What is LLVM? I Tell Huiqi About It

I've been knee deep in LLVM and in this video I talk to Huiqi about it. LLVM is a compiler infrastructure / framework. It contains a core library which you can use to build your own compiler.

Transcript

The following transcript was automatically generated by an algorithm.

00:00 : so which EW I want to tell you about
00:02 : LLVM today what is LLVM LLVM is is a
00:10 : compiler infrastructure what is a
00:14 : compiler oh let me share my screen
00:17 : that's a good idea
00:19 : so this is the LLVM web page is LLVM
00:23 : dork
00:23 : traditionally a compiler is what you
00:26 : would use in order to make a programming
00:28 : language work like a parse the language
00:34 : yes almost it parses the language and
00:38 : generates the AST but then there's more
00:41 : you and I have worked a little bit with
00:43 : AST we built some calculation evaluator
00:47 : libraries right dealt with AST if this
00:51 : is the compiler and you're feeding code
00:55 : into the compiler the first step of the
00:59 : compiler is the parser okay so you're
01:03 : taking this code and giving it to the
01:05 : parser the parser gives you an AST okay
01:09 : which stands for abstract syntax tree
01:13 : it's a representation of the syntax
01:16 : within the code that the user of the
01:19 : language wrote okay and to give a more
01:24 : concrete example I like to go to a
01:26 : scream as prima is a very good
01:31 : JavaScript parser that's also written in
01:34 : JavaScript itself so here you can use
01:38 : their demo sandbox thing you can write
01:42 : any JavaScript here and you get the ast
01:45 : of the that JavaScript on the right hand
01:48 : side here so for example this line of
01:52 : JavaScript that has a variable
01:53 : declaration equal to some calculation
01:56 : that is represented by this JavaScript
02:01 : structure which is the ast so right so
02:06 : this is this is a statement of what this
02:10 : code is at the top level what is it is a
02:12 : program
02:13 : okay a program has a body which consists
02:17 : of statements and its statement is
02:19 : represented as an object that has a type
02:22 : so this is a statement type of variable
02:27 : declaration and variable declaration in
02:30 : JavaScript can have multiple
02:31 : declarations on one line right they can
02:34 : just answer equals this comma another
02:37 : thing equals another thing but this one
02:39 : just has one of those so that's why do
02:43 : you have an array here and then the the
02:46 : variable declared her is a we're
02:50 : assigning to the identifier of answer
02:53 : and then the init is the initial value
02:56 : of that we're assigning to it and what's
02:59 : the initial value oh it's a binary
03:01 : expression it's multiplying 6 to 7 right
03:04 : so so this is the ast is what an ast
03:07 : looks like you also need yes yes except
03:16 : you have written a bunch of evaluators
03:19 : right for the ast so for binary
03:23 : expressions like + add subtract multiply
03:26 : and divide and more and function calls
03:29 : and whatnot but but for a compiler
03:32 : you don't write an evaluator as the next
03:35 : step for compiler you write a code
03:38 : generator okay what is a code generator
03:43 : good question the job of a code
03:46 : generator is to convert an ast - it's
03:50 : gonna give you machine code machine code
03:54 : what the heck is machine code machine
03:57 : code is code is binary code like all in
04:00 : ones and zeros the machine code is not
04:05 : that different conceptually from the
04:07 : code that you're writing it is still
04:09 : doing instructions line by line
04:12 : instruction by instruction but machine
04:15 : code is code that your CPU directly
04:18 : understands like your CPU is this chip
04:21 : inside your computer right and this the
04:24 : CPU usually is
04:27 : has the Intel label on it usually not
04:30 : necessarily there's other companies that
04:32 : make CPUs as well but this machine code
04:36 : is going to go into your CPU directly
04:39 : the CPU understands the instructions in
04:42 : the machine code it'll go through the
04:45 : code line by line this execute it and
04:47 : that will give you the result more or
04:50 : less it's slightly more complicated than
04:52 : that but more or less the CPU actually
04:55 : runs the code and give you the result so
04:57 : the code generator its job is to take
05:01 : this ast that's generated by the parsing
05:04 : your code and generate a code in a
05:07 : language that your CPU can directly
05:09 : understand basically once on ones and
05:15 : zeros but but to make it a little more
05:17 : concrete for you I can give you an idea
05:19 : of what machine code looks like so if
05:23 : you have a C program like you right now
05:27 : JavaScript is JavaScript is not a
05:29 : compiled language
05:31 : I'll talk maybe a little bit later about
05:33 : what how JavaScript works internally
05:36 : JavaScript is more of a dynamic
05:38 : programming language so it's slightly
05:41 : more complicated but let's talk about
