[PATCH] D56571: [RFC prototype] Implementation of asm-goto support in clang

[Jennifer Yu via Phabricator via cfe-commits]

jyu2 marked 2 inline comments as done.
jyu2 added inline comments.


================
Comment at: lib/Sema/SemaStmtAsm.cpp:470
+    if (NS->isGCCAsmGoto() &&
+        Exprs[ConstraintIdx]->getStmtClass() == Stmt::AddrLabelExprClass)
+      break;
----------------
efriedma wrote:
> jyu2 wrote:
> > efriedma wrote:
> > > jyu2 wrote:
> > > > efriedma wrote:
> > > > > jyu2 wrote:
> > > > > > jyu2 wrote:
> > > > > > > efriedma wrote:
> > > > > > > > jyu2 wrote:
> > > > > > > > > efriedma wrote:
> > > > > > > > > > jyu2 wrote:
> > > > > > > > > > > efriedma wrote:
> > > > > > > > > > > > This looks suspicious; an AddrLabelExpr could be an input or output, e.g. `"r"(&&foo)`.
> > > > > > > > > > > Syntax for asm goto:
> > > > > > > > > > >  Syntax:
> > > > > > > > > > >    asm [volatile] goto ( AssemblerTemplate
> > > > > > > > > > >                        :
> > > > > > > > > > >                        : InputOperands
> > > > > > > > > > >                        : Clobbers
> > > > > > > > > > >                        : GotoLabels)
> > > > > > > > > > > 
> > > > > > > > > > >  Only input is allowed.  Output is not allowed
> > > > > > > > > > > 
> > > > > > > > > > That doesn't really address my point here... ignore the "or output" part of the comment.
> > > > > > > > > Sorry did not realize that.  Thank you so much for catching that.  Need to add other condition "ConstraintIdx > NS->getNumInputs() - 1", change to :
> > > > > > > > > 
> > > > > > > > > if (NS->isGCCAsmGoto() && ConstraintIdx > NS->getNumInputs() - 1 &&
> > > > > > > > >         Exprs[ConstraintIdx]->getStmtClass() == Stmt::AddrLabelExprClass)
> > > > > > > > >       break;
> > > > > > > > > 
> > > > > > > > > Is this ok with you?  Thanks
> > > > > > > > That's the right idea. But I still see a few issues at that point:
> > > > > > > > 
> > > > > > > > 1. The AddrLabelExprClass check is redundant.
> > > > > > > > 2. "NS->getNumInputs() - 1" can overflow; probably should use "ConstraintIdx >= NS->getNumInputs()".
> > > > > > > > 3. "break" exits the loop completely (so it skips validating all constraints written after the label).
> > > > > > > > 4. The code needs to verify that the user correctly specified the "l" constraint modifier.
> > > > > > > Sorry not done yet.  
> > > > > > For you comment 4:
> > > > > > 
> > > > > > The code needs to verify that the user correctly specified the "l" constraint modifier.  We already emit error like following?
> > > > > > 
> > > > > > Do you mean, we need more checking here?  Thanks. 
> > > > > > 
> > > > > > n.c:4:35: error: unknown symbolic operand name in inline assembly string
> > > > > >   asm goto ("frob %%r5, %1; jc %l[error]; mov (%2), %%r5"
> > > > > >             ~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~
> > > > > > n.c:8:15: error: use of undeclared label 'error1'
> > > > > >             : error1);
> > > > > > 
> > > > > > Test is:
> > > > > > int frob(int x)
> > > > > > {
> > > > > >   int y;
> > > > > >   asm goto ("frob %%r5, %1; jc %l[error]; mov (%2), %%r5"
> > > > > >             : /* No outputs. */
> > > > > >             : "r"(x), "r"(&y)
> > > > > >             : "memory"
> > > > > >             : error1);
> > > > > >   return y;
> > > > > > error:
> > > > > >   return -1;
> > > > > > }
> > > > > > 
> > > > > > 
> > > > > I mean, there needs to be a diagnostic for the following:
> > > > > 
> > > > > ```
> > > > > asm goto ("jne %h0"::::x);
> > > > > ```
> > > > > 
> > > > > On a related note, there should also be a diagnostic for the following somewhere:
> > > > > 
> > > > > ```
> > > > > asm ("jne %l0"::"r"(0));
> > > > > ```
> > > > Hi Eli,
> > > > 
> > > > Thanks for your review.
> > > > 
> > > > For case:
> > > > asm goto ("jne %h0"::::x);
> > > > 
> > > > Without define label x, both clang and my current implementation give error of "use of undeclared label"
> > > > 
> > > > if x is defined: gcc give error 
> > > > asm_goto>!gcc
> > > > gcc n.c
> > > > n.c: Assembler messages:
> > > > n.c:4: Error: operand type mismatch for `jne'
> > > > 
> > > > My current implementation don't emit error.  I think this is need to be done in LLVM.  Am I right here?
> > > > 
> > > > For the case:
> > > > asm ("jne %l0"::"r"(0));
> > > > 
> > > > gcc don't allow any modifier 'l' with  asm stmt but it allows with asm goto.  Is that something you are look for?  Thanks.
> > > > 
> > > > So I add code in AST/Stmt.cpp to emit error.
> > > > .....
> > > >          return diag::err_asm_invalid_escape;
> > > >       } else if (!this->isGCCAsmGoto() && EscapedChar == 'l' &&
> > > >                  isDigit(*CurPtr)) {
> > > >         DiagOffs = CurPtr-StrStart;
> > > >         return diag::err_asm_invalid_operand_number;
> > > >       }
> > > > 
> > > For the first one, I was trying with Aarch64 gcc; I guess x86 doesn't emit the same error?  `void f() { x: asm goto ("jne %i0"::::x);}` should be the same for both.
> > > 
> > > > gcc don't allow any modifier 'l' with asm stmt but it allows with asm goto. Is that something you are look for? Thanks.
> > > 
> > > We should reject any use of the "l" modifier that does not point to a label in the label list.  So we should also reject `void f(){x:asm goto ("jne %l0"::"r"(&&x)::x);}`.
> > Hi Eli,
> > 
> > Thank you so much to point this out.  I add code for emit error for use of the "l" modifier that does not point to a label in the label list.
> > 
> > 
> Please also add the `void f() { x: asm goto ("jne %i0"::::x);}` testcase.
Thanks.  Jennifer


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D56571/new/

https://reviews.llvm.org/D56571