def emit_dahlia_params(fd: DahliaFuncDef) -> str:
"""Emits a comma-separated string of Dahlia
memory declarations, e.g.
`X: ubit<32> [1][10], a: ufix<8, 2>[3]`
"""
cells = []
for cell in fd.args + [fd.dest]:
cell_str = f"{cell.id.name}: {fd.data_type}"
dims = get_dims(cell.comp)
args = cell.comp.args
for i in range(0, dims):
cell_str += f"[{args[i + 1]}]"
cells.append(cell_str)
return ", ".join(cells)
def emit_dahlia_loop(control_flow: Cell, body: str) -> str:
"""Emits a Dahlia loop over `num_dims` with `body`
nested inside. Many tensor functions share the
same control flow:
(1) Perform specific looping according to `control_flow`,
(2) and do some work in the body.
For example, if body == `X`, then this
will return:
```
for (let i: ubit<X> = 0..M) {
for (let j: ubit<Y> = 0..N) {
X;
}
}
```
"""
var_name = CHARACTER_I
# Loop control flow is determined by these parameters.
num_dims = get_dims(control_flow.comp)
args = control_flow.comp.args
# Generate loop headers.
headers = []
for i in range(num_dims):
size = args[i + 1]
idx_size = args[i + 1 + num_dims]
headers.append(f"for (let __{var_name}: ubit<{idx_size}> = 0..{size})")
var_name = next_character(var_name)
headers.reverse()
# Generate loop blocks.
for i in range(num_dims):
b = body if i == 0 else headers[i - 1]
headers[i] = block(headers[i], b, sep="")
return headers[-1]