Legalize: Improve legalization of long vector extends.

When an extend more than doubles the size of the elements (e.g., a zext
from v16i8 to v16i32), the normal legalization method of splitting the
vectors will run into problems as by the time the destination vector is
legal, the source vector is illegal. The end result is the operation
often becoming scalarized, with the typical horrible performance. For
example, on x86_64, the simple input of:
define void @bar(<16 x i8> %a, <16 x i32>* %p) nounwind {
  %tmp = zext <16 x i8> %a to <16 x i32>
  store <16 x i32> %tmp, <16 x i32>*%p
  ret void
}

Generates:
  .section  __TEXT,__text,regular,pure_instructions
  .section  __TEXT,__const
  .align  5
LCPI0_0:
  .long 255                     ## 0xff
  .long 255                     ## 0xff
  .long 255                     ## 0xff
  .long 255                     ## 0xff
  .long 255                     ## 0xff
  .long 255                     ## 0xff
  .long 255                     ## 0xff
  .long 255                     ## 0xff
  .section  __TEXT,__text,regular,pure_instructions
  .globl  _bar
  .align  4, 0x90
_bar:
  vpunpckhbw  %xmm0, %xmm0, %xmm1
  vpunpckhwd  %xmm0, %xmm1, %xmm2
  vpmovzxwd %xmm1, %xmm1
  vinsertf128 $1, %xmm2, %ymm1, %ymm1
  vmovaps LCPI0_0(%rip), %ymm2
  vandps  %ymm2, %ymm1, %ymm1
  vpmovzxbw %xmm0, %xmm3
  vpunpckhwd  %xmm0, %xmm3, %xmm3
  vpmovzxbd %xmm0, %xmm0
  vinsertf128 $1, %xmm3, %ymm0, %ymm0
  vandps  %ymm2, %ymm0, %ymm0
  vmovaps %ymm0, (%rdi)
  vmovaps %ymm1, 32(%rdi)
  vzeroupper
  ret

So instead we can check if there are legal types that enable us to split
more cleverly when the input vector is already legal such that we don't
turn it into an illegal type. If the extend is such that it's more than
doubling the size of the input we check if
  - the number of vector elements is even,
  - the source type is legal,
  - the type of a split source is illegal,
  - the type of an extended (by doubling element size) source is legal, and
  - the type of that extended source when split is legal.
If the conditions are met, instead of just splitting both the
destination and the source types, we create an extend that only goes up
one "step" (doubling the element width), and the continue legalizing the
rest of the operation normally. The result is that this operates as a
new, more effecient, termination condition for the loop of "split the
operation until the destination type is legal."

With this change, the above example now compiles to:
_bar:
  vpxor %xmm1, %xmm1, %xmm1
  vpunpcklbw  %xmm1, %xmm0, %xmm2
  vpunpckhwd  %xmm1, %xmm2, %xmm3
  vpunpcklwd  %xmm1, %xmm2, %xmm2
  vinsertf128 $1, %xmm3, %ymm2, %ymm2
  vpunpckhbw  %xmm1, %xmm0, %xmm0
  vpunpckhwd  %xmm1, %xmm0, %xmm3
  vpunpcklwd  %xmm1, %xmm0, %xmm0
  vinsertf128 $1, %xmm3, %ymm0, %ymm0
  vmovaps %ymm0, 32(%rdi)
  vmovaps %ymm2, (%rdi)
  vzeroupper
  ret

This generalizes a custom lowering that was added a while back to the
ARM backend. That lowering is no longer necessary, and is removed. The
testcases for it, however, provide excellent ARM tests for this change
and so remain.

rdar://14735100

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193727 91177308-0d34-0410-b5e6-96231b3b80d8

1 files changed, 62 insertions, 3 deletions

diff --git a/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp b/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
index ad31f7e94f..a4e2fc472e 100644
--- a/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
+++ b/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
@@ -521,7 +521,6 @@ void DAGTypeLegalizer::SplitVectorResult(SDNode *N, unsigned ResNo) {
     SplitVecRes_VECTOR_SHUFFLE(cast<ShuffleVectorSDNode>(N), Lo, Hi);
     break;
 
