-
Enhancement
-
Resolution: Fixed
-
P4
-
18
-
b26
-
x86
Optimize vector mask reduction operations (firsttrue, lasttrue, tolong, truecount) on x86:
* Use vmovmskpd, vmovmskps instead of full VectorStoreMask to boolean prior to the mask reduction where possible.
* Use lzcnt/tzcnt for lasttrue/firsttrue.
* Use vmovmskpd, vmovmskps instead of full VectorStoreMask to boolean prior to the mask reduction where possible.
* Use lzcnt/tzcnt for lasttrue/firsttrue.