| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
If indexing into a type with power-of-2-sized elements we can access them
in one instruction:
x/reg1: (addr int) <- index A/reg2: (addr array int), idx/reg3: int
This translates to a single instruction because x86 instructions support
an addressing mode with left-shifts.
For non-powers-of-2, however, we need a multiply. To keep things type-safe,
it is performed like this:
x/reg1: (offset T) <- compute-offset A: (addr array T), idx: int
y/reg2: (addr T) <- index A, x
An offset is just an int that is guaranteed to be a multiple of size-of(T).
Offsets can only be used in index instructions, and the types will eventually
be required to line up.
In the process, I have to expand Input-size because mu.subx is growing
big.
|
| |
|
|
| |
Fix CI.
|
| |
|
|
|
| |
Fix CI. apps/survey was running out of space in the trace segment when
translating apps/mu.subx
|
| |
|
|
| |
Expand some buffer sizes to continue building mu.subx natively.
|
| |
|
|
|
|
|
|
|
| |
A couple more primitives now working. In the process I ran into an issue
with some buffer filling up when running ntranslate. Isolating it to survey.subx
was straightforward, but --trace ran out of RAM, and --trace --dump ran
out of (7GB of) disk. In the end what helped was just repeatedly inserting
exits at different points, and I realized there was a magic number that
hadn't been turned into a named constant.
|
|
|
This undoes 5672 in favor of a new plan:
Layers 000 - 099 are for running without syntax sugar. We use them for
building syntax-sugar passes.
Layers 100 and up are for running with all syntax sugar.
The layers are arranged in approximate order so more phases rely on earlier
layers than later ones.
I plan to not use intermediate syntax sugar (just sigils without calls,
or sigils and calls without braces) anywhere except in the specific passes
implementing them.
|