[mips] Use MTHC1 when it is available (MIPS32r2 and later) for both FP32 and FP64

Summary:
To make this work for both AFGR64 and FGR64 register sets, I've had to make the
instruction definition consistent with the white lie (that it reads the lower
32-bits of the register) when they are generated by expandBuildPairF64().

Corrected the definition of hasMips32r2() and hasMips64r2() to include
MIPS32r6 and MIPS64r6.

Depends on D3956

Reviewers: jkolek, zoran.jovanovic, vmedic

Reviewed By: vmedic

Differential Revision: http://reviews.llvm.org/D3957

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@210771 91177308-0d34-0410-b5e6-96231b3b80d8

9 files changed, 63 insertions, 34 deletions

diff --git a/lib/Target/Mips/MipsInstrFPU.td b/lib/Target/Mips/MipsInstrFPU.td
index a5961105e4..b29fa6e6da 100644
--- a/lib/Target/Mips/MipsInstrFPU.td
+++ b/lib/Target/Mips/MipsInstrFPU.td
@@ -153,6 +153,15 @@ class MTC1_FT<string opstr, RegisterOperand DstRC, RegisterOperand SrcRC,
   InstSE<(outs DstRC:$fs), (ins SrcRC:$rt), !strconcat(opstr, "\t$rt, $fs"),
          [(set DstRC:$fs, (OpNode SrcRC:$rt))], Itin, FrmFR, opstr>;
 
+class MTC1_64_FT<string opstr, RegisterOperand DstRC, RegisterOperand SrcRC,
+                 InstrItinClass Itin> :
+  InstSE<(outs DstRC:$fs), (ins DstRC:$fs_in, SrcRC:$rt),
+         !strconcat(opstr, "\t$rt, $fs"), [], Itin, FrmFR, opstr> {
+  // $fs_in is part of a white lie to work around a widespread bug in the FPU
+  // implementation. See expandBuildPairF64 for details.
+  let Constraints = "$fs = $fs_in";
+}
+
 class LW_FT<string opstr, RegisterOperand RC, InstrItinClass Itin,
             SDPatternOperator OpNode= null_frag> :
   InstSE<(outs RC:$rt), (ins mem:$addr), !strconcat(opstr, "\t$rt, $addr"),
@@ -355,8 +364,12 @@ def MTC1 : MMRel, MTC1_FT<"mtc1", FGR32Opnd, GPR32Opnd, II_MTC1,
                           bitconvert>, MFC1_FM<4>;
 def MFHC1 : MMRel, MFC1_FT<"mfhc1", GPR32Opnd, FGRH32Opnd, II_MFHC1>,
             MFC1_FM<3>, ISA_MIPS32R2;
-def MTHC1 : MMRel, MTC1_FT<"mthc1", FGRH32Opnd, GPR32Opnd, II_MTHC1>,
-            MFC1_FM<7>, ISA_MIPS32R2;
+def MTHC1_D32 : MMRel, MTC1_64_FT<"mthc1", FGR64Opnd, GPR32Opnd, II_MTHC1>,
+                MFC1_FM<7>, ISA_MIPS32R2, AdditionalRequires<[NotFP64bit]>;
+def MTHC1_D64 : MTC1_64_FT<"mthc1", AFGR64Opnd, GPR32Opnd, II_MTHC1>,
+                MFC1_FM<7>, ISA_MIPS32R2, AdditionalRequires<[IsFP64bit]> {
+  let DecoderNamespace = "Mips64";
+}
 def DMFC1 : MFC1_FT<"dmfc1", GPR64Opnd, FGR64Opnd, II_DMFC1,
             bitconvert>, MFC1_FM<1>, ISA_MIPS3;
 def DMTC1 : MTC1_FT<"dmtc1", FGR64Opnd, GPR64Opnd, II_DMTC1,
diff --git a/lib/Target/Mips/MipsSEInstrInfo.cpp b/lib/Target/Mips/MipsSEInstrInfo.cpp
index f6f364f1a3..e82c8cff72 100644
--- a/lib/Target/Mips/MipsSEInstrInfo.cpp
+++ b/lib/Target/Mips/MipsSEInstrInfo.cpp
@@ -542,20 +542,31 @@ void MipsSEInstrInfo::expandBuildPairF64(MachineBasicBlock &MBB,
   const MCInstrDesc& Mtc1Tdd = get(Mips::MTC1);
   DebugLoc dl = I->getDebugLoc();
   const TargetRegisterInfo &TRI = getRegisterInfo();
+  bool HasMTHC1 = TM.getSubtarget<MipsSubtarget>().hasMips32r2() ||
+                  TM.getSubtarget<MipsSubtarget>().hasMips32r6();
 