05:44 : how C works because the architecture of
05:46 : C mirrors more this picture I'm drawing
05:50 : okay so if you take a C program like
05:54 : this there's a calculation prints out
05:57 : the answer you compile it using the C
06:01 : compiler then it will generate this
06:03 : binary file which cannot be displayed in
06:06 : this editor because it's binary but you
06:09 : can you can use a hex editor to look at
06:13 : it it has a hex editor that's called hex
06:16 : fiend you can download it this is free
06:18 : so if you use a hex editor to look at
06:21 : the binary code it kind of looks like
06:24 : this but I'm not gonna try to explain
06:26 : what's in here because I don't
06:28 : understand myself that much how to read
06:30 : the binary instructions instead I'll
06:33 : show you this format of code okay this
06:36 : is called assembly code okay
06:39 : this is a call assembly code other than
06:41 : zeros and ones this is the lowest lowest
06:45 : level programming language you're going
06:47 : to encounter assembly code is basically
06:50 : a representation of machine code that is
06:54 : just a little bit more human readable
06:59 : usually the instruction names are three
07:02 : or four characters long so this is move
07:05 : L and the nd all prints of the
07:09 : instruction is a list like this right
07:12 : you can think of these maybe as function
07:15 : calls but we call them instructions and
07:19 : so everything basically that the CPU can
07:23 : do there's a corresponding instruction
07:26 : and each instruction has like a
07:29 : prescribed list of arguments that it's
07:32 : able to accept and has to be a certain
07:35 : format for that instruction right when
07:37 : you translate this code into actually
07:40 : ones and zeros it actually translates
07:44 : the name of the instruction into a
07:46 : particular numeric code in binary and
07:50 : then then the values for all the
07:53 : operands it also translates those into
07:56 : numeric codes as well and then just line
08:00 : up the ones and zeros one after another
08:03 : so that's how it works and that there's
08:06 : a program that translates this program
08:08 : into ones and zeros and that program is
08:11 : called an assembler and back in the day
08:15 : when the programmers that were working
08:18 : for Steve Jobs and Bill Gates at the
08:20 : early computing days they'd ever have to
08:24 : write code that's exactly like this um
08:26 : that's basically what your code
08:28 : generator has to do is to generate the
08:30 : ones and zeros if you want to make life
08:33 : a little easier for you then you as you
08:35 : generate an assembly language program
08:38 : like this and then let the assembler
08:40 : take the next step and and turn it into
08:43 : the ones and zeros right so for example
08:46 : if you write in C does it goes to
08:49 : assembly language and then it goes to
08:52 : the
08:52 : once and zero so for it directly goes to
08:55 : ones and zeros I think it goes into it
08:58 : goes into this format first inside here
09:00 : takes it into assembly language first
09:04 : and then takes it into the machine code
09:07 : so we talked about what a compiler is
09:10 : how does LLVM fit into that picture
09:14 : well LLVM has a core which you can use
09:19 : some of the pieces from the LLVM core as
09:23 : building blocks of your own compiler so
09:27 : if you want to build your own
09:28 : programming language you can like use
09:31 : some of the pieces that LLVM provides
09:35 : and i'll explain which pieces there are
09:38 : actually it has all the pieces and clang
09:42 : clang is a c compiler that the LLVM
09:46 : project provides before LLVM came along
09:49 : there was already a very widely popular
09:52 : C compiler that is free and open source
09:56 : it was called GCC right good news C
09:59 : compiler and it was very good in this
10:02 : very popular many many open source
10:04 : project used it clang is a newcomer and
10:08 : its aim is to be better than GCC it's
10:13 : it's aiming to be amazingly fast
10:15 : compiles better error messages and good
10:21 : platform for providing source level
10:23 : tools as in but you can build developer
10:26 : tools on top of this stuff easily so
10:29 : clang is one of the components of the
10:31 : LLVM project which is an entire C
10:34 : compiler it has everything that my
10:37 : picture has clang clang does it for the
10:40 : C programming language the C++
10:43 : programming language and the Objective C
10:45 : programming language so this clay used
10:49 : pieces yes Quan uses LLVM core client in
10:54 : in a sense clang is like proof of
10:57 : concept of the LLVM Court is saying C we
11:01 : can make a really awesome compiler using
11:03 : the LLVM core and this
11:05 : the proof is clang okay so so clang is a
11:10 : big part of the oil game project and
11:13 : clang is so good and so fast the machine
11:15 : code it generates is very optimized as
11:18 : in when clang generates machine code
11:21 : that machine code usually runs quite a
11:24 : bit faster than the machine code that
