Hello,
ISTR previously reporting an issue with VPCMPESTRI emulation having VEX.W ignored, and then also having got a fixed SDE version. The just released 8.49.0 looks to misbehave again with the respective test in the test harness of Xen's x86 emulator. I haven't checked register details yet, but it seems pretty likely for the earlier issue to have got re-introduced.
Thanks, Jan
Link Copied
I've now confirmed that checking for the insn results as if VEX.W was clear makes the harness pass again. So this very much looks like a regression.
The issue was miss handling of the length argument with respect to absolute saturation of the value when the input does not fit in 32 bits unsigned value. This is now fixed and it will be part of the next release.
For more complete information about compiler optimizations, see our Optimization Notice.