fixing vertex outputs

build.zig

@@ -88,10 +88,12 @@ pub fn build(b: *std.Build) !void {
     const drm = b.dependency("drm", .{}).module("drm");
 
     const logs_option: LogType = b.option(LogType, "logs", "Driver logs") orelse .none;
+    const debug_allocator_option = b.option(bool, "device-debug-allocator", "Debug device allocator") orelse false;
 
     const options = b.addOptions();
     options.addOption(std.SemanticVersion, "driver_version", driver_version);
     options.addOption(LogType, "logs", logs_option);
+    options.addOption(bool, "device_debug_allocator", debug_allocator_option);
 
     base_mod.addImport("vulkan", vulkan);
     base_mod.addImport("zmath", zmath);
@@ -274,14 +276,12 @@ fn customSoft(
 
 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);

build.zig.zon

@@ -32,8 +32,8 @@
 
         // Soft dependencies
         .SPIRV_Interpreter = .{
-            .url = "git+https://github.com/Kbz-8/SPIRV-Interpreter#445bb93296bd6da4a74aa40152aaa779170887cf",
+            .url = "git+https://github.com/Kbz-8/SPIRV-Interpreter#53da7f93154086574bf41a13130299d1bc6df670",
-            .hash = "SPIRV_Interpreter-0.0.1-ajmpn113CQDBhtMXf88no5gcMJhlricW5G_rWc7AvRS5",
+            .hash = "SPIRV_Interpreter-0.0.1-ajmpn95-CQCUUA6EfT9id_Y50N7kjcKftj0VrGLyi751",
             .lazy = true,
         },
         //.SPIRV_Interpreter = .{

src/intel/FlintInstance.zig

@@ -76,7 +76,20 @@ fn requestPhysicalDevices(interface: *Interface, allocator: std.mem.Allocator, d
             drm_device.device_info.pci.vendor_id != lib.INTEL_PCI_VENDOR_ID)
             continue;
 
-        const physical_device = try FlintPhysicalDevice.create(allocator, interface, &drm_device);
+        const version = device.getVersion(io_var, allocator) catch continue;
+        defer version.deinit(allocator);
+
+        const kmd_type: lib.KmdType = if (std.mem.eql(u8, version.name, "i915"))
+            .I915
+        else if (std.mem.eql(u8, version.name, "xe"))
+            .Xe
+        else
+            .Invalid;
+
+        if (kmd_type == .Invalid)
+            continue;
+
+        const physical_device = try FlintPhysicalDevice.create(allocator, interface, &drm_device, kmd_type);
         errdefer physical_device.interface.release(allocator) catch {};
 
         const dispatchable = try Dispatchable(base.PhysicalDevice).wrap(allocator, &physical_device.interface);

src/intel/FlintPhysicalDevice.zig

@@ -30,8 +30,9 @@ pub const EXTENSIONS = [_]vk.ExtensionProperties{
 };
 
 interface: Interface,
+kmd_type: lib.KmdType,
 
-pub fn create(allocator: std.mem.Allocator, instance: *base.Instance, drm_device: *const base.drm.Device) VkError!*Self {
+pub fn create(allocator: std.mem.Allocator, instance: *base.Instance, drm_device: *const base.drm.Device, kmd_type: lib.KmdType) VkError!*Self {
     const self = allocator.create(Self) catch return VkError.OutOfHostMemory;
     errdefer allocator.destroy(self);
 
@@ -223,6 +224,7 @@ pub fn create(allocator: std.mem.Allocator, instance: *base.Instance, drm_device
 
     self.* = .{
         .interface = interface,
+        .kmd_type = kmd_type,
     };
 
     return self;

src/phi/PhiDevice.zig

@@ -31,14 +31,7 @@ const VkError = base.VkError;
 const Self = @This();
 pub const Interface = base.Device;
 
-const DeviceAllocator = struct {
-    pub inline fn allocator(_: @This()) std.mem.Allocator {
-        return base.fallback_host_allocator;
-    }
-};
-
 interface: Interface,
-allocator: 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;
@@ -80,7 +73,6 @@ pub fn create(instance: *base.Instance, physical_device: *base.PhysicalDevice, a
 
     self.* = .{
         .interface = interface,
-        .allocator = .{},
     };
 
     try self.interface.createQueues(allocator, info);

src/software/SoftCommandBuffer.zig

@@ -195,7 +195,7 @@ pub fn beginQuery(interface: *Interface, pool: *base.QueryPool, query: u32, _: v
                     if (active.pool == impl.pool and active.query == impl.query)
                         return;
                 }
-                device.active_occlusion_queries.append(device.renderer.device.device_allocator.allocator(), .{
+                device.active_occlusion_queries.append(device.renderer.device.interface.device_allocator.allocator(), .{
                     .pool = impl.pool,
                     .query = impl.query,
                 }) catch return VkError.OutOfDeviceMemory;

src/software/SoftDevice.zig

@@ -35,14 +35,7 @@ pub const Interface = base.Device;
 
 const SpawnError = std.Thread.SpawnError;
 
