ElfScript Simple Speed Test:

Goal: get the (close to or better) speed as lua (vanilla - without jit) on "localvar test"

Lua 5.5.1: 1.243u 0.002s 0:01.25 99.2% 0+0k 0+0io 0pf+0w
Elfscript 0.7c very close : 1.354u 0.002s 0:01.36 99.2% 0+0k 0+0io 0pf+0w
Elfscript 0.6f close to lua but still about 200ms: 1.434u 0.006s 0:01.45 98.6% 0+0k 0+0io 0pf+0w
Elfscript 0.6f outperform PHP 8: 2.242u 0.013s 0:02.31 97.4% 0+0k 18792+0io 97pf+0w
ElfScript 0.4 outperform python 3: 15.768u 0.005s 0:15.83 99.5% 0+0k 0+0io 0pf+0w.
ElfScript 0.? outperform OGE3D (my Torque3D based on 3.10) : 33.268u 0.299s 0:33.61 99.8% 0+0k 0+24io 0pf+0w

Version 0.8

added class ElfScript::Lambda::InjectCaller and error handling when lamba is defined inside a function

after speed test:

Script	run 1	run 2	run 3
test_localvar.elf	1.33s	1.32s	1.33s
test_global.elf	2.42s	2.41s	2.41s
test_static.elf	4.06s	4.05s	4.06s
test_dynamic.elf	3.62s	3.64s	3.61s
test_vector_components.elf	3.83s	3.82s	3.80s

added "normal" and injected Lambda run/step to PointStorageObject ..

also make now always a speedtest to keep a eye on - if I did break something. Very good back on track.

Script	run 1	run 2	run 3
test_localvar.elf	1.35s	1.35s	1.36s
test_global.elf	2.48s	2.51s	2.44s
test_static.elf	4.02s	4.02s	4.01s
test_dynamic.elf	3.63s	3.62s	3.61s
test_vector_components.elf	3.83s	3.80s	3.79s

Removed cvLambda now using cvPointer with subType cvsLambda:

Script	run 1	run 2	run 3
test_localvar.elf	1.38s	1.38s	1.38s
test_global.elf	2.46s	2.45s	2.43s
test_static.elf	4.39s	4.40s	4.47s
test_dynamic.elf	3.67s	3.68s	3.68s
test_vector_components.elf	3.98s	3.98s	3.97s

cleanup a bit no idea how to fix the cpu prediction again, so far:

Testing crazyElf starfield with 4200 stars: ~320 fps, thats fine.

Tuple assign with pointstorage rocks here, also if i lost 100ms against 4 days ago, i dont want to roll back.

about 100 ms slower speed tests ... not funny .. must be something in the VM but what. Somewhere it breaks the CPU prediction or i added some cpu cycles. I think it's since the tuple assignment - but not sure. I changed the float format from %.9g to %#.9g but i think this is not the reason. Maybe I should clean my cpu cooler lol StarField Test still good so i should forget about get closer to lua vanilla it's replaced by jit in most cases anyway. It's not the cpu cooler - I have a binary from 4 days ago (2026-09-10) where it's much faster than now.

must be after commit => 2026-09-11 "got expoerimental func call op codes running but not really faster in my tests." so it's not CALLFUNC. I guess it's lambda change ... but can't find out why.

It's the cvLambda and cvPointer ..., when i kick em out it's fast again... Need to test this more on a faster PC since i have to change console.h all the time and everything is recompiled - bad on my slow Notebook.

Since then i added:

CALLFUNC stuff
Tuple asignment
lambda call
parent scope

Script	run 1	run 2	run 3
test_localvar.elf	1.42s	1.42s	1.42s
test_global.elf	2.53s	2.53s	2.53s
test_static.elf	4.50s	4.42s	4.42s
test_dynamic.elf	3.85s	3.83s	3.85s
test_vector_components.elf	3.92s	3.92s	3.92s

Direct FastPath on SAVE/LOAD_VAR_STR an SAVE/LOAD_LOCAL_VAR_STR... but did not speed up anything
global scope local var .. more or less useless.
pointer type ConsoleValueType
lambda variables + ConsoleValueType
tuple assignment for local variables
New CallFunc OP Codes
Cache last entry on Dictionary lookup .. this speed up global var test but there is only one variable which is not the same as in a normal enviroment
Enabled inline cache as default ON again
Added Raylib bindings here as addon - and merged it with raylib-ElfScript and elfFlux later.

Version 0.7g - stability tests for first release -

#ElfScript 0.7

One of the fastest, easy-to-embed interpreted script languages.

How it differs from TorqueScript:

Highly optimized fast Virtual Machine with Direct Threading
New ConsoleVector (4-point float) initialized with %vec = {1};
New flat Array class with constructor initialized like %arr = [1, 4.0, "Hello"];
Almost no more usage of string conversion for datatypes
Fast for (%i in ..) iterators
Inline FieldCache for fields and components
Statements can be case-insensitive
Dynamic Fields are typed (not just strings), just like local and global variables
Several fast-path implementations to directly write data
Script preprocessor to use #define constants
Included Addons (bindings): SDL3, ImGui, Neural Network (genann)
And much more ;)

SpeedTest (2026-09-09) output piped to /dev/null

Script	time
test_localvar.elf	1.357u 0.003s 0:01.36 99.2% 0+0k 0+0io 0pf+0w
test_global.elf	3.048u 0.002s 0:03.05 99.6% 0+0k 0+0io 0pf+0w
test_static.elf	4.262u 0.002s 0:04.27 99.7% 0+0k 0+0io 0pf+0w
test_dynamic.elf	3.734u 0.005s 0:03.75 99.4% 0+0k 0+0io 0pf+0w
test_vector_components.elf	4.068u 0.002s 0:04.07 99.7% 0+0k 0+0io 0pf+0w

