Allocator-local locking for JIT code allocators

Summary: Remove allocator-side `ThreadedCompileSerialize` and make the concrete allocator mutexes unconditional, so `CodeAllocatorCinder` and `MultipleSectionCodeAllocator` rely on their local mutexes for allocator state access on both GIL and FT builds.

Reviewed By: DinoV

Differential Revision: D97125443

fbshipit-source-id: 78b3ef4e4897c3625bf6cecfc21640d9de8ce9a2

Author: yoney

cinderx/Jit/code_allocator.cpp

@@ -4,7 +4,6 @@
 
 #include "cinderx/Common/log.h"
 #include "cinderx/Jit/config.h"
-#include "cinderx/Jit/threaded_compile.h"
 
 #ifdef WIN32
 #include <Windows.h>
@@ -136,10 +135,7 @@ CodeAllocatorCinder::~CodeAllocatorCinder() {
 }
 
 AllocateResult CodeAllocatorCinder::addCode(asmjit::CodeHolder* code) {
-  ThreadedCompileSerialize guard;
-#ifdef Py_GIL_DISABLED
   std::lock_guard lock{allocator_mutex_};
-#endif
 
   PROPAGATE_ERROR(code->flatten());
   PROPAGATE_ERROR(code->resolveUnresolvedLinks());
@@ -197,9 +193,7 @@ asmjit::Error CodeAllocatorCinder::releaseCode([[maybe_unused]] void* code) {
 }
 
 bool CodeAllocatorCinder::contains(const void* ptr) const {
-#ifdef Py_GIL_DISABLED
   std::lock_guard lock{allocator_mutex_};
-#endif
   for (std::span<uint8_t> alloc : allocations_) {
     if (alloc.data() <= ptr && ptr < alloc.data() + alloc.size()) {
       return true;
@@ -254,10 +248,7 @@ void MultipleSectionCodeAllocator::createSlabs() noexcept {
 }
 
 AllocateResult MultipleSectionCodeAllocator::addCode(asmjit::CodeHolder* code) {
-  ThreadedCompileSerialize guard;
-#ifdef Py_GIL_DISABLED
   std::lock_guard lock{allocator_mutex_};
-#endif
 
   if (code_sections_.empty()) {
     createSlabs();
@@ -335,13 +326,10 @@ asmjit::Error MultipleSectionCodeAllocator::releaseCode(
 }
 
 bool MultipleSectionCodeAllocator::contains(const void* ptr) const {
-  // Have to check both the hot/cold slab and the asmjit allocator.  The latter
+  // Have to check both the hot/cold slab and the asmjit allocator. The latter
   // is already thread-safe.
   {
-    ThreadedCompileSerialize guard;
-#ifdef Py_GIL_DISABLED
     std::lock_guard lock{allocator_mutex_};
-#endif
     if (code_alloc_ <= ptr && ptr < code_alloc_ + total_allocation_size_) {
       return true;
     }

cinderx/Jit/code_allocator.h

@@ -69,10 +69,8 @@ class CodeAllocatorCinder : public CodeAllocator {
   bool contains(const void* ptr) const override;
 
  private:
-#ifdef Py_GIL_DISABLED
   // Protects all allocator-owned state used by addCode()/contains().
   mutable std::mutex allocator_mutex_;
-#endif
 
   // List of chunks allocated for use in deallocation
   std::vector<std::span<uint8_t>> allocations_;
@@ -102,10 +100,8 @@ class MultipleSectionCodeAllocator : public CodeAllocator {
  private:
   void createSlabs() noexcept;
 
-#ifdef Py_GIL_DISABLED
   // Protects all allocator-owned state used by addCode()/contains().
   mutable std::mutex allocator_mutex_;
-#endif
 
   std::unordered_map<codegen::CodeSection, uint8_t*> code_sections_;
   std::unordered_map<codegen::CodeSection, size_t> code_section_free_sizes_;