llvm dev - Aug 2018 - Risc-v Assembly printer function order

Hello,

I am working on the assembly printer for RISC-V, more specifically on 
the AsmPrinter class.
I altered the RISCV Backend to print C code instead of Assembly, 
interpreted by libraries... (but that's not important)
My problem is that, for my application to work, I need to treat my 
functions in the order they are in the original C file.
I discovered that these functions are not treated in this order.

My question is : Is there a way to force llvm to treat the functions in 
the order given in the original C file ? and how?

I treat each function in the EmitFunctionBody funtion of the AsmPrinter 
class, and the function is passed through the MachineFunction passed 
trhough the runOnMachineFunction function. I didn't find where this last 
function is called.
So I wish maybe to alter the way this function is called for each 
Function to call it in the "right" order...

Thanks,
Best regards,
IG

On 8/7/2018 12:54 AM, 'Irenee GROZ IG255340' via llvm-dev
wrote:
> Hello,
>
> I am working on the assembly printer for RISC-V, more specifically on 
> the AsmPrinter class.
> I altered the RISCV Backend to print C code instead of Assembly, 
> interpreted by libraries... (but that's not important)
> My problem is that, for my application to work, I need to treat my 
> functions in the order they are in the original C file.
> I discovered that these functions are not treated in this order.
>
> My question is : Is there a way to force llvm to treat the functions 
> in the order given in the original C file ? and how?
The LLVM backend should generally print functions in the same order they 
appear in the IR.  This is an implementation detail, not a guarantee, 
but there currently aren't any backend passes that rearrange functions.

That said, clang doesn't really try to emit functions in source order: 
it will delay emitting functions for various reasons.  If that's truly 
necessary for your use-case, you'll probably have to modify clang 
somehow.  Or maybe -femit-all-decls is close enough if your input is pure C?

-Eli

-- 
Employee of Qualcomm Innovation Center, Inc.
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux
Foundation Collaborative Project

> -----Original Message-----
> From: llvm-dev [mailto:llvm-dev-bounces at lists.llvm.org] On Behalf Of
> Friedman, Eli via llvm-dev
> Sent: Tuesday, August 07, 2018 1:58 PM
> To: 'Irenee GROZ IG255340'; llvm-dev at lists.llvm.org
> Subject: Re: [llvm-dev] Risc-v Assembly printer function order
> 
> On 8/7/2018 12:54 AM, 'Irenee GROZ IG255340' via llvm-dev wrote:
> > Hello,
> >
> > I am working on the assembly printer for RISC-V, more specifically on
> > the AsmPrinter class.
> > I altered the RISCV Backend to print C code instead of Assembly,
> > interpreted by libraries... (but that's not important)
> > My problem is that, for my application to work, I need to treat my
> > functions in the order they are in the original C file.
> > I discovered that these functions are not treated in this order.
> >
> > My question is : Is there a way to force llvm to treat the functions
> > in the order given in the original C file ? and how?
> 
> The LLVM backend should generally print functions in the same order they
> appear in the IR.  This is an implementation detail, not a guarantee,
> but there currently aren't any backend passes that rearrange functions.
You could probably write a Module pass that sorted the Functions by source
location; that's the only solution that comes to mind.  You'd need to
define the ordering across files (e.g., the source location of a function
defined in a header; this happens a lot in C++, not so much in C).
--paulr
> 
> That said, clang doesn't really try to emit functions in source order:
> it will delay emitting functions for various reasons.  If that's truly
> necessary for your use-case, you'll probably have to modify clang
> somehow.  Or maybe -femit-all-decls is close enough if your input is pure
> C?
> 
> -Eli
> 
> --
> Employee of Qualcomm Innovation Center, Inc.
> Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux
> Foundation Collaborative Project
> 
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev