llvm dev - Dec 2012 - [LLVMdev] Which transform passes to apply?

Hello, I'm a new LLVM user working on a C++ EDSL for image processing. I
have
a function, which after applying createInstructionCombiningPass() and
createDeadCodeEliminationPass() looks like:

define void @jitcv_sum_64sf1001(%Matrix* %src, %Matrix* %dst, i32 %len) {
entry:
  br label %loop_i

loop_i:                                           ; preds = %loop_i_end,
%entry
  %i = phi i32 [ 0, %entry ], [ %increment_i, %loop_i_end ]
  %0 = getelementptr inbounds %Matrix* %dst, i32 0, i32 2
  %dst_columns = load i32* %0
  *%dst_yRem = urem i32 %i, %dst_columns
  %dst_y = urem i32 %i, %dst_columns
  %1 = sub i32 %i, %dst_y
  %2 = add i32 %1, %dst_yRem*
  %3 = getelementptr inbounds %Matrix* %src, i32 0, i32 0
  %4 = load i8** %3
  %src_data = bitcast i8* %4 to double*
  %5 = getelementptr double* %src_data, i32 %2
  %6 = load double* %5
  *%accumulate = fadd double %6, 0.000000e+00*
  %7 = getelementptr inbounds %Matrix* %dst, i32 0, i32 0
  %8 = load i8** %7
  %dst_data = bitcast i8* %8 to double*
  %9 = getelementptr double* %dst_data, i32 %i
  store double %accumulate, double* %9
  br label %loop_i_end

loop_i_end:                                       ; preds = %loop_i
  %increment_i = add i32 %i, 1
  %loop_i_test = icmp eq i32 %increment_i, %len
  br i1 %loop_i_test, label %loop_i_exit, label %loop_i

loop_i_exit:                                      ; preds = %loop_i_end
  ret void
}

My question is which optimization pass(es) are needed to simplify the
instructions in bold. I've tried running the same passes again and also
tried createInstructionSimplifierPass() with no luck.

Many Thanks,

Josh



--
View this message in context:
http://llvm.1065342.n5.nabble.com/Which-transform-passes-to-apply-tp52111.html
Sent from the LLVM - Dev mailing list archive at Nabble.com.

Hi Josh,

On 03/12/12 02:58, Josh Klontz wrote:
> Hello, I'm a new LLVM user working on a C++ EDSL for image processing.
I have
> a function, which after applying createInstructionCombiningPass() and
> createDeadCodeEliminationPass() looks like:
>
...
>    *%dst_yRem = urem i32 %i, %dst_columns
>    %dst_y = urem i32 %i, %dst_columns
>    %1 = sub i32 %i, %dst_y
>    %2 = add i32 %1, %dst_yRem*
...
> My question is which optimization pass(es) are needed to simplify the
> instructions in bold. I've tried running the same passes again and also
> tried createInstructionSimplifierPass() with no luck.
GVN or EarlyCSE.

Ciao, Duncan.

Thanks Duncan, GVN/EarlyCSE worked as suggested. Any pointers on how to
optimize out:
%accumulate = fadd double %6, 0.000000e+00

Using the 3.1 release and the C++ API, I can't figure out how
FPMathOperator, TargetOptions, nor IRBuilder::SetDefaultFPMathTag work. I
also don't see any floating point math transformation passes. I did see
IRBuilder::SetFastMathFlags, do I need to update to 3.2 and use this call?




--
View this message in context:
http://llvm.1065342.n5.nabble.com/Which-transform-passes-to-apply-tp52111p52211.html
Sent from the LLVM - Dev mailing list archive at Nabble.com.