workshopsoft/ludic
2
0
Fork 0
Code Issues Pull requests Projects Releases 26 Packages Wiki Activity Actions
ludic/runtime/native/vk_win.ll
Orkuncakilkaya 80764f85c1 fix(render3d): the Vulkan renderer stops leaking as it loads and draws 
- every texture upload malloc'd a CPU copy of its pixels and never freed it (236 MB over the
  valley's boot, and again for every model loaded later): the pixels are converted straight
  into the mapped staging buffer
- the per-level and per-layer views gvk_view_of made outlived their texture, and on MoltenVK a
  view keeps its Metal texture alive: a released texture takes its views with it, and the cache
  is keyed by numbers instead of a string built on every call
- the Vulkan structs filled for a draw, pass, barrier, descriptor set, buffer, allocation or
  upload (about sixty call sites) come from a reused 1 MB scratch ring (gvk_tmp)
- the descriptor-set cache and the retired buffers are emptied in place, not replaced; the grass
  cull's dispatch arguments are made once
- macOS drains an autorelease pool each frame (Vk.frame_pool, lvk_frame_pool in vk_mac.ll)
- R3D_VK_PROF reports Vulkan objects made and destroyed by kind, and every cache's length
- examples/rendering/smooth presents through render3d, so it runs on Vulkan too

The full valley on headless Vulkan loads to 2.18 GB and holds (it passed 8 GB while loading
before). ludic-dev test 305/305, selfhost-test 33/33.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-27 23:03:38 +03:00

371 lines
12 KiB
LLVM
Raw Blame History

; ============================================================================
; vk_win.ll — the Vulkan loader, for Windows, written in LLVM IR.
;
; vulkan-1.dll ships with every GPU driver on Windows 10 and 11, but it is not
; guaranteed, so it is opened AT RUN TIME and never linked: a program that asks for
; Vulkan on a machine without it gets 0 from vk_open and falls back to OpenGL
; rather than failing to start.
;
; lvk_open() -> 1 | 0 open the loader and bind every command (vk_thunks.ll)
; lvk_sym(name) -> ptr one entry point by name, null when the loader lacks it
; lvk_has(name) -> 1 | 0 whether a command is there (an extension's, say)
;
; The commands themselves, their struct layouts and constants are generated from
; the Vulkan registry by `ludic-dev vkgen` into vk_api.ludic and vk_thunks.ll.
; ============================================================================
declare ptr @LoadLibraryA(ptr)
declare ptr @GetProcAddress(ptr, ptr)
declare i32 @lvk_bind()
@.lvk_dll = private unnamed_addr constant [13 x i8] c"vulkan-1.dll\00"
@lvk_lib = internal global ptr null
define i32 @lvk_open() {
entry:
%have = load ptr, ptr @lvk_lib
%open = icmp ne ptr %have, null
br i1 %open, label %ok, label %load
load:
; Streamline: when the renderer asked for it (lvk_sl_prefer) and sl.interposer.dll is found, the
; interposer IS the loader - it proxies every Vulkan command, and intercepts the few its DLSS and
; Reflex plugins need. Without it, vulkan-1.dll as always.
%want = load i32, ptr @lvk_sl_want
%try = icmp ne i32 %want, 0
br i1 %try, label %sl, label %plain
sl:
%sll = call ptr @LoadLibraryA(ptr @.lsl_dll)
%slmiss = icmp eq ptr %sll, null
br i1 %slmiss, label %plain, label %slok
slok:
store i32 1, ptr @lvk_sl_on
store ptr %sll, ptr @lvk_lib
%sln = call i32 @lvk_bind()
br label %ok
plain:
%lib = call ptr @LoadLibraryA(ptr @.lvk_dll)
%none = icmp eq ptr %lib, null
br i1 %none, label %fail, label %bind
bind:
store ptr %lib, ptr @lvk_lib
%n = call i32 @lvk_bind()
br label %ok
ok:
ret i32 1
fail:
ret i32 0
}
define ptr @lvk_sym(ptr %name) {
entry:
%lib = load ptr, ptr @lvk_lib
%none = icmp eq ptr %lib, null
br i1 %none, label %no, label %look
look:
%p = call ptr @GetProcAddress(ptr %lib, ptr %name)
ret ptr %p
no:
ret ptr null
}
define i32 @lvk_has(ptr %name) {
entry:
%p = call ptr @lvk_sym(ptr %name)
%there = icmp ne ptr %p, null
%r = zext i1 %there to i32
ret i32 %r
}
; ---- Streamline (NVIDIA DLSS, Reflex) -----------------------------------------------------------
; lvk_sl_prefer(1) before lvk_open makes the loader try sl.interposer.dll first; lvk_sl_active says
; whether it did. The sl* thunks resolve their exports from the opened library once and return -1
; (not sl::Result::eOk) when it has none - vulkan-1.dll, or no library at all.
