Original subtitles

crisis averted yes the crisis has

been averted

hna you the crisis actually has not been

averted unfortunately the reason that

the stream is starting late today is

because well we're having a little bit

of a problem I wanted to get rid of my

taskbar I was talking about this on the

last on the last stream and I thought I

had seen some people say that you could

finally get rid of the taskbar again

used to be able to the taskbar on

Windows XP and of course when they

introduced their improvements to Windows

with Vista and 7 where they improved

everything they improved it so that it

no longer works so you can't actually

hide the taskbar it anymore basically

you only have this thing that basically

it always pops up if you go back up to

the edge of the screen and it's awful

and so you used to be able there was a

thing in here that said taskbar is

always on top or whatever and you could

uncheck that and that was awesome it was

like totally perfect and did exactly

what you want because then all of your

fullscreen windows would just be over it

which is what I wanted but I don't know

what to say so basically I was trying to

get rid of that I was looking on the web

for some of these things basically and

ya couldn't find it there's there's like

some tools you could course they got rid

of it so thoroughly that you have to

taskbar always on top Windows 7 and like

I was trying to find people who were

talking about this sort of thing and

there's like utilities you can install

where I tried installing one of them but

didn't actually work so basically like

you know when you're not using the track

bar Start menu blah blah blah right it

simulates to keep taskbar on top

unchecking that box this is what I

actually wanted like I've been surpassed

the simplest act taskbar Tweaker does

this job better than my code I recommend

you song said I installed this I do not

actually see any option for doing it so

I don't know if I'm just using the tool

wrong I haven't tried this one yet I

guess I could try it just to get a few

more viruses on my machine you know it's

always good but I don't actually know

yeah if it would actually work I could

try it and I don't I just understand why

this guy thinks that this tool does it

but he doesn't actually say how you

would do it you can see here like I've

got this thing installed if you open it

up right I don't even see an option for

it I don't even see I don't even see a

heading where the option would go but I

don't know like there just doesn't seem

to be I got I got nothing there doesn't

seem to be any any actual thing here to

click on you know I'm saying you know

what I'm saying

anyone no no one on the snow one on the

stream probably knows because I'm

guessing that people on the stream

probably have already moved to some

other thing I should probably try to

install a classic shell and see if that

has it or one of those other shell tools

and see if it you know can replace this

whole taskbar was something more usable

but yes so he says install seven-plus

taskbar speaker here is seven plus

taskbar Tweaker I do not actually see

any options that actually would work

dropping hovering thumbnails pin left

click other hi the start button that

actually does Head Start button doesn't

help show desktop button can go away

that doesn't help either although I'd be

fine with having that go I never used

that button display seconds is good but

doesn't help here reserved empty space

doesn't help doesn't do anything in our

case I just I don't know I guess I could

look at the help me the help have

something let's see

search top some of these tend office

with your tuned aspects of the tweeker

disabled tab topmost oh wait so it does

have a setting but how do you access the

setting options e^x

can use the eventual objects is it can

be open from the tray icon okay all

right all right let me see if I

right-click on the tray icon advanced

okay wait for it people this could be

exciting this could be exciting right

here apply okay and I think I also

accidentally shut this thing off so

let's see run that startup okay alright

done there we go

let's take a look that does not appear

to be disabling topmost to me but maybe

it did I don't know it doesn't look like

it let's take a look

my default password to type this option

allows change behavior always on top not

always on top but not over okay so I got

sent it to to RTFM I guess is the answer

there

try it one minute left in the stream can

we do it can we do it to return apply

okay oh my God look at it are you

looking at it look at how beautiful that

is

that is so excellent although it ruins

other top most windows like my top most

windows are now not topmost anymore so

close and yet so far

why would you taught me like that people

why well I guess we could also just

double check trying I could exit the

timer app and then start the timer app

again I suppose let's see if that helps

there we go so I guess it just it did it

to all top level windows it must have

messed up the Z order alright so what

day are we on I don't even know what is

it day 37 I want to say let's take a

look what day what actual day are we at

right now sorry we we may have actually

solved this problem so I'm pretty happy

about that but let's see what make sure

we're on the right day here video

archive day 37 was yesterday so we're on

day 38 today alright well that's pretty

exciting you have to admit that is is

pretty exciting I am relatively excited

about that hmm let's get all this stuff

straightened out again that means we no

longer have this ridiculous unused area

at the top of the screen we can actually

have everything go up to the height that

was supposed to be let's see does our

Emacs go to the hole look at that that's

fantastic well that is just great news I

am very happy about that that is super

cool you know that's the little things

you know I'm saying even if someone else

had to write them for for Microsoft at

least someone did there we go all right

I guess we can start the stream now

I hope everyone shared in that triumph I

don't know if they did or not can't see

what's going on on there yeah yeah I

think everyone's pretty except excited I

know I am so let's go ahead and get the

get the gloves on here and we will we

will go ahead and start the stream now

that we've had that that small victory I

guess I can close the old command prompt

from yesterday as well get back to our

standard standard screen and we can

actually get programming which seems

like a pretty good idea since this is

supposed to be handmade hero not

handmade fighting with the Windows user

interface get the taskbar out of the way

even though sometimes that is what

programming ends up being fighting with

Microsoft to get it out of your way is

often thing that does happen when you're

programming so you know it's not that

far off I suppose but let us go ahead

and return to the program hello everyone

welcome to day 38 of handmade hero the

show where we code a complete game from

scratch using nothing but our typing

little fingers no libraries no engines

so we can actually see the incomplete

process of making game and you can learn

how everything works right from where

the pixels are drawn on the screen all

the way up to how the game code operates

at a high level so today we happen to be

at the point where we are putting

sprites on screen and we are not doing

our full render yet because that's a

little bit more of an involved process

but basically we just wanted to have

something where we could easily put some

graphics up on the screen for our own

testing purposes as we are sort of going

through and figuring out what we want in

our engine architecture and one of the

big parts of that obviously is going to

be what we want for our rendering

architecture and it's kind of hard to

make those decisions without something

to work with so that's what we wrote

yesterday but we have a couple things

left that we need to do to it so that's

what we're gonna be doing today if you'd

like to follow along at home and you

went to hand made here at Oregon

pre-order the source code you should

have gotten a link in your email address

in your email inbox and that link will

allow you to download both the art

assets that we're using and it will also

allow you to download the source

so if you download that source code

today is day 38 so go ahead and unzip

day 37 source code out of the source

code zip file does zip files basically

has a ton of little zip files in it's

like a Russian nesting doll kind of

thing and there you go that will

basically give you the same source code

that I am starting with right now so

let's go ahead we've got a bunch of

stuff to do so let's get to it I'm gonna

open up the debugger here and just point

out where we were before if you remember

we were looking at putting the sprite up

on the screen but the colors were wrong

and so the question was what was going

on there and it turned out that since I

think it was my fault actually I had run

a little resizing tool on some of the I

had run image magic to resize some of

the pit maps and it rode them out in a

different byte order than the ones that

came straight out again and so basically

what we have to do today we figured out

this at the end of last stream but

basically the bitmap header has these

things at the bottom red green and blue

masks that we figured out and we've got

them laid out in here so we know where

they are and then in the files well that

basically tell us where the red green

and blue components go and it's I guess

assumed that the Alpha is just whatever

these aren't because there's obviously

an alpha component as well and there

isn't an alpha mask parameter as far as

I can tell although maybe there actually

is in an extended version of the file

format I'm not sure but well we were

looking on on the website of the

documentation it looks like it was only

red green and blue but doesn't really

matter because we can tell if the Alpha

is because it's where the red green and

blue isn't so what we need to do here is

we need to I don't know if you remember

this but basically we already were doing

a thing where we were swapping around

the order of the stuff that came in

because it wasn't the order that we

wanted so our pixels were packed with a

BGR which is this thing right here and

so this is actually not correct anymore

this comment we've learned is erroneous

and so basically what we need to do is

change this to say what's actually going

on which is the byte order memory is

determined by the header itself so we

have to read out the masks and convert

our pixels ourselves

right and so that's what we're gonna

have to do today so basically now we

assume that when we load one of these

BMP files we have no idea what order the

things are in and we basically have to

go ahead and figure out where they are

yeah we could for we now know we know

that basically none of the files that

we're saving you're gonna have anything

other than 8 bits per channel so we do

know that we're basically looking at an

oxf F kind of a mass but we just don't

know where it actually is in the 32 bits

we don't really know where it shows up

so what we need to do is we basically

need to get a shift value for each of

our components so if you think about it

we basically have a red shift not to be

confused with the kind of red shift that

they use in astronomy obviously that's a

different kind of red ship but we have a

red shift of green shift a blue shift

and an alpha shift right there we go and

so what we're gonna have to do is we

have to figure out what those are and

then we're gonna have to come in here

and actually get them out

right and so when we actually build our

thing we could make this a little simple

by taking the actual color value grab

that and then we could do it this way

just make it a little more easy for us

to write here we basically can figure

that we we know we need to put PAC

things such that the Alpha is on top

then the red then the green and the blue

right so what we need to do essentially

is we need to shift the red hmmm well

you know what unfortunately I don't know

that there's really any way we can do

the shift totally generically in with

