reapply: Use the new TB_NOT_REVERSABLE flag instead of special

reapply: reimplement the second half of the or/add optimization.  We should now

with no changes.  Turns out that one missing "Defs = [EFLAGS]" can upset things
a bit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@116040 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86InstrCompiler.td b/lib/Target/X86/X86InstrCompiler.td
index 38248a8d6ea766c1bfe87db5e37249b2879c595a..4fdbdc5b8b1996207ae07cdbffd05c67b5bffb7c 100644 (file)
--- a/lib/Target/X86/X86InstrCompiler.td
+++ b/lib/Target/X86/X86InstrCompiler.td
@@ -1036,21 +1036,34 @@ def ADD32rr_DB  : I<0, Pseudo, (outs GR32:$dst), (ins GR32:$src1, GR32:$src2),
 def ADD64rr_DB  : I<0, Pseudo, (outs GR64:$dst), (ins GR64:$src1, GR64:$src2),
                     "", // orq/addq REG, REG
                     [(set GR64:$dst, (or_is_add GR64:$src1, GR64:$src2))]>;
+                    
+                    
+def ADD16ri_DB  : I<0, Pseudo, (outs GR16:$dst), (ins GR16:$src1, i16imm:$src2),
+                    "", // orw/addw REG, imm
+                    [(set GR16:$dst, (or_is_add GR16:$src1, imm:$src2))]>;
+def ADD32ri_DB  : I<0, Pseudo, (outs GR32:$dst), (ins GR32:$src1, i32imm:$src2),
+                    "", // orl/addl REG, imm
+                    [(set GR32:$dst, (or_is_add GR32:$src1, imm:$src2))]>;
+def ADD64ri32_DB : I<0, Pseudo,
+                     (outs GR64:$dst), (ins GR64:$src1, i64i32imm:$src2),
+                      "", // orq/addq REG, imm
+                      [(set GR64:$dst, (or_is_add GR64:$src1,
+                                                  i64immSExt32:$src2))]>;
+                    
+def ADD16ri8_DB : I<0, Pseudo,
+                    (outs GR16:$dst), (ins GR16:$src1, i16i8imm:$src2),
+                    "", // orw/addw REG, imm8
+                    [(set GR16:$dst,(or_is_add GR16:$src1,i16immSExt8:$src2))]>;
+def ADD32ri8_DB : I<0, Pseudo,
+                    (outs GR32:$dst), (ins GR32:$src1, i32i8imm:$src2),
+                    "", // orl/addl REG, imm8
+                    [(set GR32:$dst,(or_is_add GR32:$src1,i32immSExt8:$src2))]>;
+def ADD64ri8_DB : I<0, Pseudo,
+                    (outs GR64:$dst), (ins GR64:$src1, i64i8imm:$src2),
+                    "", // orq/addq REG, imm8
+                    [(set GR64:$dst, (or_is_add GR64:$src1,
+                                                i64immSExt8:$src2))]>;
 }
-
-def : Pat<(or_is_add GR16:$src1, imm:$src2),
-          (ADD16ri GR16:$src1, imm:$src2)>;
-def : Pat<(or_is_add GR32:$src1, imm:$src2),
-          (ADD32ri GR32:$src1, imm:$src2)>;
-def : Pat<(or_is_add GR64:$src1, i64immSExt32:$src2),
-          (ADD64ri32 GR64:$src1, i64immSExt32:$src2)>;
-          
-def : Pat<(or_is_add GR16:$src1, i16immSExt8:$src2),
-          (ADD16ri8 GR16:$src1, i16immSExt8:$src2)>;
-def : Pat<(or_is_add GR32:$src1, i32immSExt8:$src2),
-          (ADD32ri8 GR32:$src1, i32immSExt8:$src2)>;
-def : Pat<(or_is_add GR64:$src1, i64immSExt8:$src2),
-          (ADD64ri8 GR64:$src1, i64immSExt8:$src2)>;
 } // AddedComplexity
 
 

