Lynne 720bc0b916 vulkan/seg_gather: avoid scratch spills and serial prefix summing ...

Compilers lowered the funnel indexer to scratch (spilling the data),
during the tight copy loop into something that spills every load.
Just use a case to select from the 2 source vectors.

Summing the offsets at the end took far longer than expected and
blocked all invocs. Just use a trivial subgroup add + atomic add
to finish the sum quickly and with less contention.

2026-07-12 13:34:57 +09:00

..

apv_decode.comp.glsl

…

apv_encode_dct.comp.glsl

…

apv_encode_tiles.comp.glsl

…

apv_idct.comp.glsl

…

common.glsl

…

dct.glsl

…

dpx_copy.comp.glsl

…

dpx_unpack.comp.glsl

…

ffv1_common.glsl

…

ffv1_dec_bayer_golomb.comp.glsl

…

ffv1_dec_bayer.comp.glsl

…

ffv1_dec_golomb.comp.glsl

…

ffv1_dec_reset_golomb.comp.glsl

…

ffv1_dec_reset.comp.glsl

…

ffv1_dec_rgb_float_golomb.comp.glsl

…

ffv1_dec_rgb_float.comp.glsl

…

ffv1_dec_rgb_golomb.comp.glsl

…

ffv1_dec_rgb.comp.glsl

…

ffv1_dec_setup.comp.glsl

…

ffv1_dec.comp.glsl

…

ffv1_enc_bayer_golomb.comp.glsl

…

ffv1_enc_bayer.comp.glsl

…

ffv1_enc_golomb.comp.glsl

…

ffv1_enc_rct_search.comp.glsl

…

ffv1_enc_remap.comp.glsl

…

ffv1_enc_reset_golomb.comp.glsl

…

ffv1_enc_reset.comp.glsl

…

ffv1_enc_rgb_float_golomb.comp.glsl

…

ffv1_enc_rgb_float.comp.glsl

…

ffv1_enc_rgb_golomb.comp.glsl

…

ffv1_enc_rgb.comp.glsl

…

ffv1_enc_setup.comp.glsl

…

ffv1_enc_sort32.comp.glsl

…

ffv1_enc.comp.glsl

…

ffv1_vlc.glsl

…

Makefile

…

prores_idct.comp.glsl

…

prores_ks_alpha_data.comp.glsl

…

prores_ks_encode_slice.comp.glsl

…

prores_ks_estimate_slice.comp.glsl

…

prores_ks_slice_data.comp.glsl

…

prores_ks_trellis_node.comp.glsl

…

prores_raw_decode.comp.glsl

…

prores_raw_idct.comp.glsl

…

prores_vld.comp.glsl

…

rangecoder.glsl

…

seg_gather.comp.glsl

…