diff options
author | Julian T <julian@jtle.dk> | 2020-02-11 11:48:53 +0100 |
---|---|---|
committer | Julian T <julian@jtle.dk> | 2020-02-11 11:48:53 +0100 |
commit | 57305119e05559c1c37e903aef89cd43f44c42c9 (patch) | |
tree | 5df95141e99550c22710a94629dca09d7185851a /sem4/embedded/emb_m2 | |
parent | bf40ddf3a1970b857adeb070a1b4d0dbc8f3c1b6 (diff) |
Added embedded assignment and python assignments
Diffstat (limited to 'sem4/embedded/emb_m2')
-rw-r--r-- | sem4/embedded/emb_m2/emb_m2.ino | 233 | ||||
-rw-r--r-- | sem4/embedded/emb_m2/myfloat_red.h | 57 |
2 files changed, 290 insertions, 0 deletions
diff --git a/sem4/embedded/emb_m2/emb_m2.ino b/sem4/embedded/emb_m2/emb_m2.ino new file mode 100644 index 0000000..c6a6a07 --- /dev/null +++ b/sem4/embedded/emb_m2/emb_m2.ino @@ -0,0 +1,233 @@ +#include "myfloat_red.h" + +// Number of numbers +#define NN 100 +#define PRINT_PRECISION 10 + +#define NEXT_OPG(c) Serial.print("\nOpgave "); Serial.println(c++); +#define AND_OPG(c) Serial.print("and "); Serial.println(c++); + +unsigned int opg_count = 1; + +void printmydoubarr(myfloat_type *arr, size_t len) { + Serial.print("[ "); + for(size_t i = 0; i < len; i++) { + Serial.print(myfloat2double( &arr[i] ), PRINT_PRECISION); + Serial.print(" "); + } + Serial.println("]"); +} + +void printarr(double *arr, size_t len) { + Serial.print("[ "); + for(size_t i = 0; i < len; i++) { + Serial.print(arr[i], PRINT_PRECISION); + Serial.print(" "); + } + Serial.println("]"); +} + +unsigned long factorial(unsigned int n) { + unsigned long res = 1; + for(unsigned int i = 2; i <= n; i++) { + res *= i; + } + + return res; +} + +double sineTayler(double x, unsigned int n) { + double sum = 0; + + for(unsigned int i = 0; i <= n; i++) { + sum += ((double)pow(-1, i) / (double)factorial(2 * i + 1)) * (double)pow(x, 2 * i + 1); + } + + + return sum; +} + +#define TAU 1.57079632679 +// After some trial and error the best TAYLORN was 2 +#define TAYLORN 2 +double sine(double x) { + + // Works by finding value in specific -pi/2 to pi/2 period. + int period = x / TAU; + + // One cycle of sin takes 4 periods. + // If 0 just return the sine approx x + // If 1 do negative sine approx on x - TAU + // If 2 do negative sine approx x + // If 3 do sine approx on period x - TAU + x = abs(x - TAU*period); + + double res = 12; + + switch( abs(period % 4) ) { + case 0: + res = sineTayler(x, TAYLORN); + break; + case 1: + res = -sineTayler(x - TAU, TAYLORN); + break; + case 2: + res = -sineTayler(x, TAYLORN); + break; + case 3: + res = sineTayler(x - TAU, TAYLORN); + break; + } + // Remember to multiply with sign + int sign = period < 0 ? -1 : 1; + + // Should never return because all cases are handled + return res * sign; + +} + +void mean_rel_error(double *da, myfloat_type *mda, size_t len, double *sign) { + // Calculate and print the relative error using the formula + // X - x + // ----- Where x is the calculated value and X is the real value( As real as double gets ) + // X + double errorsum = 0; + Serial.print("Rel Err : [ "); + for(size_t i = 0; i < len; i++) { + //Serial.println(errorsum, PRINT_PRECISION); + // Make space for sign + if( sign[i] < 0 ) { + Serial.print(" "); + } + // Handle devide by 0 + double error = da[i] ? abs(da[i] - myfloat2double(&mda[i])) / abs(da[i]) : 0; + Serial.print(error, PRINT_PRECISION); + Serial.print(" "); + + // Save sum for mean + errorsum += error; + } + Serial.println("]"); + + // Calculate mean + double mean = errorsum / len; + Serial.print("Mean Err: "); Serial.println(mean, PRINT_PRECISION); +} + +void setup() { + Serial.begin(115200); + + NEXT_OPG(opg_count); + // OPG1. Create some doubles + double da[NN]; + for(int i = 0; i < NN; i++) { + da[i] = ((double)random(-5000, 5000)) / 1000.0; + } + Serial.print("Original: "); printarr(da, NN); + + NEXT_OPG(opg_count); + // OPG2. Convert to mda + myfloat_type mda[NN]; + for(int i = 0; i < NN; i++) { + doub2mydouble(da[i], &mda[i]); + } + + Serial.print("Mydoub : "); printmydoubarr(mda, NN); + + NEXT_OPG(opg_count); + // OPG3. Calculate