1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <title>Open LLVM Projects</title>
6 <link rel="stylesheet" href="llvm.css" type="text/css">
10 <div class="doc_title">
15 <li><a href="#what">What is this?</a></li>
16 <li><a href="#improving">Improving the current system</a>
18 <li><a href="#code-cleanups">Implementing Code Cleanup bugs</a></li>
19 <li><a href="#glibc">Port glibc to LLVM</a></li>
20 <li><a href="#programs">Compile programs with the LLVM Compiler</a></li>
21 <li><a href="#llvm_ir">Extend the LLVM intermediate representation</a></li>
22 <li><a href="#target">Target backend improvements</a></li>
23 <li><a href="#misc_imp">Miscellaneous Improvements</a></li>
26 <li><a href="#new">Adding new capabilities to LLVM</a>
28 <li><a href="#newfeaturebugs">Implementing new feature PRs</a></li>
29 <li><a href="#pointeranalysis">Pointer and Alias Analysis</a></li>
30 <li><a href="#profileguided">Profile-Guided Optimization</a></li>
31 <li><a href="#xforms">New Transformations and Analyses</a></li>
32 <li><a href="#x86be">X86 Back-end Improvements</a></li>
33 <li><a href="#misc_new">Miscellaneous Additions</a></li>
37 <div class="doc_author">
38 <p>Written by the <a href="http://llvm.cs.uiuc.edu/">LLVM Team</a></p>
42 <!-- *********************************************************************** -->
43 <div class="doc_section">
44 <a name="what">What is this?</a>
46 <!-- *********************************************************************** -->
48 <div class="doc_text">
50 <p>This document is meant to be a sort of "big TODO list" for LLVM. Each
51 project in this document is something that would be useful for LLVM to have, and
52 would also be a great way to get familiar with the system. Some of these
53 projects are small and self-contained, which may be implemented in a couple of
54 days, others are larger. Several of these projects may lead to interesting
55 research projects in their own right. In any case, we welcome all
58 <p>If you are thinking about tackling one of these projects, please send a mail
59 to the <a href="http://mail.cs.uiuc.edu/mailman/listinfo/llvmdev">LLVM
60 Developer's</a> mailing list, so that we know the project is being worked on.
61 Additionally this is a good way to get more information about a specific project
62 or to suggest other projects to add to this page.
65 <p>The projects in this page are open-ended. More specific projects are
66 filed as unassigned enhancements in the <a href="http://llvm.cs.uiuc.edu/bugs/">
67 LLVM bug tracker</a>. See the <a href="http://llvm.cs.uiuc.edu/bugs/buglist.cgi?keywords_type=allwords&keywords=&bug_status=NEW&bug_status=ASSIGNED&bug_status=REOPENED&bug_severity=enhancement&emailassigned_to1=1&emailtype1=substring&email1=unassigned">list of currently outstanding issues</a> if you wish to help improve LLVM.</p>
71 <!-- *********************************************************************** -->
72 <div class="doc_section">
73 <a name="improving">Improving the current system</a>
75 <!-- *********************************************************************** -->
77 <div class="doc_text">
79 <p>Improvements to the current infrastructure are always very welcome and tend
80 to be fairly straight-forward to implement. Here are some of the key areas that
81 can use improvement...</p>
85 <!-- ======================================================================= -->
86 <div class="doc_subsection">
87 <a name="code-cleanups">Implementing Code Cleanup bugs</a>
90 <div class="doc_text">
93 The <a href="http://llvm.cs.uiuc.edu/bugs/">LLVM bug tracker</a> occasionally
95 href="http://llvm.cs.uiuc.edu/bugs/buglist.cgi?short_desc_type=allwordssubstr&short_desc=&long_desc_type=allwordssubstr&long_desc=&bug_file_loc_type=allwordssubstr&bug_file_loc=&status_whiteboard_type=allwordssubstr&status_whiteboard=&keywords_type=allwords&keywords=code-cleanup&bug_status=NEW&bug_status=ASSIGNED&bug_status=REOPENED&emailassigned_to1=1&emailtype1=substring&email1=&emailassigned_to2=1&emailreporter2=1&emailcc2=1&emailtype2=substring&email2=&bugidtype=include&bug_id=&votes=&changedin=&chfieldfrom=&chfieldto=Now&chfieldvalue=&cmdtype=doit&order=Bug+Number&field0-0-0=noop&type0-0-0=noop&value0-0-0=">"code-cleanup" bugs</a> filed in it. Taking one of these and fixing it is a good
