[PATCH] D37949: [X86][XOP] Merge rotation opcodes with AVX512 equivalents. NFCI.

[Simon Pilgrim via Phabricator via llvm-commits]

RKSimon created this revision.

The XOP rotations act as ROTL with +ve values and ROTR with -ve values, which means that we can treat them all as ROTL with unsigned modulo. We already check that we're only trying to lower as ROTL for XOP rotations.


Repository:
  rL LLVM

https://reviews.llvm.org/D37949

Files:
  lib/Target/X86/X86ISelLowering.cpp
  lib/Target/X86/X86ISelLowering.h
  lib/Target/X86/X86InstrFragmentsSIMD.td
  lib/Target/X86/X86IntrinsicsInfo.h


Index: lib/Target/X86/X86IntrinsicsInfo.h
===================================================================
--- lib/Target/X86/X86IntrinsicsInfo.h
+++ lib/Target/X86/X86IntrinsicsInfo.h
@@ -1630,14 +1630,14 @@
   X86_INTRINSIC_DATA(xop_vpermil2ps,     INTR_TYPE_4OP, X86ISD::VPERMIL2, 0),
   X86_INTRINSIC_DATA(xop_vpermil2ps_256, INTR_TYPE_4OP, X86ISD::VPERMIL2, 0),
   X86_INTRINSIC_DATA(xop_vpperm,        INTR_TYPE_3OP, X86ISD::VPPERM, 0),
-  X86_INTRINSIC_DATA(xop_vprotb,        INTR_TYPE_2OP, X86ISD::VPROT, 0),
-  X86_INTRINSIC_DATA(xop_vprotbi,       INTR_TYPE_2OP, X86ISD::VPROTI, 0),
-  X86_INTRINSIC_DATA(xop_vprotd,        INTR_TYPE_2OP, X86ISD::VPROT, 0),
-  X86_INTRINSIC_DATA(xop_vprotdi,       INTR_TYPE_2OP, X86ISD::VPROTI, 0),
-  X86_INTRINSIC_DATA(xop_vprotq,        INTR_TYPE_2OP, X86ISD::VPROT, 0),
-  X86_INTRINSIC_DATA(xop_vprotqi,       INTR_TYPE_2OP, X86ISD::VPROTI, 0),
-  X86_INTRINSIC_DATA(xop_vprotw,        INTR_TYPE_2OP, X86ISD::VPROT, 0),
-  X86_INTRINSIC_DATA(xop_vprotwi,       INTR_TYPE_2OP, X86ISD::VPROTI, 0),
+  X86_INTRINSIC_DATA(xop_vprotb,        INTR_TYPE_2OP, ISD::ROTL, 0),
+  X86_INTRINSIC_DATA(xop_vprotbi,       INTR_TYPE_2OP, X86ISD::VROTLI, 0),
+  X86_INTRINSIC_DATA(xop_vprotd,        INTR_TYPE_2OP, ISD::ROTL, 0),
+  X86_INTRINSIC_DATA(xop_vprotdi,       INTR_TYPE_2OP, X86ISD::VROTLI, 0),
+  X86_INTRINSIC_DATA(xop_vprotq,        INTR_TYPE_2OP, ISD::ROTL, 0),
+  X86_INTRINSIC_DATA(xop_vprotqi,       INTR_TYPE_2OP, X86ISD::VROTLI, 0),
+  X86_INTRINSIC_DATA(xop_vprotw,        INTR_TYPE_2OP, ISD::ROTL, 0),
+  X86_INTRINSIC_DATA(xop_vprotwi,       INTR_TYPE_2OP, X86ISD::VROTLI, 0),
   X86_INTRINSIC_DATA(xop_vpshab,        INTR_TYPE_2OP, X86ISD::VPSHA, 0),
   X86_INTRINSIC_DATA(xop_vpshad,        INTR_TYPE_2OP, X86ISD::VPSHA, 0),
   X86_INTRINSIC_DATA(xop_vpshaq,        INTR_TYPE_2OP, X86ISD::VPSHA, 0),
Index: lib/Target/X86/X86InstrFragmentsSIMD.td
===================================================================
--- lib/Target/X86/X86InstrFragmentsSIMD.td
+++ lib/Target/X86/X86InstrFragmentsSIMD.td
@@ -210,10 +210,10 @@
 def X86vrotli  : SDNode<"X86ISD::VROTLI", SDTIntShiftOp>;
 def X86vrotri  : SDNode<"X86ISD::VROTRI", SDTIntShiftOp>;
 
-def X86vprot   : SDNode<"X86ISD::VPROT",
+def X86vprot   : SDNode<"ISD::ROTL",
                         SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0,1>,
                                              SDTCisSameAs<0,2>]>>;
-def X86vproti  : SDNode<"X86ISD::VPROTI",
+def X86vproti  : SDNode<"X86ISD::VROTLI",
                         SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0,1>,
                                              SDTCisVT<2, i8>]>>;
 
Index: lib/Target/X86/X86ISelLowering.h
===================================================================
--- lib/Target/X86/X86ISelLowering.h
+++ lib/Target/X86/X86ISelLowering.h
@@ -448,8 +448,6 @@
       /// SSE4A Extraction and Insertion.
       EXTRQI, INSERTQI,
 
-      // XOP variable/immediate rotations.
-      VPROT, VPROTI,
       // XOP arithmetic/logical shifts.
       VPSHA, VPSHL,
       // XOP signed/unsigned integer comparisons.
Index: lib/Target/X86/X86ISelLowering.cpp
===================================================================
--- lib/Target/X86/X86ISelLowering.cpp
+++ lib/Target/X86/X86ISelLowering.cpp
@@ -22715,7 +22715,7 @@
   assert((Opcode == ISD::ROTL) && "Only ROTL supported");
 
   // XOP has 128-bit vector variable + immediate rotates.
-  // +ve/-ve Amt = rotate left/right.
+  // +ve/-ve Amt = rotate left/right - just need to handle ISD::ROTL.
 
   // Split 256-bit integers.
   if (VT.is256BitVector())
@@ -22728,13 +22728,13 @@
     if (auto *RotateConst = BVAmt->getConstantSplatNode()) {
       uint64_t RotateAmt = RotateConst->getAPIntValue().getZExtValue();
       assert(RotateAmt < EltSizeInBits && "Rotation out of range");
-      return DAG.getNode(X86ISD::VPROTI, DL, VT, R,
+      return DAG.getNode(X86ISD::VROTLI, DL, VT, R,
                          DAG.getConstant(RotateAmt, DL, MVT::i8));
     }
   }
 
   // Use general rotate by variable (per-element).
-  return DAG.getNode(X86ISD::VPROT, DL, VT, R, Amt);
+  return Op;
 }
 
 static SDValue LowerXALUO(SDValue Op, SelectionDAG &DAG) {
@@ -24659,8 +24659,6 @@
   case X86ISD::RDSEED:             return "X86ISD::RDSEED";
   case X86ISD::VPMADDUBSW:         return "X86ISD::VPMADDUBSW";
   case X86ISD::VPMADDWD:           return "X86ISD::VPMADDWD";
-  case X86ISD::VPROT:              return "X86ISD::VPROT";
-  case X86ISD::VPROTI:             return "X86ISD::VPROTI";
   case X86ISD::VPSHA:              return "X86ISD::VPSHA";
   case X86ISD::VPSHL:              return "X86ISD::VPSHL";
   case X86ISD::VPCOM:              return "X86ISD::VPCOM";


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D37949.115544.patch
Type: text/x-patch
Size: 4811 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20170916/54e4c28c/attachment.bin>