the global variable in the ISR, because the loop stuff was definitely taking longer than 50 cycles. So it is guaranteed that the ADC will be done before the loop body completes, at least until I optimize the loop. The included ASM is what SDCC generated. It can be greatly optimized. Rather than recalculate the effective address of s[28+y], s[29+y], etc. I will just put inline asm "inc dptr" instructions, effectively building on the previously calculated address of the previous loop iteration. But I have to first figure out the problem with the ADC. Thanks for your help.

for (y = 0; y < 224; y += 4) {
   ADCON = 8;
   s[28+y] = TH0;
   s[29+y] = TL0;
   s[30+y] = sadch >> 6;
   s[31+y] = (sadch << 2) | (sadcon >> 6);
   sadch = ADCH;
   sadcon = ADCON;
}

void adc_intr() interrupt 10 _naked using 0 {
   _asm
      push acc
   _endasm;
   ADCON &= 0xc0;
   _asm
      pop acc
   _endasm;
}

;udp_adc.c:35: for (y = 0; y < 224; y += 4) {
;     genAssign
        mov     _y,#0x00
00118$:
;     genCmpLt
;     genCmp
        clr     c
        mov     a,_y
        subb    a,#0xE0
;     genIfxJump
;       Peephole 108    removed ljmp by inverse jump logic
        jnc     00121$
00144$:
;udp_adc.c:36: ADCON = x;
;     genAssign
        mov     _ADCON,_x
;udp_adc.c:37: s[28+y] = TH0;
;     genPlus
        mov     a,#0x1C
        add     a,_y
;     genPlus
        add     a,#_s
        mov     dpl,a
;       Peephole 240    use clr instead of addc a,#0
        clr     a
        addc    a,#(_s >> 8)
        mov     dph,a
;     genPointerSet
;     genFarPointerSet
        mov     a,_TH0
        movx    @dptr,a
;udp_adc.c:38: s[29+y] = TL0;
;     genPlus
        mov     a,#0x1D
        add     a,_y
;     genPlus
        add     a,#_s
        mov     dpl,a
;       Peephole 240    use clr instead of addc a,#0
        clr     a
        addc    a,#(_s >> 8)
        mov     dph,a
;     genPointerSet
;     genFarPointerSet
        mov     a,_TL0
        movx    @dptr,a
;udp_adc.c:39: s[30+y] = sadch >> 6;
;     genPlus
        mov     a,#0x1E
        add     a,_y
;     genPlus
        add     a,#_s
        add     a,#_s
        mov     dpl,a
;       Peephole 240    use clr instead of addc a,#0
        clr     a
        addc    a,#(_s >> 8)
        mov     dph,a
;     genRightShift
;     genRightShiftLiteral
;     genrshOne
        mov     a,_sadch
        rl      a
        rl      a
        anl     a,#0x03
;     genPointerSet
;     genFarPointerSet
;       Peephole 105    removed redundant mov
        mov     r2,a
        movx    @dptr,a
;udp_adc.c:40: s[31+y] = (sadch << 2) | (sadcon >> 6);
;     genPlus
        mov     a,#0x1F
        add     a,_y
;     genPlus
        add     a,#_s
        mov     dpl,a
;       Peephole 240    use clr instead of addc a,#0
        clr     a
        addc    a,#(_s >> 8)
        mov     dph,a
;     genLeftShift
;     genLeftShiftLiteral
;     genlshOne
        mov     a,_sadch
        add     a,acc
        add     a,acc
        mov     r2,a
;     genRightShift
;     genRightShiftLiteral
;     genrshOne
        mov     a,_sadcon
        rl      a
        rl      a
        anl     a,#0x03
;     genOr
;       Peephole 105    removed redundant mov
        mov     r3,a
        orl     ar2,a
;     genPointerSet
;     genFarPointerSet
        mov     a,r2
        movx    @dptr,a
;udp_adc.c:41: sadch = ADCH;
;     genAssign
        mov     _sadch,_ADCH
;udp_adc.c:42: sadcon = ADCON;
;     genAssign
        mov     _sadcon,_ADCON
;udp_adc.c:35: for (y = 0; y < 224; y += 4) {
;     genPlus
;     genPlusIncr
        mov     a,_y
        add     a,#0x04
        mov     _y,a
;       Peephole 112.b  changed ljmp to sjmp
        sjmp    00118$
00121$:
;udp_adc.c:44: send_udp(224);
;     genCall
;       Peephole 182.b  used 16 bit load of dptr
        mov     dptr,#0x00E0
        lcall   _send_udp
        ljmp    00101$
00122$:
        ret