only one ship I'd have to think about

that because really we need to shift it

they could have to shift in either

direction right where where the red is

in our source I don't know if you guys

are already on the same page here in

terms of what I want to do but basically

I've got something where I've did you

know I've got a 32-bit value right this

is my 32-bit value and essentially I've

got the eight here the this is 8 bits 8

bits 8 bits 8 bits right this is the low

this is the high and basically what I

need to do is I've got you know some set

of things I've got the reddest some

where the greenest number that blue is

somewhere whatever so I've got my ABCD

and I basically need to rearrange these

right in some arbitrary way like I have

no idea where they actually need to go

but I know that I'm going to determine

where they need to go so what I need to

be able to do is I need to be able to

mask it out and then shift it to the

right location but I don't actually know

which directions gonna have to go it may

be that like this was the green before

and the green now has to go here so the

shift would actually be a left shift

right it'd be getting higher but it

could be the other way around where it's

actually below and so it's actually

gonna be the right shift right and since

I don't actually know if there's really

any way to write that concisely what I

think I'm gonna have to do instead is

first shift it down and then shift it up

so I know that my shifts are always

basically are always going to be to the

right for the parametric amount so

basically my take the C and I'll shift

it down by the alpha shift value right

whatever that alpha shift value is I

will end it with with the mask so that I

know that I've basically now just got

the Alpha sitting in the low eight bits

and then I'll shift it up by 24 to put

it in the position where I actually want

my alpha to be right so that is

basically what I'm gonna do for each of

my values just so I can usually a lot of

times people call this like a Swizzle

I'm not really sure why they call it a

Swizzle but a lot of times they do so i

basically want to get the red the green

and the blue by shifting them down like

this masking them and then putting them

back into the place where i actually

want them to be like that and so really

all i need to do here is figure out what

the shift would be to get whatever the

value is down into the bottom down to

the bottom a bit problem is that only

have that i have the mask and i dont

have an alpha mask technically right

whew hello that was odd what I actually

have I've got the red mask right that

comes out of the header red mask I've

got the green mask I've got the blue

mask but the Alpha mask actually isn't

there as far as we know right I don't

actually have an alpha mask so what I'm

going to do is I'm going to assume that

the Alpha mask is just wherever there

wasn't a red a green or a blue so if I

or the other masks together basically

make a composite mass that would mask

for red green and blue all at the same

time

I'm gonna assume that the knot of that

if I flip all the bits in that then I'm

gonna get basically where the Alpha is

right because that'll basically give us

wherever the other ones aren't right and

I'll just I'll dummy up these values for

a moment just so I can show you what I'm

talking about for those of you who may

not be super comfortable with bit

operations yet

I think it's amusing actually but I

think someone was actually asking just

on like a scream like yesterday or the

day before whether we how if bit

operations normally get used and today

is like all about bid operations right

so so the answer is definitely yes

anyway so if we read out the red green

the blue mask you can see them here

right there's the red mask the green

mask and the blue mask that's saying

it's basically the Reds in the high

greens in the second down the Blues in

the second from the bottom and the and

the Alpha we don't know we don't get get

that right so what I'm going to do is

gonna order them all together which will

basically produce Oh X ffffff right and

then I'm going to not it so you can see

that basically what I get is I get what

the part is that isn't filled in in any

of these other ones right and so that's

what that that's what I want I just want

to fill in that extra value right and I

can do that a little more explicitly if

you guys want just to just to see what

I'm talking about here right I can do it

in two steps I wish they allowed you

step into an expression it'd be so much

nicer god what do all the things

debuggers could do and don't so there's

the alpha mask first before I've knotted

it right you can see basically if I

order them together I get the composite

of all of those masks basically and then

the not just this flips just flips what

the actual values are right so that's

what that's what I was doing there so

that gives you all the masks but now we

have a problem which is basically we

need to actually know essentially what

the lowest bit is that set in these

values right because we need to know how

far down to shift so we're essentially

want to count how many bits there are

that are 0 starting from the first bit

and going up to the very first bit

that's set right now what's interesting

about this is this is actually something

the processor can sometimes actually

just do for you it's called a bit scan

right and basically like well you know

what I can even probably

right here right so if I look in here

it's a bit stand forward there we go if

I take a look at this here you can look

at MSDN it's actually compiler intrinsic

the compiler can actually instructions

for it basically what it says it'll

search from the least significant bit

right which if you remember if I was Li

significance at the bottom right so this

is this is we're talking about the bit

here this is the most significant this

is the least significant bit this is the

most significant one right and just in

our 32-bit value so this is bit 31 and

this is bit zero okay so basically we do

bits can forward what we can do there is

we can say okay what I'd like you to do

is give me back whatever the position

was where the first bit was set right so

tell me which bit was the first one set

which is kind of what we want because up

here somewhere we've got eight bits set

right somewhere and we want to know how

far to shift them down so that they line

up right here so they start here and so

really what we're trying to do is we're

trying to do a bit scan to find where

that value is right so we can implement

this ourselves obviously right we could

go ahead and actually search for it so I

could do for example I could come in

here and say internal and I could

basically have I guess of bool 32 for

example if I guess I don't really know

because this is the way that this

processor I don't actually remember what

the how the actual instruction works but

it looks like the way this thing works

is it basically returns a boolean that

says whether or not it found a bit at

all and then it writes to an index to

tell you where where that index where

where we did actually find it so if we

were going to mirror the way that that

worked basically it's an unsigned care

that it returns and so I don't actually

know that I care about that so much I'm

going to do a UN 32 there and then it's

going to be an unsigned int that's the

that's the index as well so basically it

looks something like this

it's can forward right and if I go ahead

in here and say all right I want to

implement this myself right you can kind

of see how we would do it it's not

particularly complicated right we just

have a UN 32 well I I can actually do

Buhl can beat you here because it's

still 3 2 value and I can you know found

equals false I can then do a search like

literally with a for loop right which is

you know tests equals zero test is less

than 32 plus plus test right and then I

can do value and one shifted up by

whatever the test amount is and if that

turns out to be set right if value and

one should set up test turns out to be

set then I will write the value of the

bit into the result and otherwise I will

not and so that will basically give me

you know which whichever one it hits now

if I find one oh sorry I've I've

actually done that slightly backwards if

we're trying to look for the first bit

that's set not the first myth that's not

set so we actually as soon as we find

one that is set we want to stop looking

and if we didn't hit one we will write

to our index value right so if it's a

zero we count that as a as a as an index

no no okay nevermind no I'm correct I

was correct the first time sorry I got I

confused my own self there as soon as we

find those in set we're actually trying

to return that index so we just set the

index equal to the one we found and we

break out everything should be fine

right and and I guess we set found equal

to true here so that we know that we

actually found one we can return the

value properly now let me take a look

here because since I confused my own

self during the explanation let's

actually go ahead and make sure that we

actually did it correctly so this is

redshift

we passed the address of redshift and we

will basically look and see red found

equals that and we can assert red found

so and then we'll compile that oh right

we already have the redshift there we go

so and what was the problem cannot

convert argument one from int 32 oh yes

right sorry UN 32 there we go

so if I go ahead in here and actually

step into it what I should find is that

I pass in here I've got a value that I'm

checking for it's all the way at the top

so I'm expecting the value to basically

be 24 right so I come in here and I keep

shifting up test to see whether it

actually hits against the value it's not

gonna hit for the first 24 times right

bla bla bla well it's what 23 times I

should say turn off X vessels you can

see 13 testing testing testing testing

testing testing up to 20 22 23 and now

24 should hit it rights back to the

index so now the index gets 24 written

to it at least that's what should happen

find out oops start index right 24 we

say that we did find it and then we

return so there we go we've done the bit

scan forward and we say that red was

found the red shift is 24 which is

exactly what we wanted so that's the bit

scan operation but we're now doing it in

a very slow way so what we'd like to do

is we'd like to actually have the

processor do it and this is the perfect

time to talk a little about the

intrinsics things that I keep alluding

to but haven't actually done yet so

maybe today we will actually do

intrinsic so that we can call the bit

scan forward okay yeah also I should

point out for those of you who aren't

familiar with these there's a bunch of

these sorts of things right so you can

actually count the other way as well

right if you want to do from the from

the high downward to count like the

basically the zeroes that come at the

top you can call the reverse version so

so you know you've got there's yeah

there's a bunch of intrinsic that will

output those sort of things that the

that the CPU can do but that don't

really have any way of specifying them

in C per se so anyway let's go ahead and

get this working first with that but

then what I think we might do is just go

ahead and do that today so we can go

ahead and do move this into the

intrinsics and call the MSB C version

and there we go and we can leave this

you know sort of slow implementation as

the reference implementation as well

which will be nice because that would

basically mean that you know we wouldn't

have sort of a yeah we wouldn't have a

problem if people wanted to compel on an

architecture where the CPU can't do that

and they just need a slow version of it

to make it work that's good too because

remember this isn't really time critical

here so at least for this function it

wouldn't matter if you could you could

use the slowest possible thing and it

wouldn't really matter right so the

other thing too is I wonder if I should

fix the API a little bit because I guess

we can always wrap the intrinsic in our

own in line that just kind of would get

you know kind of felt out a little bit I

