DISCLAIMER BLAST is new, partly experimental and under active development, bugs are to be expected and please create an issue for any issue there might be so we can handle it as quickly as possible.
Issues: /~https://github.com/nijnstein/BLAST-Documentation/issues
-A blast script is a collection of statements seperated by dotcomma's and grouped by parenthesis.
Blastscript sample:
#input position float3 0 0 0
position.x = min(100, position.x + 0.001); Full C# GameObject example:
public class Single2 : MonoBehaviour
{
BlastScript script;
void Start()
{
Blast.Initialize();
// prepare the script once
script = BlastScript.FromText("#input position float3 0 0 0\n position.x = min(100, position.x + 0.001);");
BlastError result = script.Prepare();
if (result != BlastError.success)
{
Debug.LogError($"Error during script compilation: {result}");
script = null;
}
}
void Update()
{
if (script == null) return;
// the each frame, update current position
script["position"] = (float3)transform.position;
// execute the script
BlastError result = script.Execute();
if (result == BlastError.success)
{
// and set value
transform.position = (float3)script["position"];
}
else
{
Debug.LogError($"Error during script execution: {result}");
}
}
}| keyword | description |
|---|---|
| case | switch case compound |
| default | default switch case compound |
| else | else compound of if then else statement sequence |
| if | if statement |
| input | defines an input to the script |
| for | for loop |
| output | defines an output of the script |
| return | return; terminate execution |
| switch | switch statement |
| then | then compound of if then else statement sequence |
| while | while loop |
| yield | yield execution |
| function | inline functions |
| Keyword | Value |
|---|---|
| false | 0.0 |
| true | default: 1.0, anything other then 0.0 matches to true |
Seperators: ; , ( )
Arithmetic: + - * /
Boolean: & | ^ !
Assingment: =
Comparisons: = < <= > >= !=
Indexing: .[x|y|z|w|r|g|b|a]
First off, there is no constant analysis used for removing unneeded constant statements, future versions might decide to remove constant statements leading to unused values but for now blast wont botter.
Blast uses bytecode operations to index a datasegment for variable and stack data. Any constant value that is not internally mapped to a bytecode will be packaged in the datasegment. Blast map's several often used values like 0, 1, 2, pi and others to byte sized opcodes.
In SSMD packaging mode constants are inlined into the code segment to avoid duplicate data in many datasegments at the cost of some performance while in normal packaging mode constant data is always referenced as if it was a data element. Any constant referenced multiple times is defined once in the code segment after which it is referenced by instructions: constant_ref | constant_long_ref
This has the benefit that constant data elements dont substract from the usable variable element count in the script.
| constant | source | value |
|---|---|---|
| framecount | UnityEngine.Time.frameCount | frame count since start |
| fixedtime | UnityEngine.Time.fixedTime | fixed time at start of interval |
| time | UnityEngine.Time.time | time at start of interval |
| fixeddeltatime | UnityEngine.Time.fixedDeltaTime | delta of fixed interval |
| deltatime | UnityEngine.Time.deltaTime | delta of normal interval = frametime |
| epsilon | math.EPSILON | |
| infinity | math.INFINITY | |
| negative_infinity | -math.INFINITY | |
| nan | math.NAN | |
| min_value | math.FLT_MIN_NORMAL |
An expression is a set of literals and identifiers combined with operators. Parenthesis can be used to influence the order of operation.
// numeric expressions
f = 1 * 10 * 3 * (3 + 4 * 2);
f = 10 + 5 / 1 * 2;
f = 1 * 4 + 3;
f = 4 * -9.3
// boolean expressions:
b = 1 & 0 | 1;
b = -1 & 0 | 123.22;
b = true & false | true;
// bit operations:
flags = 00100000_00000000_00000000_00000000;
if(get_bits(flags, mask)) then (set_bit(flags, 2, 0););
Vectors are defined as sets of values enclosed in parenthesis without operators combining them. A comma is used to seperate elements to remove any ambiguity when using negative constants.
Inline vector definition: (1.2 2.1, -3.3)
a = (1 2 3 4) * (5 6 7, -8);
To get the component of a vector it can be indexed: a.x
The only datatype fully supported is the float of vectorsize 1 to 4. There is partial support for bool32 operations which index a single dataelement as 32 bit flags.
- Future versions will add integer datatypes.
supported from v1.0.4
CData isnt a real datatype but just constant length data that later can be interpreted to be a fixed array of the requested type. CDATA sections will provide backing memory for constantly sized arrays. When located in the code segment they are assumed to be constant during the current frame and shared among all executions of that script.
** v1.1 ** Later versions will allow CDATA sections in the datasegment (restricted to max 127 elements) or externally pointed to by a native pointer.
