User:Asfrent/Application - Revision history

Asfrent at 16:36, 20 March 2014

2014-03-20T16:36:08Z

2014-03-20T03:03:32Z

Proof of concept (coding challenge and one step further)

2014-03-20T02:59:27Z

Roadmap and deliverables

2014-03-20T02:53:34Z

Proof of concept (coding challenge and one step further)

2014-03-20T02:52:21Z

Roadmap and deliverables

2014-03-20T02:45:05Z

About me (short summary)

2014-03-20T02:44:06Z

About me (short summary)

2014-03-20T02:35:19Z

About me (short summary)

2014-03-20T02:27:31Z

2014-03-20T01:47:44Z

Roadmap and deliverables

← Older revision		Revision as of 16:36, 20 March 2014
Line 241:		Line 241:

	[[File:Xfervm-after-caching-and-branch.png]]		[[File:Xfervm-after-caching-and-branch.png]]

			[[Category:GSoC 2014 Student proposals\|Asfrent]]

@@ Line 235: / Line 235: @@
 This is the first time when we can look at the profiler report and see that one of the slowest parts is the actual
-execution of the VM instructions:
+execution of the VM instructions. This report also gives us a hint on the interpreter design. It seems that a lot
+of time is being spent in converting strings to integers and vice versa. By looking at the code I discovered that
+the interpreter stack can only work with strings, so each time a integer is pushed or popped we need to apply
+these expensive conversions.
 [[File:Xfervm-after-caching-and-branch.png]]

@@ Line 105: / Line 105: @@
 ** ''extra: profile other parts of the Apertium, find easy to fix bottlenecks and fix them. Make Apertium FAST.''
-'''Deliverables (weeks 7-12)''': LLVM compiler for Apertium transfer, VM-transfer is faster than XML based code.
+'''Deliverables (weeks 7-12)''': LLVM compiler for Apertium transfer, VM-transfer is 2x faster than XML based code.
+* Project completed. :-)
+* Find something else to optimise and bring faster translation to the world!
 == High level optimisations ==

← Older revision		Revision as of 02:53, 20 March 2014
Line 235:		Line 235:

	[[File:Xfervm-after-caching-and-branch.png]]		[[File:Xfervm-after-caching-and-branch.png]]

	TODO more.

@@ Line 76: / Line 76: @@
 ** ''extra: make the project to compile with clang to LLVM and research LLVM code.''
 ** ''extra: improve documentation.''
+'''Deliverables (weeks 1-2)''': testing framework for VM-for-transfer, a test suite of 20 tests, tests passing.
 * '''weeks 3-4''' - work on the main algorithms and data structures that are present in the transfer code.
@@ Line 88: / Line 90: @@
 ** decide which instructions should be translated directly to LLVM code (push, pop, jnz, jz, call, test) and which will stay as C++ code.
 ** ''extra: improve documentation on each instruction.''
+'''Deliverables (weeks 3-6)''': VM-for-transfer is as fast as the XML tree walking transfer code.
 * '''weeks 7-8''' - first phase of porting the generated code to LLVM.
@@ Line 100: / Line 104: @@
 ** make the two work together.
 ** ''extra: profile other parts of the Apertium, find easy to fix bottlenecks and fix them. Make Apertium FAST.''
+'''Deliverables (weeks 7-12)''': LLVM compiler for Apertium transfer, VM-transfer is faster than XML based code.
 == High level optimisations ==

@@ Line 50: / Line 50: @@
 You can also find out more about me from my [http://asfrent.net/CV.pdf resume].
-[1] http://software.intel.com/fr-fr/articles/contest-winners-are-announced
-[2] http://intel-software-academic-program.com/tmp/certificates/4.pdf
 === About the project ===

@@ Line 39: / Line 39: @@
 [http://intel-software-academic-program.com/tmp/certificates/4.pdf 2]).
-I took a compilers course one year ago, when I had to write a compiler that translated code (see an example
+I took a compilers course one year ago, when I had to write a compiler that translated code to [http://llvm.org/ LLVM] (see an example
-[http://asfrent.net/viewgit/?a=viewblob&p=SQ2Compiler&h=3e1ef8fcf3645849e5066c08f3a1a3ffdc18ce31&hb=HEAD&f=testfiles/G_QuickSort/test.sq2 here])
+[http://asfrent.net/viewgit/?a=viewblob&p=SQ2Compiler&h=3e1ef8fcf3645849e5066c08f3a1a3ffdc18ce31&hb=HEAD&f=examples/qsort/qsort.sq2 here]
+which generated
-to LLVM. This course took me through the basics of grammars, parsing, abstract syntax trees, code generation and
+[http://asfrent.net/viewgit/?a=viewblob&p=SQ2Compiler&h=252ff92991fb53d337fdbd8d315c9babbebb8893&hb=HEAD&f=examples/qsort/qsort.lir this LLVM code]).
+This course took me through the basics of grammars, parsing, abstract syntax trees, code generation and
 optimisation. You can browse the source code [http://asfrent.net/viewgit/?a=tree&p=SQ2Compiler&hb=HEAD&f= here].

@@ Line 32: / Line 32: @@
 My name is Andrei Sfrent, I am studying for a Master's Degree in Machine Learning at the Imperial College
-and I am interested in working on one of the apertium projects in GSoC this summer.
+and I am interested in working on the VM-for-transfer Apertium project in GSoC this summer.
 I am proficient in C++ and I also have a good background in the software optimisation area (last year I
@@ Line 38: / Line 38: @@
 place worldwide [http://software.intel.com/fr-fr/articles/contest-winners-are-announced 1]
 [http://intel-software-academic-program.com/tmp/certificates/4.pdf 2]).
+I took a compilers course one year ago, when I had to write a compiler that translated code (see an example
+[http://asfrent.net/viewgit/?a=viewblob&p=SQ2Compiler&h=3e1ef8fcf3645849e5066c08f3a1a3ffdc18ce31&hb=HEAD&f=testfiles/G_QuickSort/test.sq2 here])
+to LLVM. This course took me through the basics of grammars, parsing, abstract syntax trees, code generation and
+optimisation. You can browse the source code [http://asfrent.net/viewgit/?a=tree&p=SQ2Compiler&hb=HEAD&f= here].
 I previously had two internships with Google and I had the chance to work with Map Reduce / Knowledge

@@ Line 193: / Line 193: @@
 ** transform the trie that is used for matching strings in the interpreter into a [http://en.wikipedia.org/wiki/Radix_tree Radix Tree] (a compressed version of the trie).
 ** write a memory allocator for the Radix tree in order to improve data locality (more cache hits).
-** refactor the code in order to eliminate as much overhead as possible that comes from copying structures around (such as vectors, strings). [already tried in the proof of concept].
+** ''extra: refactor the code in order to eliminate as much overhead as possible that comes from copying structures around (such as vectors, strings). [already tried in the proof of concept].''
 * '''weeks 5-6''' - interpreter analysis and optimisations.