SlotAssignNode::compile auto detect field type when field is created, but initialize with type is better!
SlotAssignNode::better ConsoleVector assign and Array constructor asign
Added Array object. A flat array holding ConsoleValues - This is now my universal Swiss Army Knife :D ! I added a constructor in lexer to do %foo = [1,2,3]; with the limitation: I can not add ConsoleVector's in this constructor. But %foo.push_back({1,2,3,4}); works fine. I removed my new explodeToObject function and added fromString method to Array which makes more sense. With this Object I was able to implement bindings like SDL_RenderPoints without fallback to String - which make the call useless. This is now a easy to use alternaive to PointStorageObject. While PointStorageObject is still a very fast way to work with a lot of Points/Rects/Colors .. everything requires a lot of 2..4 float/integer values.
Implemented direct handling getter/setter for ConsoleValue (used by ValueVector)
Added Neural Network bindings using genann - demo added to CrazyElf.
SDL3 got a "melody maker" - demo added to CrazyElf.
SDL3 got some more bindings like Clipboard text handling.
Added some methods to Simset/Group to make it more like an "std::"Vector object
Set Fieldcache default off - marked as experimental again! When working on scripts I had a strange behavoiur - cache failed and dynamic fields of just created objects where not written / could not be read. So it's fast, but not production ready. Unforually i can not reproduce it in a test enviroment, so far. This is also bad for Components speed since it's embedded into the FieldCache - but not effected. Maybe i put it in parts so i can enable Components cache without object field cache. 2026-09-04: I think i fixed this, skiping objects without ID (new Object ...) on save.
Fixed math inline commands to use getFloat instead of getFastFloat.
Added inline commands to docu generator (stub file)
SDL3 Colors (baseElf) changed to ConsoleVector
Short if: Vector (PoD) parser: true ? {0} : {0} || true ? {0} : expr || true ? expr : {0}
MiniElf Demo - showcase how to embed it - without anything else. Lib failed - need some research why.

Version 0.7f

added ConsoleVector to union << 24 byte instead of 32
print and printf do not need to pop.
ELF_ENABLE_FIELDCACHE default ON again
function toObject => %obj = toObject({"Hello", "World", 10.0}); create an object with fields TypeString v0 = "Hello", TypeString v1="World", TypeF64 v2= 10.0;
function toArray => $foo = {"Hello", "World", 10.0}; echo(toArray("$foo")); clear $foo, create variables TypeString $foo0 = "Hello", TypeString $foo1="World", TypeF64 $foo2= 10.0; and return the count of variables;
Both new function also works with a string like: "Hello World 10.0";
Auto set type on Dynamic Fields:
- $foo = {1,2,3}; %obj.foo = $foo;vardumpField("%obj.foo"); set an TypeVector
- %obj.addField("myFloat", "TypeF32", 0); %obj.myFloat[0]= 4; myFloat[0] copy the type from myFloat.

Version 0.7e

Fieldcache cleaned and fixed but still default off in cmake.

Results with FieldCache on -Ofast

Script	time
test_localvar.elf	1.388u 0.003s 0:01.39 99.2% 0+0k 0+0io 0pf+0w
test_global.elf	3.004u 0.001s 0:03.01 99.6% 0+0k 0+0io 0pf+0w
test_static.elf	4.499u 0.004s 0:04.51 99.5% 0+0k 0+0io 0pf+0w
test_dynamic.elf	3.844u 0.006s 0:03.86 99.4% 0+0k 0+0io 0pf+0w
test_vector_components.elf	4.100u 0.003s 0:04.11 99.7% 0+0k 0+0io 0pf+0w

Results with FieldCache on

Script	time
test_localvar.elf	1.412u 0.003s 0:01.42 99.2% 0+0k 0+0io 0pf+0w
test_global.elf	2.975u 0.003s 0:02.98 99.6% 0+0k 0+0io 0pf+0w
test_static.elf	4.621u 0.002s 0:04.63 99.7% 0+0k 0+0io 0pf+0w
test_dynamic.elf	4.007u 0.004s 0:04.02 99.5% 0+0k 0+0io 0pf+0w
test_vector_components.elf	4.135u 0.002s 0:04.14 99.7% 0+0k 0+0io 0pf+0w

Results with FieldCache off

Script	time
test_localvar.elf	1.420u 0.003s 0:01.43 99.3% 0+0k 0+0io 0pf+0w
test_global.elf	2.960u 0.007s 0:02.99 98.9% 0+0k 8+0io 0pf+0w
test_static.elf	9.044u 0.001s 0:09.07 99.6% 0+0k 0+0io 0pf+0w
test_dynamic.elf	11.402u 0.004s 0:11.44 99.6% 0+0k 0+0io 0pf+0w
test_vector_components.elf	6.108u 0.007s 0:06.14 99.3% 0+0k 0+0io 0pf+0w

Version 0.7 d

SDL3 added EngineUnmarshallData< Point2F >
SDL3 added EngineUnmarshallData< RectF >
Vector Component Fieldcache disabled again: AddressSanitizer: heap-use-after-free popFrame/pushFrame ... make my cache invalid!! ....added a S32 frameID no idea the cache is worth to live there !!

Version 0.7 c

Let's see if I can speedup Vector Componets as Fieldcache:
- only for local and global vars
- ok, not as fast as expected, but faster as without => 3.945u instead of 5.411u
fixme stack overflow when a new command is used but not consumed!! << added OP_POP_STK

Script	time
test_localvar.elf	1.354u 0.002s 0:01.36 99.2% 0+0k 0+0io 0pf+0w
test_global.elf	3.023u 0.004s 0:03.03 99.6% 0+0k 0+0io 0pf+0w
test_static.elf	4.196u 0.002s 0:04.21 99.5% 0+0k 0+0io 0pf+0w
test_dynamic.elf	3.923u 0.001s 0:03.93 99.7% 0+0k 0+0io 0pf+0w
test_vector_components.elf	3.837u 0.005s 0:03.85 99.4% 0+0k 0+0io 0pf+0w

localvar is also faster - nice but not sure why.
Components are much faster.
Fields are slightly slower because of my change to virtual overwrite custom types . This is ok because it speed up in real world applciation.

Version 0.7 b

Added virtual functions to overwrite static fields fastPath
Got ConsoleVector directly accessed via Console-Functions and -Methods without string convert! - Rock 'n Roll!!
this means ConsoleVector is completely implemented as a "nativ type" and i will not remove it from ConsoleValue because of 16 bytes! It's 32 bytes and disable it does not change anything in speed.
too bad i dont have a benchmark to test this ... but my ConsoleVector Math functions using reference can be changed now!