** v1.0.5 ** The SSMD interpretor will be able to loop over CDATA sections as they are constantly sized as long as the iterator is not conditionally increased and the loop condition itself is constant. As the data is expected to be constant during the current frame large optimizations become possible for the ssmd interpretor as it can expand its results over all components in vectors that result in constant operations based on the fact that CDATA is constant for the current calculation frame. (ie: deltatime is the same in all scripts during the whole frame)
Scriptcode:
#cdata a 14.42 344.4 54444.345 22190
#cdata msg 'cmd_or_msg_or_log'
b = size(a) - 1; # as float is the default datatype, size returns 5 while the bytesize is 20
b = a[b]; # again, float = default, and a is indexed as float[]
if(b = 22190) then
(
send(msg); # send data to some sink, for now the log
)Bytecode:
000| 028 040 074 000 016 082 184 102 065 051
010| 051 172 067 088 172 084 071 000 092 173
020| 070 074 000 017 099 109 100 095 111 114
030| 095 109 115 103 095 111 114 095 108 111
040| 103 001 128 073 075 000 043 009 085 000
050| 003 128 060 075 000 052 128 026 012 024
060| 128 020 129 025 024 076 075 000 046 025
Somewhat readable assembly
jump 40 cdata[16] #082 #184 #102 #065 #051 #051 #172 #067
#088 #172 #084 #071 #000 #092 #173 #070 cdata[17] #099
#109 #100 #095 #111 #114 #095 #109 #115 #103 #095 #111
#114 #095 #108 #111 #103 set b size cdataref #000 #043
- 1 nop setf b idx_N #075 #000 52 b jz 12 ( b = 22190 )
( send cdataref #000 #046 )
CData can be defined in 2 distinct ways:
Constant CDATA v1.0.4 Constant cdata is data that is inlined into the codesegment and for normal use should be constant:
#cdata a 14.42 344.4 54444.345 22190
#cdata a 'abcd'
#cdata a 11110111_11101110_10111111_00110011 11110111_11101110_10111111_00110011
Variable CDATA v1.0.5 Variable cdata is data that is allocated in the datasegment and can be set as parameter to the script through input defines.
#input msg cdata 'some message for exceptional events'
v1.0.5 CDATA arrays are stored in the smallest bytesize allowed without losing precision beyond the constantepsilon value configurable through compiler options. During compilation the compiler will decide the lowest usable bytesize for the values in the constant data. Users an also type cdata as to avoid this behaviour.
Encoding types:
| name | backing | variable type |
|---|---|---|
| None | float32 | ------------- |
| u8_fp32 | unsigned byte | float32 |
| s8_fp32 | signed byte | float32 |
| fp8_fp32 | 8 bit floating point | float32 |
| fp16_fp32 | 16 half | float32 |
| fp32_fp32 | float32 | float32 |
| u8_i32 | unsigned byte | int32 |
| i8_i32 | signed byte | int32 |
| i16_i32 | signed short | int32 |
| i32_i32 | signed int32 | int32 |
v1.0.4
CData defined to be inlined as constant is actually shared among all executions of that script using the same scriptpackage.
For now this is allowed by default but we might decide to force the user to enable this behaviour through compileroptions|defines.
- note: the optimizer will replace sequences with its equivevalent function whenever possible providing shorter code and faster execution due to reduced control flow.
