JitCache: Get physical addresses from PPCAnalyst.

So we support all kind of degenerated blocks now, not just range+length based ones.

Source/Core/Core/PowerPC/CachedInterpreter/CachedInterpreter.cpp

@@ -211,7 +211,7 @@ void CachedInterpreter::Jit(u32 address)
   b->codeSize = (u32)(GetCodePtr() - b->checkedEntry);
   b->originalSize = code_block.m_num_instructions;
 
-  m_block_cache.FinalizeBlock(*b, jo.enableBlocklink, b->checkedEntry);
+  m_block_cache.FinalizeBlock(*b, jo.enableBlocklink, code_block.m_physical_addresses);
 }
 
 void CachedInterpreter::ClearCache()

Source/Core/Core/PowerPC/Jit64/Jit.cpp

@@ -590,7 +590,8 @@ void Jit64::Jit(u32 em_address)
   }
 
   JitBlock* b = blocks.AllocateBlock(em_address);
-  blocks.FinalizeBlock(*b, jo.enableBlocklink, DoJit(em_address, &code_buffer, b, nextPC));
+  DoJit(em_address, &code_buffer, b, nextPC);
+  blocks.FinalizeBlock(*b, jo.enableBlocklink, code_block.m_physical_addresses);
 }
 
 const u8* Jit64::DoJit(u32 em_address, PPCAnalyst::CodeBuffer* code_buf, JitBlock* b, u32 nextPC)

Source/Core/Core/PowerPC/Jit64IL/JitIL.cpp

@@ -508,7 +508,8 @@ void JitIL::Jit(u32 em_address)
   }
 
   JitBlock* b = blocks.AllocateBlock(em_address);
-  blocks.FinalizeBlock(*b, jo.enableBlocklink, DoJit(em_address, &code_buffer, b, nextPC));
+  DoJit(em_address, &code_buffer, b, nextPC);
+  blocks.FinalizeBlock(*b, jo.enableBlocklink, code_block.m_physical_addresses);
 }
 
 const u8* JitIL::DoJit(u32 em_address, PPCAnalyst::CodeBuffer* code_buf, JitBlock* b, u32 nextPC)

Source/Core/Core/PowerPC/JitArm64/Jit.cpp

@@ -399,7 +399,7 @@ void JitArm64::Jit(u32)
 
   JitBlock* b = blocks.AllocateBlock(em_address);
   const u8* BlockPtr = DoJit(em_address, &code_buffer, b, nextPC);
-  blocks.FinalizeBlock(*b, jo.enableBlocklink, BlockPtr);
+  blocks.FinalizeBlock(*b, jo.enableBlocklink, code_block.m_physical_addresses);
 }
 
 const u8* JitArm64::DoJit(u32 em_address, PPCAnalyst::CodeBuffer* code_buf, JitBlock* b, u32 nextPC)

Source/Core/Core/PowerPC/JitCommon/JitCache.cpp

@@ -38,11 +38,7 @@ static void ClearCacheThreadSafe(u64 userdata, s64 cyclesdata)
 
 bool JitBlock::Overlap(u32 addr, u32 length)
 {
-  if (addr >= physicalAddress + originalSize)
-    return false;
-  if (physicalAddress >= addr + length)
-    return false;
-  return true;
+  return physical_addresses.lower_bound(addr) != physical_addresses.lower_bound(addr + length);
 }
 
 JitBaseBlockCache::JitBaseBlockCache(JitBase& jit) : m_jit{jit}
@@ -120,21 +116,21 @@ JitBlock* JitBaseBlockCache::AllocateBlock(u32 em_address)
   return &b;
 }
 
-void JitBaseBlockCache::FinalizeBlock(JitBlock& block, bool block_link, const u8* code_ptr)
+void JitBaseBlockCache::FinalizeBlock(JitBlock& block, bool block_link,
+                                      const std::set<u32>& physical_addresses)
 {
   size_t index = FastLookupIndexForAddress(block.effectiveAddress);
   fast_block_map[index] = &block;
   block.fast_block_map_index = index;
 
-  u32 block_start = block.physicalAddress;
-  u32 block_end = block_start + (block.originalSize - 1) * 4;
+  block.physical_addresses = physical_addresses;
 
