[IMPALA-4077] Populate more intrinsics in sse-util.h

• Type: Improvement
• Status: Open
• Priority: Minor
• Resolution: Unresolved
• Affects Version/s: Impala 2.7.0
• Component/s: Backend
• Labels: newbie
• Assignee: Unassigned
• Reporter: Jim Apple
• Created: 06/Sep/16 15:40
• Updated: 25/Jun/19 21:55

Description

We can write SIMD functions using Intel's intrinsics (in <immintrin.h>) now, but we have to use

__attribute__((__target__("avx2")));

at the end of the function declarations, and these functions can't be inlined, which can hurt performance quite a bit.

We should populate sse-util.h with the intrinsics we are using. A list can be found via

grep -rIohE '_mm[^( ]*' be/src/ --include '*.h' --include '*.cc' | sort | uniq
 -c | sort -n