2018-05-29----
 ❯ lldb -- /Users/anatofuz/workspace/cr/Basic/build_perl6/bin/moar nqp.moarvm examples/hello_world.nqp                                                                                                                                                               [13:52:54]
(lldb) target create "/Users/anatofuz/workspace/cr/Basic/build_perl6/bin/moar"
Current executable set to '/Users/anatofuz/workspace/cr/Basic/build_perl6/bin/moar' (x86_64).
(lldb) settings set -- target.run-args  "nqp.moarvm" "examples/hello_world.nqp"
(lldb) b add_bb_facts
Breakpoint 1: where = libmoar.dylib`add_bb_facts + 32 at facts.c:362, address = 0x0000000000118540
(lldb) c
error: invalid process
(lldb) run
Process 8479 launched: '/Users/anatofuz/workspace/cr/Basic/build_perl6/bin/moar' (x86_64)
Process 8479 stopped
* thread #2, stop reason = breakpoint 1.1
    frame #0: 0x00000001001bb540 libmoar.dylib`add_bb_facts(tc=0x0000000100802e30, g=0x0000000100f4efe0, bb=0x0000000102115e00, p=0x0000000100f514d0, cur_deopt_idx=-1) at facts.c:362
   359      MVMint32 i, is_phi;
   360
   361      /* Look for instructions that provide or propagate facts. */
-> 362      MVMSpeshIns *ins = bb->first_ins;
   363      while (ins) {
   364          /* See if there's deopt and logged annotations. Sync cur_deopt_idx
   365           * and, for logged+deopt-one, add logged facts and guards. */
Target 0: (moar) stopped.
(lldb) bt
* thread #2, stop reason = breakpoint 1.1
  * frame #0: 0x00000001001bb540 libmoar.dylib`add_bb_facts(tc=0x0000000100802e30, g=0x0000000100f4efe0, bb=0x0000000102115e00, p=0x0000000100f514d0, cur_deopt_idx=-1) at facts.c:362
    frame #1: 0x00000001001bb503 libmoar.dylib`MVM_spesh_facts_discover(tc=0x0000000100802e30, g=0x0000000100f4efe0, p=0x0000000100f514d0) at facts.c:659
    frame #2: 0x00000001001b4eb7 libmoar.dylib`MVM_spesh_candidate_add(tc=0x0000000100802e30, p=0x0000000100f514d0) at candidate.c:61
    frame #3: 0x00000001001cf991 libmoar.dylib`worker(tc=0x0000000100802e30, callsite=0x00000001006c9150, args=0x0000000000000000) at worker.c:16
    frame #4: 0x000000010014e8e2 libmoar.dylib`invoke_handler(tc=0x0000000100802e30, invokee=0x0000000102014840, callsite=0x00000001006c9150, args=0x0000000000000000) at MVMCFunction.c:9
    frame #5: 0x00000001000e8494 libmoar.dylib`thread_initial_invoke(tc=0x0000000100802e30, data=0x0000000100802050) at threads.c:59
    frame #6: 0x00000001000aefee libmoar.dylib`MVM_interp_run(tc=0x0000000100802e30, initial_invoke=(libmoar.dylib`thread_initial_invoke at threads.c:50), invoke_data=0x0000000100802050) at interp.c:93
    frame #7: 0x00000001000e7a35 libmoar.dylib`start_thread(data=0x0000000100802050) at threads.c:87
    frame #8: 0x00007fff7b9fe661 libsystem_pthread.dylib`_pthread_body + 340
    frame #9: 0x00007fff7b9fe50d libsystem_pthread.dylib`_pthread_start + 377
    frame #10: 0x00007fff7b9fdbf9 libsystem_pthread.dylib`thread_start + 13

