ARM: Add support load/store multiple VFP registers

src/arm/simulator-arm.cc

 // Copyright 2010 the V8 project authors. All rights reserved.

[Karl Klose, 2011/04/06 06:52:56]

2011

[Søren Thygesen Gjesse, 2011/04/06 08:00:09]

Done.

 // Redistribution and use in source and binary forms, with or without
 // modification, are permitted provided that the following conditions are
 // met:
 //
 //     * Redistributions of source code must retain the above copyright
 //       notice, this list of conditions and the following disclaimer.
 //     * Redistributions in binary form must reproduce the above
 //       copyright notice, this list of conditions and the following
 //       disclaimer in the documentation and/or other materials provided
 //       with the distribution.
@@ skipping 1485 matching lines @@
   while (bit_vector != 0) {
     if ((bit_vector & 1) != 0) {
       count++;
     }
     bit_vector >>= 1;
   }
   return count;
 }
 
 
-// Addressing Mode 4 - Load and Store Multiple
-void Simulator::HandleRList(Instruction* instr, bool load) {
+void Simulator::ProcessPUW(Instruction* instr,
+                           int num_regs,
+                           int reg_size,
+                           intptr_t* start_address,
+                           intptr_t* end_address) {
   int rn = instr->RnValue();
   int32_t rn_val = get_register(rn);
-  int rlist = instr->RlistValue();
-  int num_regs = count_bits(rlist);
-
-  intptr_t start_address = 0;
-  intptr_t end_address = 0;
   switch (instr->PUField()) {
     case da_x: {
       UNIMPLEMENTED();
       break;
     }
     case ia_x: {
-      start_address = rn_val;
-      end_address = rn_val + (num_regs * 4) - 4;
-      rn_val = rn_val + (num_regs * 4);
+      *start_address = rn_val;
+      *end_address = rn_val + (num_regs * reg_size) - reg_size;
+      rn_val = rn_val + (num_regs * reg_size);
       break;
     }
     case db_x: {
-      start_address = rn_val - (num_regs * 4);
-      end_address = rn_val - 4;
-      rn_val = start_address;
+      *start_address = rn_val - (num_regs * reg_size);
+      *end_address = rn_val - reg_size;
+      rn_val = *start_address;
       break;
     }
     case ib_x: {
-      start_address = rn_val + 4;
-      end_address = rn_val + (num_regs * 4);
-      rn_val = end_address;
+      *start_address = rn_val + reg_size;
+      *end_address = rn_val + (num_regs * reg_size);
+      rn_val = *end_address;
       break;
     }
     default: {
       UNREACHABLE();
       break;
     }
   }
   if (instr->HasW()) {
     set_register(rn, rn_val);
   }
+}
+
+// Addressing Mode 4 - Load and Store Multiple
+void Simulator::HandleRList(Instruction* instr, bool load) {
+  int rlist = instr->RlistValue();
+  int num_regs = count_bits(rlist);
+
+  intptr_t start_address = 0;
+  intptr_t end_address = 0;
+  ProcessPUW(instr, num_regs, kPointerSize, &start_address, &end_address);
+
   intptr_t* address = reinterpret_cast<intptr_t*>(start_address);
   int reg = 0;
   while (rlist != 0) {
     if ((rlist & 1) != 0) {
       if (load) {
         set_register(reg, *address);
       } else {
         *address = get_register(reg);
       }
       address += 1;
     }
     reg++;
     rlist >>= 1;
   }
   ASSERT(end_address == ((intptr_t)address) - 4);
 }
 
 
+// Addressing Mode 6 - Load and Store Multiple Coprocessor registers.
+void Simulator::HandleVList(Instruction* instr, bool load) {
+  VFPRegPrecision precision =
+      (instr->SzValue() == 0) ? kSinglePrecision : kDoublePrecision;
+  int operand_size = (precision == kSinglePrecision) ? 4 : 8;
+
+  int vd;
+  int num_regs;
+  if (precision == kSinglePrecision) {
+    vd = instr->VdValue() << 1 | instr->DValue();

[Rodolph Perfetta, 2011/04/05 18:16:39]

Use instr->VFPDregValue(kSinglePrecision) here.

[Søren Thygesen Gjesse, 2011/04/06 08:00:09]

Used

  vd = instr->VFPDRegValue(precision);

before the if.

