pike.git/ src/ code/ amd64.c

2012-07-16

2012-07-16 21:09:35 by Per Hedbor <ph@opera.com>

• 638a2c5cc0e5fdc8de324094be1e45b07de2377b (96 lines) (+65/-31) ^{[ Show | Annotate ]}
  Branch: 7.9

[compiler][amd64] Use new features from peep.c

Place the code for calling check_threads_etc before the function instead of
inside it, to have one branch less in tight loops.

This saves about 4% in the nested loops test, at the cost of 12 bytes extra
code-space for functions that do not actually contain loops (for functions
that contain loops 3 bytes is saved intead)

One alternative would have to place the code after the function, if it does
contain a loop, then update the relative jumps to point to the code.

That is left for later.

52:   static enum amd64_reg sp_reg = -1, fp_reg = -1, mark_sp_reg = -1;   static int dirty_regs = 0, ret_for_func = 0;   ptrdiff_t amd64_prev_stored_pc = -1; /* PROG_PC at the last point Pike_fp->pc was updated. */ - static int branch_check_threads_update_etc = -1; + static int branch_check_threads_update_etc = 0; + static int store_pc = 0;    -  +    #define MAX_LABEL_USES 6   struct label {    int n_label_uses;
875:   #define LABEL_D label(&label_D)   #define LABEL_E label(&label_E)    + static void amd64_ins_branch_check_threads_etc(int code_only); + static int func_start = 0; +  + /* Called at the start of a function. */ + void amd64_start_function(int store_lines ) + { +  store_pc = store_lines; +  branch_check_threads_update_etc = 0; +  /* +  * store_lines is true for any non-constant evaluation function +  * In reality we should only do this if we are going to be using +  * check_threads_etc (ie, if we have a loop). +  * +  * It does not really waste all that much space in the code, though. +  * +  * We could really do this one time per program instead of once per +  * function. +  * +  * But it is very hard to know when a new program starts, and +  * PIKE_PC is not reliable as a marker since constant evaluations +  * are also done using this code. So for now, waste ~20 bytes per +  * function in the program code. +  */ +  if( store_lines ) +  amd64_ins_branch_check_threads_etc(1); +  func_start = PIKE_PC; + } +  + /* Called when the current function is done */ + void amd64_end_function(int no_pc) + { +  branch_check_threads_update_etc = 0; + } +  +    /* Machine code entry prologue.    *    * On entry:
894:    push(REG_R12);    push(REG_RBX);    sub_reg_imm(REG_RSP, 8); /* Align on 16 bytes. */ -  +     mov_reg_reg(ARG1_REG, Pike_interpreter_reg); -  +     amd64_flush_code_generator_state();   }   
904:   {    sp_reg = -1;    fp_reg = -1; +  ret_for_func = 0;    mark_sp_reg = -1;    dirty_regs = 0; -  ret_for_func = 0; +     amd64_prev_stored_pc = -1; -  branch_check_threads_update_etc = -1; +    }      static void flush_dirty_regs(void)
1246:   {    static int last_prog_id=-1;    static size_t last_num_linenumbers=-1; +  if( !store_pc ) return;       if(   #ifdef PIKE_DEBUG
1266:   {    static int last_prog_id=-1;    static size_t last_num_linenumbers=-1; +  if( !store_pc ) return;       if(   #ifdef PIKE_DEBUG
1347:    amd64_add_sp( 1 );   }    - void amd64_ins_branch_check_threads_etc() + static void amd64_align()   { -  +  while( PIKE_PC & 3 ) +  ib( 0x90 ); + } +  + static void amd64_ins_branch_check_threads_etc(int code_only) + {    LABELS();    - #if 1 -  if( branch_check_threads_update_etc == -1 ) +  if( !branch_check_threads_update_etc )    { -  /* Create update + call to branch_checl_threds_etc */ +  /* Create update + call to branch_check_threds_etc */ +  if( !code_only )    jmp( &label_A ); -  mov_imm_mem32( 0, REG_RSP, 0); +     branch_check_threads_update_etc = PIKE_PC;    if( (unsigned long long)&fast_check_threads_counter < 0x7fffffffULL )    {
1373: Inside #if 1
   }    mov_imm_reg( (ptrdiff_t)branch_check_threads_etc, REG_RAX );    jmp_reg(REG_RAX); /* ret in BCTE will return to desired point. */ +  amd64_align();    } -  +  if( !code_only ) +  {    LABEL_A;    /* Use C-stack for counter. We have padding added in entry */    add_mem8_imm( REG_RSP, 0, 1 );    jno( &label_B );    call_rel_imm32( branch_check_threads_update_etc-PIKE_PC );    LABEL_B; - #else -  call_imm( &branch_check_threads_etc ); - #endif +     } -  + }         void amd64_init_interpreter_state(void)
1411:    }   }    - static void amd64_align() - { -  while( PIKE_PC & 7 ) -  ib( 0x90 ); - } -  +    void ins_f_byte(unsigned int b)   {    int flags;
1975:    ins_debug_instr_prologue(off, 0, 0); \    if (backward_jump) { \    maybe_update_pc(); \ -  amd64_ins_branch_check_threads_etc(); \ +  amd64_ins_branch_check_threads_etc(0); \    } \    } while(0)   
2047:    je(&label_A);       /* increase counter */ -  add_mem_imm( sp_reg, -1*sizeof(struct svalue)+8, 1 ); +  add_mem_imm( sp_reg, -1*(int)sizeof(struct svalue)+8, 1 );       /* get item */    mov_mem_reg( REG_RBX, OFFSETOF(array,item), REG_RBX );
2214:    add_to_program (0x74); /* jz rel8 */    add_to_program (0); /* Bytes to skip. */    skip = (INT32)PIKE_PC; -  amd64_ins_branch_check_threads_etc(); -  /* amd64_call_c_function (branch_check_threads_etc); */ +  amd64_ins_branch_check_threads_etc(0);    add_to_program (0xe9); /* jmp rel32 */    ret = DO_NOT_WARN ((INT32) PIKE_PC);    PUSH_INT (0);