fixing multithreaded cts

.gitea/workflows/1.4-CTS.yml

@@ -10,7 +10,7 @@ jobs:
 
     steps:
     - uses: actions/checkout@v4
-    - uses: mlugg/setup-zig@v2
+    - uses: https://codeberg.org/mlugg/setup-zig@v2
     - uses: actions/setup-node@v6
       with:
         node-version: 24
@@ -55,7 +55,7 @@ jobs:
         which deqp-runner && deqp-runner --version || echo "deqp-runner not found"
 
     - name: Run Vulkan CTS
-      run: zig build cts-soft -- --mustpass-list=master -j3
+      run: zig build cts-soft --release=fast -- --mustpass-list=master -j3
       continue-on-error: true
 
     - name: Verify tests

.gitea/workflows/Build.yml

@@ -17,7 +17,7 @@ jobs:
 
     steps:
     - uses: actions/checkout@v4
-    - uses: mlugg/setup-zig@v2
+    - uses: https://codeberg.org/mlugg/setup-zig@v2
     - uses: actions/setup-node@v6
       with:
         node-version: 24

.gitea/workflows/Test.yml

@@ -13,7 +13,7 @@ jobs:
 
     steps:
     - uses: actions/checkout@v4
-    - uses: mlugg/setup-zig@v2
+    - uses: https://codeberg.org/mlugg/setup-zig@v2
     - uses: actions/setup-node@v6
       with:
         node-version: 24
@@ -61,7 +61,7 @@ jobs:
         which deqp-runner && deqp-runner --version || echo "deqp-runner not found"
 
     - name: Run Vulkan CTS
-      run: zig build cts-soft -- -j3
+      run: zig build cts-soft --release=fast -- -j3
       continue-on-error: true
 
     - name: Verify tests

.gitignore

@@ -1,6 +1,7 @@
 .cache/
 .zig-cache/
 zig-out/
+zig-pkg/
 cts/
 cts_report/
 scripts/__pycache__/
@@ -14,3 +15,5 @@ scripts/__pycache__/
 *.html
 *.pyc
 *.spv
+*_table_dump.txt
+vgcore.*

README.md

@@ -10,7 +10,7 @@ It was forged for my own learning and amusement alone. Pray, do not wield it in
 
 ## Purpose
 
-To understand Vulkan — not as a humble API mere mortals call upon, but as a labyrinthine system where one may craft a driver by hand.
+To understand Vulkan - not as a humble API mere mortals call upon, but as a labyrinthine system where one may craft a driver by hand.
 It does not seek to produce a performant or production-worthy driver. \
 *The gods are merciful, but not that merciful.*
 