-  // For FP32 mode:
-  //   mtc1 Lo, $fp
-  //   mtc1 Hi, $fp + 1
-  // For FP64 mode:
+  // When mthc1 is available, use:
   //   mtc1 Lo, $fp
   //   mthc1 Hi, $fp
+  //
+  // Otherwise, for FP64:
+  //   spill + reload via ldc1
+  // This has not been implemented since FP64 on MIPS32 and earlier is not
+  // supported.
+  //
+  // Otherwise, for FP32:
+  //   mtc1 Lo, $fp
+  //   mtc1 Hi, $fp + 1
 
   BuildMI(MBB, I, dl, Mtc1Tdd, TRI.getSubReg(DstReg, Mips::sub_lo))
     .addReg(LoReg);
 
-  if (FP64) {
-    // FIXME: The .addReg(DstReg, RegState::Implicit) is a white lie used to
-    //        temporarily work around a widespread bug in the -mfp64 support.
+  if (HasMTHC1 || FP64) {
+    assert(TM.getSubtarget<MipsSubtarget>().hasMips32r2() &&
+           "MTHC1 requires MIPS32r2");
+
+    // FIXME: The .addReg(DstReg) is a white lie used to temporarily work
+    //        around a widespread bug in the -mfp64 support.
     //        The problem is that none of the 32-bit fpu ops mention the fact
     //        that they clobber the upper 32-bits of the 64-bit FPR. Fixing that
     //        requires a major overhaul of the FPU implementation which can't
@@ -565,9 +576,9 @@ void MipsSEInstrInfo::expandBuildPairF64(MachineBasicBlock &MBB,
     //        We therefore pretend that it reads the bottom 32-bits to
     //        artificially create a dependency and prevent the scheduler
     //        changing the behaviour of the code.
-    BuildMI(MBB, I, dl, get(Mips::MTHC1), TRI.getSubReg(DstReg, Mips::sub_hi))
-        .addReg(HiReg)
-        .addReg(DstReg, RegState::Implicit);
+    BuildMI(MBB, I, dl, get(FP64 ? Mips::MTHC1_D64 : Mips::MTHC1_D32), DstReg)
+        .addReg(DstReg)
+        .addReg(HiReg);
   } else
     BuildMI(MBB, I, dl, Mtc1Tdd, TRI.getSubReg(DstReg, Mips::sub_hi))
       .addReg(HiReg);
diff --git a/lib/Target/Mips/MipsSubtarget.h b/lib/Target/Mips/MipsSubtarget.h
index 373f481362..6824ebf36d 100644
--- a/lib/Target/Mips/MipsSubtarget.h
+++ b/lib/Target/Mips/MipsSubtarget.h
@@ -159,12 +159,17 @@ public:
   bool hasMips4_32() const { return HasMips4_32; }
   bool hasMips4_32r2() const { return HasMips4_32r2; }
   bool hasMips32() const { return MipsArchVersion >= Mips32; }
-  bool hasMips32r2() const { return MipsArchVersion == Mips32r2 ||
-                                   MipsArchVersion == Mips64r2; }
-  bool hasMips32r6() const { return MipsArchVersion == Mips32r6 ||
-                                   MipsArchVersion == Mips64r6; }
+  bool hasMips32r2() const {
+    return MipsArchVersion == Mips32r2 || MipsArchVersion == Mips32r6 ||
+           MipsArchVersion == Mips64r2;
+  }
+  bool hasMips32r6() const {
+    return MipsArchVersion == Mips32r6 || MipsArchVersion == Mips64r6;
+  }
   bool hasMips64() const { return MipsArchVersion >= Mips64; }
-  bool hasMips64r2() const { return MipsArchVersion == Mips64r2; }
+  bool hasMips64r2() const {
+    return MipsArchVersion == Mips64r2 || MipsArchVersion == Mips64r6;
+  }
   bool hasMips64r6() const { return MipsArchVersion == Mips64r6; }
 
   bool hasCnMips() const { return HasCnMips; }
diff --git a/test/CodeGen/Mips/2013-11-18-fp64-const0.ll b/test/CodeGen/Mips/2013-11-18-fp64-const0.ll
index f8390d9a1c..6a210a0c76 100644
--- a/test/CodeGen/Mips/2013-11-18-fp64-const0.ll
+++ b/test/CodeGen/Mips/2013-11-18-fp64-const0.ll
@@ -1,5 +1,5 @@
 ; RUN: llc -march=mips -mattr=-fp64 < %s | FileCheck -check-prefix=CHECK-FP32 %s
-; RUN: llc -march=mips -mattr=+fp64 < %s | FileCheck -check-prefix=CHECK-FP64 %s
+; RUN: llc -march=mips -mcpu=mips32r2 -mattr=+fp64 < %s | FileCheck -check-prefix=CHECK-FP64 %s
 