-  for (u32 addr = block_start / 32; addr <= block_end / 32; ++addr)
-    valid_block.Set(addr);
-
-  u32 mask = ~(BLOCK_RANGE_MAP_ELEMENTS - 1);
-  for (u32 addr = block_start & mask; addr <= (block_end & mask); addr += BLOCK_RANGE_MAP_ELEMENTS)
-    block_range_map[addr].insert(&block);
+  u32 range_mask = ~(BLOCK_RANGE_MAP_ELEMENTS - 1);
+  for (u32 addr : physical_addresses)
+  {
+    valid_block.Set(addr / 32);
+    block_range_map[addr & range_mask].insert(&block);
+  }
 
   if (block_link)
   {
@@ -207,8 +203,8 @@ void JitBaseBlockCache::InvalidateICache(u32 address, const u32 length, bool for
   if (destroy_block)
   {
     // Iterate over all macro blocks which overlap the given range.
-    u32 mask = ~(BLOCK_RANGE_MAP_ELEMENTS - 1);
-    auto start = block_range_map.lower_bound(pAddr & mask);
+    u32 range_mask = ~(BLOCK_RANGE_MAP_ELEMENTS - 1);
+    auto start = block_range_map.lower_bound(pAddr & range_mask);
     auto end = block_range_map.lower_bound(pAddr + length);
     while (start != end)
     {
@@ -221,11 +217,9 @@ void JitBaseBlockCache::InvalidateICache(u32 address, const u32 length, bool for
         {
           // If the block overlaps, also remove all other occupied slots in the other macro blocks.
           // This will leak empty macro blocks, but they may be reused or cleared later on.
-          u32 block_start = block->physicalAddress;
-          u32 block_end = block_start + (block->originalSize - 1) * 4;
-          for (u32 addr = block_start & mask; addr <= (block_end & mask); addr += BLOCK_RANGE_MAP_ELEMENTS)
-            if (addr != start->first)
-              block_range_map[addr].erase(block);
+          for (u32 addr : block->physical_addresses)
+            if ((addr & range_mask) != start->first)
+              block_range_map[addr & range_mask].erase(block);
 
           // And remove the block.
           DestroyBlock(*block);

Source/Core/Core/PowerPC/JitCommon/JitCache.h

@@ -60,6 +60,9 @@ struct JitBlock
   };
   std::vector<LinkData> linkData;
 
+  // This set stores all physical addresses of all occupied instructions.
+  std::set<u32> physical_addresses;
+
   // we don't really need to save start and stop
   // TODO (mb2): ticStart and ticStop -> "local var" mean "in block" ... low priority ;)
   u64 ticStart;    // for profiling - time.
@@ -127,7 +130,7 @@ public:
   void RunOnBlocks(std::function<void(const JitBlock&)> f);
 
   JitBlock* AllocateBlock(u32 em_address);
-  void FinalizeBlock(JitBlock& block, bool block_link, const u8* code_ptr);
+  void FinalizeBlock(JitBlock& block, bool block_link, const std::set<u32>& physical_addresses);
 
   // Look for the block in the slow but accurate way.
   // This function shall be used if FastLookupIndexForAddress() failed.

Source/Core/Core/PowerPC/MMU.cpp

@@ -384,7 +384,7 @@ TryReadInstResult TryReadInstruction(u32 address)
     auto tlb_addr = TranslateAddress<FLAG_OPCODE>(address);
     if (!tlb_addr.Success())
     {
-      return TryReadInstResult{false, false, 0};
+      return TryReadInstResult{false, false, 0, 0};
     }
     else
     {
@@ -403,7 +403,7 @@ TryReadInstResult TryReadInstruction(u32 address)
   {
     hex = PowerPC::ppcState.iCache.ReadInstruction(address);
   }
-  return TryReadInstResult{true, from_bat, hex};
+  return TryReadInstResult{true, from_bat, hex, address};
 }
 
 u32 HostRead_Instruction(const u32 address)

Source/Core/Core/PowerPC/PPCAnalyst.cpp

@@ -646,6 +646,7 @@ u32 PPCAnalyzer::Analyze(u32 address, CodeBlock* block, CodeBuffer* buffer, u32
   block->m_memory_exception = false;
   block->m_num_instructions = 0;
   block->m_gqr_used = BitSet8(0);
+  block->m_physical_addresses.clear();
 
   CodeOp* code = buffer->codebuffer;
 
@@ -676,6 +677,7 @@ u32 PPCAnalyzer::Analyze(u32 address, CodeBlock* block, CodeBuffer* buffer, u32
     code[i].branchToIndex = -1;
     code[i].skip = false;
     block->m_stats->numCycles += opinfo->numCycles;
+    block->m_physical_addresses.insert(result.physical_address);
 
     SetInstructionStats(block, &code[i], opinfo, i);
 

Source/Core/Core/PowerPC/PPCAnalyst.h

@@ -7,6 +7,7 @@
 #include <algorithm>
 #include <cstdlib>
 #include <map>
+#include <set>
 #include <string>
 #include <vector>
 
@@ -157,6 +158,9 @@ struct CodeBlock
 
   // Which GPRs this block reads from before defining, if any.
   BitSet32 m_gpr_inputs;
+
+  // Which memory locations are occupied by this block.
+  std::set<u32> m_physical_addresses;
 };
 
 class PPCAnalyzer

Source/Core/Core/PowerPC/PowerPC.h

@@ -232,6 +232,7 @@ struct TryReadInstResult
   bool valid;
   bool from_bat;
   u32 hex;
+  u32 physical_address;
 };
 TryReadInstResult TryReadInstruction(const u32 address);