<div dir="ltr">(sending this message again. Apparently it didn't make it into llvm-dev because the message was too long. Sorry for the spam).<br><br><br>
Hi Clement!

<br><div class="gmail_extra"><br><div class="gmail_quote">On Fri, Mar 2, 2018 at 12:33 PM, Clement Courbet <span dir="ltr"><<a href="mailto:courbet@google.com" target="_blank">courbet@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"><span class=""><span style="color:rgb(34,34,34);font-family:arial,sans-serif;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;font-size:12.8px;float:none;display:inline">Hi Andrea,</span><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;font-size:12.8px"><br></div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;font-size:12.8px">Thanks for this great RFC ! I've put some high-level comments here, and I'll give more focused comments in the review on Phabricator.</div><br></span></div></blockquote><div><br>
Thanks!<br> <br></div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div dir="ltr"><span class=""></span><div class="gmail_extra"><br><div class="gmail_quote"><span class="">On Thu, Mar 1, 2018 at 6:22 PM, Andrea Di Biagio <span dir="ltr"><<a href="mailto:andrea.dibiagio@gmail.com" target="_blank">andrea.dibiagio@gmail.com</a>></span> wrote:<br></span><span class=""><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div dir="ltr">Hi all,<br><br>At Sony we developed an LLVM based performance analysis tool named llvm-mca. We<br>currently use it internally to statically measure the performance of code, and<br>to help triage potential problems with target scheduling models.  We decided to<br>post this RFC because we are interested in the feedback from the community, and<br>we also believe that other people might be interested in a tool like this.<br><br></div></blockquote><div><br></div></span><span class=""><div><p dir="ltr" style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:12.8px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;line-height:1.38;margin-top:0pt;margin-bottom:0pt"><span style="font-size:9.5pt;font-family:Arial;color:rgb(0,0,0);background-color:rgb(255,255,255);font-weight:400;font-style:normal;font-variant:normal;text-decoration:none;vertical-align:baseline;white-space:pre-wrap">We are definitely interested in tools like this. As a matter of fact, we (Google) have developed a similar tool. Our approach is very similar to the one described here, so I’m confident that we can merge our tool with the one proposed here. There are some similarities and differences in the design, which I’ll quickly go over here. I’ll give my opinions about respective merits of each approach when discussing specific points below.</span></p><p dir="ltr" style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:12.8px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;line-height:1.38;margin-top:0pt;margin-bottom:0pt"><span style="font-size:9.5pt;font-family:Arial;color:rgb(0,0,0);background-color:rgb(255,255,255);font-weight:400;font-style:normal;font-variant:normal;text-decoration:none;vertical-align:baseline;white-space:pre-wrap">The main differences with the design are:</span></p><ul style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:12.8px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;margin-top:0pt;margin-bottom:0pt"><li dir="ltr" style="margin-left:15px;list-style-type:disc;font-size:9.5pt;font-family:Arial;color:rgb(0,0,0);background-color:rgb(255,255,255);font-weight:400;font-style:normal;font-variant:normal;text-decoration:none;vertical-align:baseline;white-space:pre-wrap"><p dir="ltr" style="line-height:1.38;margin-top:0pt;margin-bottom:0pt"><span style="font-size:9.5pt;font-family:Arial;color:rgb(0,0,0);background-color:rgb(255,255,255);font-weight:400;font-style:normal;font-variant:normal;text-decoration:none;vertical-align:baseline;white-space:pre-wrap">We simulate the CPU frontend too (parse/decode). We’ve found out that some kernels benefit from instruction reorderings that do not change port pressure but change how instructions are parsed and decoded (e.g. parse window can be a bottleneck).</span></p></li><li dir="ltr" style="margin-left:15px;list-style-type:disc;font-size:9.5pt;font-family:Arial;color:rgb(0,0,0);background-color:rgb(255,255,255);font-weight:400;font-style:normal;font-variant:normal;text-decoration:none;vertical-align:baseline;white-space:pre-wrap"><p dir="ltr" style="line-height:1.38;margin-top:0pt;margin-bottom:0pt"><span style="font-size:9.5pt;font-family:Arial;color:rgb(0,0,0);background-color:rgb(255,255,255);font-weight:400;font-style:normal;font-variant:normal;text-decoration:none;vertical-align:baseline;white-space:pre-wrap">Our code is more modular on the subtarget model side. Our simulator is divided in four parts: </span></p></li><ul style="margin-top:0pt;margin-bottom:0pt"><li dir="ltr" style="margin-left:15px;list-style-type:circle;font-size:9.5pt;font-family:Arial;color:rgb(0,0,0);background-color:rgb(255,255,255);font-weight:400;font-style:normal;font-variant:normal;text-decoration:none;vertical-align:baseline;white-space:pre-wrap"><p dir="ltr" style="line-height:1.38;margin-top:0pt;margin-bottom:0pt"><span style="font-size:9.5pt;font-family:Arial;color:rgb(0,0,0);background-color:rgb(255,255,255);font-weight:400;font-style:normal;font-variant:normal;text-decoration:none;vertical-align:baseline;white-space:pre-wrap">CPU components (parser/decoder/renamer/ROB/Is<wbr>sue Port/...) are modeled and unit tested individually. These are intended to be mostly target-independant.