Version 0.7 a

TestResult (best of 3 tests) in /docu/speedtest/elf_v6:

Script	time
test_localvar.elf	1.418u 0.005s 0:01.42 99.2% 0+0k 0+0io 0pf+0w
test_global.elf	2.930u 0.002s 0:02.94 99.6% 0+0k 0+0io 0pf+0w
test_static.elf	4.017u 0.002s 0:04.03 99.5% 0+0k 0+0io 0pf+0w
test_dynamic.elf	3.804u 0.004s 0:03.81 99.7% 0+0k 0+0io 0pf+0w
test_vector_components.elf	5.411u 0.002s 0:05.42 99.8% 0+0k 0+0io 0pf+0w

With IPO / -O3 optimized compiled

Script	time
test_localvar.elf	1.365u 0.003s 0:01.37 99.2% 0+0k 0+0io 0pf+0w
test_global.elf	3.487u 0.006s 0:03.50 99.4% 0+0k 0+0io 0pf+0w
test_static.elf	4.629u 0.002s 0:04.64 99.5% 0+0k 0+0io 0pf+0w
test_dynamic.elf	4.118u 0.002s 0:04.13 99.5% 0+0k 0+0io 0pf+0w
test_vector_components.elf	5.171u 0.004s 0:05.20 99.4% 0+0k 0+0io 0pf+0w

Localvar and Vector Componets is faster but global/static/dynamic slower!

With -0fast compiled:

Script	time
test_localvar.elf	1.410u 0.002s 0:01.41 100.0% 0+0k 0+0io 0pf+0w
test_global.elf	2.947u 0.003s 0:02.96 99.3% 0+0k 0+0io 0pf+0w
test_static.elf	3.908u 0.004s 0:03.92 99.4% 0+0k 0+0io 0pf+0w
test_dynamic.elf	3.784u 0.004s 0:03.80 99.4% 0+0k 0+0io 0pf+0w
test_vector_components.elf	5.699u 0.006s 0:05.73 99.3% 0+0k 0+0io 0pf+0w

Best results with this only components is slightly slower then -O2 (first one). But i would prefer -O2.

NOTE Vector Component and Static Field are float. Math.randomF is double. The ConsoleValue is also double. When float/double is converted there are rounding errors.

Version 0.6f :

Version 0.6g:

Version 0.6f :

splitted the foreach / for range OP codes but i had it fully optimized in 0.6d

Version 0.6e :

removed cvString and only using cvSTEntry

Version 0.6d :

revisit and optimize foreach/for range

Summery ElfScript 0.6d vs TorqueScript

Static Float Field: 37 times faster
Dynamic Float Field: 25 times faster
Float Component: 19 times faster
Local Float Variable: 14 times faster
Global Float Variable: 8 times faster

Version 0.6c :

revisited inline cache. removed components from cache.
fastpath setDataField/getDataField/fillFieldCache moved inline to header << over a second faster!
- The lesson I learned sometimes setting code to inline can change a lot.

Version 0.6b :

restored StringTable - there is a bug in the new implemenation, i saved it to unstable folder
ENABLE_INLINE_CACHE_LOAD/ENABLE_INLINE_CACHE_SAVE enabled again but disabled Componentfields

Version 0.6a :

rewrote StringTable but did not really speed up
added initial inline field cache ... thats an rocket in loops but need more cleanup and testing
with #define ENABLE_INLINE_CACHE but unstable on real test so ifdef'd!!!

docu/speedtest/elf/test_localvar_v6.elf

0.6g : 1.447u 0.005s 0:01.46 98.6% 0+0k 0+0io 0pf+0w
0.6f : 1.434u 0.006s 0:01.45 98.6% 0+0k 0+0io 0pf+0w
0.7a : 1.420u 0.001s 0:01.42 100.0% 0+0k 0+0io 0pf+0w

docu/speedtest/elf/test_localvar_foreach.elf

OGE3D* (current 26-08-05) : 33.268u 0.299s 0:33.61 99.8% 0+0k 0+24io 0pf+0w
0.6a ENABLE_INLINE_CACHE : 2.855u 0.003s 0:02.87 99.3% 0+0k 0+0io 0pf+0w
0.6a : 2.857u 0.004s 0:02.87 99.3% 0+0k 0+0io 0pf+0w
0.6b : 2.892u 0.007s 0:02.91 99.3% 0+0k 0+0io 0pf+0w
0.6c : 2.754u 0.002s 0:02.76 99.6% 0+0k 0+0io 0pf+0w
0.6e : 2.244u 0.004s 0:02.25 99.5% 0+0k 0+0io 0pf+0w
0.6f : 2.107u 0.004s 0:02.12 99.0% 0+0k 0+0io 0pf+0w
0.6g : 1.916u 0.003s 0:01.93 98.9% 0+0k 0+0io 0pf+0w
0.6f : 1.831u 0.005s 0:01.84 99.4% 0+0k 0+0io 0pf+0w
0.7a : 1.797u 0.003s 0:01.80 99.4% 0+0k 0+0io 0pf+0w

docu/speedtest/elf/test_globalvar_forrange.elf

OGE3D* (current 26-08-05) : 34.477u 0.299s 0:34.82 99.8% 0+0k 0+56io 0pf+0w
0.6a ENABLE_INLINE_CACHE : 4.513u 0.002s 0:04.53 99.5% 0+0k 0+0io 0pf+0w
0.6a : 4.571u 0.003s 0:04.59 99.5% 0+0k 0+0io 0pf+0w
0.6b : 4.479u 0.002s 0:04.49 99.5% 0+0k 0+0io 0pf+0w
0.6c : 4.712u 0.006s 0:04.74 99.3% 0+0k 0+0io 0pf+0w
0.6e : 3.450u 0.005s 0:03.47 99.4% 0+0k 0+0io 0pf+0w
0.6f : 3.342u 0.004s 0:03.36 99.4% 0+0k 0+0io 0pf+0w
0.6g : 3.327u 0.002s 0:03.35 99.1% 0+0k 0+0io 0pf+0w
0.6f : 3.366u 0.003s 0:03.38 99.4% 0+0k 0+0io 0pf+0w
0.7a : 3.331u 0.005s 0:03.34 99.7% 0+0k 0+0io 0pf+0w

docu/speedtest/elf/test_dynamic_fields_forrange.elf

OGE3D* using ScriptObject: 137.226u 0.296s 2:17.73 99.8% 0+0k 0+136io 0pf+0w
0.6a ENABLE_INLINE_CACHE: 5.571u 0.004s 0:05.60 99.4% 0+0k 0+0io 0pf+0w
0.6a : 14.364u 0.005s 0:14.41 99.6% 0+0k 0+0io 0pf+0w
0.6b : 6.268u 0.002s 0:06.29 99.5% 0+0k 0+0io 0pf+0w
0.6c : 5.417u 0.003s 0:05.43 99.6% 0+0k 0+0io 0pf+0w
0.6e : 4.441u 0.004s 0:04.47 99.3% 0+0k 0+0io 0pf+0w
0.6f : 4.208u 0.000s 0:04.22 99.5% 0+0k 0+0io 0pf+0w
0.6g : 4.254u 0.002s 0:04.27 99.5% 0+0k 0+0io 0pf+0w
0.6f : 4.501u 0.010s 0:04.54 99.3% 0+0k 0+0io 0pf+0w
0.7a : 4.121u 0.005s 0:04.14 99.5% 0+0k 0+0io 0pf+0w

