<div dir="ltr"><br><div>Hi Nick,</div><div><br></div><div>Thanks for your inputs.</div><div><br></div><div>I think this test case is in accordance with the new simplification we just taught it.</div><div><br></div><div>Assuming just 8 bits for explaination:</div>
<div><br></div><div>Anding or with 1 will give us:</div><div>Either 00000000 or 00000001 for the variable o depending on the LSB of x.</div><div>Adding -1 i.e 11111111 will give us:</div><div>Either 11111111 or 00000000 respectively for the variable n</div>
<div><br></div><div>this is exactly what we need in n, all zeros or all ones!</div><div><br></div><div><br></div><div>I tried a simpler test case like:</div><div><br></div><div><div style="font-family:arial,sans-serif;font-size:13px">
define i32 @test1(i32 %x) {</div><div class="im" style="font-family:arial,sans-serif;font-size:13px"> %n = or i32 %x, -1</div><div class="im" style="font-family:arial,sans-serif;font-size:13px"> %t = ashr i32 %n, 17</div>
<div style="font-family:arial,sans-serif;font-size:13px"> ret i32 %t</div><div style="font-family:arial,sans-serif;font-size:13px">}</div></div><div style="font-family:arial,sans-serif;font-size:13px"><br></div><div style="font-family:arial,sans-serif;font-size:13px">
But in this case the instruction t here will get constant folded before only.</div><div style="font-family:arial,sans-serif;font-size:13px"><br></div><div style="font-family:arial,sans-serif;font-size:13px">What do you say?? </div>
<div style="font-family:arial,sans-serif;font-size:13px"><br></div><div style="font-family:arial,sans-serif;font-size:13px">Thanks,</div><div style="font-family:arial,sans-serif;font-size:13px">Rahul</div><div><br></div>
<div>
<br></div><div><br></div><div><br></div><div><br></div></div><div class="gmail_extra"><br><br><div class="gmail_quote">On Thu, Jun 19, 2014 at 9:10 AM, Nick Lewycky <span dir="ltr"><<a href="mailto:nlewycky@google.com" target="_blank">nlewycky@google.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div dir="ltr"><div class="gmail_extra"><div class="gmail_quote"><div class="">On 18 June 2014 06:19, suyog <span dir="ltr"><<a href="mailto:suyog.sarda@samsung.com" target="_blank">suyog.sarda@samsung.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex">
<br>
Hi Nick,<br>
<br>
Added supporting testcase.<br>
<br>
Please if you could commit it for me if it looks<br>
good to you.<br></blockquote><div><br></div></div><div>Not yet, sorry. It needs a test that ensures that "opt -instsimplify" *is* doing the new simplification we just taught it. You can keep the test you wrote too if you like. Also, using "grep" is old-style, please use FileCheck. For example:</div>
<div><br></div><div><div>define i32 @test1(i32 %x) {</div><div>; CHECK-LABEL: @test1</div><div>; CHECK-NOT: ashr</div><div class=""><div> %o = and i32 %x, 1</div></div><div class=""><div> %n = add i32 %o, -1</div></div>
<div class=""><div> %t = ashr i32 %n, 17</div></div><div> ret i32 %t</div>
<div>}</div></div><span class="HOEnZb"><font color="#888888"><div><br></div><div>Nick</div></font></span><div><div class="h5"><div><br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex">
<div>Thanks,<br>
Suyog<br>
<br>
<a href="http://reviews.llvm.org/D4102" target="_blank">http://reviews.llvm.org/D4102</a><br>
<br>
Files:<br>
lib/Analysis/InstructionSimplify.cpp<br>
lib/Transforms/InstCombine/InstCombineShifts.cpp<br>
</div> test/Transforms/InstSimplify/ashr-nop.ll<br>
<div><div><br>
Index: lib/Analysis/InstructionSimplify.cpp<br>
===================================================================<br>
--- lib/Analysis/InstructionSimplify.cpp<br>
+++ lib/Analysis/InstructionSimplify.cpp<br>
@@ -1436,6 +1436,11 @@<br>
cast<OverflowingBinaryOperator>(Op0)->hasNoSignedWrap())<br>
return X;<br>
<br>
+ // Arithmetic shifting an all-sign-bit value is a no-op.<br>
+ unsigned NumSignBits = ComputeNumSignBits(Op0);<br>
+ if (NumSignBits == Op0->getType()->getScalarSizeInBits())<br>
+ return Op0;<br>
+<br>
return nullptr;<br>
}<br>
<br>
Index: lib/Transforms/InstCombine/InstCombineShifts.cpp<br>
===================================================================<br>
--- lib/Transforms/InstCombine/InstCombineShifts.cpp<br>
+++ lib/Transforms/InstCombine/InstCombineShifts.cpp<br>
@@ -820,10 +820,5 @@<br>
APInt::getSignBit(I.getType()->getScalarSizeInBits())))<br>
return BinaryOperator::CreateLShr(Op0, Op1);<br>
<br>
- // Arithmetic shifting an all-sign-bit value is a no-op.<br>
- unsigned NumSignBits = ComputeNumSignBits(Op0);<br>
- if (NumSignBits == Op0->getType()->getScalarSizeInBits())<br>
- return ReplaceInstUsesWith(I, Op0);<br>
-<br>
return nullptr;<br>
}<br>
</div></div>Index: test/Transforms/InstSimplify/ashr-nop.ll<br>
===================================================================<br>
--- test/Transforms/InstSimplify/ashr-nop.ll<br>
+++ test/Transforms/InstSimplify/ashr-nop.ll<br>
@@ -0,0 +1,8 @@<br>
+; RUN: opt < %s -instsimplify -S | not grep ashr<br>
+<br>
+define i32 @foo(i32 %x) {<br>
+ %o = and i32 %x, 1<br>
+ %n = add i32 %o, -1<br>
+ %t = ashr i32 %n, 17<br>
+ ret i32 %t<br>
+}<br>
</blockquote></div></div></div><br></div></div>
<br>_______________________________________________<br>
llvm-commits mailing list<br>
<a href="mailto:llvm-commits@cs.uiuc.edu">llvm-commits@cs.uiuc.edu</a><br>
<a href="http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits" target="_blank">http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits</a><br>
<br></blockquote></div><br></div>