llama.cpp fixes conv_transpose_2d computing only the first batch on both CPU and Metal

GitHub·medium signalXBlueskyLinkedInCopy link

Build b10676 (2026-08-28) fixes `ggml_compute_forward_conv_transpose_2d_impl`, which only computed the first batch of the destination tensor and left every subsequent batch as zero; both the src1 permutation and the main compute loop now iterate the batch dimension, and `ggml_graph_plan` scales the work buffer by the src1 batch count. The Metal kernel had the identical defect, decoding no batch from the grid, so multi-batch results diverged from the CPU reference; the grid now covers OW x OH x OC x N with nb3 passed in kernel args. A multi-batch case was added to test-backend-ops, which is what should have caught it.