... rather than being gated by -Z saturating-float-casts.
There are several reasons for this:
1. Const eval already implements this behavior.
2. Unlike with float->int casts, this behavior is uncontroversially the
right behavior and it is not as performance critical. Thus there is no
particular need to make the bug fix for u128->f32 casts opt-in.
3. Having two orthogonal features under one flag is silly, and never
should have happened in the first place.
4. Benchmarking float->int casts with the -Z flag should not pick up
performance changes due to the u128->f32 casts (assuming there are any).
Fixes#41799
This affects regular code generation as well as constant evaluation in trans,
but not the HIR constant evaluator because that one returns an error for
overflowing casts and NaN-to-int casts. That error is conservatively
correct and we should be careful to not accept more code in constant
expressions.
The changes to code generation are guarded by a new -Z flag, to be able
to evaluate the performance impact. The trans constant evaluation changes
are unconditional because they have no run time impact and don't affect
type checking either.
This commit refactors the `collect_crate_translation_items` function to only
require the `TyCtxt` instead of a `SharedCrateContext` in preparation for
query-ifying this portion of trans.
This commit moves the calculation of the `LanguageItems` structure into a
query rather than being calculated before the `TyCtxt` exists, with the eventual
end goal of removing some `CrateStore` methods.
- fixes evaluation of array length for zero-sized type referenced by
rvalue operand.
- adds test to verify fix.
Cause of the issue.
Zero-sized aggregates are handled as operands, not lvalues. Therefore while
visiting Assign statement by LocalAnalyser, mark_as_lvalue() is not called for
related Local. This behaviour is controlled by rvalue_creates_operand() method.
As result it causes error later, when rvalue operand is evaluated in
trans_rvalue_operand() while handling Rvalue::Len case. Array length evaluation
invokes trans_lvalue() which expects referenced Local to be value, not operand.
How it is fixed.
In certain cases result of Rvalue::Len can be evaluated without calling
trans_lvalue(). Method evaluate_array_len() is introduced to handle length
evaluation for zero-sized types referenced by Locals.
This elides initialization for zero-sized arrays:
* for zero-sized elements we previously emitted an empty loop
* for arrays with a length of zero we previously emitted a loop with zero
iterations
This emits llvm.memset() instead of a loop over each element when:
* all elements are zero integers
* elements are byte sized
Translate array drop glue using MIR
I was a bit lazy here and used a usize-based index instead of a pointer iteration. Do you think this is important @eddyb?
r? @eddyb
Operand: 72 -> 24 B
Statement: 192 -> 96 B
Terminator: 256 -> 112 B
librustc translation memory usage: 1795 -> 1669 MB
next step would be interning lvalues, I suppose?
The main changes around rustc::ty::Layout::struct and rustc_trans:adt:
* Added primitive_align field which stores alignment before repr align
* Always emit field padding when generating the LLVM struct fields
* Added methods for adjusting field indexes from the layout index to the
LLVM struct field index
The main user of this information is rustc_trans::adt::struct_llfields
which determines the LLVM fields to be used by LLVM, including padding
fields.
Drop of arrays is now translated in trans::block in an ugly way that I
should clean up in a later PR, and does not handle panics in the middle
of an array drop, but this commit & PR are growing too big.
These changes are in the same commit to avoid needing to adapt
meth::trans_object_shim to the new scheme.
One codegen-units test is broken because we instantiate the shims even
when they are not needed. This will be fixed in the next PR.
