Reduce JIT GC allocations (#2515)

* Turn `MemoryOperand` into a struct * Remove `IntrinsicOperation` * Remove `PhiNode` * Remove `Node` * Turn `Operand` into a struct * Turn `Operation` into a struct * Clean up pool management methods * Add `Arena` allocator * Move `OperationHelper` to `Operation.Factory` * Move `OperandHelper` to `Operand.Factory` * Optimize `Operation` a bit * Fix `Arena` initialization * Rename `NativeList<T>` to `ArenaList<T>` * Reduce `Operand` size from 88 to 56 bytes * Reduce `Operation` size from 56 to 40 bytes * Add optimistic interning of Register & Constant operands * Optimize `RegisterUsage` pass a bit * Optimize `RemoveUnusedNodes` pass a bit Iterating in reverse-order allows killing dependency chains in a single pass. * Fix PPTC symbols * Optimize `BasicBlock` a bit Reduce allocations from `_successor` & `DominanceFrontiers` * Fix `Operation` resize * Make `Arena` expandable Change the arena allocator to be expandable by allocating in pages, with some of them being pooled. Currently 32 pages are pooled. An LRU removal mechanism should probably be added to it. Apparently MHR can allocate bitmaps large enough to exceed the 16MB limit for the type. * Move `Arena` & `ArenaList` to `Common` * Remove `ThreadStaticPool` & co * Add `PhiOperation` * Reduce `Operand` size from 56 from 48 bytes * Add linear-probing to `Operand` intern table * Optimize `HybridAllocator` a bit * Add `Allocators` class * Tune `ArenaAllocator` sizes * Add page removal mechanism to `ArenaAllocator` Remove pages which have not been used for more than 5s after each reset. I am on fence if this would be better using a Gen2 callback object like the one in System.Buffers.ArrayPool<T>, to trim the pool. Because right now if a large translation happens, the pages will be freed only after a reset. This reset may not happen for a while because no new translation is hit, but the arena base sizes are rather small. * Fix `OOM` when allocating larger than page size in `ArenaAllocator` Tweak resizing mechanism for Operand.Uses and Assignemnts. * Optimize `Optimizer` a bit * Optimize `Operand.Add<T>/Remove<T>` a bit * Clean up `PreAllocator` * Fix phi insertion order Reduce codegen diffs. * Fix code alignment * Use new heuristics for degree of parallelism * Suppress warnings * Address gdkchan's feedback Renamed `GetValue()` to `GetValueUnsafe()` to make it more clear that `Operand.Value` should usually not be modified directly. * Add fast path to `ArenaAllocator` * Assembly for `ArenaAllocator.Allocate(ulong)`: .L0: mov rax, [rcx+0x18] lea r8, [rax+rdx] cmp r8, [rcx+0x10] ja short .L2 .L1: mov rdx, [rcx+8] add rax, [rdx+8] mov [rcx+0x18], r8 ret .L2: jmp ArenaAllocator.AllocateSlow(UInt64) A few variable/field had to be changed to ulong so that RyuJIT avoids emitting zero-extends. * Implement a new heuristic to free pooled pages. If an arena is used often, it is more likely that its pages will be needed, so the pages are kept for longer (e.g: during PPTC rebuild or burst sof compilations). If is not used often, then it is more likely that its pages will not be needed (e.g: after PPTC rebuild or bursts of compilations). * Address riperiperi's feedback * Use `EqualityComparer<T>` in `IntrusiveList<T>` Avoids a potential GC hole in `Equals(T, T)`.
author: FICTURE7 <FICTURE7@gmail.com> 2021-08-17 22:08:34 +0400
committer: GitHub <noreply@github.com> 2021-08-17 15:08:34 -0300
commit: 22b2cb39af00fb8881e908fd671fbf57a6e2db2a (patch)
tree: a79e2df801d7f16a33ff50b3c5bfed303cb476e9 /ARMeilleure/CodeGen/Optimizations/Optimizer.cs
parent: cd4530f29c6a4ffd1b023105350b0440fa63f47b (diff)
1 files changed, 68 insertions, 62 deletions
diff --git a/ARMeilleure/CodeGen/Optimizations/Optimizer.cs b/ARMeilleure/CodeGen/Optimizations/Optimizer.cs
index 438010a2..919e996b 100644
--- a/ARMeilleure/CodeGen/Optimizations/Optimizer.cs
+++ b/ARMeilleure/CodeGen/Optimizations/Optimizer.cs
@@ -1,8 +1,8 @@
 using ARMeilleure.IntermediateRepresentation;
 using ARMeilleure.Translation;
+using System;
 using System.Diagnostics;
-
-using static ARMeilleure.IntermediateRepresentation.OperandHelper;
+using static ARMeilleure.IntermediateRepresentation.Operand.Factory;
 
