Reapply "Make NumMicroOps a variable in the subtarget's instruction itinerary."

Reapplies r159406 with minor cleanup. The regressions appear to have been spurious.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@159541 91177308-0d34-0410-b5e6-96231b3b80d8

4 files changed, 64 insertions, 48 deletions

diff --git a/lib/Target/ARM/ARMBaseInstrInfo.cpp b/lib/Target/ARM/ARMBaseInstrInfo.cpp
index 464c9aad3c..7227e62c7c 100644
--- a/lib/Target/ARM/ARMBaseInstrInfo.cpp
+++ b/lib/Target/ARM/ARMBaseInstrInfo.cpp
@@ -2176,9 +2176,9 @@ ARMBaseInstrInfo::getNumMicroOps(const InstrItineraryData *ItinData,
 
   const MCInstrDesc &Desc = MI->getDesc();
   unsigned Class = Desc.getSchedClass();
-  unsigned UOps = ItinData->Itineraries[Class].NumMicroOps;
-  if (UOps)
-    return UOps;
+  int ItinUOps = ItinData->getNumMicroOps(Class);
+  if (ItinUOps >= 0)
+    return ItinUOps;
 
   unsigned Opc = MI->getOpcode();
   switch (Opc) {
@@ -2252,19 +2252,19 @@ ARMBaseInstrInfo::getNumMicroOps(const InstrItineraryData *ItinData,
         return 2;
       // 4 registers would be issued: 2, 2.
       // 5 registers would be issued: 2, 2, 1.
-      UOps = (NumRegs / 2);
+      int A8UOps = (NumRegs / 2);
       if (NumRegs % 2)
-        ++UOps;
-      return UOps;
+        ++A8UOps;
+      return A8UOps;
     } else if (Subtarget.isCortexA9()) {
-      UOps = (NumRegs / 2);
+      int A9UOps = (NumRegs / 2);
       // If there are odd number of registers or if it's not 64-bit aligned,
       // then it takes an extra AGU (Address Generation Unit) cycle.
       if ((NumRegs % 2) ||
           !MI->hasOneMemOperand() ||
           (*MI->memoperands_begin())->getAlignment() < 8)
-        ++UOps;
-      return UOps;
+        ++A9UOps;
+      return A9UOps;
     } else {
       // Assume the worst.
       return NumRegs;
@@ -3057,7 +3057,8 @@ unsigned ARMBaseInstrInfo::getInstrLatency(const InstrItineraryData *ItinData,
   unsigned Class = MCID.getSchedClass();
 
   // For instructions with variable uops, use uops as latency.
-  if (!ItinData->isEmpty() && !ItinData->Itineraries[Class].NumMicroOps) {
+  if (!ItinData->isEmpty() && ItinData->getNumMicroOps(Class) < 0) {
+    dbgs() << "UOPS " << getNumMicroOps(ItinData, MI) << " " << *MI << '\n';
     return getNumMicroOps(ItinData, MI);
   }
   // For the common case, fall back on the itinerary's latency.
diff --git a/lib/Target/ARM/ARMSchedule.td b/lib/Target/ARM/ARMSchedule.td
index b9a07f1ee6..81d2fa37c2 100644
--- a/lib/Target/ARM/ARMSchedule.td
+++ b/lib/Target/ARM/ARMSchedule.td
@@ -70,11 +70,11 @@ def IIC_iLoad_bh_siu : InstrItinClass;
 def IIC_iLoad_d_i  : InstrItinClass;
 def IIC_iLoad_d_r  : InstrItinClass;
 def IIC_iLoad_d_ru : InstrItinClass;
-def IIC_iLoad_m    : InstrItinClass<0>;  // micro-coded
-def IIC_iLoad_mu   : InstrItinClass<0>;  // micro-coded
-def IIC_iLoad_mBr  : InstrItinClass<0>;  // micro-coded
-def IIC_iPop       : InstrItinClass<0>;  // micro-coded
-def IIC_iPop_Br    : InstrItinClass<0>;  // micro-coded
+def IIC_iLoad_m    : InstrItinClass;
+def IIC_iLoad_mu   : InstrItinClass;
+def IIC_iLoad_mBr  : InstrItinClass;
+def IIC_iPop       : InstrItinClass;
+def IIC_iPop_Br    : InstrItinClass;
 def IIC_iLoadiALU  : InstrItinClass;
 def IIC_iStore_i   : InstrItinClass;
 def IIC_iStore_r   : InstrItinClass;
@@ -91,8 +91,8 @@ def IIC_iStore_bh_siu : InstrItinClass;
 def IIC_iStore_d_i   : InstrItinClass;