-const DeviceAllocator = struct {
-    pub inline fn allocator(_: @This()) std.mem.Allocator {
-        return base.fallback_host_allocator;
-    }
-};
-
 interface: Interface,
-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;
@@ -91,7 +84,6 @@ pub fn create(instance: *base.Instance, physical_device: *base.PhysicalDevice, a
 
     self.* = .{
         .interface = interface,
-        .device_allocator = if (config.soft_debug_allocator) .init else .{},
     };
     initialized = true;
 
@@ -102,16 +94,6 @@ 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.soft_debug_allocator) {
-        // All device memory allocations should've been freed by now
-        const leaks = self.device_allocator.detectLeaks();
-        if (leaks != 0)
-            std.log.scoped(.vkDestroyDevice).err("Device was destroyed leaking {d:.3}KB", .{@as(f32, @floatFromInt(leaks)) / 1000})
-        else
-            std.log.scoped(.vkDestroyDevice).debug("No device memory leaks detected", .{});
-        self.device_allocator.deinitWithoutLeakChecks();
-    }
-
     allocator.destroy(self);
 }
 

src/software/SoftDeviceMemory.zig

@@ -28,15 +28,14 @@ pub fn create(device: *SoftDevice, allocator: std.mem.Allocator, size: vk.Device
 
     self.* = .{
         .interface = interface,
-        .data = device.device_allocator.allocator().alloc(u8, size) catch return VkError.OutOfDeviceMemory,
+        .data = device.interface.device_allocator.allocator().alloc(u8, size) catch return VkError.OutOfDeviceMemory,
     };
     return self;
 }
 
 pub fn destroy(interface: *Interface, allocator: std.mem.Allocator) void {
     const self: *Self = @alignCast(@fieldParentPtr("interface", interface));
-    const soft_device: *SoftDevice = @alignCast(@fieldParentPtr("interface", interface.owner));
+    interface.owner.device_allocator.allocator().free(self.data);
-    soft_device.device_allocator.allocator().free(self.data);
     allocator.destroy(self);
 }
 

src/software/SoftPipeline.zig

@@ -79,11 +79,10 @@ pub fn createCompute(device: *base.Device, allocator: std.mem.Allocator, cache:
         .destroy = destroy,
     };
 
-    const soft_device: *SoftDevice = @alignCast(@fieldParentPtr("interface", device));
     const module = try NonDispatchable(ShaderModule).fromHandleObject(info.stage.module);
     const soft_module: *SoftShaderModule = @alignCast(@fieldParentPtr("interface", module));
 
-    const device_allocator = soft_device.device_allocator.allocator();
+    const device_allocator = device.device_allocator.allocator();
     const soft_cache: ?*SoftPipelineCache = if (cache) |pipeline_cache|
         @alignCast(@fieldParentPtr("interface", pipeline_cache))
     else
@@ -123,8 +122,7 @@ pub fn createGraphics(device: *base.Device, allocator: std.mem.Allocator, cache:
         .destroy = destroy,
     };
 
-    const soft_device: *SoftDevice = @alignCast(@fieldParentPtr("interface", device));
+    const device_allocator = device.device_allocator.allocator();
-    const device_allocator = soft_device.device_allocator.allocator();
     const soft_cache: ?*SoftPipelineCache = if (cache) |pipeline_cache|
         @alignCast(@fieldParentPtr("interface", pipeline_cache))
     else
@@ -185,8 +183,7 @@ pub fn createGraphics(device: *base.Device, allocator: std.mem.Allocator, cache:
 
 pub fn destroy(interface: *Interface, allocator: std.mem.Allocator) void {
     const self: *Self = @alignCast(@fieldParentPtr("interface", interface));
-    const soft_device: *SoftDevice = @alignCast(@fieldParentPtr("interface", interface.owner));
+    const device_allocator = interface.owner.device_allocator.allocator();
-    const device_allocator = soft_device.device_allocator.allocator();
 
