<div dir="ltr"><br><br><div class="gmail_quote"><div dir="ltr">On Mon, Jul 24, 2017 at 8:17 PM Frozen via llvm-dev <<a href="mailto:llvm-dev@lists.llvm.org">llvm-dev@lists.llvm.org</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div style="line-height:1.7;color:#000000;font-size:14px;font-family:Arial"><div><span style="font-family:Arial"> This email is related with the code of commit: </span><a href="http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20120409/140650.html" style="font-family:Arial;text-decoration:underline" target="_blank"><span style="font-family:Arial">http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20120409/140650.html</span></a><span style="font-family:Arial">. This commit will let our unconditional branch instruction not be deleted
when its successor is the nature layout and its basic block's
instruction only one instruction.</span></div><div><span style="font-family:Arial"></span><br></div><span style="font-family:Arial"> However, I have some concerns about it, especially in the debug mode. Let me show you some example:<br><br>[</span><span style="font-family:Arial"><span style="font-family:Arial">1.cpp</span>]<br><a href="http://1.int" target="_blank">1.int</a> main()<br>2.{<br>3. int i;<br>4.<br>5. for (i = 0; i < 256; i++)<br>6. {<br>7. continue;<br>8. }<br>9.}<br>[/<code>1.cpp</code>]<br><br>When we use clang++ to compile it and use gdb to debug it.<br><br>[Debug Part]<br></span><pre><code class="m_4318320555433844435language-text"><span style="font-family:Arial">(gdb) b main
Breakpoint 1 at 0x100000f8b: file 1.cpp, line 5.
(gdb) r
Thread 2 hit Breakpoint 1, main () at 1.cpp:5
5 for (i = 0; i < 256; i++)
(gdb) n
7 continue;
(gdb) n
5 for (i = 0; i < 256; i++)
(gdb) n
7 continue;
(gdb) n
5 for (i = 0; i < 256; i++)
(gdb) n
7 continue;<br></span></code><code class="m_4318320555433844435language-llvm"><span style="font-family:Arial">[/Debug Part]</span></code><br><code class="m_4318320555433844435language-llvm"><span style="font-family:Arial">This behavior is as expected. <br>[LLVM Part]<br>for.cond: ; preds = %for.inc, %entry<br> %0 = load i32, i32* %i, align 4, !dbg !17<br> %cmp = icmp slt i32 %0, 256, !dbg !19<br> br i1 %cmp, label %for.body, label %for.end, !dbg !20<br><br>for.body: ; preds = %for.cond<br><b> br label %for.inc, !dbg !21 ; it will keep the debug information !21</b><br><br>for.inc: ; preds = %for.body<br> %1 = load i32, i32* %i, align 4, !dbg !23<br> %inc = add nsw i32 %1, 1, !dbg !23<br> store i32 %inc, i32* %i, align 4, !dbg !23<br> br label %for.cond, !dbg !24, !llvm.loop !25<br><br>for.end: ; preds = %for.cond<br> %2 = load i32, i32* %retval, align 4, !dbg !27<br> ret i32 %2, !dbg !27<br><br>!21 = !DILocation(line: 7, column: 5, scope: !22); <br></span></code><code class="m_4318320555433844435language-llvm"><br></code><code><span style="font-family:Arial">However, let us see another source code example.<br><br>[2.cpp]</span></code><br><code><code class="m_4318320555433844435language-c"><span style="font-family:Arial"><a href="http://1.int" target="_blank">1.int</a> main()
2.{
3. int i;
4.
5. for (i = 0; i < 256; i++)
6. {
7. i++;
8. continue;
9. }
10.}</span></code></code><code><span style="font-family:Arial"><br>[/2.cpp]<br><br>If we debug it, we can see that:<br>[Debug Part]<br></span></code><br><code><code class="m_4318320555433844435language-text"><span style="font-family:Arial">(gdb) b main
Breakpoint 1 at 0x100000f7b: file 2.cpp, line 5.
(gdb) r
Thread 2 hit Breakpoint 1, main () at 2.cpp:5
5 for (i = 0; i < 256; i++)
(gdb) n
7 i++;
(gdb) n
5 for (i = 0; i < 256; i++)
(gdb) n
7 i++;
(gdb) n
5 for (i = 0; i < 256; i++)
(gdb) n
7 i++;
(gdb) n
5 for (i = 0; i < 256; i++)<br>[/Debug Part]<br><br><b>We can not stop at continue statement of line 8! </b><br>Let us see the llvm ir:<br>[LLVM]<br></span></code></code><code><code class="m_4318320555433844435language-text"><code class="m_4318320555433844435language-llvm"><span style="font-family:Arial">for.cond: ; preds = %for.inc, %entry
%0 = load i32, i32* %i, align 4, !dbg !17
%cmp = icmp slt i32 %0, 256, !dbg !19
br i1 %cmp, label %for.body, label %for.end, !dbg !20
for.body: ; preds = %for.cond
%1 = load i32, i32* %i, align 4, !dbg !21
%inc = add nsw i32 %1, 1, !dbg !21
store i32 %inc, i32* %i, align 4, !dbg !21
<b> br label %for.inc, !dbg !23; ; // Here is continue statement, but its basic block instructions size > 1 and its successor is nature layout, which will be deleted</b>
for.inc: ; preds = %for.body
%2 = load i32, i32* %i, align 4, !dbg !24
%inc1 = add nsw i32 %2, 1, !dbg !24
store i32 %inc1, i32* %i, align 4, !dbg !24
br label %for.cond, !dbg !25, !llvm.loop !26
for.end: ; preds = %for.cond
%3 = load i32, i32* %retval, align 4, !dbg !28
ret i32 %3, !dbg !28
!23 = !DILocation(line: 8, column: 5, scope: !22)</span></code></code></code></pre>As comment,<code><b> </b><code class="m_4318320555433844435language-text"><code class="m_4318320555433844435language-llvm"><span style="font-family:Arial"><b>br label %for.inc, !dbg !23;</b></span></code></code></code> will be deleted and we can not get right debug behavior. <br><br>So
I propose one changeset, if we have debug information, we don't eliminate
the unconditional branch instruction, otherwise we eliminate it.<br></div></blockquote><div><br>It is a fairly strong goal of LLVM that the presence or absence of debug information should not affect the optimization decisions/code of the final binary, unfortunately.<br> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div style="line-height:1.7;color:#000000;font-size:14px;font-family:Arial"><br> /// (fall-through) successor, and update the CFG.<br> void FastISel::fastEmitBranch(MachineBasicBlock *MSucc,<br> const DebugLoc &DbgLoc) {<br>- if (FuncInfo.MBB->getBasicBlock()->size() > 1 &&<br>- FuncInfo.MBB->isLayoutSuccessor(MSucc)) {<br>- // For more accurate line information if this is the only instruction<br>- // in the block then emit it, otherwise we have the unconditional<br>+ if (FuncInfo.MBB->isLayoutSuccessor(MSucc) && !DbgLoc) {<br>+ // For more accurate line information if this is in debug mode<br>+ // then emit it, otherwise we have the unconditional<br> // fall-through case, which needs no instructions.<br> } else {<br> // The unconditional branch case. <br><br>Wish to hear more comments and feedbacks. <br></div><br><br><span title="neteasefooter"><p> </p></span>_______________________________________________<br>
LLVM Developers mailing list<br>
<a href="mailto:llvm-dev@lists.llvm.org" target="_blank">llvm-dev@lists.llvm.org</a><br>
<a href="http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev" rel="noreferrer" target="_blank">http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev</a><br>
</blockquote></div></div>