b2llvm.org

#+ TITLE: b2llvm notes

Adding support for records and structures

Create a specific git branch “records”

Understand how records and structures are represented in BXML files.

Structures (types)

BXML: {structure}: <Struct>{field}*</Struct> {field}: <Record_Item label={name}>{type}</Record_Item>

AST: {structure}: Structure > elements : seq Field {field}: Field > id: {name} > type : {type}

Records (values)

BXML: {record}: <Record><Attributes><TypeInfo>{type}</TypeInfo></Attributes>{rec_field}*</Record> {type}: see {structure} {rec_field}: <Record_Item label={name}>{expression}</Record_Item>

AST: {record}: Record > type:{structure} > values: seq {expression}

Record item

BXML: {select}: <Binary_Expression operator=”&apos;”>{base}{elem}</Binary_Expression> {base}: {expression} (of type structure) {elem}: <Identifier value={name}></Identifier>

AST: {select}: RecField > base:{expression} (of type structure) > elem:{name}

Identify and specify the needed families of AST nodes.

Artifacts: DOCUMENTATION/specification.tex

1. record: i.e. the type of data for records
2. struct: i.e. the actual records
3. field selection
4. record assignment: assignment of full record

Get insight on the LLVM code that needs to be generated through

examples in C with struct. Artifacts: C/pg5.c, C/pg5.llvm, possibly new C and LLVM files.

Specify the code generation rules

Artifacts: DOCUMENTATION/specification.tex

1. record: aggregate type
2. struct: aggregate value
3. field selection: combination of getelementptr and load
4. record assignment: beware of atomicity issue:

r := struct(x_1: e_1; … x_n: v_n) is essentially

LET v_1 = e_1 and … e_n = v_n IN r’x1 := v_1; …; r’x_n := v_n END

Implement code generation.

1. update Makefile and coverage script to include the examples
2. ast.py: implement new families of AST nodes
3. printer.py: implement pretty-printer for the new nodes
4. loadbxml.py: extend BXML loader to include new constructs
5. translate.py: implement code generation rules