Holly - Multi Pass Map Rendering

[kevin]

    Holly - Multi Pass Map Rendering


     This is my demo of the Steve's Holly map rendering from way back in 2010.   This was written on my classic Duron 800mhz system (currently 18years old) and shows four different approaches for drawing a scrolling alpha blended map on top of high resolution backdrop.     The four methods come down to combination GPU and CPU rendering together, to improve the end through put of your program.  


     Unfortunately,  as time has gone on and Windows shows less interest in support direct draw as do hardware vendors the approach today (Windows 10 era) chokes...  



PlayBASIC Live - Running the Holly Map Opt's on 18YO Duron 800  (2021-03-28)


   

 

  Related Links

      Holly _ Platformer Adventure Demo
      Tutorial - A Crash Course In BASIC program Optimization
      Dirty Rectangles  (Combined Video & Fx Rendering)



   Script - English Captions

Code Select Expand



hi

welcome back

what you're looking at is

a phone capturing my duron 800

system which is 18 years old

it's 2002 vintage it's pretty

underpowered

but what i found on this before i'm

having to recall this

with a phone i can't see with the screen

very well

i'll go with the pb test

i don't think the framing's very good

but we'll stick with it

i'll go up here to reopen and

it's the platform and test okay

i found this version of the plat steeps

platformer demo

but this one's running or laid out

to be more optimal for the this much

older system

so single core of course

not a lot of memory in this thing low

end gpu

i think i think this one has a geforce 2

in it there's some description

and that was a replacement for one that

blew up that's right yeah

i used this system for this was all the

first versions of pb were written on

this system

and then i switched across to the athlon

system

uh some point in the i don't know when

that was now

so this thing right here i want to go

back and see how

well we can make this run on a low end

system such as this

is the right version i think it is

it looks a bit distorted on the camera

but we'll put up with it

so we have blocks uh

and keyboard to scroll you can't see the

the display on the camera i don't think

no it's kind of frames per second rating

up the top corner it's got a few

different modes for the scrolling

so this is called standard and

it's taking 52 ticks or milliseconds to

render the frame

so we're getting nine frames per second

so we've got alpha blending foreground

blocks over a background picture

in 32-bit let's try the next mode this

is called single pass occlusion

with effects four rounds so we've

improved the frame rate straight away

we're in the mid to mid-20s now

we've got 45 ticks to render the frame

so we've got blending around these kind

of edges around here

look this feathering around here that's

all

blending i'll blend it there so we're

using a trick here

because the foreground is opaque this

section here is all opaque

we can just draw that without any

blending at all and i draw the edge

cases with blending which works out to

be much easier

on the processor this is method one

i've got a method two which is called

single pass occlusion with video map

foreground now we're a game where the

24 frames per second range

uh which i'm quite happy about actually

to be honest with you

this is with video map foreground now

we'll try the next mode

this is two pass occlusion about 26

frames per second

with 38 ticks now to render the frame

it does have vsync enabled so we are

losing some time there

it's quite good

uh two boss occlusion with matt fulgrant

that yeah we're approaching about 30

frames per second now

which is pretty good for 22 sorry 18

year old system

now most of the 2d games for this kind

of system here would have been

just purely lit face

so standard mode is 16 frames per second

go forward and compare

the the quickest one

we've doubled the frame rate pretty much

it's pretty cool i'm pretty happy with

that

let's try it with our v-sync enable

i'm not sure how this demo is set up to

be honest

drawing the screen to the background

no i don't think it is yeah that's right

so kind of screen mode that we're

running this in

settings uh screen

[Music]

