kmath/kmath-ast/README.md

276 lines
8.1 KiB
Markdown
Raw Normal View History

2021-03-22 14:32:08 +03:00
# Module kmath-ast
2020-06-15 11:02:13 +03:00
Extensions to MST API: transformations, dynamic compilation and visualization.
2024-03-27 09:11:12 +03:00
- [expression-language](src/commonMain/kotlin/space/kscience/kmath/ast/parser.kt) : Expression language and its parser
- [mst-jvm-codegen](src/jvmMain/kotlin/space/kscience/kmath/asm/asm.kt) : Dynamic MST to JVM bytecode compiler
- [mst-js-codegen](src/jsMain/kotlin/space/kscience/kmath/estree/estree.kt) : Dynamic MST to JS compiler
- [rendering](src/commonMain/kotlin/space/kscience/kmath/ast/rendering/MathRenderer.kt) : Extendable MST rendering
2021-03-22 14:32:08 +03:00
## Artifact:
2024-02-18 15:05:56 +03:00
The Maven coordinates of this project are `space.kscience:kmath-ast:0.4.0-dev-3`.
2021-03-22 14:32:08 +03:00
**Gradle Kotlin DSL:**
2024-03-27 09:11:12 +03:00
2021-03-22 14:32:08 +03:00
```kotlin
repositories {
maven("https://repo.kotlin.link")
mavenCentral()
2021-03-22 14:32:08 +03:00
}
dependencies {
2024-02-18 15:05:56 +03:00
implementation("space.kscience:kmath-ast:0.4.0-dev-3")
2021-03-22 14:32:08 +03:00
}
```
## Parsing expressions
2024-03-27 09:11:12 +03:00
In this module there is a parser from human-readable strings like `"x^3-x+3"` (in the more
specific [grammar](reference/ArithmeticsEvaluator.g4)) to MST instances.
Supported literals:
2024-03-27 09:11:12 +03:00
1. Constants and variables (consist of latin letters, digits and underscores, can't start with digit): `x`, `_Abc2`.
2. Numbers: `123`, `1.02`, `1e10`, `1e-10`, `1.0e+3`—all parsed either as `kotlin.Long` or `kotlin.Double`.
Supported binary operators (from the highest precedence to the lowest one):
2024-03-27 09:11:12 +03:00
1. `^`
2. `*`, `/`
3. `+`, `-`
Supported unary operator:
2024-03-27 09:11:12 +03:00
1. `-`, e. g. `-x`
2024-03-27 09:11:12 +03:00
Arbitrary unary and binary functions are also supported: names consist of latin letters, digits and underscores, can't
start with digit. Examples:
1. `sin(x)`
2. `add(x, y)`
2020-12-22 12:00:51 +03:00
## Dynamic expression code generation
### On JVM
`kmath-ast` JVM module supports runtime code generation to eliminate overhead of tree traversal. Code generator builds a
special implementation of `Expression<T>` with implemented `invoke` function.
For example, the following code:
```kotlin
import space.kscience.kmath.asm.compileToExpression
2022-03-31 22:23:34 +03:00
import space.kscience.kmath.operations.DoubleField
2022-03-31 22:23:34 +03:00
"x^3-x+3".parseMath().compileToExpression(DoubleField)
```
&mldr; leads to generation of bytecode, which can be decompiled to the following Java class:
```java
2022-03-31 22:23:34 +03:00
import java.util.*;
import kotlin.jvm.functions.*;
import space.kscience.kmath.asm.internal.*;
import space.kscience.kmath.complex.*;
import space.kscience.kmath.expressions.*;
public final class CompiledExpression_45045_0 implements Expression<Complex> {
2020-12-22 12:00:51 +03:00
private final Object[] constants;
public Complex invoke(Map<Symbol, ? extends Complex> arguments) {
Complex var2 = (Complex)MapIntrinsics.getOrFail(arguments, "x");
return (Complex)((Function2)this.constants[0]).invoke(var2, (Complex)this.constants[1]);
}
}
```
2022-03-31 22:23:34 +03:00
For `LongRing`, `IntRing`, and `DoubleField` specialization is supported for better performance:
```java
import java.util.*;
import space.kscience.kmath.asm.internal.*;
import space.kscience.kmath.expressions.*;
public final class CompiledExpression_-386104628_0 implements DoubleExpression {
private final SymbolIndexer indexer;
public SymbolIndexer getIndexer() {
return this.indexer;
}
public double invoke(double[] arguments) {
double var2 = arguments[0];
return Math.pow(var2, 3.0D) - var2 + 3.0D;
}
public final Double invoke(Map<Symbol, ? extends Double> arguments) {
double var2 = ((Double)MapIntrinsics.getOrFail(arguments, "x")).doubleValue();
return Math.pow(var2, 3.0D) - var2 + 3.0D;
}
}
```
2024-03-27 09:11:12 +03:00
Setting JVM system property `space.kscience.kmath.ast.dump.generated.classes` to `1` makes the translator dump class
files to program's working directory, so they can be reviewed manually.