diff --git a/lib/Target/X86/X86InstrInfo.cpp b/lib/Target/X86/X86InstrInfo.cpp
index 12a5a9551526cace505000b26cca67c3d8120fee..204a187dc5d7fb37f784813d808ae005e35e71ca 100644 (file)
--- a/lib/Target/X86/X86InstrInfo.cpp
+++ b/lib/Target/X86/X86InstrInfo.cpp
@@ -68,14 +68,20 @@ X86InstrInfo::X86InstrInfo(X86TargetMachine &tm)
     { X86::ADC64rr,     X86::ADC64mr },
     { X86::ADD16ri,     X86::ADD16mi },
     { X86::ADD16ri8,    X86::ADD16mi8 },
+    { X86::ADD16ri_DB,  X86::ADD16mi  | TB_NOT_REVERSABLE },
+    { X86::ADD16ri8_DB, X86::ADD16mi8 | TB_NOT_REVERSABLE },
     { X86::ADD16rr,     X86::ADD16mr },
     { X86::ADD16rr_DB,  X86::ADD16mr | TB_NOT_REVERSABLE },
     { X86::ADD32ri,     X86::ADD32mi },
     { X86::ADD32ri8,    X86::ADD32mi8 },
+    { X86::ADD32ri_DB,  X86::ADD32mi | TB_NOT_REVERSABLE },
+    { X86::ADD32ri8_DB, X86::ADD32mi8 | TB_NOT_REVERSABLE },
     { X86::ADD32rr,     X86::ADD32mr },
     { X86::ADD32rr_DB,  X86::ADD32mr | TB_NOT_REVERSABLE },
     { X86::ADD64ri32,   X86::ADD64mi32 },
     { X86::ADD64ri8,    X86::ADD64mi8 },
+    { X86::ADD64ri32_DB,X86::ADD64mi32 | TB_NOT_REVERSABLE },
+    { X86::ADD64ri8_DB, X86::ADD64mi8 | TB_NOT_REVERSABLE },
     { X86::ADD64rr,     X86::ADD64mr },
     { X86::ADD64rr_DB,  X86::ADD64mr | TB_NOT_REVERSABLE },
     { X86::ADD8ri,      X86::ADD8mi },
@@ -263,8 +269,8 @@ X86InstrInfo::X86InstrInfo(X86TargetMachine &tm)
     { X86::DIV64r,      X86::DIV64m, 1, 0 },
     { X86::DIV8r,       X86::DIV8m, 1, 0 },
     { X86::EXTRACTPSrr, X86::EXTRACTPSmr, 0, 16 },
-    { X86::FsMOVAPDrr,  X86::MOVSDmr, 0, 0 },
-    { X86::FsMOVAPSrr,  X86::MOVSSmr, 0, 0 },
+    { X86::FsMOVAPDrr,  X86::MOVSDmr | TB_NOT_REVERSABLE , 0, 0 },
+    { X86::FsMOVAPSrr,  X86::MOVSSmr | TB_NOT_REVERSABLE , 0, 0 },
     { X86::IDIV16r,     X86::IDIV16m, 1, 0 },
     { X86::IDIV32r,     X86::IDIV32m, 1, 0 },
     { X86::IDIV64r,     X86::IDIV64m, 1, 0 },
@@ -323,18 +329,22 @@ X86InstrInfo::X86InstrInfo(X86TargetMachine &tm)
   };
 
   for (unsigned i = 0, e = array_lengthof(OpTbl0); i != e; ++i) {
-    unsigned RegOp = OpTbl0[i][0];
-    unsigned MemOp = OpTbl0[i][1];
-    unsigned Align = OpTbl0[i][3];
-    assert(!RegOp2MemOpTable0.count(RegOp) && "Duplicated entries?");
-    RegOp2MemOpTable0[RegOp] = std::make_pair(MemOp,Align);
+    unsigned RegOp      = OpTbl0[i][0];
+    unsigned MemOp      = OpTbl0[i][1] & ~TB_FLAGS;
     unsigned FoldedLoad = OpTbl0[i][2];
+    unsigned Align      = OpTbl0[i][3];
+    assert(!RegOp2MemOpTable0.count(RegOp) && "Duplicated entries?");
+    RegOp2MemOpTable0[RegOp] = std::make_pair(MemOp, Align);
+    
+    // If this is not a reversable operation (because there is a many->one)
+    // mapping, don't insert the reverse of the operation into MemOp2RegOpTable.
+    if (OpTbl0[i][1] & TB_NOT_REVERSABLE)
+      continue;
+    
     // Index 0, folded load or store.
     unsigned AuxInfo = 0 | (FoldedLoad << 4) | ((FoldedLoad^1) << 5);
-    if (RegOp != X86::FsMOVAPDrr && RegOp != X86::FsMOVAPSrr) {
-      assert(!MemOp2RegOpTable.count(MemOp) && "Duplicated entries?");
-      MemOp2RegOpTable[MemOp] = std::make_pair(RegOp, AuxInfo);
-    }
+    assert(!MemOp2RegOpTable.count(MemOp) && "Duplicated entries?");
+    MemOp2RegOpTable[MemOp] = std::make_pair(RegOp, AuxInfo);
   }
 