11:26 : GCC generates at least that's what I
11:29 : heard I mean I I don't I don't write C
11:33 : code at my day job that's kind of what I
11:36 : heard from the wild we have this
11:39 : generator and we have the parser right
11:42 : they handshake wire this format called
11:46 : the ast that's their communication point
11:49 : so I've been working through this
11:50 : tutorial which is called my first
11:53 : language front end with LLVM tutorial
11:57 : and the code is in C++ so it's a bit of
12:02 : a learning curve for me but you might be
12:04 : wondering why do they call it front end
12:08 : when they say front end they're not
12:10 : talking about like a user interface for
12:12 : that right but in in compiler terms the
12:16 : front end is is this guy okay and this
12:20 : guy is the back end and the reason that
12:25 : this tutorial says I'm going to teach
12:28 : you how to build a program language
12:30 : front-end is because you're actually
12:32 : gonna outsource the backend to LLVM
12:35 : because one of the building blocks that
12:38 : the LLVM project provides is the backend
12:42 : that's the LLVM assembler or LLVM
12:46 : compiler so i believe that's called the
12:48 : LLC the LLVM compiler you need a handle
12:56 : and then a will just generate yeah yeah
13:00 : that that's a great question yes exactly
13:03 : so if you're gonna say oh we're just
13:06 : gonna use LLVM for our back-end so that
13:10 : we don't have a back-end it's sort of
13:11 : like when your web developers they're
13:14 : like I don't even want to build a back
13:17 : end I'm just gonna use firebase or
13:19 : is exactly what you're doing like I want
13:23 : to make a programming language I don't
13:24 : even want to write a back-end is that's
13:26 : too hard I don't want to know how
13:28 : machine code works I'm just gonna bring
13:31 : in the back end and that's LLVM compiler
13:34 : so what is this how do I talk to the
13:37 : LLVM back-end that's a good question the
13:40 : answer is we still have our parser that
13:44 : we need to build because that's the
13:46 : front end so to code again could goes in
13:49 : here we still have our AST of our
13:53 : language you have to do have to just
13:55 : define what your AST format is that's
13:58 : all you but and actually we still have
14:02 : to write a generator generator number
14:04 : one is gonna spit out some code in the
14:07 : format of LLVM IR IR IR means
14:14 : intermediate representation
14:16 : what does LLVM stand for by the way on
14:19 : their website they say LLVM does not
14:21 : stand for anything it's not an acronym
14:23 : for anything but there's a weird
14:27 : statement it's kind of like a I think an
14:30 : NPM they say NPM does not stand for no
14:34 : package manager that's not what it
14:36 : stands for and then they just gives you
14:39 : like a random acronym have you seen that
14:42 : on NPM the LLVM project they they take
14:48 : the same stance they're like no LLVM
14:51 : doesn't stand for anything but I
14:53 : actually think LLVM has it actually
14:56 : stands for something I think it stands
14:58 : for low level virtual machine even
15:01 : though the authors don't want to admit
15:03 : it ok so so basically you you still have
15:06 : to write this parser and you're gonna
15:08 : need to write a generator number one
15:10 : that generates a some some code in a
15:14 : language that LLVM can understand and
15:18 : once you do that you send it through the
15:21 : LLVM compiler which is called OLC or ll
15:26 : a yes I don't know which is the more
15:28 : correct term and then that gives you a
15:30 : machine code so that's how that works
15:33 : now what does this guy look like it's a
15:36 : full-blown language I'll show you a
15:39 : little bit of that to give you a taste
15:41 : but this is essentially what's happening
15:43 : and the reason I wrote generator number
15:46 : one here is in a sense this guy is
15:49 : generator number two right okay so we
15:52 : added another intermediate step you can
15:55 : look at assembly code as yet another
15:57 : intermediate step okay
16:00 : so there's many many intermediate steps
16:02 : why is this a good idea to use LLVM as
16:05 : the backend of your programming language
16:07 : the biggest benefit is probably that
16:10 : llvm has done the hard work of
16:12 : generating machine code for variety of
16:15 : brands of CPUs so if you just know how
16:20 : to generate code in the LLVM ir format
16:24 : now you can support Intel CPUs and arm
16:28 : CPUs and SPARC CPUs and any other CPUs
16:32 : that the LLVM project can support does
16:35 : they optimize your code for you or it
16:38 : would just do whatever you code this
16:41 : optimization internally yes it does do
16:45 : optimization internally and it that's
16:47 : that's a Forte of LLVM LLVM is really
16:52 : good at optimization and that's actually
16:55 : that's the second reason that you would
16:57 : want to use our VM instead of writing
17:00 : the code generator completely yourself

What is LLVM? I Tell Huiqi About It - Part 1

Transcript