diff --git a/runtime/cfg.c b/runtime/cfg.c index 32d1a42f..360b5652 100644 --- a/runtime/cfg.c +++ b/runtime/cfg.c @@ -58,7 +58,7 @@ static const char *const known[] = { "SHADER_MEMO", "SHADOW_COMPARE", "SHADOW_RANGE", "SHADOW_RES", "SHADOW_SCALE", "SHARED_FONTS", "SHARPNESS", "SHOTS", "SHOTS_DIR", "SHOTS_FULL", "SPEED_PROBE", "SPIN", "SPIN_HANDOFF_US", "SPLIT_DRAWS", "SPLIT_US", "SPV_OVERRIDE", "STICK_DEADZONE", "STICK_SENS", "SWKBD_ANSWER", - "SET_AT", "SWKBD_TEST", "TAA", "TEST_PAD", "TEST_PAD_AT", "TEST_PREPARE_SHADERS", "TEX_ALIAS", "TEX_MEMO", "TEX_VRAM_PRESSURE", "TEX_RELEASED_MB", "TEX_COMMIT_FREE_MB", "TIMER_1MS", "TRACE", "TXD_TRANSPOSE", "UBO_SNAPSHOT", "UPSCALER", "VK_DYN_A2C", "VK_DYN_VERTEX", "VK_NOOPT", "VK_NO_FLOAT64", + "SET_AT", "SWKBD_TEST", "TAA", "TEST_PAD", "TEST_PAD_AT", "TEST_PREPARE_SHADERS", "TEX_ALIAS", "TEX_MEMO", "TEX_VRAM_PRESSURE", "TEX_RELEASED_MB", "TEX_COMMIT_FREE_MB", "VRAM_BUDGET_MB", "TIMER_1MS", "TRACE", "TXD_TRANSPOSE", "UBO_SNAPSHOT", "UPSCALER", "VK_DYN_A2C", "VK_DYN_VERTEX", "VK_NOOPT", "VK_NO_FLOAT64", "VK_PRESENT", "VK_RECONVERGE", "VK_RECORDER", "VK_SSBO_ALIGN", "VK_SUBGROUP", "VK_SUBGROUP_PIN", "VK_VALIDATION", "VOLUME", "VRAM_HEADROOM", "VS_OUTS", "WALK_KEY", "WALK_SCALE", "WB_EXACT", "WB_FENCE_LIMIT", "WB_MERGE", "WINDOW_SIZE", "XLATE", "XLATE_ID", "XLATE_KEEP", "XLATE_PROCS", "ZSTD_CHECK", "ZSTD_NATIVE", diff --git a/runtime/vk_memory.c b/runtime/vk_memory.c index e01515fc..5c425e66 100644 --- a/runtime/vk_memory.c +++ b/runtime/vk_memory.c @@ -624,7 +624,21 @@ void vk_mem_census(uint64_t frame) { window_start = frame; } -int vk_mem_vram(uint64_t *used, uint64_t *budget, int *exact) { +// NX2NAVI_VRAM_BUDGET_MB (test; unset or 0: the device's): the device-local budget every decision here and in +// vk_texture.c compares with is at most this many MiB, so a large card behaves as a small one does under pressure +// (mirrors refused or evicted, textures dropped). The heap lines still log what the driver reports. +static uint64_t budget_cap(void) { + static int64_t mb = -1; + if (mb < 0) { + const char *s = cfg_get("NX2NAVI_VRAM_BUDGET_MB"); + int64_t v = s && *s ? atoll(s) : 0; + if (v > 0) rt_log("mem: the VRAM budget is taken as at most %lld MiB (NX2NAVI_VRAM_BUDGET_MB, a test)", (long long)v); + mb = v > 0 ? v : 0; + } + return (uint64_t)mb << 20; +} + +static int vram_query(uint64_t *used, uint64_t *budget, int *exact) { VkCtx *c = vk_ctx(); *used = *budget = 0, *exact = 0; if (!c) return 0; @@ -647,6 +661,13 @@ int vk_mem_vram(uint64_t *used, uint64_t *budget, int *exact) { return *budget != 0; } +int vk_mem_vram(uint64_t *used, uint64_t *budget, int *exact) { + int ok = vram_query(used, budget, exact); + uint64_t cap = budget_cap(); + if (ok && cap && *budget > cap) *budget = cap; + return ok; +} + // ---------------------------------------------------------------- page write generations // Two-level table over the guest address space (below 2^39): 256 MiB chunks of 65,536 page generations, allocated // when a page of the chunk is first stamped. 0 = never stamped.