[codegen] Use target architecture specific branch alignment

The Assembler class (or some of them at least) have a CodeTargetAlign
method that aligns the code to a target specific value (16 byte on x86,
8 byte on arm). However, these were not used. Instead we always aligned
to 16 byte boundaries, hence wasting up to 8 bytes on arm.

Change-Id: Iee7d24ebc13a9a58002a9d7d0ce53955bee7d628
Reviewed-on: https://chromium-review.googlesource.com/c/1426125
Commit-Queue: Stephan Herhut <herhut@chromium.org>
Reviewed-by: Michael Starzinger <mstarzinger@chromium.org>
Cr-Commit-Position: refs/heads/master@{#59024}

src/arm64/assembler-arm64.cc

@@ -621,6 +621,10 @@ void Assembler::Align(int m) {
   }
 }
 
+void Assembler::CodeTargetAlign() {
+  // Preferred alignment of jump targets on some ARM chips.
+  Align(8);
+}
 
 void Assembler::CheckLabelLinkChain(Label const * label) {
 #ifdef DEBUG

src/arm64/assembler-arm64.h

@@ -287,6 +287,8 @@ class V8_EXPORT_PRIVATE Assembler : public AssemblerBase {
   // Insert the smallest number of zero bytes possible to align the pc offset
   // to a mulitple of m. m must be a power of 2 (>= 2).
   void DataAlign(int m);
+  // Aligns code to something that's optimal for a jump target for the platform.
+  void CodeTargetAlign();
 
   inline void Unreachable();
 

src/compiler/backend/code-generator.cc

@@ -191,9 +191,9 @@ void CodeGenerator::AssembleCode() {
 
   // Assemble instructions in assembly order.
   for (const InstructionBlock* block : code()->ao_blocks()) {
-    // Align loop headers on 16-byte boundaries.
+    // Align loop headers on vendor recommended boundaries.
     if (block->ShouldAlign() && !tasm()->jump_optimization_info()) {
-      tasm()->Align(16);
+      tasm()->CodeTargetAlign();
     }
     if (info->trace_turbo_json_enabled()) {
       block_starts_[block->rpo_number().ToInt()] = tasm()->pc_offset();