llvm dev - Sep 2013 - [LLVMdev] Optimisation pass to move an alloca'd array to a global constant array

Hi All,
I was wondering if there is an optimisation pass that moves a stack 
allocated array, initialised with constant values, to a global constant 
array.
And if there is such a pass, what requirements are there for it to operate?
My optimised IR is below.  As you can see an array of 5 integers is 
created with alloca, then each element is stored to in turn.  It would 
be nice if this array was transformed into a global constant array.

Thanks,
     Nick



; ModuleID = 'WinterModule'
target datalayout = 
"e-p:64:64:64-S128-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f16:16:16-f32:32:32-f64:64:64-f128:128:128-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"

; Function Attrs: nounwind readnone
define i32 @"main(int)"(i32 %i, i32* nocapture %hidden) #0 {
entry:
   %"Array literal space" = alloca [5 x i32], align 4
   %0 = getelementptr [5 x i32]* %"Array literal space", i64 0, i64 0
   store i32 1, i32* %0, align 4
   %1 = getelementptr [5 x i32]* %"Array literal space", i64 0, i64 1
   store i32 2, i32* %1, align 4
   %2 = getelementptr [5 x i32]* %"Array literal space", i64 0, i64 2
   store i32 3, i32* %2, align 4
   %3 = getelementptr [5 x i32]* %"Array literal space", i64 0, i64 3
   store i32 4, i32* %3, align 4
   %4 = getelementptr [5 x i32]* %"Array literal space", i64 0, i64 4
   store i32 5, i32* %4, align 4
   %5 = icmp ult i32 %i, 5
   br i1 %5, label %in-bounds.i, label %"elem(array<int, 5>,
int).exit"

in-bounds.i:                                      ; preds = %entry
   %6 = sext i32 %i to i64
   %7 = getelementptr [5 x i32]* %"Array literal space", i64 0, i64 %6
   %8 = load i32* %7, align 4
   br label %"elem(array<int, 5>, int).exit"

"elem(array<int, 5>, int).exit":                  ; preds =
%entry,
%in-bounds.i
   %iftmp.i = phi i32 [ %8, %in-bounds.i ], [ 0, %entry ]
   ret i32 %iftmp.i
}

attributes #0 = { nounwind readnone }

On Thu, Sep 5, 2013 at 1:46 PM, Nicholas Chapman
<admin at indigorenderer.com>wrote:
> Hi All,
> I was wondering if there is an optimisation pass that moves a stack
> allocated array, initialised with constant values, to a global constant
> array.
>
No, there is no such pass at the moment.

-Eli

> And if there is such a pass, what requirements are there for it to operate?
> My optimised IR is below.  As you can see an array of 5 integers is
> created with alloca, then each element is stored to in turn.  It would be
> nice if this array was transformed into a global constant array.
>
> Thanks,
>     Nick
>
>
>
> ; ModuleID = 'WinterModule'
> target datalayout = "e-p:64:64:64-S128-i1:8:8-i8:**
> 8:8-i16:16:16-i32:32:32-i64:**64:64-f16:16:16-f32:32:32-f64:**
> 64:64-f128:128:128-v64:64:64-**v128:128:128-a0:0:64-s0:64:64-**
> f80:128:128-n8:16:32:64"
>
> ; Function Attrs: nounwind readnone
> define i32 @"main(int)"(i32 %i, i32* nocapture %hidden) #0 {
> entry:
>   %"Array literal space" = alloca [5 x i32], align 4
>   %0 = getelementptr [5 x i32]* %"Array literal space", i64 0,
i64 0
>   store i32 1, i32* %0, align 4
>   %1 = getelementptr [5 x i32]* %"Array literal space", i64 0,
i64 1
>   store i32 2, i32* %1, align 4
>   %2 = getelementptr [5 x i32]* %"Array literal space", i64 0,
i64 2
>   store i32 3, i32* %2, align 4
>   %3 = getelementptr [5 x i32]* %"Array literal space", i64 0,
i64 3
>   store i32 4, i32* %3, align 4
>   %4 = getelementptr [5 x i32]* %"Array literal space", i64 0,
i64 4
>   store i32 5, i32* %4, align 4
>   %5 = icmp ult i32 %i, 5
>   br i1 %5, label %in-bounds.i, label %"elem(array<int, 5>,
int).exit"
>
> in-bounds.i:                                      ; preds = %entry
>   %6 = sext i32 %i to i64
>   %7 = getelementptr [5 x i32]* %"Array literal space", i64 0,
i64 %6
>   %8 = load i32* %7, align 4
>   br label %"elem(array<int, 5>, int).exit"
>
> "elem(array<int, 5>, int).exit":                  ; preds =
%entry,
> %in-bounds.i
>   %iftmp.i = phi i32 [ %8, %in-bounds.i ], [ 0, %entry ]
>   ret i32 %iftmp.i
> }
>
> attributes #0 = { nounwind readnone }
>
>
> ______________________________**_________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
>
http://lists.cs.uiuc.edu/**mailman/listinfo/llvmdev<http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20130905/60c2fe1c/attachment.html>

