[X86][AVX512] Add support for 512-bit shuffle decoding of VPERMPD/VPERMQ
llvm-svn: 274468
diff --git a/llvm/lib/Target/X86/InstPrinter/X86InstComments.cpp b/llvm/lib/Target/X86/InstPrinter/X86InstComments.cpp
index 61a9330..98c3a03 100644
--- a/llvm/lib/Target/X86/InstPrinter/X86InstComments.cpp
+++ b/llvm/lib/Target/X86/InstPrinter/X86InstComments.cpp
@@ -127,6 +127,11 @@
CASE_MASKZ_INS_COMMON(Inst, Z256, src##i) \
CASE_MASKZ_INS_COMMON(Inst, Z128, src##i)
+#define CASE_VPERM(Inst, src) \
+ CASE_AVX512_INS_COMMON(Inst, Z, src##i) \
+ CASE_AVX512_INS_COMMON(Inst, Z256, src##i) \
+ CASE_AVX_INS_COMMON(Inst, Y, src##i)
+
#define CASE_VSHUF(Inst, src) \
CASE_AVX512_INS_COMMON(SHUFF##Inst, Z, r##src##i) \
CASE_AVX512_INS_COMMON(SHUFI##Inst, Z, r##src##i) \
@@ -826,26 +831,24 @@
DestName = getRegName(MI->getOperand(0).getReg());
break;
- case X86::VPERMQYri:
- case X86::VPERMQZ256ri:
- case X86::VPERMQZ256rik:
- case X86::VPERMQZ256rikz:
- case X86::VPERMPDYri:
- case X86::VPERMPDZ256ri:
- case X86::VPERMPDZ256rik:
- case X86::VPERMPDZ256rikz:
+ CASE_VPERM(PERMPD, r)
Src1Name = getRegName(MI->getOperand(NumOperands - 2).getReg());
// FALL THROUGH.
- case X86::VPERMQYmi:
- case X86::VPERMQZ256mi:
- case X86::VPERMQZ256mik:
- case X86::VPERMQZ256mikz:
- case X86::VPERMPDYmi:
- case X86::VPERMPDZ256mi:
- case X86::VPERMPDZ256mik:
- case X86::VPERMPDZ256mikz:
+ CASE_VPERM(PERMPD, m)
if (MI->getOperand(NumOperands - 1).isImm())
- DecodeVPERMMask(MI->getOperand(NumOperands - 1).getImm(),
+ DecodeVPERMMask(getRegOperandVectorVT(MI, MVT::f64, 0),
+ MI->getOperand(NumOperands - 1).getImm(),
+ ShuffleMask);
+ DestName = getRegName(MI->getOperand(0).getReg());
+ break;
+
+ CASE_VPERM(PERMQ, r)
+ Src1Name = getRegName(MI->getOperand(NumOperands - 2).getReg());
+ // FALL THROUGH.
+ CASE_VPERM(PERMQ, m)
+ if (MI->getOperand(NumOperands - 1).isImm())
+ DecodeVPERMMask(getRegOperandVectorVT(MI, MVT::i64, 0),
+ MI->getOperand(NumOperands - 1).getImm(),
ShuffleMask);
DestName = getRegName(MI->getOperand(0).getReg());
break;
diff --git a/llvm/lib/Target/X86/Utils/X86ShuffleDecode.cpp b/llvm/lib/Target/X86/Utils/X86ShuffleDecode.cpp
index 899080a..713595d 100644
--- a/llvm/lib/Target/X86/Utils/X86ShuffleDecode.cpp
+++ b/llvm/lib/Target/X86/Utils/X86ShuffleDecode.cpp
@@ -382,12 +382,14 @@
}
}
- /// DecodeVPERMMask - this decodes the shuffle masks for VPERMQ/VPERMPD.
-/// No VT provided since it only works on 256-bit, 4 element vectors.
-void DecodeVPERMMask(unsigned Imm, SmallVectorImpl<int> &ShuffleMask) {
- for (unsigned i = 0; i != 4; ++i) {
- ShuffleMask.push_back((Imm >> (2 * i)) & 3);
- }
+/// DecodeVPERMMask - this decodes the shuffle masks for VPERMQ/VPERMPD.
+void DecodeVPERMMask(MVT VT, unsigned Imm, SmallVectorImpl<int> &ShuffleMask) {
+ assert((VT.is256BitVector() || VT.is512BitVector()) &&
+ (VT.getScalarSizeInBits() == 64) && "Unexpected vector value type");
+ unsigned NumElts = VT.getVectorNumElements();
+ for (unsigned l = 0; l != NumElts; l += 4)
+ for (unsigned i = 0; i != 4; ++i)
+ ShuffleMask.push_back(l + ((Imm >> (2 * i)) & 3));
}
void DecodeZeroExtendMask(MVT SrcScalarVT, MVT DstVT, SmallVectorImpl<int> &Mask) {
diff --git a/llvm/lib/Target/X86/Utils/X86ShuffleDecode.h b/llvm/lib/Target/X86/Utils/X86ShuffleDecode.h
index f1caa56..7caccef 100644
--- a/llvm/lib/Target/X86/Utils/X86ShuffleDecode.h
+++ b/llvm/lib/Target/X86/Utils/X86ShuffleDecode.h
@@ -105,8 +105,7 @@
SmallVectorImpl<int> &ShuffleMask);
/// Decodes the shuffle masks for VPERMQ/VPERMPD.
-/// No VT provided since it only works on 256-bit, 4 element vectors.
-void DecodeVPERMMask(unsigned Imm, SmallVectorImpl<int> &ShuffleMask);
+void DecodeVPERMMask(MVT VT, unsigned Imm, SmallVectorImpl<int> &ShuffleMask);
/// Decode a VPPERM mask from a raw array of constants such as from
/// BUILD_VECTOR.
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 9c02303..34f6f23 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -4940,7 +4940,7 @@
}
case X86ISD::VPERMI:
ImmN = N->getOperand(N->getNumOperands()-1);
- DecodeVPERMMask(cast<ConstantSDNode>(ImmN)->getZExtValue(), Mask);
+ DecodeVPERMMask(VT, cast<ConstantSDNode>(ImmN)->getZExtValue(), Mask);
IsUnary = true;
break;
case X86ISD::MOVSS: