8231349: Move intrinsic stubs generation to compiler runtime initialization code

Reviewed-by: redestad, vlivanov

Author: Vladimir Kozlov

src/hotspot/cpu/aarch64/globals_aarch64.hpp

@@ -33,9 +33,11 @@
 // (see globals.hpp)
 
 define_pd_global(bool, ImplicitNullChecks,       true);  // Generate code for implicit null checks
-define_pd_global(bool, TrapBasedNullChecks,  false);
+define_pd_global(bool, TrapBasedNullChecks,     false);
 define_pd_global(bool, UncommonNullCast,         true);  // Uncommon-trap NULLs past to check cast
 
+define_pd_global(bool, DelayCompilerStubsGeneration, COMPILER2_OR_JVMCI);
+
 define_pd_global(uintx, CodeCacheSegmentSize,    64 COMPILER1_AND_COMPILER2_PRESENT(+64)); // Tiered compilation has large code-entry alignment.
 define_pd_global(intx, CodeEntryAlignment,       64);
 define_pd_global(intx, OptoLoopAlignment,        16);

src/hotspot/cpu/aarch64/sharedRuntime_aarch64.cpp

@@ -570,26 +570,31 @@ void SharedRuntime::gen_i2c_adapter(MacroAssembler *masm,
   // caller, but with an uncorrected stack, causing delayed havoc.
 
   if (VerifyAdapterCalls &&
-      (Interpreter::code() != NULL || StubRoutines::code1() != NULL)) {
+      (Interpreter::code() != NULL || StubRoutines::final_stubs_code() != NULL)) {
 #if 0
     // So, let's test for cascading c2i/i2c adapters right now.
     //  assert(Interpreter::contains($return_addr) ||
     //         StubRoutines::contains($return_addr),
     //         "i2c adapter must return to an interpreter frame");
     __ block_comment("verify_i2c { ");
     Label L_ok;
-    if (Interpreter::code() != NULL)
+    if (Interpreter::code() != NULL) {
       range_check(masm, rax, r11,
                   Interpreter::code()->code_start(), Interpreter::code()->code_end(),
                   L_ok);
-    if (StubRoutines::code1() != NULL)
+    }
+    if (StubRoutines::initial_stubs_code() != NULL) {
       range_check(masm, rax, r11,
-                  StubRoutines::code1()->code_begin(), StubRoutines::code1()->code_end(),
+                  StubRoutines::initial_stubs_code()->code_begin(),
+                  StubRoutines::initial_stubs_code()->code_end(),
                   L_ok);
-    if (StubRoutines::code2() != NULL)
+    }
+    if (StubRoutines::final_stubs_code() != NULL) {
       range_check(masm, rax, r11,
-                  StubRoutines::code2()->code_begin(), StubRoutines::code2()->code_end(),
+                  StubRoutines::final_stubs_code()->code_begin(),
+                  StubRoutines::final_stubs_code()->code_end(),
                   L_ok);
+    }
     const char* msg = "i2c adapter must return to an interpreter frame";
     __ block_comment(msg);
     __ stop(msg);

src/hotspot/cpu/aarch64/stubGenerator_aarch64.cpp

@@ -7997,7 +7997,7 @@ class StubGenerator: public StubCodeGenerator {
 
 
   // Initialization
-  void generate_initial() {
+  void generate_initial_stubs() {
     // Generate initial stubs and initializes the entry points
 
     // entry points that exist in all platforms Note: This is code
@@ -8023,6 +8023,12 @@ class StubGenerator: public StubCodeGenerator {
       generate_throw_exception("delayed StackOverflowError throw_exception",
                                CAST_FROM_FN_PTR(address,
                                                 SharedRuntime::throw_delayed_StackOverflowError));
+
+    // Initialize table for copy memory (arraycopy) check.
+    if (UnsafeCopyMemory::_table == nullptr) {
+      UnsafeCopyMemory::create_table(8);
+    }
+
     if (UseCRC32Intrinsics) {
       // set table address before stub generation which use it
       StubRoutines::_crc_table_adr = (address)StubRoutines::aarch64::_crc_table;
@@ -8047,7 +8053,7 @@ class StubGenerator: public StubCodeGenerator {
     }
   }
 