docu/speedtest/elf/test_static_fields_localvar_forrange.elf

OGE3D* using tom2DSprite: 192.723u 0.302s 3:13.28 99.8% 0+0k 0+248io 0pf+0w
0.6a ENABLE_INLINE_CACHE: 6.656u 0.004s 0:06.68 99.5% 0+0k 0+0io 0pf+0w
0.6a : 11.753u 0.003s 0:11.79 99.6% 0+0k 0+0io 0pf+0w
0.6b : 6.853u 0.009s 0:06.89 99.4% 0+0k 0+0io 0pf+0w
0.6c : 5.274u 0.004s 0:05.29 99.6% 0+0k 0+0io 0pf+0w
0.6e : 4.634u 0.008s 0:04.67 99.1% 0+0k 0+0io 0pf+0w
0.6f : 4.500u 0.005s 0:04.52 99.5% 0+0k 0+0io 0pf+0w
0.6g : 4.486u 0.002s 0:04.50 99.5% 0+0k 0+0io 0pf+0w
0.6f : 4.718u 0.003s 0:04.73 99.5% 0+0k 0+0io 0pf+0w (slower?!)
0.7a : 4.510u 0.003s 0:04.52 99.7% 0+0k 0+0io 0pf+0w

docu/speedtest/elf/test_vector_components.elf

OGE3D* (current 26-08-05) :: 140.183u 0.322s 2:20.84 99.7% 0+0k 0+48io 0pf+0w
0.6a ENABLE_INLINE_CACHE: 6.815u 0.003s 0:06.84 99.5% 0+0k 0+0io 0pf+0w
0.6a : 7.372u 0.004s 0:07.40 99.5% 0+0k 0+0io 0pf+0w
0.6b : 7.985u 0.001s 0:08.01 99.6% 0+0k 0+0io 0pf+0w
0.6c : 7.312u 0.004s 0:07.34 99.5% 0+0k 0+0io 0pf+0w
0.6e : 6.169u 0.006s 0:06.19 99.5% 0+0k 0+0io 0pf+0w
0.6f : 6.199u 0.002s 0:06.22 99.5% 0+0k 0+0io 0pf+0w
0.6g : 6.434u 0.002s 0:06.45 99.6% 0+0k 0+0io 0pf+0w
0.6f : 6.002u 0.002s 0:06.02 99.6% 0+0k 0+0io 0pf+0w
0.7a : 6.097u 0.004s 0:06.12 99.5% 0+0k 0+0io 0pf+0w

docu/speedtest/elf/test_vector_components_global.elf

0.6a ENABLE_INLINE_CACHE: 7.474u 0.004s 0:07.51 99.4% 0+0k 0+0io 0pf+0w
0.6a : 8.119u 0.005s 0:08.14 99.6% 0+0k 0+0io 0pf+0w
0.6b : 8.669u 0.006s 0:08.71 99.4% 0+0k 0+0io 0pf+0w
0.6c : 7.636u 0.005s 0:07.66 99.6% 0+0k 0+0io 0pf+0w
0.6e : 7.298u 0.003s 0:07.33 99.4% 0+0k 0+0io 0pf+0w
0.6f : 7.036u 0.005s 0:07.06 99.5% 0+0k 0+0io 0pf+0w
0.6g : 7.286u 0.004s 0:07.31 99.5% 0+0k 0+0io 0pf+0w
0.6f : 6.953u 0.003s 0:06.97 99.7% 0+0k 0+0io 0pf+0w
0.7a : 7.005u 0.009s 0:07.04 99.4% 0+0k 0+0io 0pf+0w

docu/speedtest/elf/counter.elf << says nothing, just for fun ;)

0.6a ENABLE_INLINE_CACHE: 5.016u 0.001s 0:05.03 99.6% 0+0k 0+0io 0pf+0w
0.6a : 5.180u 0.003s 0:05.19 99.8% 0+0k 0+0io 0pf+0w
0.6b : 4.877u 0.001s 0:04.89 99.5% 0+0k 0+0io 0pf+0w
0.6f : 2.939u 0.003s 0:02.95 99.3% 0+0k 0+0io 0pf+0w
0.6g : 2.927u 0.001s 0:02.93 99.6% 0+0k 0+0io 0pf+0w
0.6f : 2.936u 0.003s 0:02.95 99.3% 0+0k 0+0io 0pf+0w
0.7a : 2.650u 0.002s 0:02.65 100.0% 0+0k 0+0io 0pf+0w

docu/speedtest/elf/test_for.elf

0.6c : 2.084u 0.002s 0:02.09 99.5% 0+0k 0+0io 0pf+0w
0.6e : 1.374u 0.007s 0:01.39 98.5% 0+0k 0+0io 0pf+0w
0.6f : 1.090u 0.004s 0:01.10 99.0% 0+0k 0+0io 0pf+0w
0.6g : 1.318u 0.004s 0:01.34 97.7% 0+0k 0+0io 0pf+0w
0.6f : 1.184u 0.002s 0:01.19 99.1% 0+0k 0+0io 0pf+0w
0.7a : 0.947u 0.004s 0:00.95 98.9% 0+0k 0+0io 0pf+0w

Crazy Elf: assets/modules/StarField.elf

0.6a : ~ 1300 FPS (GPU: 67%) falling down to 1000 FPS (GPU: 50%) Notebook handbreak ?
0.7a : ~ 1500 FPS (GPU: 76%) falling down to 1050 FPS (GPU: 55%)

OGE3D* Release Build - cant test the same scripts since it does not have the Extensions from ElfScript (for range/preprocessor/..).

As expected, the local variables followed by the global are the fastest. The
static fields are many times faster than in in the vanilla torquescript. With 0.4d it became a real rocket. Only the Dynamic Fields are lame ducs, but i started to change this.

Version 0.5d (RelWithDebug) :

0.5d is slightly slower than 0.5c - i added step to for in range which cause a different stop statement which is safer anyway

Version 0.5c (RelWithDebug) :

added foreach(%i in first..last) and foreach( %i in range start..stop) => 2.69!

