John Mather

POCL 4.0 failing with "Invalid size request on a scalable vector" on neoverse-v1

4

I believe this has to do with SVE which is present on the neoverse-v1. I have tried building POCL with both llvm 14 and 15 on Ubuntu 22.04, but both...

openlibm performance on ARM server is very poor

7

I see very poor results on a modern ARM server. Some openlibm implementations are up to 48.68x slower than their libm counterparts. This was a checkout of the main branch...

Implicit type narrowing causes crashes with large geometries

2

There is a common pattern in OpenSubdiv where integers of smaller types are passed into the resize method of std::vectors. When the incoming meshes are very large, these types overflow...

Performance differences between Windows and Linux

5

I've been running our software through our benchmark suite with both mimalloc 1.8.7 and mimalloc 2.1.7 and have noticed some differences in its behaviour across OSes. On Windows, 2.1.7 pretty...

Fix call to _mm_cvtps_ph in half.h

3

_mm_cvtps_ph doesn't support _MM_FROUND_NO_EXC. On Windows, if this is compiled, you will get the following warning: `warning C4556: value of intrinsic immediate argument '8' is out of range '0 -...