-  void generate_phase1() {
+  void generate_continuation_stubs() {
     // Continuation stubs:
     StubRoutines::_cont_thaw          = generate_cont_thaw();
     StubRoutines::_cont_returnBarrier = generate_cont_returnBarrier();
@@ -8057,7 +8063,7 @@ class StubGenerator: public StubCodeGenerator {
     JFR_ONLY(StubRoutines::_jfr_write_checkpoint = StubRoutines::_jfr_write_checkpoint_stub->entry_point();)
   }
 
-  void generate_all() {
+  void generate_final_stubs() {
     // support for verify_oop (must happen after universe_init)
     if (VerifyOops) {
       StubRoutines::_verify_oop_subroutine_entry   = generate_verify_oop();
@@ -8080,32 +8086,47 @@ class StubGenerator: public StubCodeGenerator {
                                                 SharedRuntime::
                                                 throw_NullPointerException_at_call));
 
-    if (UseSVE == 0) {
-      StubRoutines::aarch64::_vector_iota_indices = generate_iota_indices("iota_indices");
-    }
-
     // arraycopy stubs used by compilers
     generate_arraycopy_stubs();
 
-    // countPositives stub for large arrays.
-    StubRoutines::aarch64::_count_positives = generate_count_positives(StubRoutines::aarch64::_count_positives_long);
+    BarrierSetNMethod* bs_nm = BarrierSet::barrier_set()->barrier_set_nmethod();
+    if (bs_nm != NULL) {
+      StubRoutines::aarch64::_method_entry_barrier = generate_method_entry_barrier();
+    }
+
+    StubRoutines::aarch64::_spin_wait = generate_spin_wait();
+
+#if defined (LINUX) && !defined (__ARM_FEATURE_ATOMICS)
+
+    generate_atomic_entry_points();
+
+#endif // LINUX
+
+    StubRoutines::aarch64::set_completed(); // Inidicate that arraycopy and zero_blocks stubs are generated
+  }
+
+  void generate_compiler_stubs() {
+#if COMPILER2_OR_JVMCI
+
+    if (UseSVE == 0) {
+      StubRoutines::aarch64::_vector_iota_indices = generate_iota_indices("iota_indices");
+    }
 
     // array equals stub for large arrays.
     if (!UseSimpleArrayEquals) {
       StubRoutines::aarch64::_large_array_equals = generate_large_array_equals();
     }
 
+    // byte_array_inflate stub for large arrays.
+    StubRoutines::aarch64::_large_byte_array_inflate = generate_large_byte_array_inflate();
+
+    // countPositives stub for large arrays.
+    StubRoutines::aarch64::_count_positives = generate_count_positives(StubRoutines::aarch64::_count_positives_long);
+
     generate_compare_long_strings();
 
     generate_string_indexof_stubs();
 
-    // byte_array_inflate stub for large arrays.
-    StubRoutines::aarch64::_large_byte_array_inflate = generate_large_byte_array_inflate();
-
-    BarrierSetNMethod* bs_nm = BarrierSet::barrier_set()->barrier_set_nmethod();
-    if (bs_nm != NULL) {
-      StubRoutines::aarch64::_method_entry_barrier = generate_method_entry_barrier();
-    }
 #ifdef COMPILER2
     if (UseMultiplyToLenIntrinsic) {
       StubRoutines::_multiplyToLen = generate_multiplyToLen();
@@ -8192,36 +8213,33 @@ class StubGenerator: public StubCodeGenerator {
     if (UseAdler32Intrinsics) {
       StubRoutines::_updateBytesAdler32 = generate_updateBytesAdler32();
     }
-
-    StubRoutines::aarch64::_spin_wait = generate_spin_wait();
-
-#if defined (LINUX) && !defined (__ARM_FEATURE_ATOMICS)
-
-    generate_atomic_entry_points();
-
-#endif // LINUX
-
-    StubRoutines::aarch64::set_completed();
+#endif // COMPILER2_OR_JVMCI
   }
 
  public:
-  StubGenerator(CodeBuffer* code, int phase) : StubCodeGenerator(code) {
-    if (phase == 0) {
-      generate_initial();
-    } else if (phase == 1) {
-      generate_phase1(); // stubs that must be available for the interpreter
-    } else {
-      generate_all();
-    }
+  StubGenerator(CodeBuffer* code, StubsKind kind) : StubCodeGenerator(code) {
+    switch(kind) {
+    case Initial_stubs:
+      generate_initial_stubs();
+      break;
+     case Continuation_stubs:
+      generate_continuation_stubs();
+      break;
+    case Compiler_stubs:
+      generate_compiler_stubs();
+      break;
+    case Final_stubs:
+      generate_final_stubs();
+      break;
+    default:
+      fatal("unexpected stubs kind: %d", kind);
+      break;
+    };
   }
 }; // end class declaration
 