@.lsl_dll = private unnamed_addr constant [18 x i8] c"sl.interposer.dll\00"
@lvk_sl_want = internal global i32 0
@lvk_sl_on = internal global i32 0
define void @lvk_sl_prefer(i32 %on) {
entry:
store i32 %on, ptr @lvk_sl_want
ret void
}
define i32 @lvk_sl_active() {
entry:
%v = load i32, ptr @lvk_sl_on
ret i32 %v
}
@.lsl_n_slInit = private unnamed_addr constant [7 x i8] c"slInit\00"
@lsl_p_slInit = internal global ptr null
define i32 @lsl_slInit(ptr %a0, i64 %a1) {
entry:
%have = load ptr, ptr @lsl_p_slInit
%known = icmp ne ptr %have, null
br i1 %known, label %call, label %look
look:
%found = call ptr @lvk_sym(ptr @.lsl_n_slInit)
store ptr %found, ptr @lsl_p_slInit
%none = icmp eq ptr %found, null
br i1 %none, label %absent, label %call
call:
%fp = load ptr, ptr @lsl_p_slInit
%r = call i32 %fp(ptr %a0, i64 %a1)
ret i32 %r
absent:
ret i32 -1
}
@.lsl_n_slShutdown = private unnamed_addr constant [11 x i8] c"slShutdown\00"
@lsl_p_slShutdown = internal global ptr null
define i32 @lsl_slShutdown() {
entry:
%have = load ptr, ptr @lsl_p_slShutdown
%known = icmp ne ptr %have, null
br i1 %known, label %call, label %look
look:
%found = call ptr @lvk_sym(ptr @.lsl_n_slShutdown)
store ptr %found, ptr @lsl_p_slShutdown
%none = icmp eq ptr %found, null
br i1 %none, label %absent, label %call
call:
%fp = load ptr, ptr @lsl_p_slShutdown
%r = call i32 %fp()
ret i32 %r
absent:
ret i32 -1
}
@.lsl_n_slIsFeatureSupported = private unnamed_addr constant [21 x i8] c"slIsFeatureSupported\00"
@lsl_p_slIsFeatureSupported = internal global ptr null
define i32 @lsl_slIsFeatureSupported(i32 %a0, ptr %a1) {
entry:
%have = load ptr, ptr @lsl_p_slIsFeatureSupported
%known = icmp ne ptr %have, null
br i1 %known, label %call, label %look
look:
%found = call ptr @lvk_sym(ptr @.lsl_n_slIsFeatureSupported)
store ptr %found, ptr @lsl_p_slIsFeatureSupported
%none = icmp eq ptr %found, null
br i1 %none, label %absent, label %call
call:
%fp = load ptr, ptr @lsl_p_slIsFeatureSupported
%r = call i32 %fp(i32 %a0, ptr %a1)
ret i32 %r
absent:
ret i32 -1
}
@.lsl_n_slIsFeatureLoaded = private unnamed_addr constant [18 x i8] c"slIsFeatureLoaded\00"
@lsl_p_slIsFeatureLoaded = internal global ptr null
define i32 @lsl_slIsFeatureLoaded(i32 %a0, ptr %a1) {
entry:
%have = load ptr, ptr @lsl_p_slIsFeatureLoaded
%known = icmp ne ptr %have, null
br i1 %known, label %call, label %look
look:
%found = call ptr @lvk_sym(ptr @.lsl_n_slIsFeatureLoaded)
store ptr %found, ptr @lsl_p_slIsFeatureLoaded
%none = icmp eq ptr %found, null
br i1 %none, label %absent, label %call
call:
%fp = load ptr, ptr @lsl_p_slIsFeatureLoaded
%r = call i32 %fp(i32 %a0, ptr %a1)
ret i32 %r
absent:
ret i32 -1
}
@.lsl_n_slGetFeatureRequirements = private unnamed_addr constant [25 x i8] c"slGetFeatureRequirements\00"
@lsl_p_slGetFeatureRequirements = internal global ptr null
define i32 @lsl_slGetFeatureRequirements(i32 %a0, ptr %a1) {
entry:
%have = load ptr, ptr @lsl_p_slGetFeatureRequirements
%known = icmp ne ptr %have, null
br i1 %known, label %call, label %look
look:
%found = call ptr @lvk_sym(ptr @.lsl_n_slGetFeatureRequirements)
store ptr %found, ptr @lsl_p_slGetFeatureRequirements
