avr-llvm-devel

[avr-llvm-devel] Windows build

[John M. <ato...@gm...>]

I had a pain free build using Visual Studio C++ 2008 express edition and
CMake on Windows XP.

C:\_projects_\build_llvm>cmake -G "Visual Studio 9 2008"
-DLLVM_TARGETS_TO_BUILD="AVR" -DCMAKE_BUILD_TYPE=Debug -DLLVM_BUILD_TOOLS=ON
-DLLVM_BUILD_EXAMPLES=OFF -DLLVM_ENABLE_THREADS=OFF ..\llvm

~~~~~

C:\_projects_\build_llvm\bin\Debug>llc.exe --version
Low Level Virtual Machine (http://llvm.org/):
  llvm version 2.7svn
  DEBUG build with assertions.
  Built Dec  4 2009 (20:35:54).
  Host: i686-pc-win32
  Host CPU: (unknown)
  Registered Targets:
    avr - AVR [experimental]
C:\_projects_\build_llvm\bin\Debug>llc.exe -O0 -filetype=asm -mcpu=atmega128
reti8.ll -o -
        .file   "reti8.ll"

        .text
        .align  16
        .globl  func
        .type   func,@function
func:
# BB#0:
        ldi r24, #1
        mov r24, r24
        ret
        .size   func, .-func

Re: [avr-llvm-devel] Windows build

[Weddington, E. <Eri...@at...>]

 

> -----Original Message-----
> From: John Myers [mailto:ato...@gm...] 
> Sent: Friday, December 04, 2009 10:18 PM
> To: avr...@li...
> Subject: [avr-llvm-devel] Windows build
> 
>  
> I had a pain free build using Visual Studio C++ 2008 express 
> edition and CMake on Windows XP.

Congratulations!
  
> C:\_projects_\build_llvm\bin\Debug>llc.exe -O0 -filetype=asm 
> -mcpu=atmega128 reti8.ll -o -
>         .file   "reti8.ll"
> 
>         .text
>         .align  16
>         .globl  func

Can we change the directive to be the full spelling?: .global

>         .type   func,@function
> func:
> # BB#0:
>         ldi r24, #1
>         mov r24, r24

What's up with the redundant mov? Or will future optimization take care of removing this?

>         ret
>         .size   func, .-func
>  
> 

Re: [avr-llvm-devel] Windows build

[John M. <ato...@gm...>]

On Fri, Dec 4, 2009 at 10:23 PM, Weddington, Eric <Eri...@at...
> wrote:

>
>
> > -----Original Message-----
> > From: John Myers [mailto:ato...@gm...]
> > Sent: Friday, December 04, 2009 10:18 PM
> > To: avr...@li...
> > Subject: [avr-llvm-devel] Windows build
> >
> >
> > I had a pain free build using Visual Studio C++ 2008 express
> > edition and CMake on Windows XP.
>
> Congratulations!
>
> > C:\_projects_\build_llvm\bin\Debug>llc.exe -O0 -filetype=asm
> > -mcpu=atmega128 reti8.ll -o -
> >         .file   "reti8.ll"
> >
> >         .text
> >         .align  16
> >         .globl  func
>
> Can we change the directive to be the full spelling?: .global
>
>
Yeah, that's an easy change (it's in AsmPrinter\AVRAsmPrinter.cpp).


> >         .type   func,@function
> > func:
> > # BB#0:
> >         ldi r24, #1
> >         mov r24, r24
>
> What's up with the redundant mov? Or will future optimization take care of
> removing this?
>
> >         ret
> >         .size   func, .-func
> >
> >
>
Good question, I haven't looked into (with any detail) how LLVM IR is
converted to native assembly.
I will probable spend some time later this weekend checking that out.

Re: [avr-llvm-devel] Windows build

[John M. <ato...@gm...>]

On Fri, Dec 4, 2009 at 10:23 PM, Weddington, Eric <Eri...@at...
> wrote:
>         .type   func,@function
> func:
> # BB#0:
>         ldi r24, #1
>         mov r24, r24

What's up with the redundant mov? Or will future optimization take care of
> removing this?
>
> >         ret
> >         .size   func, .-func
> >
> >
>
The redudant mov is gone now. So it looks like whatever they changed  in the
LLVM trunk fixed it.

C:\_projects_\build_llvm\bin\Debug>llc --version
Low Level Virtual Machine (http://llvm.org/):
  llvm version 2.7svn
  DEBUG build with assertions.
  Built Dec  7 2009 (21:57:15).
  Host: i686-pc-win32
  Host CPU: (unknown)
  Registered Targets:
    avr - AVR [experimental]
C:\_projects_\build_llvm\bin\Debug>llc.exe -O0 -o - -filetype=asm
-mcpu=atmega128 reti8.ll
        .file   "reti8.ll"

        .text
        .align  16
        .global func
        .type   func,@function
func:
# BB#0:
        ldi r24, #1
        ret
        .size   func, .-func

Re: [avr-llvm-devel] Windows build

[Weddington, E. <Eri...@at...>]

 

> -----Original Message-----
> From: John Myers [mailto:ato...@gm...] 
> Sent: Tuesday, December 08, 2009 6:46 PM
> To: Weddington, Eric
> Cc: avr...@li...
> Subject: Re: [avr-llvm-devel] Windows build