Version 0.5a (RelWithDebug) :

opMinusMinus also got direct register write like opPlusPlus
OP_SAVE_LOCAL_VAR_FLT/OP_SAVE_LOCAL_VAR_UINT speed up
removed cleanupdata when int/float is set (good idea?) ==> No memleak!

Version 0.4h🚀 (RelWithDebug) : Direct Threading finished ... let see - compiled with -O2.

Testenviorment:

ThinkPad T570
DE: Xfce4 4.20
WM: Xfwm4 (X11)
CPU: Intel(R) Core(TM) i5-7200U (4) @ 3.10 GHz
GPU: Intel HD Graphics 620 @ 1.00 GHz [Integrated]
Memory: 1.04 GiB / 7.49 GiB (14%)

OGE3D: time ./OhmtalGame_Linux.bin -dedicated -game speedtest

Local variable:

Version 0.4a (RelWithDebug) : 15.913u 0.198s 0:16.16 99.6% 0+0k 0+0io 0pf+0w
OGE3D (current 26-08-05) : 33.268u 0.299s 0:33.61 99.8% 0+0k 0+24io 0pf+0w
Version 0.4d 🚀 (RelWithDebug) : 5.199u 0.000s 0:05.20 99.8% 0+0k 0+0io 0pf+0w
Version 0.4f (RelWithDebug) : 4.509u 0.003s 0:04.52 99.5% 0+0k 0+0io 0pf+0w
Version 0.4g (RelWithDebug) : 4.753u 0.006s 0:04.77 99.5% 0+0k 0+0io 0pf+0w
Version 0.4h (RelWithDebug) : 4.345u 0.004s 0:04.36 99.5% 0+0k 0+0io 0pf+0w
Version 0.5a (RelWithDebug) :
- 3.988u 0.010s 0:04.02 99.2% 0+0k 0+0io 0pf+0w
- 3.664u 0.002s 0:03.68 99.4% 0+0k 0+0io 0pf+0w
- Invalid ::: 3.252u 0.004s 0:03.27 99.3% 0+0k 0+0io 0pf+0w
Version 0.5c (using foreach range)): 2.696u 0.003s 0:02.70 99.6% 0+0k 0+0io 0pf+0w
Version 0.5d (using for range)): 2.876u 0.002s 0:02.89 99.3% 0+0k 0+0io 0pf+0w
Version 0.6a (for) : 3.750u 0.002s 0:03.77 99.4% 0+0k 0+0io 0pf+0w
Version 0.6a (for range) : 2.868u 0.003s 0:02.88 99.3% 0+0k 0+0io 0pf+0w

Mission impossible: Lua 5.5: 1.215u 0.002s 0:01.22 99.1% 0+0k 24+0io 1pf+0w

for OGE3D i had to replace the #define's with global variables!, 200 ms get lost at startup (console)

#define JLOOPS 25
#define ILOOPS 1000000

%sum = 0;
for (%j = 0; %j < JLOOPS; %j++) {
    for (%i = 0; %i < ILOOPS; %i++) {
        %sum ++;
    }
    echo("SUM (++) IS:" SPC %sum);
    for (%i = 0; %i < ILOOPS; %i++) {
        %sum --;
    }
    echo("SUM (--) IS:" SPC %sum);
    %sum = 66;
    echo("set Sum to 66 == " SPC %sum * 1);
    for (%i = 0; %i < ILOOPS; %i++) {
        %sum *= %i + 1;
        %sum /= %i + 1;
    }
    echo("SUM (*/ %i+1) IS:" SPC %sum);
    for (%i = 0; %i < ILOOPS / 2.0; %i++) {
        %ran = getRandomF(%i);
        %sum -= %ran;
        %sum += %ran;
    }
    echo("SUM (rand +-) IS:" SPC %sum);
}
%sum -= 66.0 ;
echo("---------------------");
echo("---------------------");
echo("Final sum should be 0 == ", %sum);
echo("---------------------");
echo("---------------------");

Global variable :

Version 0.4a (RelWithDebug) :17.481u 0.182s 0:17.68 99.8% 0+0k 0+0io 0pf+0w
OGE3D (current 26-08-05) : 34.477u 0.299s 0:34.82 99.8% 0+0k 0+56io 0pf+0w
Version 0.4d 🚀 (RelWithDebug) : 6.622u 0.013s 0:06.64 99.8% 0+0k 0+0io 0pf+0w
Version 0.4f 🚀 (RelWithDebug) : 6.345u 0.003s 0:06.35 99.8% 0+0k 0+0io 0pf+0w
Version 0.4g🚀 (RelWithDebug) : 6.618u 0.009s 0:06.63 99.6% 0+0k 0+0io 0pf+0w
Version 0.4h🚀 (RelWithDebug) : 6.059u 0.003s 0:06.08 99.5% 0+0k 8+0io 0pf+0w
Version 0.5d (using for range): 4.350u 0.004s 0:04.37 99.5% 0+0k 0+0io 0pf+0w
Version 0.6a (using for range): 4.812u 0.007s 0:04.84 99.3% 0+0k 0+0io 0pf+0w

#define JLOOPS 25
#define ILOOPS 1000000

$sum = 0;
for (%j = 0; %j < JLOOPS; %j++) {
    for (%i = 0; %i < ILOOPS; %i++) {
        $sum ++;
    }
    echo("SUM (++) IS:" SPC $sum);
    for (%i = 0; %i < ILOOPS; %i++) {
        $sum --;
    }
    echo("SUM (--) IS:" SPC $sum);
    $sum = 66;
    echo("set Sum to 66 == " SPC $sum * 1);
    for (%i = 0; %i < ILOOPS; %i++) {
        $sum *= %i + 1;
        $sum /= %i + 1;
    }
    echo("SUM (*/ %i+1) IS:" SPC $sum);
    for (%i = 0; %i < ILOOPS / 2.0; %i++) {
        %ran = getRandomF(%i);
        $sum -= %ran;
        $sum += %ran;
    }
    echo("SUM (rand +-) IS:" SPC $sum);
}
$sum -= 66.0 ;
echo("---------------------");
echo("---------------------");
echo("Final sum should be 0 == ", $sum);
echo("---------------------");
echo("---------------------");

