Remove the pmulld intrinsic and autoupdate it as a vector multiply.

Rewrite the pmulld patterns, and make sure that they fold in loads of
arguments into the instruction.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@99910 91177308-0d34-0410-b5e6-96231b3b80d8

3 files changed, 39 insertions, 3 deletions

diff --git a/lib/Target/X86/X86InstrInfo.cpp b/lib/Target/X86/X86InstrInfo.cpp
index 5def78737f..614a21182b 100644
--- a/lib/Target/X86/X86InstrInfo.cpp
+++ b/lib/Target/X86/X86InstrInfo.cpp
@@ -597,7 +597,6 @@ X86InstrInfo::X86InstrInfo(X86TargetMachine &tm)
     { X86::PMULHUWrr,       X86::PMULHUWrm, 16 },
     { X86::PMULHWrr,        X86::PMULHWrm, 16 },
     { X86::PMULLDrr,        X86::PMULLDrm, 16 },
-    { X86::PMULLDrr_int,    X86::PMULLDrm_int, 16 },
     { X86::PMULLWrr,        X86::PMULLWrm, 16 },
     { X86::PMULUDQrr,       X86::PMULUDQrm, 16 },
     { X86::PORrr,           X86::PORrm, 16 },
diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index 077d92de4f..e207598144 100644
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -3448,8 +3448,28 @@ let Constraints = "$src1 = $dst" in {
                        OpSize;
   }
 }
-defm PMULLD       : SS41I_binop_patint<0x40, "pmulld", v4i32, mul,
-                                       int_x86_sse41_pmulld, 1>;
+
+/// SS48I_binop_rm - Simple SSE41 binary operator.
+let Constraints = "$src1 = $dst" in {
+multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
+                        ValueType OpVT, bit Commutable = 0> {
+  def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), 
+                                 (ins VR128:$src1, VR128:$src2),
+               !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
+               [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>,
+               OpSize {
+    let isCommutable = Commutable;
+  }
+  def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), 
+                                 (ins VR128:$src1, i128mem:$src2),
+               !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
+               [(set VR128:$dst, (OpNode VR128:$src1,
+                                  (bc_v4i32 (memopv2i64 addr:$src2))))]>,
+               OpSize;
+}
+}
+
+defm PMULLD         : SS48I_binop_rm<0x40, "pmulld", mul, v4i32, 1>;
 
 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
 let Constraints = "$src1 = $dst" in {
diff --git a/lib/VMCore/AutoUpgrade.cpp b/lib/VMCore/AutoUpgrade.cpp
index 5e4c9fb766..b9aa5c3467 100644
--- a/lib/VMCore/AutoUpgrade.cpp
+++ b/lib/VMCore/AutoUpgrade.cpp
@@ -225,7 +225,12 @@ static bool UpgradeIntrinsicFunction1(Function *F, Function *&NewFn) {
       // Calls to these intrinsics are transformed into ShuffleVector's.
       NewFn = 0;
       return true;
+    } else if (Name.compare(5, 16, "x86.sse41.pmulld", 16) == 0) {
+      // Calls to these intrinsics are transformed into vector multiplies.
+      NewFn = 0;
+      return true;
     }
+    
 
     break;
   }
@@ -355,6 +360,18 @@ void llvm::UpgradeIntrinsicCall(CallInst *CI, Function *NewFn) {
       
       //  Clean up the old call now that it has been completely upgraded.
       CI->eraseFromParent();
+    } else if (F->getName() == "llvm.x86.sse41.pmulld") {
+      // Upgrade this set of intrinsics into vector multiplies.
+      Instruction *Mul = BinaryOperator::CreateMul(CI->getOperand(1),
+                                                   CI->getOperand(2),
+                                                   CI->getName(),
+                                                   CI);
+      // Fix up all the uses with our new multiply.
+      if (!CI->use_empty())
+        CI->replaceAllUsesWith(Mul);
+        
+      // Remove upgraded multiply.
+      CI->eraseFromParent();
     } else {
       llvm_unreachable("Unknown function for CallInst upgrade.");
     }