Files
NX2navi/runtime/vk_fg.cpp
chrg6 501d5bd084 Release pass: fix crashes, freezes and races, cut VRAM use, and translate more shaders.
Crashes: the render-target heap (HeapDef DynamicTextureAllocatorMem and its parent) grows with the widescreen
picture, so the pause menu no longer reads a null block at 5120x2160; texture uploads keep the game's memory mapped
while they copy it, so UnmapPhysicalMemory no longer pulls it away mid-copy; DLSS Frame Generation no longer crashes
in Streamline's set-up (physical devices and the window surface go through its proxies, its swapchain images are
counted first, resources are tagged per frame); a stale core-wait target is cleared before its thread is freed.

Freezes and races: a resize during a swapchain rebuild is kept, a failed frame generation swapchain is retried
instead of replaced by a plain one, three acquire timeouts in a row rebuild the swapchain, device waits take the
queue lock, a translator child that stops answering times out, an in-flight compute pipeline builds at normal
priority, a save Commit started before the window closes finishes and none starts after, a settings change just
before closing is saved, and a movie sample taken before a flush is dropped.

HDR: Enhanced HDR extends the game's own tonemap so highlights go above the picture's white (the hue slider,
which no longer applies, is gone from the menu); frame generation gets a HUD-less picture, the HDR flag and the HDR10
metadata under HDR too.

Memory: ASTC textures are encoded to BC3 unless the PC has 20 GiB of VRAM and 48 GiB of RAM, as Windows charges video
memory to the commit charge (a long session held 17 GiB of RGBA8 textures and 45.7 GiB of commit on a 32 GiB PC).

Shaders: the translator lowers spill-mode programs with more than eight uniform register groups in a wider spill
layout, and local loads and stores with a uniform-register address (36029 of the game's 36031 variants translate,
20 failed before).

The launcher release ships the shader tools and AMD's FidelityFX library (make_launcher.py --ffx); windres there gets
the toolchain on PATH.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-07 16:02:30 -05:00

