intel

Trace files are big and compress really well, 5x-ish in my experience. Could magic-trace write out compressed traces instead of raw traces?

Perfetto already has support for loading gzipped traces. But gzip is really slow (fingers crossed for Sapphire Rapids, but that's years from being commonplace).

As far as I can tell, we have two options here:

• teach magic-trace to write gzipped trace

varying float<3> col = { 0, 0, 0 };
uniform float <3> reduce_col;
reduce_col.x = reduce_add( col.x );
reduce_col.y = reduce_add( col.y );
reduce_col.z = reduce_add( col.z );
reduce_col.w = reduce_add( col.w );

This currently does not produce an error. I expect attempting to access the w component of a three component type to be an error.

Describe the bug
A clear and concise description of what the bug is.

To Reproduce
Steps to reproduce the behavior:

1. Download install.sh
2. Make it executable
3. Run install.sh

Expected behavior
Honoring XDG Base Directory specifications and not installing binaries and all of the data files into $HOME/.config which is default $XDG_CONFIG_HOME and shouldn't be hardco

If you speak another language, I would appreciate your help in translating the README.md.

For tables, checklists, or other data that might change, please indicate that that information is in the main README. Otherwise every change to the main README will need to be replicated to the other READMEs.

^ I've tried to do this a bit in the French README. The only thing you need to replicate when

There are couple more cases to filter out from isEntryPoint() check: __builtin* and __esimd*.

Originally posted by @kbobrovs in intel/llvm#5858 (comment)