ludic/runtime/native/vk_mac.ll

; ============================================================================
; vk_mac.ll — the Vulkan loader, for macOS, written in LLVM IR.
;
; macOS has no Vulkan of its own: the LunarG loader and MoltenVK (Vulkan over
; Metal) come with the Vulkan SDK or with a game that bundles them. So the loader
; is looked for at run time - beside the executable, in the usual library
; directories, then MoltenVK on its own (a bundle's Contents/Frameworks), then under
; $VULKAN_SDK - and a Mac without any gets 0 from vk_open and stays on OpenGL. The contract is vk_win.ll's:
;
;   lvk_open() -> 1 | 0      lvk_sym(name) -> ptr      lvk_has(name) -> 1 | 0
;
; MoltenVK is a portability driver: an instance must ask for
; VK_KHR_portability_enumeration (and the ENUMERATE_PORTABILITY flag) to see it.
; ============================================================================

declare ptr @dlopen(ptr, i32)
declare ptr @dlsym(ptr, ptr)
declare ptr @getenv(ptr)
declare i32 @snprintf(ptr, i64, ptr, ...)
declare i32 @lvk_bind()

@.lvk_c0 = private unnamed_addr constant [18 x i8] c"libvulkan.1.dylib\00"
@.lvk_c1 = private unnamed_addr constant [35 x i8] c"@executable_path/libvulkan.1.dylib\00"
@.lvk_c2 = private unnamed_addr constant [33 x i8] c"/usr/local/lib/libvulkan.1.dylib\00"
@.lvk_c3 = private unnamed_addr constant [36 x i8] c"/opt/homebrew/lib/libvulkan.1.dylib\00"
@.lvk_env = private unnamed_addr constant [11 x i8] c"VULKAN_SDK\00"
@.lvk_fmt = private unnamed_addr constant [25 x i8] c"%s/lib/libvulkan.1.dylib\00"
@.lvk_fmtm = private unnamed_addr constant [25 x i8] c"%s/lib/libMoltenVK.dylib\00"
@.lvk_m0 = private unnamed_addr constant [49 x i8] c"@executable_path/../Frameworks/libMoltenVK.dylib\00"
@.lvk_m1 = private unnamed_addr constant [35 x i8] c"@executable_path/libMoltenVK.dylib\00"
@.lvk_m2 = private unnamed_addr constant [18 x i8] c"libMoltenVK.dylib\00"
@lvk_lib  = internal global ptr null

; RTLD_NOW (2) | RTLD_LOCAL (4)
define internal ptr @lvk_try(ptr %path) {
entry:
  %h = call ptr @dlopen(ptr %path, i32 6)
  ret ptr %h
}

; the candidates in order: a Vulkan loader first (the SDK's, so the validation layer can be
; stacked in development), then MoltenVK itself, which exports every vk* entry point and is what
; a bundle ships in Contents/Frameworks - no loader, no ICD json.
@lvk_paths = internal constant [7 x ptr] [ptr @.lvk_c0, ptr @.lvk_c1, ptr @.lvk_c2, ptr @.lvk_c3, ptr @.lvk_m0, ptr @.lvk_m1, ptr @.lvk_m2]

define i32 @lvk_open() {
entry:
  %have = load ptr, ptr @lvk_lib
  %open = icmp ne ptr %have, null
  br i1 %open, label %ok, label %loop
loop:
  %i = phi i32 [ 0, %entry ], [ %i1, %next ]
  %slot = getelementptr [7 x ptr], ptr @lvk_paths, i32 0, i32 %i
  %path = load ptr, ptr %slot
  %h = call ptr @lvk_try(ptr %path)
  %miss = icmp eq ptr %h, null
  br i1 %miss, label %next, label %got
next:
  %i1 = add i32 %i, 1
  %more = icmp slt i32 %i1, 7
  br i1 %more, label %loop, label %sdk
got:
  store ptr %h, ptr @lvk_lib
  br label %bind
sdk:
  %sdkp = call ptr @getenv(ptr @.lvk_env)
  %nosdk = icmp eq ptr %sdkp, null
  br i1 %nosdk, label %fail, label %s1
s1:
  %buf = alloca [1024 x i8]
  %w = call i32 (ptr, i64, ptr, ...) @snprintf(ptr %buf, i64 1024, ptr @.lvk_fmt, ptr %sdkp)
  %h4 = call ptr @lvk_try(ptr %buf)
  %n4 = icmp eq ptr %h4, null
  br i1 %n4, label %s2, label %got4
s2:
  %w2 = call i32 (ptr, i64, ptr, ...) @snprintf(ptr %buf, i64 1024, ptr @.lvk_fmtm, ptr %sdkp)
  %h5 = call ptr @lvk_try(ptr %buf)
  %n5 = icmp eq ptr %h5, null
  br i1 %n5, label %fail, label %got5
got4:
  store ptr %h4, ptr @lvk_lib
  br label %bind
got5:
  store ptr %h5, ptr @lvk_lib
  br label %bind
bind:
  %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 @dlsym(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: Windows only. The same symbols, so a program that calls them links everywhere. ----
define void @lvk_sl_prefer(i32 %on) {
entry:
  ret void
}
define i32 @lvk_sl_active() {
entry:
  ret i32 0
}
define i32 @lsl_slInit(ptr %a0, i64 %a1) {
entry:
  ret i32 -1
}
define i32 @lsl_slShutdown() {
entry:
  ret i32 -1
}
define i32 @lsl_slIsFeatureSupported(i32 %a0, ptr %a1) {
entry:
  ret i32 -1
}
define i32 @lsl_slIsFeatureLoaded(i32 %a0, ptr %a1) {
entry:
  ret i32 -1
}
define i32 @lsl_slGetFeatureRequirements(i32 %a0, ptr %a1) {
entry:
  ret i32 -1
}
define i32 @lsl_slGetFeatureFunction(i32 %a0, ptr %a1, ptr %a2) {
entry:
  ret i32 -1
}
define i32 @lsl_slGetNewFrameToken(ptr %a0, ptr %a1) {
entry:
  ret i32 -1
}
define i32 @lsl_slSetTagForFrame(ptr %a0, ptr %a1, ptr %a2, i32 %a3, ptr %a4) {
entry:
  ret i32 -1
}
define i32 @lsl_slSetConstants(ptr %a0, ptr %a1, ptr %a2) {
entry:
  ret i32 -1
}
define i32 @lsl_slEvaluateFeature(i32 %a0, ptr %a1, ptr %a2, i32 %a3, ptr %a4) {
entry:
  ret i32 -1
}
define i32 @lsl_slFreeResources(i32 %a0, ptr %a1) {
entry:
  ret i32 -1
}
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 --------------------------------------------------------------
; MoltenVK and CAMetalLayer autorelease objects on every frame (the drawable, each render pass's
; descriptor), and a program that pumps its own events never drains a pool: without this they are
; kept for the life of the process - 160 MB/s of small allocations in a windowed valley.
; lvk_frame_pool() pops the pool the last frame pushed and pushes the next one.
declare ptr @objc_autoreleasePoolPush()
declare void @objc_autoreleasePoolPop(ptr)
@lvk_pool = internal global ptr null

define void @lvk_frame_pool() {
entry:
  %old = load ptr, ptr @lvk_pool
  %have = icmp ne ptr %old, null
  br i1 %have, label %pop, label %push
pop:
  call void @objc_autoreleasePoolPop(ptr %old)
  br label %push
push:
  %p = call ptr @objc_autoreleasePoolPush()
  store ptr %p, ptr @lvk_pool
  ret void
}