 def IIC_iStore_d_r   : InstrItinClass;
 def IIC_iStore_d_ru  : InstrItinClass;
-def IIC_iStore_m   : InstrItinClass<0>;  // micro-coded
-def IIC_iStore_mu  : InstrItinClass<0>;  // micro-coded
+def IIC_iStore_m   : InstrItinClass;
+def IIC_iStore_mu  : InstrItinClass;
 def IIC_Preload    : InstrItinClass;
 def IIC_Br         : InstrItinClass;
 def IIC_fpSTAT     : InstrItinClass;
@@ -126,12 +126,12 @@ def IIC_fpSQRT32   : InstrItinClass;
 def IIC_fpSQRT64   : InstrItinClass;
 def IIC_fpLoad32   : InstrItinClass;
 def IIC_fpLoad64   : InstrItinClass;
-def IIC_fpLoad_m   : InstrItinClass<0>;  // micro-coded
-def IIC_fpLoad_mu  : InstrItinClass<0>;  // micro-coded
+def IIC_fpLoad_m   : InstrItinClass;
+def IIC_fpLoad_mu  : InstrItinClass;
 def IIC_fpStore32  : InstrItinClass;
 def IIC_fpStore64  : InstrItinClass;
-def IIC_fpStore_m  : InstrItinClass<0>;  // micro-coded
-def IIC_fpStore_mu : InstrItinClass<0>;  // micro-coded
+def IIC_fpStore_m  : InstrItinClass;
+def IIC_fpStore_mu : InstrItinClass;
 def IIC_VLD1       : InstrItinClass;
 def IIC_VLD1x2     : InstrItinClass;
 def IIC_VLD1x3     : InstrItinClass;
diff --git a/lib/Target/ARM/ARMScheduleA8.td b/lib/Target/ARM/ARMScheduleA8.td
index eb1083ca23..61de00a208 100644
--- a/lib/Target/ARM/ARMScheduleA8.td
+++ b/lib/Target/ARM/ARMScheduleA8.td
@@ -155,28 +155,30 @@ def CortexA8Itineraries : MultiIssueItineraries<
   // Load multiple, def is the 5th operand. Pipeline 0 only.
   // FIXME: A8_LSPipe cycle time is dynamic, this assumes 3 to 4 registers.
   InstrItinData<IIC_iLoad_m  , [InstrStage<2, [A8_Pipe0], 0>,
-                                InstrStage<2, [A8_LSPipe]>], [1, 1, 1, 1, 3]>,
+                                InstrStage<2, [A8_LSPipe]>],
+                [1, 1, 1, 1, 3], [], -1>, // dynamic uops
   //
   // Load multiple + update, defs are the 1st and 5th operands.
   InstrItinData<IIC_iLoad_mu , [InstrStage<3, [A8_Pipe0], 0>,
-                                InstrStage<3, [A8_LSPipe]>], [2, 1, 1, 1, 3]>,
+                                InstrStage<3, [A8_LSPipe]>],
+                [2, 1, 1, 1, 3], [], -1>, // dynamic uops
   //
   // Load multiple plus branch
   InstrItinData<IIC_iLoad_mBr, [InstrStage<3, [A8_Pipe0], 0>,
                                 InstrStage<3, [A8_LSPipe]>,
                                 InstrStage<1, [A8_Pipe0, A8_Pipe1]>],
-                               [1, 2, 1, 1, 3]>,
+                              [1, 2, 1, 1, 3], [], -1>, // dynamic uops
   //
   // Pop, def is the 3rd operand.
   InstrItinData<IIC_iPop  ,    [InstrStage<3, [A8_Pipe0], 0>,
-                                InstrStage<3, [A8_LSPipe]>], [1, 1, 3]>,
+                                InstrStage<3, [A8_LSPipe]>],
+                [1, 1, 3], [], -1>, // dynamic uops
   //
   // Push, def is the 3th operand.
   InstrItinData<IIC_iPop_Br,   [InstrStage<3, [A8_Pipe0], 0>,
                                 InstrStage<3, [A8_LSPipe]>,
                                 InstrStage<1, [A8_Pipe0, A8_Pipe1]>],
-                               [1, 1, 3]>,
-
+                               [1, 1, 3], [], -1>, // dynamic uops
   //
   // iLoadi + iALUr for t2LDRpci_pic.
   InstrItinData<IIC_iLoadiALU, [InstrStage<1, [A8_Pipe0, A8_Pipe1], 0>,
@@ -231,12 +233,13 @@ def CortexA8Itineraries : MultiIssueItineraries<
   // Store multiple. Pipeline 0 only.
   // FIXME: A8_LSPipe cycle time is dynamic, this assumes 3 to 4 registers.
