Paul Rigge

Remove proc inlining

2

Proc inlining as currently envisioned is complex, needs to re-implement some of what codegen does, and in many cases produces procs which cannot be pipelined well. With multi-proc codegen #950...

### What's hard to do? (limit 100 words) It's hard to write reusable components like an arbiter because you ideally want it to be type-generic, e.g. ``` proc Arb {...

### What's hard to do? (limit 100 words) Currently, module_builder puts functions in each module. With multi-proc codegen, this means you end up redefining a bunch of functions in many...

### What's hard to do? (limit 100 words) You might want to do any of the following: - Generate Verilog but avoid all SystemVerilog keywords b/c the file will be...

### What's hard to do? (limit 100 words) If you have a DUT proc, you might want to use a DSLX model (ex: a memory model) and attach it to...

**Describe the bug** Extern types (e.g. `#[sv_type(...)]`) don't codegen correctly for array types. **To Reproduce** Given an sv type like `typedef logic[7:0][3] u8_3_t;`: ``` #[sv_type("u8_3_t")] type u8_3_t = u8[3]; fn...

Currently, we have protos in channels that map "logical" (e.g. `VALID`) ports to "physical" (e.g. `chan_vld`) port names. Signature generation goes through the channels and populates the signature, validates that...

JIT packed wrapper should use loops for arrays

1

Currently, the packed wrapper generates loads/stores for each array element, which for large arrays can cause LLVM to fall down. Packing/unpacking arrays should use loops.

`Node::SetId()` has a TODO to drop it. Ideally, this would be private to the parser or not even exist. Node IDs are usually OK to treat as immutable, but they...

Node DFS and Mutation is Tricky

1

As I write this issue, I am concurrently landing a change which makes Node DFS iterative rather than recursive. The discussion in code review revealed some ways in which a...