96 way to get your feet wet in the LLVM code and discover how some of its components
102 <!-- ======================================================================= -->
103 <div class="doc_subsection">
104 <a name="glibc">Port glibc to LLVM</a>
107 <div class="doc_text">
109 <p>It would be very useful to <a
110 href="http://www.gnu.org/software/libc/manual/html_node/Porting.html">port</a> <a
111 href="http://www.gnu.org/software/libc/">glibc</a> to LLVM. This would allow a
112 variety of interprocedural algorithms to be much more effective in the face of
113 library calls. The most important pieces to port are things like the string
114 library and the <tt>stdio</tt> related functions... low-level system calls like
115 '<tt>read</tt>' should stay unimplemented in LLVM.</p>
119 <!-- ======================================================================= -->
120 <div class="doc_subsection">
121 <a name="programs">Compile programs with the LLVM Compiler</a>
124 <div class="doc_text">
126 <p>We are always looking for new testcases and benchmarks for use with LLVM. In
127 particular, it is useful to try compiling your favorite C source code with LLVM.
128 If it doesn't compile, try to figure out why or report it to the <a
129 href="http://mail.cs.uiuc.edu/pipermail/llvmbugs/">llvm-bugs</a> list. If you
130 get the program to compile, it would be extremely useful to convert the build
131 system to be compatible with the LLVM Programs testsuite so that we can check it
132 into CVS and the automated tester can use it to track progress of the
135 <p>When testing a code, try running it with a variety of optimizations, and with
136 all the back-ends: CBE, llc, and lli.</p>
140 <!-- ======================================================================= -->
141 <div class="doc_subsection">
142 <a name="llvm_ir">Extend the LLVM intermediate representation</a>
145 <div class="doc_text">
148 <li>Add support for platform-independent prefetch support. The GCC <a
149 href="http://gcc.gnu.org/projects/prefetch.html">prefetch project</a> page
150 has a good survey of the prefetching capabilities of a variety of modern
157 <!-- ======================================================================= -->
158 <div class="doc_subsection">
159 <a name="target">Target backend improvements</a>
162 <div class="doc_text">
165 <li>Find benchmarks either using our <a
166 href="http://llvm.cs.uiuc.edu/testresults/">test results</a> or on your own,
167 where LLVM code generators do not produce optimal code or simply where another
168 compiler produces better code. Try to minimize the test case that
169 demonstrates the issue. Then, either <a
170 href="http://llvm.cs.uiuc.edu/bugs/">submit a bug</a> with your testcase and
171 the code that LLVM produces vs. the code that it <em>should</em> produce, or
172 even better, see if you can improve the code generator and submit a
178 <!-- ======================================================================= -->
179 <div class="doc_subsection">
180 <a name="misc_imp">Miscellaneous Improvements</a>
183 <div class="doc_text">
186 <li>Someone needs to look into getting the <tt>ranlib</tt> tool to index LLVM
187 bytecode files, so that linking in .a files is not hideously slow. They
188 would also then have to implement the reader for this index in
191 <li>Rework the PassManager to be more flexible</li>
193 <li>Some transformations and analyses only work on reducible flow graphs. It
194 would be nice to have a transformation which could be "required" by these passes
195 which makes irreducible graphs reducible. This can easily be accomplished
196 through code duplication. See <a
197 href="http://citeseer.nj.nec.com/janssen97making.html">Making Graphs Reducible
198 with Controlled Node Splitting</a> and perhaps <a
199 href="http://doi.acm.org/10.1145/262004.262005">Nesting of Reducible and
200 Irreducible Loops</a>.</li>
206 <!-- *********************************************************************** -->
207 <div class="doc_section">
208 <a name="new">Adding new capabilities to LLVM</a>
210 <!-- *********************************************************************** -->
212 <div class="doc_text">
214 <p>Sometimes creating new things is more fun than improving existing things.