 ; This test case is a simplified version of an llvm-stress generated test with
 ; seed=3718491962.
diff --git a/test/CodeGen/Mips/Fast-ISel/simplestorefp1.ll b/test/CodeGen/Mips/Fast-ISel/simplestorefp1.ll
index a374470d89..6759c01c77 100644
--- a/test/CodeGen/Mips/Fast-ISel/simplestorefp1.ll
+++ b/test/CodeGen/Mips/Fast-ISel/simplestorefp1.ll
@@ -28,9 +28,9 @@ entry:
 ; CHECK:  ori  $[[REG2a:[0-9]+]], $[[REG1a]], 49353
 ; CHECK:  lui  $[[REG1b:[0-9]+]], 21403
 ; CHECK:  ori  $[[REG2b:[0-9]+]], $[[REG1b]], 34951
-; CHECK:  mtc1  $[[REG2b]], $f[[REG3b:[0-9]+]]
-; CHECK:  mtc1  $[[REG2a]], $f[[REG3a:[0-9]+]]
-; CHECK:  sdc1  $f[[REG3b]], 0(${{[0-9]+}})
+; CHECK:  mtc1  $[[REG2b]], $f[[REG3:[0-9]+]]
+; CHECK:  mthc1  $[[REG2a]], $f[[REG3]]
+; CHECK:  sdc1  $f[[REG3]], 0(${{[0-9]+}})
 ; CHECK:  .end  d1
   ret void
 }
diff --git a/test/CodeGen/Mips/buildpairextractelementf64.ll b/test/CodeGen/Mips/buildpairextractelementf64.ll
index b9bf2b60a6..88d1d07e29 100644
--- a/test/CodeGen/Mips/buildpairextractelementf64.ll
+++ b/test/CodeGen/Mips/buildpairextractelementf64.ll
@@ -1,7 +1,7 @@
 ; RUN: llc -march=mipsel < %s | FileCheck %s -check-prefix=FP32 -check-prefix=CHECK
 ; RUN: llc -march=mips  < %s | FileCheck %s -check-prefix=FP32 -check-prefix=CHECK
-; RUN: llc -march=mipsel -mattr=+fp64 < %s | FileCheck %s -check-prefix=FP64 -check-prefix=CHECK
-; RUN: llc -march=mips -mattr=+fp64 < %s | FileCheck %s -check-prefix=FP64 -check-prefix=CHECK
+; RUN: llc -march=mipsel -mcpu=mips32r2 -mattr=+fp64 < %s | FileCheck %s -check-prefix=FP64 -check-prefix=CHECK
+; RUN: llc -march=mips -mcpu=mips32r2 -mattr=+fp64 < %s | FileCheck %s -check-prefix=FP64 -check-prefix=CHECK
 
 @a = external global i32
 
diff --git a/test/CodeGen/Mips/fcopysign.ll b/test/CodeGen/Mips/fcopysign.ll
index 44c4117510..3a9d9c73b2 100644
--- a/test/CodeGen/Mips/fcopysign.ll
+++ b/test/CodeGen/Mips/fcopysign.ll
@@ -17,7 +17,7 @@ entry:
 
 ; 32R2: ext  $[[EXT:[0-9]+]], ${{[0-9]+}}, 31, 1
 ; 32R2: ins  $[[INS:[0-9]+]], $[[EXT]], 31, 1
-; 32R2: mtc1 $[[INS]], $f1
+; 32R2: mthc1 $[[INS]], $f0
 
 ; 64: daddiu $[[T0:[0-9]+]], $zero, 1
 ; 64: dsll   $[[MSK1:[0-9]+]], $[[T0]], 63
diff --git a/test/CodeGen/Mips/fmadd1.ll b/test/CodeGen/Mips/fmadd1.ll
index bd672ef4b0..271631efb4 100644
--- a/test/CodeGen/Mips/fmadd1.ll
+++ b/test/CodeGen/Mips/fmadd1.ll
@@ -196,8 +196,8 @@ entry:
 
