aboutsummaryrefslogtreecommitdiff
path: root/qwen_kernels_impl.h
diff options
context:
space:
mode:
authorOnur Gündüz <ogunduz@gmail.com>2026-06-15 08:40:11 +0200
committerOnur Gündüz <ogunduz@gmail.com>2026-06-15 08:40:11 +0200
commitc4a1c97a09da6d8d88c2ac834bf6234506fc41bc (patch)
tree81ce5675ce886aad0309a41f4d6a696826e0a061 /qwen_kernels_impl.h
parentb377ba8656bfb807ed8dda6c9f26942b50ce626d (diff)
downloadffwd-c4a1c97a09da6d8d88c2ac834bf6234506fc41bc.tar.gz
ffwd-c4a1c97a09da6d8d88c2ac834bf6234506fc41bc.zip
chore: run clang-format
Diffstat (limited to 'qwen_kernels_impl.h')
-rw-r--r--qwen_kernels_impl.h30
1 files changed, 15 insertions, 15 deletions
diff --git a/qwen_kernels_impl.h b/qwen_kernels_impl.h
index 43e4fbe..217f1e8 100644
--- a/qwen_kernels_impl.h
+++ b/qwen_kernels_impl.h
@@ -7,47 +7,47 @@
#include <stdint.h>
-void qwen_bf16_matvec_fused_generic(float *y, const float *x, const uint16_t *W_bf16,
- const float *bias, int in_dim, int out_dim);
+void qwen_bf16_matvec_fused_generic(
+ float *y, const float *x, const uint16_t *W_bf16, const float *bias, int in_dim, int out_dim);
float qwen_dot_f32_generic(const float *a, const float *b, int n);
void qwen_vec_scale_inplace_generic(float *dst, float scale, int n);
void qwen_vec_axpy_inplace_generic(float *dst, const float *src, float alpha, int n);
void qwen_vec_scale_add_generic(float *dst, const float *src, float correction, int n);
#ifdef __ARM_NEON
-void qwen_bf16_matvec_fused_neon(float *y, const float *x, const uint16_t *W_bf16,
- const float *bias, int in_dim, int out_dim);
+void qwen_bf16_matvec_fused_neon(
+ float *y, const float *x, const uint16_t *W_bf16, const float *bias, int in_dim, int out_dim);
float qwen_dot_f32_neon(const float *a, const float *b, int n);
void qwen_vec_scale_inplace_neon(float *dst, float scale, int n);
void qwen_vec_axpy_inplace_neon(float *dst, const float *src, float alpha, int n);
void qwen_vec_scale_add_neon(float *dst, const float *src, float correction, int n);
#define qwen_bf16_matvec_fused_impl qwen_bf16_matvec_fused_neon
-#define qwen_dot_f32_impl qwen_dot_f32_neon
+#define qwen_dot_f32_impl qwen_dot_f32_neon
#define qwen_vec_scale_inplace_impl qwen_vec_scale_inplace_neon
-#define qwen_vec_axpy_inplace_impl qwen_vec_axpy_inplace_neon
-#define qwen_vec_scale_add_impl qwen_vec_scale_add_neon
+#define qwen_vec_axpy_inplace_impl qwen_vec_axpy_inplace_neon
+#define qwen_vec_scale_add_impl qwen_vec_scale_add_neon
#elif defined(__AVX2__) && defined(__FMA__)
-void qwen_bf16_matvec_fused_avx(float *y, const float *x, const uint16_t *W_bf16,
- const float *bias, int in_dim, int out_dim);
+void qwen_bf16_matvec_fused_avx(
+ float *y, const float *x, const uint16_t *W_bf16, const float *bias, int in_dim, int out_dim);
float qwen_dot_f32_avx(const float *a, const float *b, int n);
void qwen_vec_scale_inplace_avx(float *dst, float scale, int n);
void qwen_vec_axpy_inplace_avx(float *dst, const float *src, float alpha, int n);
void qwen_vec_scale_add_avx(float *dst, const float *src, float correction, int n);
#define qwen_bf16_matvec_fused_impl qwen_bf16_matvec_fused_avx
-#define qwen_dot_f32_impl qwen_dot_f32_avx
+#define qwen_dot_f32_impl qwen_dot_f32_avx
#define qwen_vec_scale_inplace_impl qwen_vec_scale_inplace_avx
-#define qwen_vec_axpy_inplace_impl qwen_vec_axpy_inplace_avx
-#define qwen_vec_scale_add_impl qwen_vec_scale_add_avx
+#define qwen_vec_axpy_inplace_impl qwen_vec_axpy_inplace_avx
+#define qwen_vec_scale_add_impl qwen_vec_scale_add_avx
#else
#define qwen_bf16_matvec_fused_impl qwen_bf16_matvec_fused_generic
-#define qwen_dot_f32_impl qwen_dot_f32_generic
+#define qwen_dot_f32_impl qwen_dot_f32_generic
#define qwen_vec_scale_inplace_impl qwen_vec_scale_inplace_generic
-#define qwen_vec_axpy_inplace_impl qwen_vec_axpy_inplace_generic
-#define qwen_vec_scale_add_impl qwen_vec_scale_add_generic
+#define qwen_vec_axpy_inplace_impl qwen_vec_axpy_inplace_generic
+#define qwen_vec_scale_add_impl qwen_vec_scale_add_generic
#endif
#endif /* QWEN_KERNELS_IMPL_H */