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

feat: increase work_ctx memory buffer size (#814)

Browse Source
This commit is contained in:
leejet 2025-09-14 13:19:20 +08:00 committed by GitHub
parent a6a8569ea0
commit dc46993b55
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
5 changed files with 7 additions and 21 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
conditioner.hpp
View File

@ -115,7 +115,7 @@ struct FrozenCLIPEmbedderWithCustomWords : public Conditioner {
return true;
}
struct ggml_init_params params;
params.mem_size = 10 * 1024 * 1024; // max for custom embeddings 10 MB
params.mem_size = 100 * 1024 * 1024; // max for custom embeddings 100 MB
params.mem_buffer = NULL;
params.no_alloc = false;
struct ggml_context* embd_ctx = ggml_init(params);

2
preprocessing.hpp
View File

@ -164,7 +164,7 @@ void threshold_hystersis(struct ggml_tensor* img, float high_threshold, float lo
uint8_t* preprocess_canny(uint8_t* img, int width, int height, float high_threshold, float low_threshold, float weak, float strong, bool inverse) {
struct ggml_init_params params;
params.mem_size = static_cast<size_t>(10 * 1024 * 1024); // 10
params.mem_size = static_cast<size_t>(10 * 1024 * 1024); // 10MB
params.mem_buffer = NULL;
params.no_alloc = false;
struct ggml_context* work_ctx = ggml_init(params);

17
stable-diffusion.cpp
View File

@ -2196,19 +2196,7 @@ sd_image_t* generate_image(sd_ctx_t* sd_ctx, const sd_img_gen_params_t* sd_img_g
}
struct ggml_init_params params;
params.mem_size = static_cast<size_t>(10 * 1024 * 1024); // 10 MB
if (sd_version_is_sd3(sd_ctx->sd->version)) {
params.mem_size *= 3;
}
if (sd_version_is_flux(sd_ctx->sd->version)) {
params.mem_size *= 4;
}
if (sd_ctx->sd->stacked_id) {
params.mem_size += static_cast<size_t>(10 * 1024 * 1024); // 10 MB
}
params.mem_size += width * height * 3 * sizeof(float) * 3;
params.mem_size += width * height * 3 * sizeof(float) * 3 * sd_img_gen_params->ref_images_count;
params.mem_size *= sd_img_gen_params->batch_count;
params.mem_size = static_cast<size_t>(1024 * 1024) * 1024; // 1G
params.mem_buffer = NULL;
params.no_alloc = false;
// LOG_DEBUG("mem_size %u ", params.mem_size);
@ -2448,8 +2436,7 @@ SD_API sd_image_t* generate_video(sd_ctx_t* sd_ctx, const sd_vid_gen_params_t* s
}
struct ggml_init_params params;
params.mem_size = static_cast<size_t>(200 * 1024) * 1024; // 200 MB
params.mem_size += width * height * frames * 3 * sizeof(float) * 2;
params.mem_size = static_cast<size_t>(1024 * 1024) * 1024; // 1GB
params.mem_buffer = NULL;
params.no_alloc = false;
// LOG_DEBUG("mem_size %u ", params.mem_size);

5
upscaler.cpp
View File

@ -69,8 +69,7 @@ struct UpscalerGGML {
input_image.width, input_image.height, output_width, output_height);
struct ggml_init_params params;
params.mem_size = output_width * output_height * 3 * sizeof(float) * 2;
params.mem_size += 2 * ggml_tensor_overhead();
params.mem_size = static_cast<size_t>(1024 * 1024) * 1024; // 1G
params.mem_buffer = NULL;
params.no_alloc = false;
@ -80,7 +79,7 @@ struct UpscalerGGML {
LOG_ERROR("ggml_init() failed");
return upscaled_image;
}
LOG_DEBUG("upscale work buffer size: %.2f MB", params.mem_size / 1024.f / 1024.f);
// LOG_DEBUG("upscale work buffer size: %.2f MB", params.mem_size / 1024.f / 1024.f);
ggml_tensor* input_image_tensor = ggml_new_tensor_4d(upscale_ctx, GGML_TYPE_F32, input_image.width, input_image.height, 3, 1);
sd_image_to_tensor(input_image.data, input_image_tensor);

2
wan.hpp
View File

@ -1219,7 +1219,7 @@ namespace WAN {
void test() {
struct ggml_init_params params;
params.mem_size = static_cast<size_t>(1000 * 1024 * 1024); // 10 MB
params.mem_size = static_cast<size_t>(1024 * 1024) * 1024; // 1G
params.mem_buffer = NULL;
params.no_alloc = false;