LLVM is a low-level language (whose purpose is a compiler backend) that looks a lot like AT&T assembly, if not 10x worse. Here's an example:
define i32 @add_sub(i32 %x, i32 %y, i32 %z) {
entry:
%tmp = add i32 %x, %y
%tmp2 = sub i32 %tmp, %z
ret i32 %tmp2
}
This is roughly equilavent with the following hand-written x86 assembly:
; Body
mov eax, edi
add eax, esi
sub eax, edx
ret
LLVM llc 3.3 generates the following code (indented differently for readability):
.file "add_sub.ll"
.text
.globl add_sub
.align 16, 0x90
.type add_sub,@function
add_sub: # @add_sub
.cfi_startproc
# BB#0: # %entry
lea EAX, DWORD PTR [RDI + RSI]
sub EAX, EDX
ret
.Ltmp0:
.size add_sub, .Ltmp0-add_sub
.cfi_endproc
.section ".note.GNU-stack","",@progbits
The relevant code is this:
lea EAX, DWORD PTR [RDI + RSI]
sub EAX, EDX
ret
As you can see, LLVM has a very powerful optimizer. It is probably the closest that you're going to get.