Making use of VFP / NEON floating point multiply-accumulate / subtraction is
difficult on current ARM implementations for a few reasons.
1. Even though a single vmla has latency that is one cycle shorter than a pair
   of vmul + vadd, a RAW hazard during the first (4? on Cortex-a8) can cause
   additional pipeline stall. So it's frequently better to single codegen
   vmul + vadd.
2. A vmla folowed by a vmul, vmadd, or vsub causes the second fp instruction to
   stall for 4 cycles. We need to schedule them apart.
3. A vmla followed vmla is a special case. Obvious issuing back to back RAW
   vmla + vmla is very bad. But this isn't ideal either:
     vmul
     vadd
     vmla
   Instead, we want to expand the second vmla:
     vmla
     vmul
     vadd
   Even with the 4 cycle vmul stall, the second sequence is still 2 cycles
   faster.

Up to now, isel simply avoid codegen'ing fp vmla / vmls. This works well enough
but it isn't the optimial solution. This patch attempts to make it possible to
use vmla / vmls in cases where it is profitable.

A. Add missing isel predicates which cause vmla to be codegen'ed.
B. Make sure the fmul in (fadd (fmul)) has a single use. We don't want to
   compute a fmul and a fmla.
C. Add additional isel checks for vmla, avoid cases where vmla is feeding into
   fp instructions (except for the #3 exceptional case).
D. Add ARM hazard recognizer to model the vmla / vmls hazards.
E. Add a special pre-regalloc case to expand vmla / vmls when it's likely the
   vmla / vmls will trigger one of the special hazards.

Work in progress, only A+B are enabled.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@120960 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/lib/Target/ARM/ARMHazardRecognizer.cpp b/lib/Target/ARM/ARMHazardRecognizer.cpp
new file mode 100644
index 0000000..317934f
--- /dev/null
+++ b/lib/Target/ARM/ARMHazardRecognizer.cpp
@@ -0,0 +1,114 @@
+//===-- ARMHazardRecognizer.cpp - ARM postra hazard recognizer ------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "ARMHazardRecognizer.h"
+#include "ARMBaseInstrInfo.h"
+#include "ARMSubtarget.h"
+#include "llvm/CodeGen/MachineInstr.h"
+#include "llvm/CodeGen/ScheduleDAG.h"
+#include "llvm/Target/TargetRegisterInfo.h"
+using namespace llvm;
+
+static bool hasRAWHazard(MachineInstr *DefMI, MachineInstr *MI,
+                         const TargetRegisterInfo &TRI) {
+  // FIXME: Detect integer instructions properly.
+  const TargetInstrDesc &TID = MI->getDesc();
+  unsigned Domain = TID.TSFlags & ARMII::DomainMask;
+  if (Domain == ARMII::DomainVFP) {
+    unsigned Opcode = MI->getOpcode();
+    if (Opcode == ARM::VSTRS || Opcode == ARM::VSTRD ||
+        Opcode == ARM::VMOVRS || Opcode == ARM::VMOVRRD)
+      return false;
+  } else if (Domain == ARMII::DomainNEON) {
+    if (MI->getDesc().mayStore() || MI->getDesc().mayLoad())
+      return false;
+  } else
+    return false;
+  return MI->readsRegister(DefMI->getOperand(0).getReg(), &TRI);
+}
+
+ScheduleHazardRecognizer::HazardType
+ARMHazardRecognizer::getHazardType(SUnit *SU) {
+  MachineInstr *MI = SU->getInstr();
+
+  if (!MI->isDebugValue()) {
+    if (ITBlockSize && MI != ITBlockMIs[ITBlockSize-1])
+      return Hazard;
+
+    // Look for special VMLA / VMLS hazards. A VMUL / VADD / VSUB following
+    // a VMLA / VMLS will cause 4 cycle stall.
+    const TargetInstrDesc &TID = MI->getDesc();
+    if (LastMI && (TID.TSFlags & ARMII::DomainMask) != ARMII::DomainGeneral) {
+      MachineInstr *DefMI = LastMI;
+      const TargetInstrDesc &LastTID = LastMI->getDesc();
+      // Skip over one non-VFP / NEON instruction.
+      if (!LastTID.isBarrier() &&
+          (LastTID.TSFlags & ARMII::DomainMask) == ARMII::DomainGeneral) {
+        MachineBasicBlock::iterator I = LastMI;
+        if (I != LastMI->getParent()->begin()) {
+          I = llvm::prior(I);
+          DefMI = &*I;
+        }
+      }
+
+      if (TII.isFpMLxInstruction(DefMI->getOpcode()) &&
+          (TII.canCauseFpMLxStall(MI->getOpcode()) ||
+           hasRAWHazard(DefMI, MI, TRI))) {
+        // Try to schedule another instruction for the next 4 cycles.
+        if (Stalls == 0)
+          Stalls = 4;
+        return Hazard;
+      }
+    }
+  }
+
+  return PostRAHazardRecognizer::getHazardType(SU);
+}
+
+void ARMHazardRecognizer::Reset() {
+  LastMI = 0;
+  Stalls = 0;
+  ITBlockSize = 0;
+  PostRAHazardRecognizer::Reset();
+}
+
+void ARMHazardRecognizer::EmitInstruction(SUnit *SU) {
+  MachineInstr *MI = SU->getInstr();
+  unsigned Opcode = MI->getOpcode();
+  if (ITBlockSize) {
+    --ITBlockSize;
+  } else if (Opcode == ARM::t2IT) {
+    unsigned Mask = MI->getOperand(1).getImm();
+    unsigned NumTZ = CountTrailingZeros_32(Mask);
+    assert(NumTZ <= 3 && "Invalid IT mask!");
+    ITBlockSize = 4 - NumTZ;
+    MachineBasicBlock::iterator I = MI;
+    for (unsigned i = 0; i < ITBlockSize; ++i) {
+      // Advance to the next instruction, skipping any dbg_value instructions.
+      do {
+        ++I;
+      } while (I->isDebugValue());
+      ITBlockMIs[ITBlockSize-1-i] = &*I;
+    }
+  }
+
+  if (!MI->isDebugValue()) {
+    LastMI = MI;
+    Stalls = 0;
+  }
+
+  PostRAHazardRecognizer::EmitInstruction(SU);
+}
+
+void ARMHazardRecognizer::AdvanceCycle() {
+  if (Stalls && --Stalls == 0)
+    // Stalled for 4 cycles but still can't schedule any other instructions.
+    LastMI = 0;
+  PostRAHazardRecognizer::AdvanceCycle();
+}