%none = icmp eq ptr %found, null
br i1 %none, label %absent, label %call
call:
%fp = load ptr, ptr @lsl_p_slGetFeatureRequirements
%r = call i32 %fp(i32 %a0, ptr %a1)
ret i32 %r
absent:
ret i32 -1
}
@.lsl_n_slGetFeatureFunction = private unnamed_addr constant [21 x i8] c"slGetFeatureFunction\00"
@lsl_p_slGetFeatureFunction = internal global ptr null
define i32 @lsl_slGetFeatureFunction(i32 %a0, ptr %a1, ptr %a2) {
entry:
%have = load ptr, ptr @lsl_p_slGetFeatureFunction
%known = icmp ne ptr %have, null
br i1 %known, label %call, label %look
look:
%found = call ptr @lvk_sym(ptr @.lsl_n_slGetFeatureFunction)
store ptr %found, ptr @lsl_p_slGetFeatureFunction
%none = icmp eq ptr %found, null
br i1 %none, label %absent, label %call
call:
%fp = load ptr, ptr @lsl_p_slGetFeatureFunction
%r = call i32 %fp(i32 %a0, ptr %a1, ptr %a2)
ret i32 %r
absent:
ret i32 -1
}
@.lsl_n_slGetNewFrameToken = private unnamed_addr constant [19 x i8] c"slGetNewFrameToken\00"
@lsl_p_slGetNewFrameToken = internal global ptr null
define i32 @lsl_slGetNewFrameToken(ptr %a0, ptr %a1) {
entry:
%have = load ptr, ptr @lsl_p_slGetNewFrameToken
%known = icmp ne ptr %have, null
br i1 %known, label %call, label %look
look:
%found = call ptr @lvk_sym(ptr @.lsl_n_slGetNewFrameToken)
store ptr %found, ptr @lsl_p_slGetNewFrameToken
%none = icmp eq ptr %found, null
br i1 %none, label %absent, label %call
call:
%fp = load ptr, ptr @lsl_p_slGetNewFrameToken
%r = call i32 %fp(ptr %a0, ptr %a1)
ret i32 %r
absent:
ret i32 -1
}
@.lsl_n_slSetTagForFrame = private unnamed_addr constant [17 x i8] c"slSetTagForFrame\00"
@lsl_p_slSetTagForFrame = internal global ptr null
define i32 @lsl_slSetTagForFrame(ptr %a0, ptr %a1, ptr %a2, i32 %a3, ptr %a4) {
entry:
%have = load ptr, ptr @lsl_p_slSetTagForFrame
%known = icmp ne ptr %have, null
br i1 %known, label %call, label %look
look:
%found = call ptr @lvk_sym(ptr @.lsl_n_slSetTagForFrame)
store ptr %found, ptr @lsl_p_slSetTagForFrame
%none = icmp eq ptr %found, null
br i1 %none, label %absent, label %call
call:
%fp = load ptr, ptr @lsl_p_slSetTagForFrame
%r = call i32 %fp(ptr %a0, ptr %a1, ptr %a2, i32 %a3, ptr %a4)
ret i32 %r
absent:
ret i32 -1
}
@.lsl_n_slSetConstants = private unnamed_addr constant [15 x i8] c"slSetConstants\00"
@lsl_p_slSetConstants = internal global ptr null
define i32 @lsl_slSetConstants(ptr %a0, ptr %a1, ptr %a2) {
entry:
%have = load ptr, ptr @lsl_p_slSetConstants
%known = icmp ne ptr %have, null
br i1 %known, label %call, label %look
look:
%found = call ptr @lvk_sym(ptr @.lsl_n_slSetConstants)
store ptr %found, ptr @lsl_p_slSetConstants
%none = icmp eq ptr %found, null
br i1 %none, label %absent, label %call
call:
%fp = load ptr, ptr @lsl_p_slSetConstants
%r = call i32 %fp(ptr %a0, ptr %a1, ptr %a2)
ret i32 %r
absent:
ret i32 -1
}
@.lsl_n_slEvaluateFeature = private unnamed_addr constant [18 x i8] c"slEvaluateFeature\00"
@lsl_p_slEvaluateFeature = internal global ptr null
define i32 @lsl_slEvaluateFeature(i32 %a0, ptr %a1, ptr %a2, i32 %a3, ptr %a4) {
entry:
%have = load ptr, ptr @lsl_p_slEvaluateFeature
%known = icmp ne ptr %have, null
br i1 %known, label %call, label %look
look:
