8299162: Refactor shared trampoline emission logic

Reviewed-by: fyang, adinn, luhenry

src/hotspot/cpu/aarch64/codeBuffer_aarch64.cpp

@@ -42,6 +42,8 @@ void CodeBuffer::share_trampoline_for(address dest, int caller_offset) {
   _finalize_stubs = true;
 }
 
+#define __ masm.
+
 static bool emit_shared_trampolines(CodeBuffer* cb, CodeBuffer::SharedTrampolineRequests* requests) {
   if (requests == nullptr) {
     return true;
@@ -49,39 +51,35 @@ static bool emit_shared_trampolines(CodeBuffer* cb, CodeBuffer::SharedTrampoline
 
   MacroAssembler masm(cb);
 
-  bool p_succeeded = true;
   auto emit = [&](address dest, const CodeBuffer::Offsets &offsets) {
-    masm.set_code_section(cb->stubs());
+    assert(cb->stubs()->remaining() >= MacroAssembler::max_trampoline_stub_size(), "pre-allocated trampolines");
-    if (!is_aligned(masm.offset(), wordSize)) {
-      if (cb->stubs()->maybe_expand_to_ensure_remaining(NativeInstruction::instruction_size) && cb->blob() == NULL) {
-        ciEnv::current()->record_failure("CodeCache is full");
-        p_succeeded = false;
-        return p_succeeded;
-      }
-      masm.align(wordSize);
-    }
-
     LinkedListIterator<int> it(offsets.head());
     int offset = *it.next();
-    for (; !it.is_empty(); offset = *it.next()) {
+    address stub = __ emit_trampoline_stub(offset, dest);
-      masm.relocate(trampoline_stub_Relocation::spec(cb->insts()->start() + offset));
+    assert(stub, "pre-allocated trampolines");
-    }
-    masm.set_code_section(cb->insts());
 
-    address stub = masm.emit_trampoline_stub(offset, dest);
+    address reloc_pc = cb->stubs()->end() - NativeCallTrampolineStub::instruction_size;
-    if (stub == nullptr) {
+    while (!it.is_empty()) {
-      ciEnv::current()->record_failure("CodeCache is full");
+      offset = *it.next();
-      p_succeeded = false;
+      address caller_pc = cb->insts()->start() + offset;
+      cb->stubs()->relocate(reloc_pc, trampoline_stub_Relocation::spec(caller_pc));
     }
-
+    return true;
-    return p_succeeded;
   };
 
-  requests->iterate(emit);
+  assert(requests->number_of_entries() >= 1, "at least one");
-
+  const int total_requested_size = MacroAssembler::max_trampoline_stub_size() * requests->number_of_entries();
-  return p_succeeded;
+  if (cb->stubs()->maybe_expand_to_ensure_remaining(total_requested_size) && cb->blob() == NULL) {
+    ciEnv::current()->record_failure("CodeCache is full");
+    return false;
   }
 
+  requests->iterate(emit);
+  return true;
+}
+
+#undef __
+
 bool CodeBuffer::pd_finalize_stubs() {
   return emit_shared_stubs_to_interp<MacroAssembler>(this, _shared_stub_to_interp_requests)
       && emit_shared_trampolines(this, _shared_trampoline_requests);

src/hotspot/cpu/aarch64/compiledIC_aarch64.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1997, 2021, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2023, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2014, 2018, Red Hat Inc. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
@@ -71,15 +71,14 @@ address CompiledStaticCall::emit_to_interp_stub(CodeBuffer &cbuf, address mark)
 #undef __
 
 int CompiledStaticCall::to_interp_stub_size() {
-  // isb; movk; movz; movz; movk; movz; movz; br
+  return MacroAssembler::static_call_stub_size();
-  return 8 * NativeInstruction::instruction_size;
 }
 
 int CompiledStaticCall::to_trampoline_stub_size() {
   // Somewhat pessimistically, we count 3 instructions here (although
   // there are only two) because we sometimes emit an alignment nop.
   // Trampoline stubs are always word aligned.
-  return 3 * NativeInstruction::instruction_size + wordSize;
+  return MacroAssembler::max_trampoline_stub_size();
 }
 
 // Relocation entries for call stub, compiled java to interpreter.

src/hotspot/cpu/aarch64/macroAssembler_aarch64.cpp

@@ -926,8 +926,7 @@ address MacroAssembler::trampoline_call(Address entry) {
 address MacroAssembler::emit_trampoline_stub(int insts_call_instruction_offset,
                                              address dest) {
   // Max stub size: alignment nop, TrampolineStub.
-  address stub = start_a_stub(NativeInstruction::instruction_size
+  address stub = start_a_stub(max_trampoline_stub_size());
-                   + NativeCallTrampolineStub::instruction_size);
   if (stub == NULL) {
     return NULL;  // CodeBuffer::expand failed
   }
@@ -959,6 +958,11 @@ address MacroAssembler::emit_trampoline_stub(int insts_call_instruction_offset,
   return stub_start_addr;
 }
 
