<div dir="ltr"><div>Thanks for your help! Providing data layout works for the example I have. </div><div><br></div><div>I also feel there is still something else going on during the investigation. </div><div><br></div><div>I have another simple ir, which basically does the same thing as the original example except initialize the array element by element instead of by an constant array.</div><div><br></div><div>---</div><div><div>define void @f(i32* nocapture %l0) {</div><div>entry:</div><div> %fc_ = alloca [1 x i32]</div><div> %0 = getelementptr inbounds [1 x i32]* %fc_, i32 0, i32 0</div><div> store i32 1, i32* %0, align 4</div><div> %1 = getelementptr [1 x i32]* %fc_, i64 0, i64 0</div><div> %2 = load i32* %1</div><div> %tobool = icmp eq i32 %2, 0</div><div> br i1 %tobool, label %4, label %3</div><div><br></div><div>; <label>:3 ; preds = %entry</div><div> store i32 1, i32* %l0</div><div> br label %4</div><div><br></div><div>; <label>:4 ; preds = %entry, %3</div><div> %storemerge = phi i32 [ 1, %3 ], [ 0, %entry ]</div><div> store i32 %storemerge, i32* %l0</div><div> ret void</div><div>}</div></div><div>---</div><div><br></div><div>For this ir, without target data layout,</div><div>opt -O2 got the expected optmization:</div><div><br></div><div>---</div><div><div>; ModuleID = 't1.txt'</div><div><br></div><div>; Function Attrs: nounwind</div><div>define void @f(i32* nocapture %l0) #0 {</div><div> store i32 1, i32* %l0</div><div> ret void</div><div>}</div><div><br></div><div>attributes #0 = { nounwind }</div></div><div>---</div><div><br></div><div>By checking the ir after each transformation, I see that gvn removes the load expression and the transformations before do nothing. </div><div><br></div><div>So, I ran opt -gvn on the ir, but did not get load expression eliminated.</div><div><br></div><div>Checking into the gvn code. Looks like the memory dependency computation got different results for the load expression with -O2 or -gvn. With O2, the load is not clobbered, but with gvn alone, it says clobbered.</div><div><br></div><div>Does that sound expected?</div><div><br></div><div><br></div><div><br></div><div><br></div><div><br></div></div><div class="gmail_extra"><br><div class="gmail_quote">On Wed, Sep 10, 2014 at 12:50 PM, Philip Reames <span dir="ltr"><<a href="mailto:listmail@philipreames.com" target="_blank">listmail@philipreames.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">I came in to an email this morning that said basically the same thing for the reduced example we were looking at. However, the original IR it came from (before hand reduction) had the data layout set correctly, so there's probably still *something* going on. It's just not what I thought at first. :)<span class="HOEnZb"><font color="#888888"><br>
<br>
Philip</font></span><div class="HOEnZb"><div class="h5"><br>
<br>
<br>
On 09/10/2014 02:26 AM, Roel Jordans wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
Looking at the -debug output of opt shows that SROA was skipped due to missing target data.<br>
<br>
Adding something like:<br>
<br>
target datalayout = "e-p:32:32:32-i32:32:32"<br>
<br>
to the top seems sufficient to fix the issue at -O3.<br>
<br>
By defining the size and storage requirements for i32 SROA is capable of rewriting the array load into a constant scalar load which can then be further optimized.<br>
<br>
Cheers,<br>
Roel<br>
<br>
On 09/09/14 18:30, Peng Cheng wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