 namespace ARMeilleure.CodeGen.Optimizations
 {
@@ -10,62 +10,60 @@ namespace ARMeilleure.CodeGen.Optimizations
     {
         public static void RunPass(ControlFlowGraph cfg)
         {
+            // Scratch buffer used to store uses.
+            Span<Operation> buffer = default;
+
             bool modified;
 
             do
             {
                 modified = false;
 
-                for (BasicBlock block = cfg.Blocks.First; block != null; block = block.ListNext)
+                for (BasicBlock block = cfg.Blocks.Last; block != null; block = block.ListPrevious)
                 {
-                    Node node = block.Operations.First;
+                    Operation node;
+                    Operation prevNode;
 
-                    while (node != null)
+                    for (node = block.Operations.Last; node != default; node = prevNode)
                     {
-                        Node nextNode = node.ListNext;
+                        prevNode = node.ListPrevious;
 
-                        bool isUnused = IsUnused(node);
-
-                        if (!(node is Operation operation) || isUnused)
+                        if (IsUnused(node))
                         {
-                            if (isUnused)
-                            {
-                                RemoveNode(block, node);
-
-                                modified = true;
-                            }
+                            RemoveNode(block, node);
 
-                            node = nextNode;
+                            modified = true;
 
                             continue;
                         }
+                        else if (node.Instruction == Instruction.Phi)
+                        {
+                            continue;
+                        }
 
-                        ConstantFolding.RunPass(operation);
-
-                        Simplification.RunPass(operation);
+                        ConstantFolding.RunPass(node);
+                        Simplification.RunPass(node);
 
-                        if (DestIsLocalVar(operation))
+                        if (DestIsLocalVar(node))
                         {   
-                            if (IsPropagableCompare(operation))
+                            if (IsPropagableCompare(node))
                             {
-                                modified |= PropagateCompare(operation);
+                                modified |= PropagateCompare(ref buffer, node);
 
-                                if (modified && IsUnused(operation))
+                                if (modified && IsUnused(node))
                                 {
                                     RemoveNode(block, node);
                                 }
                             }
-                            else if (IsPropagableCopy(operation))
+                            else if (IsPropagableCopy(node))
                             {
-                                PropagateCopy(operation);
+                                PropagateCopy(ref buffer, node);
 
                                 RemoveNode(block, node);
 
                                 modified = true;
                             }
                         }
-
-                        node = nextNode;
                     }
                 }
             }
@@ -80,13 +78,14 @@ namespace ARMeilleure.CodeGen.Optimizations
             {
                 modified = false;
 
-                for (BasicBlock block = cfg.Blocks.First; block != null; block = block.ListNext)
+                for (BasicBlock block = cfg.Blocks.Last; block != null; block = block.ListPrevious)
                 {
-                    Node node = block.Operations.First;
+                    Operation node;
+                    Operation prevNode;
 
-                    while (node != null)
+                    for (node = block.Operations.Last; node != default; node = prevNode)
                     {
-                        Node nextNode = node.ListNext;
+                        prevNode = node.ListPrevious;
 
                         if (IsUnused(node))
                         {
@@ -94,15 +93,27 @@ namespace ARMeilleure.CodeGen.Optimizations
 
                             modified = true;
                         }
-
-                        node = nextNode;
                     }
                 }
             }
             while (modified);
         }
 
-        private static bool PropagateCompare(Operation compOp)
+        private static Span<Operation> GetUses(ref Span<Operation> buffer, Operand operand)
+        {
+            ReadOnlySpan<Operation> uses = operand.Uses;
+
+            if (buffer.Length < uses.Length)
+            {
+                buffer = Allocators.Default.AllocateSpan<Operation>((uint)uses.Length);
+            }
+
+            uses.CopyTo(buffer);
+
+            return buffer.Slice(0, uses.Length);
+        }
+
+        private static bool PropagateCompare(ref Span<Operation> buffer, Operation compOp)
         {
             // Try to propagate Compare operations into their BranchIf uses, when these BranchIf uses are in the form
             // of:
@@ -149,17 +160,12 @@ namespace ARMeilleure.CodeGen.Optimizations
 
             Comparison compType = (Comparison)comp.AsInt32();
 