@@ -22,15 +22,9 @@ zig build
 ```
 
 Then ensure thy Vulkan loader is pointed toward the ICD manifest.
-The precise ritual varies by system — consult the tomes of your operating system, or wander the web's endless mausoleum of documentation.
+The precise ritual varies by system - consult the tomes of your operating system, or wander the web's endless mausoleum of documentation.
 
-Use at your own risk. If thy machine shudders, weeps, or attempts to flee — know that it was warned.
+Use at your own risk. If thy machine shudders, weeps, or attempts to flee - know that it was warned.
-\
-\
-Thou may also conjure forth a tome of compile commands by doing thus:
-```
-zig build cdb
-```
 
 ## Vulkan 1.0 specification
 
@@ -61,12 +55,12 @@ vkCmdClearAttachments                          | ⚙️ wip
 vkCmdClearColorImage                           | ⚙️ WIP
 vkCmdClearDepthStencilImage                    | ⚙️ WIP
 vkCmdCopyBuffer                                | ✅ Implemented
-vkCmdCopyBufferToImage                         | ⚙️ WIP
+vkCmdCopyBufferToImage                         | ✅ Implemented
-vkCmdCopyImage                                 | ⚙️ WIP
+vkCmdCopyImage                                 | ✅ Implemented
-vkCmdCopyImageToBuffer                         | ⚙️ WIP
+vkCmdCopyImageToBuffer                         | ✅ Implemented
 vkCmdCopyQueryPoolResults                      | ⚙️ WIP
 vkCmdDispatch                                  | ✅ Implemented
-vkCmdDispatchIndirect                          | ⚙️ WIP
+vkCmdDispatchIndirect                          | ✅ Implemented
 vkCmdDraw                                      | ⚙️ WIP
 vkCmdDrawIndexed                               | ⚙️ WIP
 vkCmdDrawIndexedIndirect                       | ⚙️ WIP
@@ -109,7 +103,7 @@ vkCreateImage                                  | ✅ Implemented
 vkCreateImageView                              | ✅ Implemented
 vkCreateInstance                               | ✅ Implemented
 vkCreatePipelineCache                          | ⚙️ WIP
-vkCreatePipelineLayout                         | ⚙️ WIP
+vkCreatePipelineLayout                         | ✅ Implemented
 vkCreateQueryPool                              | ⚙️ WIP
 vkCreateRenderPass                             | ⚙️ WIP
 vkCreateSampler                                | ⚙️ WIP
@@ -129,7 +123,7 @@ vkDestroyImageView                             | ✅ Implemented
 vkDestroyInstance                              | ✅ Implemented
 vkDestroyPipeline                              | ✅ Implemented
 vkDestroyPipelineCache                         | ⚙️ WIP
-vkDestroyPipelineLayout                        | ⚙️ WIP
+vkDestroyPipelineLayout                        | ✅ Implemented
 vkDestroyQueryPool                             | ⚙️ WIP
 vkDestroyRenderPass                            | ⚙️ WIP
 vkDestroySampler                               | ⚙️ WIP
@@ -157,8 +151,8 @@ vkGetImageSparseMemoryRequirements             | ⚙️ WIP
 vkGetImageSubresourceLayout                    | ⚙️ WIP
 vkGetInstanceProcAddr                          | ✅ Implemented
 vkGetPhysicalDeviceFeatures                    | ✅ Implemented
-vkGetPhysicalDeviceFormatProperties            | ⚙️ WIP
+vkGetPhysicalDeviceFormatProperties            | ✅ Implemented
-vkGetPhysicalDeviceImageFormatProperties       | ⚙️ WIP
+vkGetPhysicalDeviceImageFormatProperties       | ✅ Implemented
 vkGetPhysicalDeviceMemoryProperties            | ✅ Implemented
 vkGetPhysicalDeviceProperties                  | ✅ Implemented
 vkGetPhysicalDeviceQueueFamilyProperties       | ✅ Implemented
@@ -173,8 +167,8 @@ vkQueueBindSparse                              | ⚙️ WIP
 vkQueueSubmit                                  | ✅ Implemented
 vkQueueWaitIdle                                | ✅ Implemented
 vkResetCommandBuffer                           | ✅ Implemented
-vkResetCommandPool                             | ⚙️ WIP
+vkResetCommandPool                             | ✅ Implemented
-vkResetDescriptorPool                          | ⚙️ WIP
+vkResetDescriptorPool                          | ✅ Implemented
 vkResetEvent                                   | ⚙️ WIP
 vkResetFences                                  | ✅ Implemented
 vkSetEvent                                     | ⚙️ WIP

build.zig

@@ -1,13 +1,12 @@
 const std = @import("std");
 const Step = std.Build.Step;
-const zcc = @import("compile_commands");
 const builtin = @import("builtin");
 
 const ImplementationDesc = struct {
     name: []const u8,
     root_source_file: []const u8,
     vulkan_version: std.SemanticVersion,
-    custom: ?*const fn (*std.Build, *std.Build.Step.Compile) anyerror!void = null,
+    custom: ?*const fn (*std.Build, *std.Build.Step.Compile, *std.Build.Step.Options) anyerror!void = null,
 };
 
 const implementations = [_]ImplementationDesc{
@@ -19,6 +18,12 @@ const implementations = [_]ImplementationDesc{
     },
 };
 
+const RunningMode = enum {
+    normal,
+    gdb,
+    valgrind,
+};
+
 pub fn build(b: *std.Build) !void {
     const target = b.standardTargetOptions(.{});
     const optimize = b.standardOptimizeOption(.{});
@@ -27,11 +32,8 @@ pub fn build(b: *std.Build) !void {
         .root_source_file = b.path("src/vulkan/lib.zig"),
         .target = target,
         .optimize = optimize,
-        .link_libc = true,
     });
 
-    const zdt = b.dependency("zdt", .{}).module("zdt");
-    const zigrc = b.dependency("zigrc", .{}).module("zigrc");
     const vulkan_headers = b.dependency("vulkan_headers", .{});
     const vulkan_utility_libraries = b.dependency("vulkan_utility_libraries", .{});
 
@@ -39,19 +41,21 @@ pub fn build(b: *std.Build) !void {
         .registry = vulkan_headers.path("registry/vk.xml"),
     }).module("vulkan-zig");
 
-    base_mod.addImport("zdt", zdt);