215 These projects tend to be more involved and perhaps require more work, but can
216 also be very rewarding.</p>
221 <!-- ======================================================================= -->
222 <div class="doc_subsection">
223 <a name="newfeaturebugs">Implementing new feature PRs</a>
226 <div class="doc_text">
228 <p>Many ideas for feature requests are stored in LLVM bugzilla. Just <a
229 href="http://llvm.cs.uiuc.edu/bugs/buglist.cgi?short_desc_type=allwordssubstr&short_desc=&long_desc_type=allwordssubstr&long_desc=&bug_file_loc_type=allwordssubstr&bug_file_loc=&status_whiteboard_type=allwordssubstr&status_whiteboard=&keywords_type=allwords&keywords=new-feature&bug_status=UNCONFIRMED&bug_status=NEW&bug_status=ASSIGNED&bug_status=REOPENED&emailassigned_to1=1&emailtype1=substring&email1=&emailassigned_to2=1&emailreporter2=1&emailcc2=1&emailtype2=substring&email2=&bugidtype=include&bug_id=&votes=&changedin=&chfieldfrom=&chfieldto=Now&chfieldvalue=&cmdtype=doit&namedcmd=All+PRs&newqueryname=&order=Bug+Number&field0-0-0=noop&type0-0-0=noop&value0-0-0=">search for bugs with a "new-feature" keyword</a>.</p>
233 <!-- ======================================================================= -->
234 <div class="doc_subsection">
235 <a name="pointeranalysis">Pointer and Alias Analysis</a>
238 <div class="doc_text">
240 <p>We have a <a href="AliasAnalysis.html">strong base for development</a> of
241 both pointer analysis based optimizations as well as pointer analyses
242 themselves. It seems natural to want to take advantage of this...</p>
245 <li>Implement a flow-sensitive context-sensitive alias analysis algorithm<br>
246 - Pick one of the somewhat efficient algorithms, but strive for maximum
249 <li>Implement a flow-sensitive context-insensitive alias analysis algorithm<br>
250 - Just an efficient local algorithm perhaps?</li>
252 <li>Implement alias-analysis-based optimizations:
260 <!-- ======================================================================= -->
261 <div class="doc_subsection">
262 <a name="profileguided">Profile-Guided Optimization</a>
265 <div class="doc_text">
267 <p>We now have a unified infrastructure for writing profile-guided
268 transformations, which will work either at offline-compile-time or in the JIT,
269 but we don't have many transformations. We would welcome new profile-guided
270 transformations as well as improvements to the current profiling system.
