summaryrefslogtreecommitdiffstats
path: root/llvm/lib/Target/X86
diff options
context:
space:
mode:
Diffstat (limited to 'llvm/lib/Target/X86')
-rw-r--r--llvm/lib/Target/X86/X86ISelLowering.cpp9
-rw-r--r--llvm/lib/Target/X86/X86InstrAVX512.td17
2 files changed, 23 insertions, 3 deletions
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 74299e92d4a..3a013a24111 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -13102,10 +13102,13 @@ SDValue X86TargetLowering::LowerTRUNCATE(SDValue Op, SelectionDAG &DAG) const {
}
// vpmovqb/w/d, vpmovdb/w, vpmovwb
- if (((!InVT.is512BitVector() && Subtarget->hasVLX()) || InVT.is512BitVector()) &&
- (InVT.getVectorElementType() != MVT::i16 || Subtarget->hasBWI()))
+ if (Subtarget->hasAVX512()) {
+ // word to byte only under BWI
+ if (InVT == MVT::v16i16 && !Subtarget->hasBWI()) // v16i16 -> v16i8
+ return DAG.getNode(X86ISD::VTRUNC, DL, VT,
+ DAG.getNode(X86ISD::VSEXT, DL, MVT::v16i32, In));
return DAG.getNode(X86ISD::VTRUNC, DL, VT, In);
-
+ }
if ((VT == MVT::v4i32) && (InVT == MVT::v4i64)) {
// On AVX2, v4i64 -> v4i32 becomes VPERMD.
if (Subtarget->hasInt256()) {
diff --git a/llvm/lib/Target/X86/X86InstrAVX512.td b/llvm/lib/Target/X86/X86InstrAVX512.td
index ca124ed253c..60ba687da62 100644
--- a/llvm/lib/Target/X86/X86InstrAVX512.td
+++ b/llvm/lib/Target/X86/X86InstrAVX512.td
@@ -6036,6 +6036,23 @@ defm VPMOVWB : avx512_trunc_wb<0x30, "vpmovwb", X86vtrunc>;
defm VPMOVSWB : avx512_trunc_sat_wb<0x20, "s", X86vtruncs>;
defm VPMOVUSWB : avx512_trunc_sat_wb<0x10, "us", X86vtruncus>;
+let Predicates = [HasAVX512, NoVLX] in {
+def: Pat<(v8i16 (X86vtrunc (v8i32 VR256X:$src))),
+ (v8i16 (EXTRACT_SUBREG
+ (v16i16 (VPMOVDWZrr (v16i32 (SUBREG_TO_REG (i32 0),
+ VR256X:$src, sub_ymm)))), sub_xmm))>;
+def: Pat<(v4i32 (X86vtrunc (v4i64 VR256X:$src))),
+ (v4i32 (EXTRACT_SUBREG
+ (v8i32 (VPMOVQDZrr (v8i64 (SUBREG_TO_REG (i32 0),
+ VR256X:$src, sub_ymm)))), sub_xmm))>;
+}
+
+let Predicates = [HasBWI, NoVLX] in {
+def: Pat<(v16i8 (X86vtrunc (v16i16 VR256X:$src))),
+ (v16i8 (EXTRACT_SUBREG (VPMOVWBZrr (v32i16 (SUBREG_TO_REG (i32 0),
+ VR256X:$src, sub_ymm))), sub_xmm))>;
+}
+
multiclass avx512_extend_common<bits<8> opc, string OpcodeStr,
X86VectorVTInfo DestInfo, X86VectorVTInfo SrcInfo,
X86MemOperand x86memop, PatFrag LdFrag, SDNode OpNode>{
OpenPOWER on IntegriCloud