initial @expect() implementation #3364
Conversation
shawnl
force-pushed
the
expect
branch
3 times, most recently
from
87ae657
to
327599c
Compare
|
@andrewrk Frendly ping. :) |
| Provide optimization hint to the compiler. Can be used on {#link|integers|Integers#}, {#syntax#}bools{#endsyntax#}, and {#link|enums|enum#}. | ||
| </p> | ||
| <p> | ||
| Tell the compiler that {#syntax#}value{#endsyntax#} is very likely to be {#syntax#}expected{#endsyntax#}. The compiler may ignore this. This may override the branch predictor rather than just provide its default state, and can result in reduced performance (PowerPC disabled branch prediction override by default due to misuse of such hints). |
There was a problem hiding this comment.
(PowerPC disabled branch prediction override by default due to misuse of such hints).
It's a good warning, but please cite source or remove assertion.
| if (const_val->special == ConstValSpecialUndef) { | ||
| return (uint32_t)0xc0ffee00; | ||
| } |
There was a problem hiding this comment.
This looks like it might be a problem. Why was this necessary?
| uint32_t int_size; | ||
| ZigType *type = instruction->base.value.type; | ||
|
|
||
| switch (type->id) { |
There was a problem hiding this comment.
This function should work on all types. If LLVM does not support lowering it, that's fine, just ignore it and return the value undisturbed.
| case ZigTypeIdPointer: | ||
| case ZigTypeIdOptional: | ||
| case ZigTypeIdErrorUnion: | ||
| zig_panic("TODO. more types for @expect()"); |
There was a problem hiding this comment.
@expect should work on all types. Instead of panicking here, just return value.
| if (type_is_invalid(expect_instr->value.type)) | ||
| return ira->codegen->invalid_instruction; | ||
|
|
||
| switch (value_type->id) { |
There was a problem hiding this comment.
I don't think this switch is needed at all. Implicit cast the expected value to the correct type, resolve it as comptime, and then either return value (in the case value is comptime known) or return an expect instruction.
|
Implementation is not quite right, and it's a low-priority feature (compared to other features in this release cycle). Closing until a more mergeable implementation is available. |
Closes: #489