the mean relative error + mean_rel_error(da, mda, NN, da); + + { + NEXT_OPG(opg_count); + // OPG4. Compute some stuff + double da2[NN]; + for(int i = 0; i < NN; i++) { + da2[i] = da[i] * da[i]; + } + Serial.print("Orig pow: "); printarr(da2, NN); + + NEXT_OPG(opg_count); + // OPG5. + myfloat_type mda2[NN]; + for(int i = 0; i < NN; i++) { + // This thing sometimes does weird stuff -4.06*(-4.06) -> -16.00 + mult_float(&mda[i], &mda[i], &mda2[i]); + } + Serial.print("Mda pow: "); printmydoubarr(mda2, NN); + + NEXT_OPG(opg_count); + // OPG6. Doing the mean error thing + mean_rel_error(da2, mda2, NN, da); + } + + { + // OPG7. Measure execution time + NEXT_OPG(opg_count); + double a = da[1]; + unsigned long begin = micros(); + for(int i = 1; i < NN; i++) { + a *= da[i]; + } + + unsigned long time1 = micros() - begin; + Serial.print("Exe time: "); Serial.println(time1); + Serial.print("a : "); Serial.println(a); + + NEXT_OPG(opg_count); + // Convert to myfloat + myfloat_type f1; + doub2mydouble(a, &f1); + + Serial.print("f1 : "); Serial.println(myfloat2double(&f1)); + + NEXT_OPG(opg_count); + // OPG8-9. + begin = micros(); + for(int i = 0; i < NN; i++) { + // Multiply + myfloat_type f; + mult_float(&f1, &mda[i], &f); + + // Copy back + memcpy(&f1, &f, sizeof(myfloat_type)); + } + + unsigned long time2 = micros() - begin; + Serial.print("Exe time: "); Serial.println(time2); + Serial.print("f1 : "); Serial.println(myfloat2double(&f1)); + Serial.println("Overflow"); + + NEXT_OPG(opg_count); + // OPG10. Difference relative to biggest result + + unsigned long absdiff = abs( time1 - time2 ); + Serial.print("Abs diff: "); Serial.println(absdiff); + double reldiff = (double)absdiff / max(abs(time1), abs(time2)); + Serial.print("Rel diff: "); Serial.println(reldiff); + } + + NEXT_OPG(opg_count); + // OPG11. Impl sine + // Hardest value should be TAU as it's the furthest from center. + unsigned long begin = micros(); + double res = sine(TAU); + unsigned long time1 = micros() - begin; + Serial.print("own sin : "); Serial.println(res, PRINT_PRECISION); + + begin = micros(); + double correct = sin(TAU); + unsigned long time2 = micros() - begin; + Serial.print("ard sin : "); Serial.println(correct, PRINT_PRECISION); + + double error = correct ? abs(correct - res) / abs(correct) : 0; + Serial.print("rel err : "); Serial.println(error, PRINT_PRECISION); + + Serial.print("\nown time: "); Serial.println(time1); + Serial.print("ard time: "); Serial.println(time2); + +} + +void loop() { + +} diff --git a/sem4/embedded/emb_m2/myfloat_red.h b/sem4/embedded/emb_m2/myfloat_red.h new file mode 100644 index 0000000..b553233 --- /dev/null +++ b/sem4/embedded/emb_m2/myfloat_red.h @@ -0,0 +1,57 @@ +/* + Low level precision floating point lib + Author: Henrik Schiøler +*/ + +//low precision floating pt type +typedef struct myfloat +{ + signed char mantissa; + signed char exponent; +} myfloat_type; + + +//convert from double to low precision type +void doub2mydouble(double arg, myfloat_type *res) +{ + int exponent; + double temp; + exponent = ceil(log(abs(arg))/log(2)); //base 2 logarithm + temp=arg*pow(2,7-exponent); + res->mantissa = (signed char)temp; + res->exponent = exponent-7; +} + +//convert from low precision type to double +double myfloat2double(myfloat_type *arg1) +{ + double res = (double)(arg1->mantissa) * pow(2,arg1->exponent); + return res; +} + +//multiply to low precision types +void mult_float(myfloat_type *arg1,myfloat_type *arg2,myfloat_type *result) +{ + int temp; + unsigned char sign; + + sign=0x80 & ((unsigned char)arg1-> mantissa ^ (unsigned char)arg2-> mantissa); //find sign of result + + char i=0; + temp = (int)(arg1-> mantissa) * (int)(arg2-> mantissa); + + temp = temp & 0x7f00; //take away sign from product + + while(abs(temp)>128) + { + i++; + temp=temp>>1; + } + + result->mantissa = (unsigned char) temp; + + result->mantissa = result->mantissa | sign; //add recorded sign + + result->exponent = arg1->exponent + arg2->exponent + i; + +} |