     var it = self.stages.iterator();
     while (it.next()) |entry| {
@@ -486,10 +483,22 @@ fn readImageFloat4(context: *anyopaque, dim: spv.SpvDim, x: i32, y: i32, z: i32,
             .@"2d", .@"2d_array", .cube, .cube_array => .{ .x = x, .y = y, .z = 0 },
             else => .{ .x = x, .y = y, .z = z },
         };
+        const sample_count = image.interface.samples.toInt();
+        const sample_index: u32 = if (sample_count > 1) blk: {
+            if (image_view.interface.view_type == .@"2d_array" or image_view.interface.view_type == .cube_array)
+                break :blk @intCast(@mod(z, @as(i32, @intCast(sample_count))));
+            break :blk @intCast(z);
+        } else 0;
+        const array_z: i32 = if (sample_count > 1 and
+            (image_view.interface.view_type == .@"2d_array" or image_view.interface.view_type == .cube_array))
+            @divFloor(z, @as(i32, @intCast(sample_count)))
+        else
+            z;
         const array_layer = image_view.interface.subresource_range.base_array_layer + switch (image_view.interface.view_type) {
             .@"1d_array" => @as(u32, @intCast(y)),
-            .@"2d_array" => @as(u32, @intCast(z)),
+            .@"2d_array" => @as(u32, @intCast(array_z)),
             .cube => cube_face,
+            .cube_array => @as(u32, @intCast(array_z)),
             else => 0,
         };
         const aspect_mask = imageReadAspect(image_view, false);
@@ -498,7 +507,6 @@ fn readImageFloat4(context: *anyopaque, dim: spv.SpvDim, x: i32, y: i32, z: i32,
             .mip_level = mip_level,
             .array_layer = array_layer,
         };
-        const sample_index: u32 = if (image.interface.samples.toInt() > 1) @intCast(z) else 0;
         const format = base.format.fromAspect(image_view.interface.format, aspect_mask);
         const raw_pixel = if (dim == .SubpassData) blk: {
             if (findInputAttachmentSnapshot(image_view, subresource)) |snapshot| {
@@ -541,10 +549,22 @@ fn readImageInt4(context: *anyopaque, dim: spv.SpvDim, x: i32, y: i32, z: i32, l
             .@"2d", .@"2d_array", .cube, .cube_array => .{ .x = x, .y = y, .z = 0 },
             else => .{ .x = x, .y = y, .z = z },
         };
+        const sample_count = image.interface.samples.toInt();
+        const sample_index: u32 = if (sample_count > 1) blk: {
+            if (image_view.interface.view_type == .@"2d_array" or image_view.interface.view_type == .cube_array)
+                break :blk @intCast(@mod(z, @as(i32, @intCast(sample_count))));
+            break :blk @intCast(z);
+        } else 0;
+        const array_z: i32 = if (sample_count > 1 and
+            (image_view.interface.view_type == .@"2d_array" or image_view.interface.view_type == .cube_array))
+            @divFloor(z, @as(i32, @intCast(sample_count)))
+        else
+            z;
         const array_layer = image_view.interface.subresource_range.base_array_layer + switch (image_view.interface.view_type) {
             .@"1d_array" => @as(u32, @intCast(y)),
-            .@"2d_array" => @as(u32, @intCast(z)),
+            .@"2d_array" => @as(u32, @intCast(array_z)),
             .cube => cube_face,
+            .cube_array => @as(u32, @intCast(array_z)),
             else => 0,
         };
         const aspect_mask = imageReadAspect(image_view, true);
@@ -553,7 +573,6 @@ fn readImageInt4(context: *anyopaque, dim: spv.SpvDim, x: i32, y: i32, z: i32, l
             .mip_level = mip_level,
             .array_layer = array_layer,
         };
-        const sample_index: u32 = if (image.interface.samples.toInt() > 1) @intCast(z) else 0;
         const format = base.format.fromAspect(image_view.interface.format, aspect_mask);
         const raw_pixel = if (dim == .SubpassData) blk: {
             if (findInputAttachmentSnapshot(image_view, subresource)) |snapshot| {

src/software/SoftPipelineCache.zig

@@ -67,8 +67,7 @@ pub fn create(device: *base.Device, allocator: std.mem.Allocator, info: *const v
 
 pub fn destroy(interface: *Interface, allocator: std.mem.Allocator) void {
     const self: *Self = @alignCast(@fieldParentPtr("interface", interface));
-    const soft_device: *SoftDevice = @alignCast(@fieldParentPtr("interface", interface.owner));
+    const data_allocator = interface.owner.device_allocator.allocator();
-    const data_allocator = soft_device.device_allocator.allocator();
 
     for (self.shaders.items) |*shader| {
         shader.deinit(allocator, data_allocator);

src/software/SoftQueue.zig

@@ -69,7 +69,7 @@ pub fn submit(interface: *Interface, infos: []Interface.SubmitInfo, p_fence: ?*b
     const self: *Self = @alignCast(@fieldParentPtr("interface", interface));
     const soft_device: *SoftDevice = @alignCast(@fieldParentPtr("interface", interface.owner));
     const io = interface.owner.io();
-    const allocator = soft_device.device_allocator.allocator();
+    const allocator = interface.owner.device_allocator.allocator();
 
     const data = allocator.create(TaskData) catch return VkError.OutOfDeviceMemory;
     errdefer allocator.destroy(data);
@@ -111,7 +111,7 @@ fn executeSubmitInfo(soft_device: *SoftDevice, info: Interface.SubmitInfo) VkErr
 
     var execution_device: ExecutionDevice = undefined;
     execution_device.setup(soft_device);
-    defer execution_device.deinit(soft_device.device_allocator.allocator());
+    defer execution_device.deinit(soft_device.interface.device_allocator.allocator());
 
     for (info.command_buffers.items) |command_buffer| {
         const soft_command_buffer: *SoftCommandBuffer = @alignCast(@fieldParentPtr("interface", command_buffer));
@@ -125,7 +125,7 @@ fn executeSubmitInfo(soft_device: *SoftDevice, info: Interface.SubmitInfo) VkErr
 
 fn taskRunner(data: *TaskData) void {
     const io = data.queue.interface.owner.io();
-    const allocator = data.soft_device.device_allocator.allocator();
+    const allocator = data.soft_device.interface.device_allocator.allocator();
 
     defer {
         deinitSubmitInfos(allocator, &data.infos);

src/software/SoftSampler.zig

@@ -276,7 +276,6 @@ pub fn queryImageLod(image: *SoftImage, image_view: *SoftImageView, sampler: *Se
     const clamped_lod = std.math.clamp(biased_lod, sampler.interface.min_lod, sampler.interface.max_lod);
     const max_level: f32 = @floatFromInt(viewMipCount(image_view) - 1);
     const level = std.math.clamp(mipmapModeLevel(sampler, clamped_lod), 0.0, max_level);
-
     return .{ level, lod, 0.0, 0.0 };
 }
 

src/software/SoftShaderModule.zig

@@ -25,8 +25,7 @@ pub fn create(device: *base.Device, allocator: std.mem.Allocator, info: *const v
 
     var interface = try Interface.init(device, allocator, info);
 
