When the input is a constant, we use a fairly inefficient sequence of
fmov+fcvt+dup or, when the double isn't encodable in fmov,
adr+ldr+fcvt+dup.
Instead, we can use the same lowering as X64 when the input is a
constant, and load the vector from memory. However, if the constant is
encodable via fmov, we can use a vector fmov instead (which is just one
instruction and doesn't need constant space).
Fortunately the bit encoding of fmov for 32-bit floating point numbers
matches that of 64-bit: the decoding algorithm is a little different
because it expands into a larger exponent, but the values are
compatible, so if a double can be encoded into a scalar fmov with a
given abcdefgh pattern, the same pattern should encode the same float;
due to the very limited number of mantissa and exponent bits, all values
that are encodable are also exact in both 32-bit and 64-bit floats.
This strategy is ~same as what gcc uses. For complex vectors, we
previously used 4 instructions and 8 bytes of constant storage, and now
we use 2 instructions and 16 bytes of constant storage, so the memory
footprint is the same; for simple vectors we just need 1 instruction (4
bytes).
clang lowers vector constants a little differently, opting to synthesize
a 64-bit integer using 4 instructions (mov/movk) and then move it to the
vector register - this requires 5 instructions and 20 bytes, vs ours/gcc
2 instructions and 8+16=24 bytes. I tried a simpler version of this that
would be more compact - synthesize a 32-bit integer constant with
mov+movk, and move it to vector register via dup.4s - but this was a
little slower on M2, so for now we prefer the slightly larger version as
it's not a regression vs current implementation.
On the vector approximation benchmark we get:
- Before this PR (flag=false): ~7.85 ns/op
- After this PR (flag=true): ~7.74 ns/op
- After this PR, with 0.125 instead of 0.123 in the benchmark code (to
use fmov): ~7.52 ns/op
- Not part of this PR, but the mov/dup strategy described above: ~8.00
ns/op
- Updated Roblox copyright to 2023
- Floor division operator `//` (implements #832)
- Autocomplete now shows `end` within `do` blocks
- Restore BraceType when using `Lexer::lookahead` (fixes#1019)
# New typechecker
- Subtyping tests between metatables and tables
- Subtyping tests between string singletons and tables
- Subtyping tests for class types
# Native codegen
- Fixed macOS test failure (wrong spill restore offset)
- Fixed clobbering of non-volatile xmm registers on Windows
- Fixed wrong storage location of SSA reg spills
- Implemented A64 support for add/sub extended
- Eliminated zextReg from A64 lowering
- Remove identical table slot lookups
- Propagate values from predecessor into the linear block
- Disabled reuse slot optimization
- Keep `LuaNode::val` check for nil when optimizing `CHECK_SLOT_MATCH`
- Implemented IR translation of `table.insert` builtin
- Fixed mmap error handling on macOS/Linux
# Tooling
- Used `|` as a column separator instead of `+` in `bench.py`
- Added a `table.sort` micro-benchmark
- Switched `libprotobuf-mutator` to a less problematic version