   static const unsigned OpTbl1[][3] = {
@@ -352,8 +362,8 @@ X86InstrInfo::X86InstrInfo(X86TargetMachine &tm)
     { X86::CVTTSD2SIrr,     X86::CVTTSD2SIrm, 0 },
     { X86::CVTTSS2SI64rr,   X86::CVTTSS2SI64rm, 0 },
     { X86::CVTTSS2SIrr,     X86::CVTTSS2SIrm, 0 },
-    { X86::FsMOVAPDrr,      X86::MOVSDrm, 0 },
-    { X86::FsMOVAPSrr,      X86::MOVSSrm, 0 },
+    { X86::FsMOVAPDrr,      X86::MOVSDrm | TB_NOT_REVERSABLE , 0 },
+    { X86::FsMOVAPSrr,      X86::MOVSSrm | TB_NOT_REVERSABLE , 0 },
     { X86::IMUL16rri,       X86::IMUL16rmi, 0 },
     { X86::IMUL16rri8,      X86::IMUL16rmi8, 0 },
     { X86::IMUL32rri,       X86::IMUL32rmi, 0 },
@@ -449,17 +459,20 @@ X86InstrInfo::X86InstrInfo(X86TargetMachine &tm)
 
   for (unsigned i = 0, e = array_lengthof(OpTbl1); i != e; ++i) {
     unsigned RegOp = OpTbl1[i][0];
-    unsigned MemOp = OpTbl1[i][1];
+    unsigned MemOp = OpTbl1[i][1] & ~TB_FLAGS;
     unsigned Align = OpTbl1[i][2];
     assert(!RegOp2MemOpTable1.count(RegOp) && "Duplicate entries");
-    RegOp2MemOpTable1[RegOp] = std::make_pair(MemOp,Align);
+    RegOp2MemOpTable1[RegOp] = std::make_pair(MemOp, Align);
+    
+    // If this is not a reversable operation (because there is a many->one)
+    // mapping, don't insert the reverse of the operation into MemOp2RegOpTable.
+    if (OpTbl1[i][1] & TB_NOT_REVERSABLE)
+      continue;
     
     // Index 1, folded load
     unsigned AuxInfo = 1 | (1 << 4);
-    if (RegOp != X86::FsMOVAPDrr && RegOp != X86::FsMOVAPSrr) {
-      assert(!MemOp2RegOpTable.count(MemOp) && "Duplicate entries");
-      MemOp2RegOpTable[MemOp] = std::make_pair(RegOp, AuxInfo);
-    }
+    assert(!MemOp2RegOpTable.count(MemOp) && "Duplicate entries");
+    MemOp2RegOpTable[MemOp] = std::make_pair(RegOp, AuxInfo);
   }
 
   static const unsigned OpTbl2[][3] = {
@@ -671,7 +684,6 @@ X86InstrInfo::X86InstrInfo(X86TargetMachine &tm)
     assert(!RegOp2MemOpTable2.count(RegOp) && "Duplicate entry!");
     RegOp2MemOpTable2[RegOp] = std::make_pair(MemOp, Align);
     
-    
     // If this is not a reversable operation (because there is a many->one)
     // mapping, don't insert the reverse of the operation into MemOp2RegOpTable.
     if (OpTbl2[i][1] & TB_NOT_REVERSABLE)
@@ -1154,6 +1166,8 @@ X86InstrInfo::convertToThreeAddressWithLEA(unsigned MIOpc,
     break;
   case X86::ADD16ri:
   case X86::ADD16ri8:
+  case X86::ADD16ri_DB:
+  case X86::ADD16ri8_DB:
     addRegOffset(MIB, leaInReg, true, MI->getOperand(2).getImm());    
     break;
   case X86::ADD16rr:
@@ -1418,6 +1432,8 @@ X86InstrInfo::convertToThreeAddress(MachineFunction::iterator &MFI,
     }
     case X86::ADD64ri32:
     case X86::ADD64ri8:
+    case X86::ADD64ri32_DB:
+    case X86::ADD64ri8_DB:
       assert(MI->getNumOperands() >= 3 && "Unknown add instruction!");
       NewMI = addRegOffset(BuildMI(MF, MI->getDebugLoc(), get(X86::LEA64r))
                               .addReg(Dest, RegState::Define |
@@ -1425,7 +1441,9 @@ X86InstrInfo::convertToThreeAddress(MachineFunction::iterator &MFI,
                               Src, isKill, MI->getOperand(2).getImm());
       break;
     case X86::ADD32ri:
-    case X86::ADD32ri8: {
+    case X86::ADD32ri8:
+    case X86::ADD32ri_DB:
+    case X86::ADD32ri8_DB: {
       assert(MI->getNumOperands() >= 3 && "Unknown add instruction!");
       unsigned Opc = is64Bit ? X86::LEA64_32r : X86::LEA32r;
       NewMI = addRegOffset(BuildMI(MF, MI->getDebugLoc(), get(Opc))
@@ -1436,6 +1454,8 @@ X86InstrInfo::convertToThreeAddress(MachineFunction::iterator &MFI,
     }
     case X86::ADD16ri:
     case X86::ADD16ri8:
+    case X86::ADD16ri_DB:
+    case X86::ADD16ri8_DB:
       if (DisableLEA16)
         return is64Bit ? convertToThreeAddressWithLEA(MIOpc, MFI, MBBI, LV) : 0;
       assert(MI->getNumOperands() >= 3 && "Unknown add instruction!");

