[FastISel][X86] Emit more efficient instructions for integer constant materialization.

This mostly affects the i64 value type, which always resulted in an 15byte
mobavsq instruction to materialize any constant. The custom code checks the
value of the immediate and tries to use a different and smaller mov
instruction when possible.

This fixes <rdar://problem/17420988>.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@215593 91177308-0d34-0410-b5e6-96231b3b80d8

Author: ributzka

lib/Target/X86/X86FastISel.cpp

@@ -3105,7 +3105,34 @@ X86FastISel::TargetSelectInstruction(const Instruction *I)  {
 unsigned X86FastISel::X86MaterializeInt(const ConstantInt *CI, MVT VT) {
   if (VT > MVT::i64)
     return 0;
-  return FastEmit_i(VT, VT, ISD::Constant, CI->getZExtValue());
+
+  uint64_t Imm = CI->getZExtValue();
+  unsigned Opc = 0;
+  switch (VT.SimpleTy) {
+  default: llvm_unreachable("Unexpected value type");
+  case MVT::i1:  VT = MVT::i8; // fall-through
+  case MVT::i8:  Opc = X86::MOV8ri;  break;
+  case MVT::i16: Opc = X86::MOV16ri; break;
+  case MVT::i32: Opc = X86::MOV32ri; break;
+  case MVT::i64: {
+    if (isUInt<32>(Imm))
+      Opc = X86::MOV32ri;
+    else if (isInt<32>(Imm))
+      Opc = X86::MOV64ri32;
+    else
+      Opc = X86::MOV64ri;
+    break;
+  }
+  }
+  if (VT == MVT::i64 && Opc == X86::MOV32ri) {
+    unsigned SrcReg = FastEmitInst_i(Opc, &X86::GR32RegClass, Imm);
+    unsigned ResultReg = createResultReg(&X86::GR64RegClass);
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
+            TII.get(TargetOpcode::SUBREG_TO_REG), ResultReg)
+      .addImm(0).addReg(SrcReg).addImm(X86::sub_32bit);
+    return ResultReg;
+  }
+  return FastEmitInst_i(Opc, TLI.getRegClassFor(VT), Imm);
 }
 
 unsigned X86FastISel::X86MaterializeFP(const ConstantFP *CFP, MVT VT) {

test/CodeGen/X86/object-size.ll

@@ -1,4 +1,4 @@
-; RUN: llc -O0 < %s -march=x86-64 | FileCheck %s -check-prefix=X64
+; RUN: llc -O0 < %s -march=x86-64 | FileCheck %s
 
 ; ModuleID = 'ts.c'
 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-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"
@@ -12,8 +12,8 @@ entry:
   %tmp = load i8** @p                             ; <i8*> [#uses=1]
   %0 = call i64 @llvm.objectsize.i64.p0i8(i8* %tmp, i1 0) ; <i64> [#uses=1]
   %cmp = icmp ne i64 %0, -1                       ; <i1> [#uses=1]
-; X64: movabsq $-1, [[RAX:%r..]]
-; X64: cmpq    $-1, [[RAX]]
+; CHECK: movq $-1, [[RAX:%r..]]
+; CHECK: cmpq $-1, [[RAX]]
   br i1 %cmp, label %cond.true, label %cond.false
 
 cond.true:                                        ; preds = %entry