There's no such a pass in LLVM.
I wrote one 5 years ago but never benchmarked it properly, so I never 
commited it (and because eventually clang got smarter).
The patch is here: 
http://web.ist.utl.pt/nuno.lopes/llvm_constant_extractor.txt  Again, the 
patch is 5 years old and was written for LLVM 2, so it won't work out of the
box for sure, but it should work with a few tweaks.

Nuno


----- Original Message ----- 
From: "Nicholas Chapman" <admin at indigorenderer.com>
To: <llvmdev at cs.uiuc.edu>
Sent: Thursday, September 05, 2013 9:46 PM
Subject: [LLVMdev] Optimisation pass to move an alloca'd array to a global 
constant array

> Hi All,
> I was wondering if there is an optimisation pass that moves a stack 
> allocated array, initialised with constant values, to a global constant 
> array.
> And if there is such a pass, what requirements are there for it to 
> operate?
> My optimised IR is below.  As you can see an array of 5 integers is 
> created with alloca, then each element is stored to in turn.  It would be 
> nice if this array was transformed into a global constant array.
>
> Thanks,
>     Nick
>
>
>
> ; ModuleID = 'WinterModule'
> target datalayout = 
>
"e-p:64:64:64-S128-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f16:16:16-f32:32:32-f64:64:64-f128:128:128-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"
>
> ; Function Attrs: nounwind readnone
> define i32 @"main(int)"(i32 %i, i32* nocapture %hidden) #0 {
> entry:
>   %"Array literal space" = alloca [5 x i32], align 4
>   %0 = getelementptr [5 x i32]* %"Array literal space", i64 0,
i64 0
>   store i32 1, i32* %0, align 4
>   %1 = getelementptr [5 x i32]* %"Array literal space", i64 0,
i64 1
>   store i32 2, i32* %1, align 4
>   %2 = getelementptr [5 x i32]* %"Array literal space", i64 0,
i64 2
>   store i32 3, i32* %2, align 4
>   %3 = getelementptr [5 x i32]* %"Array literal space", i64 0,
i64 3
>   store i32 4, i32* %3, align 4
>   %4 = getelementptr [5 x i32]* %"Array literal space", i64 0,
i64 4
>   store i32 5, i32* %4, align 4
>   %5 = icmp ult i32 %i, 5
>   br i1 %5, label %in-bounds.i, label %"elem(array<int, 5>,
int).exit"
>
> in-bounds.i:                                      ; preds = %entry
>   %6 = sext i32 %i to i64
>   %7 = getelementptr [5 x i32]* %"Array literal space", i64 0,
i64 %6
>   %8 = load i32* %7, align 4
>   br label %"elem(array<int, 5>, int).exit"
>
> "elem(array<int, 5>, int).exit":                  ; preds =
%entry,
> %in-bounds.i
>   %iftmp.i = phi i32 [ %8, %in-bounds.i ], [ 0, %entry ]
>   ret i32 %iftmp.i
> }
>
> attributes #0 = { nounwind readnone }
>
>
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev