/* RETRO ------------------------------------------------------ A personal, minimalistic forth Copyright (c) 2016 - 2019 Charles Childers ---------------------------------------------------------- */ #include #include #include #include #include #include #include #include #include #include #include #include #include typedef void (*Handler)(void); /*--------------------------------------------------------------------- First, a few things that relate to the Nga virtual machine that RETRO runs on. ---------------------------------------------------------------------*/ #define CELL int32_t /* Cell size (32 bit, signed integer */ #define IMAGE_SIZE 524288 * 48 /* Amount of RAM. 12MiB by default. */ #define ADDRESSES 2048 /* Depth of address stack */ #define STACK_DEPTH 512 /* Depth of data stack */ extern CELL sp, rp, ip; /* Stack & instruction pointers */ extern CELL data[STACK_DEPTH]; /* The data stack */ extern CELL address[ADDRESSES]; /* The address stack */ extern CELL memory[IMAGE_SIZE + 1]; /* The memory for the image */ extern char string_data[]; #define TOS data[sp] /* Shortcut for top item on stack */ #define NOS data[sp-1] /* Shortcut for second item on stack */ #define TORS address[rp] /* Shortcut for top item on address stack */ /*--------------------------------------------------------------------- Function prototypes. ---------------------------------------------------------------------*/ CELL stack_pop(); void stack_push(CELL value); char *string_extract(CELL at); CELL string_inject(char *str, CELL buffer); /*--------------------------------------------------------------------- I have a stack of floating point values ("floats") and a stack pointer (`fsp`). ---------------------------------------------------------------------*/ double Floats[8192]; CELL fsp; double AFloats[8192]; CELL afsp; /*--------------------------------------------------------------------- The first two functions push a float to the stack and pop a value off the stack. ---------------------------------------------------------------------*/ void float_push(double value) { fsp++; Floats[fsp] = value; } double float_pop() { fsp--; return Floats[fsp + 1]; } void float_to_alt() { afsp++; AFloats[afsp] = float_pop(); } void float_from_alt() { afsp--; float_push(AFloats[afsp + 1]); } /*--------------------------------------------------------------------- RETRO operates on 32-bit signed integer values. This function just pops a number from the data stack, casts it to a float, and pushes it to the float stack. ---------------------------------------------------------------------*/ void float_from_number() { float_push((double)stack_pop()); } /*--------------------------------------------------------------------- To get a float from a string in the image, I provide this function. I cheat: using `atof()` takes care of the details, so I don't have to. ---------------------------------------------------------------------*/ void float_from_string() { float_push(atof(string_extract(stack_pop()))); } /*--------------------------------------------------------------------- Converting a floating point into a string is slightly more work. Here I pass it off to `snprintf()` to deal with. ---------------------------------------------------------------------*/ void float_to_string() { snprintf(string_data, 8192, "%f", float_pop()); string_inject(string_data, stack_pop()); } /*--------------------------------------------------------------------- Converting a floating point back into a standard number requires a little care due to the signed nature. This makes adjustments for the max & min value, and then casts (rounding) the float back to a normal number. ---------------------------------------------------------------------*/ void float_to_number() { double a = float_pop(); if (a > 2147483647) a = 2147483647; if (a < -2147483648) a = -2147483648; stack_push((CELL)round(a)); } /*--------------------------------------------------------------------- Now I get to define a bunch of functions that operate on floats. These provide the basic math, and wrappers around functionality in libm. ---------------------------------------------------------------------*/ void float_add() { double a = float_pop(); double b = float_pop(); float_push(a+b); } void float_sub() { double a = float_pop(); double b = float_pop(); float_push(b-a); } void float_mul() { double a = float_pop(); double b = float_pop(); float_push(a*b); } void float_div() { double a = float_pop(); double b = float_pop(); float_push(b/a); } void float_floor() { float_push(floor(float_pop())); } void float_ceil() { float_push(ceil(float_pop())); } void float_eq() { double a = float_pop(); double b = float_pop(); if (a == b) stack_push(-1); else stack_push(0); } void float_neq() { double a = float_pop(); double b = float_pop(); if (a != b) stack_push(-1); else stack_push(0); } void float_lt() { double a = float_pop(); double b = float_pop(); if (b < a) stack_push(-1); else stack_push(0); } void float_gt() { double a = float_pop(); double b = float_pop(); if (b > a) stack_push(-1); else stack_push(0); } void float_depth() { stack_push(fsp); } void float_dup() { double a = float_pop(); float_push(a); float_push(a); } void float_drop() { float_pop(); } void float_swap() { double a = float_pop(); double b = float_pop(); float_push(a); float_push(b); } void float_log() { double a = float_pop(); double b = float_pop(); float_push(log(b) / log(a)); } void float_sqrt() { float_push(sqrt(float_pop())); } void float_pow() { double a = float_pop(); double b = float_pop(); float_push(pow(b, a)); } void float_sin() { float_push(sin(float_pop())); } void float_cos() { float_push(cos(float_pop())); } void float_tan() { float_push(tan(float_pop())); } void float_asin() { float_push(asin(float_pop())); } void float_acos() { float_push(acos(float_pop())); } void float_atan() { float_push(atan(float_pop())); } /*--------------------------------------------------------------------- With this finally done, I implement the FPU instructions. ---------------------------------------------------------------------*/ Handler FloatHandlers[] = { float_from_number, float_from_string, float_to_number, float_to_string, float_add, float_sub, float_mul, float_div, float_floor, float_ceil, float_sqrt, float_eq, float_neq, float_lt, float_gt, float_depth, float_dup, float_drop, float_swap, float_log, float_pow, float_sin, float_tan, float_cos, float_asin, float_acos, float_atan, float_to_alt, float_from_alt }; void io_floatingpoint_query() { stack_push(1); stack_push(2); } void io_floatingpoint_handler() { FloatHandlers[stack_pop()](); }