> 
> 
> 
> The redudant mov is gone now. So it looks like whatever they 
> changed  in the LLVM trunk fixed it.

That's great!

Question: Is the assembler output of llvm supposed to closely match the assembler output of avr-gcc? Is that what we're aiming for?
  
> C:\_projects_\build_llvm\bin\Debug>llc.exe -O0 -o - 
> -filetype=asm -mcpu=atmega128 reti8.ll
>         .file   "reti8.ll"
> 
>         .text
>         .align  16
>         .global func
>         .type   func,@function
> func:
> # BB#0:
>         ldi r24, #1
>         ret
>         .size   func, .-func

If so, then we need to drop the .align directive above. This is the equivalent from avr-gcc:

	.file	"test.c"
__SREG__ = 0x3f
__SP_H__ = 0x3e
__SP_L__ = 0x3d
__CCP__  = 0x34
__tmp_reg__ = 0
__zero_reg__ = 1
	.text
.global	func
	.type	func, @function
func:
/* prologue: function */
/* frame size = 0 */
	ldi r24,lo8(1)
/* epilogue start */
	ret
	.size	func, .-func

Re: [avr-llvm-devel] Windows build

[John M. <ato...@gm...>]

On Tue, Dec 8, 2009 at 7:04 PM, Weddington, Eric
<Eri...@at...>wrote:

>
>
> > -----Original Message-----
> > From: John Myers [mailto:ato...@gm...]
> > Sent: Tuesday, December 08, 2009 6:46 PM
> > To: Weddington, Eric
> > Cc: avr...@li...
> > Subject: Re: [avr-llvm-devel] Windows build
> >
> >
> >
> > The redudant mov is gone now. So it looks like whatever they
> > changed  in the LLVM trunk fixed it.
>
> That's great!
>
> Question: Is the assembler output of llvm supposed to closely match the
> assembler output of avr-gcc? Is that what we're aiming for?
>

Yeah, it's suppose to be an assembly printer for gas / avr-gcc. Some targets
actually produce multiple asm types for different assembler brands.

>
> > C:\_projects_\build_llvm\bin\Debug>llc.exe -O0 -o -
> > -filetype=asm -mcpu=atmega128 reti8.ll
> >         .file   "reti8.ll"
> >
> >         .text
> >         .align  16
> >         .global func
> >         .type   func,@function
> > func:
> > # BB#0:
> >         ldi r24, #1
> >         ret
> >         .size   func, .-func
>
> If so, then we need to drop the .align directive above. This is the
> equivalent from avr-gcc:
>

I'll try to find were the .align directive is printed to remove it. Will it
mess things up if it's there?


>
>        .file   "test.c"
> __SREG__ = 0x3f
> __SP_H__ = 0x3e
> __SP_L__ = 0x3d
> __CCP__  = 0x34
> __tmp_reg__ = 0
> __zero_reg__ = 1
>        .text
> .global func
>        .type   func, @function
> func:
> /* prologue: function */
> /* frame size = 0 */
>        ldi r24,lo8(1)
> /* epilogue start */
>        ret
>        .size   func, .-func
>

Re: [avr-llvm-devel] Windows build

[Weddington, E. <Eri...@at...>]

 

> -----Original Message-----
> From: John Myers [mailto:ato...@gm...] 
> Sent: Tuesday, December 08, 2009 8:30 PM
> To: Weddington, Eric
> Cc: avr...@li...
> Subject: Re: [avr-llvm-devel] Windows build
> 
> 
>  
> I'll try to find were the .align directive is printed to 
> remove it. Will it mess things up if it's there?
>  

It might. I'm not sure if .align 16 is correct for the AVR target. Thanks for looking into this.

AFAIK, the AVR default linker script for GNU ld take care of aligning the functions on a word (16-bit) boundary.

Re: [avr-llvm-devel] Windows build

[Josef E. <za...@za...>]

John Myers wrote:
>     What's up with the redundant mov? Or will future optimization take
>     care of removing this?
> 
>     >         ret
>     >         .size   func, .-func
>     >
>     >
> 
> The redudant mov is gone now. So it looks like whatever they changed  in
> the LLVM trunk fixed it.

Nope. You fixed it in r53 ;)! The isMoveInstr() allows llvm to detect
redundant MOVs.
BTW: great work!

Josef

Re: [avr-llvm-devel] Windows build

[John M. <ato...@gm...>]

LOL, I must have had a bad build after I added that because original I
didn't see any changes.

On Wed, Dec 9, 2009 at 7:08 AM, Josef Eisl <za...@za...> wrote:

> John Myers wrote:
> >     What's up with the redundant mov? Or will future optimization take
> >     care of removing this?
> >
> >     >         ret
> >     >         .size   func, .-func
> >     >
> >     >
> >
> > The redudant mov is gone now. So it looks like whatever they changed  in
> > the LLVM trunk fixed it.
>
> Nope. You fixed it in r53 ;)! The isMoveInstr() allows llvm to detect
> redundant MOVs.
> BTW: great work!
>
> Josef
>