+    const logs_option = b.option(bool, "logs", "Driver logs") orelse false;
-    base_mod.addImport("zigrc", zigrc);
+
+    const options = b.addOptions();
+    options.addOption(bool, "logs", logs_option);
+
     base_mod.addImport("vulkan", vulkan);
     base_mod.addSystemIncludePath(vulkan_headers.path("include"));
     base_mod.addSystemIncludePath(vulkan_utility_libraries.path("include"));
 
-    for (implementations) |impl| {
+    const use_llvm = b.option(bool, "use-llvm", "LLVM build") orelse (b.release_mode != .off);
-        var targets = std.ArrayList(*std.Build.Step.Compile){};
 
+    for (implementations) |impl| {
         const lib_mod = b.createModule(.{
             .root_source_file = b.path(impl.root_source_file),
             .target = target,
-            .link_libc = true,
             .optimize = optimize,
             .imports = &.{
                 .{ .name = "base", .module = base_mod },
@@ -65,11 +69,11 @@ pub fn build(b: *std.Build) !void {
             .name = b.fmt("vulkan_{s}", .{impl.name}),
             .root_module = lib_mod,
             .linkage = .dynamic,
-            .use_llvm = true, // Fixes some random bugs happenning with custom backend. Investigations needed
+            .use_llvm = use_llvm,
         });
 
         if (impl.custom) |custom| {
-            custom(b, lib) catch continue;
+            custom(b, lib, options) catch continue;
         }
 
         const icd_file = b.addWriteFile(
@@ -98,21 +102,29 @@ pub fn build(b: *std.Build) !void {
         const test_step = b.step(b.fmt("test-{s}", .{impl.name}), b.fmt("Run libvulkan_{s} tests", .{impl.name}));
         test_step.dependOn(&run_tests.step);
 
-        const c_test = addCTest(b, target, optimize, vulkan_headers, &impl, lib) catch continue;
+        (try addCTS(b, target, &impl, lib, .normal)).dependOn(&lib_install.step);
-
+        (try addCTS(b, target, &impl, lib, .gdb)).dependOn(&lib_install.step);
-        try targets.append(b.allocator, c_test);
+        (try addCTS(b, target, &impl, lib, .valgrind)).dependOn(&lib_install.step);
-        try targets.append(b.allocator, lib);
-        _ = zcc.createStep(b, "cdb", try targets.toOwnedSlice(b.allocator));
-
-        (try addCTestRunner(b, &impl, c_test, false)).dependOn(&lib_install.step);
-        (try addCTestRunner(b, &impl, c_test, true)).dependOn(&lib_install.step);
-
-        (try addCTS(b, target, &impl, lib, false)).dependOn(&lib_install.step);
-        (try addCTS(b, target, &impl, lib, true)).dependOn(&lib_install.step);
 
         (try addMultithreadedCTS(b, target, &impl, lib)).dependOn(&lib_install.step);
+
+        const impl_autodoc_test = b.addObject(.{
+            .name = "lib",
+            .root_module = lib_mod,
+        });
+
+        const impl_install_docs = b.addInstallDirectory(.{
+            .source_dir = impl_autodoc_test.getEmittedDocs(),
+            .install_dir = .prefix,
+            .install_subdir = b.fmt("docs-{s}", .{impl.name}),
+        });
+
+        const impl_docs_step = b.step(b.fmt("docs-{s}", .{impl.name}), b.fmt("Build and install the documentation for lib_vulkan_{s}", .{impl.name}));
+        impl_docs_step.dependOn(&impl_install_docs.step);
     }
 
+    base_mod.addOptions("config", options);
+
     const autodoc_test = b.addObject(.{
         .name = "lib",
         .root_module = base_mod,
@@ -128,13 +140,10 @@ pub fn build(b: *std.Build) !void {
     docs_step.dependOn(&install_docs.step);
 }
 
-fn customSoft(b: *std.Build, lib: *std.Build.Step.Compile) !void {
+fn customSoft(b: *std.Build, lib: *std.Build.Step.Compile, options: *std.Build.Step.Options) !void {
     const cpuinfo = b.lazyDependency("cpuinfo", .{}) orelse return error.UnresolvedDependency;
-    lib.addSystemIncludePath(cpuinfo.path("include"));
+    lib.root_module.addSystemIncludePath(cpuinfo.path("include"));
-    lib.linkLibrary(cpuinfo.artifact("cpuinfo"));
+    lib.root_module.linkLibrary(cpuinfo.artifact("cpuinfo"));
-
-    const interface = b.lazyDependency("interface", .{}) orelse return error.UnresolvedDependency;
-    lib.root_module.addImport("interface", interface.module("interface"));
 
     const spv = b.dependency("SPIRV_Interpreter", .{
         .@"no-example" = true,
@@ -143,57 +152,20 @@ fn customSoft(b: *std.Build, lib: *std.Build.Step.Compile) !void {
     }).module("spv");
     lib.root_module.addImport("spv", spv);
 
-    const debug_allocator_option = b.option(bool, "debug-allocator", "debug device allocator") 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 options = b.addOptions();
     options.addOption(bool, "debug_allocator", debug_allocator_option);
-    lib.root_module.addOptions("config", options);
+    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);
 }
 
-fn addCTest(b: *std.Build, target: std.Build.ResolvedTarget, optimize: std.builtin.OptimizeMode, vulkan_headers: *std.Build.Dependency, impl: *const ImplementationDesc, impl_lib: *std.Build.Step.Compile) !*std.Build.Step.Compile {
+fn addCTS(b: *std.Build, target: std.Build.ResolvedTarget, impl: *const ImplementationDesc, impl_lib: *std.Build.Step.Compile, comptime mode: RunningMode) !*std.Build.Step {
-    const volk = b.lazyDependency("volk", .{}) orelse return error.DepNotFound;
-    const kvf = b.lazyDependency("kvf", .{}) orelse return error.DepNotFound;
-    const stb = b.lazyDependency("stb", .{}) orelse return error.DepNotFound;
-
-    const exe = b.addExecutable(.{
-        .name = b.fmt("c_test_vulkan_{s}", .{impl.name}),
-        .root_module = b.createModule(.{
-            .target = target,
-            .optimize = optimize,
-            .link_libc = true,
-        }),
-    });
-
-    exe.root_module.addSystemIncludePath(volk.path(""));