Static Float Field

Version 0.4a (RelWithDebug) : 29.871u 0.159s 0:30.06 99.8% 0+0k 0+0io 0pf+0w
OGE3D using tom2DSprite: 192.723u 0.302s 3:13.28 99.8% 0+0k 0+248io 0pf+0w
Version 0.4c 🚀 (RelWithDebug) : 31.224u 0.193s 0:31.45 99.8% 0+0k 0+0io 0pf+0w
Version 0.4d 🚀 (RelWithDebug) : 21.784u 0.009s 0:21.82 99.8% 0+0k 0+0io 0pf+0w
Version 0.4f 🚀 (RelWithDebug) : 21.170u 0.003s 0:21.20 99.8% 0+0k 0+0io 0pf+0w
Version 0.4g 🚀 (RelWithDebug) : 21.010u 0.009s 0:21.06 99.7% 0+0k 0+0io 0pf+0w
Version 0.4h🚀 (RelWithDebug) : 19.916u 0.002s 0:19.97 99.6% 0+0k 8+0io 0pf+0w
Version 0.6a (RelWithDebug) : 12.193u 0.003s 0:12.23 99.6% 0+0k 0+0io 0pf+0w

Here 0.4c is slower than 0.4a (no idea why) - but 0.4b was at the same speed so rocket change did not cause it. Use local var is now the fastest. global is slightly slower than local but still faster than named.

OGE3D :: it's slower than a dynamic field ? - here ElfScript fastpath really shine :)

#define JLOOPS 25
#define ILOOPS 1000000

new PointStorageObject(sto);

sto.x = 0;
for (%j = 0; %j < JLOOPS; %j++) {
    for (%i = 0; %i < ILOOPS; %i++) {
        sto.x ++;
    }
    echo("SUM (++) IS:" SPC sto.x);
    for (%i = 0; %i < ILOOPS; %i++) {
        sto.x --;
    }
    echo("SUM (--) IS:" SPC sto.x);
    sto.x = 66;
    echo("set Sum to 66 == " SPC sto.x * 1);
    for (%i = 0; %i < ILOOPS; %i++) {
        sto.x *= %i + 1;
        sto.x /= %i + 1;
    }
    echo("SUM (*/ %i+1) IS:" SPC sto.x);
    for (%i = 0; %i < ILOOPS / 2.0; %i++) {
        %ran = getRandomF(%i);
        sto.x -= %ran;
        sto.x += %ran;
    }
    echo("SUM (rand +-) IS:" SPC sto.x);
}
sto.x -= 66.0 ;
echo("---------------------");
echo("---------------------");
echo("Final sum should be 0 == ", sto.x);
echo("---------------------");
echo("---------------------");

Static Float Field but with objectID (local var)

Version 0.4a (RelWithDebug) : 34.982u 1.909s 0:36.97 99.7% 0+0k 0+0io 0pf+0w
Version 0.4c 🚀 (RelWithDebug) : 24.884u 0.202s 0:25.12 99.8% 0+0k 0+0io 0pf+0w
Version 0.4d 🚀 (RelWithDebug) : 13.967u 0.003s 0:13.99 99.7% 0+0k 0+0io 0pf+0w
Version 0.4f 🚀 (RelWithDebug) : 14.015u 0.009s 0:14.04 99.7% 0+0k 0+0io 0pf+0w
Version 0.4g 🚀 (RelWithDebug) : 14.773u 0.006s 0:14.80 99.7% 0+0k 0+0io 0pf+0w
Version 0.4h🚀 (RelWithDebug) : 13.582u 0.002s 0:13.62 99.7% 0+0k 0+0io 0pf+0w
Version 0.5d (using for range): 12.359u 0.007s 0:12.40 99.5% 0+0k 0+0io 0pf+0w
Version 0.6a (using for range): 6.578u 0.000s 0:06.60 99.5% 0+0k 0+0io 0pf+0w

So why was this slower (same with a global var):

1. Object is created and %stoObj is set to unsigned int (OP_SAVEVAR_UINT). fine
1. %stoObj.x = 0; It need to lookup the object and does OP_LOADIMMED_STR => OP_SETCURVAR => OP_LOADVAR_STR => OP_SETCUROBJECT So it need to convert the integer to string every time we use %stoObj.xxx.
  SimObject* findObject(const char* name) is used in OP_SETCUROBJECT. But there are alternativ: SimObject* findObject(ConsoleValue*); or SimObject* findObject(const char* name);
1. Concept to proof: OP_LOADIMMED_STR =>

Example call :

% sto.x = 1;
0: OP_LOADIMMED_STR stk=+1 str=1
2: OP_LOADIMMED_IDENT stk=+1 str=sto
5: OP_SETCUROBJECT stk=0
6: OP_SETCURFIELD stk=0 field=x
9: OP_POP_STK stk=-1
10: OP_SAVEFIELD_FASTPATH stk=-1 (curCodeIP: 69)
11: OP_POP_STK stk=-1
12: OP_RETURN_VOID stk=0
% $sto.y = 1;
0: OP_LOADIMMED_STR stk=+1 str=1  << This saves the value "1" to the Stack
2: OP_SETCURVAR stk=0 var=$sto    << loopup variable
5: OP_LOADVAR_STR stk=+1          << set the content of $sto to the stack 
6: OP_SETCUROBJECT stk=0          << now it uses the slow int=>string and the slow findObject(const char*)
7: OP_SETCURFIELD stk=0 field=y
10: OP_POP_STK stk=-1
11: OP_SAVEFIELD_FASTPATH stk=-1 (curCodeIP: 69)
12: OP_POP_STK stk=-1
13: OP_RETURN_VOID stk=0

ElfScript 0.4c changed OP_SETCUROBJECT to 🚀 rocket mode ;)

After my change to directly call findObject ===> 24.916u 0.166s 0:25.11 99.8% 0+0k 0+0io 0pf+0w

But i in script i don't get an advantage out of this. No matter how i try. The assigned var becomes string

% %sto = $sto;
0: OP_SETCURVAR stk=0 var=$sto
3: OP_LOADVAR_STR stk=+1
4: OP_SAVE_LOCAL_VAR_STR stk=0 reg=2  << 💩  my rocket crash here too. $sto is integer . guess this can be fixed 
6: OP_POP_STK stk=-1
7: OP_RETURN_VOID stk=0

fixed assign from global int var .. still call OP_SAVE_LOCAL_VAR_STR but check which type the stack var have and set int/float/string ... my BaseElf starfield changed from named to local var assigned by the global object var raised up from 550 to 900 fps - holy cow.

So next one - DefineEngineMethod( SimObject, getId, S32 ....

let see what the stack says in OP_SAVE_LOCAL_VAR_STR .. .. let me double check but we get an integer --- thats fine!!!

