adding VK_KHR_device_group functions; adding single ICD output lib for all backends

README.md

@@ -48,6 +48,7 @@ Assume thou that functions lacking in this array are, for now, not intended to b
 
 Name                                             | Status
 -------------------------------------------------|--------
+vkAcquireNextImage2KHR                           | ✅ Implemented
 vkAcquireNextImageKHR                            | ✅ Implemented
 vkAllocateCommandBuffers                         | ✅ Implemented
 vkAllocateDescriptorSets                         | ✅ Implemented
@@ -71,6 +72,7 @@ vkCmdCopyImage                                   | ✅ Implemented
 vkCmdCopyImageToBuffer                           | ✅ Implemented
 vkCmdCopyQueryPoolResults                        | ✅ Implemented
 vkCmdDispatch                                    | ✅ Implemented
+vkCmdDispatchBaseKHR                             | ✅ Implemented
 vkCmdDispatchIndirect                            | ✅ Implemented
 vkCmdDraw                                        | ✅ Implemented
 vkCmdDrawIndexed                                 | ✅ Implemented
@@ -89,6 +91,7 @@ vkCmdResolveImage                                | ✅ Implemented
 vkCmdSetBlendConstants                           | ✅ Implemented
 vkCmdSetDepthBias                                | ⚙️ WIP
 vkCmdSetDepthBounds                              | ⚙️ WIP
+vkCmdSetDeviceMaskKHR                            | ✅ Implemented
 vkCmdSetEvent                                    | ✅ Implemented
 vkCmdSetLineWidth                                | ⚙️ WIP
 vkCmdSetScissor                                  | ✅ Implemented
@@ -153,13 +156,16 @@ vkEnumerateDeviceExtensionProperties             | ✅ Implemented
 vkEnumerateDeviceLayerProperties                 | ✅ Implemented
 vkEnumerateInstanceExtensionProperties           | ✅ Implemented
 vkEnumerateInstanceLayerProperties               | ✅ Implemented
-vkEnumeratePhysicalDevices                       | ✅ Implemented
 vkEnumeratePhysicalDeviceGroupsKHR               | ✅ Implemented
+vkEnumeratePhysicalDevices                       | ✅ Implemented
 vkFlushMappedMemoryRanges                        | ✅ Implemented
 vkFreeCommandBuffers                             | ✅ Implemented
 vkFreeDescriptorSets                             | ✅ Implemented
 vkFreeMemory                                     | ✅ Implemented
 vkGetBufferMemoryRequirements                    | ✅ Implemented
+vkGetDeviceGroupPeerMemoryFeaturesKHR            | ✅ Implemented
+vkGetDeviceGroupPresentCapabilitiesKHR           | ✅ Implemented
+vkGetDeviceGroupSurfacePresentModesKHR           | ✅ Implemented
 vkGetDeviceMemoryCommitment                      | ⚙️ WIP
 vkGetDeviceProcAddr                              | ✅ Implemented
 vkGetDeviceQueue                                 | ✅ Implemented

build.zig

@@ -4,24 +4,37 @@ const builtin = @import("builtin");
 
 const ImplementationDesc = struct {
     name: []const u8,
+    icd_name: ?[]const u8 = null,
     root_source_file: []const u8,
     vulkan_version: std.SemanticVersion,
     custom: ?*const fn (
         *std.Build,
         *Step.Compile,
+        *std.Build.Module,
+        *std.Build.Module,
+        *std.Build.Module,
+        *std.Build.Module,
         std.Build.ResolvedTarget,
         std.builtin.OptimizeMode,
-        *Step.Options,
         bool,
     ) anyerror!void = null,
+    options: ?*const fn (*std.Build, *Step.Options) anyerror!void = null,
 };
 
 const implementations = [_]ImplementationDesc{
+    .{
+        .name = "ape",
+        .icd_name = "ape",
+        .root_source_file = "src/ape/lib.zig",
+        .vulkan_version = .{ .major = 1, .minor = 0, .patch = 0 },
+        .custom = customApe,
+    },
     .{
         .name = "soft",
         .root_source_file = "src/soft/lib.zig",
         .vulkan_version = .{ .major = 1, .minor = 0, .patch = 0 },
         .custom = customSoft,
+        .options = optionsSoft,
     },
 };
 
@@ -79,7 +92,8 @@ pub fn build(b: *std.Build) !void {
         base_c_includes.link_libc = true;
     }
 
-    base_mod.addImport("base_c", base_c_includes.createModule());
+    const base_c_mod = base_c_includes.createModule();
+    base_mod.addImport("base_c", base_c_mod);
 
     const use_llvm = b.option(bool, "use-llvm", "LLVM build") orelse (b.release_mode != .off);
 
@@ -104,12 +118,22 @@ pub fn build(b: *std.Build) !void {
             .use_llvm = use_llvm,
         });
 
-        if (impl.custom) |custom| {
-            custom(b, lib, target, optimize, options, use_llvm) catch continue;
+        if (impl.custom) |func| {
+            func(b, lib, lib_mod, base_mod, vulkan, base_c_mod, target, optimize, use_llvm) catch continue;
+        }
+
+        if (impl.options) |func| {
+            func(b, options) catch continue;
         }
 
         const icd_file = b.addWriteFile(
-            b.getInstallPath(.lib, b.fmt("vk_ape_{s}.json", .{impl.name})),
+            b.getInstallPath(
+                .lib,
+                if (impl.icd_name) |icd_name|
+                    b.fmt("vk_{s}.json", .{icd_name})
+                else
+                    b.fmt("vk_ape_{s}.json", .{impl.name}),
+            ),
             b.fmt(
                 \\{{
                 \\    "file_format_version": "1.0.1",
@@ -171,12 +195,48 @@ pub fn build(b: *std.Build) !void {
     docs_step.dependOn(&install_docs.step);
 }
 
