Kristof Roomp comments

Results 13 comments of


                                            Kristof Roomp

move_mask for integer types

This is implemented now, can close the issue

Tests failed on fedora koji on s390x arch

This is fixed now and the build verification now includes MIPS bigendian

improve JPEG encode_block perf by about 25% with faster integer encoding

Looks like NonZeroU32::unsigned_abs was stabilized in 1.64... let's see if there's a workaround

improve JPEG encode_block perf by about 25% with faster integer encoding

Rebased since this change had sat there for a while

improve JPEG encode_block perf by about 25% with faster integer encoding

There's still a CI test_toolchain with 1.63 that needs to be upgraded before this can be integrated

NEON instructions on `aarch64`?

aarch64 neon support is now in version 0.7.7

Adds boolean methods to u32x4 and u32x8

Do you think it would be cleaner to use the i32 versions of these since the implementation is identical, avoid code duplication ie: ``` #[inline] #[must_use] pub fn move_mask(self) ->...

Use wrapping shifts in fallback implementations

These are a bit tricky since the SIMD shifts don't behave like wrapping shifts. For example, SIMD shift of a 1u32

`wide::u8x32` is only slower than `wide::u8x16` on i9-13900K/i9-14900K

Isn't that just u8x16::max?

Feature request: shuffle methods

What stable intrinsics would this translate to? Sadly most of the < 32 bit shuffle intrinsics are kind of special case