DarthAffe/stable-diffusion.cpp
1
0 Fork 0
mirror of https://github.com/leejet/stable-diffusion.cpp.git synced 2026-11-11 03:24:30 +00:00
Code Issues Packages Projects Releases Wiki Activity

perf: accelerate INT8 diffusion inference with CUDA fusions (#2235)

Browse Source
This commit is contained in:
leejet authored and GitHub committed 2026-10-28 22:02:46 +08:00
1 parent 8cbe905fdd
commit 77b72a2efe
12 files changed
+113 -128
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

src/core/ggml_runner.h
+2 -2
View File
@@ -141,7 +141,7 @@ private:
size_t logged_segment_count_ = 0;
ggml_status last_compute_status_ = GGML_STATUS_SUCCESS;
sd::ComputeWorkspace::Measurement measure(ggml_cgraph* graph, size_t direct_bytes);
sd::ComputeWorkspace::Measurement measure(ggml_cgraph* graph);
std::vector<DeviceMemoryRequest> memory_requests(const std::vector<sd::BackendBufferSize>& sizes,
size_t pending_cache_bytes) const;
bool fits(const std::vector<DeviceMemoryRequest>& requests,
@@ -161,6 +161,7 @@ protected:
sd::GraphCutTensorCache cut_cache_;
sd::ComputeWorkspace workspace_;
ggml_context* compute_ctx = nullptr;
std::unique_ptr<std::max_align_t[]> compute_ctx_buffer_;
bool runner_started_ = false;
bool graph_active_ = false;
@@ -169,7 +170,6 @@ protected:
std::vector<size_t> graph_cut_layer_split_backend_vram_limits_;
std::vector<ggml_backend_t> extra_runtime_backends; // borrowed (SDBackendManager-owned)
bool multi_device_eval_callback_warned = false;
std::shared_ptr<WeightAdapter> weight_adapter = nullptr;
std::weak_ptr<DeviceResidencyManager> residency_manager;