#define VECTREX 1 // Although I have another file with generic divide code, this is specific to dividing by 10 // which is frequently needed when displaying numbers on screen, and worth optimizing. // That said, I have not yet put this code on the clock to confirm timings. // This file contains a 16-bit signed divide by 10, // and 8-bit signed and unsigned divides by 10 (with two different // implementations of each - time them and use the faster ones) #ifdef VECTREX #include <vectrex.h> #include "muldiv.h" // high is 0 on intel, 1 on 6809 #define HIGH 0 #define LOW 1 #define int8_t int #define uint8_t unsigned int #define int16_t long #define uint16_t unsigned long #define USE_MUL 1 #define EXIT_SUCCESS 0 #define EXIT_FAILURE 1 #else #include <stdio.h> #include <stdlib.h> // high is 0 on intel, 1 on 6809 #define HIGH 0 #define LOW 1 #define int8_t signed char #define uint8_t unsigned char #define int16_t signed short #define uint16_t unsigned short #define USE_MUL 1 #define TEST #endif // If we read and write our parameter and result from a volatile pointer, // we can ensure the compiler does not optimise expressions out of existence // when the input is known at compile time! static long long P1forced_access; static volatile long long *P1LongLong = (volatile long long *)&P1forced_access; static volatile long *P1Long = (volatile long *)&P1forced_access; static volatile int *P1Int = (volatile int *)&P1forced_access; static long long P2forced_access; static volatile long long *P2LongLong = (volatile long long *)&P2forced_access; static volatile long *P2Long = (volatile long *)&P2forced_access; static volatile int *P2Int = (volatile int *)&P2forced_access; static void SHOW_NUM(char *dest, long num) { // left-aligned #define show_digit(x) *dest++ = (char)((x)+'0') int8_t digit, zeroes; // This replaces code that used divide by 10 and modulo 10. Much faster. // handles full 16 bit range of -32768:32767 - Uses negative numbers to avoid the issue of negating -32768 if (num >= 0) num = -num; else *dest++ = '-'; digit = 0; zeroes = 1; // CLRing is shorter // max 11 add/subtracts... if (num <= -20000) { num += 20000; digit += 2; zeroes = 0; } if (num <= -10000) { num += 10000; digit += 1; zeroes = 0; } if (!zeroes) show_digit(digit); digit = 0; if (num <= -8000) { num += 8000; digit += 8; zeroes = 0; } else if (num <= -4000) { num += 4000; digit += 4; zeroes = 0; } if (num <= -2000) { num += 2000; digit += 2; zeroes = 0; } if (num <= -1000) { num += 1000; digit += 1; zeroes = 0; } if (!zeroes) show_digit(digit); digit = 0; if (num <= -800) { num += 800; digit += 8; zeroes = 0; } else if (num <= -400) { num += 400; digit += 4; zeroes = 0; } if (num <= -200) { num += 200; digit += 2; zeroes = 0; } if (num <= -100) { num += 100; digit += 1; zeroes = 0; } if (!zeroes) show_digit(digit); digit = 0; if (num <= -80) { num += 80; digit += 8; zeroes = 0; } else if (num <= -40) { num += 40; digit += 4; zeroes = 0; } if (num <= -20) { num += 20; digit += 2; zeroes = 0; } if (num <= -10) { num += 10; digit += 1; zeroes = 0; } if (!zeroes) show_digit(digit); show_digit((int8_t)-num); *dest++ = ' '; *dest = 0x80; } // An accurate signed 16 bit division by 10 // Taken from http://homepage.divms.uiowa.edu/~jones/bcd/divide.html static inline int16_t sdiv10_16(int16_t A) { int16_t Q; /* the quotient */ int16_t R; /* the remainder */ // Avoid using divide: Q = ((A >> 1) + A) >> 1; /* Q = A*0.11 */ Q = ((Q >> 4) + Q) ; /* Q = A*0.110011 */ Q = ((Q >> 8) + Q) >> 3; /* Q = A*0.00011001100110011 */ /* either Q = A/10 or Q+1 = A/10 for -87,381 < A < 534,890 */ #ifdef