-    const soft_device: *SoftDevice = @alignCast(@fieldParentPtr("interface", device));
+    const device_allocator = device.device_allocator.allocator();
-    const device_allocator = soft_device.device_allocator.allocator();
 
     interface.vtable = &.{
         .destroy = destroy,
@@ -62,8 +61,7 @@ pub fn destroy(interface: *Interface, allocator: std.mem.Allocator) void {
 }
 
 pub fn drop(self: *Self, allocator: std.mem.Allocator) void {
-    const soft_device: *SoftDevice = @alignCast(@fieldParentPtr("interface", self.interface.owner));
+    const device_allocator = self.interface.owner.device_allocator.allocator();
-    const device_allocator = soft_device.device_allocator.allocator();
 
     self.module.deinit(device_allocator);
 

src/software/device/ComputeDispatcher.zig

@@ -75,7 +75,7 @@ pub fn dispatchBase(self: *Self, base_group_x: u32, base_group_y: u32, base_grou
     const spv_module = &shader.module.module;
     self.batch_size = if (spv_module.reflection_infos.has_atomics) 1 else shader.runtimes.len;
 
-    const allocator = self.device.device_allocator.allocator();
+    const allocator = self.device.interface.device_allocator.allocator();
     const local_size = try getLocalSize(&shader.runtimes[0].rt, allocator, spv_module);
     const local_size_xy = std.math.mul(usize, local_size[0], local_size[1]) catch return VkError.ValidationFailed;
     const invocations_per_workgroup = std.math.mul(usize, local_size_xy, local_size[2]) catch return VkError.ValidationFailed;
@@ -124,7 +124,7 @@ fn runWrapper(data: RunData) void {
 }
 
 inline fn run(data: RunData) !void {
-    const allocator = data.self.device.device_allocator.allocator();
+    const allocator = data.self.device.interface.device_allocator.allocator();
     const io = data.self.device.interface.io();
 
     const shader = data.pipeline.stages.getPtrAssertContains(.compute);
@@ -233,7 +233,8 @@ fn runBarrierWorkgroup(
     group_count: @Vector(3, u32),
     group_id: @Vector(3, u32),
 ) !void {
-    const allocator = data.self.device.device_allocator.allocator();
+    const allocator = data.self.device.interface.device_allocator.allocator();
+
     const workgroup_memory = try runtimes[0].createWorkgroupMemory(allocator);
     defer runtimes[0].destroyWorkgroupMemory(allocator, workgroup_memory);
     for (runtimes, 0..) |*rt, i| {
@@ -292,21 +293,25 @@ fn dumpResultsTable(allocator: std.mem.Allocator, io: std.Io, rt: *spv.Runtime,
 }
 
 fn setupWorkgroupBuiltins(self: *Self, rt: *spv.Runtime, local_size: @Vector(3, u32), group_count: @Vector(3, u32), group_id: @Vector(3, u32)) spv.Runtime.RuntimeError!void {
-    rt.writeBuiltIn(self.device.device_allocator.allocator(), std.mem.asBytes(&local_size), .WorkgroupSize) catch |err| switch (err) {
+    const allocator = self.device.interface.device_allocator.allocator();
+
+    rt.writeBuiltIn(allocator, std.mem.asBytes(&local_size), .WorkgroupSize) catch |err| switch (err) {
         SpvRuntimeError.NotFound => {},
         else => return err,
     };
-    rt.writeBuiltIn(self.device.device_allocator.allocator(), std.mem.asBytes(&group_count), .NumWorkgroups) catch |err| switch (err) {
+    rt.writeBuiltIn(allocator, std.mem.asBytes(&group_count), .NumWorkgroups) catch |err| switch (err) {
         SpvRuntimeError.NotFound => {},
         else => return err,
     };
-    rt.writeBuiltIn(self.device.device_allocator.allocator(), std.mem.asBytes(&group_id), .WorkgroupId) catch |err| switch (err) {
+    rt.writeBuiltIn(allocator, std.mem.asBytes(&group_id), .WorkgroupId) catch |err| switch (err) {
         SpvRuntimeError.NotFound => {},
         else => return err,
     };
 }
 
 fn setupSubgroupBuiltins(self: *Self, rt: *spv.Runtime, local_size: @Vector(3, u32), group_id: @Vector(3, u32), local_invocation_index: usize) spv.Runtime.RuntimeError!void {
+    const allocator = self.device.interface.device_allocator.allocator();
+
     const local_base = local_size * group_id;
     var local_invocation = @Vector(3, u32){ 0, 0, 0 };
 
