Reland "[vm, compiler] Reduce alignment of Instructions and remove some debugging trap instructions."

Further adjustments for simarm_x64.

Flutter Gallery:
ARM32
Instructions(CodeSize): 6979424 ->  6771856 (-2.97%)
Total(CodeSize):       10855163 -> 10363319 (-4.53%)
ARM64
Instructions(CodeSize): 7334368 ->  6817312 (-7.05%)
Total(CodeSize):       11505069 -> 10771431 (-6.37%)

Bug: #37103
Bug: #38452
Change-Id: I2bb44e7b2c002d53830b66cf4aedc4455c815326
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/119440
Commit-Queue: Ryan Macnak <rmacnak@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>

Author: rmacnak-google

runtime/vm/compiler/backend/code_statistics.cc

@@ -167,9 +167,10 @@ void CodeStatistics::Finalize() {
   intptr_t function_size = assembler_->CodeSize();
   unaccounted_bytes_ = function_size - instruction_bytes_;
   ASSERT(unaccounted_bytes_ >= 0);
+
+  const intptr_t unaligned_bytes = Instructions::HeaderSize() + function_size;
   alignment_bytes_ =
-      Utils::RoundUp(function_size, OS::PreferredCodeAlignment()) -
-      function_size;
+      Utils::RoundUp(unaligned_bytes, kObjectAlignment) - unaligned_bytes;
   assembler_ = NULL;
 }
 

runtime/vm/compiler/backend/flow_graph_compiler_arm.cc

@@ -952,7 +952,9 @@ void FlowGraphCompiler::CompileGraph() {
 
   VisitBlocks();
 
+#if defined(DEBUG)
   __ bkpt(0);
+#endif
 
   if (!skip_body_compilation()) {
     ASSERT(assembler()->constant_pool_allowed());

runtime/vm/compiler/backend/flow_graph_compiler_arm64.cc

@@ -920,7 +920,9 @@ void FlowGraphCompiler::CompileGraph() {
 
   VisitBlocks();
 
+#if defined(DEBUG)
   __ brk(0);
+#endif
 
   if (!skip_body_compilation()) {
     ASSERT(assembler()->constant_pool_allowed());

runtime/vm/compiler/backend/flow_graph_compiler_ia32.cc

@@ -839,7 +839,9 @@ void FlowGraphCompiler::CompileGraph() {
   VisitBlocks();
 
   if (!skip_body_compilation()) {
+#if defined(DEBUG)
     __ int3();
+#endif
     GenerateDeferredCode();
   }
 }

runtime/vm/compiler/backend/flow_graph_compiler_x64.cc

@@ -935,7 +935,9 @@ void FlowGraphCompiler::CompileGraph() {
   ASSERT(!block_order().is_empty());
   VisitBlocks();
 
+#if defined(DEBUG)
   __ int3();
+#endif
 
   if (!skip_body_compilation()) {
     ASSERT(assembler()->constant_pool_allowed());

runtime/vm/compiler/backend/il_arm.cc

@@ -510,7 +510,9 @@ static void EmitAssertBoolean(Register reg,
   compiler->GenerateRuntimeCall(token_pos, deopt_id,
                                 kNonBoolTypeErrorRuntimeEntry, 1, locs);
   // We should never return here.
+#if defined(DEBUG)
   __ bkpt(0);
+#endif
   __ Bind(&done);
 }
 

runtime/vm/compiler/backend/il_arm64.cc

@@ -499,7 +499,9 @@ static void EmitAssertBoolean(Register reg,
   compiler->GenerateRuntimeCall(token_pos, deopt_id,
                                 kNonBoolTypeErrorRuntimeEntry, 1, locs);
   // We should never return here.
+#if defined(DEBUG)
   __ brk(0);
+#endif
   __ Bind(&done);
 }
 

runtime/vm/compiler/backend/il_ia32.cc

@@ -398,7 +398,9 @@ static void EmitAssertBoolean(Register reg,
   compiler->GenerateRuntimeCall(token_pos, deopt_id,
                                 kNonBoolTypeErrorRuntimeEntry, 1, locs);
   // We should never return here.
+#if defined(DEBUG)
   __ int3();
+#endif
   __ Bind(&done);
 }
 

runtime/vm/compiler/backend/il_x64.cc

@@ -527,7 +527,9 @@ static void EmitAssertBoolean(Register reg,
   compiler->GenerateRuntimeCall(token_pos, deopt_id,
                                 kNonBoolTypeErrorRuntimeEntry, 1, locs);
   // We should never return here.
+#if defined(DEBUG)
   __ int3();
+#endif
   __ Bind(&done);
 }
 

runtime/vm/compiler/relocation.cc

@@ -21,9 +21,12 @@ DEFINE_FLAG(bool,
             false,
             "Generate always trampolines (for testing purposes).");
 
-// The trampolines will have a 1-word object header in front of them.
+// The trampolines will be disguised as FreeListElement objects, with a 1-word
+// object header in front of the jump code.
 const intptr_t kOffsetInTrampoline = kWordSize;
-const intptr_t kTrampolineSize = OS::kMaxPreferredCodeAlignment;
+const intptr_t kTrampolineSize = Utils::RoundUp(
+    kOffsetInTrampoline + PcRelativeTrampolineJumpPattern::kLengthInBytes,
+    kObjectAlignment);
 
 CodeRelocator::CodeRelocator(Thread* thread,
                              GrowableArray<RawCode*>* code_objects,
@@ -408,7 +411,12 @@ bool CodeRelocator::IsTargetInRangeFor(UnresolvedCall* unresolved_call,
 static void MarkAsFreeListElement(uint8_t* trampoline_bytes,
                                   intptr_t trampoline_length) {
   uint32_t tags = 0;
+#if defined(IS_SIMARM_X64)
+  // Account for difference in kObjectAlignment between host and target.
+  tags = RawObject::SizeTag::update(trampoline_length * 2, tags);
+#else
   tags = RawObject::SizeTag::update(trampoline_length, tags);
+#endif
   tags = RawObject::ClassIdTag::update(kFreeListElement, tags);
   tags = RawObject::OldBit::update(true, tags);
   tags = RawObject::OldAndNotMarkedBit::update(true, tags);
@@ -464,9 +472,6 @@ void CodeRelocator::BuildTrampolinesForAlmostOutOfRangeCalls() {
       // buffer.
       auto trampoline_bytes = new uint8_t[kTrampolineSize];
       memset(trampoline_bytes, 0x00, kTrampolineSize);
-      ASSERT((kOffsetInTrampoline +
-              PcRelativeTrampolineJumpPattern::kLengthInBytes) <
-             kTrampolineSize);
       auto unresolved_trampoline = new UnresolvedTrampoline{
           unresolved_call->callee,
           unresolved_call->offset_into_target,