-    exe.root_module.addSystemIncludePath(kvf.path(""));
-    exe.root_module.addSystemIncludePath(stb.path(""));
-    exe.root_module.addSystemIncludePath(vulkan_headers.path("include"));
-
-    exe.root_module.addCSourceFile(.{
-        .file = b.path("test/c/main.c"),
-        .flags = &.{b.fmt("-DLIBVK=\"{s}\"", .{impl_lib.name})},
-    });
-
-    const install = b.addInstallArtifact(exe, .{});
-    install.step.dependOn(&impl_lib.step);
-
-    return exe;
-}
-
-fn addCTestRunner(b: *std.Build, impl: *const ImplementationDesc, exe: *std.Build.Step.Compile, comptime gdb: bool) !*std.Build.Step {
-    const run = b.addRunArtifact(exe);
-    if (gdb) {
-        try run.argv.insert(b.allocator, 0, .{ .bytes = b.fmt("gdb", .{}) }); // Hacky
-    }
-    run.step.dependOn(&exe.step);
-
-    const run_step = b.step(b.fmt("test-c-{s}{s}", .{ impl.name, if (gdb) "-gdb" else "" }), b.fmt("Run libvulkan_{s} C test{s}", .{ impl.name, if (gdb) " within GDB" else "" }));
-    run_step.dependOn(&run.step);
-
-    return &run.step;
-}
-
-fn addCTS(b: *std.Build, target: std.Build.ResolvedTarget, impl: *const ImplementationDesc, impl_lib: *std.Build.Step.Compile, comptime gdb: bool) !*std.Build.Step {
     const cts = b.dependency("cts_bin", .{});
 
     const cts_exe_name = cts.path(b.fmt("deqp-vk-{s}", .{
@@ -212,12 +184,23 @@ fn addCTS(b: *std.Build, target: std.Build.ResolvedTarget, impl: *const Implemen
 
     const cts_exe_path = try cts_exe_name.getPath3(b, null).toString(b.allocator);
 
-    const run = b.addSystemCommand(&[_][]const u8{if (gdb) "gdb" else cts_exe_path});
+    const run = b.addSystemCommand(&[_][]const u8{switch (mode) {
+        .normal => cts_exe_path,
+        .gdb => "gdb",
+        .valgrind => "valgrind",
+    }});
     run.step.dependOn(&impl_lib.step);
 
-    if (gdb) {
+    switch (mode) {
-        run.addArg("--args");
+        .gdb => {
-        run.addArg(cts_exe_path);
+            run.addArg("--args");
+            run.addArg(cts_exe_path);
+        },
+        .valgrind => {
+            run.addArg("--track-origins=yes");
+            run.addArg(cts_exe_path);
+        },
+        else => {},
     }
 
     run.addArg(b.fmt("--deqp-archive-dir={s}", .{try cts.path("").getPath3(b, null).toString(b.allocator)}));
@@ -237,7 +220,24 @@ fn addCTS(b: *std.Build, target: std.Build.ResolvedTarget, impl: *const Implemen
         run.addArg(b.fmt("--deqp-caselist-file={s}", .{mustpass}));
     }
 
-    const run_step = b.step(b.fmt("raw-cts-{s}{s}", .{ impl.name, if (gdb) "-gdb" else "" }), b.fmt("Run Vulkan conformance tests for libvulkan_{s}{s}", .{ impl.name, if (gdb) " within GDB" else "" }));
+    const run_step = b.step(
+        b.fmt("raw-cts-{s}{s}", .{
+            impl.name,
+            switch (mode) {
+                .normal => "",
+                .gdb => "-gdb",
+                .valgrind => "-valgrind",
+            },
+        }),
+        b.fmt("Run Vulkan conformance tests for libvulkan_{s}{s}", .{
+            impl.name,
+            switch (mode) {
+                .normal => "",
+                .gdb => " within GDB",
+                .valgrind => " within Valgrind",
+            },
+        }),
+    );
     run_step.dependOn(&run.step);
 
     return &run.step;
@@ -246,18 +246,6 @@ fn addCTS(b: *std.Build, target: std.Build.ResolvedTarget, impl: *const Implemen
 fn addMultithreadedCTS(b: *std.Build, target: std.Build.ResolvedTarget, impl: *const ImplementationDesc, impl_lib: *std.Build.Step.Compile) !*std.Build.Step {
     const cts = b.dependency("cts_bin", .{});
 
-    // Some systems may need a manual path management to get to packages (e.g. Github Actions)
-    const cache_path = blk: {
-        if (std.process.getEnvVarOwned(b.allocator, "ZIG_GLOBAL_CACHE_DIR")) |cache_path| {
-            break :blk b.fmt("{s}/../", .{cache_path});
-        } else |err| switch (err) {
-            error.EnvironmentVariableNotFound => {
-                break :blk "";
-            },
-            else => unreachable,
-        }
-    };
-
     const cts_exe_name = cts.path(b.fmt("deqp-vk-{s}", .{
         switch (if (target.query.os_tag) |tag| tag else builtin.target.os.tag) {
             .linux => "linux.x86_64",
@@ -294,19 +282,21 @@ fn addMultithreadedCTS(b: *std.Build, target: std.Build.ResolvedTarget, impl: *c
 
     run.addArg("run");
     run.addArg("--deqp");
-    run.addArg(b.fmt("{s}{s}", .{ cache_path, cts_exe_path }));
+    run.addArg(cts_exe_path);
     run.addArg("--caselist");
-    run.addArg(b.fmt("{s}{s}", .{ cache_path, mustpass_path }));
+    run.addArg(mustpass_path);
     run.addArg("--output");
     run.addArg("./cts");
+    run.addArg("--timeout");
+    run.addArg("300");
     if (jobs_count) |count| {
         run.addArg(b.fmt("-j{d}", .{count}));
     }
     run.addArg("--");
-    run.addArg(b.fmt("--deqp-archive-dir={s}{s}", .{ cache_path, try cts.path("").getPath3(b, null).toString(b.allocator) }));
+    run.addArg(b.fmt("--deqp-archive-dir={s}", .{try cts.path("").getPath3(b, null).toString(b.allocator)}));
     run.addArg(b.fmt("--deqp-vk-library-path={s}", .{b.getInstallPath(.lib, impl_lib.out_lib_filename)}));
 
-    const run_step = b.step(b.fmt("cts-{s}", .{impl.name}), b.fmt("Run Vulkan conformance tests in a multithreaded environment for libvulkan_{s}", .{impl.name}));
+    const run_step = b.step(b.fmt("cts-{s}", .{impl.name}), b.fmt("Run Vulkan conformance tests for libvulkan_{s} in a multithreaded environment", .{impl.name}));
     run_step.dependOn(&run.step);
 
     return &run.step;

build.zig.zon

@@ -2,65 +2,33 @@
     .name = .VulkanDriver,
     .version = "0.0.1",
     .fingerprint = 0x52cb73649f1107de,
-    .minimum_zig_version = "0.15.2",
+    .minimum_zig_version = "0.16.0",
 
     .dependencies = .{
