Tentatively manually encoding non-finite f16 values.

Issue #988.

lib/srfi/160/mini-test.sld

@@ -1,6 +1,6 @@
 
 (define-library (srfi 160 mini-test)
-  (import (scheme base)
+  (import (scheme base) (scheme inexact)
           (srfi 160 base) (srfi 160 f8) (srfi 160 f16)
           (chibi test))
   (export run-tests)
@@ -82,7 +82,11 @@
         (test '#f16(1 2)
           (vector->f16vector '#(0 1 2 3) 1 3))
         (test '#(1.0 2.0)
-          (f16vector->vector '#f16(0 1 2 3) 1 3))
+            (f16vector->vector '#f16(0 1 2 3) 1 3))
+        (test '(-inf.0 -1.0 -0.0 0.0 1.0 +inf.0)
+            (f16vector->list
+             '#f16(-inf.0 -1.0 -0.0 0.0 1.0 +inf.0)))
+        (test-assert (nan? (f16vector-ref '#f16(+nan.0) 0)))
         )
 
       (test-end))))

sexp.c

@@ -3206,16 +3206,23 @@ static float int_as_float(const unsigned int n) {
 
 /* https://arxiv.org/abs/2112.08926 */
 double sexp_half_to_double(unsigned short x) {
-  unsigned int e = (x&0x7C00)>>10,
-    m = (x&0x03FF)<<13,
-    v = float_as_int((float)m)>>23;
+  unsigned int e, m, v;
+  if (x == 31744) return INFINITY;
+  if (x == 32767) return NAN;
+  if (x == 64512) return -INFINITY;
+  e = (x&0x7C00)>>10;
+  m = (x&0x03FF)<<13;
+  v = float_as_int((float)m)>>23;
   return int_as_float((x&0x8000)<<16 | (e!=0)*((e+112)<<23|m) | ((e==0)&(m!=0))*((v-37)<<23|((m<<(150-v))&0x007FE000)));
 }
 
 unsigned short sexp_double_to_half(double x) {
-  unsigned int b = float_as_int(x)+0x00001000,
-    e = (b&0x7F800000)>>23,
-    m = b&0x007FFFFF;
+  unsigned int b, e, m;
+  if (isnan(x)) return 32767;
+  if (isinf(x)) return x < 0 ? 64512 : 31744;
+  b = float_as_int(x)+0x00001000;
+  e = (b&0x7F800000)>>23;
+  m = b&0x007FFFFF;
   return (b&0x80000000)>>16 | (e>112)*((((e-112)<<10)&0x7C00)|m>>13) | ((e<113)&(e>101))*((((0x007FF000+m)>>(125-e))+1)>>1) | (e>143)*0x7FFF;
 }
 #endif