<div dir="ltr">Thanks for testing out LLVM PGO and evaluated the performance.<div><br></div><div>We are currently still more focused on infrastructure improvement which is the foundation for performance improvement. We are making great progress in this direction, but there are still some key missing pieces such as profile data in inliner etc. We are working on that. Once those are done, more focus will be on making more passes profile aware, make existing profile aware passes better (e.g, code layout etc).</div><div><br></div><div>I looked at this particular example. GCC PGO can reduce the runtime by half, while LLVM's PGO makes no performance difference as you noticed.</div><div><br></div><div>For GCC case, PGO itself contributes about 15% performance boost. The majority of the performance improvement comes from loop vectorization. Note that trunk GCC does not turn on vectorization at O2, but O3 or O2 with PGO.</div><div><br></div><div>LLVM also vectorizes the key loops. However compared with GCC's vectorizor, LLVM's auto-vectorizer produces worse code (e.g, long sequence of instructions to do sign extension etc): ~6.5instr/iter vs ~9instr/iter. GCC also does loop unroll after vectorization which also helped a little more. LLVM's vectorization actually hurts performance a little.</div><div><br></div><div>We will look into this issue.</div><div><br></div><div>thanks,</div><div><br></div><div>David</div></div><div class="gmail_extra"><br><div class="gmail_quote">On Fri, May 6, 2016 at 2:04 PM, Jie Chen <span dir="ltr"><<a href="mailto:Jie.Chen@mathworks.com" target="_blank">Jie.Chen@mathworks.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div style="word-wrap:break-word;color:rgb(0,0,0);font-size:14px;font-family:Calibri,sans-serif">
<div>Hi David,</div>
<div><br>
</div>
<div>I am a performance engineer from MathWorks. I am currently exploring building our products with PGO on the Mac platform. While searching for llvm PGO solutions, I came across your name many times. So I thought you were probably the guy behind llvm’s PGO
implementation! :-) Here is what confused me regarding the llvm PGO capability. I started with a small code (see my code at the end of this email) which I saw more than 10% performance improvement with PGO on Linux GCC (g++ -O2, -profile-geneate, -profile-use).
I wrote this code based on the assumption that llvm would rearrange the hot/code branches based on profile run. But when tried with Apple Clang and Clang on ubuntu, I did not see any performance improvement. Since I do not know the implementation detail of
llvm PGO, I am confused by not seeing performance improvement as I saw it with GCC (probably with Visual Studio PGO as well). Could you please offer me some insights into the issue? Or on a further question, what kind of code would benefit from llvm PGO optimization? </div>
<div><br>
</div>
<div>Best,</div>
<div><br>
</div>
<div>Jie Chen</div>
<div>MathWorks</div>
<div><br>
</div>
<div><br>
</div>
<div>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';color:rgb(195,55,32)">
<span style="color:#d53bd3">#include
</span><iostream></p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';color:rgb(195,55,32)">
<span style="color:#d53bd3">#include
</span><stdlib.h></p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';min-height:13px">
<br>
</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';color:rgb(52,189,38)">
<span style="color:#ce7924">using</span><span style="color:#000000">
</span>namespace<span style="color:#000000"> std;</span></p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';min-height:13px">
<br>
</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
<span style="color:#34bd26">long</span>
<span style="color:#34bd26">long</span> hot() {</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
<span style="color:#34bd26">long</span>
<span style="color:#34bd26">long</span> x =
<span style="color:#c33720">0</span>;</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';min-height:13px">
<br>
</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
<span style="color:#ce7924">for</span> (<span style="color:#34bd26">int</span> i =
<span style="color:#c33720">0</span>; i <
<span style="color:#c33720">1000</span>; i++) {</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
x += i^<span style="color:#c33720">2</span>;</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
}</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';min-height:13px">
<br>
</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
<span style="color:#ce7924">return</span> x;</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
}</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';min-height:13px">
<br>
</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
<span style="color:#34bd26">long</span>
<span style="color:#34bd26">long</span> cold() {</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
<span style="color:#34bd26">long</span>
<span style="color:#34bd26">long</span> y =
<span style="color:#c33720">0</span>;</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';min-height:13px">
<br>
</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
<span style="color:#ce7924">for</span> (<span style="color:#34bd26">int</span> i =
<span style="color:#c33720">0</span>; i <
<span style="color:#c33720">1000</span>; i++) {</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
y += i^<span style="color:#c33720">2</span>;</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
}</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';min-height:13px">
<br>
</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
<span style="color:#ce7924">return</span> y;</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';min-height:13px">
<br>
</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
}</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';min-height:13px">
<br>
</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
<span style="color:#34bd26">long</span>
<span style="color:#34bd26">long</span> foo() {</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
<span style="color:#34bd26">long</span>
<span style="color:#34bd26">long</span> y =
<span style="color:#c33720">0</span>;</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';min-height:13px">
<br>
</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
<span style="color:#ce7924">for</span> (<span style="color:#34bd26">int</span> i =
<span style="color:#c33720">0</span>; i <
<span style="color:#c33720">1000</span>; i++) {</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
y *= i^<span style="color:#c33720">2</span>;</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
}</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';min-height:13px">
<br>
</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