-fn customSoft(
+fn customApe(
     b: *std.Build,
     lib: *Step.Compile,
+    lib_mod: *std.Build.Module,
+    base_mod: *std.Build.Module,
+    vulkan: *std.Build.Module,
+    base_c_mod: *std.Build.Module,
+    target: std.Build.ResolvedTarget,
+    optimize: std.builtin.OptimizeMode,
+    use_llvm: bool,
+) !void {
+    for (implementations) |impl| {
+        if (std.mem.eql(u8, impl.name, "ape"))
+            continue;
+
+        const mod = b.createModule(.{
+            .root_source_file = b.path(impl.root_source_file),
+            .target = target,
+            .optimize = optimize,
+            .imports = &.{
+                .{ .name = "base", .module = base_mod },
+                .{ .name = "vulkan", .module = vulkan },
+            },
+        });
+
+        if (impl.custom) |func| {
+            func(b, lib, mod, base_mod, vulkan, base_c_mod, target, optimize, use_llvm) catch continue;
+        }
+
+        lib_mod.addImport(impl.name, mod);
+    }
+}
+
+fn customSoft(
+    b: *std.Build,
+    _: *Step.Compile,
+    lib_mod: *std.Build.Module,
+    _: *std.Build.Module,
+    _: *std.Build.Module,
+    base_c_mod: *std.Build.Module,
     _: std.Build.ResolvedTarget,
     _: std.builtin.OptimizeMode,
-    options: *Step.Options,
     use_llvm: bool,
 ) !void {
     const spv = b.lazyDependency("SPIRV_Interpreter", .{
@@ -184,23 +244,25 @@ fn customSoft(
         .@"no-test" = true,
         .@"use-llvm" = use_llvm,
     }) orelse return error.UnresolvedDependency;
-    lib.root_module.addImport("spv", spv.module("spv"));
 
-    const single_threaded_option = b.option(bool, "single-threaded", "Single threaded runtime mode") orelse false;
-    const debug_allocator_option = b.option(bool, "debug-allocator", "Debug device allocator") orelse false;
-    const shaders_simd_option = b.option(bool, "shader-simd", "Shaders SIMD acceleration") orelse true;
-    const single_threaded_compute_option = b.option(bool, "single-threaded-compute", "Single threaded compute shaders execution") orelse true;
-    const compute_dump_early_results_table_option = b.option(u32, "compute-dump-early-results-table", "Dump compute shaders results table before invocation");
-    const compute_dump_final_results_table_option = b.option(u32, "compute-dump-final-results-table", "Dump compute shaders results table after invocation");
-    const approxiamte_rgb_option = b.option(bool, "approximates-rgb", "Approximate sRGB <-> RGB conversions") orelse true;
+    lib_mod.addImport("soft_c", base_c_mod);
+    lib_mod.addImport("spv", spv.module("spv"));
+}
 
-    options.addOption(bool, "single_threaded", single_threaded_option);
-    options.addOption(bool, "debug_allocator", debug_allocator_option);
-    options.addOption(bool, "shaders_simd", shaders_simd_option);
-    options.addOption(bool, "single_threaded_compute", single_threaded_compute_option);
-    options.addOption(?u32, "compute_dump_early_results_table", compute_dump_early_results_table_option);
-    options.addOption(?u32, "compute_dump_final_results_table", compute_dump_final_results_table_option);
-    options.addOption(bool, "approximates_rgb", approxiamte_rgb_option);
+fn optionsSoft(b: *std.Build, options: *Step.Options) !void {
+    const single_threaded_option = b.option(bool, "soft-single-threaded", "Single threaded runtime mode") orelse false;
+    const debug_allocator_option = b.option(bool, "soft-debug-allocator", "Debug device allocator") orelse false;
+    const shaders_simd_option = b.option(bool, "soft-shader-simd", "Shaders SIMD acceleration") orelse true;
+    const compute_dump_early_results_table_option = b.option(u32, "soft-compute-dump-early-results-table", "Dump compute shaders results table before invocation");
+    const compute_dump_final_results_table_option = b.option(u32, "soft-compute-dump-final-results-table", "Dump compute shaders results table after invocation");
+    const approxiamte_rgb_option = b.option(bool, "soft-approximates-rgb", "Approximate sRGB <-> RGB conversions") orelse true;
+
+    options.addOption(bool, "soft_single_threaded", single_threaded_option);
+    options.addOption(bool, "soft_debug_allocator", debug_allocator_option);
+    options.addOption(bool, "soft_shaders_simd", shaders_simd_option);
+    options.addOption(?u32, "soft_compute_dump_early_results_table", compute_dump_early_results_table_option);
+    options.addOption(?u32, "soft_compute_dump_final_results_table", compute_dump_final_results_table_option);
+    options.addOption(bool, "soft_approximates_rgb", approxiamte_rgb_option);
 }
 
 fn addCTS(b: *std.Build, target: std.Build.ResolvedTarget, impl: *const ImplementationDesc, impl_lib: *Step.Compile, comptime mode: RunningMode) !*Step {

build.zig.zon

@@ -26,7 +26,7 @@
             .hash = "N-V-__8AAF9uOh0I4P_99za7N822J3JwsDaqONrFVrcEQo59",
         },
         .SPIRV_Interpreter = .{
-            .url = "git+https://git.kbz8.me/kbz_8/SPIRV-Interpreter#52ded6b490a47d843ad401cbd2c41b4dbdc7dca1",
+            .url = "git+https://git.kbz8.me/kbz_8/SPIRV-Interpreter#42554a5cc1069e8174b377e95ce2ac2d802a44ff",
             .hash = "SPIRV_Interpreter-0.0.1-ajmpn0GXBwD01BnmV_Kf8EeNqTDoyuq7UvHVaq-WoBP0",
             .lazy = true,
         },