-        .compile_commands = .{
-            .url = "git+https://github.com/the-argus/zig-compile-commands#f74e2d13e43fafab3a71e19557a0e1cfbf0f2e1b",
-            .hash = "zig_compile_commands-0.0.1-OZg5-a3CAACM-h32Kjb1obTMqrKGs9YoDhorVZ8-LGle",
-        },
         .vulkan_headers = .{
             .url = "git+https://github.com/KhronosGroup/Vulkan-Headers?ref=v1.4.330#ee3b5caaa7e372715873c7b9c390ee1c3ca5db25",
             .hash = "N-V-__8AAFXYAQKsK51AAGXB9HziPDFjS_DVUq6_QHVxHrBM",
         },
         .vulkan_zig = .{
-            .url = "git+https://github.com/catmeow72/vulkan-zig/#8961518db28f88d2cf09ea68e146923de2cfa7f0",
+            .url = "git+https://github.com/Snektron/vulkan-zig#be16fc99f7f794cba90334978c6c5aadeda8094f",
-            .hash = "vulkan-0.0.0-r7Ytx6hBAwD8X_TN32qlkzul4riK6vFvjtK9fZfRvALg",
+            .hash = "vulkan-0.0.0-r7Ytx-V9AwCwHxS2TaZ-KJCp_3axidu3ASZUg3B512-k",
         },
         .vulkan_utility_libraries = .{
             .url = "git+https://github.com/KhronosGroup/Vulkan-Utility-Libraries.git#ba452bad58bb4d4c64d7fbd872bf69f70141510e",
             .hash = "N-V-__8AAE42fwC1FFw26LNZ8AaSuGMdgG4vfYkfV_227sET",
         },
-        .zdt = .{
-            .url = "git+https://github.com/FObersteiner/zdt/?ref=v0.8.1#8b551a0a3e5ae64a32b5bad0e6a93119787b43af",
-            .hash = "zdt-0.8.1-xr0_vAxUDwCJRDh9pcAS_mdZBIsvcGTtN-K8JJSWY4I6",
-        },
         .cts_bin = .{
             .url = "git+https://git.kbz8.me/kbz_8/Vulkan-CTS-bin#19ce2da05f8176348064a9fc6688847e5f76a46e",
             .hash = "N-V-__8AAHDV0xtS93nAGaYd7YWxBLnvHDEplwIpC29izSGa",
         },
-        .zigrc = .{
-            .url = "https://github.com/Aandreba/zigrc/archive/refs/tags/1.1.0.tar.gz",
-            .hash = "zigrc-1.0.0-lENlWzvQAACulrbkL9PVhWjFsWSkYhi7AmfSbCM-2Xlh",
-        },
         .cpuinfo = .{
-            .url = "git+https://github.com/Kbz-8/cpuinfo#4883954cfcec3f6c9ca9c4aaddfc26107e08726f",
+            .url = "git+https://github.com/Kbz-8/cpuinfo.git#c9bea4f6c166a495ee0ce117821f9627d4aed118",
-            .hash = "cpuinfo-0.0.1-RLgIQTLRMgF4dLo8AJ-HvnpFsJe6jmXCJjMWWjil6RF1",
+            .hash = "cpuinfo-0.0.1-RLgIQYrTMgGqfQMOd1nAa2EuglXOh5gR9bNzwMzQTemt",
-            .lazy = true,
-        },
-        .volk = .{
-            .url = "git+https://github.com/zeux/volk/#8f53cc717f50f142db4736f401d0b61956cd78f9",
-            .hash = "N-V-__8AAPn9BwCBHnaxOC_rffCpFI7QRfi5qBCLvov9EYK3",
-            .lazy = true,
-        },
-        .kvf = .{
-            .url = "git+https://git.kbz8.me/kbz_8/KVF#98b845f876bea94f7bf1b9d30588cf617bf93452",
-            .hash = "N-V-__8AAEGKAgC2cGDnxmAIFKkaICxS_ogfVYWH83Re29zN",
-            .lazy = true,
-        },
-        .stb = .{
-            .url = "git+https://github.com/nothings/stb#f1c79c02822848a9bed4315b12c8c8f3761e1296",
-            .hash = "N-V-__8AABQ7TgCnPlp8MP4YA8znrjd6E-ZjpF1rvrS8J_2I",
-            .lazy = true,
-        },
-        .interface = .{
-            .url = "git+https://github.com/nilslice/zig-interface#8c0fe8fa9fd0702eee43f50cb75dce1cc5a7e1f4",
-            .hash = "interface-0.0.2-GFlWJ1mcAQARS-V4xJ7qDt5_cutxOHSEz6H9yiK-Sw0A",
             .lazy = true,
         },
         .SPIRV_Interpreter = .{
-            .url = "git+https://git.kbz8.me/kbz_8/SPIRV-Interpreter#e8a08d78851dd07e72de7b40821cd2af10f38866",
+            .url = "git+https://git.kbz8.me/kbz_8/SPIRV-Interpreter#664ea9b92bf84bc97ec4a062c171562bf6628263",
-            .hash = "SPIRV_Interpreter-0.0.1-ajmpn9DJAwAaQ_vqdY_c-GX__VX9YQZ5viIBxsgPUlBk",
+            .hash = "SPIRV_Interpreter-0.0.1-ajmpn1qKBACshq_ncUUF-zXJzpdNLRzIAPcWRQL57W8l",
         },
     },
 

src/soft/SoftCommandBuffer.zig

@@ -18,9 +18,14 @@ const ExecutionDevice = @import("device/Device.zig");
 const Self = @This();
 pub const Interface = base.CommandBuffer;
 
-const Command = InterfaceFactory(.{
+const Command = struct {
-    .execute = fn (*ExecutionDevice) VkError!void,
+    const VTable = struct {
-}, null);
+        execute: *const fn (*anyopaque, *ExecutionDevice) VkError!void,
+    };
+
+    ptr: *anyopaque,
+    vtable: *const VTable,
+};
 
 interface: Interface,
 
@@ -41,6 +46,7 @@ pub fn create(device: *base.Device, allocator: std.mem.Allocator, info: *const v
         .begin = begin,
         .bindDescriptorSets = bindDescriptorSets,
         .bindPipeline = bindPipeline,
+        .blitImage = blitImage,
         .clearColorImage = clearColorImage,
         .copyBuffer = copyBuffer,
         .copyBufferToImage = copyBufferToImage,
@@ -49,6 +55,7 @@ pub fn create(device: *base.Device, allocator: std.mem.Allocator, info: *const v
         .dispatch = dispatch,
         .dispatchIndirect = dispatchIndirect,
         .end = end,
+        .executeCommands = executeCommands,
         .fillBuffer = fillBuffer,
         .pipelineBarrier = pipelineBarrier,
         .reset = reset,
@@ -69,6 +76,7 @@ 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));
     allocator.destroy(self);
+    _ = self.command_allocator.reset(.free_all);
 }
 
 pub fn execute(self: *Self, device: *ExecutionDevice) void {
@@ -76,10 +84,10 @@ pub fn execute(self: *Self, device: *ExecutionDevice) void {
     defer self.interface.finish() catch {};
 
     for (self.commands.items) |command| {
-        command.vtable.execute(command.ptr, device) catch |err| {
+        command.vtable.execute(@ptrCast(command.ptr), device) catch |err| {
             base.errors.errorLoggerContext(err, "the software execution device");
             if (@errorReturnTrace()) |trace| {
-                std.debug.dumpStackTrace(trace.*);
+                std.debug.dumpErrorReturnTrace(trace);
             }
             return; // Should we return or continue ? Maybe device lost ?
         };