<span style="color:#ce7924">return</span> y*<span style="color:#c33720">2</span>;</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';min-height:13px">
<br>
</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
}</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';min-height:13px">
<br>
</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
<span style="color:#34bd26">long</span>
<span style="color:#34bd26">long</span> bar() {</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
<span style="color:#34bd26">long</span>
<span style="color:#34bd26">long</span> y =
<span style="color:#c33720">0</span>;</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';min-height:13px">
<br>
</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
<span style="color:#ce7924">for</span> (<span style="color:#34bd26">int</span> i =
<span style="color:#c33720">0</span>; i <
<span style="color:#c33720">1000</span>; i++) {</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
y *= i^<span style="color:#c33720">2</span>;</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
}</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';min-height:13px">
<br>
</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
<span style="color:#ce7924">return</span> y*<span style="color:#c33720">3</span>;</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';min-height:13px">
<br>
</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
}</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';min-height:13px">
<br>
</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';color:rgb(213,59,211)">
#define SIZE <span style="color:#c33720">
10000000</span></p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';min-height:13px">
<br>
</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
<span style="color:#34bd26">int</span> main() {</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';min-height:13px">
<br>
</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
<span style="color:#34bd26">int</span>* a = (<span style="color:#34bd26">int</span> *)calloc(SIZE,
<span style="color:#ce7924">sizeof</span>(<span style="color:#34bd26">int</span>));</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';min-height:13px">
<br>
</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
a[<span style="color:#c33720">100</span>] =
<span style="color:#c33720">1</span>;</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';min-height:13px">
<br>
</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
<span style="color:#34bd26">long</span>
<span style="color:#34bd26">long</span> sum =
<span style="color:#c33720">0</span>;</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';min-height:13px">
<br>
</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
<span style="color:#ce7924">for</span> (<span style="color:#34bd26">int</span> i =
<span style="color:#c33720">0</span>; i < SIZE; i++) {</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
<span style="color:#ce7924">
if</span> (a[i] == <span style="color:#c33720">
1</span>) {</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
sum += cold();</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
} <span style="color:#ce7924">
else</span> <span style="color:#ce7924">
if</span> (a[i] > <span style="color:#c33720">
1</span>) {</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
sum += bar();</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
sum += foo();</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
} <span style="color:#ce7924">
else</span> <span style="color:#ce7924">
if</span> (a[i] < <span style="color:#c33720">
1</span>) {</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
sum += hot();</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
}</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
}</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';min-height:13px">
<br>
</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
cout << sum << endl;</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';min-height:13px">
<br>
</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
<span style="color:#ce7924">return</span>
<span style="color:#c33720">0</span>;</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
}</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
<br>
</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
Makefile to compile the above code on Mac:</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
<br>
</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
<span style="color:#ce7924">.PHONY:</span> clean</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';min-height:13px">
<br>
</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
<span style="color:#34bbc7">regular:</span> main.cpp</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';color:rgb(195,55,32)">
clang++ -O2 main.cpp -o main.regular</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';min-height:13px">
<br>
</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
<span style="color:#34bbc7">hand:</span> main2.cpp</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';color:rgb(195,55,32)">
clang++ -O2 main2.cpp -o main.regular2</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';min-height:13px">
<br>
</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
<span style="color:#34bbc7">instr:</span> main.cpp</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';color:rgb(195,55,32)">
clang++ -O2 -fprofile-instr-generate main.cpp -o main.instr</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';min-height:13px">
<br>
</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
<span style="color:#34bbc7">profile:</span> main.instr</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';color:rgb(195,55,32)">
./main.instr</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';min-height:13px">
<br>
</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
<span style="color:#34bbc7">merge:</span> default.profraw</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';color:rgb(195,55,32)">
xcrun llvm-profdata merge -output default.profdata default.profraw</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';min-height:13px">
<br>
</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
<span style="color:#34bbc7">optimize:</span> default.profdata</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';color:rgb(195,55,32)">
clang++ -O2 -fprofile-instr-use=default.profdata main.cpp -o main.optimized</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';min-height:13px">
<br>
</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';color:rgb(52,187,199)">
clean:</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console'">
</p>
<p style="margin:0px;font-size:13px;line-height:normal;font-family:'Lucida Console';color:rgb(195,55,32)">
<span style="color:#34bbc7">$(RM)</span> default.* main.instr main.optimized main.regular</p>
</div>
</div>
</blockquote></div><br></div>