RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-24 03:33:20 +01:00
Code Issues Projects Releases Wiki Activity

[X86] Add X86ISD::VTRUNC to computeKnownBitsForTargetNode.

Browse Source 
We have to take special care to avoid the cases where the result of the truncate would be padded with zero elements.

Ideally we'd just use ISD::TRUNCATE for these cases instead.

llvm-svn: 322454
This commit is contained in:
Craig Topper 2018-01-14 08:11:33 +00:00
parent 767d0f3bfe
commit f1f9a8c7f7
2 changed files with 14 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
lib/Target/X86/X86ISelLowering.cpp
View File

@ -27837,6 +27837,18 @@ void X86TargetLowering::computeKnownBitsForTargetNode(const SDValue Op,
Known.Zero.setBitsFrom(InBitWidth);
break;
}
case X86ISD::VTRUNC: {
// TODO: Add DemandedElts support.
SDValue N0 = Op.getOperand(0);
// We can only handle cases with the same number of elements. Otherwise
// the truncate fills with zero elements.
// TODO: Maybe we could just discard any 1s we found instead of skipping?
if (VT.getVectorNumElements() != N0.getValueType().getVectorNumElements())
break;
DAG.computeKnownBits(N0, Known, Depth+1);
Known = Known.trunc(BitWidth);
break;
}
case X86ISD::CMOV: {
DAG.computeKnownBits(Op.getOperand(1), Known, Depth+1);
// If we don't know any bits, early out.

4
test/CodeGen/X86/avx512-trunc.ll
View File

@ -748,7 +748,7 @@ define <16 x i8> @usat_trunc_db_256(<8 x i32> %x) {
; KNL-NEXT: vpbroadcastd {{.*#+}} ymm1 = [255,255,255,255,255,255,255,255]
; KNL-NEXT: vpminud %ymm1, %ymm0, %ymm0
; KNL-NEXT: vpmovdw %zmm0, %ymm0
; KNL-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u]
; KNL-NEXT: vpackuswb %xmm0, %xmm0, %xmm0
; KNL-NEXT: vzeroupper
; KNL-NEXT: retq
;
@ -756,7 +756,7 @@ define <16 x i8> @usat_trunc_db_256(<8 x i32> %x) {
; SKX: ## %bb.0:
; SKX-NEXT: vpminud {{.*}}(%rip){1to8}, %ymm0, %ymm0
; SKX-NEXT: vpmovdw %ymm0, %xmm0
; SKX-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u]
; SKX-NEXT: vpackuswb %xmm0, %xmm0, %xmm0
; SKX-NEXT: vzeroupper
; SKX-NEXT: retq
%tmp1 = icmp ult <8 x i32> %x, <i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255, i32 255>