@@ -320,15 +325,15 @@ fn setupSubgroupBuiltins(self: *Self, rt: *spv.Runtime, local_size: @Vector(3, u
     const global_invocation_index = local_base + local_invocation;
     const local_invocation_index_u32: u32 = @intCast(local_invocation_index);
 
-    rt.writeBuiltIn(self.device.device_allocator.allocator(), std.mem.asBytes(&local_invocation), .LocalInvocationId) catch |err| switch (err) {
+    rt.writeBuiltIn(allocator, std.mem.asBytes(&local_invocation), .LocalInvocationId) catch |err| switch (err) {
         SpvRuntimeError.NotFound => {},
         else => return err,
     };
-    rt.writeBuiltIn(self.device.device_allocator.allocator(), std.mem.asBytes(&local_invocation_index_u32), .LocalInvocationIndex) catch |err| switch (err) {
+    rt.writeBuiltIn(allocator, std.mem.asBytes(&local_invocation_index_u32), .LocalInvocationIndex) catch |err| switch (err) {
         SpvRuntimeError.NotFound => {},
         else => return err,
     };
-    rt.writeBuiltIn(self.device.device_allocator.allocator(), std.mem.asBytes(&global_invocation_index), .GlobalInvocationId) catch |err| switch (err) {
+    rt.writeBuiltIn(allocator, std.mem.asBytes(&global_invocation_index), .GlobalInvocationId) catch |err| switch (err) {
         SpvRuntimeError.NotFound => {},
         else => return err,
     };

src/software/device/Renderer.zig

@@ -187,7 +187,7 @@ pub fn init(device: *SoftDevice, state: *PipelineState, active_occlusion_queries
 }
 
 pub fn resetInputAttachmentSnapshots(self: *Self) void {
-    const allocator = self.device.device_allocator.allocator();
+    const allocator = self.device.interface.device_allocator.allocator();
     for (self.input_attachment_snapshots) |snapshot| {
         allocator.free(snapshot.data);
     }
@@ -196,12 +196,12 @@ pub fn resetInputAttachmentSnapshots(self: *Self) void {
 }
 
 pub fn draw(self: *Self, vertex_count: usize, instance_count: usize, first_vertex: usize, first_instance: usize) VkError!void {
-    var bounded_allocator: BoundedAllocator = .init(self.device.device_allocator.allocator(), 4 * @"1GiB");
+    var bounded_allocator: BoundedAllocator = .init(self.device.interface.device_allocator.allocator(), 4 * @"1GiB");
     try self.drawCall(&bounded_allocator, vertex_count, instance_count, first_vertex, first_instance, null, null);
 }
 
 pub fn drawIndexed(self: *Self, index_count: usize, instance_count: usize, first_index: usize, first_instance: usize, vertex_offset: i32) VkError!void {
-    var bounded_allocator: BoundedAllocator = .init(self.device.device_allocator.allocator(), 4 * @"1GiB");
+    var bounded_allocator: BoundedAllocator = .init(self.device.interface.device_allocator.allocator(), 4 * @"1GiB");
     const allocator = bounded_allocator.allocator();
 
     const indexed_draw = try self.readIndexBuffer(allocator, index_count, first_index, vertex_offset);
@@ -266,6 +266,9 @@ fn drawCall(self: *Self, bounded_allocator: *BoundedAllocator, vertex_count: usi
         return VkError.Unknown;
     };
 
+    if (pipeline.interface.mode.graphics.rasterization.rasterizer_discard_enable)
+        return;
+
     draw_call.viewport = try self.resolveViewport(0);
     draw_call.scissor = try self.resolveScissor(0);
 

src/software/device/fragment.zig

@@ -67,10 +67,8 @@ pub fn shaderInvocation(
         else => return err,
     };
     if (rt.getBuiltinResult(.FragCoord)) |frag_coord_word| {
-        const frag_x: i32 = @intFromFloat(position[0]);
+        const frag_coord_dx = zm.f32x4(1.0, 0.0, 0.0, 0.0);
-        const frag_y: i32 = @intFromFloat(position[1]);
+        const frag_coord_dy = zm.f32x4(0.0, 1.0, 0.0, 0.0);
-        const frag_coord_dx = zm.f32x4(if (@mod(frag_x, 2) == 0) 1.0 else -1.0, 0.0, 0.0, 0.0);
-        const frag_coord_dy = zm.f32x4(0.0, if (@mod(frag_y, 2) == 0) 1.0 else -1.0, 0.0, 0.0);
         try rt.setDerivativeFromMemory(allocator, frag_coord_word, std.mem.asBytes(&frag_coord_dx), std.mem.asBytes(&frag_coord_dy));
     }
     if (point_coord) |coord| {
@@ -170,7 +168,9 @@ pub fn shaderInvocation(
 
             if (input.size != 0) {
                 const input_bytes = input.blob[0..input.size];
-                if (!has_result_value or input.size < memory_size)
+                if (has_result_value and input.size < memory_size)
+                    try writeAggregateInputLocations(allocator, rt, fragment_inputs, result_word, location, component, memory_size)
+                else if (!has_result_value)
                     try rt.writeInputLocation(input_bytes, @intCast(location))
                 else
                     try rt.writeInput(allocator, input_bytes, result_word);
@@ -262,6 +262,70 @@ pub fn shaderInvocation(
     };
 }
 
