https://bugs.llvm.org/show_bug.cgi?id=32160
Bug ID: 32160
Summary: [x86, SSE] poor codegen splatting from illegal
truncated vector type
Product: libraries
Version: trunk
Hardware: PC
OS: All
Status: NEW
Severity: enhancement
Priority: P
Component: Backend: X86
Assignee: unassignedb...@nondot.org
Reporter: spatel+l...@rotateright.com
CC: llvm-bugs@lists.llvm.org
This is fallout noted by Eli from the proposed InstCombine in:
https://reviews.llvm.org/D30123
define <8 x i16> @narrow_splat1(<8 x i32> %x) {
%shuf = trunc <8 x i32> %x to <8 x i16>
%trunc = shufflevector <8 x i16> %shuf, <8 x i16> undef, <8 x i32> <i32 2,
i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2>
ret <8 x i16> %trunc
}
$ ./llc -o - widesplat.ll
pslld $16, %xmm1
psrad $16, %xmm1
pslld $16, %xmm0
psrad $16, %xmm0
packssdw %xmm1, %xmm0
pshuflw $170, %xmm0, %xmm0 ## xmm0 = xmm0[2,2,2,2,4,5,6,7]
pshufd $80, %xmm0, %xmm0 ## xmm0 = xmm0[0,0,1,1]
retq
Something (demandedbits?) should tell us that %xmm1 isn't necessary here.
This could have been (if the shuffle was before the trunc):
pshufd $170, %xmm0, %xmm0 ## xmm0 = xmm0[2,2,2,2]
pslld $16, %xmm0
psrad $16, %xmm0
packssdw %xmm0, %xmm0
retq
Or some combination of pslldq + pshuflw + pshufhw?
The DAG looks like this before any x86 shuffling kicks in:
t2: v4i32,ch = CopyFromReg t0, Register:v4i32 %vreg0
t4: v4i32,ch = CopyFromReg t0, Register:v4i32 %vreg1
t5: v8i32 = concat_vectors t2, t4
t6: v8i16 = truncate t5
t8: v8i16 = vector_shuffle<2,2,2,2,2,2,2,2> t6, undef:v8i16
--
You are receiving this mail because:
You are on the CC list for the bug._______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