%found = call ptr @lvk_sym(ptr @.lsl_n_slEvaluateFeature)
store ptr %found, ptr @lsl_p_slEvaluateFeature
%none = icmp eq ptr %found, null
br i1 %none, label %absent, label %call
call:
%fp = load ptr, ptr @lsl_p_slEvaluateFeature
%r = call i32 %fp(i32 %a0, ptr %a1, ptr %a2, i32 %a3, ptr %a4)
ret i32 %r
absent:
ret i32 -1
}
@.lsl_n_slFreeResources = private unnamed_addr constant [16 x i8] c"slFreeResources\00"
@lsl_p_slFreeResources = internal global ptr null
define i32 @lsl_slFreeResources(i32 %a0, ptr %a1) {
entry:
%have = load ptr, ptr @lsl_p_slFreeResources
%known = icmp ne ptr %have, null
br i1 %known, label %call, label %look
look:
%found = call ptr @lvk_sym(ptr @.lsl_n_slFreeResources)
store ptr %found, ptr @lsl_p_slFreeResources
%none = icmp eq ptr %found, null
br i1 %none, label %absent, label %call
call:
%fp = load ptr, ptr @lsl_p_slFreeResources
%r = call i32 %fp(i32 %a0, ptr %a1)
ret i32 %r
absent:
ret i32 -1
}
; a feature's own functions (slDLSSSetOptions, slReflexSleep ...) come from slGetFeatureFunction
define i32 @lsl_call_p(ptr %fn, ptr %a0) {
entry:
%r = call i32 %fn(ptr %a0)
ret i32 %r
}
define i32 @lsl_call_pp(ptr %fn, ptr %a0, ptr %a1) {
entry:
%r = call i32 %fn(ptr %a0, ptr %a1)
ret i32 %r
}
define i32 @lsl_call_ppp(ptr %fn, ptr %a0, ptr %a1, ptr %a2) {
entry:
%r = call i32 %fn(ptr %a0, ptr %a1, ptr %a2)
ret i32 %r
}
define i32 @lsl_call_ip(ptr %fn, i32 %a0, ptr %a1) {
entry:
%r = call i32 %fn(i32 %a0, ptr %a1)
ret i32 %r
}
; a command-buffer command with three counts, through a pointer (vkCmdDrawMeshTasksEXT: the Streamline
; interposer exports no such command, so it is looked up per device with vkGetDeviceProcAddr)
define void @lsl_call_piii(ptr %fn, ptr %a0, i32 %a1, i32 %a2, i32 %a3) {
entry:
call void %fn(ptr %a0, i32 %a1, i32 %a2, i32 %a3)
ret void
}
; ---- acceleration structures: commands the Streamline interposer does not export --------------------
; vkCreateAccelerationStructureKHR(device, info, allocator, out) -> VkResult
define i32 @lsl_call_pppp(ptr %fn, ptr %a0, ptr %a1, ptr %a2, ptr %a3) {
entry:
%r = call i32 %fn(ptr %a0, ptr %a1, ptr %a2, ptr %a3)
ret i32 %r
}
; vkDestroyAccelerationStructureKHR(device, handle, allocator)
define void @lsl_call_plp(ptr %fn, ptr %a0, i64 %a1, ptr %a2) {
entry:
call void %fn(ptr %a0, i64 %a1, ptr %a2)
ret void
}
; vkGetAccelerationStructureBuildSizesKHR(device, buildType, info, maxPrimitiveCounts, sizes)
define void @lsl_call_pippp(ptr %fn, ptr %a0, i32 %a1, ptr %a2, ptr %a3, ptr %a4) {
entry:
call void %fn(ptr %a0, i32 %a1, ptr %a2, ptr %a3, ptr %a4)
ret void
}
; vkCmdBuildAccelerationStructuresKHR(commandBuffer, infoCount, infos, buildRangeInfos)
define void @lsl_call_pipp(ptr %fn, ptr %a0, i32 %a1, ptr %a2, ptr %a3) {
entry:
call void %fn(ptr %a0, i32 %a1, ptr %a2, ptr %a3)
ret void
}
; vkGetAccelerationStructureDeviceAddressKHR(device, info) -> VkDeviceAddress
define i64 @lsl_call_pp_l(ptr %fn, ptr %a0, ptr %a1) {
entry:
%r = call i64 %fn(ptr %a0, ptr %a1)
ret i64 %r
}
; the frame's autorelease pool is a macOS matter (vk_mac.ll); Windows has nothing to drain
define void @lvk_frame_pool() {
entry:
ret void
}
Reference in a new issue View git blame Copy permalink