248 lines
11 KiB
C++

// Frame generation with AMD FSR 3.1 (NX2NAVI_FRAMEGEN = fsr3; applies after a restart): AMD's frame generation
// swapchain (amd_fidelityfx_vk.dll, vk_ffx.cpp loads it) takes the place of gfx_present.c's swapchain. Every frame
// the game presents, the swapchain presents a generated frame between it and the one before, on its own present
// queue and pacing threads, so the screen shows twice the game's frame rate.
//
// The generated frame is built from the presented picture and the scene's depth and camera motion vectors, which
// vk_dlaa.c makes for DLSS and FSR 3.1 upscaling: frame generation needs one of those on (DLAA or Native AA included).
// The HUD is interpolated with the picture (it hardly moves); the menu is not, since generation pauses while it is
// open. The device creates two more queues of the graphics family for it (vk_device.c, vk_fg_queue_count).
#ifdef _WIN32
#include <windows.h>
#endif
#include <stdbool.h>
#include <stdio.h>
#include <string.h>
#include <wchar.h>
#include "third_party/ffx_api/ffx_api.h"
#include "third_party/ffx_api/ffx_framegeneration.h"
#include "third_party/ffx_api/vk/ffx_api_vk.h"
extern "C" {
#include "cfg.h"
#include "host.h"
#include "platform/platform.h"
#include "vk.h"
#include "vk_dlaa.h"
#include "vk_ffx.h"
#include "vk_fg.h"
#include "ui/menu.h"
}
namespace {
struct {
int wanted = -1;
VkQueue present = VK_NULL_HANDLE, acquire = VK_NULL_HANDLE;
ffxContext sc = nullptr; // the swapchain context
ffxContext fg = nullptr; // the frame generation context
VkExtent2D fg_ext = {0, 0};
VkFormat fg_fmt = VK_FORMAT_UNDEFINED;
ffxQueryDescSwapchainReplacementFunctionsVK fn = {};
uint64_t frame = 0;
int failed = 0;
uint64_t generated = 0, paused = 0, no_inputs = 0, hudless = 0;
} F;
VkResult submit_game(uint32_t n, const VkSubmitInfo *si, VkFence fence) {
vk_queue_lock(); // the game's queue is shared with the recorder and the present path
VkResult r = vkQueueSubmit(vk_ctx()->gfx_queue, n, si, fence);
vk_queue_unlock();
return r;
}
ffxReturnCode_t generate(ffxDispatchDescFrameGeneration *d, void *user) {
F.generated++;
return vk_ffx_dispatch((ffxContext *)user, &d->header);
}
FfxApiResource image(const VkFfxImage *i) {
VkImageCreateInfo ci = {};
ci.sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO;
ci.imageType = VK_IMAGE_TYPE_2D;
ci.format = i->format;
ci.extent = {i->width, i->height, 1};
ci.mipLevels = 1;
ci.arrayLayers = 1;
return ffxApiGetResourceVK((void *)i->image, ffxApiGetImageResourceDescriptionVK(i->image, ci, 0),
FFX_API_RESOURCE_STATE_UNORDERED_ACCESS);
}
void fail(const char *what, unsigned r) {
rt_log("fg: %s failed (%u): frame generation off for this run", what, r);
F.failed = 1;
}
} // namespace
extern "C" int vk_fg_wanted(void) {
if (F.wanted < 0) {
F.wanted = !strcmp(cfg_value("NX2NAVI_FRAMEGEN"), "fsr3");
if (F.wanted && !vk_dlaa_wanted()) {
rt_log("fg: frame generation needs DLSS or FSR 3.1 upscaling (any mode): off");
F.wanted = 0;
}
}
return F.wanted;
}
extern "C" uint32_t vk_fg_queue_count(void) { return vk_fg_wanted() ? 3 : 1; }
extern "C" void vk_fg_queues(VkQueue present, VkQueue acquire) { F.present = present, F.acquire = acquire; }
extern "C" int vk_fg_active(void) { return F.sc != nullptr && !F.failed; }
extern "C" VkResult vk_fg_create_swapchain(const VkSwapchainCreateInfoKHR *ci, VkSwapchainKHR *sc) {
