[cfe-dev] Variable declaration on custom target

[Lorenzo De Carli]

That fixed it, thanks! And sorry for being inaccurate, I haven't been
working with LLVM in a long time and I am not very familiar with the
assembly anymore.

Lorenzo

On Wed, Nov 10, 2010 at 3:09 PM, Eli Friedman <eli.friedman at gmail.com> wrote:
> On Wed, Nov 10, 2010 at 12:18 PM, Lorenzo De Carli <lorenzo at cs.wisc.edu> wrote:
>> Hi list,
>>
>> I am working on a custom 16-bit target for Clang. I have defined an
>> appropriate TargetInfo object in Targets.cpp, and I am able to get
>> 16-bit LLVM assembly from Clang. However, I noticed that variables
>> declared in a function body are treated differently when compiling for
>> X86 and for my architecture. In particular, on X86 such variables are
>> treated as virtual registers while on the custom target they are
>> declared as auto-variables. For example, if I compile the following
>> code:
>>
>> int myfunc(int param) {
>>  int val;
>>  int result;
>>
>>  val = 5;
>>  result = val * param;
>>
>>  return result;
>> }
>>
>> on X86 I get:
>>
>> ; ModuleID = 'myfunc.c'
>> target datalayout =
>> "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"
>> target triple = "x86_64-unknown-linux-gnu"
>>
>> define i32 @myfunc(i32 %param) nounwind {
>>  %1 = alloca i32, align 4
>>  %val = alloca i32, align 4
>>  %result = alloca i32, align 4
>>  store i32 %param, i32* %1, align 4
>>  store i32 5, i32* %val, align 4
>>  %2 = load i32* %val, align 4
>>  %3 = load i32* %1, align 4
>>  %4 = mul nsw i32 %2, %3
>>  store i32 %4, i32* %result, align 4
>>  %5 = load i32* %result, align 4
>>  ret i32 %5
>> }
>>
>> while on my custom target I get:
>>
>> target datalayout =
>> "e-p:16:16:16-i16:16:16-f16:16:16-n16:16:16-v16:16:16-a16:16:16"
>> target triple = "myarch-myvendor-myos"
>>
>> @myfunc.auto.val = internal global i16 0, align 2
>> @myfunc.auto.result = internal global i16 0, align 2
>>
>> define i16 @myfunc(i16 %param) nounwind {
>>  %1 = alloca i16, align 2
>>  store i16 %param, i16* %1, align 2
>>  store i16 5, i16* @myfunc.auto.val, align 2
>>  %2 = load i16* @myfunc.auto.val, align 2
>>  %3 = load i16* %1, align 2
>>  %4 = mul nsw i16 %2, %3
>>  store i16 %4, i16* @myfunc.auto.result, align 2
>>  %5 = load i16* @myfunc.auto.result, align 2
>>  ret i16 %5
>> }
>>
>> Could someone explain why this is happening? (I already asked a
>> similar question on LLVM-dev, but Eli recommended
>> me to re-post on cfe-dev).
>
> Oh, that looks like you copy-pasted the
> "useGlobalsForAutomaticVariables()" definition from the PIC16 target.
> You're not really describing the problem accurately though; they're
> globals, not virtual registers.
>
> -Eli
>