[PowerPC] Recognize consecutive memory accesses from intrinsics

When generating unaligned vector loads, we need to search for other loads or
stores nearby offset by one vector width. If we find one, then we know that we
can safely generate another aligned load at that address. Otherwise, we must
generate the next load using an offset of the vector width minus one byte (so
we don't read off the end of the allocation if the base unaligned address
happened to be aligned at runtime). We had previously done this using only
other vector loads and stores, but did not consider the PowerPC-specific vector
load/store intrinsics. Now we'll also consider vector intrinsics. By itself,
this change is a feature enhancement, but is a necessary step toward fixing the
underlying problem behind PR19991.

llvm-svn: 214469

2 files changed, 111 insertions, 9 deletions

diff --git a/llvm/lib/Target/PowerPC/PPCISelLowering.cpp b/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
index dc6197d5d95..d25c9e68471 100644
--- a/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
+++ b/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
@@ -7549,16 +7549,12 @@ SDValue PPCTargetLowering::DAGCombineFastRecipFSQRT(SDValue Op,
   return SDValue();
 }
 
-// Like SelectionDAG::isConsecutiveLoad, but also works for stores, and does
-// not enforce equality of the chain operands.
-static bool isConsecutiveLS(LSBaseSDNode *LS, LSBaseSDNode *Base,
+static bool isConsecutiveLSLoc(SDValue Loc, EVT VT, LSBaseSDNode *Base,
                             unsigned Bytes, int Dist,
                             SelectionDAG &DAG) {
-  EVT VT = LS->getMemoryVT();
   if (VT.getSizeInBits() / 8 != Bytes)
     return false;
 
-  SDValue Loc = LS->getBasePtr();
   SDValue BaseLoc = Base->getBasePtr();
   if (Loc.getOpcode() == ISD::FrameIndex) {
     if (BaseLoc.getOpcode() != ISD::FrameIndex)
@@ -7589,6 +7585,64 @@ static bool isConsecutiveLS(LSBaseSDNode *LS, LSBaseSDNode *Base,
   return false;
 }
 
+// Like SelectionDAG::isConsecutiveLoad, but also works for stores, and does
+// not enforce equality of the chain operands.
+static bool isConsecutiveLS(SDNode *N, LSBaseSDNode *Base,
+                            unsigned Bytes, int Dist,
+                            SelectionDAG &DAG) {
+  if (LSBaseSDNode *LS = dyn_cast<LSBaseSDNode>(N)) {
+    EVT VT = LS->getMemoryVT();
+    SDValue Loc = LS->getBasePtr();
+    return isConsecutiveLSLoc(Loc, VT, Base, Bytes, Dist, DAG);
+  }
+
+  if (N->getOpcode() == ISD::INTRINSIC_W_CHAIN) {
+    EVT VT;
+    switch (cast<ConstantSDNode>(N->getOperand(1))->getZExtValue()) {
+    default: return false;
+    case Intrinsic::ppc_altivec_lvx:
+    case Intrinsic::ppc_altivec_lvxl:
+      VT = MVT::v4i32;
+      break;
+    case Intrinsic::ppc_altivec_lvebx:
+      VT = MVT::i8;
+      break;
+    case Intrinsic::ppc_altivec_lvehx:
+      VT = MVT::i16;
+      break;
+    case Intrinsic::ppc_altivec_lvewx:
+      VT = MVT::i32;
+      break;
+    }
+
+    return isConsecutiveLSLoc(N->getOperand(2), VT, Base, Bytes, Dist, DAG);
+  }
+
+  if (N->getOpcode() == ISD::INTRINSIC_VOID) {
+    EVT VT;
+    switch (cast<ConstantSDNode>(N->getOperand(1))->getZExtValue()) {
+    default: return false;
+    case Intrinsic::ppc_altivec_stvx:
+    case Intrinsic::ppc_altivec_stvxl:
+      VT = MVT::v4i32;
+      break;
+    case Intrinsic::ppc_altivec_stvebx:
+      VT = MVT::i8;
+      break;
+    case Intrinsic::ppc_altivec_stvehx:
+      VT = MVT::i16;
+      break;
+    case Intrinsic::ppc_altivec_stvewx:
+      VT = MVT::i32;
+      break;
+    }
+
+    return isConsecutiveLSLoc(N->getOperand(3), VT, Base, Bytes, Dist, DAG);
+  }
+
+  return false;
+}
+
 // Return true is there is a nearyby consecutive load to the one provided
 // (regardless of alignment). We search up and down the chain, looking though
 // token factors and other loads (but nothing else). As a result, a true result
@@ -7610,7 +7664,7 @@ static bool findConsecutiveLoad(LoadSDNode *LD, SelectionDAG &DAG) {
     if (!Visited.insert(ChainNext))
       continue;
 
