Add function to populate optstruct from format name

Author: mfasi

src/cpfloat_definitions.h

@@ -24,6 +24,7 @@
 #define _CHOPFAST_DEFINITIONS_
 
 // #define _CRT_RAND_S
+#include <stdbool.h>
 #include <stdlib.h>
 #include <stdint.h>
 
@@ -384,6 +385,21 @@ typedef struct {
   cpfloat_randseed_t *randseed;
 } optstruct;
 
+/**
+ @brief Populate fields of @ref optstruct from `format` field.
+
+ @details This function populates the fields of @p fpopts based on the
+ string in the @p fpopts->format.
+
+ @param[in] fpopts Parameters describing the target format, the rounding mode,
+ and the probability of soft errors striking the rounded values.
+
+ @return The function returns @b 0 if the string in @p fpopts->format is a
+ valid format name, @b 1 if the string is empty or @p NULL, @b 2 if the string
+ is "custom" or "c", and @b -1 if the string is not a valid format name.<p/>
+ */
+int cpfloat_populate_optstruct_from_format(optstruct *fpopts);
+
 /**
  @brief Allocate @ref optstruct struct to store parameters of target format.
 

src/cpfloat_template.h

@@ -101,6 +101,63 @@ static inline BITTYPE PRNG_ADVANCE_BIT(BITSEEDTYPE *seed, size_t delta) {
  * Prototypes are in cpfloat_definitions.
  */
 
+int cpfloat_populate_optstruct_from_format(optstruct *fpopts) {
+  if (fpopts == NULL)
+    return -1;
+  const char *fmt = fpopts->format;
+  bool format_is_valid = true;
+  bool is_subn_rnd_default = false;
+  bool is_inf_no_default = false;
+  if (fmt == NULL || fmt[0] == '\0')
+    return 1;
+  else if (strcmp(fmt, "custom") == 0 || strcmp(fmt, "c") == 0) {
+    return 2;
+  } else if (strcmp(fmt, "q43") == 0 || strcmp(fmt, "e4m3") == 0 || strcmp(fmt, "E4M3") == 0) {
+    fpopts->precision = 4;
+    fpopts->emin = -6;
+    fpopts->emax = 8;
+    is_inf_no_default = true;
+  } else if (strcmp(fmt, "q52") == 0 || strcmp(fmt, "e5m2") == 0 || strcmp(fmt, "E5M2") == 0) {
+    fpopts->precision = 3;
+    fpopts->emin = -14;
+    fpopts->emax = 15;
+  } else if (strcmp(fmt, "b") == 0 || strcmp(fmt, "bf16") == 0 || strcmp(fmt, "bfloat16") == 0) {
+    fpopts->precision = 8;
+    fpopts->emin = -126;
+    fpopts->emax = 127;
+    is_subn_rnd_default = true;
+  } else if (strcmp(fmt, "h") == 0 || strcmp(fmt, "fp16") == 0 || strcmp(fmt, "binary16") == 0 || strcmp(fmt, "half") == 0) {
+    fpopts->precision = 11;
+    fpopts->emin = -14;
+    fpopts->emax = 15;
+  } else if (strcmp(fmt, "t") == 0 || strcmp(fmt, "tf32") == 0 || strcmp(fmt, "TensorFloat-32") == 0) {
+    fpopts->precision = 11;
+    fpopts->emin = -126;
+    fpopts->emax = 127;
+  } else if (strcmp(fmt, "s") == 0 || strcmp(fmt, "fp32") == 0 || strcmp(fmt, "binary32") == 0 || strcmp(fmt, "single") == 0) {
+    fpopts->precision =  24;
+    fpopts->emin = -126;
+    fpopts->emax = 127;
+  } else if (strcmp(fmt, "d") == 0 || strcmp(fmt, "fp64") == 0 || strcmp(fmt, "binary64") == 0 || strcmp(fmt, "double") == 0) {
+    fpopts->precision =   53;
+    fpopts->emin = -1022;
+    fpopts->emax = 1023;
+  } else {
+    format_is_valid = false;
+  }
+  fpopts->subnormal = is_subn_rnd_default ? CPFLOAT_SUBN_RND : CPFLOAT_SUBN_USE;
+  fpopts->explim = CPFLOAT_EXPRANGE_TARG;
+  fpopts->infinity = is_inf_no_default ? CPFLOAT_INF_NO : CPFLOAT_INF_USE;
+  fpopts->round = CPFLOAT_RND_NE;
+  fpopts->saturation = CPFLOAT_SAT_NO;
+  fpopts->flip = CPFLOAT_SOFTERR_NO;
+  fpopts->p = 0.0;
+  if (format_is_valid)
+    return 0;
+  else
+    return -1;
+}
+
 optstruct *init_optstruct() {
   optstruct *fpopts = malloc(sizeof(*fpopts));
   fpopts->bitseed = NULL;