-            Node[] uses = dest.Uses.ToArray();
+            Span<Operation> uses = GetUses(ref buffer, dest);
 
-            foreach (Node use in uses)
+            foreach (Operation use in uses)
             {
-                if (!(use is Operation operation))
-                {
-                    continue;
-                }
-
                 // If operation is a BranchIf and has a constant value 0 in its RHS or LHS source operands.
-                if (IsZeroBranch(operation, out Comparison otherCompType))
+                if (IsZeroBranch(use, out Comparison otherCompType))
                 {
                     Comparison propCompType;
 
@@ -176,9 +182,9 @@ namespace ARMeilleure.CodeGen.Optimizations
                         continue;
                     }
 
-                    operation.SetSource(0, src1);
-                    operation.SetSource(1, src2);
-                    operation.SetSource(2, Const((int)propCompType));
+                    use.SetSource(0, src1);
+                    use.SetSource(1, src2);
+                    use.SetSource(2, Const((int)propCompType));
 
                     modified = true;
                 }
@@ -187,15 +193,15 @@ namespace ARMeilleure.CodeGen.Optimizations
             return modified;
         }
 
-        private static void PropagateCopy(Operation copyOp)
+        private static void PropagateCopy(ref Span<Operation> buffer, Operation copyOp)
         {
             // Propagate copy source operand to all uses of the destination operand.
             Operand dest   = copyOp.Destination;
             Operand source = copyOp.GetSource(0);
 
-            Node[] uses = dest.Uses.ToArray();
+            Span<Operation> uses = GetUses(ref buffer, dest);
 
-            foreach (Node use in uses)
+            foreach (Operation use in uses)
             {
                 for (int index = 0; index < use.SourcesCount; index++)
                 {
@@ -207,7 +213,7 @@ namespace ARMeilleure.CodeGen.Optimizations
             }
         }
 
-        private static void RemoveNode(BasicBlock block, Node node)
+        private static void RemoveNode(BasicBlock block, Operation node)
         {
             // Remove a node from the nodes list, and also remove itself
             // from all the use lists on the operands that this node uses.
@@ -215,31 +221,31 @@ namespace ARMeilleure.CodeGen.Optimizations
 
             for (int index = 0; index < node.SourcesCount; index++)
             {
-                node.SetSource(index, null);
+                node.SetSource(index, default);
             }
 
-            Debug.Assert(node.Destination == null || node.Destination.Uses.Count == 0);
+            Debug.Assert(node.Destination == default || node.Destination.UsesCount == 0);
 
-            node.Destination = null;
+            node.Destination = default;
         }
 
-        private static bool IsUnused(Node node)
+        private static bool IsUnused(Operation node)
         {
-            return DestIsLocalVar(node) && node.Destination.Uses.Count == 0 && !HasSideEffects(node);
+            return DestIsLocalVar(node) && node.Destination.UsesCount == 0 && !HasSideEffects(node);
         }
 
-        private static bool DestIsLocalVar(Node node)
+        private static bool DestIsLocalVar(Operation node)
         {
-            return node.Destination != null && node.Destination.Kind == OperandKind.LocalVariable;
+            return node.Destination != default && node.Destination.Kind == OperandKind.LocalVariable;
         }
 
-        private static bool HasSideEffects(Node node)
+        private static bool HasSideEffects(Operation node)
         {
-            return (node is Operation operation) && (operation.Instruction == Instruction.Call
-                || operation.Instruction == Instruction.Tailcall
-                || operation.Instruction == Instruction.CompareAndSwap
-                || operation.Instruction == Instruction.CompareAndSwap16
-                || operation.Instruction == Instruction.CompareAndSwap8);
+            return node.Instruction == Instruction.Call
+                || node.Instruction == Instruction.Tailcall
+                || node.Instruction == Instruction.CompareAndSwap
+                || node.Instruction == Instruction.CompareAndSwap16
+                || node.Instruction == Instruction.CompareAndSwap8;
         }
 
         private static bool IsPropagableCompare(Operation operation)
author	FICTURE7 <FICTURE7@gmail.com>	2021-08-17 22:08:34 +0400
committer	GitHub <noreply@github.com>	2021-08-17 15:08:34 -0300
commit	22b2cb39af00fb8881e908fd671fbf57a6e2db2a (patch)
tree	a79e2df801d7f16a33ff50b3c5bfed303cb476e9 /ARMeilleure/CodeGen/Optimizations/Optimizer.cs
parent	cd4530f29c6a4ffd1b023105350b0440fa63f47b (diff)