src/ape/ApeInstance.zig

@@ -0,0 +1,92 @@
+const std = @import("std");
+const vk = @import("vulkan");
+const base = @import("base");
+const soft = @import("soft");
+
+const Dispatchable = base.Dispatchable;
+const VkError = base.VkError;
+
+const Self = @This();
+pub const Interface = base.Instance;
+
+interface: Interface,
+backend_instances: std.ArrayList(*Interface),
+
+pub const EXTENSIONS = soft.Instance.EXTENSIONS;
+
+pub fn create(allocator: std.mem.Allocator, infos: *const vk.InstanceCreateInfo) VkError!*Interface {
+    const self = allocator.create(Self) catch return VkError.OutOfHostMemory;
+    errdefer allocator.destroy(self);
+
+    self.* = .{
+        .interface = try base.Instance.init(allocator, infos),
+        .backend_instances = .empty,
+    };
+    errdefer destroyBackendInstances(self, allocator);
+
+    self.interface.dispatch_table = &.{
+        .destroy = destroy,
+    };
+    self.interface.vtable = &.{
+        .requestPhysicalDevices = requestPhysicalDevices,
+        .releasePhysicalDevices = releasePhysicalDevices,
+        .io = io,
+    };
+
+    const soft_instance = try soft.Instance.create(allocator, infos);
+    errdefer soft_instance.deinit(allocator) catch {};
+    self.backend_instances.append(allocator, soft_instance) catch return VkError.OutOfHostMemory;
+
+    return &self.interface;
+}
+
+fn destroy(interface: *Interface, allocator: std.mem.Allocator) VkError!void {
+    const self: *Self = @alignCast(@fieldParentPtr("interface", interface));
+    destroyBackendInstances(self, allocator);
+    self.backend_instances.deinit(allocator);
+    allocator.destroy(self);
+}
+
+fn requestPhysicalDevices(interface: *Interface, allocator: std.mem.Allocator) VkError!void {
+    const self: *Self = @alignCast(@fieldParentPtr("interface", interface));
+
+    for (self.backend_instances.items) |backend| {
+        try appendBackendPhysicalDevices(self, allocator, backend);
+    }
+}
+
+fn appendBackendPhysicalDevices(self: *Self, allocator: std.mem.Allocator, backend: *Interface) VkError!void {
+    try backend.requestPhysicalDevices(allocator);
+    errdefer backend.releasePhysicalDevices(allocator) catch {};
+
+    self.interface.physical_devices.appendSlice(allocator, backend.physical_devices.items) catch return VkError.OutOfHostMemory;
+    backend.physical_devices.deinit(allocator);
+    backend.physical_devices = .empty;
+}
+
+fn releasePhysicalDevices(interface: *Interface, allocator: std.mem.Allocator) VkError!void {
+    for (interface.physical_devices.items) |physical_device| {
+        try physical_device.object.releasePhysicalDevice(allocator);
+        physical_device.destroy(allocator);
+    }
+
+    interface.physical_devices.deinit(allocator);
+    interface.physical_devices = .empty;
+}
+
+fn destroyBackendInstances(self: *Self, allocator: std.mem.Allocator) void {
+    for (self.backend_instances.items) |backend| {
+        backend.dispatch_table.destroy(backend, allocator) catch |err| {
+            base.errors.errorLogger(err);
+        };
+    }
+    self.backend_instances.clearRetainingCapacity();
+}
+
+fn io(interface: *Interface) std.Io {
+    const self: *Self = @alignCast(@fieldParentPtr("interface", interface));
+    if (self.backend_instances.items.len != 0) {
+        return self.backend_instances.items[0].io();
+    }
+    unreachable;
+}

src/ape/lib.zig

@@ -0,0 +1,28 @@
+const std = @import("std");
+const vk = @import("vulkan");
+pub const base = @import("base");
+pub const soft = @import("soft");
+
+pub const c = base.c;
+
+pub const ApeInstance = @import("ApeInstance.zig");
+
+pub const Instance = ApeInstance;
+
+pub const DRIVER_NAME = "Ape";
+
+pub const VULKAN_VERSION = vk.makeApiVersion(0, 1, 0, 0);
+pub const DRIVER_VERSION = vk.makeApiVersion(0, 0, 0, 1);
+
+pub const std_options = base.std_options;
+
+comptime {
+    _ = base;
+    _ = soft;
+}
+
+test {
+    std.testing.refAllDecls(ApeInstance);
+    std.testing.refAllDecls(soft);
+    std.testing.refAllDecls(base);
+}

src/soft/SoftCommandBuffer.zig

@@ -61,6 +61,7 @@ pub fn create(device: *base.Device, allocator: std.mem.Allocator, info: *const v
         .copyImageToBuffer = copyImageToBuffer,
         .copyQueryPoolResults = copyQueryPoolResults,
         .dispatch = dispatch,
+        .dispatchBase = dispatchBase,
         .dispatchIndirect = dispatchIndirect,
         .draw = draw,
         .drawIndexed = drawIndexed,
@@ -80,6 +81,7 @@ pub fn create(device: *base.Device, allocator: std.mem.Allocator, info: *const v
         .resolveImage = resolveImage,
         .setEvent = setEvent,
         .setBlendConstants = setBlendConstants,
+        .setDeviceMask = setDeviceMask,
         .setScissor = setScissor,
         .setStencilCompareMask = setStencilCompareMask,
         .setStencilReference = setStencilReference,
@@ -782,25 +784,35 @@ pub fn copyQueryPoolResults(interface: *Interface, pool: *base.QueryPool, first:
 }
 
 pub fn dispatch(interface: *Interface, group_count_x: u32, group_count_y: u32, group_count_z: u32) VkError!void {
+    try dispatchBase(interface, 0, 0, 0, group_count_x, group_count_y, group_count_z);
+}
+
+pub fn dispatchBase(interface: *Interface, base_group_x: u32, base_group_y: u32, base_group_z: u32, group_count_x: u32, group_count_y: u32, group_count_z: u32) VkError!void {
     const self: *Self = @alignCast(@fieldParentPtr("interface", interface));
     const allocator = self.command_allocator.allocator();
 
     const CommandImpl = struct {
         const Impl = @This();
 
