ludic-dev shaders: a variant whose first file is *.mesh compiles that stage with glslang -S mesh for Vulkan 1.3, without the vertex stage's depth-remap wrapper or invariant gl_Position (a mesh shader writes an array of positions and remaps depth itself). Its SPIR-V keeps the .vert name, so the manifest and the loader are unchanged. The 51 existing programs build identical SPIR-V. runtime: lsl_call_piii(fn, ptr, i32, i32, i32) calls a command-buffer command through a pointer. NVIDIA Streamline's interposer exports no vkCmdDrawMeshTasksEXT, so it is to be looked up per device with vkGetDeviceProcAddr. Both runtimes assemble. Nothing uses either yet. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
192 lines
5.1 KiB
LLVM
192 lines
5.1 KiB
LLVM
; ============================================================================
|
|
; 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 under $VULKAN_SDK - and a Mac without it 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_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
|
|
}
|
|
|
|
define i32 @lvk_open() {
|
|
entry:
|
|
%have = load ptr, ptr @lvk_lib
|
|
%open = icmp ne ptr %have, null
|
|
br i1 %open, label %ok, label %t0
|
|
t0:
|
|
%h0 = call ptr @lvk_try(ptr @.lvk_c0)
|
|
%n0 = icmp eq ptr %h0, null
|
|
br i1 %n0, label %t1, label %got0
|
|
got0:
|
|
store ptr %h0, ptr @lvk_lib
|
|
br label %bind
|
|
t1:
|
|
%h1 = call ptr @lvk_try(ptr @.lvk_c1)
|
|
%n1 = icmp eq ptr %h1, null
|
|
br i1 %n1, label %t2, label %got1
|
|
got1:
|
|
store ptr %h1, ptr @lvk_lib
|
|
br label %bind
|
|
t2:
|
|
%h2 = call ptr @lvk_try(ptr @.lvk_c2)
|
|
%n2 = icmp eq ptr %h2, null
|
|
br i1 %n2, label %t3, label %got2
|
|
got2:
|
|
store ptr %h2, ptr @lvk_lib
|
|
br label %bind
|
|
t3:
|
|
%h3 = call ptr @lvk_try(ptr @.lvk_c3)
|
|
%n3 = icmp eq ptr %h3, null
|
|
br i1 %n3, label %t4, label %got3
|
|
got3:
|
|
store ptr %h3, ptr @lvk_lib
|
|
br label %bind
|
|
t4:
|
|
%sdk = call ptr @getenv(ptr @.lvk_env)
|
|
%nosdk = icmp eq ptr %sdk, null
|
|
br i1 %nosdk, label %fail, label %t4b
|
|
t4b:
|
|
%buf = alloca [1024 x i8]
|
|
%w = call i32 (ptr, i64, ptr, ...) @snprintf(ptr %buf, i64 1024, ptr @.lvk_fmt, ptr %sdk)
|
|
%h4 = call ptr @lvk_try(ptr %buf)
|
|
%n4 = icmp eq ptr %h4, null
|
|
br i1 %n4, label %fail, label %got4
|
|
got4:
|
|
store ptr %h4, 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
|
|
}
|