zhangyiss/gguf-tools
1
0
Fork 0
mirror of https://github.com/antirez/gguf-tools.git synced 2025-09-18 21:08:07 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Merge pull request #16 from jart/ftz

Browse Source 
Remove flush to zero from bf16
This commit is contained in:
Salvatore Sanfilippo
2025-01-09 16:46:11 +01:00
committed by GitHub
parent bd64d6e812 918234ce80
commit 8fa6eb6523
1 changed files with 0 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
bf16.h
View File

@@ -52,10 +52,6 @@ static inline float from_brain(uint16_t h) {
/** /**
* Converts float32 to brain16. * Converts float32 to brain16.
*
* This function is binary identical to AMD Zen4 VCVTNEPS2BF16.
* Subnormals shall be flushed to zero, and NANs will be quiet.
* This code should vectorize nicely if using modern compilers.
*/ */
static inline uint16_t to_brain(float s) { static inline uint16_t to_brain(float s) {
uint16_t h; uint16_t h;
@@ -68,10 +64,6 @@ static inline uint16_t to_brain(float s) {
h = (u.i >> 16) | 64; /* force to quiet */ h = (u.i >> 16) | 64; /* force to quiet */
return h; return h;
} }
if (!(u.i & 0x7f800000)) { /* subnormal */
h = (u.i & 0x80000000) >> 16; /* flush to zero */
return h;
}
return (u.i + (0x7fff + ((u.i >> 16) & 1))) >> 16; return (u.i + (0x7fff + ((u.i >> 16) & 1))) >> 16;
} }