Files
ARMSX3/3rdparty/lsfg/armsx3_lsfg_shim.h
T
jpolo1224 43dcf6d0ca Framegen: wait on a sync fd instead of idling framegen's whole device
Every generated frame ended with vkDeviceWaitIdle on framegen's device. It is
the hardest sync primitive Vulkan has and it ran once per frame, which is a
large part of why frame generation cost more than it returned.

It was there because the two devices share no semaphore. The upstream
semaphore path cannot supply one on Android: framegen's device deliberately
does not enable VK_KHR_external_semaphore_fd -- it shares through
AHardwareBuffer instead -- so vkImportSemaphoreFdKHR resolves to nullptr there
and every semaphore handed across is -1. That is not a flag that can be
flipped.

A fence can be exported where a semaphore cannot. framegen already creates one
per generation pass, so this adds VK_KHR_external_fence_fd, exports a sync fd
after the passes are submitted, and hands it back through a new
armsx3_lsfg_present_fenced. Our side polls that fd instead of idling a device.

Three things this deliberately does not do:

The existing completion fence is not the one exported. SYNC_FD export has copy
transference, which resets the source fence, so exporting it would leave the
slot-reuse wait at the top of present() blocking on a fence with no pending
signal -- a hang traded for a stall. A dedicated fence signalled by a
zero-batch submit is used instead, which signals once all previously submitted
queue work completes.

The extension is probed, not required. framegen's required-extension list
throws on a miss, so a driver without it would fail vkCreateDevice and lose
frame generation altogether -- a regression rather than a degradation.

The device wait stays as the fallback. A null entry point, a failed export, a
poll timeout or a poll error all fall back to it, and the warning for that is
emitted once per session rather than once per frame.

armsx3_lsfg_present keeps its signature and forwards to the new path with a
null fd, so the two bodies cannot drift. ABI goes to 3; the version check is
left exact rather than loosened.
2026-08-19 23:28:22 -04:00