USE_MUL // Need to time both cases on 6809 to see which is faster. R = Q * 10; #else R = ((Q << 2) + Q) << 1; #endif R = A - R; /* R = A - 10*Q */ if (R >= 10) { R -= 10; Q += 1; } /* Q = A/10 for -87,381 < A < 534,890 */ if ((Q < 0) && (R != 0)) Q += 1; #ifdef TEST if (Q != A/10) { fprintf(stderr, "A: %d A/10: %d Q: %d\n", A, A/10, Q); exit(EXIT_FAILURE); } #endif return Q; } // ditto with remainder exported to user static inline int16_t sdiv10_16_with_remainder(int16_t A, int16_t *Remainder) { int16_t Q; /* the quotient */ int16_t R; /* the remainder */ // Avoid using divide: Q = ((A >> 1) + A) >> 1; /* Q = A*0.11 */ Q = ((Q >> 4) + Q) ; /* Q = A*0.110011 */ Q = ((Q >> 8) + Q) >> 3; /* Q = A*0.00011001100110011 */ /* either Q = A/10 or Q+1 = A/10 for -87,381 < A < 534,890 */ #ifdef USE_MUL // Need to time both cases on 6809 to see which is faster. R = Q * 10; #else R = ((Q << 2) + Q) << 1; #endif R = -R + A; /* R = A - 10*Q */ if (R >= 10) { R -= 10; Q += 1; } /* Q = A/10 for -87,381 < A < 534,890 */ if ((Q < 0) && (R != 0)) { R -= 10; Q += 1; } *Remainder = R; #ifdef TEST if ((Q != A/10) || (R != A%10)) { fprintf(stderr, "A: %d A/10: %d Q: %d R: %d A%%10: %d\n", A, A/10, Q, R, A%10); exit(EXIT_FAILURE); } #endif return Q; } // divide signed 8-bit quantity by 10 using 16 bit intermediate at most static inline int8_t s8divided_by_10(int8_t x) { int8_t rem; int8_t result; result = (int8_t)(((int16_t)x * 25L) >> 8L); // A*B should be the only 16 bit quantity #ifdef USE_MUL //rem = -(result*10)+x; // *10 is marginally faster than shift+add. rem = (-10*result)+x; // yet to check if r*-10 faster than -(r*10). Probably is. #else rem = -(((result << 2) + result) << 1) + x; #endif if (rem > 0) { if ((x < 0) || (rem >= 10)) result++; } #ifdef TEST if (result != x/10) { printf("s: %d %d %d rem %d\n", x, x/10, result, rem); exit(EXIT_FAILURE); } #endif return result; } // divide unsigned 8-bit quantity by 10 using 16 bit intermediate at most static inline uint8_t u8divided_by_10(uint8_t x) { uint8_t rem; uint8_t result; result = (uint8_t)(((uint16_t)x * 25UL) >> 8UL); // Again, A*B should be the only 16 bit quantity rem = x - result*10; // yet to check if *10 faster than shift+add. Probably is. if (rem >= 10) result++; #ifdef TEST if (result != x/10U) { printf("u: %u %u %u rem %u\n", x, x/10U,result,rem); exit(EXIT_FAILURE); } #endif return result; } static inline uint8_t divu10_8(uint8_t i) { uint8_t result, remainder; union twobytes { // an experiment to see if the 6809 code generator can be forced to avoid using 8 double-byte LSR instructions uint16_t sixteenbit; // Despite being lengthier source code this should be tighter machine code. uint8_t eightbit[2]; } temp; temp.sixteenbit = (i+1)*51; result = temp.eightbit[LOW]>>1; remainder = i - result*10; #ifdef TEST if (result != i/10) { printf("U: %u/10 = %u? %u rem %u\n", i, i/10, result, i - result*10); exit(EXIT_FAILURE); } #endif return result; } static inline int8_t divs10_8(int8_t i) { int8_t result; union twobytes { // an experiment to see if the 6809 code generator can be forced to avoid using 8 double-byte ASR instructions int16_t sixteenbit; int8_t eightbit[2]; } temp; temp.sixteenbit = (i+1)*51; result = temp.eightbit[LOW]>>1; if ((i < 0) && (i > result*10)) result += 1; // there's probably a bit-twiddling expression to avoid two comparisons... // (but just casting i to uint8_t won't work) #ifdef TEST if (result != i/10) { printf("S: %d/10 = %d? %d rem %d\n", i, i/10, result, i - result*10); exit(EXIT_FAILURE); } #endif return result; } #define IRQ_6809_MASK 0x10 #define ENABLE_IRQ_6809 