+        base_group_x: u32,
+        base_group_y: u32,
+        base_group_z: u32,
         group_count_x: u32,
         group_count_y: u32,
         group_count_z: u32,
 
         pub fn execute(context: *anyopaque, device: *ExecutionDevice) VkError!void {
             const impl: *Impl = @ptrCast(@alignCast(context));
-            try device.compute.dispatch(impl.group_count_x, impl.group_count_y, impl.group_count_z);
+            try device.compute.dispatchBase(impl.base_group_x, impl.base_group_y, impl.base_group_z, impl.group_count_x, impl.group_count_y, impl.group_count_z);
         }
     };
 
     const cmd = allocator.create(CommandImpl) catch return VkError.OutOfHostMemory;
     errdefer allocator.destroy(cmd);
     cmd.* = .{
+        .base_group_x = base_group_x,
+        .base_group_y = base_group_y,
+        .base_group_z = base_group_z,
         .group_count_x = group_count_x,
         .group_count_y = group_count_y,
         .group_count_z = group_count_z,
@@ -808,6 +820,10 @@ pub fn dispatch(interface: *Interface, group_count_x: u32, group_count_y: u32, g
     self.commands.append(allocator, .{ .ptr = cmd, .vtable = &.{ .execute = CommandImpl.execute } }) catch return VkError.OutOfHostMemory;
 }
 
+pub fn setDeviceMask(_: *Interface, device_mask: u32) VkError!void {
+    if (device_mask != 1) return VkError.ValidationFailed;
+}
+
 pub fn dispatchIndirect(interface: *Interface, buffer: *base.Buffer, offset: vk.DeviceSize) VkError!void {
     const self: *Self = @alignCast(@fieldParentPtr("interface", interface));
     const allocator = self.command_allocator.allocator();

src/soft/SoftDevice.zig

@@ -42,7 +42,7 @@ const DeviceAllocator = struct {
 };
 
 interface: Interface,
-device_allocator: if (config.debug_allocator) std.heap.DebugAllocator(.{}) else DeviceAllocator,
+device_allocator: if (config.soft_debug_allocator) std.heap.DebugAllocator(.{}) else DeviceAllocator,
 
 pub fn create(instance: *base.Instance, physical_device: *base.PhysicalDevice, allocator: std.mem.Allocator, info: *const vk.DeviceCreateInfo) VkError!*Self {
     const self = allocator.create(Self) catch return VkError.OutOfHostMemory;
@@ -84,11 +84,14 @@ pub fn create(instance: *base.Instance, physical_device: *base.PhysicalDevice, a
         .createSemaphore = createSemaphore,
         .createShaderModule = createShaderModule,
         .destroy = destroy,
+        .getDeviceGroupPeerMemoryFeatures = getDeviceGroupPeerMemoryFeatures,
+        .getDeviceGroupPresentCapabilitiesKHR = getDeviceGroupPresentCapabilitiesKHR,
+        .getDeviceGroupSurfacePresentModesKHR = getDeviceGroupSurfacePresentModesKHR,
     };
 
     self.* = .{
         .interface = interface,
-        .device_allocator = if (config.debug_allocator) .init else .{},
+        .device_allocator = if (config.soft_debug_allocator) .init else .{},
     };
     initialized = true;
 
@@ -99,7 +102,7 @@ pub fn create(instance: *base.Instance, physical_device: *base.PhysicalDevice, a
 pub fn destroy(interface: *Interface, allocator: std.mem.Allocator) VkError!void {
     const self: *Self = @alignCast(@fieldParentPtr("interface", interface));
 
-    if (config.debug_allocator) {
+    if (config.soft_debug_allocator) {
         // All device memory allocations should've been freed by now
         const leaks = self.device_allocator.detectLeaks();
         if (leaks != 0)
@@ -212,3 +215,25 @@ pub fn createShaderModule(interface: *Interface, allocator: std.mem.Allocator, i
     const module = try SoftShaderModule.create(interface, allocator, info);
     return &module.interface;
 }
+
+pub fn getDeviceGroupPeerMemoryFeatures(interface: *Interface, heap_index: u32, local_device_index: u32, remote_device_index: u32) VkError!vk.PeerMemoryFeatureFlags {
+    if (heap_index >= interface.physical_device.mem_props.memory_heap_count) return VkError.ValidationFailed;
+    if (local_device_index != 0 or remote_device_index != 0) return VkError.ValidationFailed;
+
+    return .{
+        .copy_src_bit = true,
+        .copy_dst_bit = true,
+        .generic_src_bit = true,
+        .generic_dst_bit = true,
+    };
+}
+
+pub fn getDeviceGroupPresentCapabilitiesKHR(_: *Interface, capabilities: *vk.DeviceGroupPresentCapabilitiesKHR) VkError!void {
+    capabilities.present_mask = @splat(0);
+    capabilities.present_mask[0] = 1;
+    capabilities.modes = .{ .local_bit_khr = true };
+}
+
+pub fn getDeviceGroupSurfacePresentModesKHR(_: *Interface, _: *base.SurfaceKHR) VkError!vk.DeviceGroupPresentModeFlagsKHR {
+    return .{ .local_bit_khr = true };
+}

src/soft/SoftInstance.zig

@@ -36,7 +36,7 @@ pub fn create(allocator: std.mem.Allocator, infos: *const vk.InstanceCreateInfo)
     errdefer allocator.destroy(self);
 
     self.allocator = std.heap.smp_allocator;
-    self.threaded = if (comptime base.config.single_threaded) .init_single_threaded else std.Io.Threaded.init(self.allocator, .{});
+    self.threaded = if (comptime base.config.soft_single_threaded) .init_single_threaded else std.Io.Threaded.init(self.allocator, .{});
     self.io_impl = self.threaded.io();
 
     self.interface = try base.Instance.init(allocator, infos);

src/soft/SoftPhysicalDevice.zig

@@ -23,6 +23,7 @@ fn castExtension(comptime ext: vk.ApiInfo) vk.ExtensionProperties {
 }
 
 pub const EXTENSIONS = [_]vk.ExtensionProperties{
+    castExtension(vk.extensions.khr_device_group),
     castExtension(vk.extensions.khr_swapchain),
 };
 

src/soft/SoftShaderModule.zig

@@ -37,7 +37,7 @@ pub fn create(device: *base.Device, allocator: std.mem.Allocator, info: *const v
     self.* = .{
         .interface = interface,
         .module = spv.Module.init(device_allocator, code, .{
-            .use_simd_vectors_specializations = base.config.shaders_simd,
+            .use_simd_vectors_specializations = base.config.soft_shaders_simd,
         }) catch |err| switch (err) {
             spv.Module.ModuleError.OutOfMemory => return VkError.OutOfHostMemory,
             else => {

