feat: add CUDA execution for custom INT8 operators

This commit is contained in:
leejet committed 2026-10-31 20:24:30 +08:00
1 parent 2f68a92505
commit b71db9a314
5 files changed
+592 -7

No files matched your search

+2
View File
@@ -10,6 +10,7 @@
#include <unordered_set>
#ifdef SD_CUDA_EXTENSIONS
void sd_ggml_register_cuda_i8(ggml_backend_dev_t device);
void sd_ggml_register_cuda_rope(ggml_backend_dev_t device);
#endif
#ifdef SD_CUDA_CUTE_ATTENTION
@@ -179,6 +180,7 @@ void sd_ggml_register_extensions(ggml_backend_t backend) {
}
if (cuda) {
#ifdef SD_CUDA_EXTENSIONS
sd_ggml_register_cuda_i8(device);
sd_ggml_register_cuda_rope(device);
#endif
#ifdef SD_CUDA_CUTE_ATTENTION