+fn writeAggregateInputLocations(
+    allocator: std.mem.Allocator,
+    rt: *spv.Runtime,
+    inputs: [spv.SPIRV_MAX_OUTPUT_LOCATIONS]VertexInterpolationLocation,
+    result_word: spv.SpvWord,
+    location: usize,
+    component: usize,
+    memory_size: usize,
+) SpvRuntimeError!void {
+    const value = try rt.results[result_word].getConstValue();
+    switch (value.*) {
+        .Array, .Matrix => {},
+        else => {
+            const input = inputs[location][component];
+            if (input.size != 0)
+                try rt.writeInputLocation(input.blob[0..input.size], @intCast(location));
+            return;
+        },
+    }
+
+    const bytes = allocator.alloc(u8, memory_size + INTERFACE_BLOB_PADDING) catch return SpvRuntimeError.OutOfMemory;
+    defer allocator.free(bytes);
+    @memset(bytes, 0);
+
+    var source_location = location;
+    var offset: usize = 0;
+    try copyAggregateInputLocations(value, inputs, &source_location, component, bytes[0..memory_size], &offset);
+    try rt.writeInput(allocator, bytes[0..memory_size], result_word);
+}
+
+fn copyAggregateInputLocations(
+    value: anytype,
+    inputs: [spv.SPIRV_MAX_OUTPUT_LOCATIONS]VertexInterpolationLocation,
+    source_location: *usize,
+    component: usize,
+    bytes: []u8,
+    offset: *usize,
+) SpvRuntimeError!void {
+    switch (value.*) {
+        .Array => |array| {
+            for (array.values) |*element|
+                try copyAggregateInputLocations(element, inputs, source_location, component, bytes, offset);
+        },
+        .Matrix => |columns| {
+            for (columns) |*column|
+                try copyAggregateInputLocations(column, inputs, source_location, component, bytes, offset);
+        },
+        else => {
+            if (source_location.* >= inputs.len or component >= 4)
+                return SpvRuntimeError.OutOfBounds;
+            const input = inputs[source_location.*][component];
+            const element_size = try value.getPlainMemorySize();
+            if (offset.* > bytes.len or element_size > bytes.len - offset.*)
+                return SpvRuntimeError.OutOfBounds;
+            if (input.size != 0) {
+                const copy_size = @min(element_size, input.size);
+                @memcpy(bytes[offset.* .. offset.* + copy_size], input.blob[0..copy_size]);
+            }
+            offset.* += element_size;
+            source_location.* += 1;
+        },
+    }
+}
+
 fn readFragmentOutput(
     allocator: std.mem.Allocator,
     rt: anytype,

src/software/device/rasterizer.zig

@@ -107,7 +107,7 @@ pub fn processThenFragmentStage(renderer: *Renderer, allocator: std.mem.Allocato
     const pipeline_data = (renderer.state.pipeline orelse return VkError.InvalidHandleDrv).interface.mode.graphics;
     const topology = pipeline_data.input_assembly.topology;
     if (renderer.input_attachment_snapshots.len == 0) {
-        renderer.input_attachment_snapshots = try snapshotInputAttachments(renderer.device.device_allocator.allocator(), draw_call);
+        renderer.input_attachment_snapshots = try snapshotInputAttachments(renderer.device.interface.device_allocator.allocator(), draw_call);
     }
     draw_call.input_attachment_snapshots = renderer.input_attachment_snapshots;
 

src/software/device/rasterizer/common.zig

@@ -612,7 +612,8 @@ pub fn writeToTargets(
             const color = maybe_color orelse continue;
             const color_offset = targetSampleOffset(color, x, y, sample_index) orelse continue;
             if (base.format.isUnnormalizedInteger(color.format)) {
-                blitter.writeInt4(std.mem.bytesToValue(U32x4, &outputs[location]), color.base[color_offset..], color.format);
+                const value = std.mem.bytesToValue(U32x4, &outputs[location]);
+                blitter.writeInt4(value, color.base[color_offset..], color.format);
             } else {
                 const src = fragmentOutputFloat4(outputs[location], color.format);
                 const encoded_dst = blitter.readFloat4(color.base[color_offset..], color.format);

src/software/device/rasterizer/edge_function.zig

@@ -381,7 +381,7 @@ inline fn run(data: RunData) !void {
             if (early_depth.mask == 0)
                 continue;
 
-            const interpolation_barycentrics = if (data.sample_count > 1) blk: {
+            const interpolation_barycentrics = if (data.sample_count > 1 and data.fragment_uses_centroid) blk: {
                 const sample_pos = firstCoveredSamplePosition(data.sample_count, early_depth.mask);
                 const centroid_p = zm.f32x4(
                     @as(f32, @floatFromInt(x)) + sample_pos.x,

src/software/device/vertex_dispatcher.zig

@@ -115,6 +115,7 @@ inline fn run(data: RunData) !void {
 
         try readPosition(rt, std.mem.asBytes(&output.position));
         try readPointSize(rt, &output.point_size);
+        try readActiveInterfaceOutputs(data, output, rt, entry);
 
         for (0..spv.SPIRV_MAX_OUTPUT_LOCATIONS) |location| {
             const location_result = rt.getResultByLocation(@intCast(location), .output) catch |err| switch (err) {
@@ -122,7 +123,8 @@ inline fn run(data: RunData) !void {
                 else => return err,
             };
 
