Thread: I can't spot my mistake...

Originally Posted by JaredPuffin

Thank You!
Finaly a nice and angumented answer

Yeaaah, thing is I already did that before I started, doesn't mean I get a full understanding from documents alone, sometimes you just have to get on with failing a few dozen times before you finally understand correctly (especially if what you're dealing with doesn't a nice fixed size like they where working with)

Gotta get ready for work so will just dump the code and output for now (uploaded all code to alu project on github just now so can search through that whilst waiting for me to get a chance to repond to questions)

Code:

int_t alup__neg_int( alup_t const * const _SRC )
{
	int ret = alup_not( _SRC );
	
	if ( ret == 0 )
		return alup__inc_int( _SRC );
		
	alu_error( ret );
	return ret;
}

int_t alup_neg( alup_t const * const _SRC )
{
	if ( alup_floating( _SRC ) )
	{
		alup_t _MAN;
		alub_t sign = alup_final_bit( _SRC );
		bool_t neg = !(*(sign.ptr) & sign.mask);
		
		*(sign.ptr) &= ~(sign.mask);
		*(sign.ptr) |= IFTRUE( neg, sign.mask );
		
		alup_init_mantissa( _SRC, _MAN );
		
		return alup__neg_int( &_MAN );
	}
	
	return alup__neg_int( _SRC );
}

Code:

make check.run
...
Running suite(s): ALU
../tests/check_alu.c:670: test_alup_op1_floating_incremental_fn() 192, -(192.000000) = -192.000000, got 192.000000
../tests/check_alu.c:680: test_alup_op1_floating_incremental_fn() &_TESTNO = 0 000'0000'0000'0000'0000'0000'1100'0000
../tests/check_alu.c:681: test_alup_op1_floating_incremental_fn() &_BEFORE = 0 +0007 10000110 100'0000'0000'0000'0000'0000
../tests/check_alu.c:682: test_alup_op1_floating_incremental_fn() &_EXPECT = 1 +0007 10000110 100'0000'0000'0000'0000'0000
../tests/check_alu.c:683: test_alup_op1_floating_incremental_fn() &_RESULT = 0 +0007 10000110 100'0000'0000'0000'0000'0000
99%: Checks: 3561, Failures: 1, Errors: 0
...
Compilation finished successfully.

Originally Posted by awsdert

Gotta get ready for work so will just dump the code and output for now (uploaded all code to alu project on github just now so can search through that whilst waiting for me to get a chance to repond to questions)

Code:

int_t alup__neg_int( alup_t const * const _SRC )
{
	int ret = alup_not( _SRC );
	
	if ( ret == 0 )
		return alup__inc_int( _SRC );
		
	alu_error( ret );
	return ret;
}

int_t alup_neg( alup_t const * const _SRC )
{
	if ( alup_floating( _SRC ) )
	{
		alup_t _MAN;
		alub_t sign = alup_final_bit( _SRC );
		bool_t neg = !(*(sign.ptr) & sign.mask);
		
		*(sign.ptr) &= ~(sign.mask);
		*(sign.ptr) |= IFTRUE( neg, sign.mask );
		
		alup_init_mantissa( _SRC, _MAN );
		
		return alup__neg_int( &_MAN );
	}
	
	return alup__neg_int( _SRC );
}

Code:

make check.run
...
Running suite(s): ALU
../tests/check_alu.c:670: test_alup_op1_floating_incremental_fn() 192, -(192.000000) = -192.000000, got 192.000000
../tests/check_alu.c:680: test_alup_op1_floating_incremental_fn() &_TESTNO = 0 000'0000'0000'0000'0000'0000'1100'0000
../tests/check_alu.c:681: test_alup_op1_floating_incremental_fn() &_BEFORE = 0 +0007 10000110 100'0000'0000'0000'0000'0000
../tests/check_alu.c:682: test_alup_op1_floating_incremental_fn() &_EXPECT = 1 +0007 10000110 100'0000'0000'0000'0000'0000
../tests/check_alu.c:683: test_alup_op1_floating_incremental_fn() &_RESULT = 0 +0007 10000110 100'0000'0000'0000'0000'0000
99%: Checks: 3561, Failures: 1, Errors: 0
...
Compilation finished successfully.

Turned out to be as simple as a noob mistake on length handling, renamed a parameter of alup_t to make it less likely I'll make that mistake in future, finally got back to FPN math bugs, current one is in multiplication though that is probably due to a change I made in alup.h after noticing my mistake, I'll check another day as I will soon go to bed

slightly off topic though still important for the ALU, finally found an example of those annoyingly complex math formulas for random numbers converted into code, tried my own variant (just a change in numbers) which seems to produce the static for a *.ppm file but when I tried to install dieharder to test the quality the only accessible one was from the AUR which isn't linking due to multiple definitions of symbols, can someone with dieharder compile the attached file and post the results please.
In case you wanna play with the code the main function to consider is this one:

Code:

size_t rng( size_t seed )
{
#if 0
	/* Lehmer method */
	size_t base = seed + 0xE120FC15;
	size_t temp = base * 0x4A39b70D;
	size_t num1 = (temp >> 32) ^ temp;
	temp = num1 * 0x12FAD5C9;
	return (temp >> 32) ^ temp;
#else
	size_t base = seed + 0xEA57;
	size_t temp = base * 0xDEAD;
	size_t num1 = (temp >> (bitsof(size_t)/2)) ^ temp;
	temp = num1 * 0xC0DE;
	return (temp >> (bitsof(size_t)/2)) ^ temp;
#endif
}

the original code I saw was for 32 bit numbers which somehow worked for 64 bit but I wanted 16bit compatible also hence the shorter numbers and a calculation of bits to move by