nvn: reactive memory watchdog, 16-way view cache, and texture upload batching

dm/gfx/nvn/hle_nvn.c

@@ -26,9 +26,21 @@
 #include <TargetConditionals.h>
 #include <unistd.h>
 #include <sys/mman.h>
+#include <sys/sysctl.h>
 #include <errno.h>
 #if TARGET_OS_OSX
-#define DM_MEM_CAP_DEFAULT_GB 24
+static inline long dm_host_mem_cap_gb(void) {
+    uint64_t mem = 0; size_t len = sizeof mem;
+    if (sysctlbyname("hw.memsize", &mem, &len, NULL, 0) == 0 && mem > 0) {
+        long gb = (long)(mem >> 30);
+        if (gb >= 64) return gb - 16;   // 64 GB Mac -> 48 GB; 128 GB -> 112 GB
+        if (gb >= 36) return gb - 8;    // 36-48 GB Mac -> 28-40 GB
+        if (gb >= 16) return gb - 4;    // 16-24 GB Mac -> 12-20 GB
+        return (gb * 3) / 4;
+    }
+    return 24;
+}
+#define DM_MEM_CAP_DEFAULT_GB dm_host_mem_cap_gb()
 #else
 #define DM_MEM_CAP_DEFAULT_GB 0
 #endif

dm/gfx/nvn/nvn_mtl.m

@@ -3268,7 +3268,7 @@ static void rb_swap_rows(uint8_t *p, size_t wb, size_t rows, size_t nz, size_t r
                     if(nvn_deswz_gpu_direct_m4(g_up_ce,g_up_ct,params,poff,(__bridge id<MTLBuffer>)mb,mo,&t->d,l,view)){ok=0;break;}
                     g_direct_dispatches++;
                 }
-                if(ok){[g_up_texs addObject:mt];g_direct_uploads++;if(!g_cb)up_flush();return mt;}
+                if(ok){[g_up_texs addObject:mt];g_direct_uploads++;if(!g_cb && (g_direct_uploads % 32 == 0))up_flush();return mt;}
                 // Any encoded writes must finish before fallback copies replace
                 // them; both versions read the same source and write all texels.
                 [g_up_ce barrierAfterEncoderStages:MTLStageDispatch beforeEncoderStages:MTLStageDispatch visibilityOptions:MTL4VisibilityOptionDevice];
@@ -3290,7 +3290,7 @@ static void rb_swap_rows(uint8_t *p, size_t wb, size_t rows, size_t nz, size_t r
         }
         if (enc_ok) {
             [g_up_texs addObject:mt];
-            if (!g_cb) up_flush();
+            if (!g_cb && (g_up_texs.count >= 32)) up_flush();
             return mt;
         }
         g_nup_blits = nb0;   // Roll back the copies already recorded for this texture: the already-encoded deswizzle only writes the staging area and is harmless; this texture takes the CPU path below instead
@@ -6056,7 +6056,7 @@ static void treg_finish_submission(void) {
 // registrations ended in a new Metal texture view, 45% of the replay's time. A hit is checked against the current base texture (re-uploads and format-view upgrades replace it).
 // Only live textures are cached; entries go when the texture is finalized (mtl_tex_release) or its pool is (treg_drop_pool_views), and all of them past VC_MAX textures. Under g_rmu
 typedef struct { uint32_t pf, l0, nl, s0, ns; uint8_t tt, sw[4], pad[3]; } vc_key;
-enum { VC_WAYS = 4, VC_MAX = 16384 };
+enum { VC_WAYS = 16, VC_MAX = 16384 };
 typedef struct { const void *pool; unsigned next; vc_key k[VC_WAYS]; void *view[VC_WAYS];   // view: retained
                  void *rows, *rows_of; uint64_t rows_rid, rows_n; } vc_set;   // rows: a texture buffer as 2D rows (tbuf_as_2d; retained, resident), rows_of / rows_n: the buffer texture it was made from and its element count
 static nvn_hmap g_vc = {.klen = 8};   // texture serial → vc_set

docs/env_switches.tsv

@@ -96,7 +96,7 @@ DM_LONGFRAME	diagnostic	`0` turns off the long-frame log (on by default), a line
 DM_MEASURE_LOCK	path	`=<file>`: `scripts/regress.py`'s default `--lock`, the measuring lock `scripts/bench.py --lock` shares: held only while a game runs, and the harness's shader builds pause while another process holds it
 DM_MEMSTAT	diagnostic	`=<tsv>` writes one line per second of memory accounting: footprint, Metal allocation and resident and charged memory by region class (guest, code, malloc, gpu, …)
 DM_MEMSTAT_DUMP	diagnostic	`=<seconds>`: with `DM_MEMSTAT`, at that second writes every VM region of 1 MiB or more to `<tsv>.regions`
-DM_MEM_CAP_GB	switch	`=<GB>`: the memory watchdog exits when resident memory exceeds that many GB (default 24 on the Mac, none on iOS; `0` turns it off)
+DM_MEM_CAP_GB	switch	`=<GB>`: the memory watchdog exits when resident memory exceeds that many GB (default sized from host RAM on the Mac: 48 GB on 64 GB, 28-40 on 36-48 GB, 12 on 16 GB; none on iOS; `0` turns it off)
 DM_METAL_CAPTURE	diagnostic	`=1` makes `build_ipa.sh` produce a build that can write GPU frame captures (`MetalCaptureEnabled` in `Info.plist`), for `DM_GPU_CAPTURE`
 DM_METAL_HUD	player	`1` shows Apple's Metal performance HUD over the game (frame rate, frame and GPU time, memory)
 DM_MOTION	switch	`0` ignores live motion: every six-axis sensor lies flat as before (a shake still plays), and neither controllers' motion sensors nor the phone's device motion are switched on (`hle_hid.c`, `motion_apple.h`, `DmGamepad.m`); the NSP script's `--no-motion` (`MOTION = False`) builds it in