-  case ISD::ANY_EXTEND:
   case ISD::CONVERT_RNDSAT:
   case ISD::CTLZ:
   case ISD::CTTZ:
@@ -548,14 +547,18 @@ void DAGTypeLegalizer::SplitVectorResult(SDNode *N, unsigned ResNo) {
   case ISD::FSIN:
   case ISD::FSQRT:
   case ISD::FTRUNC:
-  case ISD::SIGN_EXTEND:
   case ISD::SINT_TO_FP:
   case ISD::TRUNCATE:
   case ISD::UINT_TO_FP:
-  case ISD::ZERO_EXTEND:
     SplitVecRes_UnaryOp(N, Lo, Hi);
     break;
 
+  case ISD::ANY_EXTEND:
+  case ISD::SIGN_EXTEND:
+  case ISD::ZERO_EXTEND:
+    SplitVecRes_ExtendOp(N, Lo, Hi);
+    break;
+
   case ISD::ADD:
   case ISD::SUB:
   case ISD::MUL:
@@ -921,6 +924,62 @@ void DAGTypeLegalizer::SplitVecRes_UnaryOp(SDNode *N, SDValue &Lo,
   }
 }
 
+void DAGTypeLegalizer::SplitVecRes_ExtendOp(SDNode *N, SDValue &Lo,
+                                            SDValue &Hi) {
+  SDLoc dl(N);
+  EVT SrcVT = N->getOperand(0).getValueType();
+  EVT DestVT = N->getValueType(0);
+  EVT LoVT, HiVT;
+  GetSplitDestVTs(DestVT, LoVT, HiVT);
+
+  // We can do better than a generic split operation if the extend is doing
+  // more than just doubling the width of the elements and the following are
+  // true:
+  //   - The number of vector elements is even,
+  //   - the source type is legal,
+  //   - the type of a split source is illegal,
+  //   - the type of an extended (by doubling element size) source is legal, and
+  //   - the type of that extended source when split is legal.
+  //
+  // This won't necessarily completely legalize the operation, but it will
+  // more effectively move in the right direction and prevent falling down
+  // to scalarization in many cases due to the input vector being split too
+  // far.
+  unsigned NumElements = SrcVT.getVectorNumElements();
+  if ((NumElements & 1) == 0 &&
+      SrcVT.getSizeInBits() * 2 < DestVT.getSizeInBits()) {
+    LLVMContext &Ctx = *DAG.getContext();
+    EVT NewSrcVT = EVT::getVectorVT(
+        Ctx, EVT::getIntegerVT(
+                 Ctx, SrcVT.getVectorElementType().getSizeInBits() * 2),
+        NumElements);
+    EVT SplitSrcVT =
+        EVT::getVectorVT(Ctx, SrcVT.getVectorElementType(), NumElements / 2);
+    EVT SplitLoVT, SplitHiVT;
+    GetSplitDestVTs(NewSrcVT, SplitLoVT, SplitHiVT);
+    if (TLI.isTypeLegal(SrcVT) && !TLI.isTypeLegal(SplitSrcVT) &&
+        TLI.isTypeLegal(NewSrcVT) && TLI.isTypeLegal(SplitLoVT)) {
+      DEBUG(dbgs() << "Split vector extend via incremental extend:";
+            N->dump(&DAG); dbgs() << "\n");
+      // Extend the source vector by one step.
+      SDValue NewSrc =
+          DAG.getNode(N->getOpcode(), dl, NewSrcVT, N->getOperand(0));
+      // Get the low and high halves of the new, extended one step, vector.
+      Lo = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, SplitLoVT, NewSrc,
+                       DAG.getConstant(0, TLI.getVectorIdxTy()));
+      Hi = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, SplitHiVT, NewSrc,
+                       DAG.getConstant(SplitLoVT.getVectorNumElements(),
+                                       TLI.getVectorIdxTy()));
+      // Extend those vector halves the rest of the way.
+      Lo = DAG.getNode(N->getOpcode(), dl, LoVT, Lo);
+      Hi = DAG.getNode(N->getOpcode(), dl, HiVT, Hi);
+      return;
+    }
+  }
+  // Fall back to the generic unary operator splitting otherwise.
+  SplitVecRes_UnaryOp(N, Lo, Hi);
+}
+
 void DAGTypeLegalizer::SplitVecRes_VECTOR_SHUFFLE(ShuffleVectorSDNode *N,
                                                   SDValue &Lo, SDValue &Hi) {
   // The low and high parts of the original input give four input vectors.