could make it so you don't have to pass

an address there basically so I could do

something more like this which is like a

bit scan result or something like that

which is you know whether or not it was

found and the new in 32 that was the

index right something like this and then

that could do something that looks more

like that in line bit scan results so

bit scan result and that would equal you

know nothing at first

having some typing problems today is

true and so then when we actually go

through we will then return the result

here and we will do results that index

equals tests and results ound equals

true like that and break out and then

we'll return the result that way and so

that would be a little handier than them

having to do sort of passing the address

nonsense which doesn't seem particularly

useful to me I don't I don't really

think there's any reason to structure it

like that I was just mirroring the way

they were doing it any other one so now

you can basically do something like this

where we basically have these so we'll

just get the bit scan results out of our

mass values directly like that and we'll

do the bit scan for words here and you

know the other thing I might do since

I'm kind of in charge now in Microsoft

bit scan for words kind of a weird term

maybe I would do something like find

lowest find least significant one

bit person least-significant set bit

something like that right because that

actually says what it is the bits came

forward well what do you mean by forward

at least now we know we're trying to

find the least significant one that's

set and so that actually seems maybe a

little more saying to me I don't know I

guess it would depend on endianness

things and blah blah blah but I think

that might be a little bit better

because at least I understand what that

is when I read it at least on one

architecture it makes sense so that

seems like priced a better idea so let's

go ahead and do that will now get each

of them out blue mask and alpha mask so

now we have found those and now we just

shift those by their indices like so

index index index index and we can also

do an assertion on each of them because

basically we're not supporting loading

anything that doesn't have all of these

panels at the moment maybe in the future

we would but for now we want at least

ensure that we don't actually

accidentally load something that doesn't

have one of these channels so let's take

a look at where that puts us on the

scale of fixing our color problems hey

it looks exactly right to me now the

only thing that we're not handling now

is of course the alpha blending around

those the edges and on the background

and that's the next thing we have to do

anyway so that seems like a pretty good

thing to do I'm gonna go ahead and say

though that since we're we've got a fair

bit of timer I'm just going to go ahead

and give the intrinsic thing a shot just

because it's about time that we started

talking about those sorts of things we

don't have to do very much it's not very

complicated it's a very simple thing to

do so it might be a good one to start

with so basically what I want to do in

here is I want to move this bit scan

result I want to move this out and I

want to move it to here right and so

what I want to do is be able to have

this dysfunction such that if we are on

a compiler that can do it really quickly

it will do it and and not have to go

through this for loop it'll instead just

tell the processor to basically tell the

processor to do it you know in a much

more expedient way if that makes sense

so really what I need to do here is I

need to be able to figure out whether or

not you know basically whether I'm on it

welcome

however I'm alone if that makes sense

and as you know as I said I talked about

quite a bit at the beginning you want to

keep all of your stuff segregated as

much as possible in terms of keeping

platform code out of

platform-independent code right so

you'll notice that we buried everything

inside win32 handmade and you'll notice

none of this stuff leaks out into our

main code all the win32 specific code is

all in here right but there are limits

to what we can do there if we don't want

to sacrifice performance like what I

could do here right is I could call a

function back in the platform layer that

did this operation right and that would

keep it and I could even pass in a

pointer to that function so it was

completely isolated I could do all kinds

of stuff but the problem is the only

reason why you ever care about intrinsic

sin the first place is for a performance

if we didn't care about performance we

just run this for loop right so the

intrinsics file the whole point of this

intrinsics file is going to be the this

is the place where we make a compromise

and say that this is the place where we

allowed to do platform specific stuff

that can be directly included right into

the code of platform nonspecific stuff

if that makes sense and again the only

reason we're doing that is for

performance if we didn't care about

performance at all if we literally just

had no performance concerns then there

would be no reason to do this there'd be

no point you would just always do things

the way we had been doing them but in

the case of intrinsics which are which

you're only using really when you're

caring about performance in the first

place this is where that sort of stuff

that comes in and so what you want to do

is you basically want to have it

isolated again you know even though it's

you're sort of saying it's okay to mix

things up a little bit here you still

want it as isolated as possible and so

you want to basically boil it down to

like a file or two that have compiler

switches in them and that bring in that

sort of non independent parts of the

code right so that's basically what

we're going to do here so what we're

going to do is basically say you know if

the compiler is msec

right then we know that we can issue

this thing because this

compiler happens to support this bits

can forward thing right and we want to

do from lowes LSB to MSB so we're

basically going to call this one we know

that in this case we're scanning a

32-bit value so basically what we're

gonna do here is we're going to to call

this bits can forward the result found

is going to be equal to that return

value we're going to pass the address of

the index that we wanted to be set and

it'll set it and then we're gonna pass

that value and so that's just gonna

allow it to do it right there we're

gonna make sure this is smart to in line

so hopefully if the compiler isn't dumb

it will go ahead and say oh yes we can

go ahead and in line this directly and

turn this into a instruction right so

that's about all we have to do there but

we have to do this compiler of

determination and so I'm gonna go back

to our handmade believe it's have a

platform right yeah in handmade platform

what I'm going to introduce at the top

kind of like you know we have this sort

of basic this is like the basic setup

for the the code this is what has to

happen in the code I'm going to do a

little thing here which basically says

you know compiler m SVC is zero and

maybe compiler LLVM because i know those

are the two things at least that way

we'll compile under right and what we

will do is we'll sort of do an if here

on some kind of thing basically if you

know if not compiler m SVC I should do

this way if n dev compiler MSD C and if

and if and F LLVM this you know I could

write these out a little bit more

cleanly so you can kind of see what

they're doing it's a short if end death

is like a shorthand do you find that so

what I'm going to do is if the compiler

variables and maybe I'll even make a

little section here like this and this

can be you know note AC pipes like this

Thank You Emacs for indenting me it's

finding the C++ bracket they're doing

the indentation but I didn't actually

want to do what are you gonna do

compilers so basically what I want to do

here is I want to basically say okay

if the build system whoever is building

us our batch file whatever if they want

to explicitly set what the compiler is

then we'll just let them set it so

basically we're only going to set these

if they're not already set then so we're

gonna and we're going to clear them to

zero or actually I guess what we could

do is if they're not already set right

then what we do is at the end here no I

still need to do this so we need to do

something like if the if the compiler is

not set at all right if we have no

compiler set whatsoever right this sort

of thing then inside here is where I'm

going to actually try to determine what

the compiler actually is so if neither

the compiler things have been set by

anyone then I'm going to go ahead and

try to see what the compiler actually is

and MSV see for example will

automatically define something that

tells us what it is I don't remember

it's like MS Sivir I mean I think it's

it's something like that but I want to

actually go see what it actually is

before I make that clock Proclamation

underscore I'm a sefer yeah almost

underscore MS Sivir so basically I want

to do that and say if that's true then

compiler MSB c equals one right and then

what we'll also do is we'll do an else

here and in this case we'll just assume

it's LLVM for now but we'll put it to do

here which is basically like more

compilers or something right maybe even

like that and so eventually we'll do

something a little smarter than that as

we go but right now we're only compiling

nm SVC so you know for now that should

be good enough so when we actually do

that let's take a look at what happens

here oh right we have to undefined

compiler it because we define it to zero

and you can't just change it

unfortunately you got a undef it first

like so and then when I got one one

missing Kendra so then there's a end if

there alright so now it's coming that

bits can forward is not defined and that

may be true let's see what has to be

defined in order for that to exist

somewhere let's

well it looks like we just need to do

that or do we actually have to pound

include in trindade I'm not actually

sure if we need to do that or not we

could take a look I guess the first

thing we could do is try this first and

see what it does so if we're defining

MSB see we can basically now have our

little section where we do the things

that we want to do if ms DC is here such

as that pragma in terms of function not

declared so okay so I guess we have to

do include this file as well which is

the intrinsics and then we'll see okay

bits can forward cannot convert argument

one from you at 32 start to unsigned

long star why not types pointing to or

unrelated is that really true why would

those be unrelated

I guess unsigned in and unsigned long or

technically different types oh the

mystery is of C++ I can program it for

so long and yet things like that always

still confused very much what actually

do you believe the technical difference

to be between those types and I have

absolutely no idea

oops put that in the wrong place there

we go so I believe that should probably

work let's find out and see if it does

and hey guess what it does and now let's

take a look in there and see what it

actually did differently now when we

step into it so if we step in here you

can see that this is actually using this

this line that we talked about and if we

go to the disassembly we can take a look

at what it does so essentially what

happens here is do you see this BSF

that's bit scan forward it's literally

one assembly language instruction that

does all the work of that for loop now

don't let that fool you because the

assembly language terms can still take a

while to execute sometimes but the point

here is basically that like that is so

much more concise certainly to the

processor and it's probably going to be

much faster than if we were to actually

have to micro to code the loop ourselves

because if this is executing something

in micro code that's still gonna be a

lot faster right so basically what

happens here in fact you can even watch

it work

what its gonna do here is going to move

the value this this thing right here

it's going to load that value into the