diff --git a/lib/Target/X86/X86MCInstLower.cpp b/lib/Target/X86/X86MCInstLower.cpp
index 662717bae713fc441b025de7a277ace295d06606..8f90d872829185f29870084f8eff963018f0360f 100644 (file)
--- a/lib/Target/X86/X86MCInstLower.cpp
+++ b/lib/Target/X86/X86MCInstLower.cpp
@@ -437,9 +437,15 @@ ReSimplify:
   // These are pseudo-ops for OR to help with the OR->ADD transformation.  We do
   // this with an ugly goto in case the resultant OR uses EAX and needs the
   // short form.
-  case X86::ADD16rr_DB: OutMI.setOpcode(X86::OR16rr); goto ReSimplify;
-  case X86::ADD32rr_DB: OutMI.setOpcode(X86::OR32rr); goto ReSimplify;
-  case X86::ADD64rr_DB: OutMI.setOpcode(X86::OR64rr); goto ReSimplify;
+  case X86::ADD16rr_DB:   OutMI.setOpcode(X86::OR16rr); goto ReSimplify;
+  case X86::ADD32rr_DB:   OutMI.setOpcode(X86::OR32rr); goto ReSimplify;
+  case X86::ADD64rr_DB:   OutMI.setOpcode(X86::OR64rr); goto ReSimplify;
+  case X86::ADD16ri_DB:   OutMI.setOpcode(X86::OR16ri); goto ReSimplify;
+  case X86::ADD32ri_DB:   OutMI.setOpcode(X86::OR32ri); goto ReSimplify;
+  case X86::ADD64ri32_DB: OutMI.setOpcode(X86::OR64ri32); goto ReSimplify;
+  case X86::ADD16ri8_DB:  OutMI.setOpcode(X86::OR16ri8); goto ReSimplify;
+  case X86::ADD32ri8_DB:  OutMI.setOpcode(X86::OR32ri8); goto ReSimplify;
+  case X86::ADD64ri8_DB:  OutMI.setOpcode(X86::OR64ri8); goto ReSimplify;
       
   // The assembler backend wants to see branches in their small form and relax
   // them to their large form.  The JIT can only handle the large form because

diff --git a/test/CodeGen/X86/3addr-or.ll b/test/CodeGen/X86/3addr-or.ll
index 30a1f36850de16926b94318b0b04583ac445dc33..912bdc215474ee2b90890489f1ab10548a9152f7 100644 (file)
--- a/test/CodeGen/X86/3addr-or.ll
+++ b/test/CodeGen/X86/3addr-or.ll
@@ -1,9 +1,9 @@
 ; RUN: llc < %s -mtriple=x86_64-apple-darwin | FileCheck %s
 ; rdar://7527734
 
-define i32 @test(i32 %x) nounwind readnone ssp {
+define i32 @test1(i32 %x) nounwind readnone ssp {
 entry:
-; CHECK: test:
+; CHECK: test1:
 ; CHECK: leal 3(%rdi), %eax
   %0 = shl i32 %x, 5                              ; <i32> [#uses=1]
   %1 = or i32 %0, 3                               ; <i32> [#uses=1]
@@ -25,3 +25,37 @@ define i64 @test2(i8 %A, i8 %B) nounwind {
   %H = or i64 %G, %E                              ; <i64> [#uses=1]
   ret i64 %H
 }
+
+;; Test that OR is only emitted as LEA, not as ADD.
+
+define void @test3(i32 %x, i32* %P) nounwind readnone ssp {
+entry:
+; No reason to emit an add here, should be an or.
+; CHECK: test3:
+; CHECK: orl $3, %edi
+  %0 = shl i32 %x, 5
+  %1 = or i32 %0, 3
+  store i32 %1, i32* %P
+  ret void
+}
+
+define i32 @test4(i32 %a, i32 %b) nounwind readnone ssp {
+entry:
+  %and = and i32 %a, 6
+  %and2 = and i32 %b, 16
+  %or = or i32 %and2, %and
+  ret i32 %or
+; CHECK: test4:
+; CHECK: leal  (%rsi,%rdi), %eax
+}
+
+define void @test5(i32 %a, i32 %b, i32* nocapture %P) nounwind ssp {
+entry:
+  %and = and i32 %a, 6
+  %and2 = and i32 %b, 16
+  %or = or i32 %and2, %and
+  store i32 %or, i32* %P, align 4
+  ret void
+; CHECK: test5:
+; CHECK: orl
+}