3 3 yield yield nop
4 4 pop pop nop
5 5 peek peek nop
6 6 seed ex_op seed
7 7 push push nop
8 8 pushf pushf nop
9 9 pushc pushc nop
10 10 pushv pushv nop
11 11 debug ex_op debug
12 12 debugstack ex_op debugstack
13 13 validate ex_op validate
32 32 return ret nop
33 33 abs abs nop
34 34 min min nop
35 35 max max nop
36 36 mina mina nop
37 37 maxa maxa nop
38 38 any any nop
39 39 all all nop
40 40 adda adda nop
41 41 suba suba nop
42 42 diva diva nop
43 43 mula mula nop
44 44 fma fma nop
45 45 fmod ex_op fmod
46 46 trunc trunc nop
47 47 csum csum nop
48 48 select select nop
49 49 random random nop
50 50 idx index_x nop
51 51 idy index_y nop
52 52 idz index_z nop
53 53 idw index_w nop
54 54 idxn index_n nop
55 55 expand2 expand_v2 nop
56 56 expand3 expand_v3 nop
57 57 expand4 expand_v4 nop
58 58 sin ex_op sin
59 59 cos ex_op cos
60 60 tan ex_op tan
61 61 atan ex_op atan
62 62 atan2 ex_op atan2
63 63 cosh ex_op cosh
64 64 sinh ex_op sinh
65 65 degrees ex_op degrees
66 66 rad ex_op radians
67 67 sqrt ex_op sqrt
68 68 rsqrt ex_op rsqrt
69 69 pow ex_op pow
70 70 normalize ex_op normalize
71 71 saturate ex_op saturate
72 72 clamp ex_op clamp
73 73 log2 ex_op log2
74 74 log10 ex_op log10
75 75 log ex_op logn
76 76 exp ex_op exp
77 77 exp10 ex_op exp10
78 78 cross ex_op cross
79 79 dot ex_op dot
80 80 ceil ex_op ceil
81 81 floor ex_op floor
82 82 frac ex_op frac
83 83 lerp ex_op lerp
84 84 slerp ex_op slerp
85 85 nlerp ex_op nlerp
86 86 unlerp ex_op unlerp
87 87 remap ex_op remap
88 88 ceillog2 ex_op ceillog2
89 89 floorlog2 ex_op floorlog2
90 90 ceilpow2 ex_op ceilpow2
91 91 set_bits ex_op set_bits
92 92 set_bit ex_op set_bit
93 93 get_bits ex_op get_bits
94 94 get_bit ex_op get_bit
95 95 count_bits ex_op count_bits
96 96 reverse_bits ex_op reverse_bits
97 97 shl ex_op shl
98 98 shr ex_op shr
99 99 rol ex_op rol
100 100 ror ex_op ror
101 101 lzcnt ex_op lzcnt
102 102 tzcnt ex_op tzcnt
103 103 zero zero nop
104 104 clear_bits zero nop
105 105 reinterpret_int32 ex_op reinterpret_int32
106 106 reinterpret_bool32 ex_op reinterpret_bool32
107 107 reinterpret_float32 ex_op reinterpret_float
108 108 send ex_op send
109 109 size size nop
110 110 length ex_op length
111 111 lengthsq ex_op lengthsq
112 112 square ex_op square
113 113 distance ex_op distance
114 114 distancesq ex_op distancesq
115 115 reflect ex_op reflect
116 116 project ex_op project
117 117 up ex_op up
118 118 down ex_op down
119 119 forward ex_op forward
120 120 back ex_op back
121 121 left ex_op left
122 122 right ex_op right
123 123 eulerxyz ex_op EulerXYZ
124 124 eulerxzy ex_op EulerXZY
125 125 euleryxz ex_op EulerYXZ
126 126 euleryzx ex_op EulerYZX
127 127 eulerzxy ex_op EulerZXY
128 128 eulerzyx ex_op EulerZYX
129 129 euler ex_op Euler
130 130 quaternionxyz ex_op QuaternionXYZ
131 131 quaternionxzy ex_op QuaternionXZY
132 132 quaternionyxz ex_op QuaternionYXZ
133 133 quaternionyzx ex_op QuaternionYZX
134 134 quaternionzxy ex_op QuaternionZXY
135 135 quaternionzyx ex_op QuaternionZYX
136 136 quaternion ex_op Quaternion
137 137 lookrotation ex_op LookRotation
138 138 lookrotationsafe ex_op LookRotationSafe
139 139 rotate ex_op Rotate
140 140 angle ex_op Angle
141 141 mul ex_op Mul
#input flags bool32 11001100_11001100_11001100_11001100;
# operations that return true|false
a = count_bits(flags);
a = lzcnt(flags);
a = tzcnt(flags);
a = get_bit(flags, 1);
a = get_bits(flags, 11111111_00000000_11111111_00000000, true);
# both any and all support multiple parameters as long as all are bool32
a = any(flags);
a = all(flags);
# operations that work in place
set_bit(flags, 1, true);
set_bit(flags, 2, a);
set_bits(flags, 11111111_00000000_11111111_00000000, true);
reverse_bits(flags);
rol(flags, 1);
ror(flags, 1);
shr(flags, 1);
shl(flags, 1);
# reinterpret the data at a as a bool32, set metadata of a to type Bool32 and vectorsize 1
reinterpret_bool32(a);Blast allows the user to define inline functions, a function consists of the function keyword, a parameterlist (that even if empty may not be ommited) and a body:
function f1(a, b)
(
c = a + b;
return (a + b) / c;
);Blast functions will however work differently then in for example C#. There is NO scope as the function is inlined at the callsite. It operates directly on the used variables and it does not make any data copies, making all parameters passed by reference. Returning data will also work directly on the same data elements. Assigning c within a function assigns c in global scope.
Inlined Functions may call other inline functions.
Inlined Functions may not declare new variables
This should be viewed as a utility function, for recurring needs please consider implementing user defined external function calls that can be called natively.