if (F.sc) return F.fn.pOutCreateSwapchainFFXAPI(vk_ctx()->device, ci, nullptr, sc, F.sc);
if (!vk_fg_wanted() || F.failed || !F.present || !vk_ffx_load()) return VK_ERROR_FEATURE_NOT_PRESENT;
const VkCtx *c = vk_ctx();
ffxCreateContextDescFrameGenerationSwapChainVK d = {};
d.header.type = FFX_API_CREATE_CONTEXT_DESC_TYPE_FGSWAPCHAIN_VK;
d.physicalDevice = c->phys;
d.device = c->device;
VkSwapchainKHR none = VK_NULL_HANDLE;
d.swapchain = &none;
d.createInfo = *ci;
d.createInfo.oldSwapchain = VK_NULL_HANDLE;
d.gameQueue = {c->gfx_queue, c->gfx_family, submit_game};
d.asyncComputeQueue = {VK_NULL_HANDLE, c->gfx_family, nullptr};
d.presentQueue = {F.present, c->gfx_family, nullptr};
d.imageAcquireQueue = {F.acquire, c->gfx_family, nullptr};
ffxReturnCode_t r = vk_ffx_create(&F.sc, &d.header);
if (r != FFX_API_RETURN_OK) {
F.sc = nullptr;
fail("the frame generation swapchain", r);
return VK_ERROR_INITIALIZATION_FAILED;
}
F.fn.header.type = FFX_API_QUERY_DESC_TYPE_FGSWAPCHAIN_FUNCTIONS_VK;
if (vk_ffx_query(&F.sc, &F.fn.header) != FFX_API_RETURN_OK || !F.fn.pOutQueuePresentKHR) {
fail("the swapchain functions", 0);
return VK_ERROR_INITIALIZATION_FAILED;
}
*sc = none;
rt_log("fg: AMD FSR 3.1 frame generation swapchain, %ux%u", ci->imageExtent.width, ci->imageExtent.height);
return VK_SUCCESS;
}
extern "C" void vk_fg_destroy_swapchain(VkSwapchainKHR sc) {
if (F.sc && sc) F.fn.pOutDestroySwapchainFFXAPI(vk_ctx()->device, sc, nullptr, F.sc);
}
extern "C" VkResult vk_fg_get_images(VkSwapchainKHR sc, uint32_t *n, VkImage *img) {
return F.fn.pOutGetSwapchainImagesKHR(vk_ctx()->device, sc, n, img);
}
extern "C" VkResult vk_fg_acquire(VkSwapchainKHR sc, uint64_t timeout, VkSemaphore sem, uint32_t *idx) {
return F.fn.pOutAcquireNextImageKHR(vk_ctx()->device, sc, timeout, sem, VK_NULL_HANDLE, idx);
}
extern "C" VkResult vk_fg_present(const VkPresentInfoKHR *pi) {
return F.fn.pOutQueuePresentKHR(vk_ctx()->gfx_queue, pi);
}
extern "C" uint64_t vk_fg_presents(VkSwapchainKHR sc) {
return vk_fg_active() && F.fn.pOutGetLastPresentCountFFXAPI && sc ? F.fn.pOutGetLastPresentCountFFXAPI(sc) : 0;
}
// The HDR10 metadata goes to AMD's swapchain through its own call (vk_hdr.c vk_hdr_metadata).
extern "C" PFN_vkSetHdrMetadataEXT vk_fg_hdr_metadata_fn(void) {
return vk_fg_active() ? F.fn.pOutSetHdrMetadataEXT : nullptr;
}
// An HDR10 swapchain (10-bit PQ or 16-bit float): frame generation is told its inputs are HDR.
static bool hdr_format(VkFormat f) {
return f == VK_FORMAT_A2B10G10R10_UNORM_PACK32 || f == VK_FORMAT_A2R10G10B10_UNORM_PACK32 ||
f == VK_FORMAT_R16G16B16A16_SFLOAT;
}
extern "C" void vk_fg_frame(VkCommandBuffer cb, VkSwapchainKHR sc, VkFormat fmt, VkExtent2D ext, VkImage hudless,
VkImageView hudless_view) {
(void)hudless_view;
if (!vk_fg_active()) return;
vk_dlaa_hudless_wanted = 1; // the upscaling pass keeps a copy of the picture before the HUD (vk_dlaa.c hudless)
const VkCtx *c = vk_ctx();
if (F.fg && (F.fg_ext.width != ext.width || F.fg_ext.height != ext.height || F.fg_fmt != fmt)) {
vk_queue_lock(), vkDeviceWaitIdle(c->device), vk_queue_unlock(); // a new window size: the context is made for its display size
vk_ffx_destroy(&F.fg);
F.fg = nullptr;
}
if (!F.fg) {
ffxCreateBackendVKDesc be = {};
be.header.type = FFX_API_CREATE_CONTEXT_DESC_TYPE_BACKEND_VK;