166 lines
8.8 KiB
C++

// C ABI for Lossless Scaling frame generation.
//
// framegen CANNOT be linked into libarmsx3-core.so. It links volk, which defines 655 globals
// named vkCreateImage, vkQueueSubmit, ... and 124 of those are byte-for-byte the names our own
// Vulkan loader declares in rpcs3/Emu/RSX/VK/vk_android_loader.h -- every single symbol the RSX
// renderer uses. Two ways that goes wrong, and the second is the one that costs a week:
//
// 1. duplicate symbol at link time (clang defaults to -fno-common), or
// 2. the linker merges them, and framegen's volkLoadDevice(itsOwnDevice) then repoints every
// entry point the renderer uses at framegen's VkDevice. Every later vkCmdDraw goes to the
// wrong device, and it presents as a driver crash with nothing pointing at frame generation.
//
// So framegen and volk live in their own libarmsx3_lsfg.so, reached by dlopen + dlsym through
// this header. Nothing here is C++: the CMake project builds ANDROID_STL=c++_static, so each .so
// carries its own libc++ and an std::vector or std::function crossing the boundary would be two
// unrelated types that happen to share a name. The shim builds those on its own side.
//
// framegen also throws (LSFG::vulkan_error and friends). Exceptions must not cross a dlopen
// boundary either, so every entry point here catches everything and returns a code; the message
// is retrievable with armsx3_lsfg_last_error().
#pragma once
#include <stdint.h>
#ifdef __cplusplus
extern "C" {
#endif
// Bump when anything below changes shape. The loader refuses a library whose version it does not
// recognise, so a stale libarmsx3_lsfg.so on a user's device fails loudly at load instead of
// quietly passing mismatched structs.
#define ARMSX3_LSFG_ABI_VERSION 3u
// Mark the exported surface explicitly.
//
// The library is built -fvisibility=hidden so framegen's and volk's symbols stay in, and a
// version script narrows the dynamic table further. Neither of those can PROMOTE a symbol: a
// function hidden at compile time is local in the object, and `global:` in the linker script
// cannot bring it back. Without this attribute the .so builds and exports nothing at all, and
// the failure only shows up as dlsym returning null at runtime.
#if defined(__GNUC__) || defined(__clang__)
#define ARMSX3_LSFG_API __attribute__((visibility("default")))
#else
#define ARMSX3_LSFG_API
#endif
enum armsx3_lsfg_result
{
ARMSX3_LSFG_OK = 0,
ARMSX3_LSFG_ERR_UNKNOWN = -1,
ARMSX3_LSFG_ERR_NOT_INITIALIZED = -2,
ARMSX3_LSFG_ERR_BAD_ARGUMENT = -3,
ARMSX3_LSFG_ERR_SHADERS = -4,
ARMSX3_LSFG_ERR_VULKAN = -5,
};
// Hand back the SPIR-V for a named shader.
//
// framegen does NOT read Lossless.dll -- it asks for shaders by name and expects SPIR-V back.
// Extracting them from the user's own copy (PE resource -> DXBC -> SPIR-V) is the caller's job,
// which is deliberate: the shaders are THS's property and nothing here ships or downloads them.
//
// Return ARMSX3_LSFG_OK and set *out_data / *out_size on success. The buffer must stay valid
// until the initialize() call that triggered this returns. Any other return means "no such
// shader" and fails initialization.
typedef int (*armsx3_lsfg_shader_loader)(const char* name, const uint8_t** out_data,
uint32_t* out_size, void* user);
// Version of the loaded library. Call first; anything else on a mismatched library is undefined.
ARMSX3_LSFG_API uint32_t armsx3_lsfg_abi_version(void);
// Bring up framegen on the adapter identified by device_uuid (VkPhysicalDeviceIDProperties
// deviceUUID, 16 bytes, passed as the first 8 -- that is what framegen matches on).
//
// framegen creates its OWN VkDevice on that adapter. It does not share ours, which is why images
// have to be handed over as AHardwareBuffer below rather than as VkImage.
// performance selects framegen's 3.1p shader family instead of 3.1: a cheaper pipeline at lower
// quality, which is the difference between usable and not on a mobile GPU. It is fixed for the
// lifetime of the library state -- every context, present and teardown after this call goes to the
// family chosen here, because the two keep separate contexts and separate device state.
//
// flow_scale is the optical-flow resolution as a fraction of full: 1.0 is upstream's default and
// lower is cheaper. Note the sense is inverted from upstream's own config file, which stores a
// divisor and passes 1.0f/value here.
ARMSX3_LSFG_API int armsx3_lsfg_initialize(uint64_t device_uuid, int is_hdr, float flow_scale,
uint64_t generation_count, int performance, armsx3_lsfg_shader_loader loader, void* user);
// Create a context over a set of shared images.
//
// AHardwareBuffer rather than the FD path framegen also offers, because Adreno and Mali both
// refuse vkGetMemoryFdKHR(OPAQUE_FD) on AHB-imported memory -- the FD path simply does not work
// on the hardware this port runs on.
//
// The caller keeps ownership of every AHardwareBuffer and must keep them alive until the context
// is destroyed. Returns a context id >= 0, or a negative armsx3_lsfg_result.
ARMSX3_LSFG_API int32_t armsx3_lsfg_create_context_ahb(void* in0, void* in1, void* const* out_n,
uint32_t out_count, uint32_t width, uint32_t height, int32_t format);
// Generate frames for one presented pair.
//
// Semaphores are sync file descriptors, not VkSemaphore: framegen is on a different device and a
// VkSemaphore handle would be meaningless to it. in_sem is waited on before generation starts;
// each out_sems[i] is signalled when output image i is ready. Pass -1 for an unused slot.
ARMSX3_LSFG_API int armsx3_lsfg_present(int32_t ctx, int in_sem, const int* out_sems, uint32_t out_count);
// Generate frames for one presented pair, and hand back a fence for the result.
//
// Identical to armsx3_lsfg_present in every respect except that *out_fence_fd receives a sync file
// descriptor that becomes readable once the generation this call submitted has finished. The
// caller owns that fd and must close(2) it.
//
// This is the answer to armsx3_lsfg_wait_idle() below being the only completion signal on offer.
// framegen renders on its OWN VkDevice, so the caller cannot wait on its queues; before this
// entry point existed the only way to know the generated images were ready -- and, more
// importantly, that framegen had finished READING the caller's input images -- was a
// vkDeviceWaitIdle on framegen's device, once per presented frame. A sync fd can be waited on
// with poll(2) instead, which parks a thread rather than draining a GPU.
//
// *out_fence_fd is set to -1 whenever a descriptor is not available: an older library, a driver
// without VK_KHR_external_fence_fd, or work that had already completed by the time it was asked
// for. -1 is not an error and the return code is still ARMSX3_LSFG_OK -- the caller must fall
// back to armsx3_lsfg_wait_idle(), which is always correct.
//
// Added in ABI 3. Resolve it with dlsym rather than assuming it: this is the one entry point a
// caller can do without.
ARMSX3_LSFG_API int armsx3_lsfg_present_fenced(int32_t ctx, int in_sem, const int* out_sems,
uint32_t out_count, int* out_fence_fd);
ARMSX3_LSFG_API int armsx3_lsfg_destroy_context(int32_t ctx);
// Read the user's own Lossless.dll and keep the shaders it contains.
//
// Nothing is bundled or downloaded: the shaders are THS's property and the user must supply a
// legitimately purchased copy. Only the extracted SPIR-V is kept -- the DLL itself is not needed
// afterwards and the caller may delete its copy.
//
// The work is PE resource walk -> DXBC -> SPIR-V, and it is slow enough to be worth doing once
// and caching rather than at every boot. Returns the number of shaders extracted, or a negative
// armsx3_lsfg_result; armsx3_lsfg_last_error() explains a failure in terms a user can act on
// ("is Lossless Scaling up to date?" rather than a resource id).
ARMSX3_LSFG_API int armsx3_lsfg_import_shaders(const char* dll_path);
// How many shaders are currently held. Zero means frame generation cannot start.
ARMSX3_LSFG_API int armsx3_lsfg_shader_count(void);
// Serve a previously imported shader by name, for initialize()'s loader.
//
// Pass a null loader to armsx3_lsfg_initialize to use these instead of supplying your own.
ARMSX3_LSFG_API int armsx3_lsfg_get_shader(const char* name, const uint8_t** out_data, uint32_t* out_size);
// Block until framegen's device is idle.
//
// Needed on Android because framegen's device reads AHBs that OUR device writes, and there is no
// semaphore shared between the two. Without this the read races the write. It is also the reason
// frame generation cannot be free here: this is a device-level stall, not a queue wait.
ARMSX3_LSFG_API void armsx3_lsfg_wait_idle(void);
ARMSX3_LSFG_API void armsx3_lsfg_finalize(void);
// Message for the last failing call on this thread, or "" if none. Never null.
ARMSX3_LSFG_API const char* armsx3_lsfg_last_error(void);
#ifdef __cplusplus
}
#endif