+    num_regs = instr->Immed8Value();
+  } else {
+    vd = instr->VdValue();
+    num_regs = instr->Immed8Value() / 2;
+  }
+
+  intptr_t start_address = 0;
+  intptr_t end_address = 0;
+  ProcessPUW(instr, num_regs, operand_size, &start_address, &end_address);
+
+  intptr_t* address = reinterpret_cast<intptr_t*>(start_address);
+  for (int reg = vd; reg < vd + num_regs; reg++) {
+    if (precision == kSinglePrecision) {
+      if (load) {
+        set_s_register_from_sinteger(
+            reg, ReadW(reinterpret_cast<int32_t>(address), instr));
+      } else {
+        WriteW(reinterpret_cast<int32_t>(address),
+               get_sinteger_from_s_register(reg), instr);
+      }
+      address += 1;
+    } else {
+      if (load) {
+        set_s_register_from_sinteger(
+            2 * reg, ReadW(reinterpret_cast<int32_t>(address), instr));
+        set_s_register_from_sinteger(
+            2 * reg + 1, ReadW(reinterpret_cast<int32_t>(address + 1), instr));
+      } else {
+        WriteW(reinterpret_cast<int32_t>(address),
+               get_sinteger_from_s_register(2 * reg), instr);
+        WriteW(reinterpret_cast<int32_t>(address + 1),
+               get_sinteger_from_s_register(2 * reg + 1), instr);
+      }
+      address += 2;
+    }
+  }
+  ASSERT_EQ(((intptr_t)address) - operand_size, end_address);
+}
+
+
 // Calls into the V8 runtime are based on this very simple interface.
 // Note: To be able to return two values from some calls the code in runtime.cc
 // uses the ObjectPair which is essentially two 32-bit values stuffed into a
 // 64-bit value. With the code below we assume that all runtime calls return
 // 64 bits of result. If they don't, the r1 result register contains a bogus
 // value, which is fine because it is caller-saved.
 typedef int64_t (*SimulatorRuntimeCall)(int32_t arg0,
                                         int32_t arg1,
                                         int32_t arg2,
                                         int32_t arg3,
@@ skipping 1363 matching lines @@
         int32_t address = get_register(rn) + 4 * offset;
         if (instr->HasL()) {
           // Load double from memory: vldr.
           set_s_register_from_sinteger(vd, ReadW(address, instr));
         } else {
           // Store double to memory: vstr.
           WriteW(address, get_sinteger_from_s_register(vd), instr);
         }
         break;
       }
-      default:
-        UNIMPLEMENTED();  // Not used by V8.
+      default: {
+        BlockAddrMode am =
+            static_cast<BlockAddrMode>(
+                instr->BitField(24, 21) & kBlockAddrModeMask);
+        if (am == ia || am == ia_w || am == db_w) {
+          bool to_arm_register = (instr->VLValue() == 0x1);

[Rodolph Perfetta, 2011/04/05 18:16:39]

Since the instruction encoding is given to HandleVList, is it worth passing the
L bit separately?

[Søren Thygesen Gjesse, 2011/04/06 08:00:09]

No - argument removed.

+          if (to_arm_register) {
+            // Load multiple single from memory: vldm.
+            HandleVList(instr, true);
+          } else {
+            HandleVList(instr, false);
+          }
+        } else {
+          UNIMPLEMENTED();  // Not used by V8.
+        }
         break;
+      }
     }
   } else if (instr->CoprocessorValue() == 0xB) {
     switch (instr->OpcodeValue()) {
       case 0x2:
         // Load and store double to two GP registers
         if (instr->Bits(7, 4) != 0x1) {
           UNIMPLEMENTED();  // Not used by V8.
         } else {
           int rt = instr->RtValue();
           int rn = instr->RnValue();
@@ skipping 26 matching lines @@
           // Load double from memory: vldr.
           set_s_register_from_sinteger(2*vd, ReadW(address, instr));
           set_s_register_from_sinteger(2*vd + 1, ReadW(address + 4, instr));
         } else {
           // Store double to memory: vstr.
           WriteW(address, get_sinteger_from_s_register(2*vd), instr);
           WriteW(address + 4, get_sinteger_from_s_register(2*vd + 1), instr);
         }
         break;
       }
-      default:
-        UNIMPLEMENTED();  // Not used by V8.
+      default: {
+        BlockAddrMode am =
+            static_cast<BlockAddrMode>(
+                instr->BitField(24, 21) & kBlockAddrModeMask);
+        if (am == ia || am == ia_w || am == db_w) {
+          bool to_arm_register = (instr->VLValue() == 0x1);

[Rodolph Perfetta, 2011/04/05 18:16:39]

Ditto.

[Søren Thygesen Gjesse, 2011/04/06 08:00:09]

Done.

+          if (to_arm_register) {
+            // Load multiple double from memory: vldm.
+            HandleVList(instr, true);
+          } else {
+            HandleVList(instr, false);
+          }
+        } else {
+          UNIMPLEMENTED();  // Not used by V8.
+        }
         break;
+      }
     }
   } else {
     UNIMPLEMENTED();  // Not used by V8.
   }
 }
 
 
 // Executes the current instruction.
 void Simulator::InstructionDecode(Instruction* instr) {
   if (v8::internal::FLAG_check_icache) {
@@ skipping 196 matching lines @@
   uintptr_t address = *stack_slot;
   set_register(sp, current_sp + sizeof(uintptr_t));
   return address;
 }
 
 } }  // namespace v8::internal
 
 #endif  // USE_SIMULATOR
 
 #endif  // V8_TARGET_ARCH_ARM