asm("andcc #0xef") #define DISABLE_IRQ_6809 asm("orcc #0x10") //static int Ready = 0; // does not return, since we use the // actual waitRecal "shortcut" to jump to "set_refresh" and return from there __attribute__((noinline)) void iWaitRecal(void) // if inlined, the JMP -> RTS goes astray!!! { asm("jmp 0xF1A2"); // Set_Refresh } volatile unsigned int8_t *rand = (volatile unsigned int *)0xc87b; volatile unsigned int8_t *timer_lo = (volatile unsigned int *)0xD004; static unsigned int16_t udiv16(unsigned int16_t qq, unsigned int16_t d) { unsigned int32_t x, q=qq; unsigned int16_t b, res, t; t = msb16(d); //fprintf(stdout, "q: %d t: %d d: %d\n", q, t, d); if (t != d) t <<= 1; // t >= d //fprintf(stdout, "t: %d >= d: %d\n", t, d); b = 1; while (t < q) { //fprintf(stdout, "t: %d <= q: %d b: %d\n", t, q, b); t <<= 1; b <<= 1; if (t == 0) break; } //fprintf(stdout, "... t: %d q: %d b: %d\n", t, q, b); res = b; for (;;) { x = d*res; // assumes 16 bit quotient was formed as the product of 2 8-bit factors. If not, use umultiply16. //fprintf(stdout, "trying %d * %d = %d against %d\n", d, res, x, q); b >>= 1; if (x == q) { //fprintf(stdout, "%d / %d = %d (actual %d)\n", q, d, res, q / d); return res; } else if ((x <= q) && (x+d > q)) { //fprintf(stdout, "%d / %d = %d rem %d (actual %d rem %d)\n", q, d, res, q-x, q / d, q % d); return res; } else if (x < q) { res += b; } else if (x > q) { res -= b; } } } int main(void) { static unsigned long t1,t2; static char debug[12]; Vec_Rfrsh = 0L; Vec_Loop_Count = 0L; for (;;) { Wait_Recal(); Reset0Ref(); Intensity_a(0x3f); { asm(" nop "); // initialisation: *P1Long = 12345L; *P2Long = 10L; long p1 = *(long *)P1Long; // use of volatile forces code to be executed long p2 = *(long *)P2Long; // insert code to be tested here: t1 = dp_VIA_t2; //======================================================================================= //asm(" nop "); // 4 cycles (not 3 like https://atjs.mbnet.fi/mc6809/Information/6809.htm ) //asm(" adda #0 "); // 4 cycles //p1 = sdiv16_by_16(p1,p2); // 17626 cycles //p1 = sdiv16_by_8(p1,(int)p2); // 17634 cycles //p1 = (long)udiv16_by_16((unsigned long)p1,(unsigned long)p2); // 17180 cycles //p1 = (long)udiv16_by_8((unsigned long)p1,10); // 17138 cycles //p1 = p1/10L; // 2364 cycles //p1 = p1/p2; // 2360 cycles //p1 = sdiv10_16(p1); // 352 cycles p1 = (long)udiv16((unsigned long)p1,(unsigned long)p2); // 9760 cycles //======================================================================================= t2 = dp_VIA_t2; t1 = (t1<<8) | ((t1>>8)&255); t2 = (t2<<8) | ((t2>>8)&255); *P1Long = p1; // force results to be calculated and not discarded *P2Long = p2; } SHOW_NUM(debug, (long)((t1-t2-13L)<<1)); // cycle count of operation above Print_Str_d(80, -40, debug); //SHOW_NUM(debug, (long)t1); // display the result as a check! SHOW_NUM(debug, (long int)*P1Long); // display the result as a check! Print_Str_d(40, -40, debug); continue; (void)(*P1LongLong); (void)(*P1Long); (void)(*P1Int); (void)(*P2LongLong); (void)(*P2Long); (void)(*P2Int); } #ifdef NEVER { // signed 16-bit divide int16_t Q,R,A = 0; do Q = sdiv10_16_with_remainder(A++,&R); while (A); do Q = sdiv10_16(A++); while (A); } { // unsigned 8-bit divide uint8_t i, r2, r5; i = 0; for (;;) { if (i) { r2 = u8divided_by_10(i); r5 = divu10_8(i); } i++; if (i == 0) break; } } { // signed 8-bit divide int8_t i, r1, r2, r5; i = -128; for (;;) { if (i) { r1 = i/10; r2 = s8divided_by_10(i); r5 = divs10_8(i); } i++; if (i == -128) break; } } #ifdef TEST printf("All test passed\n"); exit(EXIT_SUCCESS); #endif #endif return (EXIT_FAILURE); }