be.vkDevice = c->device;
be.vkPhysicalDevice = c->phys;
be.vkDeviceProcAddr = vkGetDeviceProcAddr;
// the HUD-less picture has the swapchain's format (gfx_present.c hudless_record)
ffxCreateContextDescFrameGenerationHudless hl = {};
hl.header.type = FFX_API_CREATE_CONTEXT_DESC_TYPE_FRAMEGENERATION_HUDLESS;
hl.header.pNext = &be.header;
hl.hudlessBackBufferFormat = ffxApiGetSurfaceFormatVK(fmt);
ffxCreateContextDescFrameGeneration d = {};
d.header.type = FFX_API_CREATE_CONTEXT_DESC_TYPE_FRAMEGENERATION;
d.header.pNext = &hl.header;
d.flags = FFX_FRAMEGENERATION_ENABLE_DEPTH_INFINITE | (hdr_format(fmt) ? FFX_FRAMEGENERATION_ENABLE_HIGH_DYNAMIC_RANGE : 0);
d.displaySize = {ext.width, ext.height};
d.maxRenderSize = {ext.width, ext.height};
d.backBufferFormat = ffxApiGetSurfaceFormatVK(fmt);
ffxReturnCode_t r = vk_ffx_create(&F.fg, &d.header);
if (r != FFX_API_RETURN_OK) return F.fg = nullptr, fail("the frame generation context", r);
F.fg_ext = ext, F.fg_fmt = fmt;
rt_log("fg: frame generation context for %ux%u%s", ext.width, ext.height, hdr_format(fmt) ? " (HDR)" : "");
}
VkFfxImage depth, mv;
float jx = 0, jy = 0;
int reset = 0;
int have = vk_dlaa_fg_inputs(&depth, &mv, &jx, &jy, &reset);
int on = have && !menu_is_open();
if (!have) F.no_inputs++;
else if (!on) F.paused++;
F.frame++;
ffxConfigureDescFrameGeneration cfg = {};
cfg.header.type = FFX_API_CONFIGURE_DESC_TYPE_FRAMEGENERATION;
cfg.swapChain = (void *)sc;
cfg.frameGenerationCallback = generate;
cfg.frameGenerationCallbackUserContext = &F.fg;
cfg.frameGenerationEnabled = on != 0;
cfg.allowAsyncWorkloads = false;
cfg.generationRect = {0, 0, (int32_t)ext.width, (int32_t)ext.height};
if (on && hudless) {
VkFfxImage h = {hudless, fmt, ext.width, ext.height, 1};
cfg.HUDLessColor = image(&h);
F.hudless++;
}
cfg.frameID = F.frame;
ffxReturnCode_t r = vk_ffx_configure(&F.fg, &cfg.header);
if (r != FFX_API_RETURN_OK) return fail("frame generation configuration", r);
if (!on) return;
ffxDispatchDescFrameGenerationPrepare p = {};
p.header.type = FFX_API_DISPATCH_DESC_TYPE_FRAMEGENERATION_PREPARE;
p.frameID = F.frame;
p.commandList = cb;
p.renderSize = {depth.width, depth.height};
p.jitterOffset = {jx, jy};
p.motionVectorScale = {1.0f, 1.0f}; // render pixels, current to previous (vk_dlaa.c mv_pass), as FSR takes them
static uint64_t last_ns;
uint64_t now = plat_now_ns();
double dt = last_ns ? (now - last_ns) / 1e6 : 16.67;
last_ns = now;
p.frameTimeDelta = dt > 0 && dt < 100 ? (float)dt : 16.67f;
p.cameraNear = 0.1f; // Ocarina's projection: near 0.1, far at infinity (vk_ffx.cpp)
p.cameraFar = 1e8f;
p.cameraFovAngleVertical = 1.0f;
p.viewSpaceToMetersFactor = 0.01f;
p.depth = image(&depth);
p.motionVectors = image(&mv);
r = vk_ffx_dispatch(&F.fg, &p.header);
if (r != FFX_API_RETURN_OK) return fail("frame generation preparation", r);
if (F.frame == 2 || F.frame % 3600 == 0)
rt_log("fg: %llu frames presented, %llu generated (%llu with the HUD kept out); paused %llu (menu), %llu without depth "
"and motion", (unsigned long long)F.frame, (unsigned long long)F.generated, (unsigned long long)F.hudless,
(unsigned long long)F.paused, (unsigned long long)F.no_inputs);
}
extern "C" void vk_fg_release(void) {
if (F.fg) vk_ffx_destroy(&F.fg), F.fg = nullptr;
if (F.sc) vk_ffx_destroy(&F.sc), F.sc = nullptr;
}