% %sto = sto.getId();                                         
0: OP_PUSH_FRAME stk=0 count=1
2: OP_LOADIMMED_IDENT stk=+1 str=sto
5: OP_PUSH stk=-1
6: OP_CALLFUNC stk=+1 name=getId nspace=(null) callType=MethodCall
12: OP_SAVE_LOCAL_VAR_STR stk=0 reg=2    << 💩  my rocket crash here 
14: OP_POP_STK stk=-1
15: OP_RETURN_VOID stk=0

Fine Fine but to make it complete - i also need to change global vars: OP_SAVEVAR_STR

🚀 Second rocket stage ignited: OP_SAVE_LOCAL_VAR_STR and OP_SAVEVAR_STR*

Code:

#define JLOOPS 25
#define ILOOPS 1000000

%stoObj = new PointStorageObject();

%stoObj.x = 0;
for (%j = 0; %j < JLOOPS; %j++) {
    for (%i = 0; %i < ILOOPS; %i++) {
        %stoObj.x ++;
    }
    echo("SUM (++) IS:" SPC %stoObj.x);
    for (%i = 0; %i < ILOOPS; %i++) {
        %stoObj.x --;
    }
    echo("SUM (--) IS:" SPC %stoObj.x);
    %stoObj.x = 66;
    echo("set Sum to 66 == " SPC %stoObj.x * 1);
    for (%i = 0; %i < ILOOPS; %i++) {
        %stoObj.x *= %i + 1;
        %stoObj.x /= %i + 1;
    }
    echo("SUM (*/ %i+1) IS:" SPC %stoObj.x);
    for (%i = 0; %i < ILOOPS / 2.0; %i++) {
        %ran = getRandomF(%i);
        %stoObj.x -= %ran;
        %stoObj.x += %ran;
    }
    echo("SUM (rand +-) IS:" SPC %stoObj.x);
}
%stoObj.x -= 66.0 ;
echo("---------------------");
echo("---------------------");
echo("Final sum should be 0 == ", %stoObj.x);
echo("---------------------");
echo("---------------------");

Dynamic (string) Field - named again

Version 0.4a (RelWithDebug) : 127.546u 2.019s 2:09.75 99.8% 0+0k 0+0io 0pf+0w
OGE3D using ScriptObject: 137.226u 0.296s 2:17.73 99.8% 0+0k 0+136io 0pf+0w
Version 0.4c 🚀 (RelWithDebug) : 125.068u 1.963s 2:07.38 99.7% 0+0k 1416+0io 8pf+0w
Version 0.4d 🚀 (RelWithDebug) : 124.605u 1.840s 2:06.71 99.7% 0+0k 0+0io 0pf+0w
Version 0.4e 🚀 (RelWithDebug) : 114.157u 1.692s 1:56.06 99.8% 0+0k 0+0io 0pf+0w
Version 0.4f 🚀 (RelWithDebug) : 22.012u 0.000s 0:22.04 99.8% 0+0k 0+0io 0pf+0w
Version 0.4g 🚀 (RelWithDebug) : 22.104u 0.000s 0:22.12 99.9% 0+0k 0+0io 0pf+0w
Version 0.4h🚀 (RelWithDebug) : 21.901u 0.005s 0:21.97 99.6% 0+0k 8+0io 0pf+0w

same test but using local var (docu/speedtest/elf/test_dynamic_fields_localvar.elf)

Version 0.4h🚀 (RelWithDebug) : 14.863u 0.001s 0:14.90 99.7% 0+0k 0+0io 0pf+0w

same test but using local var / for range (docu/speedtest/elf/test_dynamic_fields_forrange.elf)

Version 0.6a (RelWithDebug) : 5.571u 0.004s 0:05.60 99.4% 0+0k 0+0io 0pf+0w

Note: dynX must have TypeF32 set, else it fall back to string.

OGE3D here nearly on same speed, since ElfScript have no Dyanmic Field optimations

0.4c/d This my problem child:

Dynamic fields in the object is one of the coolest features in Elf(Torque)Script.
I could get a better result with local var instead of named now, but is not a big difference.
It store the values in a fast Map as strings but i need the values also as float and int.
value is stored in SimFieldDictionary: char *value;
- void SimFieldDictionary::setFieldValue(StringTableEntry slotName, const char *value)
- SimFieldDictionary::Entry SimFieldDictionary::addEntry(U32 bucket, StringTableEntry slotName, ConsoleBaseType type, char* value)
- const char *SimFieldDictionary::getFieldValue(StringTableEntry slotName)
- void SimFieldDictionary::printFields(SimObject *obj)
- Entry() : slotName(StringTable->EmptyString()), value(NULL), next(NULL), type(NULL) {};
- char *value;

% $sto = new PointStorageObject(sto);$Debug::DumpByteCode=1; 
% $sto.self = $sto;
0: OP_SETCURVAR stk=0 var=$sto
3: OP_LOADVAR_STR stk=+1
4: OP_SETCURVAR stk=0 var=$sto
7: OP_LOADVAR_STR stk=+1
8: OP_SETCUROBJECT stk=0
9: OP_SETCURFIELD stk=0 field=self
12: OP_POP_STK stk=-1
13: OP_SAVEFIELD_FASTPATH stk=-1 (curCodeIP: 69)
14: OP_POP_STK stk=-1
15: OP_RETURN_VOID stk=0

OP_SAVEFIELD_FASTPATH is fine but curObject->findField(curField); is nullptr because it's dynamic field also fine. it enter SimObject::setDataField so we are in slow string mode. It jumps to DynField.. but now it check for if(!array) only. Not if it's also have an entry.

test_dynamic_fields_localvar.elf ==> 106.306u 1.783s 1:48.25 99.8% 0+0k 0+0io 0pf+0w

same on OP_LOADFIELD_STR => if(!array)

test_dynamic_fields_localvar.elf ==> 106.732u 1.770s 1:48.67 99.8% 0+0k 0+0io 0pf+0w

??? slower ??? should be faster .. maybe my laptop is tired ...