-#define UCM_TABLE_MAX_ENTRIES 8
-void StubGenerator_generate(CodeBuffer* code, int phase) {
-  if (UnsafeCopyMemory::_table == NULL) {
-    UnsafeCopyMemory::create_table(UCM_TABLE_MAX_ENTRIES);
-  }
-  StubGenerator g(code, phase);
+void StubGenerator_generate(CodeBuffer* code, StubCodeGenerator::StubsKind kind) {
+  StubGenerator g(code, kind);
 }
 
 

src/hotspot/cpu/aarch64/stubRoutines_aarch64.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2003, 2020, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2003, 2023, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2014, Red Hat Inc. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
@@ -35,8 +35,11 @@ static bool    returns_to_call_stub(address return_pc)   {
 }
 
 enum platform_dependent_constants {
-  code_size1 = 19000,          // simply increase if too small (assembler will crash if too small)
-  code_size2 = 45000           // simply increase if too small (assembler will crash if too small)
+  // simply increase sizes if too small (assembler will crash if too small)
+  _initial_stubs_code_size      = 10000,
+  _continuation_stubs_code_size =  2000,
+  _compiler_stubs_code_size     = 30000,
+  _final_stubs_code_size        = 20000
 };
 
 class aarch64 {

src/hotspot/cpu/arm/globals_arm.hpp

@@ -34,6 +34,8 @@ define_pd_global(bool,  ImplicitNullChecks,       true);  // Generate code for i
 define_pd_global(bool,  UncommonNullCast,         true);  // Uncommon-trap nulls past to check cast
 define_pd_global(bool,  TrapBasedNullChecks,      false); // Not needed
 
+define_pd_global(bool,  DelayCompilerStubsGeneration, false); // No need - only few compiler's stubs
+
 define_pd_global(uintx, CodeCacheSegmentSize, 64 COMPILER1_AND_COMPILER2_PRESENT(+64)); // Tiered compilation has large code-entry alignment.
 define_pd_global(intx,  CodeEntryAlignment,       16);
 define_pd_global(intx,  OptoLoopAlignment,        16);

src/hotspot/cpu/arm/stubGenerator_arm.cpp

@@ -3077,7 +3077,7 @@ class StubGenerator: public StubCodeGenerator {
   //---------------------------------------------------------------------------
   // Initialization
 
-  void generate_initial() {
+  void generate_initial_stubs() {
     // Generates all stubs and initializes the entry points
 
     //------------------------------------------------------------------------------------------------------------------------
@@ -3094,6 +3094,10 @@ class StubGenerator: public StubCodeGenerator {
     // stub for throwing stack overflow error used both by interpreter and compiler
     StubRoutines::_throw_StackOverflowError_entry  = generate_throw_exception("StackOverflowError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError));
 
+    if (UnsafeCopyMemory::_table == nullptr) {
+      UnsafeCopyMemory::create_table(32);
+    }
+
     // integer division used both by interpreter and compiler
     StubRoutines::Arm::_idiv_irem_entry = generate_idiv_irem();
 
@@ -3106,7 +3110,7 @@ class StubGenerator: public StubCodeGenerator {
 
   }
 
-  void generate_phase1() {
+  void generate_continuation_stubs() {
     // Continuation stubs:
     StubRoutines::_cont_thaw          = generate_cont_thaw();
     StubRoutines::_cont_returnBarrier = generate_cont_returnBarrier();
@@ -3116,14 +3120,9 @@ class StubGenerator: public StubCodeGenerator {
     JFR_ONLY(StubRoutines::_jfr_write_checkpoint = StubRoutines::_jfr_write_checkpoint_stub->entry_point();)
   }
 
-  void generate_all() {
+  void generate_final_stubs() {
     // Generates all stubs and initializes the entry points
 
