Add fast sine and cosine definitions in common.h for CUDA templates (#865)

86aaf3c1 · Tong WU · GitHub · 48c9a352 · 86aaf3c1
Unverified Commit 86aaf3c1 authored Sep 23, 2025 by Tong WU Committed by GitHub Sep 23, 2025
Show whitespace changes
Inline Side-by-side

Showing with 2 additions and 0 deletions

src/tl_templates/cuda/common.h src/tl_templates/cuda/common.h +2 -0

No files found.
--- a/src/tl_templates/cuda/common.h
+++ b/src/tl_templates/cuda/common.h
@@ -18,6 +18,8 @@ using int4_t = int4;
 #define hexp cutlass::fast_exp
 #define hlog cutlass::fast_log
 #define hsqrt cutlass::fast_sqrt
+#define hsin cutlass::fast_sin
+#define hcos cutlass::fast_cos
 #define htanh cutlass::fast_tanh
 #define hpow powf