==> 114.157u 1.692s 1:56.06 99.8% 0+0k 0+0io 0pf+0w overall 10sec faster ... when i want more i need to add float/int to SimFieldDictionary or instad of value a ConsoleValue object ?! << this would make sense or not ?

replaced const char* value with ConsoleValue mValue - ==> "named" speed test: 114.462u 1.952s 1:56.65 99.7% 0+0k 0+0io 0pf+0w thats fine same as before but with ConsoleValue in place

use types in mValue SAVE: bool pushDataField(StringTableEntry slotName, const char array, ConsoleValue stackP, S32 type_OP_path);

called from OP_SAVEFIELD_FASTPATH

  - implemented status quo
  - wrote new implemenation based on the incomming stack, the data is set.
  - added also ConsoleValueType in SimFieldDictionary::setFieldType (float/int/string)
  - bad: 112.777u 1.854s 1:54.86 99.7%   0+0k 0+0io 0pf+0w

use types in mValue LOAD: - implemented fast set of stack - 113.513u 1.814s 1:55.47 99.8% 0+0k 0+0io 0pf+0w
Find what is missing I made get and set and also added typed but is still slow !!!!
- also added FLT_/INT_ oto this path
- => shit => 115.705u 1.714s 1:57.69 99.7% 0+0k 192+0io 2pf+0w
Everything seams to work fine but somewhere it gets slow down .... i break the test script in parts :
```
  - sto.DynX ++; : 20.060u 0.365s 0:20.46 99.8%    0+0k 0+0io 0pf+0w
  - sto.DynX -- ; : 19.625u 0.398s 0:20.06 99.7%    0+0k 0+0io 0pf+0w
  - */ : 37.276u 0.683s 0:38.00 99.8%    0+0k 0+0io 0pf+0w
  - rand +-: 27.374u 0.296s 0:27.71 99.8%    0+0k 0+0io 0pf+0w
```
All slow nothing special
- setrand only: 10.443u 0.236s 0:10.69 99.8% 0+0k 0+0io 0pf+0w
👾 LOL! I had threaded on in releasebuild where the changes are not implemented so far => 22.338u 0.009s 0:22.39 99.7% 0+0k 0+0io 0pf+0w

🐞 the values are all ZERO ?! << FIXED

22.151u 0.000s 0:22.17 99.9% 0+0k 0+0io 0pf+0w

==> SAME SPEED AS STATIC FIELDS - ROCK AND ROLL

Version 0.4f:

lot of testing << some but will be continued

cleanup compiled eval or completly switch to threaded!

  - [X] make threaded as default add and test the latest changes!!! 
      -  OP_LOADFIELD_UINT
      -  OP_LOADFIELD_FLT
      -  OP_LOADFIELD_STR
      -  OP_SAVEFIELD_FLT:
      -  OP_SAVEFIELD_UINT:
      -  OP_SAVEFIELD_FASTPATH:
  - [~] remove GarbageCollection shit
  - [X] remove fast path ifdefs and make it default 
  - [X] ELFSCRIPT_PREPROCESSOR also as default

Make a memleak check after that !!!!!!!!!!!

Code:

#define JLOOPS 25
#define ILOOPS 1000000

%sto = new PointStorageObject(){
    TypeF32 DynX = 0; // here is the beef!
};

sto.DynX = 0;
for (%j = 0; %j < JLOOPS; %j++) {
    for (%i = 0; %i < ILOOPS; %i++) {
        sto.DynX ++;
    }
    echo("SUM (++) IS:" SPC sto.DynX);
    for (%i = 0; %i < ILOOPS; %i++) {
        sto.DynX --;
    }
    echo("SUM (--) IS:" SPC sto.DynX);
    sto.DynX = 66;
    echo("set Sum to 66 == " SPC sto.DynX * 1);
    for (%i = 0; %i < ILOOPS; %i++) {
        sto.DynX *= %i + 1;
        sto.DynX /= %i + 1;
    }
    echo("SUM (*/ %i+1) IS:" SPC sto.DynX);
    for (%i = 0; %i < ILOOPS / 2.0; %i++) {
        %ran = getRandomF(%i);
        sto.DynX -= %ran;
        sto.DynX += %ran;
    }
    echo("SUM (rand +-) IS:" SPC sto.DynX);
}
sto.DynX -= 66.0 ;
echo("---------------------");
echo("---------------------");
echo("Final sum should be 0 == ", sto.DynX);
echo("---------------------");
echo("---------------------");

Speed template saved here ;)

Script	time
test_localvar.elf
test_global.elf
test_static.elf
test_dynamic.elf
test_vector_components.elf

FilesExpand file tree

BetterSpeedTest.md

Latest commit

History

BetterSpeedTest.md

File metadata and controls

ElfScript Simple Speed Test:

Goal: get the (close to or better) speed as lua (vanilla - without jit) on "localvar test"

Version 0.8

Version 0.7g - stability tests for first release -

Version 0.7f

Version 0.7e

Version 0.7 d

Version 0.7 c

Version 0.7 b

Version 0.7 a

Version 0.6f :

Version 0.6g:

Version 0.6f :

Version 0.6e :

Version 0.6d :

Summery ElfScript 0.6d vs TorqueScript

Version 0.6c :

Version 0.6b :

Version 0.6a :

docu/speedtest/elf/test_localvar_v6.elf

docu/speedtest/elf/test_localvar_foreach.elf

docu/speedtest/elf/test_globalvar_forrange.elf

docu/speedtest/elf/test_dynamic_fields_forrange.elf

docu/speedtest/elf/test_static_fields_localvar_forrange.elf

docu/speedtest/elf/test_vector_components.elf

docu/speedtest/elf/test_vector_components_global.elf

docu/speedtest/elf/counter.elf << says nothing, just for fun ;)

docu/speedtest/elf/test_for.elf

Crazy Elf: assets/modules/StarField.elf

Version 0.5d (RelWithDebug) :

Version 0.5c (RelWithDebug) :

Version 0.5a (RelWithDebug) :

Version 0.4h🚀 (RelWithDebug) : Direct Threading finished ... let see - compiled with -O2.

Testenviorment:

Local variable:

Global variable :

Static Float Field

Static Float Field but with objectID (local var)

So why was this slower (same with a global var):

ElfScript 0.4c changed OP_SETCUROBJECT to 🚀 rocket mode ;)

🚀 Second rocket stage ignited: OP_SAVE_LOCAL_VAR_STR and OP_SAVEVAR_STR*

Code:

Dynamic (string) Field - named again

same test but using local var (docu/speedtest/elf/test_dynamic_fields_localvar.elf)

same test but using local var / for range (docu/speedtest/elf/test_dynamic_fields_forrange.elf)

Note: dynX must have TypeF32 set, else it fall back to string.

0.4c/d This my problem child:

Version 0.4f:

Code: