Update machine models. Specify buffer sizes for OOO processors.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@184033 91177308-0d34-0410-b5e6-96231b3b80d8

lib/Target/ARM/ARMScheduleA9.td

@@ -1883,7 +1883,7 @@ def CortexA9Itineraries : ProcessorItineraries<
 // Cortex-A9 machine model for scheduling and other instruction cost heuristics.
 def CortexA9Model : SchedMachineModel {
   let IssueWidth = 2; // 2 micro-ops are dispatched per cycle.
-  let MinLatency = 0; // Data dependencies are allowed within dispatch groups.
+  let MicroOpBufferSize = 56; // Based on available renamed registers.
   let LoadLatency = 2; // Optimistic load latency assuming bypass.
                        // This is overriden by OperandCycles if the
                        // Itineraries are queried instead.
@@ -1901,7 +1901,7 @@ def A9UnitALU : ProcResource<2>;
 def A9UnitMul : ProcResource<1> { let Super = A9UnitALU; }
 def A9UnitAGU : ProcResource<1>;
 def A9UnitLS  : ProcResource<1>;
-def A9UnitFP  : ProcResource<1>;
+def A9UnitFP  : ProcResource<1> { let BufferSize = 0; }
 def A9UnitB   : ProcResource<1>;
 
 //===----------------------------------------------------------------------===//

lib/Target/ARM/ARMScheduleSwift.td

@@ -1076,7 +1076,7 @@ def SwiftItineraries : ProcessorItineraries<
 // Swift machine model for scheduling and other instruction cost heuristics.
 def SwiftModel : SchedMachineModel {
   let IssueWidth = 3; // 3 micro-ops are dispatched per cycle.
-  let MinLatency = 0; // Data dependencies are allowed within dispatch groups.
+  let MicroOpBufferSize = 45; // Based on NEON renamed registers.
   let LoadLatency = 3;
   let MispredictPenalty = 14; // A branch direction mispredict.
 

lib/Target/X86/X86SchedHaswell.td

@@ -16,7 +16,7 @@ def HaswellModel : SchedMachineModel {
   // All x86 instructions are modeled as a single micro-op, and HW can decode 4
   // instructions per cycle.
   let IssueWidth = 4;
-  let MinLatency = 0; // 0 = Out-of-order execution.
+  let MicroOpBufferSize = 192; // Based on the reorder buffer.
   let LoadLatency = 4;
   let MispredictPenalty = 16;
 }

lib/Target/X86/X86SchedSandyBridge.td

@@ -17,7 +17,7 @@ def SandyBridgeModel : SchedMachineModel {
   // instructions per cycle.
   // FIXME: Identify instructions that aren't a single fused micro-op.
   let IssueWidth = 4;
-  let MinLatency = 0; // 0 = Out-of-order execution.
+  let MicroOpBufferSize = 168; // Based on the reorder buffer.
   let LoadLatency = 4;
   let MispredictPenalty = 16;
 }

lib/Target/X86/X86Schedule.td

@@ -550,8 +550,9 @@ def IIC_NOP : InstrItinClass;
 // Resources beyond the decoder operate on micro-ops and are bufferred
 // so adjacent micro-ops don't directly compete.
 //
-// MinLatency=0 indicates that RAW dependencies can be decoded in the
-// same cycle.
+// MicroOpBufferSize > 1 indicates that RAW dependencies can be
+// decoded in the same cycle. The value 32 is a reasonably arbitrary
+// number of in-flight instructions.
 //
 // HighLatency=10 is optimistic. X86InstrInfo::isHighLatencyDef
 // indicates high latency opcodes. Alternatively, InstrItinData
@@ -562,7 +563,7 @@ def IIC_NOP : InstrItinClass;
 // The GenericModel contains no instruciton itineraries.
 def GenericModel : SchedMachineModel {
   let IssueWidth = 4;
-  let MinLatency = 0;
+  let MicroOpBufferSize = 32;
   let LoadLatency = 4;
   let HighLatency = 10;
 }

lib/Target/X86/X86ScheduleAtom.td

@@ -525,8 +525,7 @@ def AtomItineraries : ProcessorItineraries<
 // Atom machine model.
 def AtomModel : SchedMachineModel {
   let IssueWidth = 2;  // Allows 2 instructions per scheduling group.
-  let MinLatency = 1;  // InstrStage cycles overrides MinLatency.
-                       // OperandCycles may be used for expected latency.
+  let MicroOpBufferSize = 0; // In-order execution, always hide latency.
   let LoadLatency = 3; // Expected cycles, may be overriden by OperandCycles.
   let HighLatency = 30;// Expected, may be overriden by OperandCycles.