src/soft/device/ComputeDispatcher.zig

@@ -19,6 +19,9 @@ const RunData = struct {
     self: *Self,
     batch_id: usize,
     group_count: usize,
+    base_group_x: usize,
+    base_group_y: usize,
+    base_group_z: usize,
     group_count_x: usize,
     group_count_y: usize,
     group_count_z: usize,
@@ -41,12 +44,16 @@ pub fn init(device: *SoftDevice, state: *PipelineState) Self {
         .state = state,
         .batch_size = 0,
         .invocation_index = .init(0),
-        .early_dump = base.config.compute_dump_early_results_table,
-        .final_dump = base.config.compute_dump_final_results_table,
+        .early_dump = base.config.soft_compute_dump_early_results_table,
+        .final_dump = base.config.soft_compute_dump_final_results_table,
     };
 }
 
 pub fn dispatch(self: *Self, group_count_x: u32, group_count_y: u32, group_count_z: u32) VkError!void {
+    try self.dispatchBase(0, 0, 0, group_count_x, group_count_y, group_count_z);
+}
+
+pub fn dispatchBase(self: *Self, base_group_x: u32, base_group_y: u32, base_group_z: u32, group_count_x: u32, group_count_y: u32, group_count_z: u32) VkError!void {
     const group_count_xy = std.math.mul(usize, group_count_x, group_count_y) catch return VkError.ValidationFailed;
     const group_count = std.math.mul(usize, group_count_xy, group_count_z) catch return VkError.ValidationFailed;
 
@@ -73,6 +80,9 @@ pub fn dispatch(self: *Self, group_count_x: u32, group_count_y: u32, group_count
             .self = self,
             .batch_id = batch_id,
             .group_count = group_count,
+            .base_group_x = @as(usize, @intCast(base_group_x)),
+            .base_group_y = @as(usize, @intCast(base_group_y)),
+            .base_group_z = @as(usize, @intCast(base_group_z)),
             .group_count_x = @as(usize, @intCast(group_count_x)),
             .group_count_y = @as(usize, @intCast(group_count_y)),
             .group_count_z = @as(usize, @intCast(group_count_z)),
@@ -80,9 +90,6 @@ pub fn dispatch(self: *Self, group_count_x: u32, group_count_y: u32, group_count
             .pipeline = pipeline,
         };
 
-        if (comptime base.config.single_threaded_compute)
-            runWrapper(run_data)
-        else
         wg.async(self.device.interface.io(), runWrapper, .{run_data});
     }
     wg.await(self.device.interface.io()) catch return VkError.DeviceLost;
@@ -151,9 +158,9 @@ inline fn run(data: RunData) !void {
             @as(u32, @intCast(data.group_count_z)),
         };
         const group_id_vec = @Vector(3, u32){
-            @as(u32, @intCast(group_x)),
-            @as(u32, @intCast(group_y)),
-            @as(u32, @intCast(group_z)),
+            @as(u32, @intCast(data.base_group_x + group_x)),
+            @as(u32, @intCast(data.base_group_y + group_y)),
+            @as(u32, @intCast(data.base_group_z + group_z)),
         };
 
         if (uses_control_barrier) {
@@ -168,9 +175,9 @@ inline fn run(data: RunData) !void {
             const invocation_index = data.self.invocation_index.fetchAdd(1, .monotonic);
 
             try setupSubgroupBuiltins(data.self, rt, .{
-                @as(u32, @intCast(group_x)),
-                @as(u32, @intCast(group_y)),
-                @as(u32, @intCast(group_z)),
+                @as(u32, @intCast(data.base_group_x + group_x)),
+                @as(u32, @intCast(data.base_group_y + group_y)),
+                @as(u32, @intCast(data.base_group_z + group_z)),
             }, i);
 
             if (data.self.early_dump != null and data.self.early_dump.? == invocation_index) {

src/vulkan/CommandBuffer.zig

@@ -33,6 +33,7 @@ pool: *CommandPool,
 state: State,
 begin_info: ?vk.CommandBufferBeginInfo,
 usage_flags: vk.CommandBufferUsageFlags,
+device_mask: u32,
 host_allocator: VulkanAllocator,
 state_mutex: std.Io.Mutex,
 
@@ -57,6 +58,7 @@ pub const DispatchTable = struct {
     copyImageToBuffer: *const fn (*Self, *Image, vk.ImageLayout, *Buffer, []const vk.BufferImageCopy) VkError!void,
     copyQueryPoolResults: *const fn (*Self, *QueryPool, u32, u32, *Buffer, vk.DeviceSize, vk.DeviceSize, vk.QueryResultFlags) VkError!void,
     dispatch: *const fn (*Self, u32, u32, u32) VkError!void,
+    dispatchBase: *const fn (*Self, u32, u32, u32, u32, u32, u32) VkError!void,
     dispatchIndirect: *const fn (*Self, *Buffer, vk.DeviceSize) VkError!void,
     draw: *const fn (*Self, usize, usize, usize, usize) VkError!void,
     drawIndexed: *const fn (*Self, usize, usize, usize, i32, usize) VkError!void,
@@ -76,6 +78,7 @@ pub const DispatchTable = struct {
     resolveImage: *const fn (*Self, *Image, vk.ImageLayout, *Image, vk.ImageLayout, vk.ImageResolve) VkError!void,
     setEvent: *const fn (*Self, *Event, vk.PipelineStageFlags) VkError!void,
     setBlendConstants: *const fn (*Self, [4]f32) VkError!void,
+    setDeviceMask: *const fn (*Self, u32) VkError!void,
     setScissor: *const fn (*Self, u32, []const vk.Rect2D) VkError!void,
     setStencilCompareMask: *const fn (*Self, vk.StencilFaceFlags, u32) VkError!void,
     setStencilReference: *const fn (*Self, vk.StencilFaceFlags, u32) VkError!void,
@@ -96,6 +99,7 @@ pub fn init(device: *Device, allocator: std.mem.Allocator, info: *const vk.Comma
         .state = .Initial,
         .begin_info = null,
         .usage_flags = .{},
+        .device_mask = 1,
         .host_allocator = VulkanAllocator.from(allocator).cloneWithScope(.object),
         .state_mutex = .init,
         .vtable = undefined,
@@ -152,6 +156,7 @@ pub fn resetFromPool(self: *Self, flags: vk.CommandBufferResetFlags) VkError!voi
     try self.dispatch_table.reset(self, flags);
     self.begin_info = null;
     self.usage_flags = .{};
+    self.device_mask = 1;
 }
 
 pub fn submit(self: *Self) VkError!void {
@@ -247,6 +252,10 @@ pub inline fn dispatch(self: *Self, group_count_x: u32, group_count_y: u32, grou
     try self.dispatch_table.dispatch(self, group_count_x, group_count_y, group_count_z);
 }
 
