DarthAffe/stable-diffusion.cpp
1
0 Fork 0
mirror of https://github.com/leejet/stable-diffusion.cpp.git synced 2026-11-12 03:49:09 +00:00
Code Issues Packages Projects Releases Wiki Activity

perf: optimize H3 inference and segmented weight prefetch (#2240)

Browse Source
This commit is contained in:
leejet authored and GitHub committed 2026-10-30 00:38:14 +08:00
1 parent e7b7ec6d6f
commit 37bc88b5e9
17 files changed
+367 -210
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

src/model_manager.cpp
+10 -4
View File
@@ -1767,10 +1767,16 @@ ModelManager::CapacityCheck ModelManager::check_capacity(
auto add = [](size_t a, size_t b) { return b > SIZE_MAX - a ? SIZE_MAX : a + b; };
const size_t missing = compute_backend_alloc_size(states, true);
// Backend scratch buffers and pipelines are not included in graph measurements.
constexpr size_t safety_margin = 512ULL * 1024ULL * 1024ULL;
result.required_device_bytes = add(add(request.pending_allocation_bytes, missing), safety_margin);
result.required_budget_bytes = add(request.runtime_peak_bytes(), missing);
auto available_device_bytes = [&](ggml_backend_t backend) {
size_t safety_margin = 512ULL * 1024ULL * 1024ULL;
#ifdef _WIN32
if (sd_backend_is(request.compute_backend, "CUDA")) {
// Avoid WDDM allocation churn when transient buffers compete with reusable staging buffers.
safety_margin = 1024ULL * 1024ULL * 1024ULL;
}
#endif
result.required_device_bytes = add(add(request.pending_allocation_bytes, missing), safety_margin);
result.required_budget_bytes = add(request.runtime_peak_bytes(), missing);
auto available_device_bytes = [&](ggml_backend_t backend) {
auto device = ggml_backend_get_device(backend);
if (device == nullptr) {
return SIZE_MAX;