@@ -98,8 +106,7 @@ pub fn end(interface: *Interface) VkError!void {
 
 pub fn reset(interface: *Interface, _: vk.CommandBufferResetFlags) VkError!void {
     const self: *Self = @alignCast(@fieldParentPtr("interface", interface));
-    const allocator = self.command_allocator.allocator();
+    self.commands.clearAndFree(self.command_allocator.allocator());
-    self.commands.clearAndFree(allocator);
     _ = self.command_allocator.reset(.free_all);
 }
 
@@ -117,7 +124,8 @@ pub fn bindDescriptorSets(interface: *Interface, bind_point: vk.PipelineBindPoin
         sets: [base.VULKAN_MAX_DESCRIPTOR_SETS]?*base.DescriptorSet,
         dynamic_offsets: []const u32,
 
-        pub fn execute(impl: *const Impl, device: *ExecutionDevice) VkError!void {
+        pub fn execute(context: *anyopaque, device: *ExecutionDevice) VkError!void {
+            const impl: *Impl = @ptrCast(@alignCast(context));
             for (impl.first_set.., impl.sets[0..]) |i, set| {
                 if (set == null)
                     break;
@@ -134,7 +142,7 @@ pub fn bindDescriptorSets(interface: *Interface, bind_point: vk.PipelineBindPoin
         .sets = sets,
         .dynamic_offsets = dynamic_offsets,
     };
-    self.commands.append(allocator, Command.from(cmd)) catch return VkError.OutOfHostMemory;
+    self.commands.append(allocator, .{ .ptr = cmd, .vtable = &.{ .execute = CommandImpl.execute } }) catch return VkError.OutOfHostMemory;
 }
 
 pub fn bindPipeline(interface: *Interface, bind_point: vk.PipelineBindPoint, pipeline: *base.Pipeline) VkError!void {
@@ -147,7 +155,8 @@ pub fn bindPipeline(interface: *Interface, bind_point: vk.PipelineBindPoint, pip
         bind_point: vk.PipelineBindPoint,
         pipeline: *SoftPipeline,
 
-        pub fn execute(impl: *const Impl, device: *ExecutionDevice) VkError!void {
+        pub fn execute(context: *anyopaque, device: *ExecutionDevice) VkError!void {
+            const impl: *Impl = @ptrCast(@alignCast(context));
             device.pipeline_states[@intCast(@intFromEnum(impl.bind_point))].pipeline = impl.pipeline;
         }
     };
@@ -158,10 +167,41 @@ pub fn bindPipeline(interface: *Interface, bind_point: vk.PipelineBindPoint, pip
         .bind_point = bind_point,
         .pipeline = @alignCast(@fieldParentPtr("interface", pipeline)),
     };
-    self.commands.append(allocator, Command.from(cmd)) catch return VkError.OutOfHostMemory;
+    self.commands.append(allocator, .{ .ptr = cmd, .vtable = &.{ .execute = CommandImpl.execute } }) catch return VkError.OutOfHostMemory;
 }
 
-pub fn clearColorImage(interface: *Interface, image: *base.Image, layout: vk.ImageLayout, color: *const vk.ClearColorValue, range: vk.ImageSubresourceRange) VkError!void {
+pub fn blitImage(interface: *Interface, src: *base.Image, _: vk.ImageLayout, dst: *base.Image, _: vk.ImageLayout, regions: []const vk.ImageBlit, filter: vk.Filter) VkError!void {
+    const self: *Self = @alignCast(@fieldParentPtr("interface", interface));
+    const allocator = self.command_allocator.allocator();
+
+    const CommandImpl = struct {
+        const Impl = @This();
+
+        src: *const SoftImage,
+        dst: *SoftImage,
+        regions: []const vk.ImageBlit,
+        filter: vk.Filter,
+
+        pub fn execute(context: *anyopaque, device: *ExecutionDevice) VkError!void {
+            const impl: *Impl = @ptrCast(@alignCast(context));
+            for (impl.regions[0..]) |region| {
+                try device.blitter.blitRegion(impl.src, impl.dst, region, impl.filter);
+            }
+        }
+    };
+
+    const cmd = allocator.create(CommandImpl) catch return VkError.OutOfHostMemory;
+    errdefer allocator.destroy(cmd);
+    cmd.* = .{
+        .src = @alignCast(@fieldParentPtr("interface", src)),
+        .dst = @alignCast(@fieldParentPtr("interface", dst)),
+        .regions = allocator.dupe(vk.ImageBlit, regions) catch return VkError.OutOfHostMemory, // Will be freed on cmdbuf reset or destroy
+        .filter = filter,
+    };
+    self.commands.append(allocator, .{ .ptr = cmd, .vtable = &.{ .execute = CommandImpl.execute } }) catch return VkError.OutOfHostMemory;
+}
+
+pub fn clearColorImage(interface: *Interface, image: *base.Image, _: vk.ImageLayout, color: *const vk.ClearColorValue, range: vk.ImageSubresourceRange) VkError!void {
     const self: *Self = @alignCast(@fieldParentPtr("interface", interface));
     const allocator = self.command_allocator.allocator();
 
@@ -169,12 +209,13 @@ pub fn clearColorImage(interface: *Interface, image: *base.Image, layout: vk.Ima
         const Impl = @This();
 
         image: *SoftImage,
-        layout: vk.ImageLayout,
         clear_color: vk.ClearColorValue,
         range: vk.ImageSubresourceRange,
 
-        pub fn execute(impl: *const Impl, _: *ExecutionDevice) VkError!void {
+        pub fn execute(context: *anyopaque, device: *ExecutionDevice) VkError!void {
-            impl.image.clearRange(impl.clear_color, impl.range);
+            const impl: *Impl = @ptrCast(@alignCast(context));
+            const clear_format = try impl.image.getClearFormat();
+            try device.blitter.clear(.{ .color = impl.clear_color }, clear_format, impl.image, impl.image.interface.format, impl.range, null);
         }
     };
 
@@ -182,11 +223,10 @@ pub fn clearColorImage(interface: *Interface, image: *base.Image, layout: vk.Ima
     errdefer allocator.destroy(cmd);
     cmd.* = .{
         .image = @alignCast(@fieldParentPtr("interface", image)),
-        .layout = layout,
         .clear_color = color.*,
         .range = range,
     };
-    self.commands.append(allocator, Command.from(cmd)) catch return VkError.OutOfHostMemory;
+    self.commands.append(allocator, .{ .ptr = cmd, .vtable = &.{ .execute = CommandImpl.execute } }) catch return VkError.OutOfHostMemory;
