[mlgo] Add latency modeling topic to 2023 GSoC

[asl]

LGTM provided there will be a Discourse link :)

[aidengrossman]

LGTM

[ondrasej]

OpenProjects.html

778–783

I'd go with something like:

Better modeling of the execution environment by including additional runtime/profiling information, such as additional PMU data, LLC miss probabilities or branch mispredictions. This involves (1) building a data collection pipeline that covers additional runtime information, (2) modifying the ML models to allow processing this data, and (3) modifying the training and inference process for the models to make use this data.

Today, the models are almost pure static analysis; they see the instructions, but they make one-size-fits-all assumptions about the execution environment and the runtime behavior of the code. The goal of this project is to move from static analysis towards more dynamic models that better represent code the way it actually executes.

[mtrofin]

feedback

[mtrofin]

OpenProjects.html

778–783

thanks - done!

[ondrasej]

LGTM

OpenProjects.html

 <!--#include virtual="header.incl" -->
 
 <div class="www_sectiontitle">Open LLVM Projects</div>
 
 <ul>
   <li>Google Summer of Code Ideas & Projects
     <ul>
       <li>
         <a href="#gsoc23">Google Summer of Code 2023</a>
         <ul>
           <li>
             <b>LLVM Core</b>
             <ul>
                <li><a href="#llvm_improving_compile_times">Improving compile times</a></li>
                <li><a href="#llvm_addressing_rust_optimization_failures">Addressing Rust optimization failures</a></li>
                <li><a href="#remove_ub_tests">Remove undefined behavior from tests</a></li>
+               <li><a href="#llvm_mlgo_latency_model">Better performance models for MLGO training</a></li>
             </ul>
           </li>
           <li>
             <b>LLD</b>
             <ul>
               <li><a href="#llvm_lld_embedded">LLD Linker Improvements for Embedded Targets</a></li>
             </ul>
           </li>
<div class="www_text">
   <p><b>Discourse</b>
     <a href="https://discourse.llvm.org/t/instrumentation-of-clang-llvm-for-compile-time">
       URL</a>
   </p>
 </div>
 
 <!-- *********************************************************************** -->
 <div class="www_subsubsection">
+  <a id="llvm_mlgo_latency_model">Better performance models for MLGO training</a>
+</div>
+<!-- *********************************************************************** -->
+<div class="www_text">
+  <p><b>Description of the project</b>
+    We are using machine-guided compiler optimizations ("MLGO") for register allocation eviction and inlining for size, in
+    real-life deployments. The ML models have been trained with reinforcement learning algorithms. Expanding to more
+    performance areas is currently impeded by the poor prediction quality of our performance estimation models. Improving
+    those is critical to the effectiveness of reinforcement learning training algorithms, and therefore to enabling applying
+    MLGO systematically to more optimizations.
+  </p>
+  <p><b>Project size:</b> either 175 or 350 hr.</p>
+  <p><b>Difficulty:</b> Medium</p>
+  <p><b>Skills:</b> C/C++, some compiler experience, some Python. ML experience is a bonus.</p>
+  <p><b>Expected outcomes</b>: Better modeling of the execution environment by including additional runtime/profiling
+    information, such as additional PMU data, LLC miss probabilities or branch mispredictions. This involves (1) building
+    a data collection pipeline that covers additional runtime information, (2) modifying the ML models to allow processing
+    this data, and (3) modifying the training and inference process for the models to make use this data.
+  <p>Today, the models are almost pure static analysis; they see the instructions, but they make one-size-fits-all
+    assumptions about the execution environment and the runtime behavior of the code. The goal of this project is to move
+    from static analysis towards more dynamic models that better represent code the way it actually executes.</p>
+  <p><b>Mentors</b>
+    Ondrej Sykora, Mircea Trofin, Aiden Grossman
+  </p>
+  <p>
+  <b>Discourse</b>
+  <a href="https://discourse.llvm.org/t/better-performance-models-for-mlgo-training/68219">URL</a>
+  </p>
+</div>
+
+<!-- *********************************************************************** -->
+<div class="www_subsubsection">
   <a id="llvm_mlgo_passes">Machine Learning Guided Ordering of Compiler Optimization Passes</a>
 </div>
 <!-- *********************************************************************** -->
 <div class="www_text">
   <p><b>Description of the project</b>
   This continues the work of GSoC 2020 and <a href="https://summerofcode.withgoogle.com/archive/2021/projects/6411038932598784">2021</a>.
 
   Developers generally use standard optimization pipelines like -O2 and -O3 to