DarthAffe/stable-diffusion.cpp
1
0 Fork 0
mirror of https://github.com/leejet/stable-diffusion.cpp.git synced 2026-11-11 11:26:34 +00:00
Code Issues Packages Projects Releases Wiki Activity

feat: add native CUDA SageAttention support (#2005)

Browse Source
This commit is contained in:
leejet authored and GitHub committed 2026-09-21 01:16:59 +08:00
1 parent 15f335daa5
commit 187b2561ea
13 files changed
+176 -5
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

src/core/ggml_runner.h
+10
View File
@@ -68,6 +68,7 @@ struct GGMLRunnerContext {
ggml_backend_t backend = nullptr;
ggml_context* ggml_ctx = nullptr;
bool flash_attn_enabled = false;
bool sage_attn_enabled = false;
float linear_scale = 0.f;
float attn_scale = 0.f;
bool conv2d_direct_enabled = false;
@@ -176,6 +177,7 @@ protected:
const std::string final_result_name = "ggml_runner_final_result_tensor";
bool flash_attn_enabled = false;
bool sage_attn_enabled = false;
float linear_scale = 0.f;
float attn_scale = 0.f;
bool conv2d_direct_enabled = false;
@@ -337,6 +339,14 @@ public:
flash_attn_enabled = enabled;
}
void set_sage_attention_enabled(bool enabled) {
if (sage_attn_enabled != enabled) {
free_cache_ctx_and_buffer();
graph_cut_plan_cache_.graph_cut_plans.clear();
sage_attn_enabled = enabled;
}
}
void set_scale_overrides(float linear_scale, float attn_scale) {
this->linear_scale = linear_scale;
this->attn_scale = attn_scale;