 }
 
 pub fn copyBuffer(interface: *Interface, src: *base.Buffer, dst: *base.Buffer, regions: []const vk.BufferCopy) VkError!void {
@@ -200,7 +240,8 @@ pub fn copyBuffer(interface: *Interface, src: *base.Buffer, dst: *base.Buffer, r
         dst: *SoftBuffer,
         regions: []const vk.BufferCopy,
 
-        pub fn execute(impl: *const Impl, _: *ExecutionDevice) VkError!void {
+        pub fn execute(context: *anyopaque, _: *ExecutionDevice) VkError!void {
+            const impl: *Impl = @ptrCast(@alignCast(context));
             try impl.src.copyBuffer(impl.dst, impl.regions);
         }
     };
@@ -212,7 +253,7 @@ pub fn copyBuffer(interface: *Interface, src: *base.Buffer, dst: *base.Buffer, r
         .dst = @alignCast(@fieldParentPtr("interface", dst)),
         .regions = allocator.dupe(vk.BufferCopy, regions) catch return VkError.OutOfHostMemory, // Will be freed on cmdbuf reset or destroy
     };
-    self.commands.append(allocator, Command.from(cmd)) catch return VkError.OutOfHostMemory;
+    self.commands.append(allocator, .{ .ptr = cmd, .vtable = &.{ .execute = CommandImpl.execute } }) catch return VkError.OutOfHostMemory;
 }
 
 pub fn copyBufferToImage(interface: *Interface, src: *base.Buffer, dst: *base.Image, dst_layout: vk.ImageLayout, regions: []const vk.BufferImageCopy) VkError!void {
@@ -227,7 +268,8 @@ pub fn copyBufferToImage(interface: *Interface, src: *base.Buffer, dst: *base.Im
         dst_layout: vk.ImageLayout,
         regions: []const vk.BufferImageCopy,
 
-        pub fn execute(impl: *const Impl, _: *ExecutionDevice) VkError!void {
+        pub fn execute(context: *anyopaque, _: *ExecutionDevice) VkError!void {
+            const impl: *Impl = @ptrCast(@alignCast(context));
             for (impl.regions[0..]) |region| {
                 try impl.dst.copyFromBuffer(impl.src, region);
             }
@@ -242,10 +284,10 @@ pub fn copyBufferToImage(interface: *Interface, src: *base.Buffer, dst: *base.Im
         .dst = @alignCast(@fieldParentPtr("interface", dst)),
         .regions = allocator.dupe(vk.BufferImageCopy, regions) catch return VkError.OutOfHostMemory, // Will be freed on cmdbuf reset or destroy
     };
-    self.commands.append(allocator, Command.from(cmd)) catch return VkError.OutOfHostMemory;
+    self.commands.append(allocator, .{ .ptr = cmd, .vtable = &.{ .execute = CommandImpl.execute } }) catch return VkError.OutOfHostMemory;
 }
 
-pub fn copyImage(interface: *Interface, src: *base.Image, src_layout: vk.ImageLayout, dst: *base.Image, dst_layout: vk.ImageLayout, regions: []const vk.ImageCopy) VkError!void {
+pub fn copyImage(interface: *Interface, src: *base.Image, _: vk.ImageLayout, dst: *base.Image, _: vk.ImageLayout, regions: []const vk.ImageCopy) VkError!void {
     const self: *Self = @alignCast(@fieldParentPtr("interface", interface));
     const allocator = self.command_allocator.allocator();
 
@@ -253,13 +295,14 @@ pub fn copyImage(interface: *Interface, src: *base.Image, src_layout: vk.ImageLa
         const Impl = @This();
 
         src: *const SoftImage,
-        src_layout: vk.ImageLayout,
         dst: *SoftImage,
-        dst_layout: vk.ImageLayout,
         regions: []const vk.ImageCopy,
 
-        pub fn execute(impl: *const Impl, _: *ExecutionDevice) VkError!void {
+        pub fn execute(context: *anyopaque, _: *ExecutionDevice) VkError!void {
-            try impl.src.copyImage(impl.src_layout, impl.dst, impl.dst_layout, impl.regions);
+            const impl: *Impl = @ptrCast(@alignCast(context));
+            for (impl.regions[0..]) |region| {
+                try impl.src.copyToImage(impl.dst, region);
+            }
         }
     };
 
@@ -267,12 +310,10 @@ pub fn copyImage(interface: *Interface, src: *base.Image, src_layout: vk.ImageLa
     errdefer allocator.destroy(cmd);
     cmd.* = .{
         .src = @alignCast(@fieldParentPtr("interface", src)),
-        .src_layout = src_layout,
         .dst = @alignCast(@fieldParentPtr("interface", dst)),
-        .dst_layout = dst_layout,
         .regions = allocator.dupe(vk.ImageCopy, regions) catch return VkError.OutOfHostMemory, // Will be freed on cmdbuf reset or destroy
     };
-    self.commands.append(allocator, Command.from(cmd)) catch return VkError.OutOfHostMemory;
+    self.commands.append(allocator, .{ .ptr = cmd, .vtable = &.{ .execute = CommandImpl.execute } }) catch return VkError.OutOfHostMemory;
 }
 
 pub fn copyImageToBuffer(interface: *Interface, src: *base.Image, src_layout: vk.ImageLayout, dst: *base.Buffer, regions: []const vk.BufferImageCopy) VkError!void {
@@ -287,7 +328,8 @@ pub fn copyImageToBuffer(interface: *Interface, src: *base.Image, src_layout: vk
         dst: *SoftBuffer,
         regions: []const vk.BufferImageCopy,
 
-        pub fn execute(impl: *const Impl, _: *ExecutionDevice) VkError!void {
+        pub fn execute(context: *anyopaque, _: *ExecutionDevice) VkError!void {
+            const impl: *Impl = @ptrCast(@alignCast(context));
             for (impl.regions[0..]) |region| {
                 try impl.src.copyToBuffer(impl.dst, region);
             }
@@ -302,7 +344,7 @@ pub fn copyImageToBuffer(interface: *Interface, src: *base.Image, src_layout: vk
         .dst = @alignCast(@fieldParentPtr("interface", dst)),
         .regions = allocator.dupe(vk.BufferImageCopy, regions) catch return VkError.OutOfHostMemory, // Will be freed on cmdbuf reset or destroy
     };
-    self.commands.append(allocator, Command.from(cmd)) catch return VkError.OutOfHostMemory;
+    self.commands.append(allocator, .{ .ptr = cmd, .vtable = &.{ .execute = CommandImpl.execute } }) catch return VkError.OutOfHostMemory;
 }
 
