1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
|
Prevent implicit conversion of float to double to avoid precision
error on i686.
https://github.com/Maratyszcza/FP16/issues/20
Taken from Debian:
https://salsa.debian.org/deeplearning-team/fp16/-/blob/master/debian/patches/ftbfs-i386.patch
Index: fp16/include/fp16/fp16.h
===================================================================
--- fp16.orig/include/fp16/fp16.h
+++ fp16/include/fp16/fp16.h
@@ -228,7 +228,8 @@ static inline uint16_t fp16_ieee_from_fp
const float scale_to_inf = fp32_from_bits(UINT32_C(0x77800000));
const float scale_to_zero = fp32_from_bits(UINT32_C(0x08800000));
#endif
- float base = (fabsf(f) * scale_to_inf) * scale_to_zero;
+ const volatile float base_inf = fabsf(f) * scale_to_inf;
+ float base = base_inf * scale_to_zero;
const uint32_t w = fp32_to_bits(f);
const uint32_t shl1_w = w + w;
|
