ARM: Fix RFE macrop.

This changes the RFE macroop into 3 microops:

URa = [sp]; URb = [sp+4]; // load CPSR,PC values from stack
sp = sp + offset;         // optionally auto-increment
PC = URa; CPSR = URb;     // write to the PC and CPSR.

Importantly:
- writing to PC is handled in the last micro-op.
- loading occurs prior to state changes.

2 files changed, 50 insertions, 9 deletions

diff --git a/src/arch/arm/isa/templates/macromem.isa b/src/arch/arm/isa/templates/macromem.isa
index b7ca7fa48..a7f7f0da8 100644
--- a/src/arch/arm/isa/templates/macromem.isa
+++ b/src/arch/arm/isa/templates/macromem.isa
@@ -109,6 +109,41 @@ def template MicroNeonMemDeclare {{
 
 ////////////////////////////////////////////////////////////////////
 //
+// PC   = Integer(ura)
+// CPSR = Integer(urb)
+//
+
+def template MicroSetPCCPSRDeclare {{
+    class %(class_name)s : public %(base_class)s
+    {
+      public:
+        %(class_name)s(ExtMachInst machInst,
+                       IntRegIndex _ura,
+                       IntRegIndex _urb,
+                       IntRegIndex _urc);
+        %(BasicExecDeclare)s
+    };
+}};
+
+def template MicroSetPCCPSRConstructor {{
+    %(class_name)s::%(class_name)s(ExtMachInst machInst,
+                                   IntRegIndex _ura,
+                                   IntRegIndex _urb,
+                                   IntRegIndex _urc)
+          : %(base_class)s("%(mnemonic)s", machInst, %(op_class)s,
+                           _ura, _urb, _urc)
+    {
+        %(constructor)s;
+        if (!(condCode == COND_AL || condCode == COND_UC)) {
+            for (int x = 0; x < _numDestRegs; x++) {
+                _srcRegIdx[_numSrcRegs++] = _destRegIdx[x];
+            }
+        }
+    }
+}};
+
+////////////////////////////////////////////////////////////////////
+//
 // Integer = Integer op Integer microops
 //
 
diff --git a/src/arch/arm/isa/templates/mem.isa b/src/arch/arm/isa/templates/mem.isa
index 3d073b322..dcfd47ace 100644
--- a/src/arch/arm/isa/templates/mem.isa
+++ b/src/arch/arm/isa/templates/mem.isa
@@ -917,9 +917,9 @@ def template CompleteAccDeclare {{
 
 def template RfeConstructor {{
     inline %(class_name)s::%(class_name)s(ExtMachInst machInst,
-            uint32_t _base, int _mode, bool _wb)
-         : %(base_class)s("%(mnemonic)s", machInst, %(op_class)s,
-                 (IntRegIndex)_base, (AddrMode)_mode, _wb)
+                                          uint32_t _base, int _mode, bool _wb)
+        : %(base_class)s("%(mnemonic)s", machInst, %(op_class)s,
+                         (IntRegIndex)_base, (AddrMode)_mode, _wb)
     {
         %(constructor)s;
         if (!(condCode == COND_AL || condCode == COND_UC)) {
@@ -928,12 +928,18 @@ def template RfeConstructor {{
             }
         }
 #if %(use_uops)d
-        assert(numMicroops >= 2);
-        uops = new StaticInstPtr[numMicroops];
-        uops[0] = new %(acc_name)s(machInst, _base, _mode, _wb);
-        uops[0]->setDelayedCommit();
-        uops[1] = new %(wb_decl)s;
-        uops[1]->setLastMicroop();
+        uops = new StaticInstPtr[1 + %(use_wb)d + %(use_pc)d];
+        int uopIdx = 0;
+        uops[uopIdx] = new %(acc_name)s(machInst, _base, _mode, _wb);
+        uops[uopIdx]->setDelayedCommit();
+#if %(use_wb)d
+        uops[++uopIdx] = new %(wb_decl)s;
+        uops[uopIdx]->setDelayedCommit();
+#endif
+#if %(use_pc)d
+        uops[++uopIdx] = new %(pc_decl)s;
+#endif
+        uops[uopIdx]->setLastMicroop();
 #endif
     }
 }};