see also FaqDOS.
The segment registers (
cs,
ds,
es,
fs,
gs) should not be changed from an
Asm block, except in certain cases with the DOS port (note that they do NOT work the same way as in real-mode DOS, see also FaqDOS). The operating system or DPMI host is responsible for memory management; the meaning of segments (selectors) in protected mode is very different from real-mode memory addressing.
Note that those "unsafe" instructions are not guaranteed to raise a "visible" crash even when ran with insufficient privilege - the OS or DPMI host can decide to "emulate" them, either functionally (reading from some CRx works under HDPMI32), or "dummy" (nothing happens, instruction will pass silently, like a NOP).