On a 6 GB card (RTX 3060 Laptop, RTX 2060) Auto gave the game's memory pools an
eighth of the card (768 MB) and the Low (6GB) preset, every 6 GB card's from its
first start, 512 MB. The game's largest pools then stayed in system memory and
the GPU read them over PCIe.

- gpu/vk/mirror_budget.h: Auto's rule (a quarter, at most 3 GB, an eighth only
  on cards of 4.5 GB or less) and a preset's cap (never more than Auto's).
- gpu/vk/vk_memory.cpp: mirror_budget() uses them.
- gpu/vk/vk_menu.cpp: the 3.6 GB VRAM budget step (Low) mirrors 1536 MB.
- tests/test_mirror_budget.{py,cpp}: the rules and their use; the VRAM budget
  row check updated.

Ported from https://github.com/NotRexed/Hookshot/pull/2834

Co-Authored-By: Claude Opus 5.5 <[email protected]>

@@ -0,0 +22,4 @@
+/// The mirrors when a Quality preset set the budgets (by_preset): never more than Auto gives the card, so a preset made for a
+/// larger card (Low's 1536 MB on a 4 GB one, preset_caps_memory lets Low's budgets through up to 6.75 GB) keeps a small card at
+/// Auto's. A budget chosen on the VRAM budget row itself is the player's on any card, as chosen
+inline uint64_t preset_bytes(uint64_t asked, bool by_preset, uint64_t vram) {

@@ -2486,3 +2483,1 @@
-        // percent and the frame stuttering)
-        if (vram && vram <= (6ull << 30) + (512ull << 20)) return vram / 8;
-        return vram ? std::min<uint64_t>(vram / 4, 3072ull << 20) : (uint64_t)(3072ull << 20);
+        return mirror_rule::auto_bytes(device_vram_bytes());

@@ -853,0 +854,4 @@
+// PCIe, 30-32 GB of bound ranges every 600 submits in Hyrule Field on an RTX 3060 Laptop (6 GB), the GPU's frame 27 ms. A preset's
+// mirrors are never more than Auto's on the card (mirror_budget.h preset_bytes): Low on a 4 GB card, which preset_caps_memory lets
+// through, keeps 512
+constexpr int kVramMirrorMb[kVramBudgets] = {0, 256, 1536, 1024, 1536, 2048, 2560, 3072};

@@ -0,0 +23,4 @@
+    CHECK(preset_bytes(256 * MB, true, 4 * GB) == 256 * MB);    // (Minimum: its own, smaller)
+    CHECK(preset_bytes(1024 * MB, true, 8 * GB) == 1024 * MB);  // (Balanced on an 8 GB card: its own)
+    CHECK(preset_bytes(3072 * MB, true, 16 * GB) == 3072 * MB); // (Ultra on a 16 GB card)
+    CHECK(preset_bytes(1536 * MB, false, 4 * GB) == 1536 * MB); // (3.6 GB chosen on the row on a 4 GB card: the player's)

@@ -0,0 +30,4 @@
+/// a small card at Auto's. Chosen on the VRAM budget row itself: the player's, except on a small card, where it is capped at Auto's
+/// too (the row's 3.6 GB step on a 4 GB card was 512 MB before Low's mirrors went to 1536; its steps of 4.8 GB and more, past
+/// what such a card has, now keep 512 as well)
+inline uint64_t preset_bytes(uint64_t asked, bool by_preset, uint64_t vram) {

@@ -251,3 +251,3 @@
 | `RX_VK_NATIVE_FADD=0` or `1` | The shaders' float adds to storage buffers: 1 the GPU's own atomic where it has `VK_EXT_shader_atomic_float`'s `shaderBufferFloat32AtomicAdd`, 0 a compare-exchange loop on every GPU, the extension not enabled. By default the GPU's own on NVIDIA's driver only, the one it was tried on in the game; on AMD and Intel the loop until runs there show the native add works (the log's `vk: float atomic add on storage buffers:` line says which, and why). The loop gives up after 1024 attempts, and an add that gives up is lost |
 | `RX_VK_UB_WORD_HASH=0` | A/B test: use the old low-address-bit index for cached uniform words (default: mix upper address bits to reduce aligned-buffer collisions) |
-| `RX_VK_MIRROR=0`, `RX_VK_MIRROR_MB=<MB>` | No VRAM mirrors of the game's memory pools (on for discrete cards); their budget |
+| `RX_VK_MIRROR=0`, `RX_VK_MIRROR_MB=<MB>` | No VRAM mirrors of the game's memory pools (on for discrete cards); their budget (by default a quarter of the card, at most 3 GB, an eighth on cards of 4.5 GB or less, and a Quality preset's never more than that: `gpu/vk/mirror_budget.h`) |

@@ -854,3 +854,3 @@
 constexpr int kVramMinimumOnly = 1;   // (2.2 GB: set by the Minimum (4GB) preset only; the row's own steps go past it)
 constexpr int kVramTextureMb[kVramBudgets] = {0, 2200, 3600, 4800, 6144, 7500, 9216, 11000};
-constexpr int kVramMirrorMb[kVramBudgets] = {0, 256, 512, 1024, 1536, 2048, 2560, 3072};
+// (longshot) 3.6 GB's mirrors, the Low (6GB) preset's and so every 6 GB card's from its first start (menu_first_start_for_gpu): 1536

@@ -0,0 +9,4 @@
+import tempfile
+from pathlib import Path
+
+RUN_ALL_ARGS = ["--cxx", "g++"]