perf: fuse Z-Image GPU operations and optimize BF16 GEMM (#2265)

This commit is contained in:
leejet authored and GitHub committed 2026-11-03 23:59:21 +08:00
1 parent 6225bf8c88
commit d30854463d
9 files changed
+530 -5

No files matched your search

+4
View File
@@ -12,6 +12,8 @@ namespace SD_CUDA_NAMESPACE {
void register_convrot(ggml_backend_dev_t device);
void register_affine(ggml_backend_dev_t device);
void register_modulate(ggml_backend_dev_t device);
void register_bf16(ggml_backend_dev_t device);
void register_rms_affine(ggml_backend_dev_t device);
#if !defined(SD_CUDA_VENDOR_HIP) && !defined(SD_CUDA_VENDOR_MUSA)
void register_sage_attention(ggml_backend_dev_t device);
void register_sol_attention(ggml_backend_dev_t device);
@@ -25,6 +27,8 @@ namespace SD_CUDA_NAMESPACE {
register_i8(device);
register_convrot(device);
register_rope(device);
register_bf16(device);
register_rms_affine(device);
#if !defined(SD_CUDA_VENDOR_HIP) && !defined(SD_CUDA_VENDOR_MUSA)
register_sage_attention(device);
register_sol_attention(device);