 ; 32R2:          ldc1 $[[T0:f[0-9]+]], 16($sp)
 ; 32R2:          madd.d $[[T1:f[0-9]+]], $[[T0]], $f12, $f14
-; 32R2:          mtc1 $zero, $[[T2:f[02468]+]]
-; 32R2:          mtc1 $zero, ${{f[13579]+}}
+; 32R2:          mtc1 $zero, $[[T2:f[0-9]+]]
+; 32R2:          mthc1 $zero, $[[T2]]
 ; 32R2:          add.d $f0, $[[T1]], $[[T2]]
 
 ; 32R6-DAG:      ldc1 $[[T0:f[0-9]+]], 16($sp)
@@ -238,8 +238,8 @@ entry:
 
 ; 32R2:          ldc1 $[[T0:f[0-9]+]], 16($sp)
 ; 32R2:          msub.d $[[T1:f[0-9]+]], $[[T0]], $f12, $f14
-; 32R2:          mtc1 $zero, $[[T2:f[02468]+]]
-; 32R2:          mtc1 $zero, ${{f[13579]+}}
+; 32R2:          mtc1 $zero, $[[T2:f[0-9]+]]
+; 32R2:          mthc1 $zero, $[[T2]]
 ; 32R2:          add.d $f0, $[[T1]], $[[T2]]
 
 ; 32R6-DAG:      ldc1 $[[T0:f[0-9]+]], 16($sp)
@@ -283,8 +283,8 @@ entry:
 
 ; 32R2-NAN:      ldc1 $[[T0:f[0-9]+]], 16($sp)
 ; 32R2-NAN:      madd.d $[[T1:f[0-9]+]], $[[T0]], $f12, $f14
-; 32R2-NAN:      mtc1 $zero, $[[T2:f[02468]+]]
-; 32R2-NAN:      mtc1 $zero, ${{f[13579]+}}
+; 32R2-NAN:      mtc1 $zero, $[[T2:f[0-9]+]]
+; 32R2-NAN:      mthc1 $zero, $[[T2]]
 ; 32R2-NAN:      sub.d $f0, $[[T2]], $[[T1]]
 
 ; 32R6-DAG:      ldc1 $[[T0:f[0-9]+]], 16($sp)
@@ -330,8 +330,8 @@ entry:
 
 ; 32R2-NAN:      ldc1 $[[T0:f[0-9]+]], 16($sp)
 ; 32R2-NAN:      msub.d $[[T1:f[0-9]+]], $[[T0]], $f12, $f14
-; 32R2-NAN:      mtc1 $zero, $[[T2:f[02468]+]]
-; 32R2-NAN:      mtc1 $zero, ${{f[13579]+}}
+; 32R2-NAN:      mtc1 $zero, $[[T2:f[0-9]+]]
+; 32R2-NAN:      mthc1 $zero, $[[T2]]
 ; 32R2-NAN:      sub.d $f0, $[[T2]], $[[T1]]
 
 ; 32R6-DAG:      ldc1 $[[T0:f[0-9]+]], 16($sp)
diff --git a/test/CodeGen/Mips/mno-ldc1-sdc1.ll b/test/CodeGen/Mips/mno-ldc1-sdc1.ll
index f4854f8805..ffc977a182 100644
--- a/test/CodeGen/Mips/mno-ldc1-sdc1.ll
+++ b/test/CodeGen/Mips/mno-ldc1-sdc1.ll
@@ -13,7 +13,7 @@
 ; LE-PIC-DAG: lw $[[R0:[0-9]+]], 0(${{[0-9]+}})
 ; LE-PIC-DAG: lw $[[R1:[0-9]+]], 4(${{[0-9]+}})
 ; LE-PIC-DAG: mtc1 $[[R0]], $f0
-; LE-PIC-DAG: mtc1 $[[R1]], $f1
+; LE-PIC-DAG: mthc1 $[[R1]], $f0
 ; LE-STATIC-LABEL: test_ldc1:
 ; LE-STATIC-DAG: lui $[[R0:[0-9]+]], %hi(g0)
 ; LE-STATIC-DAG: lw $[[R1:[0-9]+]], %lo(g0)($[[R0]])
@@ -66,7 +66,7 @@ entry:
 ; LE-PIC-DAG: lw $[[R0:[0-9]+]], 0(${{[0-9]+}})
 ; LE-PIC-DAG: lw $[[R1:[0-9]+]], 4(${{[0-9]+}})
 ; LE-PIC-DAG: mtc1 $[[R0]], $f0
-; LE-PIC-DAG: mtc1 $[[R1]], $f1
+; LE-PIC-DAG: mthc1 $[[R1]], $f0
 ; CHECK-LDC1-SDC1-LABEL: test_ldxc1:
 ; CHECK-LDC1-SDC1: ldxc1 $f{{[0-9]+}}