273 <p>Ideas for profile-guided transformations:</p>
276 <li>Superblock formation (with many optimizations)</li>
277 <li>Loop unrolling/peeling</li>
278 <li>Profile directed inlining</li>
283 <p>Improvements to the existing support:</p>
286 <li>The current block and edge profiling code that gets inserted is very simple
287 and inefficient. Through the use of control-dependence information, many fewer
288 counters could be inserted into the code. Also, if the execution count of a
289 loop is known to be a compile-time or runtime constant, all of the counters in
290 the loop could be avoided.</li>
292 <li>You could implement one of the "static profiling" algorithms which analyze a
293 piece of code an make educated guesses about the relative execution frequencies
294 of various parts of the code.</li>
296 <li>You could add path profiling support, or adapt the existing LLVM path
297 profiling code to work with the generic profiling interfaces.</li>
302 <!-- ======================================================================= -->
303 <div class="doc_subsection">
304 <a name="xforms">New Transformations and Analyses</a>
307 <div class="doc_text">
310 <li>Implement <a href="http://www.cs.purdue.edu/homes/vandrutj/">GVN-PRE</a>, a
311 powerful and simple Partial Redundancy Elimination algorithm for SSA form</li>
312 <li>Implement a Dependence Analysis Infrastructure<br>
313 - Design some way to represent and query dep analysis</li>
314 <li>Implement a strength reduction pass</li>
315 <li>Value range propagation pass</li>
320 <!-- ======================================================================= -->
321 <div class="doc_section">
322 <a name="x86be">X86 Back-end Improvements</a>
325 <div class="doc_text">
328 <li>Implement a better instruction selector</li>
329 <li>Implement support for the "switch" instruction without requiring the
330 lower-switches pass.</li>
331 <li>Implement interprocedural register allocation. The CallGraphSCCPass can be
332 used to implement a bottom-up analysis that will determine the *actual*
333 registers clobbered by a function. Use the pass to fine tune register usage
334 in callers based on *actual* registers used by the callee.</li>
339 <!-- ======================================================================= -->
340 <div class="doc_section">
341 <a name="misc_new">Miscellaneous Additions</a>
344 <div class="doc_text">
347 <li>Port the <a href="http://www-sop.inria.fr/mimosa/fp/Bigloo/">Bigloo</A>
348 Scheme compiler, from Manuel Serrano at INRIA Sophia-Antipolis, to
349 output LLVM bytecode. It seems that it can already output .NET
350 bytecode, JVM bytecode, and C, so LLVM would ostensibly be another good
352 <li>Write a new frontend for C/C++ <b>in</b> C++, giving us the ability to
353 directly use LLVM C++ classes from within a compiler rather than use
354 C-based wrapper functions a la llvm-gcc. One possible starting point is the <a
355 href="http://www.parashift.com/c++-faq-lite/compiler-dependencies.html#faq-37.11">C++
356 yacc grammar by Ed Willink</a>.</li>
357 <li>Write a new frontend for some other language (Java? OCaml? Forth?)</li>
358 <li>Write a new backend for a target (IA64? MIPS? MMIX?)</li>
359 <li>Write a disassembler for machine code that would use TableGen to output
360 <tt>MachineInstr</tt>s for transformations, optimizations, etc.</li>
361 <li>Random test vector generator: Use a C grammar to generate random C code;
362 run it through llvm-gcc, then run a random set of passes on it using opt.
363 Try to crash opt. When opt crashes, use bugpoint to reduce the test case and
364 mail the result to yourself. Repeat ad infinitum.</li>
365 <li>Design a simple, recognizable logo.</li>
366 <li>Improve the usefulness and utility of the Skeleton target backend:
368 <li>Convert the non-functional Skeleton target to become an abstract machine
369 target (choose some simple instructions, a register set, etc). This will
370 become a much more useful example of a backend since it would be a simple
371 but <em>functional</em> backend. Examples of such architectures include MIX,
373 href="http://www.cs.cinvestav.mx/SC/prof_personal/adiaz/vhdl/DLX/">DLX</a>,
374 or come up with your own!</li>
375 <li>Use the new Skeleton backend in the Interpreter: compile LLVM to Skeleton
376 target, and then interpret that code instead of LLVM. Performance win would
377 be the primary goal, as the number of registers would be a small constant
378 instead of unbounded, for example.</li>
384 <!-- *********************************************************************** -->
388 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
389 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
390 <a href="http://validator.w3.org/check/referer"><img
391 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
393 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
394 <a href="http://llvm.cs.uiuc.edu">LLVM Compiler Infrastructure</a><br>
395 Last modified: $Date$