   InstrItinData<IIC_iStore_m , [InstrStage<2, [A8_Pipe0], 0>,
-                                InstrStage<2, [A8_LSPipe]>]>,
+                                InstrStage<2, [A8_LSPipe]>],
+                [], [], -1>, // dynamic uops
   //
   // Store multiple + update
   InstrItinData<IIC_iStore_mu, [InstrStage<2, [A8_Pipe0], 0>,
-                                InstrStage<2, [A8_LSPipe]>], [2]>,
-
+                                InstrStage<2, [A8_LSPipe]>],
+                [2], [], -1>, // dynamic uops
   //
   // Preload
   InstrItinData<IIC_Preload, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2]>,
@@ -397,14 +400,16 @@ def CortexA8Itineraries : MultiIssueItineraries<
                                InstrStage<1, [A8_NLSPipe], 0>,
                                InstrStage<1, [A8_LSPipe]>,
                                InstrStage<1, [A8_NLSPipe], 0>,
-                               InstrStage<1, [A8_LSPipe]>], [1, 1, 1, 2]>,
+                               InstrStage<1, [A8_LSPipe]>],
+                [1, 1, 1, 2], [], -1>, // dynamic uops
   //
   // FP Load Multiple + update
   InstrItinData<IIC_fpLoad_mu,[InstrStage<1, [A8_Pipe0, A8_Pipe1], 0>,
                                InstrStage<1, [A8_NLSPipe], 0>,
                                InstrStage<1, [A8_LSPipe]>,
                                InstrStage<1, [A8_NLSPipe], 0>,
-                               InstrStage<1, [A8_LSPipe]>], [2, 1, 1, 1, 2]>,
+                               InstrStage<1, [A8_LSPipe]>],
+                [2, 1, 1, 1, 2], [], -1>, // dynamic uops
   //
   // Single-precision FP Store
   InstrItinData<IIC_fpStore32,[InstrStage<1, [A8_Pipe0, A8_Pipe1], 0>,
@@ -423,15 +428,16 @@ def CortexA8Itineraries : MultiIssueItineraries<
                                InstrStage<1, [A8_NLSPipe], 0>,
                                InstrStage<1, [A8_LSPipe]>,
                                InstrStage<1, [A8_NLSPipe], 0>,
-                               InstrStage<1, [A8_LSPipe]>], [1, 1, 1, 1]>,
+                               InstrStage<1, [A8_LSPipe]>],
+                [1, 1, 1, 1], [], -1>, // dynamic uops
   //
   // FP Store Multiple + update
   InstrItinData<IIC_fpStore_mu,[InstrStage<1, [A8_Pipe0, A8_Pipe1], 0>,
                                 InstrStage<1, [A8_NLSPipe], 0>,
                                 InstrStage<1, [A8_LSPipe]>,
                                 InstrStage<1, [A8_NLSPipe], 0>,
-                                InstrStage<1, [A8_LSPipe]>], [2, 1, 1, 1, 1]>,
-
+                                InstrStage<1, [A8_LSPipe]>],
+                [2, 1, 1, 1, 1], [], -1>, // dynamic uops
   // NEON
   // Issue through integer pipeline, and execute in NEON unit.
   //
diff --git a/lib/Target/ARM/ARMScheduleA9.td b/lib/Target/ARM/ARMScheduleA9.td
index a00577bf3d..1677ba6a98 100644
--- a/lib/Target/ARM/ARMScheduleA9.td
+++ b/lib/Target/ARM/ARMScheduleA9.td
@@ -284,7 +284,8 @@ def CortexA9Itineraries : MultiIssueItineraries<
                                 InstrStage<2, [A9_AGU], 1>,
                                 InstrStage<2, [A9_LSUnit]>],
                                [1, 1, 1, 1, 3],
-                         [NoBypass, NoBypass, NoBypass, NoBypass, A9_LdBypass]>,
+                         [NoBypass, NoBypass, NoBypass, NoBypass, A9_LdBypass],
+                         -1>, // dynamic uops
   //
   // Load multiple + update, defs are the 1st and 5th operands.
   InstrItinData<IIC_iLoad_mu , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
@@ -292,7 +293,8 @@ def CortexA9Itineraries : MultiIssueItineraries<
                                 InstrStage<2, [A9_AGU], 1>,
                                 InstrStage<2, [A9_LSUnit]>],
                                [2, 1, 1, 1, 3],
-                         [NoBypass, NoBypass, NoBypass, NoBypass, A9_LdBypass]>,
+                         [NoBypass, NoBypass, NoBypass, NoBypass, A9_LdBypass],
+                         -1>, // dynamic uops
   //
   // Load multiple plus branch
   InstrItinData<IIC_iLoad_mBr, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
@@ -301,7 +303,8 @@ def CortexA9Itineraries : MultiIssueItineraries<
                                 InstrStage<2, [A9_LSUnit]>,
                                 InstrStage<1, [A9_Branch]>],
                                [1, 2, 1, 1, 3],
-                         [NoBypass, NoBypass, NoBypass, NoBypass, A9_LdBypass]>,
+                         [NoBypass, NoBypass, NoBypass, NoBypass, A9_LdBypass],
+                         -1>, // dynamic uops
   //
   // Pop, def is the 3rd operand.