well there you go we're running 1024 by

768

now on this system here

there are very few old 2d games that run

in that resolution

uh let's just try and play the screen

mode there a bit we'll try and drop it

to 30

16 bit

uh and just see what we get

the desktop of the system is a mess now

complete mess

okay so we've doubled the frame rate

by going a 16-bit frame buffer

we barely noticed the difference on

screen at all

this is mode 0 which is the slowest

method

single pass is now 43 frames per second

44 frames per second

the other single pass occlusion with

within

with video map four ram about the same

48 frames per second now

it's quite smooth and

2 pass occlusion with here

that's what that means i'm not quite

sure now i noticed

that um

the principal idea here is is to

pre-calc

what blocks you need to draw and how you

need to draw them

so mode four yeah i've got a nice

44 45 frames per second

it's pretty cool so what it's doing

is that it's actually it's bleeding the

background picture

in one asynchronous bleed it's using the

video hardware

to do that then the foreground here this

opaque section

inside the tiles this is just tiled and

drawn as video blocks as well

so if these tiles are too small that's a

bit of a clutch for the gpu but keeping

them with

decent size and a power of two helps but

these edges

that need alpha they're blended to the

background so

this this portion around here is drawn

to an effects buffer

blended with the effects buffer and then

copied to video memory

because that's starting 32-bit and

scaled down

we don't actually have any problem with

that

you would notice that the colored

limitations of 16 bit more

if you had you know a linear gradient

from one point to another

i don't know if this runs with if i

change the resolution

we'll soon find out let's have a look

if you just go back to 600

uh go to 960

960 is probably good compromise actually

if this monitor will support it i'm not

sure it does

there's your answer it doesn't so we

have no

there's no full screen exclusive uh

supported for this monitor so go back to

800 by 600 which is the default so we're

in 16 bit

800 by 600 uh

with foreground layer with alpha effect

with alpha blending on it anyway

yeah i think our uh i think our

loader is organized to work in uh

it expects the video frame to be

whatever was 1024 wide

so this was with no optimization at all

getting 35

frames per second on the 18 year old

system

just straight to the bleeding

a bit of messing around and now you've

got you know 60 frames per second

a single pass occlusion

about the same for for the second

variant

now for too fast occlusion you've got

frame rates in the

70s and 80s

now your rate will vary given the amount

of work that's included in the frame

stands to reason i would have thought

hmm

i don't know what version of pb we're

actually using i think it's

164 variant of some type anyway

so it's not as this is not as quick as

like a 164p for example

so there's probably some performance

laying on the table just through the

interpreter

itself

probably not a lot but some yeah

hmm that's pretty cool sorry you didn't

test

mode four

cool once you get

the blending effects off-screen you're

doing frame rates of almost 200 frames

per second but there's no

they're irrelevant because just timing

how fast

the blue can draw the background

there you go uh a bit of a trip to

memory lane

this example here was written in 2010.

i'm going to copy this now so i'll put

it on the

well put it i'll copy across to the

forum etc so you can have a

play with it yourself i don't know if i

release this or not i've no idea

not a clue i hope so

sometimes i mean to do those things but

don't get around to doing it

yeah which happens a lot happens way too

frequently unfortunately

but that's why i've got the system set

up at the moment i'm sort of copying

making sure i've got everything off this

machine

there's a whole bunch of files on here

that i think are

probably the only copies of those files

left

mostly things like you know downloads

from the forums that have now been

deleted

by because they're stored externally

but it means that those old threads

become useless because

all of a sudden some thread that's

talking about something you've just

searched for

has nothing in it

yeah the copy here just uses copyright

nothing special about that at all mouse

button

draw foreground over the background fill

in holes

oh that is interesting so i'll run that

again

and apparently if i hold the mouse

button down

it'll give us an indication of when

we're using the fill modes um

what what blocks are being

i'll just change the resolution back to

whatever it was

it's a bit of a massive system now it's

got stuff

everywhere so go back to 1024x768

in 16 bit

so over here go through one of these

block methods

there you go so if i hold down the right

mouse button

it shows the foreground blocks that are

currently being drawn opaque now if we

use it

we've used a smaller block size

we use two maps which is what what i

would probably do

and i would use two maps because these

blocks here

could really be blocks i think these are

i think the tiles are about 50 by 50 or

something

like like a weird size

so these blocks here could be you know

a nice even number like if they're 50

wide we make them 10 by 10

this would mean that this block here

could

we have much less fragments of it being

blended

and there would be more opportunity for

these background blocks here to be

just drawn opaque

yeah so all those methods have the same

limitation

same sorry effect in them

so we're not really even here we're

doing a lot of software blending

on this scene and we can improve this

by you know by altering

the block size which would actually help

them things like collision and storage

and

all those side side

sorry side effects as well

i not that it matters now really

but it's an interesting observation to

make in terms of optimization if you

want to draw maps efficiently

this is why we added some occlusion

functions into pv

for you to help you

if you've got multiple layers of blocks

and you want to clear out

backlash from being drawn you can use

the foreground occlusion to do that

anyway they'll have to do us thanks for

watching uh it's very late and i'll put

this together together quickly

uh in the morning or tomorrow at least

anyway

i'll get back to copying backing up the

files

so hopefully that was interesting i

don't know if it was or not

i hope you have a good time coding and

have fun for easter all that kind of

stuff and

we'll see you soon bye




  Download

      Download Holly Multi Pass Map Test (login required)

[stevmjon]

hey kev

thanks for the demo. you used #include maps. i didn't use this, but i did have multi-pass method set up.

i originally had everything drawing to the video screen, which was super slow (fx blending to video screen).
then i learnt about fx memory (fx & afx images), then had everything draw to fx image (faster but solid tiles were drawn by cpu).
then i came across your demo on multi-pass (a mini platformer you made), then i made a multi-pass map too (divide fx to cpu and video to gpu, yay).

to run the holly demo: project2_5 on PB v1.65 , i needed to comment out line 694 in declare tab [seems it doesn't like ExitFunction there???].
***  to see multi-pass run game, then press F5 , then press 3 (above letters only) , then choose between pressing H , J , L.

it will display boxes around fx / video tiles for multi-pass.

also for some reason the game runs slower in normal mode, then runs faster when i press F5 where it displays text on screen ???
i may have to look into this. i will also check the newer version holly demo: project3_5 to see if it runs different, as this was a re-write.

[kevin]

thanks for the demo. you used #include maps. i didn't use this, but i did have multi-pass method set up.

  The map slib includes the Load/Save Map functions,  I guess it was just dropped into your original code in order to save the map out and load into this..  not that i remember now

to run the holly demo: project2_5 on PB v1.65 , i needed to comment out line 694 in declare tab [seems it doesn't like ExitFunction there???].
***  to see multi-pass run game, then press F5 , then press 3 (above letters only) , then choose between pressing H , J , L.

   That's a pretty old build,  I think it ran without changes in the current editions.  Which woold be C2 beta20 or whatever it was from a few weeks ago.

also for some reason the game runs slower in normal mode, then runs faster when i press F5 where it displays text on screen Huh
i may have to look into this. i will also check the newer version holly demo: project3_5 to see if it runs different, as this was a re-write.

    This relies upon the GPU doing the bliting asynchronously, which just doesn't seem to be supported in Win10 and I guess Win8 as well...   On my Win10 boxes, a pure FX scene runs quicker. 

     Porting it to g2D would get a better result for the adventurous