R12+0x40=IISFIFO_WR
R9=p (pointer to sample)
R10=working register

         "ldr r10, [r9], #4    \n\t" /* load two samples */
         "mov r10, r10, ror #16\n\t" /* put left sample at the top bits */
         "str r10, [r12, #0x40]\n\t" /* write top sample, lower sample ignored */
         "mov r10, r10, lsl #16\n\t" /* shift lower sample up */
         "str r10, [r12, #0x40]\n\t" /* then write it */