+pub inline fn dispatchBase(self: *Self, base_group_x: u32, base_group_y: u32, base_group_z: u32, group_count_x: u32, group_count_y: u32, group_count_z: u32) VkError!void {
+    try self.dispatch_table.dispatchBase(self, base_group_x, base_group_y, base_group_z, group_count_x, group_count_y, group_count_z);
+}
+
 pub inline fn dispatchIndirect(self: *Self, buffer: *Buffer, offset: vk.DeviceSize) VkError!void {
     try self.dispatch_table.dispatchIndirect(self, buffer, offset);
 }
@@ -329,6 +338,13 @@ pub inline fn setBlendConstants(self: *Self, constants: [4]f32) VkError!void {
     try self.dispatch_table.setBlendConstants(self, constants);
 }
 
+pub inline fn setDeviceMask(self: *Self, device_mask: u32) VkError!void {
+    if (device_mask != 1)
+        return VkError.ValidationFailed;
+    self.device_mask = device_mask;
+    try self.dispatch_table.setDeviceMask(self, device_mask);
+}
+
 pub inline fn setScissor(self: *Self, first: u32, scissor: []const vk.Rect2D) VkError!void {
     try self.dispatch_table.setScissor(self, first, scissor);
 }

src/vulkan/Device.zig

@@ -72,6 +72,9 @@ pub const DispatchTable = struct {
     createSemaphore: *const fn (*Self, std.mem.Allocator, *const vk.SemaphoreCreateInfo) VkError!*BinarySemaphore,
     createShaderModule: *const fn (*Self, std.mem.Allocator, *const vk.ShaderModuleCreateInfo) VkError!*ShaderModule,
     destroy: *const fn (*Self, std.mem.Allocator) VkError!void,
+    getDeviceGroupPeerMemoryFeatures: *const fn (*Self, u32, u32, u32) VkError!vk.PeerMemoryFeatureFlags,
+    getDeviceGroupPresentCapabilitiesKHR: *const fn (*Self, *vk.DeviceGroupPresentCapabilitiesKHR) VkError!void,
+    getDeviceGroupSurfacePresentModesKHR: *const fn (*Self, *SurfaceKHR) VkError!vk.DeviceGroupPresentModeFlagsKHR,
 };
 
 pub fn init(allocator: std.mem.Allocator, instance: *Instance, physical_device: *const PhysicalDevice, _: *const vk.DeviceCreateInfo) VkError!Self {
@@ -215,3 +218,15 @@ pub fn waitIdle(self: *Self) VkError!void {
         }
     }
 }
+
+pub inline fn getDeviceGroupPeerMemoryFeatures(self: *Self, heap_index: u32, local_device_index: u32, remote_device_index: u32) VkError!vk.PeerMemoryFeatureFlags {
+    return self.dispatch_table.getDeviceGroupPeerMemoryFeatures(self, heap_index, local_device_index, remote_device_index);
+}
+
+pub inline fn getDeviceGroupPresentCapabilitiesKHR(self: *Self, capabilities: *vk.DeviceGroupPresentCapabilitiesKHR) VkError!void {
+    try self.dispatch_table.getDeviceGroupPresentCapabilitiesKHR(self, capabilities);
+}
+
+pub inline fn getDeviceGroupSurfacePresentModesKHR(self: *Self, surface: *SurfaceKHR) VkError!vk.DeviceGroupPresentModeFlagsKHR {
+    return self.dispatch_table.getDeviceGroupSurfacePresentModesKHR(self, surface);
+}

src/vulkan/lib_vulkan.zig

@@ -137,6 +137,7 @@ const device_pfn_map = block: {
     @setEvalBranchQuota(65535);
     break :block std.StaticStringMap(vk.PfnVoidFunction).initComptime(.{
         functionMapEntryPoint("vkAcquireNextImageKHR"),
+        functionMapEntryPoint("vkAcquireNextImage2KHR"),
         functionMapEntryPoint("vkAllocateCommandBuffers"),
         functionMapEntryPoint("vkAllocateDescriptorSets"),
         functionMapEntryPoint("vkAllocateDescriptorSets"),
@@ -160,6 +161,8 @@ const device_pfn_map = block: {
         functionMapEntryPoint("vkCmdCopyImageToBuffer"),
         functionMapEntryPoint("vkCmdCopyQueryPoolResults"),
         functionMapEntryPoint("vkCmdDispatch"),
+        functionMapEntryPoint("vkCmdDispatchBase"),
+        functionMapEntryPoint("vkCmdDispatchBaseKHR"),
         functionMapEntryPoint("vkCmdDispatchIndirect"),
         functionMapEntryPoint("vkCmdDraw"),
         functionMapEntryPoint("vkCmdDrawIndexed"),
@@ -178,6 +181,8 @@ const device_pfn_map = block: {
         functionMapEntryPoint("vkCmdSetBlendConstants"),
         functionMapEntryPoint("vkCmdSetDepthBias"),
         functionMapEntryPoint("vkCmdSetDepthBounds"),
+        functionMapEntryPoint("vkCmdSetDeviceMask"),
+        functionMapEntryPoint("vkCmdSetDeviceMaskKHR"),
         functionMapEntryPoint("vkCmdSetEvent"),
         functionMapEntryPoint("vkCmdSetLineWidth"),
         functionMapEntryPoint("vkCmdSetScissor"),
@@ -236,6 +241,10 @@ const device_pfn_map = block: {
         functionMapEntryPoint("vkFreeMemory"),
         functionMapEntryPoint("vkGetBufferMemoryRequirements"),
         functionMapEntryPoint("vkGetDeviceMemoryCommitment"),
+        functionMapEntryPoint("vkGetDeviceGroupPeerMemoryFeatures"),
+        functionMapEntryPoint("vkGetDeviceGroupPeerMemoryFeaturesKHR"),
+        functionMapEntryPoint("vkGetDeviceGroupPresentCapabilitiesKHR"),
+        functionMapEntryPoint("vkGetDeviceGroupSurfacePresentModesKHR"),
         functionMapEntryPoint("vkGetDeviceProcAddr"),
         functionMapEntryPoint("vkGetDeviceQueue"),
         functionMapEntryPoint("vkGetEventStatus"),
@@ -402,16 +411,12 @@ pub export fn apeEnumeratePhysicalDevices(p_instance: vk.Instance, count: *u32,
     return .success;
 }
 