I have the following simplified llvm ir, which basically returns value<br>
based on the first value of a constant array.<br>
<br>
----<br>
; ModuleID = 'simple_ir3.txt'<br>
<br>
@f.b = constant [1 x i32] [i32 1], align 4 ; constant array<br>
with value 1 at the first element<br>
<br>
define void @f(i32* nocapture %l0) {<br>
entry:<br>
%fc_ = alloca [1 x i32]<br>
%f.b.v = load [1 x i32]* @f.b<br>
store [1 x i32] %f.b.v, [1 x i32]* %fc_<br>
%0 = getelementptr [1 x i32]* %fc_, i64 0, i64 0 ; load the first<br>
element of the constant array, which is actually 1<br>
%1 = load i32* %0<br>
%tobool = icmp ne i32 %1, 0 ; check the first element to<br>
see if it is 1, which is actually always true since the first element of<br>
constant array is 1<br>
br i1 %tobool, label %2, label %4<br>
<br>
; <label>:2 ; true branch<br>
store i32 1, i32* %l0;<br>
%3 = load i32* %l0;<br>
br label %4<br>
<br>
; <label>:4<br>
%storemerge = phi i32 [ %3, %2 ], [ 0, %entry ]<br>
store i32 %storemerge, i32* %l0<br>
ret void<br>
}<br>
---<br>
<br>
I ran opt -O3 simple_ir.txt -S, and got:<br>
<br>
---<br>
; ModuleID = 'simple_ir3.txt'<br>
<br>
@f.b = constant [1 x i32] [i32 1], align 4<br>
<br>
; Function Attrs: nounwind<br>
define void @f(i32* nocapture %l0) #0 {<br>
entry:<br>
%fc_ = alloca [1 x i32]<br>
store [1 x i32] [i32 1], [1 x i32]* %fc_<br>
%0 = getelementptr [1 x i32]* %fc_, i64 0, i64 0<br>
%1 = load i32* %0<br>
%tobool = icmp eq i32 %1, 0<br>
br i1 %tobool, label %3, label %2<br>
<br>
; <label>:2 ; preds = %entry<br>
store i32 1, i32* %l0<br>
br label %3<br>
<br>
; <label>:3 ; preds = %entry, %2<br>
%storemerge = phi i32 [ 1, %2 ], [ 0, %entry ]<br>
store i32 %storemerge, i32* %l0<br>
ret void<br>
}<br>
<br>
attributes #0 = { nounwind }<br>
---<br>
<br>
I would expect that the constant folding, or some other transformations,<br>
would be able to fold the constant to get the following ir:<br>
<br>
---<br>
define void @f(i32* nocapture %l0) #0 {<br>
store i32 1, i32* %l0<br>
ret void<br>
}<br>
---<br>
<br>
How could I get the expected optimized ir? update the original ir, or<br>
use different set of transformations?<br>
<br>
Any suggestions or comments?<br>
<br>
<br>
Thanks,<br>
-Peng<br>
<br>
<br>
______________________________<u></u>_________________<br>
LLVM Developers mailing list<br>
<a href="mailto:LLVMdev@cs.uiuc.edu" target="_blank">LLVMdev@cs.uiuc.edu</a> <a href="http://llvm.cs.uiuc.edu" target="_blank">http://llvm.cs.uiuc.edu</a><br>
<a href="http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev" target="_blank">http://lists.cs.uiuc.edu/<u></u>mailman/listinfo/llvmdev</a><br>
<br>
</blockquote>
______________________________<u></u>_________________<br>
LLVM Developers mailing list<br>
<a href="mailto:LLVMdev@cs.uiuc.edu" target="_blank">LLVMdev@cs.uiuc.edu</a> <a href="http://llvm.cs.uiuc.edu" target="_blank">http://llvm.cs.uiuc.edu</a><br>
<a href="http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev" target="_blank">http://lists.cs.uiuc.edu/<u></u>mailman/listinfo/llvmdev</a><br>
</blockquote>
<br>
______________________________<u></u>_________________<br>
LLVM Developers mailing list<br>
<a href="mailto:LLVMdev@cs.uiuc.edu" target="_blank">LLVMdev@cs.uiuc.edu</a> <a href="http://llvm.cs.uiuc.edu" target="_blank">http://llvm.cs.uiuc.edu</a><br>
<a href="http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev" target="_blank">http://lists.cs.uiuc.edu/<u></u>mailman/listinfo/llvmdev</a><br>
</div></div></blockquote></div><br></div>