#### Limitations
2024-03-27 09:11:12 +03:00
- The same classes may be generated and loaded twice, so it is recommended to cache compiled expressions to avoid class
loading overhead.
- This API is not supported by non-dynamic JVM implementations like TeaVM or GraalVM Native Image because they may not
support class loaders.
2020-12-22 12:00:51 +03:00
### On JS
A similar feature is also available on JS.
```kotlin
import space.kscience.kmath.expressions.Symbol.Companion.x
import space.kscience.kmath.expressions.*
import space.kscience.kmath.operations.*
import space.kscience.kmath.estree.*
MstField { x + 2 }.compileToExpression(DoubleField)
2020-12-22 12:00:51 +03:00
```
The code above returns expression implemented with such a JS function:
```js
var executable = function (constants, arguments) {
return constants[1](constants[0](arguments, "x"), 2);
2020-12-22 12:00:51 +03:00
};
```
JS also supports experimental expression optimization with [WebAssembly](https://webassembly.org/) IR generation.
Currently, only expressions inside `DoubleField` and `IntRing` are supported.
```kotlin
import space.kscience.kmath.expressions.Symbol.Companion.x
import space.kscience.kmath.expressions.*
import space.kscience.kmath.operations.*
import space.kscience.kmath.wasm.*
MstField { x + 2 }.compileToExpression(DoubleField)
```
2021-04-23 04:49:01 +03:00
An example of emitted Wasm IR in the form of WAT:
```lisp
2022-03-31 22:23:34 +03:00
(func \$executable (param \$0 f64) (result f64)
(f64.add
2022-03-31 22:23:34 +03:00
(local.get \$0)
(f64.const 2)
)
)
```
#### Limitations
2020-12-22 12:00:51 +03:00
- ESTree expression compilation uses `eval` which can be unavailable in several environments.
- WebAssembly isn't supported by old versions of browsers (see https://webassembly.org/roadmap/).
2021-03-25 19:57:47 +03:00
## Rendering expressions
kmath-ast also includes an extensible engine to display expressions in LaTeX or MathML syntax.
2021-03-25 19:57:47 +03:00
Example usage:
```kotlin
import space.kscience.kmath.ast.*
import space.kscience.kmath.ast.rendering.*
import space.kscience.kmath.misc.*
2021-03-25 19:57:47 +03:00
@OptIn(UnstableKMathAPI::class)
2021-03-25 19:57:47 +03:00
public fun main() {
2021-05-08 14:36:37 +03:00
val mst = "exp(sqrt(x))-asin(2*x)/(2e10+x^3)/(12)+x^(2/3)".parseMath()
2021-03-25 19:57:47 +03:00
val syntax = FeaturedMathRendererWithPostProcess.Default.render(mst)
val latex = LatexSyntaxRenderer.renderWithStringBuilder(syntax)
println("LaTeX:")
println(latex)
println()
val mathML = MathMLSyntaxRenderer.renderWithStringBuilder(syntax)
println("MathML:")
println(mathML)
}
```
Result LaTeX:
2021-03-25 19:57:47 +03:00
2024-03-27 09:11:12 +03:00
$$\operatorname{exp}\\,\left(\sqrt{x}\right)-\frac{\frac{\operatorname{arcsin}\\,\left(2\\,x\right)
}{2\times10^{10}+x^{3}}}{12}+x^{2/3}$$
2021-03-25 19:57:47 +03:00
2021-05-08 14:36:37 +03:00
Result MathML (can be used with MathJax or other renderers):
2021-03-25 19:57:47 +03:00
<details>
2021-03-25 19:57:47 +03:00
```html
<math xmlns="https://www.w3.org/1998/Math/MathML">
<mrow>
<mo>exp</mo>
<mspace width="0.167em"></mspace>
<mfenced open="(" close=")" separators="">
<msqrt>
<mi>x</mi>
</msqrt>
</mfenced>
<mo>-</mo>
<mfrac>
<mrow>
<mfrac>
<mrow>
<mo>arcsin</mo>
<mspace width="0.167em"></mspace>
<mfenced open="(" close=")" separators="">
<mn>2</mn>
<mspace width="0.167em"></mspace>
<mi>x</mi>
</mfenced>
</mrow>
<mrow>
<mn>2</mn>
<mo>&times;</mo>
<msup>
<mrow>
<mn>10</mn>
</mrow>
<mrow>
<mn>10</mn>
</mrow>
</msup>
<mo>+</mo>
<msup>
<mrow>
<mi>x</mi>
</mrow>
<mrow>
<mn>3</mn>
</mrow>
</msup>
</mrow>
</mfrac>
</mrow>
<mrow>
<mn>12</mn>
</mrow>
</mfrac>
2021-05-08 14:36:37 +03:00
<mo>+</mo>
<msup>
<mrow>
<mi>x</mi>
</mrow>
<mrow>
<mn>2</mn>
<mo>/</mo>
<mn>3</mn>
</mrow>
</msup>
</mrow>
</math>
2021-03-25 19:57:47 +03:00
```
</details>
It is also possible to create custom algorithms of render, and even add support of other markup languages
2021-03-25 19:57:47 +03:00
(see API reference).