-pub export fn apeEnumeratePhysicalDeviceGroups(
-    p_instance: vk.Instance,
-    count: *u32,
-    p_groups: ?[*]vk.PhysicalDeviceGroupProperties,
-) callconv(vk.vulkan_call_conv) vk.Result {
+pub export fn apeEnumeratePhysicalDeviceGroups(p_instance: vk.Instance, count: *u32, p_groups: ?[*]vk.PhysicalDeviceGroupProperties) callconv(vk.vulkan_call_conv) vk.Result {
     entryPointBeginLogTrace(.vkEnumeratePhysicalDeviceGroups);
     defer entryPointEndLogTrace();
 
     const instance = Dispatchable(Instance).fromHandleObject(p_instance) catch |err| return toVkResult(err);
-    const available: u32 = 1;
+    const available: u32 = @intCast(instance.physical_devices.items.len);
 
     if (p_groups) |groups| {
         const write_count = @min(count.*, available);
@@ -420,16 +425,15 @@ pub export fn apeEnumeratePhysicalDeviceGroups(
             return .incomplete;
         }
 
-        var group = groups[0];
-        group.physical_device_count = @intCast(instance.physical_devices.items.len);
+        for (groups[0..write_count], instance.physical_devices.items[0..write_count]) |*group, physical_device| {
+            group.physical_device_count = 1;
             group.physical_devices = @splat(.null_handle);
-        for (instance.physical_devices.items, 0..) |physical_device, i| {
-            group.physical_devices[i] = physical_device.toVkHandle(vk.PhysicalDevice);
-        }
+            group.physical_devices[0] = physical_device.toVkHandle(vk.PhysicalDevice);
             group.subset_allocation = .false;
-        groups[0] = group;
+        }
 
         count.* = write_count;
+        if (write_count < available) return .incomplete;
         return .success;
     }
 
@@ -437,12 +441,8 @@ pub export fn apeEnumeratePhysicalDeviceGroups(
     return .success;
 }
 
-pub export fn apeEnumeratePhysicalDeviceGroupsKHR(
-    p_instance: vk.Instance,
-    count: *u32,
-    p_groups: ?[*]vk.PhysicalDeviceGroupProperties,
-) callconv(vk.vulkan_call_conv) vk.Result {
-    return apeEnumeratePhysicalDeviceGroups(p_instance, count, p_groups);
+pub export fn apeEnumeratePhysicalDeviceGroupsKHR(p_instance: vk.Instance, count: *u32, p_groups: ?[*]vk.PhysicalDeviceGroupProperties) callconv(vk.vulkan_call_conv) vk.Result {
+    return @call(.always_inline, apeEnumeratePhysicalDeviceGroups, .{ p_instance, count, p_groups });
 }
 
 // Physical Device functions =================================================================================================================================
@@ -1412,6 +1412,60 @@ pub export fn apeGetDeviceMemoryCommitment(p_device: vk.Device, p_memory: vk.Dev
     _ = committed_memory;
 }
 
+pub export fn apeGetDeviceGroupPeerMemoryFeatures(
+    p_device: vk.Device,
+    heap_index: u32,
+    local_device_index: u32,
+    remote_device_index: u32,
+    p_peer_memory_features: *vk.PeerMemoryFeatureFlags,
+) callconv(vk.vulkan_call_conv) void {
+    entryPointBeginLogTrace(.vkGetDeviceGroupPeerMemoryFeatures);
+    defer entryPointEndLogTrace();
+
+    const device = Dispatchable(Device).fromHandleObject(p_device) catch |err| return errorLogger(err);
+    p_peer_memory_features.* = device.getDeviceGroupPeerMemoryFeatures(heap_index, local_device_index, remote_device_index) catch |err| return errorLogger(err);
+}
+
+pub export fn apeGetDeviceGroupPeerMemoryFeaturesKHR(
+    p_device: vk.Device,
+    heap_index: u32,
+    local_device_index: u32,
+    remote_device_index: u32,
+    p_peer_memory_features: *vk.PeerMemoryFeatureFlags,
+) callconv(vk.vulkan_call_conv) void {
+    apeGetDeviceGroupPeerMemoryFeatures(p_device, heap_index, local_device_index, remote_device_index, p_peer_memory_features);
+}
+
+pub export fn apeGetDeviceGroupPresentCapabilitiesKHR(
+    p_device: vk.Device,
+    p_device_group_present_capabilities: *vk.DeviceGroupPresentCapabilitiesKHR,
+) callconv(vk.vulkan_call_conv) vk.Result {
+    entryPointBeginLogTrace(.vkGetDeviceGroupPresentCapabilitiesKHR);
+    defer entryPointEndLogTrace();
+
+    if (p_device_group_present_capabilities.s_type != .device_group_present_capabilities_khr) {
+        return .error_validation_failed;
+    }
+
+    const device = Dispatchable(Device).fromHandleObject(p_device) catch |err| return toVkResult(err);
+    device.getDeviceGroupPresentCapabilitiesKHR(p_device_group_present_capabilities) catch |err| return toVkResult(err);
+    return .success;
+}
+
+pub export fn apeGetDeviceGroupSurfacePresentModesKHR(
+    p_device: vk.Device,
+    p_surface: vk.SurfaceKHR,
+    p_modes: *vk.DeviceGroupPresentModeFlagsKHR,
+) callconv(vk.vulkan_call_conv) vk.Result {
+    entryPointBeginLogTrace(.vkGetDeviceGroupSurfacePresentModesKHR);
+    defer entryPointEndLogTrace();
+
+    const device = Dispatchable(Device).fromHandleObject(p_device) catch |err| return toVkResult(err);
+    const surface = NonDispatchable(SurfaceKHR).fromHandleObject(p_surface) catch |err| return toVkResult(err);
+    p_modes.* = device.getDeviceGroupSurfacePresentModesKHR(surface) catch |err| return toVkResult(err);
+    return .success;
+}
+
 pub export fn apeGetDeviceProcAddr(p_device: vk.Device, p_name: ?[*:0]const u8) callconv(vk.vulkan_call_conv) vk.PfnVoidFunction {
     defer entryPointEndLogTrace();
 
