ast: Add support for array-to-array assignment

[apullin]

This commit adds support for SystemVerilog array-to-array assignment operations that were previously unsupported:

1. Direct array assignment: b = a;
2. Array ternary expressions: out = sel ? a : b;

Both single-dimensional and multi-dimensional unpacked arrays are supported. The implementation expands these array operations during AST simplification into element-wise assignments.

Example of now-supported syntax:

wire [7:0] state_regs[8];
wire [7:0] r[8];
wire [7:0] sel[8];
assign sel = condition ? state_regs : r;

I was generating some verilog with Google XLS, and this pattern came up, and I found that it was not supported by Yosys.

Tests are added in tests/svtypes/array_assign.sv.

This collides with PR #2331, but that PR has been up for > 5 years. This implementation also addresses the review feedback from that PR, including support for array ternary expressions
(out = sel ? a : b) and multi-dimensional unpacked arrays.

[whitequark]

This looks okay to me, although I'm far from the expert on the Verilog frontend.

This approach will generate quite a lot of code if the arrays have large dimensions. Any thoughts on addressing that or (more likely) at least surfacing it as something other than OS resource exhaustion?

[apullin]

quite a lot

AFAIK, RTLIL just has to be a ton of entries, wires, ports - there's no bulk memory copy primitive, so combinational array assignment fundamentally requires N parallel paths.

But large code gen is an issue: arrays of ~100K elements take ~3s, and ~1M elements can take 60s+.

I added a warning for arrays over 10,000 elements.

[whitequark]

I'm happy with a warning, and I don't think you can do better here, either.

[apullin]

@widlarizer bump

[widlarizer]

• Integrate this case as a test
• Cover the // Wrap in AST_BLOCK for procedural case with a test

Otherwise looks good

[apullin]

• Added the requested test case. Sure enough, it failed. Resolved by mapping array assignment expansion by element position (range-direction aware), instead of reusing raw numeric indices across LHS/RHS.
• Also added the procedural ternary array-assignment case.

[widlarizer]

Looks good

[apullin]

Yay. Can probably mark #2331 as superseded ?