-#ifdef COMPILER2
-    // Generate partial_subtype_check first here since its code depends on
-    // UseZeroBaseCompressedOops which is defined after heap initialization.
-    StubRoutines::Arm::_partial_subtype_check                = generate_partial_subtype_check();
-#endif
     // These entry points require SharedInfo::stack0 to be set up in non-core builds
     // and need to be relocatable, so they each fabricate a RuntimeStub internally.
     StubRoutines::_throw_AbstractMethodError_entry         = generate_throw_exception("AbstractMethodError throw_exception",          CAST_FROM_FN_PTR(address, SharedRuntime::throw_AbstractMethodError));
@@ -3144,6 +3143,14 @@ class StubGenerator: public StubCodeGenerator {
       StubRoutines::Arm::_method_entry_barrier = generate_method_entry_barrier();
     }
 
+  }
+
+  void generate_compiler_stubs() {
+#ifdef COMPILER2
+    // Generate partial_subtype_check first here since its code depends on
+    // UseZeroBaseCompressedOops which is defined after heap initialization.
+    StubRoutines::Arm::_partial_subtype_check                = generate_partial_subtype_check();
+
 #ifdef COMPILE_CRYPTO
     // generate AES intrinsics code
     if (UseAESIntrinsics) {
@@ -3154,25 +3161,31 @@ class StubGenerator: public StubCodeGenerator {
       StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptAESCrypt();
     }
 #endif // COMPILE_CRYPTO
+#endif // COMPILER2
   }
 
-
  public:
-  StubGenerator(CodeBuffer* code, int phase) : StubCodeGenerator(code) {
-    if (phase == 0) {
-      generate_initial();
-    } else if (phase == 1) {
-      generate_phase1();
-    } else {
-      generate_all();
-    }
+  StubGenerator(CodeBuffer* code, StubsKind kind) : StubCodeGenerator(code) {
+    switch(kind) {
+    case Initial_stubs:
+      generate_initial_stubs();
+      break;
+     case Continuation_stubs:
+      generate_continuation_stubs();
+      break;
+    case Compiler_stubs:
+      generate_compiler_stubs();
+      break;
+    case Final_stubs:
+      generate_final_stubs();
+      break;
+    default:
+      fatal("unexpected stubs kind: %d", kind);
+      break;
+    };
   }
 }; // end class declaration
 
-#define UCM_TABLE_MAX_ENTRIES 32
-void StubGenerator_generate(CodeBuffer* code, int phase) {
-  if (UnsafeCopyMemory::_table == nullptr) {
-    UnsafeCopyMemory::create_table(UCM_TABLE_MAX_ENTRIES);
-  }
-  StubGenerator g(code, phase);
+void StubGenerator_generate(CodeBuffer* code, StubCodeGenerator::StubsKind kind) {
+  StubGenerator g(code, kind);
 }

src/hotspot/cpu/arm/stubRoutines_arm.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2008, 2019, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2008, 2023, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -30,8 +30,11 @@
 // extend it.
 
 enum platform_dependent_constants {
-  code_size1 =  9000,           // simply increase if too small (assembler will crash if too small)
-  code_size2 = 22000            // simply increase if too small (assembler will crash if too small)
+  // simply increase sizes if too small (assembler will crash if too small)
+  _initial_stubs_code_size      =  9000,
+  _continuation_stubs_code_size =  2000,
+  _compiler_stubs_code_size     = 22000,
+  _final_stubs_code_size        = 22000
 };
 
 class Arm {

src/hotspot/cpu/ppc/globals_ppc.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2002, 2020, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2002, 2023, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2012, 2020 SAP SE. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
@@ -36,6 +36,8 @@ define_pd_global(bool, ImplicitNullChecks,    true);  // Generate code for impli
 define_pd_global(bool, TrapBasedNullChecks,   true);
 define_pd_global(bool, UncommonNullCast,      true);  // Uncommon-trap NULLs passed to check cast.
 
+define_pd_global(bool, DelayCompilerStubsGeneration, COMPILER2_OR_JVMCI);
+
 #define DEFAULT_STACK_YELLOW_PAGES (2)
 #define DEFAULT_STACK_RED_PAGES (1)
 // Java_java_net_SocketOutputStream_socketWrite0() uses a 64k buffer on the