-            try readVertexOutput(data, output, rt, location, 0, location_result);
+            if (output.outputs[location][0] == null)
+                try readVertexOutput(data, output, rt, location, 0, location_result);
 
             for (1..4) |component| {
                 const component_result = rt.getResultByLocationComponent(@intCast(location), @intCast(component), .output) catch |err| switch (err) {
@@ -133,12 +135,11 @@ inline fn run(data: RunData) !void {
                 if (component_result == location_result)
                     continue;
 
-                try readVertexOutput(data, output, rt, location, component, component_result);
+                if (output.outputs[location][component] == null)
+                    try readVertexOutput(data, output, rt, location, component, component_result);
             }
         }
 
-        try readActiveInterfaceOutputs(data, output, rt, entry);
-
         try rt.flushDescriptorSets(data.allocator);
     }
 }
@@ -170,9 +171,8 @@ fn readActiveInterfaceOutputs(data: RunData, output: *Renderer.Vertex, rt: *spv.
         if (rt.results[type_word].variant) |type_variant| switch (type_variant) {
             .Type => |t| switch (t) {
                 .Structure => |structure| {
-                    const base_location = location orelse continue;
                     for (structure.members_type_word, 0..) |_, member_index| {
-                        const member_location = interfaceMemberLocation(rt, type_word, base_location, @intCast(member_index));
+                        const member_location = explicitInterfaceMemberLocation(rt, type_word, @intCast(member_index)) orelse continue;
                         if (member_location >= spv.SPIRV_MAX_OUTPUT_LOCATIONS)
                             continue;
                         if (output.outputs[member_location][component] != null)
@@ -199,8 +199,16 @@ fn readActiveInterfaceOutputs(data: RunData, output: *Renderer.Vertex, rt: *spv.
 }
 
 fn readVertexOutput(data: RunData, output: *Renderer.Vertex, rt: *spv.Runtime, location: usize, component: usize, result_word: spv.SpvWord) !void {
+    const value = try rt.results[result_word].getConstValue();
     const memory_size = try rt.getResultMemorySize(result_word);
     const interpolation_type = vertexOutputInterpolationType(data, rt, location, component, result_word);
+    switch (value.*) {
+        .Array, .Matrix => {
+            try readVertexOutputAggregate(data, output, value, location, component, interpolation_type);
+            return;
+        },
+        else => {},
+    }
 
     output.outputs[location][component] = .{
         .interpolation_type = interpolation_type,
@@ -211,6 +219,66 @@ fn readVertexOutput(data: RunData, output: *Renderer.Vertex, rt: *spv.Runtime, l
     try rt.readOutput(output.outputs[location][component].?.blob, result_word);
 }
 
+fn readVertexOutputAggregate(
+    data: RunData,
+    output: *Renderer.Vertex,
+    value: anytype,
+    location: usize,
+    component: usize,
+    interpolation_type: Renderer.InterpolationType,
+) !void {
+    var target_location = location;
+    switch (value.*) {
+        .Array => |array| {
+            for (array.values) |*element| {
+                try readVertexOutputAggregate(data, output, element, target_location, component, interpolation_type);
+                target_location += vertexOutputValueLocationCount(element);
+            }
+        },
+        .Matrix => |columns| {
+            for (columns) |*column| {
+                try writeVertexOutputValue(data, output, column, target_location, component, interpolation_type);
+                target_location += 1;
+            }
+        },
+        else => try writeVertexOutputValue(data, output, value, location, component, interpolation_type),
+    }
+}
+
+fn writeVertexOutputValue(
+    data: RunData,
+    output: *Renderer.Vertex,
+    value: anytype,
+    location: usize,
+    component: usize,
+    interpolation_type: Renderer.InterpolationType,
+) !void {
+    if (location >= spv.SPIRV_MAX_OUTPUT_LOCATIONS or component >= 4)
+        return VkError.ValidationFailed;
+
+    const memory_size = try value.getPlainMemorySize();
+    output.outputs[location][component] = .{
+        .interpolation_type = interpolation_type,
+        .blob = data.allocator.alloc(u8, memory_size + INTERFACE_BLOB_PADDING) catch return VkError.OutOfDeviceMemory,
+        .size = memory_size,
+    };
+    @memset(output.outputs[location][component].?.blob, 0);
+    _ = try value.read(output.outputs[location][component].?.blob[0..memory_size]);
+}
+
+fn vertexOutputValueLocationCount(value: anytype) usize {
+    return switch (value.*) {
+        .Array => |array| count: {
+            var location_count: usize = 0;
+            for (array.values) |*element|
+                location_count += vertexOutputValueLocationCount(element);
+            break :count location_count;
+        },
+        .Matrix => |columns| columns.len,
+        else => 1,
+    };
+}
+
 fn vertexOutputInterpolationType(data: RunData, rt: *spv.Runtime, location: usize, component: usize, result_word: spv.SpvWord) Renderer.InterpolationType {
     const result_is_integer = resultIsInteger(rt, result_word);
 
