- User Since
- Sep 7 2020, 6:07 AM (6 w, 5 d)
Mon, Oct 19
@RKSimon I ran it through llvm-mca now. It gives me a reciprocal throughput of 3.5 for Silvermont and 3 for Haswell for the new instruction sequence.
@RKSimon Thank you for your very speedy response.
Sat, Oct 10
Can you commit the changes for me?
Tue, Oct 6
Mon, Oct 5
Unfortunately I couldn't get the "update_analyze_test_checks.py" script to work no matter what arguments I tried (which included --force-update, --opt-binary=T:\llvm-project\llvm\NATIVE\Debug\bin\opt.exe, different quotation mark placement and so on). Most of the time it did just nothing without any error or it would sometimes error with "ERROR: Unexpected opt name: opt.exe".
Sun, Oct 4
This patch is now just about v8i16 and I updated the array in "getTypeBasedIntrinsicInstrCost".
Tue, Sep 29
Thank you for your comments.
Sep 21 2020
I published a followup patch D88026 that fixes the horizontal reduction regression in the way suggested by craig.topper.