-    if (LoadSDNode *ChainLD = dyn_cast<LoadSDNode>(ChainNext)) {
+    if (MemSDNode *ChainLD = dyn_cast<MemSDNode>(ChainNext)) {
       if (isConsecutiveLS(ChainLD, LD, VT.getStoreSize(), 1, DAG))
         return true;
 
@@ -7641,14 +7695,14 @@ static bool findConsecutiveLoad(LoadSDNode *LD, SelectionDAG &DAG) {
       if (!Visited.insert(LoadRoot))
         continue;
 
-      if (LoadSDNode *ChainLD = dyn_cast<LoadSDNode>(LoadRoot))
+      if (MemSDNode *ChainLD = dyn_cast<MemSDNode>(LoadRoot))
         if (isConsecutiveLS(ChainLD, LD, VT.getStoreSize(), 1, DAG))
           return true;
 
       for (SDNode::use_iterator UI = LoadRoot->use_begin(),
            UE = LoadRoot->use_end(); UI != UE; ++UI)
-        if (((isa<LoadSDNode>(*UI) &&
-            cast<LoadSDNode>(*UI)->getChain().getNode() == LoadRoot) ||
+        if (((isa<MemSDNode>(*UI) &&
+            cast<MemSDNode>(*UI)->getChain().getNode() == LoadRoot) ||
             UI->getOpcode() == ISD::TokenFactor) && !Visited.count(*UI))
           Queue.push_back(*UI);
     }
diff --git a/llvm/test/CodeGen/PowerPC/unal-altivec-wint.ll b/llvm/test/CodeGen/PowerPC/unal-altivec-wint.ll
new file mode 100644
index 00000000000..7e0963f54b3
--- /dev/null
+++ b/llvm/test/CodeGen/PowerPC/unal-altivec-wint.ll
@@ -0,0 +1,48 @@
+; RUN: llc -mcpu=pwr7 < %s | FileCheck %s
+target datalayout = "E-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-f128:128:128-v128:128:128-n32:64"
+target triple = "powerpc64-unknown-linux-gnu"
+
+declare <4 x i32> @llvm.ppc.altivec.lvx(i8*) #1
+
+define <4 x i32> @test1(<4 x i32>* %h) #0 {
+entry:
+  %h1 = getelementptr <4 x i32>* %h, i64 1
+  %hv = bitcast <4 x i32>* %h1 to i8*
+  %vl = call <4 x i32> @llvm.ppc.altivec.lvx(i8* %hv)
+
+  %v0 = load <4 x i32>* %h, align 8
+
+  %a = add <4 x i32> %v0, %vl
+  ret <4 x i32> %a
+
+; CHECK-LABEL: @test1
+; CHECK: li [[REG:[0-9]+]], 16
+; CHECK-NOT: li {{[0-9]+}}, 15
+; CHECK-DAG: lvx {{[0-9]+}}, 0, 3
+; CHECK-DAG: lvx {{[0-9]+}}, 3, [[REG]]
+; CHECK: blr
+}
+
+declare void @llvm.ppc.altivec.stvx(<4 x i32>, i8*) #0
+
+define <4 x i32> @test2(<4 x i32>* %h, <4 x i32> %d) #0 {
+entry:
+  %h1 = getelementptr <4 x i32>* %h, i64 1
+  %hv = bitcast <4 x i32>* %h1 to i8*
+  call void @llvm.ppc.altivec.stvx(<4 x i32> %d, i8* %hv)
+
+  %v0 = load <4 x i32>* %h, align 8
+
+  ret <4 x i32> %v0
+
+; CHECK-LABEL: @test2
+; CHECK: li [[REG:[0-9]+]], 16
+; CHECK-NOT: li {{[0-9]+}}, 15
+; CHECK-DAG: lvx {{[0-9]+}}, 0, 3
+; CHECK-DAG: lvx {{[0-9]+}}, 3, [[REG]]
+; CHECK: blr
+}
+
+attributes #0 = { nounwind }
+attributes #1 = { nounwind readonly }
+