@@ -259,7 +327,7 @@ fn interfaceLocationMemberHasDecoration(rt: *const spv.Runtime, location: usize,
             continue;
 
         const type_word = pointerTargetType(rt, variable.type_word) orelse continue;
-        const base_location = resultLocation(rt, @intCast(id)) orelse continue;
+        const base_location = resultLocation(rt, @intCast(id));
         const type_result = rt.results[type_word];
         const type_variant = type_result.variant orelse continue;
         switch (type_variant) {
@@ -292,14 +360,21 @@ fn resultLocation(rt: *const spv.Runtime, result_word: spv.SpvWord) ?usize {
     return null;
 }
 
-fn interfaceMemberLocation(rt: *const spv.Runtime, type_word: spv.SpvWord, base_location: usize, member_index: spv.SpvWord) usize {
+fn interfaceMemberLocation(rt: *const spv.Runtime, type_word: spv.SpvWord, base_location: ?usize, member_index: spv.SpvWord) ?usize {
+    if (explicitInterfaceMemberLocation(rt, type_word, member_index)) |location|
+        return location;
+
+    return if (base_location) |location| location + @as(usize, @intCast(member_index)) else null;
+}
+
+fn explicitInterfaceMemberLocation(rt: *const spv.Runtime, type_word: spv.SpvWord, member_index: spv.SpvWord) ?usize {
     if (type_word < rt.results.len) {
         for (rt.results[type_word].decorations.items) |decoration| {
             if (decoration.rtype == .Location and decoration.index == member_index)
                 return decoration.literal_1;
         }
     }
-    return base_location + @as(usize, @intCast(member_index));
+    return null;
 }
 
 fn interfaceMemberHasDecoration(rt: *const spv.Runtime, variable_word: spv.SpvWord, member_index: spv.SpvWord, decoration: anytype) bool {

src/vulkan/Device.zig

@@ -1,5 +1,7 @@
 const std = @import("std");
 const vk = @import("vulkan");
+const lib = @import("lib.zig");
+const config = lib.config;
 
 const Dispatchable = @import("Dispatchable.zig").Dispatchable;
 const NonDispatchable = @import("NonDispatchable.zig").NonDispatchable;
@@ -37,10 +39,17 @@ const SwapchainKHR = @import("wsi/SwapchainKHR.zig");
 const Self = @This();
 pub const ObjectType: vk.ObjectType = .device;
 
+const DeviceAllocator = struct {
+    pub inline fn allocator(_: @This()) std.mem.Allocator {
+        return lib.fallback_host_allocator;
+    }
+};
+
 instance: *Instance,
 physical_device: *const PhysicalDevice,
 queues: std.AutoArrayHashMapUnmanaged(u32, std.ArrayList(*Dispatchable(Queue))),
 host_allocator: VulkanAllocator,
+device_allocator: if (config.device_debug_allocator) std.heap.DebugAllocator(.{}) else DeviceAllocator,
 enabled_khr_swapchain: bool,
 enabled_khr_device_group: bool,
 
@@ -110,6 +119,7 @@ pub fn init(allocator: std.mem.Allocator, instance: *Instance, physical_device:
         .host_allocator = VulkanAllocator.from(allocator).cloneWithScope(.object),
         .enabled_khr_swapchain = enabled_khr_swapchain,
         .enabled_khr_device_group = enabled_khr_device_group,
+        .device_allocator = if (config.device_debug_allocator) .init else .{},
         .dispatch_table = undefined,
         .vtable = undefined,
     };
@@ -154,6 +164,17 @@ pub inline fn destroy(self: *Self, allocator: std.mem.Allocator) VkError!void {
         family.deinit(allocator);
     }
     self.queues.deinit(allocator);
+
+    if (config.device_debug_allocator) {
+        // All device memory allocations should've been freed by now
+        const leaks = self.device_allocator.detectLeaks();
+        if (leaks != 0)
+            std.log.scoped(.vkDestroyDevice).err("Device was destroyed leaking {d:.3}KB", .{@as(f32, @floatFromInt(leaks)) / 1000})
+        else
+            std.log.scoped(.vkDestroyDevice).debug("No device memory leaks detected", .{});
+        self.device_allocator.deinitWithoutLeakChecks();
+    }
+
     try self.dispatch_table.destroy(self, allocator);
 }
 

src/vulkan/Pipeline.zig

@@ -45,6 +45,7 @@ mode: union(enum) {
             scissor: ?[]vk.Rect2D,
         },
         rasterization: struct {
+            rasterizer_discard_enable: bool,
             polygon_mode: vk.PolygonMode,
             cull_mode: vk.CullModeFlags,
             front_face: vk.FrontFace,
@@ -194,6 +195,7 @@ pub fn initGraphics(device: *Device, allocator: std.mem.Allocator, cache: ?*Pipe
                     },
                 },
                 .rasterization = .{
+                    .rasterizer_discard_enable = rasterizer_discard_enable,
                     .polygon_mode = if (info.p_rasterization_state) |state| state.polygon_mode else return VkError.ValidationFailed,
                     .cull_mode = if (info.p_rasterization_state) |state| state.cull_mode else return VkError.ValidationFailed,
                     .front_face = if (info.p_rasterization_state) |state| state.front_face else return VkError.ValidationFailed,