DarthAffe/stable-diffusion.cpp
1
0 Fork 0
mirror of https://github.com/leejet/stable-diffusion.cpp.git synced 2026-11-14 12:56:33 +00:00
Code Issues Packages Projects Releases Wiki Activity

refactor: unify ggml operation helpers around runner context (#2314)

Browse Source
This commit is contained in:
leejet authored and GitHub committed 2026-11-10 01:58:04 +08:00
1 parent 73fef08224
commit a3c49b13be
76 files changed
+1983 -1831
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

src/pipeline/diffusion_engine.cpp
+9 -7
View File
@@ -856,11 +856,13 @@ bool StableDiffusionGGML::set_sage_attention_enabled(bool enabled) {
if (ctx == nullptr) {
return false;
}
auto* q = ggml_new_tensor_4d(ctx, GGML_TYPE_F32, 128, 128, 1, 1);
auto* k = ggml_new_tensor_4d(ctx, GGML_TYPE_F32, 128, 128, 1, 1);
auto* v = ggml_new_tensor_4d(ctx, GGML_TYPE_F16, 128, 128, 1, 1);
auto* op = ggml_sage_attn(ctx, q, k, v, 1.f / sqrtf(128.f), GGML_SAGE_ATTN_AUTO);
bool supported = true;
auto* q = ggml_new_tensor_4d(ctx, GGML_TYPE_F32, 128, 128, 1, 1);
auto* k = ggml_new_tensor_4d(ctx, GGML_TYPE_F32, 128, 128, 1, 1);
auto* v = ggml_new_tensor_4d(ctx, GGML_TYPE_F16, 128, 128, 1, 1);
GGMLRunnerContext probe_ctx;
probe_ctx.ggml_ctx = ctx;
auto* op = ggml_ext_sage_attn(&probe_ctx, q, k, v, 1.f / sqrtf(128.f), GGML_SAGE_ATTN_AUTO);
bool supported = true;
for (auto backend : backend_manager.runtime_backends(SDBackendModule::DIFFUSION)) {
if (!ggml_backend_supports_op(backend, op)) {
LOG_ERROR("SageAttention is unavailable on %s; it requires patched GGML, CUDA Toolkit 12.0 or newer, and SM80 or newer kernels",
@@ -2442,9 +2444,9 @@ sd::Tensor<float> StableDiffusionGGML::sample(const std::shared_ptr<DiffusionMod
#ifdef SD_USE_UPSTREAM_GGML
// workaround for sd.cpp #2149
const auto diffusion_backends = backend_manager.runtime_backends(SDBackendModule::DIFFUSION);
const auto diffusion_backends = backend_manager.runtime_backends(SDBackendModule::DIFFUSION);
const bool batch_cfg_backend_ok = std::none_of(diffusion_backends.begin(), diffusion_backends.end(),
[](ggml_backend_t backend) { return sd_backend_is(backend, "MTL"); });
[](ggml_backend_t backend) { return sd_backend_is(backend, "MTL"); });
#else
const bool batch_cfg_backend_ok = true;
#endif