[W1nge]

Reuse the prompt embedding allocation for the later half output (bo). Embedding gathers, image-row uploads, the last embedding consumer (gr_broadcast), output projections, MoE combine and HC writes are ordered on m.cs. A following chunk reloads its embeddings; a stage receiving hand_in_ does not read emb. PLE embeddings retain their separate allocation.

Remove the corresponding bo allocation from bytes_needed_impl as well as carve. This saves chunk * 2560 * sizeof(float) payload bytes (320 MiB at 32768 rows), with allocator rounding handled by the existing allocator. No new flags, default chunk changes, or numerical kernels. Independent of #1451.

Validation: reviewed all m.emb/m.bo accesses in current main d5ea713; prefill.cpp compiles with MSVC 19.44/CUDA 12.4 headers both without native/MMQ and with native/MMQ/fused enabled; git diff --check passes.

Draft: this alias was part of the locally validated 32K fork, but this isolated upstream port has not had a full-model, peer/stage or other-backend runtime comparison. The allocation saving is not an isolated throughput measurement. No broader HC scratch overlay is included.

Validation update (2026-10-08): a complete current-main integration build including #1451, #1454, #1455, #1457 and #1465 passed four real IQ3_XXS model baseline/integration pairs on RTX 2080 Ti: default, GR_UNFUSED, PREFILL_BF16X2 and legacy RING_BYTES modes. Each used INT8 KV, 194 input tokens, two prefill chunks and 8 output tokens; all runs exited successfully and output IDs matched within each pair. A 256K configured-context startup and short generation also passed (this did not fill a 256K prompt). These are combined regression checks, not an isolated performance result or exhaustive state equality. Other hardware/backends remain untested. The earlier draft-only status reflected the absence of any current-main model run; now requesting review with these limits explicit.

[Niko1221]

Thanks, W1nge. The VRAM reuse is in 0.1.41, with your authorship kept: the half output now shares the embedding storage, about 320 MiB less at 32,768 rows, checked identical on one GPU and on a 2-GPU layer split. On a VRAM-limited card the saved bytes would also let the prompt planner pick a slightly larger chunk (an RTX 3060 went from 6,400 to 6,656 tokens, about +3.9% prompt speed), but that changes the prompt's rounding, so by default the planner keeps the old accounting and the output stays byte-identical. STRATA_EMB_REUSE_ACCOUNT=1 turns the planner gain on.

Closing as ported.