(lldb) n
Process 8479 stopped
* thread #2, stop reason = step over
    frame #0: 0x00000001001bb579 libmoar.dylib`add_bb_facts(tc=0x0000000100802e30, g=0x0000000100f4efe0, bb=0x0000000102115e00, p=0x0000000100f514d0, cur_deopt_idx=-1) at facts.c:370
   367          MVMSpeshAnn *ann_deopt_one = NULL;
   368          MVMSpeshAnn *ann_logged = NULL;
   369          MVMint32 is_deopt_ins = 0;
-> 370          while (ann) {
   371              switch (ann->type) {
   372                  case MVM_SPESH_ANN_DEOPT_ONE_INS:
   373                      ann_deopt_one = ann;
Target 0: (moar) stopped.
(lldb)
Process 8479 stopped
* thread #2, stop reason = step over
    frame #0: 0x00000001001bb620 libmoar.dylib`add_bb_facts(tc=0x0000000100802e30, g=0x0000000100f4efe0, bb=0x0000000102115e00, p=0x0000000100f514d0, cur_deopt_idx=-1) at facts.c:385
   382              }
   383              ann = ann->next;
   384          }
-> 385          if (ann_deopt_one && ann_logged)
   386              log_facts(tc, g, bb, ins, p, ann_deopt_one, ann_logged);
   387
   388          /* Look through operands for reads and writes. */
Target 0: (moar) stopped.
(lldb) l
   389          is_phi = ins->info->opcode == MVM_SSA_PHI;
   390          for (i = 0; i < ins->info->num_operands; i++) {
   391              /* Reads need usage tracking; if the read is after a deopt point
   392               * relative to the write then give it an extra usage bump. */
   393              if ((is_phi && i > 0)
   394                  || (!is_phi && (ins->info->operands[i] & MVM_operand_rw_mask) == MVM_operand_read_reg)) {
   395                  MVMSpeshFacts *facts = &(g->facts[ins->operands[i].reg.orig][ins->operands[i].reg.i]);
(lldb) l
   396                  facts->usages += facts->deopt_idx == cur_deopt_idx ? 1 : 2;
   397              }
   398
   399              /* Writes need the current deopt index and the writing instruction
   400               * to be specified. A write that's on a deopt instruction bumps
   401               * the usage too. */
   402              if ((is_phi && i == 0)
(lldb) n
Process 8479 stopped
* thread #2, stop reason = step over
    frame #0: 0x00000001001bb65b libmoar.dylib`add_bb_facts(tc=0x0000000100802e30, g=0x0000000100f4efe0, bb=0x0000000102115e00, p=0x0000000100f514d0, cur_deopt_idx=-1) at facts.c:389
   386              log_facts(tc, g, bb, ins, p, ann_deopt_one, ann_logged);
   387
   388          /* Look through operands for reads and writes. */
-> 389          is_phi = ins->info->opcode == MVM_SSA_PHI;
   390          for (i = 0; i < ins->info->num_operands; i++) {
   391              /* Reads need usage tracking; if the read is after a deopt point
   392               * relative to the write then give it an extra usage bump. */
Target 0: (moar) stopped.
(lldb) p MVM_SSA_PHI
error: use of undeclared identifier 'MVM_SSA_PHI'
(lldb) nexr
error: 'nexr' is not a valid command.
error: Unrecognized command 'nexr'.
(lldb) n
Process 8479 stopped
* thread #2, stop reason = step over
    frame #0: 0x00000001001bb677 libmoar.dylib`add_bb_facts(tc=0x0000000100802e30, g=0x0000000100f4efe0, bb=0x0000000102115e00, p=0x0000000100f514d0, cur_deopt_idx=-1) at facts.c:390
   387
   388          /* Look through operands for reads and writes. */
   389          is_phi = ins->info->opcode == MVM_SSA_PHI;
-> 390          for (i = 0; i < ins->info->num_operands; i++) {
   391              /* Reads need usage tracking; if the read is after a deopt point
   392               * relative to the write then give it an extra usage bump. */
   393              if ((is_phi && i > 0)
Target 0: (moar) stopped.
(lldb) p is_phi
(MVMint32) $14 = 0
(lldb) p ins
(MVMSpeshIns *) $15 = 0x0000000102115e60
(lldb) p *ins
(MVMSpeshIns) $16 = {
  info = 0x00000001005757c0
  operands = 0x0000000000000000
  prev = 0x0000000000000000
  next = 0x0000000000000000
  annotations = 0x0000000000000000
}
(lldb) p *ins->info
(MVMOpInfo) $17 = {
  opcode = 0
  name = 0x00000001002def92 "no_op"
  mark = {
    [0] = ' '
    [1] = ' '
  }
  num_operands = 0
  pure = '\0'
  deopt_point = '\0'
  logged = '\0'
  no_inline = '\0'
  jittivity = '\0'
  uses_hll = '\0'
  operands = ([0] = '\0', [1] = '\0', [2] = '\0', [3] = '\0', [4] = '\0', [5] = '\0', [6] = '\0', [7] = '\0')
}


* profilingの箇所を見ていく

* perlスクリプトをperl6に移植する
* markdownのparserみたいなのを書き換えてみるなど
* タイトループを早くするのか，ここのハッシュを早くするのか，オブジェクトのオペレーションを早くするのか…
* 手近な場所をとにかく見ていきたい

* native codeをやるのかvmをやるのか

* bytecode interpureter, gc関連を書き直すという手もある

* gcc vs clang

* llvm側でtail coll呼び出す時はフラグを見ている
    * code grarであるというフラグを渡している感じ

* script言語--> 上手く並列化出来ない

* targetとしてCbCのコードを吐く
    * 継続コード(meta level)
    * 継続コードを見て計算結果を見て変えていく
        * MoarVMのJIT，nqpのコンパイラにcbcを生成する部分を入れる
    * perlccみたいなものをいれる
        * targetのアプリケーション...whileでstring matchをする部分を最適化する

* 正規表現の箇所をターゲットとして書くのも手…!?
    * fileをmmapを使うかどうかなど
* 並列のgrep
    * 早くなるには早くなる
    * cache prefetchの処理でも高速化出来る


----------
2018-06-05----
MoarVMが実際にインタプリタとして走っているのはMVM_interp_runという関数らしい．
これはスレッド毎に用意されるらしい

定義自体は `src/core/interp.c`に書かれている

 MVMInstancとかいう構造体に全ての情報が入っている

 MVM_cu_map_from_fileでスレッド情報とファイルから生成した結果をMVMCompUnit構造体として生成している

 DISPATCHでNEXT_OPを判断している，これはオペランドリストとswitch文が同じオーダーで書かれているので
 最適化しやすいらしい

 MVM_interp_runの後DISPATCHというマクロでNEXT_OPというマクロに応じて処理を行っているが
 これがMoarVMのバイトコードを一対一対応しているので5000行のcase文が生成されている


----------