EAX register right and that's this

register right here it's a little

confusing because we're in x64 mode

which means that it uses our ax to

denote the ax register because it's 64

bits long but really what we're talking

about is we're talking about the bottom

portion of it here right we're talking

about the bottom 32 bit value so when we

do that you see how it loaded it in

there there it is it just loaded it

right there and now it's going to call

that BS F function on the value in the

register to load it out right it goes

ahead and does that and the destination

is also e ax so it's going to write the

result back in there right and you can

see that it written wrote the result

back in there and that's the result that

we were expecting right that's what we

wanted to actually have again it's in

hexadecimal which is why it looks like

an 18 instead of a 24 but you know

that's that's just well you know I can I

can here you go

so Oh X 18 right you can see that that's

just the hexadecimal encoding for 24 so

there it is and then that is the

entirety that right there did like the

entirety of that this loop right so

that's intrinsics in a nutshell they're

really pretty simple and basically what

we want to do is we want to do those for

anything where there is a convenience

CPU instructions so that we can access

it because we don't want to go around

especially when we start getting more

time critical code we don't want to go

around wasting a lot of that stuff that

the processor could do for us by Optus

getting it now it's possible that the

compiler might have actually been smart

enough sometimes depending on the

compiler to actually realize that this

is a bit scam forward but why make it do

that when we could just go ahead and say

this is f what we want right and so now

we've done that and we're in good shape

so now we're loading our stuff and we're

pretty happy but if you actually take a

look at what happens here you'll notice

that we would like to be able to

composite this guy on top of things and

right now we're not actually doing that

and so that brings up use of the Alpha

channel which we should talk about now

it's actually very very easy for us to

make this work but you know well you

know what I'll do first I'll basically

show you the really simple version first

and then we'll talk about it a little

bit so as you know the Alpha Channel

right if you remember how this is how

this works in the actual art tool let me

see if I can actually

how would I do this I'll go to a

hand-made data test and I will load up

that hero front head there we go edit

with GIMP so if we look at this here you

can see that basically you've got your

red green and blue color channels but

you also have a fourth channel and that

channel is called the Alpha Channel now

most people probably sort of you know

even if you don't know exactly what that

is

you know intuitively now what it is

because you've probably seen it in paint

programs like you've probably worked

with it before right and so if we take a

look at the channels dialog let's take a

look here inside GIMP there's a channels

dialog right which is this you can

actually see I can turn off the channels

and we can actually see the Alpha

Channel by itself right instead of the

other channels which is basically all

the Alpha channel is is it's basically

just something that says where the image

exists and where it doesn't

and it's got partial values around the

edges that basically say well the it was

transitioning at this point right it was

transitioning its anti-aliasing that

edge basically right it's giving a

partial value because the shape

obviously can't be rep is not perfectly

pixelated the shape was actually

supposed to be a smooth line that went

through there and so these alpha values

are basically there to kind of smooth

that out and to sort of give you that

gradual fall off so it's not a hard

cookie cutter edge right it just kind of

feathers the things into the background

for you that's the way the art was set

up to work so right now we're drawing

those red green and blue channels but

we're not doing anything with the Alpha

Channel and obviously since the Alpha

Channel is the only thing that tells us

where the image exists we need to start

doing something with the Alpha Channel

to display the person such that display

things on top of other things so that

like bottom layers show through upper

layers right and so inside our renderer

this is going to work well yeah I won't

say how it's going to work in the

renderer necessarily that's that's a

little premature so let's not say that

word let's say a different word what I'm

actually going to say it's not that what

I'm going to say is let's talk about one

way we could easily implement this

for our testing purposes for our testing

purposes what we can do is basically do

exactly what Photoshop or anything else

does where we basically just think about

things in terms of layers we could draw

something first we could draw down you

know our trees or whatever in the

background right and then we want to

stick a guy on top of that what we would

do is we'd essentially think of creating

a separate layer right which has our guy

on it and basically what we'll do is

when we draw that layer will only draw

in the places where the Alpha Channel

says there was actually a shape and we

won't touch what's underneath in the

places where it's not so well basically

do compositing very much the way you

would think about it in Photoshop where

you have one layer and you're basically

drawing layers on top of each other and

you just don't draw where the Alpha

Channel says there isn't anything and

that gives you this ER through so

basically like it's like taking pieces

of paper cutouts and stacking them on

top of each other wherever it was cut

out you could still see through to the

thing that was underneath right so that

alpha layer at the Alpha channel is

basically telling us where to do the

cutout it's basically what tells us how

to do the cutout so how do we actually

implement that right well while we are

drawing right we basically have

something you know we've got our trees

you know they're there and then we've

got our guy and he's got the Alpha

channel on and says where things exist

and basically it's zero everywhere where

he does not exist it's 255 everywhere

that he does and at the edges it's

somewhere between those two values right

so it's basically a brightness value you

might think of it between 0 and 25 that

says what you know how much he exists at

that particular point so what we would

like to do right is we'd like to think

about that essentially for now as an

opacity value basically to say how much

of a guy blends through and by the way

we could use this alpha channel as well

for partial transparency like if we have

something that's supposed to be sort of

ghost-like that we want to sort of only

partially blend on just like in

Photoshop how you could paint the Alpha

Channel to be some middle value like 128

or something that's partially

translucent we might want to do that too

so what we'd like to do is use this

value essentially as a way of blending

in between the two right so I don't want

to talk about that

yet because let's just do the simplest

possible thing let's do something that

just for starters

uses zero or 255 or sort of rather on or

off so we'll basically say if it's zero

we won't draw anything and if it's 255

or sorry if it's anything from 1 to 255

anything in between we will I guess

another thing we could do is say 128 is

the cutoff so basically anything less

than 128 we won't draw anything greater

than 128 we will draw right you can

pretty much see how we could do that

very simply right because we're

iterating over all the pixels of the guy

in our drawbitmap regime right we we

iterate over all the values here to do

this dest plus plus equal source plus of

us right so what we could do instead is

we could go ahead and change the way

this works we could do both we could

always do both advances right and we can

isolate just the copy like that and what

we could say is if that source value the

source part that's coming in extract the

alpha right we shift it down by 24 to

get just the alpha by itself if that

value is greater than 128 then we do the

copy otherwise we don't make sense right

so if we did that and we ran it we now

get much more of what we would expect to

get right we now get something which

doesn't draw the guy everywhere where he

had an alpha value less than 128 and it

does draw the guy everywhere there's a

value over 128 and all we did was just

add that one simple test right that's

all we were doing now what you'll notice

is it has a very hard edge around him

right if I was to do the the zoom in on

this right let's say we go in tikrit a--

there i don't actually know how creative

work sir can i do a paste as new image

paste into new image i don't know i

don't really know how to use this

program Oh does this what I want I don't

even know all right yeah I guess it is

it's doing what I wanted okay good

enough so if you're looking here you can

see you know there's a lot of

anti-aliasing this is very smooth

graphics there's a lot of anti aliasing

everywhere in here there's a lot of kind

of you know blurring of those edges to

make them a little smoother that's kind

of the way the art works but when we get

to this guy you'll notice there's

no smooth there's no blending there at

all so it's a very hard pixel edge along

the guy and the reason for that again is

because we're not using any partial

alpha values at all right we're not

doing anything to try and use those

values that are in between 0 and 255

we're just making a hard cutoff now what

this is often called in fact you'd see

this called this in 3d graphics lot it's

called alpha test is what this is called

which is basically a hard clamp limit

which basically just looks at the Alpha

value and says I'm either gonna do an

operation or not based on whether the

Alpha exceeds a threshold or not and

that is exactly what we implemented this

is basically alpha test right here but

what we'd like to do is something a

little bit better than alpha test we'd

rather do something that's actually some

kind of an alpha blend a blend that uses

the alpha value to compute the the

resulting color that we'll use what's

underneath

to actually and show through

what's underneath in case the alpha

value is not 255 that makes sense now

we are going to do this in the slowest

and dumbest possible way because I'd

like to make show the math being

explicit and so this will not be fast by

any stretch of the imagination it will

be very slow but again we are not

writing the render at the moment what

we're trying to do is write stuff that

we can use and so what I'd rather do is

just well you know what no I'm not going

to talk about how to do any of the fast

stuff we're just going to do it in the

slowest possible way so what we'd like

to do now is we'd like to go ahead and

and see about how to actually do that

alpha blend now I probably shouldn't

spend too much time belabor this because

there's actually an entire video in fact

of me explaining interpolation and I

don't know whether we want to rehash

that entirely on the stream or just did

that just crash I think frita just

totally crashed on me it did didn't it

uh yes it did

creative just totally just crashed it

just full-on it full-on went away

completely well what are you gonna do

watch it back up come on you can do it

it was working so well there for a while

see I tried to make a new image I guess

the thing to remember has never tried to

make a new image people that's the rules

you don't ever want to make a new image

all right that's fine make the image

please there we go hello all right okay

so here we go so what we need to

essentially do here right is I want to

make something that allows us to

smoothly blend between the two values so

we have a background value right I don't

know why that's always the hand shape

just is I guess the hand was what they

thought was good there so I have a

background color that's coming in right

I've got a hand apparently that's

drawing this now I've got a background

