pike.git/ src/ code/ amd64.c

2014-08-31

2014-08-31 15:55:36 by Per Hedbor <ph@opera.com>

• 0eb13c31e72f55bfcab741b63469f7f25d734887 (21 lines) (+17/-4) ^{[ Show | Annotate ]}
  Branch: 8.0

Added x86_64 support for F_MOD

The divide instruciton also does mod.
But for negative numbers pike is somewhat different.
So, be rather restrictive.

457:    xor_reg_reg( reg, reg );   }    - #if 0 +    static void neg_reg( enum amd64_reg reg )   {    rex(1,0,0,reg);    opcode(0xf7);    modrm(3,3,reg);   } - #endif +       static void mov_imm_reg( long imm, enum amd64_reg reg )   {
1951:    return;       case F_DIVIDE: +  case F_MOD:    {    LABELS();    if_not_two_int(&label_A,0);    mov_mem_reg(sp_reg, SVAL(-1).value, P_REG_RBX );    mov_mem_reg(sp_reg, SVAL(-2).value, P_REG_RAX ); -  +  if( b+F_OFFSET == F_MOD ) +  { +  cmp_reg_imm(P_REG_RAX, 0 ); +  /* FIXME: Emulate pikes % functionality. */ +  jl(&label_A); +  }    cmp_reg_imm(P_REG_RBX,0);    je(&label_A);    cmp_reg_imm(P_REG_RBX,-1);
1966:       Int.native_min / -1 -> exception.    +  This checks for -<whatever> / -1. +     There is bound to be a better way.    Somehow disable the exception?    -  Some SSE alternative? +  Some SSE alternative that does not throw? +  +  Perhaps a 64/64->64 instead of the current 128/64->64?    */    LABEL_C;    div_reg_reg(P_REG_RAX,P_REG_RBX);    jo(&label_A);    mov_imm_mem(PIKE_T_INT, sp_reg, SVAL(-2).type); -  +  if( b+F_OFFSET == F_MOD ) +  mov_reg_mem(P_REG_RDX, sp_reg, SVAL(-2).value); +  else    mov_reg_mem(P_REG_RAX, sp_reg, SVAL(-2).value);    amd64_add_sp(-1);    jmp(&label_B);