ASM Frontend

The ASM frontend is a strict GPU-friendly subset frontend, not a generic JVM decompiler.

Its intended architecture is:

your AST -> GPU-friendly ASM -> JavaToGpu ASM frontend -> IR -> OpenCL

Use It For

• compiler integrations
• generator pipelines
• canonical structured control flow

Current Supported Direction

Treat the ASM input as a purpose-built subset with:

• static methods
• explicit temporary-heavy lowering
• predictable labels and control flow
• INVOKESTATIC helper and intrinsic calls
• supported wrappers, vectors, and structs
• canonical if / else, while, do-while, and switch

Not For

• arbitrary JVM bytecode
• virtual dispatch
• exception-heavy code

Practical Rule

If your generator has a choice between:

• compact JVM bytecode
• explicit GPU-friendly bytecode

choose the second one.

JavaToGpu is designed to lift canonical bytecode shapes back into structured IR, not to reverse-engineer every possible JVM pattern.

Optional Helper APIs

The project also contains optional helper layers for generator authors:

• a builder-style API for canonical structured ASM generation
• a tiny experimental neutral AST that lowers into canonical ASM

These helpers are conveniences, not mandatory user-facing API.

What To Generate

Preferred patterns:

• explicit locals
• simple stack usage
• direct numeric opcodes
• direct branch layouts
• switches with predictable case bodies

Avoid:

• dup_x* stack tricks
• dynamic dispatch
• exception tables
• non-canonical merged CFG shapes

See Also

• Kernel Programming Model
• Examples and References