</span></p></li><li dir="ltr" style="margin-left:15px;list-style-type:circle;font-size:9.5pt;font-family:Arial;color:rgb(0,0,0);background-color:rgb(255,255,255);font-weight:400;font-style:normal;font-variant:normal;text-decoration:none;vertical-align:baseline;white-space:pre-wrap"><p dir="ltr" style="line-height:1.38;margin-top:0pt;margin-bottom:0pt"><span style="font-size:9.5pt;font-family:Arial;color:rgb(0,0,0);background-color:rgb(255,255,255);font-weight:400;font-style:normal;font-variant:normal;text-decoration:none;vertical-align:baseline;white-space:pre-wrap">Each subtarget describes how to assemble these components together. This is where knowledge about the target is injected.</span></p></li><li dir="ltr" style="margin-left:15px;list-style-type:circle;font-size:9.5pt;font-family:Arial;color:rgb(0,0,0);background-color:rgb(255,255,255);font-weight:400;font-style:normal;font-variant:normal;text-decoration:none;vertical-align:baseline;white-space:pre-wrap"><p dir="ltr" style="line-height:1.38;margin-top:0pt;margin-bottom:0pt"><span style="font-size:9.5pt;font-family:Arial;color:rgb(0,0,0);background-color:rgb(255,255,255);font-weight:400;font-style:normal;font-variant:normal;text-decoration:none;vertical-align:baseline;white-space:pre-wrap">The framework drives the components in a subtarget-agnostic way. Events are logged to a `SimulationLog` object. This is similar to the `HWEventListener` interface in llvm-mca.</span></p></li><li dir="ltr" style="margin-left:15px;list-style-type:circle;font-size:9.5pt;font-family:Arial;color:rgb(0,0,0);background-color:rgb(255,255,255);font-weight:400;font-style:normal;font-variant:normal;text-decoration:none;vertical-align:baseline;white-space:pre-wrap"><p dir="ltr" style="line-height:1.38;margin-top:0pt;margin-bottom:0pt"><span style="font-size:9.5pt;font-family:Arial;color:rgb(0,0,0);background-color:rgb(255,255,255);font-weight:400;font-style:normal;font-variant:normal;text-decoration:none;vertical-align:baseline;white-space:pre-wrap">Analysis passes analyze the `SimulationLog` to extract whatever metric they care about (e.g. port pressure or IPC), or generate an annotated trace. A similar functionality is provided in llvm-mca `XXView` implementations. </span></p></li></ul></ul><p dir="ltr" style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:12.8px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;line-height:1.38;margin-top:0pt;margin-bottom:0pt"><span style="font-size:9.5pt;font-family:Arial;color:rgb(0,0,0);background-color:rgb(255,255,255);font-weight:400;font-style:normal;font-variant:normal;text-decoration:none;vertical-align:baseline;white-space:pre-wrap">We also have a IACA-like binary that displays analysis results.</span></p><p dir="ltr" style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:12.8px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;line-height:1.38;margin-top:0pt;margin-bottom:0pt"><span style="font-size:9.5pt;font-family:Arial;color:rgb(0,0,0);background-color:rgb(255,255,255);font-weight:400;font-style:normal;font-variant:normal;text-decoration:none;vertical-align:baseline;white-space:pre-wrap">For reference, our code can be found here: <a href="https://github.com/google/EXEgesis/tree/master/llvm_sim" style="color:rgb(17,85,204)" target="_blank">https://github.com/google/EXEg<wbr>esis/tree/master/llvm_sim</a></span></p></div></span></div></div></div></blockquote><div><br>
<div>Thanks for the info. I will definitely look into your code.<br></div><div><br></div><div>I
 had a look at some of your comments in the review, and I agree that the
 design can/should be improved.Our plan is also to move at one point the analysis into a 
library in order to make it accessible from the compiler (this was 
requested by Simon (Pilgrim) many times :-)).<br>It would be great if we
 could work together on this tool, and I like your idea of integrating 
features from your tool into this one.<br></div><div>We could start by 
further modularizing the tool (as you have already suggested in the 
phabricator review). We could then integrate all the missing analysis 
(plus improvements) from your code into this.<div class="gmail-yj6qo gmail-ajU"><div id="gmail-:374" class="gmail-ajR" tabindex="0"><img class="gmail-ajT" src="https://ssl.gstatic.com/ui/v1/icons/mail/images/cleardot.gif"></div></div></div>

Cheers,<br></div><div>-Andrea<br></div></div></div></div>