@@ -1876,6 +1930,18 @@ pub export fn apeCmdDispatch(p_cmd: vk.CommandBuffer, group_count_x: u32, group_
     cmd.dispatch(group_count_x, group_count_y, group_count_z) catch |err| return errorLogger(err);
 }
 
+pub export fn apeCmdDispatchBase(p_cmd: vk.CommandBuffer, base_group_x: u32, base_group_y: u32, base_group_z: u32, group_count_x: u32, group_count_y: u32, group_count_z: u32) callconv(vk.vulkan_call_conv) void {
+    entryPointBeginLogTrace(.vkCmdDispatchBase);
+    defer entryPointEndLogTrace();
+
+    const cmd = Dispatchable(CommandBuffer).fromHandleObject(p_cmd) catch |err| return errorLogger(err);
+    cmd.dispatchBase(base_group_x, base_group_y, base_group_z, group_count_x, group_count_y, group_count_z) catch |err| return errorLogger(err);
+}
+
+pub export fn apeCmdDispatchBaseKHR(p_cmd: vk.CommandBuffer, base_group_x: u32, base_group_y: u32, base_group_z: u32, group_count_x: u32, group_count_y: u32, group_count_z: u32) callconv(vk.vulkan_call_conv) void {
+    @call(.always_inline, apeCmdDispatchBase, .{ p_cmd, base_group_x, base_group_y, base_group_z, group_count_x, group_count_y, group_count_z });
+}
+
 pub export fn apeCmdDispatchIndirect(p_cmd: vk.CommandBuffer, p_buffer: vk.Buffer, offset: vk.DeviceSize) callconv(vk.vulkan_call_conv) void {
     entryPointBeginLogTrace(.vkCmdDispatchIndirect);
     defer entryPointEndLogTrace();
@@ -2073,6 +2139,18 @@ pub export fn apeCmdSetDepthBounds(p_cmd: vk.CommandBuffer, min: f32, max: f32)
     _ = max;
 }
 
+pub export fn apeCmdSetDeviceMask(p_cmd: vk.CommandBuffer, device_mask: u32) callconv(vk.vulkan_call_conv) void {
+    entryPointBeginLogTrace(.vkCmdSetDeviceMask);
+    defer entryPointEndLogTrace();
+
+    const cmd = Dispatchable(CommandBuffer).fromHandleObject(p_cmd) catch |err| return errorLogger(err);
+    cmd.setDeviceMask(device_mask) catch |err| return errorLogger(err);
+}
+
+pub export fn apeCmdSetDeviceMaskKHR(p_cmd: vk.CommandBuffer, device_mask: u32) callconv(vk.vulkan_call_conv) void {
+    @call(.always_inline, apeCmdSetDeviceMask, .{ p_cmd, device_mask });
+}
+
 pub export fn apeCmdSetEvent(p_cmd: vk.CommandBuffer, p_event: vk.Event, stage_mask: vk.PipelineStageFlags) callconv(vk.vulkan_call_conv) void {
     entryPointBeginLogTrace(.vkCmdSetEvent);
     defer entryPointEndLogTrace();
@@ -2221,6 +2299,27 @@ pub export fn apeAcquireNextImageKHR(p_device: vk.Device, p_swapchain: vk.Swapch
     return .success;
 }
 
+pub export fn apeAcquireNextImage2KHR(p_device: vk.Device, p_acquire_info: *const vk.AcquireNextImageInfoKHR, image_index: *u32) callconv(vk.vulkan_call_conv) vk.Result {
+    entryPointBeginLogTrace(.vkAcquireNextImage2KHR);
+    defer entryPointEndLogTrace();
+
+    if (p_acquire_info.s_type != .acquire_next_image_info_khr) {
+        return .error_validation_failed;
+    }
+
+    if (p_acquire_info.device_mask != 1) {
+        return .error_validation_failed;
+    }
+
+    Dispatchable(Device).checkHandleValidity(p_device) catch |err| return toVkResult(err);
+
+    const swapchain = NonDispatchable(SwapchainKHR).fromHandleObject(p_acquire_info.swapchain) catch |err| return toVkResult(err);
+    const semaphore = if (p_acquire_info.semaphore != .null_handle) NonDispatchable(BinarySemaphore).fromHandleObject(p_acquire_info.semaphore) catch |err| return toVkResult(err) else null;
+    const fence = if (p_acquire_info.fence != .null_handle) NonDispatchable(Fence).fromHandleObject(p_acquire_info.fence) catch |err| return toVkResult(err) else null;
+    swapchain.getNextImage(p_acquire_info.timeout, semaphore, fence, image_index) catch |err| return toVkResult(err);
+    return .success;
+}
+
 pub export fn apeCreateSwapchainKHR(p_device: vk.Device, info: *const vk.SwapchainCreateInfoKHR, callbacks: ?*const vk.AllocationCallbacks, p_swapchain: *vk.SwapchainKHR) callconv(vk.vulkan_call_conv) vk.Result {
     entryPointBeginLogTrace(.vkCreateSwapchainKHR);
     defer entryPointEndLogTrace();