DarthAffe/stable-diffusion.cpp
1
0
Fork 0
mirror of https://github.com/leejet/stable-diffusion.cpp.git synced 2025-12-12 21:38:58 +00:00
Code Issues Packages Projects Releases Wiki Activity

refactor: move tiling cacl and debug print into the tiling code branch (#833)

Browse Source
This commit is contained in:
Erik Scholz 2025-09-16 16:46:56 +02:00 committed by GitHub
parent 8376dfba2a
commit 8909523e92
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 13 additions and 13 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

26
stable-diffusion.cpp
View File

@ -1362,15 +1362,15 @@ public:
} }
if (!use_tiny_autoencoder) { if (!use_tiny_autoencoder) {
float tile_overlap;
int tile_size_x, tile_size_y;
// multiply tile size for encode to keep the compute buffer size consistent
get_tile_sizes(tile_size_x, tile_size_y, tile_overlap, vae_tiling_params, W, H, 1.30539f);
LOG_DEBUG("VAE Tile size: %dx%d", tile_size_x, tile_size_y);
process_vae_input_tensor(x); process_vae_input_tensor(x);
if (vae_tiling_params.enabled && !encode_video) { if (vae_tiling_params.enabled && !encode_video) {
float tile_overlap;
int tile_size_x, tile_size_y;
// multiply tile size for encode to keep the compute buffer size consistent
get_tile_sizes(tile_size_x, tile_size_y, tile_overlap, vae_tiling_params, W, H, 1.30539f);
LOG_DEBUG("VAE Tile size: %dx%d", tile_size_x, tile_size_y);
auto on_tiling = [&](ggml_tensor* in, ggml_tensor* out, bool init) { auto on_tiling = [&](ggml_tensor* in, ggml_tensor* out, bool init) {
first_stage_model->compute(n_threads, in, false, &out, work_ctx); first_stage_model->compute(n_threads, in, false, &out, work_ctx);
}; };
@ -1507,15 +1507,15 @@ public:
} }
int64_t t0 = ggml_time_ms(); int64_t t0 = ggml_time_ms();
if (!use_tiny_autoencoder) { if (!use_tiny_autoencoder) {
float tile_overlap;
int tile_size_x, tile_size_y;
get_tile_sizes(tile_size_x, tile_size_y, tile_overlap, vae_tiling_params, x->ne[0], x->ne[1]);
LOG_DEBUG("VAE Tile size: %dx%d", tile_size_x, tile_size_y);
process_latent_out(x); process_latent_out(x);
// x = load_tensor_from_file(work_ctx, "wan_vae_z.bin"); // x = load_tensor_from_file(work_ctx, "wan_vae_z.bin");
if (vae_tiling_params.enabled && !decode_video) { if (vae_tiling_params.enabled && !decode_video) {
float tile_overlap;
int tile_size_x, tile_size_y;
get_tile_sizes(tile_size_x, tile_size_y, tile_overlap, vae_tiling_params, x->ne[0], x->ne[1]);
LOG_DEBUG("VAE Tile size: %dx%d", tile_size_x, tile_size_y);
// split latent in 32x32 tiles and compute in several steps // split latent in 32x32 tiles and compute in several steps
auto on_tiling = [&](ggml_tensor* in, ggml_tensor* out, bool init) { auto on_tiling = [&](ggml_tensor* in, ggml_tensor* out, bool init) {
first_stage_model->compute(n_threads, in, true, &out, NULL); first_stage_model->compute(n_threads, in, true, &out, NULL);