color it's funny to have it be a hand

like that that helped it all nope that

doesn't help at all welcome to the

mysteries of the universe ladies and

gentlemen I've got a background pixel so

this is one pixel right I've got a

background pixel and I've got a

foreground pixel right that this is the

sprogis I'll call this the sprite pixel

the bitmap pixel that we're drawing

right so this is whatever is in the

buffer and this is whatever is in the

thing we're putting on top right and so

in this code here we've got a source and

a desk right so basically this is the

desk this is the thing we're writing to

and we've got the source here this is

the source this is the thing that we're

trying to draw down this is the guy's

head right so this is the trees pixels

this is the the head pixels right so now

if we do this what we need to do is this

is some color

you know let's call that color a and

this is another color we'll call that

color B Wow

this is awful like I can't even click

down anymore ah I tell you computers

these days nothing ever works anymore

okay so I've got these two colors a and

B and what I want to do is I want to

produce a third color out of these right

I want to produce a C that's some blend

between a and B and what I wanted to do

is I want it to

blend based on that alpha channel right

I want the alpha channel to blend it

such that when it's zero I get a and

when it's 255 right I get B does that

make sense so when I actually go to do

this what I need to do is figure out

some some mathematical way of getting

from A to B smoothly based on another

value right I need to figure out some

way to pick because if you think about

the number line right you know I've got

255 at the high end for each Channel and

I've got 0 at the low end right

essentially what I'm doing here is I'm

saying I've got two points on this and I

don't really know where they are in fact

they don't even have to be in a Pickler

order a could be higher than B or vice

versa or whatever right so this is let's

say and this is when I say colors

obviously I mean one channel so we could

think of this as just the red Channel

right or the green channel or the blue

channel just one of them so when I'm

going to do this I basically have

whatever was in the background whatever

that channels color was right that's

here and then I have whatever the the

new color is that I'm blinging on top

whatever that the sprite is whatever is

in the guy and that's here and what I

need to do is figure out a way to

produce all of these values in between

here right I need to be able to produce

all of those values based on what the

Alpha channel is based on what the Alpha

Channel says and if the Alpha Channel

Zero that I'm going to leave it where it

was so I'm not gonna touch the

background at all at the Alpha Channel

typically five I want to drive it all

the way to be right so essentially the

way you could think about this and the

way that I usually like to sort of

explain it to begin with is that you can

think of it as being based around one of

the numbers we can think of it as being

based around the number that's that's

zero primal so the number where we have

zero yields that value which in this

case is the background color right we

can think of starting there and what we

want to do is you want to start there

and we want to sort of go some distance

towards be where that distance is going

to be 255 if we went all the way right

and zero if we didn't so how do we

produce what that with that actual

distance is right what's that value well

if you know rudimentary math you already

know what that is

it's just B minus a right B minus a is

if you will almost and we haven't talked

about vectors yet but it is essentially

a one-dimensional vector but it's

basically the Delta right it's the it's

the difference between a and B such that

if I were to take a star today and add

back in some you know this this full

portion I would get B and it's pretty

trivial to see why algebra wise right a

plus B minus a well the to the a

positive and a negative cancel and you

just are left with B which is the value

that I said we would get right so this

is basically the core the core of the

thing that we're going to use to

determine how to plan something right

because basically all I really need to

do is take this expression the a plus

the B minus a right and say as if I can

figure out some way to only add part of

this in there right all I really need to

do is stick some kind of a percentage in

here right some kind of a value that's

like a percentage from zero to a hundred

percent right if I could just multiply

this by some percentage then I could go

any percentage of the way from A to B

just by changing what P is right a plus

some percent of the distance from A to B

is going to give me exactly what I want

right so if we go ahead and build on

that assumption right if I want to say

I've got a plus some percentage here

right and also by convention this is

often called T I'll just point that out

to you we can use that cuz you'll see it

use us this way for a often but some

percentage of the distance from A to B

this would basically give me that value

that I wanted this would be that result

in color this would be the color that I

wanted out right so really all I have to

do is figure out how to do that

now what you'll also notice is I can

rearrange this expression in a couple of

different ways if you know algebra which

I'm hoping everyone on the stream does

because well honestly I've never thought

about how to teach algebra before so I

don't suspect that would give a very

good explanation but assuming you know

algebra then you also know that I can

pretty trivially use the distributive

property here to distribute T into this

expression and I would end up with these

terms right a plus TB minus T a right

and I can rearrange those terms any way

I want to so I could group them this way

right I could group them here by just

removing this term this term here moves

here right and this term stays where it

is this term slides there if I do that

rearrangement what you can see here is

these two front terms both have an A in

it so I can pull out the 1 minus T there

times a plus TB okay and now you have

what's usually the more canonical

version of what's called a linear blend

right this is usually called a linear

blend and what it is in either form

doesn't matter which one you find more

attractive they're just different ways

of computing the exact same thing right

algebra it's the same thing all they are

are a way of taking one thing in that's

just you know your starting value one

thing in that your ending value and

basically being able to produce anything

in between them by plugging in this

percentage value here but we haven't

actually talked about how to actually

make this percentage value work in

practice I just kind of magically waved

my hand and said we had a percentage

value here so we now need to talk a

little bit about just how that guy works

right well literally in the math sense

and if you're doing this entirely in

floating point math it could literally

just be a percentage right you know how

to translate percentages again if you've

had high school math or whatever you

know how to track translate percentages

into floating point math 100 percent

right is 1.0 50 percent is 0.5 right

zero percent is zero right so the number

is between zero and one all inclusive in

there are just those four

so really in this expression if T were

to range between zero and one including

all the fractional values in between it

then no matter what a and B are I don't

care what they are they could be colors

that could be anything positions in

space whatever doesn't matter this will

give us a blend between them right this

will allow us to move smoothly between

the two of them okay and so if you're

interested about in more about this sort

of stuff as well I would also recommend

if you do want to go to YouTube on the

hand mate in fact YouTube Casey Mira

Torre I think is what you would type in

here let's see here

Casey Mira Tory on YouTube there we go

you can actually see in my videos page

there is a witness Wednesday part 10

interpolation video and that is a full

on 50 minute explanation of

interpolation and how it works in a lot

of different ways so basically if you

want to watch that for more information

you can do that as well so where are we

at here we've only got seven minutes

left so let's go ahead and slam in the

floating-point version of this so that

we can actually see it in action and

then maybe we'll go ahead and fix it to

the fixed point version later but

basically you can see exactly what I'm

talking about here right so if I was to

go in here and write the floating-point

version of this let's go ahead and see

what we get

right what I would get first of all is I

gotta extract my read write and so my

read channel is going to basically be

shifted down by by 16 whatever's in that

source rank and then masked off so there

we go

and convert it to a float so that's a

number between 0 and 255 there we go

doop doop so then we have green I guess

I could just do our G and B for short

all right the green is the 8 and that's

0 then the alpha channel right I need I

just said I need it to be a number

between 0 and 1 well this will be a

number between 0 and 255 right so if I

want to turn a number between 0 to 255

into a number between 0 and 1 I need to

divide by that range but I need to

divide by 255 so that way 0 will map

0-255 will map-21 and everything in

between will map somewhere in between

right so now I've got an Arg and a bead

that are actually in floating-point now

I need to do the exact equation I just

said I need to create my new read so

I'll say that my read is now going to be

equal to oh I forgot something

I need the desk as well so I have my ice

or SARS and my Death Star's right I've

got I've got to pull them out of here

so that I can blend with them right so I

pull out the Death Star of the desk G

the desk be and I've got my sorcerer and

source' G and and to be fair I guess I

don't actually even care what these are

we don't even care if these are our G

and B we just care that their channels 0

1 & 2 right we don't even care so we

could be wrong about these shifts and it

wouldn't matter as long as we do them

all so anyway now I'm going to actually

produce my result red and my result red

is just going to be the exact equation

but I just specified it's going to be a

plus T a minus B if I wanted to do it

that way or I can do it this way over

here which is 1 minus T plus TB

right so I can do it this way right

times a plus T times B so we'll leave it

that way for now and we'll basically

just say okay if I do one

what's my T well my T is the Alpha value

that's the thing that's that's actually

causing it to work there and then what's

my a might be my a is going to be my

destination red times my source red

because I'm trying to get to the source

red right so that source red is the B

color and the the back and color is the

a color right all right I believe that's

what we said it was and so off we go now

we'll go ahead in and just do that for

all of them I've got to move that up a

bit so it's not obscured by the

five-minute warning the dreaded

five-minute warning be a B be like that

and then once we're done we have to then

reassemble the resulting value back into

something that we can actually use so

what I have to do now is actually do

sort of that that build back again for

the thing that we're going to actually

write to death so we have to actually

write something to death and what we're

going to write to death is going to be

our board with our G or with our B but

we have to now turn them back into

actual actual rounded values right so we

have to actually round the values each

time now so like that I suppose we could

just go ahead since we know these are

always positive we could also do this as

a truncation which would be a little

cleaner right I could do it this way

where we just actually truncate the

values plus the rounding amount like so

which I think you know probably is a

little stainer here since we are

actually doing this per pixel per pixel

things are very expensive so we do have

