From: Nate Begeman <natebegeman@mac.com>
Date: Fri, 25 Jul 2008 19:05:58 +0000 (+0000)
Subject: Disable mov{L, LP, HP, HLP, *DUP} shuffles for mmx
X-Git-Url: http://demsky.eecs.uci.edu/git/?a=commitdiff_plain;h=fb8ead0c20f347fc1f0385f801ee0636d1ff6289;p=oota-llvm.git

Disable mov{L, LP, HP, HLP, *DUP} shuffles for mmx
mmx needs its own fancy shuffle logic based on unpack; for now we get correct but awful code.

Also commit Mon Ping's VSETCC patch


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@54039 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index e8e1b2e7ea0..00c8eeade5b 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -3881,14 +3881,15 @@ X86TargetLowering::LowerVECTOR_SHUFFLE(SDOperand Op, SelectionDAG &DAG) {
       return V2;
     if (ISD::isBuildVectorAllZeros(V1.Val))
       return getVZextMovL(VT, VT, V2, DAG, Subtarget);
-    return Op;
+    if (!isMMX)
+      return Op;
   }
 
-  if (X86::isMOVSHDUPMask(PermMask.Val) ||
-      X86::isMOVSLDUPMask(PermMask.Val) ||
-      X86::isMOVHLPSMask(PermMask.Val) ||
-      X86::isMOVHPMask(PermMask.Val) ||
-      X86::isMOVLPMask(PermMask.Val))
+  if (!isMMX && (X86::isMOVSHDUPMask(PermMask.Val) ||
+                 X86::isMOVSLDUPMask(PermMask.Val) ||
+                 X86::isMOVHLPSMask(PermMask.Val) ||
+                 X86::isMOVHPMask(PermMask.Val) ||
+                 X86::isMOVLPMask(PermMask.Val)))
     return Op;
 
   if (ShouldXformToMOVHLPS(PermMask.Val) ||
@@ -4772,6 +4773,7 @@ SDOperand X86TargetLowering::LowerVSETCC(SDOperand Op, SelectionDAG &DAG) {
 
     switch (SetCCOpcode) {
     default: break;
+    case ISD::SETOEQ:
     case ISD::SETEQ:  SSECC = 0; break;
     case ISD::SETOGT: 
     case ISD::SETGT: Swap = true; // Fallthrough
@@ -4782,7 +4784,7 @@ SDOperand X86TargetLowering::LowerVSETCC(SDOperand Op, SelectionDAG &DAG) {
     case ISD::SETLE:
     case ISD::SETOLE: SSECC = 2; break;
     case ISD::SETUO:  SSECC = 3; break;
-    case ISD::SETONE:
+    case ISD::SETUNE:
     case ISD::SETNE:  SSECC = 4; break;
     case ISD::SETULE: Swap = true;
     case ISD::SETUGE: SSECC = 5; break;
@@ -4793,15 +4795,21 @@ SDOperand X86TargetLowering::LowerVSETCC(SDOperand Op, SelectionDAG &DAG) {
     if (Swap)
       std::swap(Op0, Op1);
 
-    // In the one special case we can't handle, emit two comparisons.
+    // In the two special cases we can't handle, emit two comparisons.
     if (SSECC == 8) {
-      SDOperand UNORD, EQ;
-
-      assert(SetCCOpcode == ISD::SETUEQ && "Illegal FP comparison");
-      
-      UNORD = DAG.getNode(Opc, VT, Op0, Op1, DAG.getConstant(3, MVT::i8));
-      EQ = DAG.getNode(Opc, VT, Op0, Op1, DAG.getConstant(0, MVT::i8));
-      return DAG.getNode(ISD::OR, VT, UNORD, EQ);
+      if (SetCCOpcode == ISD::SETUEQ) {
+        SDOperand UNORD, EQ;
+        UNORD = DAG.getNode(Opc, VT, Op0, Op1, DAG.getConstant(3, MVT::i8));
+        EQ = DAG.getNode(Opc, VT, Op0, Op1, DAG.getConstant(0, MVT::i8));
+        return DAG.getNode(ISD::OR, VT, UNORD, EQ);
+      }
+      else if (SetCCOpcode == ISD::SETONE) {
+        SDOperand ORD, NEQ;
+        ORD = DAG.getNode(Opc, VT, Op0, Op1, DAG.getConstant(7, MVT::i8));
+        NEQ = DAG.getNode(Opc, VT, Op0, Op1, DAG.getConstant(4, MVT::i8));
+        return DAG.getNode(ISD::AND, VT, ORD, NEQ);
+      }
+      assert(0 && "Illegal FP comparison");
     }
     // Handle all other FP comparisons here.
     return DAG.getNode(Opc, VT, Op0, Op1, DAG.getConstant(SSECC, MVT::i8));
diff --git a/test/CodeGen/X86/vec_insert-7.ll b/test/CodeGen/X86/vec_insert-7.ll
new file mode 100644
index 00000000000..93af9f4d414
--- /dev/null
+++ b/test/CodeGen/X86/vec_insert-7.ll
@@ -0,0 +1,8 @@
+; RUN: llvm-as < %s | llc -march=x86 -mattr=+mmx -mtriple=i686-apple-darwin9 -stats -info-output-file - | grep asm-printer | grep punpckldq
+
+define <2 x i32> @mmx_movzl(<2 x i32> %x) nounwind  {
+entry:
+	%tmp3 = insertelement <2 x i32> %x, i32 32, i32 0		; <<2 x i32>> [#uses=1]
+	%tmp8 = insertelement <2 x i32> %tmp3, i32 0, i32 1		; <<2 x i32>> [#uses=1]
+	ret <2 x i32> %tmp8
+}