Merge main into the test line: the pull requests taken there and two render target fixes.

Main brought the fixes of eleven pull requests (the texture pool set sealed when an early replay part is submitted, every location reflected in interface arrays, RGB9E5 texels recognised for integer storage views, the upscaler a frame started with kept, GPU pass timestamps without vendor breadcrumbs, the Linux build, window and NGX fixes), render targets released when drawn over at another address, the vertex layout kept in the pipeline on newer AMD drivers, and the upscaler choice read once per evaluation.

Four hunks conflicted and take both sides: pool_sync returns for a sealed set before the per-frame creation budget's clock (vk_exec_draw.c); make_module keeps the test line's early_ok argument and main's module hash (vk_pipeline.c); a texture taken back counts as sampled again and records how long it was away, and the texture line logs the disuse counters and the overwritten render targets (vk_texture.c).

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
This commit is contained in:
chrg6
2026-11-09 21:10:23 -05:00
27 changed files with 1205 additions and 144 deletions

View File

@@ -873,7 +873,8 @@ static void set_rt(VkExec *ex, const p_set_rt *r) {
// command-buffer state is lost there (bindings and dynamic state: vk_exec_draw_new_cb, ex->dirty); barriers, layouts and
// queries are not: a barrier's first scope is everything earlier in submission order, image layouts belong to the image,
// and occlusion queries end with their pass. Fences, completion callbacks and reports stay per submission: only the last
// part signals the value.
// part signals the value. The texture pool set a part binds is sealed when the part goes (vk_exec_pool_split): later
// parts write and bind another, and a part is not split off while no other set is free of this submission.
// A part also spans at least NX2NAVI_SPLIT_US (1000) of replay time: a split costs about 40 us of the render thread
// (vkQueueSubmit2, a command buffer, the mirror prologue), and a fight frame of 2,000 draws made 15 parts on the 7900X;
// a millisecond of recording is a fraction of a millisecond of GPU work, so the GPU still starts early.
@@ -925,8 +926,8 @@ static void maybe_split(VkExec *ex) {
}
split_draws = 0;
split_at = split_n;
if (idle && ex->recorded != part_start) n_idle_parts++;
if (ex->recorded == part_start) return;
if (ex->recorded == part_start || !vk_exec_pool_split(ex->fence)) return;
if (idle) n_idle_parts++;
if (ex->in_pass) vk_exec_pass_end(ex), n_pass_splits++; // the next draw begins the pass again, in the new part
ex->cb = vk_sync_split();
part_start = ex->recorded;