+int MacroAssembler::max_trampoline_stub_size() {
+  // Max stub size: alignment nop, TrampolineStub.
+  return NativeInstruction::instruction_size + NativeCallTrampolineStub::instruction_size;
+}
+
 void MacroAssembler::emit_static_call_stub() {
   // CompiledDirectStaticCall::set_to_interpreted knows the
   // exact layout of this stub.
@@ -971,6 +975,11 @@ void MacroAssembler::emit_static_call_stub() {
   br(rscratch1);
 }
 
+int MacroAssembler::static_call_stub_size() {
+  // isb; movk; movz; movz; movk; movz; movz; br
+  return 8 * NativeInstruction::instruction_size;
+}
+
 void MacroAssembler::c2bool(Register x) {
   // implements x == 0 ? 0 : 1
   // note: must only look at least-significant byte of x

src/hotspot/cpu/aarch64/macroAssembler_aarch64.hpp

@@ -638,7 +638,9 @@ public:
     return false;
   }
   address emit_trampoline_stub(int insts_call_instruction_offset, address target);
+  static int max_trampoline_stub_size();
   void emit_static_call_stub();
+  static int static_call_stub_size();
 
   // The following 4 methods return the offset of the appropriate move instruction
 

src/hotspot/cpu/riscv/codeBuffer_riscv.cpp

@@ -44,6 +44,8 @@ void CodeBuffer::share_trampoline_for(address dest, int caller_offset) {
   _finalize_stubs = true;
 }
 
+#define __ masm.
+
 static bool emit_shared_trampolines(CodeBuffer* cb, CodeBuffer::SharedTrampolineRequests* requests) {
   if (requests == nullptr) {
     return true;
@@ -51,39 +53,35 @@ static bool emit_shared_trampolines(CodeBuffer* cb, CodeBuffer::SharedTrampoline
 
   MacroAssembler masm(cb);
 
-  bool p_succeeded = true;
   auto emit = [&](address dest, const CodeBuffer::Offsets &offsets) {
-    masm.set_code_section(cb->stubs());
+    assert(cb->stubs()->remaining() >= MacroAssembler::max_trampoline_stub_size(), "pre-allocated trampolines");
-    if (!is_aligned(masm.offset() + NativeCallTrampolineStub::data_offset, wordSize)) {
-      if (cb->stubs()->maybe_expand_to_ensure_remaining(NativeInstruction::instruction_size) && cb->blob() == NULL) {
-        ciEnv::current()->record_failure("CodeCache is full");
-        p_succeeded = false;
-        return p_succeeded;
-      }
-      masm.align(wordSize, NativeCallTrampolineStub::data_offset);
-    }
-
     LinkedListIterator<int> it(offsets.head());
     int offset = *it.next();
-    for (; !it.is_empty(); offset = *it.next()) {
+    address stub = __ emit_trampoline_stub(offset, dest);
-      masm.relocate(trampoline_stub_Relocation::spec(cb->insts()->start() + offset));
+    assert(stub, "pre-allocated trampolines");
-    }
-    masm.set_code_section(cb->insts());
 
-    address stub = masm.emit_trampoline_stub(offset, dest);
+    address reloc_pc = cb->stubs()->end() - NativeCallTrampolineStub::instruction_size;
-    if (stub == nullptr) {
+    while (!it.is_empty()) {
-      ciEnv::current()->record_failure("CodeCache is full");
+      offset = *it.next();
-      p_succeeded = false;
+      address caller_pc = cb->insts()->start() + offset;
+      cb->stubs()->relocate(reloc_pc, trampoline_stub_Relocation::spec(caller_pc));
     }
-
+    return true;
-    return p_succeeded;
   };
 
-  requests->iterate(emit);
+  assert(requests->number_of_entries() >= 1, "at least one");
-
+  const int total_requested_size = MacroAssembler::max_trampoline_stub_size() * requests->number_of_entries();
-  return p_succeeded;
+  if (cb->stubs()->maybe_expand_to_ensure_remaining(total_requested_size) && cb->blob() == NULL) {
+    ciEnv::current()->record_failure("CodeCache is full");
+    return false;
   }
 
