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/model/diffusion/fibo2.hpp
+9 -11
View File
@@ -108,14 +108,14 @@ namespace Fibo2 {
};
ggml_tensor *q, *k, *v;
if (fused_qkv) {
auto parts = ggml_ext_chunk(c, project("qkv", x), 3, 0);
auto parts = ggml_ext_chunk(ctx, project("qkv", x), 3, 0);
q = parts[0];
k = parts[1];
v = parts[2];
} else {
q = project("to_q", x);
if (fused_kv) {
auto parts = ggml_ext_chunk(c, project("to_kv", kv), 2, 0);
auto parts = ggml_ext_chunk(ctx, project("to_kv", kv), 2, 0);
k = parts[0];
v = parts[1];
} else {
@@ -158,9 +158,7 @@ namespace Fibo2 {
return ZImage::FeedForward::forward(ctx, x);
}
auto w2 = std::dynamic_pointer_cast<Linear>(blocks["w2"]);
if (sd_backend_is(ctx->backend, "Vulkan") || sd_backend_is(ctx->backend, "ROCm")) {
w2->set_force_prec_f32(true);
}
w2->set_precision_flags(GGML_PRECISION_F32_VULKAN | GGML_PRECISION_F32_ROCM);
x = std::dynamic_pointer_cast<Linear>(blocks["w13"])->forward(ctx, x);
return w2->forward(ctx, ggml_swiglu(ctx->ggml_ctx, x));
}
@@ -206,7 +204,7 @@ namespace Fibo2 {
}
ggml_tensor* forward(GGMLRunnerContext* ctx, ggml_tensor* text, int gist_count) {
auto x = ggml_ext_slice(ctx->ggml_ctx, params["latent_queries"], 1, 0, gist_count);
auto x = ggml_ext_slice(ctx, params["latent_queries"], 1, 0, gist_count);
for (int i = 0; i < config.perceiver_depth; ++i) {
x = std::dynamic_pointer_cast<PerceiverLayer>(blocks["layers." + std::to_string(i)])->forward(ctx, x, text);
}
@@ -244,8 +242,8 @@ namespace Fibo2 {
return std::dynamic_pointer_cast<RMSNorm>(blocks[name])->forward(ctx, h);
};
auto modulation = std::dynamic_pointer_cast<Linear>(blocks["adaLN_modulation.0"]);
auto mn = ggml_ext_chunk(c, modulation->forward(ctx, noisy), 4, 0);
auto mc = ggml_ext_chunk(c, modulation->forward(ctx, clean), 4, 0);
auto mn = ggml_ext_chunk(ctx, modulation->forward(ctx, noisy), 4, 0);
auto mc = ggml_ext_chunk(ctx, modulation->forward(ctx, clean), 4, 0);
auto select = [&](int index, bool gate) {
auto n = gate ? ggml_tanh(c, mn[index]) : mn[index];
auto k = gate ? ggml_tanh(c, mc[index]) : mc[index];
@@ -311,12 +309,12 @@ namespace Fibo2 {
auto embedder = std::dynamic_pointer_cast<TimestepEmbedder>(blocks["t_embedder"]);
auto noisy = embedder->forward(ctx, t);
auto clean = embedder->forward(ctx, ggml_scale(c, t, 0.f));
auto image_pe = ggml_ext_slice(c, pe, 3, gist_count, pe->ne[3]);
auto image_pe = ggml_ext_slice(ctx, pe, 3, gist_count, pe->ne[3]);
for (int i = 0; i < config.refiner_depth; ++i) {
x = std::dynamic_pointer_cast<TransformerBlock>(blocks["noise_refiner." + std::to_string(i)])->forward(ctx, x, image_pe, noisy, clean, 0, image_count);
sd::ggml_graph_cut::mark_graph_cut(x, "fibo2.noise_refiner." + std::to_string(i), "x");
}
auto bundle = [&](int i) { return ggml_ext_slice(c, text, 0, i * config.text_dim, (i + 1) * config.text_dim); };
auto bundle = [&](int i) { return ggml_ext_slice(ctx, text, 0, i * config.text_dim, (i + 1) * config.text_dim); };
auto gist = std::dynamic_pointer_cast<Perceiver>(blocks["perceiver"])->forward(ctx, bundle(0), gist_count);
x = ggml_concat(c, gist, x, 1);
int bundle_index = 1;
@@ -326,7 +324,7 @@ namespace Fibo2 {
x = block->forward(ctx, x, pe, noisy, clean, gist_count, image_count, context);
sd::ggml_graph_cut::mark_graph_cut(x, "fibo2.layers." + std::to_string(i), "x");
}
x = ggml_ext_slice(c, x, 1, gist_count, gist_count + image_count);
x = ggml_ext_slice(ctx, x, 1, gist_count, gist_count + image_count);
x = std::dynamic_pointer_cast<ZImage::FinalLayer>(blocks["final_layer"])->forward(ctx, x, noisy);
x = ggml_cont(c, ggml_permute(c, x, 1, 0, 2, 3));
return ggml_reshape_4d(c, x, width, height, config.channels, 1);