A few notes on the implementation:
- Any unsigned power of two integer type less than 64 bits in size is supported
as a Limb type.
- The algorithms used are kept simple for the moment. More complicated
algorithms are generally only more useful as integer sizes increase a
lot and I don't expect our current usage to be used for this purpose
just yet.
- All branches (practically) have been covered by tests.
See 986a2b3243/bench
for rough performance comparison numbers.
Closes#364.
See #770
To help automatically translate code, see the
zig-fmt-pointer-reform-2 branch.
This will convert all & into *. Due to the syntax
ambiguity (which is why we are making this change),
even address-of & will turn into *, so you'll have
to manually fix thes instances. You will be guaranteed
to get compile errors for them - expected 'type', found 'foo'
start using zig-fmt-pointer-reform branch build of zig fmt
to fix code to use the new syntax
all of test/cases/* are processed, but there are more left
to be done - all the std lib used by the behavior tests
The purpose of this is:
* Only one way to do things
* Changing a function with void return type to return a possible
error becomes a 1 character change, subtly encouraging
people to use errors.
See #632
Here are some imperfect sed commands for performing this update:
remove arrow:
```
sed -i 's/\(\bfn\b.*\)-> /\1/g' $(find . -name "*.zig")
```
add void:
```
sed -i 's/\(\bfn\b.*\))\s*{/\1) void {/g' $(find ../ -name "*.zig")
```
Some cleanup may be necessary, but this should do the bulk of the work.
Before:
* IR basic blocks are in arbitrary order
* when doing an IR pass, when a block is encountered, code
must look at all the instructions in the old basic block,
determine what blocks are referenced, and queue up those
old basic blocks first.
* This had a bug (See #667)
Now:
* IR basic blocks are required to be in an order that guarantees
they will be referenced by a branch, before any instructions
within are referenced.
ir pass1 is updated to meet this constraint.
* When doing an IR pass, we iterate over old basic blocks
in the order they appear. Blocks which have not been
referenced are discarded.
* After the pass is complete, we must iterate again to look
for old basic blocks which now point to incomplete new
basic blocks, due to comptime code generation.
* This last part can probably be optimized - most of the time
we don't need to iterate over the basic block again.
closes#667
previously we used the bigfloat abstraction to do all
compile-time float math. but runtime code and comptime code
are supposed to get the same result. so now if you add a
f32 to a f32 at compile time it does it with f32 math
instead of the bigfloat. float literals still get the
bigfloat math.
closes#424
* add u3, u4, u5, u6, u7 and i3, i4, i5, i6, i7
* shift operations shift amount parameter type is
integer with log2 bit width of other param
- This enforces not violating undefined behavior on
shift amount >= bit width with the type system
* clean up math.log, math.ln, math.log2, math.log10
closes#403
Before:
* << is left shift, not allowed to shift 1 bits out
* <<% is left shift, allowed to shift 1 bits out
* >> is right shift, allowed to shift 1 bits out
After:
* << is left shift, allowed to shift 1 bits out
* >> is right shift, allowed to shift 1 bits out
* @shlExact is left shift, not allowed to shift 1 bits out
* @shrExact is right shift, not allowed to shift 1 bits out
Closes#413
