From 77a52e457719c23dd9921839170f01e120f2afeb Mon Sep 17 00:00:00 2001 From: ipavlin98 Date: Mon, 16 Nov 2026 03:55:56 +0300 Subject: [PATCH] TOTK_WB_MERGE=0 also turns off the shadow storage's word merge, so the two can be compared --- docs/settings.md | 2 +- runtime/vk_memory.c | 22 ++++++++++++++-------- 2 files changed, 15 insertions(+), 9 deletions(-) diff --git a/docs/settings.md b/docs/settings.md index 1b0193fe..36318fbc 100644 --- a/docs/settings.md +++ b/docs/settings.md @@ -153,7 +153,7 @@ Many rows below are developer tools that only act together with another setting | `vram_headroom` (default `1`) | `0` turns off the spare half-block of video memory that a background thread keeps per memory type. | Texture allocations on the render thread can wait for a new block again. | Caution | | `vs_outs` | `all` declares all 32 vertex shader output locations; otherwise only the union of the paired fragment inputs. | Larger vertex shaders and new pipelines; unwritten outputs read the same as before. | Caution | | `wb_exact` (default `1`) | `0` decides write-back of GPU-written textures by page only. | A target that shares a page with data the CPU writes can go unwritten, so a name label next to the reticle can blink. | Caution | -| `wb_merge` (default `1`) | `0` copies GPU-written ranges back as plain copies instead of writing word by word the words that changed. | In mirror memory mode the GPU's copy can overwrite words the CPU wrote, which breaks effects such as falling rain and waterfall splashes. | Caution | +| `wb_merge` (default `1`) | `0` copies GPU-written ranges back as plain copies instead of writing word by word the words that changed; with `shadow = 1` it also leaves ranges the GPU has yet to write back out of the copy into the shadow. | In mirror memory mode, and with `shadow = 1`, the GPU's copy can overwrite words the CPU wrote, which breaks effects such as falling rain and waterfall splashes. | Caution | ## Motion vectors and DLSS diff --git a/runtime/vk_memory.c b/runtime/vk_memory.c index 21c3f7b7..8329abb0 100644 --- a/runtime/vk_memory.c +++ b/runtime/vk_memory.c @@ -996,7 +996,8 @@ static int query_stamp(uint64_t lo, uint64_t hi, uint32_t g) { // spawns, and a CPU copy that skipped pending ranges never let them reach the GPU: falling rain showed only its // splashes. Each pending range keeps a base, the words guest memory and the shadow last agreed on (the shadow's bytes // when the range became pending). The CPU copy takes the words where guest memory differs from the base; the fence -// copy back takes the words where the shadow differs from it. A word both changed takes the GPU's. +// copy back takes the words where the shadow differs from it. A word both changed takes the GPU's. TOTK_WB_MERGE=0 +// skips the pending ranges in the CPU copy and copies them back whole, as before. // TOTK_SHADOW=1/0 (shadow in totk.ini) forces the mode on or off; the default is on for Linux and Wine. typedef struct { uint64_t lo, hi; uint8_t *base; int refs; } SRange; static PlatMutex shadow_lock = PLAT_MUTEX_INIT; // after pool_lock; before gen_lock @@ -1068,6 +1069,16 @@ static void shadow_forget(PoolRec *r) { plat_mutex_unlock(&shadow_lock); } +// TOTK_WB_MERGE: the write-back merge of mirrors (wb_merge_record) and the word merge of the shadow's pending ranges. +static int merge_on = -1; +static int merge_wanted(void) { + if (merge_on < 0) { + merge_on = cfg_bool("TOTK_WB_MERGE"); + rt_log("mem: write-back merge %s (TOTK_WB_MERGE)", merge_on ? "on" : "off"); + } + return merge_on; +} + // A run of differing words from *at on, as byte offsets [*at, *end) of x and y (n bytes, the last word may be shorter); // 0 when none is left. static int diff_run(const uint8_t *x, const uint8_t *y, uint64_t n, uint64_t *at, uint64_t *end) { @@ -1311,7 +1322,7 @@ void vk_mem_gpu_pending(uint64_t addr, uint64_t size) { spend[i].refs++; } else { uint8_t *base = NULL; - if (r && r->smap && lo >= r->import_lo && hi <= r->import_hi && (base = malloc(size))) { + if (r && r->smap && lo >= r->import_lo && hi <= r->import_hi && merge_wanted() && (base = malloc(size))) { memcpy(base, r->smap + (lo - r->import_lo), size); for (int k = 0; k < nspend; k++) { const SRange *p = &spend[k]; @@ -1893,7 +1904,6 @@ static PlatMutex snap_pool_lock = PLAT_MUTEX_INIT; static SnapChunk snap_pool[SNAP_POOL_N]; static int nsnap_pool; static uint64_t snap_pool_bytes; -static int merge_on = -1; static struct { int tried, ok; PFN_vkCmdPushDescriptorSetKHR push; @@ -1936,11 +1946,7 @@ fail: } static int merging(void) { - if (merge_on < 0) { - merge_on = cfg_bool("TOTK_WB_MERGE"); - rt_log("mem: write-back merge %s (TOTK_WB_MERGE)", merge_on ? "on" : "off"); - } - if (!merge_on) return 0; + if (!merge_wanted()) return 0; if (!WM.tried) merge_init(vk_ctx()); return WM.ok; }