-
Enhancement
-
Resolution: Fixed
-
P4
-
21
-
b19
-
x86
Currently, abs nodes are implemented using the sequence:
mov tmp, src
sar tmp, 31
mov dst, src
xor dst, tmp
sub dst, tmp
This can be optimized into:
xor dst, dst
sub dst, src
cmovl dst, src
This saves 2 instructions and 1 temp register.
mov tmp, src
sar tmp, 31
mov dst, src
xor dst, tmp
sub dst, tmp
This can be optimized into:
xor dst, dst
sub dst, src
cmovl dst, src
This saves 2 instructions and 1 temp register.
- relates to
-
JDK-8222074 Enhance auto vectorization for x86
-
- Resolved
-