Blast uses function pointers to connect to other parts of its environment, these will be supplied with the following data by the BLAST engine:
| data | source | description |
|---|---|---|
| engine | blast | pointer to engine data, contains constants, functionpointers and randomizer root. |
| environment | user | optional pointer to native data containing data to be used by all scripts |
| caller | user | optional pointer to native data containing all data supplied by the callsite |
| parameters | blast | blast will call the external function using the parameters supplied to it by script |
supports only single floats, future versions will expand on this
Blast will use all components of the supplied input parameters to satify the functions parameter list:
a = (1 1 1);
external(a); Is equivalent to:
a = (1 1 1);
external(a.x, a.y, a.z); // full paramater list (including enviroment data), up until 16 float params
public delegate float BlastDelegate_f1(IntPtr engine, IntPtr data, IntPtr caller, float a);
public delegate float BlastDelegate_f11(IntPtr engine, IntPtr data, IntPtr caller, float a, float b);
public delegate float BlastDelegate_f111(IntPtr engine, IntPtr data, IntPtr caller, float a, float b, float c);
// ......
public delegate float BlastDelegate_f1111(IntPtr engine, IntPtr data, IntPtr caller, float a, float b, float c, float d...... , float 16);
// short external defines without environment pointers
public delegate float BlastDelegate_f0_s();
public delegate float BlastDelegate_f1_s(float a);
public delegate float BlastDelegate_f11_s(float a, float b);Blast uses the input and output keywords to define input or output variables. These will be prepared in the compiled package for fast IO syncs, the sync method however depends heavily on the packaging mode and its usage. Samples for each mode (Normal, SSMD) will be provided with the package.
Blast determines the size of the stack to allocate as follows:
- First; if stack_size is defined as:
#define stack_size 12blast will fix the stacksize to 12 bytes, additional space will be added if yield is to be supported. - Second; if compileroptions specify EstimateStack then during compilation the package is executed once using default data and the stack size is determined from analysis of the stack. Note that this will force an extra compilation step if the packagemode is ssmd as currently stack estimation is only supported by the normal package mode which will always use more stack then the ssmd packager. (which should not matter too much if the no-stack option is used to use the threadlocal stack in the interpretor and not actually package the stack in the package memory)
- Third; if by now no stacksize is known, blast will count overlapping pairs of push-pop pairs and estimates stack from that. If no push command is used then the stacksize will be set to 0; If Yield is used in the package, 20 bytes are additionally allocated to support stacking internal state on yield.
Use methods provided by the BlastScript class to read or write to script variables exposed via input and/or output, it is not necessary to use input nor output defines but doing so forces their memoryorder regardless the code written and should be considered good practice, the can be omitted but you will have to directly write to the datasegments to set variables by name.
BlastScript:
#input a float3 3 3 3
#input b float 3
a = b * a; Unity:
// create script, set a
BlastScript script = BlastScript.FromResource("/bs");
script.Set("a", new Vector3(4, 4, 4));
script.Set("b", 8);
script.Execute();
Vector3 result = (Vector3)script["a"];The script's datasegment is directly mapped against an IComponentData record:
For a script defined as:
#input a float3 3 3 3
#input b float 3
a = b * a; The IComponentData should be like:
public struct data : IComponentData
{
public float3 ValueA;
public float ValueB;
}The script can then very efficiently use this as a datasegment.
If yield is used, all memory used for script execution must map to the IComponentData struct. This includes any stackmemory that is used: yield reserves 20 bytes in each segment to allow it to store its internal execution state.
The scripts datasegment is packed as with entities but not matched to any structure, it might also not be complete depending on any constant data in it. In SSMD mode there is no stack allocated in the datasegment and it can inline any constant data in the code datastream to maximize memory efficiency when running many billions of scripts each second.
Yield is not supported yet.
During development we have the need to test a lot and there is some support for automatic testing in the form of validation script defines: #validate a 1 These allow the script to set values that blast can match to the output of the script given default input.
Blast can (depending on compilersettings) then validate the script during compilation in the same run it uses to determine the stackspace it needs to reserve in the compiled package. It proved extremely usefull during development and it possibly can catch bugs early in deployment.
Example script with validation defines:
#define result_1 11111
#define result_2 22222
a = 10;
b = 2;
switch(a > b)
(
case 1:
(
e = result_1;
)
default:
(
e = result_2;
)
);
#validate e 11111 To validate values in runtime use the Validate(id, value) function. This will compare the value to the value of the id and if different it will raise an error during interpretation in TRACE and DEBUG modes. This function will not be compiled in release mode.
A typical testing script in our unittests using Validate()
#input a float 0
b = 00000001_00000000_00000000_00000000;
a = get_bit(b, 7);
debug(a);
debug(b);
validate(a, 1);