I'm trying to understand an LLVM CodeGen/Generic testcase so I can get it passing for a backend for a new processor.
This is the testcase (llvm/test/CodeGen/Generic/2007-04-08-MultipleFrameIndices.ll):
; RUN: llc -no-integrated-as < %s
; XFAIL: sparc-sun-solaris2
; PR1308
; PR1557
define i32 @stuff(i32, ...) {
%foo = alloca i8*
%bar = alloca i32*
%A = call i32 asm sideeffect "inline asm $0 $2 $3 $4", "=r,0,i,m,m"( i32 0, i32 1, i8** %foo, i32** %bar )
ret i32 %A
}
I'm specifically wondering about this part:
%A = call i32 asm sideeffect "inline asm $0 $2 $3 $4", "=r,0,i,m,m"( i32 0, i32 1, i8** %foo, i32** %bar )
What exactly should that inline asm accomplish?
When I run it through llc as indicated in the comment, I see this on the output (for x86_64):
.text
.file "<stdin>"
.globl stuff
.align 16, 0x90
.type stuff,@function
stuff: # @stuff
.cfi_startproc
# BB#0:
leaq -8(%rsp), %rax
movq %rax, -24(%rsp)
leaq -16(%rsp), %rax
movq %rax, -32(%rsp)
xorl %eax, %eax
#APP
inline asm %eax $1 -24(%rsp) -32(%rsp)
#NO_APP
retq
.Ltmp0:
.size stuff, .Ltmp0-stuff
.cfi_endproc
.section ".note.GNU-stack","",@progbits
It seems to have resulting in the following line in the output asm code:
inline asm %eax $1 -24(%rsp) -32(%rsp)
...but I'm not entirely sure how that was generated from the input. Where does the %eax come from, for example? What does this generated line mean/do?
答案 0 :(得分:1)
I am guessing that it came to be because input operand $1
is the same place as output operand $0
, and must be in a register. It's also specified as a constant zero. The compiler picked %eax
for $0
and thus $1
too, and zeroed it beforehand by the xorl %eax, %eax
.
The generated code is invalid, I assume it is not meant to be assembled, presumably the test engine just checks whether the output assembly is the expected one.