Yes - shifting the 32 bit is probably smaller (and gives a more stable timing if you need that). Maybe try this:

#include <8052.h>

unsigned long f ()
{
  unsigned long t; /* no need to clear */
  unsigned char i = 32;

  do
  {
     t <<= 1;
     if (P0_1)
        t |= 0x00000001;
  } while (--i);

  return t;
}

passed over "sdcc -c 32bit_shift.c" gives:
<PRE>
330 ;Allocation info for local variables in function 'f'
331 ;------------------------------------------------------------
332 ;t Allocated with name '_f_t_1_1'
333 ;i Allocated to registers r2
334 ;------------------------------------------------------------
335 ; 32bit_shift.c:3: unsigned long f ()
336 ; -----------------------------------------

0000 339 _f:
348 ; 32bit_shift.c:8: do
0000 7A 20 349 mov r2,#0x20
0002 350 00103$:
351 ; 32bit_shift.c:10: t <<= 1;
0002 E5*00 352 mov a,_f_t_1_1
0004 25*00 353 add a,_f_t_1_1
0006 F5*00 354 mov _f_t_1_1,a
0008 E5*01 355 mov a,(_f_t_1_1 + 1)
000A 33 356 rlc a
000B F5*01 357 mov (_f_t_1_1 + 1),a
000D E5*02 358 mov a,(_f_t_1_1 + 2)
000F 33 359 rlc a
0010 F5*02 360 mov (_f_t_1_1 + 2),a
0012 E5*03 361 mov a,(_f_t_1_1 + 3)
0014 33 362 rlc a
0015 F5*03 363 mov (_f_t_1_1 + 3),a
364 ; 32bit_shift.c:11: if (P0_1)
0017 30 81 03 365 jnb _P0_1,00104$
366 ; 32bit_shift.c:12: t |= 0x00000001;
001A 43r00 01 367 orl _f_t_1_1,#0x01
001D 368 00104$:
369 ; 32bit_shift.c:13: } while (--i);
001D DA E3 370 djnz r2,00103$
371 ; 32bit_shift.c:15: return t;


001F 85*00 82 372 mov dpl,_f_t_1_1
0022 85*01 83 373 mov dph,(_f_t_1_1 + 1)
0025 85*02 F0 374 mov b,(_f_t_1_1 + 2)
0028 E5*03 375 mov a,(_f_t_1_1 + 3)
002A 22 376 ret


Good luck with your project!)

Frieder