+  requests->iterate(emit);
+  return true;
+}
+
+#undef __
+
 bool CodeBuffer::pd_finalize_stubs() {
   return emit_shared_stubs_to_interp<MacroAssembler>(this, _shared_stub_to_interp_requests)
          && emit_shared_trampolines(this, _shared_trampoline_requests);

src/hotspot/cpu/riscv/compiledIC_riscv.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1997, 2019, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2023, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2014, 2018, Red Hat Inc. All rights reserved.
  * Copyright (c) 2020, 2021, Huawei Technologies Co., Ltd. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
@@ -69,15 +69,14 @@ address CompiledStaticCall::emit_to_interp_stub(CodeBuffer &cbuf, address mark)
 #undef __
 
 int CompiledStaticCall::to_interp_stub_size() {
-  // (lui, addi, slli, addi, slli, addi) + (lui, addi, slli, addi, slli) + jalr
+  return MacroAssembler::static_call_stub_size();
-  return 12 * NativeInstruction::instruction_size;
 }
 
 int CompiledStaticCall::to_trampoline_stub_size() {
   // Somewhat pessimistically, we count 4 instructions here (although
   // there are only 3) because we sometimes emit an alignment nop.
   // Trampoline stubs are always word aligned.
-  return NativeInstruction::instruction_size + NativeCallTrampolineStub::instruction_size;
+  return MacroAssembler::max_trampoline_stub_size();
 }
 
 // Relocation entries for call stub, compiled java to interpreter.

src/hotspot/cpu/riscv/macroAssembler_riscv.cpp

@@ -3142,8 +3142,8 @@ address MacroAssembler::ic_call(address entry, jint method_index) {
 
 address MacroAssembler::emit_trampoline_stub(int insts_call_instruction_offset,
                                              address dest) {
-  address stub = start_a_stub(NativeInstruction::instruction_size
+  // Max stub size: alignment nop, TrampolineStub.
-                            + NativeCallTrampolineStub::instruction_size);
+  address stub = start_a_stub(max_trampoline_stub_size());
   if (stub == NULL) {
     return NULL;  // CodeBuffer::expand failed
   }
@@ -3183,6 +3183,16 @@ address MacroAssembler::emit_trampoline_stub(int insts_call_instruction_offset,
   return stub_start_addr;
 }
 
+int MacroAssembler::max_trampoline_stub_size() {
+  // Max stub size: alignment nop, TrampolineStub.
+  return NativeInstruction::instruction_size + NativeCallTrampolineStub::instruction_size;
+}
+
+int MacroAssembler::static_call_stub_size() {
+  // (lui, addi, slli, addi, slli, addi) + (lui, addi, slli, addi, slli) + jalr
+  return 12 * NativeInstruction::instruction_size;
+}
+
 Address MacroAssembler::add_memory_helper(const Address dst, Register tmp) {
   switch (dst.getMode()) {
     case Address::base_plus_offset:

src/hotspot/cpu/riscv/macroAssembler_riscv.hpp

@@ -412,7 +412,9 @@ class MacroAssembler: public Assembler {
   }
 
   address emit_trampoline_stub(int insts_call_instruction_offset, address target);
+  static int max_trampoline_stub_size();
   void emit_static_call_stub();
+  static int static_call_stub_size();
 
   // The following 4 methods return the offset of the appropriate move instruction
 

src/hotspot/share/asm/codeBuffer.inline.hpp

@@ -29,6 +29,8 @@
 #include "ci/ciEnv.hpp"
 #include "code/compiledIC.hpp"
 
+#define __ masm.
+
 template <typename MacroAssembler, int relocate_format = 0>
 bool emit_shared_stubs_to_interp(CodeBuffer* cb, SharedStubToInterpRequests* shared_stub_to_interp_requests) {
   if (shared_stub_to_interp_requests == NULL) {
@@ -46,7 +48,7 @@ bool emit_shared_stubs_to_interp(CodeBuffer* cb, SharedStubToInterpRequests* sha
   shared_stub_to_interp_requests->sort(by_shared_method);
   MacroAssembler masm(cb);
   for (int i = 0; i < shared_stub_to_interp_requests->length();) {
-    address stub = masm.start_a_stub(CompiledStaticCall::to_interp_stub_size());
+    address stub = __ start_a_stub(CompiledStaticCall::to_interp_stub_size());
     if (stub == NULL) {
       ciEnv::current()->record_failure("CodeCache is full");
       return false;
@@ -55,13 +57,15 @@ bool emit_shared_stubs_to_interp(CodeBuffer* cb, SharedStubToInterpRequests* sha
     ciMethod* method = shared_stub_to_interp_requests->at(i).shared_method();
     do {
       address caller_pc = cb->insts_begin() + shared_stub_to_interp_requests->at(i).call_offset();
-      masm.relocate(static_stub_Relocation::spec(caller_pc), relocate_format);
+      __ relocate(static_stub_Relocation::spec(caller_pc), relocate_format);
       ++i;
     } while (i < shared_stub_to_interp_requests->length() && shared_stub_to_interp_requests->at(i).shared_method() == method);
-    masm.emit_static_call_stub();
+    __ emit_static_call_stub();
-    masm.end_a_stub();
+    __ end_a_stub();
   }
   return true;
 }
 
+#undef __
+
 #endif // SHARE_ASM_CODEBUFFER_INLINE_HPP