to watch out even in our test code so

that we won't get too slow and not be

able for not for us to use it but we may

replace this we probably don't want to

be doing this necessarily in floating

point math the way that we're doing it

here right now anyway but anyway point

being so here we go we're gonna actually

take these values and pack them back

into the places that they normally go

right like so we basically unpack them

and pack them oops

8:16 like that the argot shifted back up

thank God to put everything in there I

don't know I wrote that in pretty

quickly there we'll have to step through

this and see if it works

but I believe if we typed it correctly

we should have I'm not gonna open I'm

not gonna paste things into crea anymore

lesson learned we're gonna paste things

into GIMP from now on that way at the

crashes we can still draw in the

background so now let's take a look at

what happened can you see and hear how

that how that looks now we're you've

basically got you know that kind of

smooth that nice smooth edge now that

we're being the alpha Channel so that's

a much nicer comp right that's just a

much nicer thing to look at right and

yeah I mean that's that's alpha blending

in a nutshell right and so what kind of

alpha blending actually is this this is

a straight linear blend between the two

there's other like I said before on the

previous stream there's lots of

different things you could mean we say

alpha Bunny it depends on what that

blend equation is right so basically

this guy here this is our blend equation

that we're using there are other blend

equations so what I'll also say here is

to do KC

someday we need to talk about

premultiplied alpha because we are not

talking about that at the moment but now

is not the time to start that discussion

suffice to say this is not be multiplied

alpha but I'll just leave that for a

little thing that we can talk about

later on because at the moment it's

really not something that you need to

think about whether it's premultiplied

or or not but it is something that we

will have to think about a little later

on because it affects the math in very

important ways and we will need to

switch to using p mult multiplied alpha

for our final assets and so on so since

we are basically out of time here I

believe yes we have one minute left I'm

gonna call that done for the day we have

our alpha pudding working oh you know

what I'm not gonna quite call it done

for the day I might do one more thing

although I think this this actually will

cause us to have a couple bugs but I'm

gonna go ahead and make our our hero

head actually draw at the location of

the player now yeah so there we go so

there's our hero head very exciting he's

not actually in the right position but

at least he moves around there with the

guy very much fun yay yay for alpha

blending we're pretty close now so I

think basically tomorrow we'll have a

little guy walking around instead of our

rectangle because we're pretty close

all right everyone that is it for now

the Q & A has begun remember I only

start looking for Q colons when we start

the Q&A so please if you have a question

that you like answered now is the time

to post a chip to the twitch chat put a

Q : in front of it and please keep the

questions here in the post stream to

stuff that's actually out the code that

we wrote today or in previous days if

it's something that's off topic and

about something else please come before

our stream sometime oftentimes I try to

be on the stream about 15 minutes early

to answer just any off-topic questions

that you want answered and so I do

answer those that it's not on the po

stream because this is the one that kind

of goes into the recorded handmade hero

stream for people to learn on later and

I don't want to have lots of off-topic

stuff that they have to skip through

okay tiny typo in function name finally

significant set bit is missing an eye

okay thank you gasps blizzard let's take

a look at that intrinsic find least

significant there we go

that's better let's go ahead and click

that finally sick okay thank you very

much that is a definitely a typo yes

that's true

so mosaic Oh was talking about asserting

the compression mode is true we only

load one type of compression now which

is compression mode three so while we

are asserting the masks we should

probably also go ahead and assert that

the header compression mode is still in

that three mode because we don't load

any of the other compression modes at

present so we'd want to know if we got a

file that didn't conform to that

compression mode so that we could be

warned about it it's a good point so

also someone was saying we don't need

that intrinsic definition anymore well

we do actually need so I that's not

actually true so um I'll tell you what

the person saying and then I want to

make sure that is actually the case that

I'm not sure it is

so we'll yeah we'll talk about that so

basically what the person was saying

they were saying that the you don't need

this pragma intrinsic if you include

intrinsic but I'm not 100% certain if

that's the case let's find out so let's

say we do that right

now let's go take a look at the code

that gets generated just so I can see if

it actually still does that is intrinsic

it might but I just want to make sure

because I don't actually know sometimes

it doesn't for other functions I'm

pretty sure it doesn't but for this

function because they may not have done

a separate implementation of it then it

might not so that's that's definitely

possible all right what am I looking for

here I'm looking for a load bitmap right

load BMP so basically in here what we

want to do is take a look at this and

see what it generated go to disassembly

nope so does it okay so that's totally

fine so then it is fine without it

sometimes the reason that I don't that I

sometimes if you if you have if you

don't put the private intrinsic it'll

only use it in optimized builds or

whatever and so I wanted to make sure it

was I we there's no reason for us not to

always use it but as long as it always

does do that then that's totally fine

would it be worthwhile to avoid

converting to float when doing the alpha

blend yeah well so this is one of those

things that basically what we have to

say is who knows so technically I think

it probably would be in this particular

case right but we're so far away from

actually doing performance oriented

stuff that who knows it may be that it

really just doesn't matter for our

purposes at all in this particular case

but for something this trivial you can

do it entirely without moving to float

but you know first of all you'd never do

this you'd load the whole thing as a

Sindhi vector instead of doing the

channels individually like this is just

not we aren't thinking about performance

right now so thinking about whether you

could convert to float or not is just

not that relevant because there's so

many things that are wrong with this and

from a performance standpoint but it's

it's just not really that useful to

think about it I mean it's an I I

actually would suspect probably that at

the end of the day we won't even be

storing things in this format we won't

even have a RGB ease we might have like

a buffer of A's a buffer of ours and a

buffer of G's and so on right so things

might it's it's it's worth being aware

of how your code is performing but it's

also worth knowing when you don't really

care about some differences and in that

case I think that's a difference we

probably wouldn't care about that much

it might be interesting to do it just to

show how you would do it without

converting it to float but yeah do I

expect this to be a major performance

thing no not really because

already performing so poorly probably

that I'm not sure how much worse this

would really make it you know it's hard

to say why is the windows logo still

visit I have no idea because I don't

know if you caught at the beginning of

the stream I was doing that that tweak

thing that seven-plus task for our

Tweaker thing and it's the thing that

that's causing the windows to not be on

top it might be it might be possible

that that's a thing that was intentional

so let me see if it's in the Advanced

Options I'm not sure let's see let's see

why that's doing that you know that's in

the the help over here

so see option Z X and fix is about when

the classic theme is used clicking on

the string crooner no customize

horizontal paging it's a showing trick

lock allows to customize your desktop

button its rest reverse the direction

yeah I don't I don't actually know if

there's a thing I'll look for it later

but I don't see unfortunate I don't see

a way to get it to stop doing that

so I think that's might just happen for

a while until we figure out how to get

rid of it

let's see are there other ways to use

out or are there other ways to alpha

blend and should you be taking into

account which colors are easier to see

like green appears to be brighter than

red so assuming that I understand your

question correctly I think what you're

asking is are there other ways to blend

between two colors such that you're not

just using per channel so the first

thing I will say is that yes there are

many ways you could consider blending

between two things this is a linear

blend for starters and it's linear in

whatever space the colors are in which

themselves may not be linear so we

haven't we'll talk about this a lot we

get to the renderer but basically what

you have to understand is that the

colors that are displayed out of the

monitor are actually not linear so

basically you know we think about these

colors and we're doing this math the

math actually is assuming that that

colors are are linear a lot of times but

actually the way things work is there's

actually a curve so basically you know

there's a color value that you set so

there's basically a value from 0 to 255

that we are setting and we're calling

that the color right so what happens

though is that when that value is given

over to the display it actually goes

through a number of mappings potentially

once potentially on the graphics card

and once potentially by the monitor that

actually changes the perceived

brightness of that to not be linear so

then there's brightness here right and

so what actually happens is there some

curve that happens here right that you

know there's there's some curve you can

imagine where as you change

the actual numerical value of the color

instead of it getting you know if you

change it from between zero and ten the

difference in brightness may be much

more than the same distance right 240 to

250 or something might be a lot less of

a difference in brightness right let's

say or vice versa depending on how those

curves are set I think they're usually

set in this way but I can't quite

remember their exponentiated right yeah

so well exponentially they went through

this I don't know let's take a look

so just to just to give you the it's

been a lot this is the first graphics

programming I've done in a long time

actually so you know a lot of this stuff

is rusty for me to be honest with you so

let's see srgb gama curve can you look

at that can we get that up here

somewhere gamma curve there we go so

linear space and there's well that's not

what I actually wanted I wanted one on

one side and one with the other this is

not giving me the values I want I just

want brightness on one side and value on

the other I'm not looking for gamma

correction I'm actually looking for just

that doesn't anyone bake it down I just

want someone who's baked it down for me

that's really all I want well let's take

a look and see if this has it so srgb

that's actually got the full Clark you

here we go that's RGB intensities versus

HIV numerical values right so basically

let's see which was red so srgb

numerical values are red curve these are

the numerical values and I assume that's

the intensity showing exact low a

compressed value observe as linear

intensity or at a linear intensity as

this curve is linear so the gamma is 1

which curve are we talking about all

right this is not what I wanted I just

want why can't I get what I want maybe

I'm misinterpreting these diagrams but I