   InstrItinData<IIC_iPop  ,    [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
@@ -309,7 +312,8 @@ def CortexA9Itineraries : MultiIssueItineraries<
                                 InstrStage<2, [A9_AGU], 1>,
                                 InstrStage<2, [A9_LSUnit]>],
                                [1, 1, 3],
-                               [NoBypass, NoBypass, A9_LdBypass]>,
+                               [NoBypass, NoBypass, A9_LdBypass],
+                               -1>, // dynamic uops
   //
   // Pop + branch, def is the 3rd operand.
   InstrItinData<IIC_iPop_Br,   [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
@@ -318,8 +322,8 @@ def CortexA9Itineraries : MultiIssueItineraries<
                                 InstrStage<2, [A9_LSUnit]>,
                                 InstrStage<1, [A9_Branch]>],
                                [1, 1, 3],
-                               [NoBypass, NoBypass, A9_LdBypass]>,
-
+                               [NoBypass, NoBypass, A9_LdBypass],
+                               -1>, // dynamic uops
   //
   // iLoadi + iALUr for t2LDRpci_pic.
   InstrItinData<IIC_iLoadiALU, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
@@ -413,14 +417,15 @@ def CortexA9Itineraries : MultiIssueItineraries<
   InstrItinData<IIC_iStore_m , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
                                 InstrStage<1, [A9_MUX0], 0>,
                                 InstrStage<1, [A9_AGU], 0>,
-                                InstrStage<2, [A9_LSUnit]>]>,
+                                InstrStage<2, [A9_LSUnit]>],
+                [], [], -1>, // dynamic uops
   //
   // Store multiple + update
   InstrItinData<IIC_iStore_mu, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
                                 InstrStage<1, [A9_MUX0], 0>,
                                 InstrStage<1, [A9_AGU], 0>,
-                                InstrStage<2, [A9_LSUnit]>], [2]>,
-
+                                InstrStage<2, [A9_LSUnit]>],
+                [2], [], -1>, // dynamic uops
   //
   // Preload
   InstrItinData<IIC_Preload,   [InstrStage<1, [A9_Issue0, A9_Issue1]>], [1, 1]>,
@@ -717,7 +722,8 @@ def CortexA9Itineraries : MultiIssueItineraries<
                                InstrStage<1, [A9_DRegsVFP], 0, Required>,
                                InstrStage<2, [A9_DRegsN],   0, Reserved>,
                                InstrStage<1, [A9_NPipe], 0>,
-                               InstrStage<2, [A9_LSUnit]>], [1, 1, 1, 1]>,
+                               InstrStage<2, [A9_LSUnit]>],
+                [1, 1, 1, 1], [], -1>, // dynamic uops
   //
   // FP Load Multiple + update
   // FIXME: assumes 2 doubles which requires 2 LS cycles.
@@ -726,7 +732,8 @@ def CortexA9Itineraries : MultiIssueItineraries<
                                InstrStage<1, [A9_DRegsVFP], 0, Required>,
                                InstrStage<2, [A9_DRegsN],   0, Reserved>,
                                InstrStage<1, [A9_NPipe], 0>,
-                               InstrStage<2, [A9_LSUnit]>], [2, 1, 1, 1]>,
+                               InstrStage<2, [A9_LSUnit]>],
+                [2, 1, 1, 1], [], -1>, // dynamic uops
   //
   // Single-precision FP Store
   InstrItinData<IIC_fpStore32,[InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
@@ -753,7 +760,8 @@ def CortexA9Itineraries : MultiIssueItineraries<
                                InstrStage<1, [A9_DRegsVFP], 0, Required>,
                                InstrStage<2, [A9_DRegsN],   0, Reserved>,
                                InstrStage<1, [A9_NPipe], 0>,
-                               InstrStage<2, [A9_LSUnit]>], [1, 1, 1, 1]>,
+                               InstrStage<2, [A9_LSUnit]>],
+                [1, 1, 1, 1], [], -1>, // dynamic uops
   //
   // FP Store Multiple + update
   // FIXME: assumes 2 doubles which requires 2 LS cycles.
@@ -762,7 +770,8 @@ def CortexA9Itineraries : MultiIssueItineraries<
                                 InstrStage<1, [A9_DRegsVFP], 0, Required>,
                                 InstrStage<2, [A9_DRegsN],   0, Reserved>,
                                 InstrStage<1, [A9_NPipe], 0>,
-                                InstrStage<2, [A9_LSUnit]>], [2, 1, 1, 1]>,
+                                InstrStage<2, [A9_LSUnit]>],
+                [2, 1, 1, 1], [], -1>, // dynamic uops
   // NEON
   // VLD1
   InstrItinData<IIC_VLD1,     [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,