http://llvm.org/bugs/show_bug.cgi?id=22412
Bug ID: 22412
Summary: New vector shuffle legality: turns blendp+permp to
permp+permilp+blendp.
Product: libraries
Version: trunk
Hardware: PC
OS: All
Status: NEW
Severity: normal
Priority: P
Component: Backend: X86
Assignee: [email protected]
Reporter: [email protected]
CC: [email protected]
Classification: Unclassified
Created attachment 13784
--> http://llvm.org/bugs/attachment.cgi?id=13784&action=edit
IR testcase
For the attached IR, we used to generate (targeting core-avx2):
vblendpd $1, %ymm0, %ymm1, %ymm0 ## ymm0 = ymm0[0],ymm1[1,2,3]
vmovaps LCPI0_0(%rip), %ymm1 ## ymm1 = [1,0,7,6,5,4,3,2]
vpermps %ymm0, %ymm1, %ymm0
With the new all-shuffles-are-legal lowering, we now generate:
vmovaps LCPI0_0(%rip), %ymm2 ## ymm2 = <u,u,7,6,5,4,3,2>
vpermps %ymm1, %ymm2, %ymm1
vpermilps $225, %ymm0, %ymm0 ## ymm0 = ymm0[1,0,2,3,5,4,6,7]
vblendpd $1, %ymm0, %ymm1, %ymm0 ## ymm0 = ymm0[0],ymm1[1,2,3]
--
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
LLVMbugs mailing list
[email protected]
http://lists.cs.uiuc.edu/mailman/listinfo/llvmbugs