literally just want what I just drew

I wanted something that's the value that

you put into the buffer that's coming

out and how bright it appears on the

monitor that's all I wanted I don't know

why I can't get I just want one graph

that's just got that curve on it so that

I could show you but I don't know okay

so we can't get that curve maybe we'll

look up that curve I'll look up that

curve outside the stream so I won't have

to waste your time watching me try to

read through documentation but basically

point being it's nonlinear so it's

entirely possible that that is a bad

thing for your blend right you may be

doing alpha blending and you want this

this alpha was expecting it to be you

know let's say a smooth linear

transition so as you change alpha you

get a smooth step all along the way but

instead you're actually if you're doing

it in this exponentiated space

unbeknownst to you a little bit of the

blend actually goes a long way in terms

of what the perceived blend towards the

new color is things like that so there's

alpha corrected versions you can do some

of these things there's other blend

equations that do other operations that

aren't designed to basically map one

thing on to the other there's pre

multiplied versus not putting on

premultiplied alpha which is basically

getting rid of this 1.0 minus a term and

baking it into the to the I'm sorry gets

rid of this a term and bakes it into the

sr term which is another thing

altogether so there is a lot of stuff

that goes on here I don't know if that's

exactly what you were asking but that's

how that works

whether you take into account the fact

that channels work differently I don't

think so because the images themselves

have already had that taken into account

and you need to smoothly blend between

them so you could blend in a different

color space like you could blend a set

of blending in RGB you could blend in

HSV or you could blend in log or you

could do other things like this I don't

know whether that's really such a good

idea though because then what you'll get

is you're not going to get smooth

changes in the colors of the pixel

elements on the screen since they are RG

and B so I think that's probably where

you want to be for the operation that

we're doing but if you were trying to

alpha blend if you're trying to do an

alpha line that was trying to do

something else like smoothly blend hues

of things together then you might want

to blend in a different

space one that wasn't based on our gene

be one that had different properties I

don't know sorry I don't have a good

answer to that question really I don't

know yeah I don't think I answered that

very well but there are different ways

of blending things but I don't know if

they are exactly the way that maybe

you're thinking about it are we at all

concerned with the destinations alpha or

blending with a transparent destination

so well we are and we aren't basically

we're not concerned the reason we're not

concerned with the destinations alpha

channel at the moment is a because we

don't so two reasons first of all we

never really right the destinations

alpha Channel

because we don't care about it at all

right so you can see in this expression

we don't even write it it's always zero

right the destination just always gonna

have a zero alpha channel always now

that doesn't mean you have to do that

you can do rendering that uses the

destination alpha Channel to store

important information that it will use

and this is basically something that we

may explore later when we're actually

doing our actual renderer because

there's reasons why you might want to

use that summation alpha to do things

but right now we don't have a

destination alpha however what I do want

to point out is this doesn't have

anything to do with blending with other

transparent things because that actually

just always works because basically

since we are since we're basically since

we are always drawing things in

Photoshop style like I was saying before

where we're going to layer things down

such that we start with the bottom layer

right which has just whatever our base

code is right at some grass or something

right and then we're gonna like right on

to that a tree or whatever and then

we're going to write on top of that the

guy that's in front of the tree since we

know we're always going to be doing that

photoshop on top up on top up on top up

on top of compositing we will always

work because just like Photoshop layers

always stack on top of each other

properly we will always that on top of

each other properly if we were to draw

out of order right so in other words if

we're trying to draw stuff stacked up

right I'm gonna try and do a really bad

stacking diagram here so if this is our

background and this is sprite one

and this is sprite two if we're always

drawing in this order right from the

bottom most thing to the top then

stamping down we know that all we're

doing is writing in the colors of this

thing where they exist and blending them

with the background whatever it was so

another transparent thing that comes on

top doesn't have to care that the

destination buffer is actually already

the combination of two things

it doesn't care it'll just stack on top

of it properly so it doesn't matter to

this guy whether this guy had a

transparent portion or not because

that's already been resolved by this guy

by this operation right by the time this

guy writes to it it's as if they're only

ever was one buffer because all the

things that came before and have already

been slammed in there right if we were

going out of order if we were gonna try

and render the top thing like the this

bottom thing then this top thing then

the middle thing and this has to insert

between them then that's the point where

you start worrying about what the what

was actually written into the buffer

before right and that's all a huge mess

so basically what we want to do as much

as possible since we are a 2-d game we

have the luxury of probably being able

to do always to always know the order in

which we're drawing things and that will

let us do a bunch of things that are

actually extremely difficult in fact

sometimes impossible in 3d renderers

which is basically order independent

transparency is what they have to do and

they have to do all kinds of work to

make that stuff happen properly it's

really a nightmare for them it's why

games really hate stuff like looking

through several sets of windows at some

smokey particles right having lots of

translucence is actually really hard in

renderers for a number of reasons so

they try to avoid those sorts of things

or do them with sort of what they'll

usually do is they'll they'll actually

break apart the pieces of the scene that

are transparent partially and they'll

sort they actually will sort those

they'll draw everything that's not

transparent first and then will actually

sort the transparent pieces and draw

those in passes just like we're gonna do

for our renderer but we have the luxury

of being able to sort all of our stuff

so everything will just work for free in

that way

but they don't have the luxury of that

because they have a much higher

primitive account they have a much

higher primitive per pixel count than we

do and so they don't they can't get away

with that right so yeah well the final

blend favored the source over the

destination or is linear good enough I'm

not sure what that means we're really

not favoring anybody we're just using

the Alpha Channel to tell us how much of

the foreground should come in at that

point right how much of the source

should be used so I'm not sure can you

restate that question let's see how

would you deal with other picture

formats such as PNG JPEG GIF etc well we

just write a loader for whatever the

other format was or if you're lazy and

don't want to write another loader you

could just run a conversion tool and

then load them that way right you could

just run some freely available

conversion to a batch conversion tool

like image magic to batch convert your

stuff from whatever your source format

is into BMP so you could either load

them yourself review that or like I said

I'm hand made hero we write everything

from scratch but you don't have to write

everything from scratch you know there's

no reason that it's you can't use a

image loading library if you'd like to

and there are several that are available

there's thing called free image that

loads images you know your mileage may

vary but there's libraries you can use

so it's entirely up to you yeah

why is the math so complicated to use I

assume the BMP format does that because

it is somehow easier than doing

something else why isn't the math they

specified something that can just be

ANDed with the value or something so I

don't know why they chose that it seems

like a really bad idea to me what I

would have done is I would have

specified the order just hard-coded the

order and if I really wanted two

different orders like one for

little-endian one for big any and I

would have had a flag but you're right

the masks and BMP are complicated to use

for no real reason like why do we have

two bits can them because they're just

it was not a very smart format or store

the shift in there to write like store

the shift and they're like what spend an

extra 16 bits at the front of you're at

the front of your file format right to

tell what the ships were

something else that came up earlier the

stream did you mean to call the arc zone

of caprara LLVM or clang I always call

it LLVM because really it's it's not

clang that really is the thing that

determines what the intrinsics are it's

LLVM for the most part because LLVM

either has those intrinsic sin the bike

code or doesn't and if LLVM doesn't have

the intrinsic in the byte code then it's

like playing is just I mean playing just

a thing that like spits out LLVM byte

code right like and so if LLVM byte code

has intrinsic then it has it and if not

it doesn't so I typically always think

of it as LLVM but you could call it

playing if you wanted to it's up to you

in the bits can function what about

returning the index and have it sets a

negative one if not found

I see search functions doing that a lot

yes you could do that I find that to

just be a little more error-prone so I

would rather have there be a clear rule

that you can switch on because the

compiler is is probably going to

reorganize that stuff anyway and so it

just seems like an easier thing to do

can you go over the linear alpha blend

diagram again sure I certainly can so a

linear blend is actually something that

you can think about working between any

two things even if you don't really even

know what they were as long as they obey

a few simple properties right you could

almost think of it as just any two

points in any space right as long as

that space obey some linearity things

which let's not let's not get too crazy

for the moment but basically if I have a

point that I'm going to and a point that

I'm starting from right and I want a

linear blend between these all I need to

do to be able to do a linear blend is

have a way of producing the sort of the

the difference between those right I

need a way of being able to say a minus

B now I need to be able to produce that

and what that is like a minus B right

B minus a because we're going to be so B

minus a is basically telling me this

motion it's telling me how to go from A

to B right so on a simple number line

right if if I'm talking about just 5 the

number 5 is B and the number 2 is a I

subtract them right and I get 3 it's the

amount of steps it would take the

distance it would take to get from one

to the other right but this works in all

dimensions not just one so I'm just

gonna say like okay it's something it

doesn't matter two dimensions three

dimensions four dimensions I don't care

as long as we're talking about is you

know basically Euclidian stuff here as

long as I can produce this I can do a

linear blend and the way I do that

linear blend is you can think about it

exactly like you think about starting

someplace and walking somewhere else

right if I start a day and I add in this

difference this move movement right here

