(original) (raw)

We encounter a inline bug in generating call, hard to fix.

I want to ask some questions here.

1^st Does the IR generate from FE is wrong ?

PLS refer t.ll line3, it use “\*m” constrain for call operand “void (...)\* @sincos”. I think the right IR should be (“m” + “void (...)\* @sincos”) or (“\*m” + “void (...)\*\* @sincos\_ptr”)

related patch in FE https://reviews.llvm.org/D107523

2^nd How we fix it in Back End ?

I find it is hard to fix in BE, because the current arch of inline\_asm let us hard to match the variable to its instruction.

I create a unbeautiful patch at BE too https://reviews.llvm.org/D108024

The following code will generate one more time load for “sincos” with cmd “clang -fasm-blocks t.c -fpic -S -emit-llvm” + “llc t.ll”

t.c:

1 extern void sincos ();

2 void foo (){

3 \_\_asm{

4 call sincos

5 ret };

6 }

t.ll:

1 define void @foo() #0 {

2 entry:

3 call void asm sideeffect inteldialect "call qword ptr ${0:P}\\0A\\09ret", "\*m,\~{dirflag},\~{fpsr},\~{flags}"(void (...)\* @sincos) #2, !srcloc !5

4 ret void

5 }

t.s:

1 movq sincos@GOTPCREL(%rip), %rax // now the rax has been the function address “sincos”

2 #APP

3 callq \*(%rax) // the call should directly read the %rax, should not “load” it again, it should be “callq \*%rax” or better code “call sincos@plt”

4 retq

5 #NO\_APP

BR!

Thank you!

Xiang