JIT: morph blocks in RPO

When optimizing, process blocks in RPO. Disallow creation of new blocks
and new flow edges (the latter with certain preapproved exceptions).

Morph does not yet take advantage of the RPO to enable more optimization.

Contributes to dotnet#93246.

Author: AndyAyersMS

src/coreclr/jit/block.h

@@ -440,6 +440,7 @@ enum BasicBlockFlags : unsigned __int64
     BBF_TAILCALL_SUCCESSOR             = MAKE_BBFLAG(40), // BB has pred that has potential tail call
     BBF_RECURSIVE_TAILCALL             = MAKE_BBFLAG(41), // Block has recursive tailcall that may turn into a loop
     BBF_NO_CSE_IN                      = MAKE_BBFLAG(42), // Block should kill off any incoming CSE
+    BBF_CAN_ADD_PRED                   = MAKE_BBFLAG(43), // Ok to add pred edge to this block, even when "safe" edge creation disabled
 
     // The following are sets of flags.
 

src/coreclr/jit/compiler.h

@@ -3215,10 +3215,11 @@ class Compiler
 //=========================================================================
 // BasicBlock functions
 #ifdef DEBUG
-    // This is a debug flag we will use to assert when creating block during codegen
-    // as this interferes with procedure splitting. If you know what you're doing, set
-    // it to true before creating the block. (DEBUG only)
+    // When false, assert when creating a new basic block.
     bool fgSafeBasicBlockCreation;
+
+    // When false, assert when creating a new flow edge
+    bool fgSafeFlowEdgeCreation;
 #endif
 
     /*

src/coreclr/jit/fgbasic.cpp

@@ -84,6 +84,7 @@ void Compiler::fgInit()
 
 #ifdef DEBUG
     fgSafeBasicBlockCreation = true;
+    fgSafeFlowEdgeCreation   = true;
 #endif // DEBUG
 
     fgLocalVarLivenessDone = false;

src/coreclr/jit/fgflow.cpp

@@ -162,6 +162,11 @@ FlowEdge* Compiler::fgAddRefPred(BasicBlock* block, BasicBlock* blockPred, FlowE
     }
     else
     {
+        // Create a new edge
+        //
+        // We may be disallowing edge creation, except for edges targeting special blocks.
+        //
+        assert(fgSafeFlowEdgeCreation || ((block->bbFlags & BBF_CAN_ADD_PRED) != 0));
 
 #if MEASURE_BLOCK_SIZE
         genFlowNodeCnt += 1;

src/coreclr/jit/flowgraph.cpp

@@ -2228,7 +2228,7 @@ class MergedReturns
         // Add 'return' expression to the return block
         comp->fgNewStmtAtEnd(newReturnBB, returnExpr);
         // Flag that this 'return' was generated by return merging so that subsequent
-        // return block morhping will know to leave it alone.
+        // return block merging will know to leave it alone.
         returnExpr->gtFlags |= GTF_RET_MERGED;
 
 #ifdef DEBUG

src/coreclr/jit/morph.cpp

@@ -7472,7 +7472,6 @@ void Compiler::fgMorphRecursiveFastTailCallIntoLoop(BasicBlock* block, GenTreeCa
 //------------------------------------------------------------------------------
 // fgAssignRecursiveCallArgToCallerParam : Assign argument to a recursive call to the corresponding caller parameter.
 //
-//
 // Arguments:
 //    arg  -  argument to assign
 //    late  -  whether to use early or late arg
@@ -13892,13 +13891,70 @@ PhaseStatus Compiler::fgMorphBlocks()
 
     // Morph all blocks.
     //
-    // Note morph can add blocks downstream from the current block,
-    // and alter (but not null out) the current block's bbNext;
-    // this iterator ensures they all get visited.
-    //
-    for (BasicBlock* block : Blocks())
+    if (!optLocalAssertionProp)
     {
-        fgMorphBlock(block);
+        // If we aren't optimizing, we just morph in normal bbNext order.
+        //
+        // Note morph can add blocks downstream from the current block,
+        // and alter (but not null out) the current block's bbNext;
+        // this iterator ensures they all get visited.
+        //
+        for (BasicBlock* block : Blocks())
+        {
+            fgMorphBlock(block);
+        }
+    }
+    else
+    {
+        // We are optimizing. Process in RPO.
+        //
+        fgRenumberBlocks();
+        EnsureBasicBlockEpoch();
+        fgComputeEnterBlocksSet();
+        fgDfsReversePostorder();
+
+        // Disallow general creation of new blocks or edges as it
+        // would invalidate RPO.
+        //
+        // Removal of edges, or altering dup counts, is OK.
+        //
+        INDEBUG(fgSafeBasicBlockCreation = false;);
+        INDEBUG(fgSafeFlowEdgeCreation = false;);
+
+        // Allow edge creation to genReturnBB (target of return merging)
+        // and the scratch block successor (target for tail call to loop).
+        // This will also disallow dataflow into these blocks.
+        //
+        if (genReturnBB != nullptr)
+        {
+            genReturnBB->bbFlags |= BBF_CAN_ADD_PRED;
+        }
+        if (fgFirstBBisScratch())
+        {
+            fgFirstBB->Next()->bbFlags |= BBF_CAN_ADD_PRED;
+        }
+
+        unsigned const bbNumMax = fgBBNumMax;
+        for (unsigned i = 1; i <= bbNumMax; i++)
+        {
+            BasicBlock* const block = fgBBReversePostorder[i];
+            fgMorphBlock(block);
+        }
+
+        // Re-enable block and edge creation, and revoke
+        // special treatment of genReturnBB and the "first" bb
+        //
+        INDEBUG(fgSafeBasicBlockCreation = true;);
+        INDEBUG(fgSafeFlowEdgeCreation = true;);
+
+        if (genReturnBB != nullptr)
+        {
+            genReturnBB->bbFlags &= ~BBF_CAN_ADD_PRED;
+        }
+        if (fgFirstBBisScratch())
+        {
+            fgFirstBB->Next()->bbFlags &= ~BBF_CAN_ADD_PRED;
+        }
     }
 
     // Under OSR, we no longer need to specially protect the original method entry