that B minus a right and I add in B

minus a I will get to B that's just how

it worked like you can think of it very

simply if I take the distance between

these two things and I and I walk it I'm

gonna end up at the other one right now

technically it's not quite a distance

it's actually a directional distance

right I know what direction I'm walking

to and that's why it's important that

we're talking about sign the numbers

here

right B you know if I say like I said

before let's say B was was actually

three and a was the five are sorry just

to was what I said before let's say B

was two when a was the five it's

important that I don't end up with three

here when I do this operation I need to

end up with negative three because it's

a direction I need to know which way I'm

going

right but basically as long as you can

produce that you could end up with this

which is basically just I mean it's it's

as simple as as this it's B equals a

plus B minus a right this this equation

right here if you can produce that

equation for the things that you're

talking about linear blending is as

simple as introducing this right you

could produce that new point by saying

instead of walking the whole distance

between a and B I'm going to walk some

portion of the distance between a and B

some fractional value that can be

whatever it is it could be 50%

it could be 25% what

ever it is and I will get however far

along there I need to go and that is a

linear blend like I said you can use

algebra to rearrange it if you want to

you can multiply this out if I have a

plus T B minus a equals some new value I

know that if I rearrange that I can use

the distributive property t be minus T

hey like that and I can go ahead and

regroup the terms these two a terms I

have a I have a 1 here in front of my a

coefficient right and I have a minus T

here so that's just 1 minus T times a

plus TB I just regroup the terms I

grouped this term in this term into one

term and then the TB still there and

that's the other form of the linear

blend that you often see so you've got

those two different formats there right

and so that's linear blending in a

nutshell it's very very simple

extremely powerful this one equation is

so powerful actually that you will see

it underlie almost everything we do in

this game there will be almost no part

of the code that does not at some point

compute some variation of this then so

trust me on this one you will have if

you don't quite understand it now you

will have a lot of opportunity to

practice because this is literally this

is probably if I had to say and I don't

know I haven't thought this through I'm

just saying this off the cuff but if I

had to say this is probably the most

important equation in all of gaming like

games need this equation probably more

than any other equation and there's tons

of equations that are important to

gaming but if you could only have one

this is probably the one you would need

because games can't do anything without

a linear blend because everything else

is built on top of it nonlinear blending

is built on top of it

all rasterization is built on top of it

all of animation is built on top of it

it's just incredibly incredibly

important

I'm unexperienced with Windows I know

we'd do a large virtual Alec at the

beginning is this basically the same as

an EM map in Linux yes for the most part

I'm not super experienced with M map so

I'm not prepared to say exactly but what

I would say is I would encourage you if

you have questions like that the best

place to go at the moment since I

haven't really looked at the Linux stuff

and haven't done an implementation of it

yet if you go to hand-made here or click

on news and forums there's a coding

resources page if you go to that coding

resources page go to the Linux listing

and click on handmade penguin SDL port

because I believe and I'm sorry if I it

turns out to be wrong but I believe he

actually explains how to do all of the

stuff that I did in in in Linux so you

can kind of go through here and look so

there's that M map and it appears to be

in the same place as our virtual Alex so

so take a look at that page handmade

penguin because that that would probably

be the best place to go for an

exploration since I know he's actually

worked it all out already

let's take a look here what bit value if

is 50% brightness well I'd be you have

zero and you have 255 so you can either

think of it as 127 well okay so here's

how to think about it in our actual

thing right here's how to answer that

question basically so you know that the

Alpha value that we're pulling out is

going to be whatever the alpha value in

in the bitmap is divided by 2 times 5

right so if you think about it what we

would want to do is say all right if we

have 128 divided by 255 what is that

well it's point 500 something something

something right not quite exactly 50 127

divided by 255 is 0.49 8 something

something something something right so

the actual answer is that there is no

actual value for exactly 50 127 is a

little bit low 128 is a little bit high

what are you gonna do you could adjust

this math to make it exact

you know if you wanted to write but but

the problem is you want to make sure

that one is exact so you need to make

sure that to be 5 by 2 to 5 is 0 like if

we were to do 128 divided by 56 you

could get an exact so if we divided by

triple 86 or you can get exact but you

could never get 1 because then your

highest value is typically 625 you would

end up never quite being able to get one

either and that has consequences as well

worst consequences in fact let's see

here let's see

what else we got does the display blit

operation actually do anything with the

Alpha Channel no it just writes zeros to

the Alpha Channel it's only using the

Alpha channels of control to basically

control how the blitt operates so it's

reading it but it never writes it if

that's what you're asking which i think

is what you're asking them now the solid

curve on Wikipedia's gamma correction is

the linear space intensity should be

what you are looking for okay so you're

saying go to Wikipedia and look up gamma

correction did you say or something else

what I'm sorry I just spaced out there

let's see oh we're out of time let's

just grab it then saw curve on gamma

correction I think so gamma correction

look at you so you're saying there's a

solid curve on here okay so you're

saying this curve right here hopefully

is what you're saying so I didn't do the

reading yet but assuming that that's the

solid curve basically is this what are

the two axes though the dotted line it's

a linear transfer function so I should

CRT there's difference what are the axes

by the way which axis is which hi there

guys can you tell me what the axes are

because which one is which one is the

brightness and which one is the input

value enable sheriff you have a

correction the dotted line indicates a

linear transfer function those solid

line shows how a typical CRT behaves the

dashed line represents the inverse

function so yeah so I guess I don't

actually know it doesn't say I assume

this is brightness and this is the input

value but I guess I don't actually know

assuming it was and you can see that's

what the curve looks like so it actually

has a lot more kind of resolution there

at the bottom then at the top there's

not a lot of resolution in the

brightness but I don't know I don't know

if that's reading this right because it

doesn't actually have the thing so we'll

have to look at that a little more

detail later all right

let's see I think that's about it for

stuff we can cover today

we have gotten this is pretty good we

have gotten to the point where we can

now do layered graphics on the screen so

I'm thinking it's a you know it's a it's

kind of a question to some degree about

where we go next like you know what what

is the next thing that we should be

doing if we're actually talking about

moving forwards on the game and so what

I was thinking of doing next is is

working on our player motion code

because I think that's important because

it's going to talk it's going to

basically speak to how we need to do our

collision detection which speaks to how

we need to interpret the tile map so I

was thinking of doing is basically

saying all right maybe we do a little

cleanup tomorrow and Friday of stuff

that still needs to kind of get cleaned

up a little bit just looking looking at

things like that see what we want to do

in terms of that sort of stuff but maybe

what I was thinking is so we'll start

next week on like actually talking about

linear algebra and vectors and we'll

actually start writing the movement code

for the guy for real so that it feels

good and we'll talk talking about

collision detection for real so that we

can kind of see what we need from our

engine for that because that's a really

important thing to spec that's one

direction to go the other direction to

go is we need to talk about how to

actually build these background bitmaps

right which is to say take the tile map

and kind of go ahead and build the

background bitmaps I don't know which

one we should do next it's kind of a

tough thing to say so it's a toss-up

maybe we'll flip a coin maybe we'll take

a poll I'm not sure both of them have to

get done before we know what our what

our engine spec really truly is so it's

something to think about but for now we

can always be happy that we now have a

nice living engine we can use so we can

put stuff up on the screen which is kind

of nice and from now on we'll always

have graphics to back us up which is

much nicer than looking at a bunch of

rectangles in my opinion so thank you

everyone for joining me for another

episode of handmade hero it has been a

pleasure coding with you as always and

it was a pleasure talking about the

linear blend which like I said is one of

the most important things you will see

it come up again and again and again on

the screen so it was nice to finally get

to that I think that's a pretty big

milestone starting to actually use the

linear blend and again like I said we'll

talk about it a lot more

in the future and we'll see a lot more

about it and we'll kind of learn more

and more about how interesting it is and

how intricate you can basically how

amazing it is that you can do so much

for such a simple function and basically

how it underlies everything if you will

so anyway thank you very much for

joining me if you would like to follow

along at home with the source code you

can pre-order the game and it comes with

full source code already every night I

update the source code you'll basically

a link in your email that you can use to

download the source code it also has the

art assets that we've been using on the

stream there's a separate zip file you

can download if you want to use like you

know the little guy and the backgrounds

and that sort of stuff for doing your

development and so you know anything you

basically completely follow along at

home and all the assets that you need to

follow along completely at home come

with the with with that pre-order so

check that out if you're interested we

also have a patreon if you want to

support the video series you can always

subscribe there we also have that news

and forums site which I showed just a

few minutes ago which has a ton of

resources for people who are trying to

learn on the stream too a lot of great

stuff the schedule for the week's always

posted here there's also the episode

guide which allows you to catch up with

old episodes got nice annotated video

series on there coding resources has the

ports like I showed before and other

resources you might need like good

documentation of stuff and there's a

forum you can post in when you need to

ask questions so please check that out

if you're trying to learn from the

series it's a great resource and there's

a lot of very helpful people on there

who are very nice on the forums who will

also try to answer your questions if I'm

not around to do so

so thank you very much for joining me I

will see you tomorrow same time saying

place 8:00 p.m. Pacific Standard Time on

Twitch so I hope to see you all there

and if I don't have an excellent

Thursday see you later everybody