 pub fn dispatch(interface: *Interface, group_count_x: u32, group_count_y: u32, group_count_z: u32) VkError!void {
@@ -316,7 +358,8 @@ pub fn dispatch(interface: *Interface, group_count_x: u32, group_count_y: u32, g
         group_count_y: u32,
         group_count_z: u32,
 
-        pub fn execute(impl: *const Impl, device: *ExecutionDevice) VkError!void {
+        pub fn execute(context: *anyopaque, device: *ExecutionDevice) VkError!void {
+            const impl: *Impl = @ptrCast(@alignCast(context));
             try device.compute_routines.dispatch(impl.group_count_x, impl.group_count_y, impl.group_count_z);
         }
     };
@@ -328,7 +371,7 @@ pub fn dispatch(interface: *Interface, group_count_x: u32, group_count_y: u32, g
         .group_count_y = group_count_y,
         .group_count_z = group_count_z,
     };
-    self.commands.append(allocator, Command.from(cmd)) catch return VkError.OutOfHostMemory;
+    self.commands.append(allocator, .{ .ptr = cmd, .vtable = &.{ .execute = CommandImpl.execute } }) catch return VkError.OutOfHostMemory;
 }
 
 pub fn dispatchIndirect(interface: *Interface, buffer: *base.Buffer, offset: vk.DeviceSize) VkError!void {
@@ -341,13 +384,11 @@ pub fn dispatchIndirect(interface: *Interface, buffer: *base.Buffer, offset: vk.
         buffer: *SoftBuffer,
         offset: vk.DeviceSize,
 
-        pub fn execute(impl: *const Impl, device: *ExecutionDevice) VkError!void {
+        pub fn execute(context: *anyopaque, device: *ExecutionDevice) VkError!void {
+            const impl: *Impl = @ptrCast(@alignCast(context));
             const size = 3 * @sizeOf(u32);
             const memory = if (impl.buffer.interface.memory) |memory| memory else return VkError.InvalidDeviceMemoryDrv;
             const map: []u32 = @as([*]u32, @ptrCast(@alignCast(try memory.map(impl.offset, size))))[0..3];
-
-            std.debug.print("{any}\n", .{map});
-
             try device.compute_routines.dispatch(map[0], map[1], map[2]);
         }
     };
@@ -358,7 +399,30 @@ pub fn dispatchIndirect(interface: *Interface, buffer: *base.Buffer, offset: vk.
         .buffer = @alignCast(@fieldParentPtr("interface", buffer)),
         .offset = offset,
     };
-    self.commands.append(allocator, Command.from(cmd)) catch return VkError.OutOfHostMemory;
+    self.commands.append(allocator, .{ .ptr = cmd, .vtable = &.{ .execute = CommandImpl.execute } }) catch return VkError.OutOfHostMemory;
+}
+
+pub fn executeCommands(interface: *Interface, commands: *Interface) VkError!void {
+    const self: *Self = @alignCast(@fieldParentPtr("interface", interface));
+    const allocator = self.command_allocator.allocator();
+
+    const CommandImpl = struct {
+        const Impl = @This();
+
+        cmd: *Self,
+
+        pub fn execute(context: *anyopaque, device: *ExecutionDevice) VkError!void {
+            const impl: *Impl = @ptrCast(@alignCast(context));
+            impl.cmd.execute(device);
+        }
+    };
+
+    const cmd = allocator.create(CommandImpl) catch return VkError.OutOfHostMemory;
+    errdefer allocator.destroy(cmd);
+    cmd.* = .{
+        .cmd = @alignCast(@fieldParentPtr("interface", commands)),
+    };
+    self.commands.append(allocator, .{ .ptr = cmd, .vtable = &.{ .execute = CommandImpl.execute } }) catch return VkError.OutOfHostMemory;
 }
 
 pub fn fillBuffer(interface: *Interface, buffer: *base.Buffer, offset: vk.DeviceSize, size: vk.DeviceSize, data: u32) VkError!void {
@@ -373,7 +437,8 @@ pub fn fillBuffer(interface: *Interface, buffer: *base.Buffer, offset: vk.Device
         size: vk.DeviceSize,
         data: u32,
 
-        pub fn execute(impl: *const Impl, _: *ExecutionDevice) VkError!void {
+        pub fn execute(context: *anyopaque, _: *ExecutionDevice) VkError!void {
+            const impl: *Impl = @ptrCast(@alignCast(context));
             try impl.buffer.fillBuffer(impl.offset, impl.size, impl.data);
         }
     };
@@ -386,7 +451,7 @@ pub fn fillBuffer(interface: *Interface, buffer: *base.Buffer, offset: vk.Device
         .size = size,
         .data = data,
     };
-    self.commands.append(allocator, Command.from(cmd)) catch return VkError.OutOfHostMemory;
+    self.commands.append(allocator, .{ .ptr = cmd, .vtable = &.{ .execute = CommandImpl.execute } }) catch return VkError.OutOfHostMemory;
 }
 
 pub fn pipelineBarrier(interface: *Interface, src_stage: vk.PipelineStageFlags, dst_stage: vk.PipelineStageFlags, dependency: vk.DependencyFlags, memory_barriers: []const vk.MemoryBarrier, buffer_barriers: []const vk.BufferMemoryBarrier, image_barriers: []const vk.ImageMemoryBarrier) VkError!void {
@@ -396,15 +461,15 @@ pub fn pipelineBarrier(interface: *Interface, src_stage: vk.PipelineStageFlags,
     const CommandImpl = struct {
         const Impl = @This();
 
-        pub fn execute(_: *const Impl, _: *ExecutionDevice) VkError!void {
+        pub fn execute(_: *anyopaque, _: *ExecutionDevice) VkError!void {
-            // TODO: implement synchronization for rasterizations stages
+            // TODO: implement synchronization for rasterization stages
         }
     };
 
     const cmd = allocator.create(CommandImpl) catch return VkError.OutOfHostMemory;
     errdefer allocator.destroy(cmd);
     cmd.* = .{};
-    self.commands.append(allocator, Command.from(cmd)) catch return VkError.OutOfHostMemory;
+    self.commands.append(allocator, .{ .ptr = cmd, .vtable = &.{ .execute = CommandImpl.execute } }) catch return VkError.OutOfHostMemory;
 
     _ = src_stage;
     _ = dst_stage;

src/soft/SoftDescriptorPool.zig

@@ -14,6 +14,8 @@ pub const Interface = base.DescriptorPool;
 
 interface: Interface,
 
+list: std.ArrayList(*SoftDescriptorSet),
+
 pub fn create(device: *base.Device, allocator: std.mem.Allocator, info: *const vk.DescriptorPoolCreateInfo) VkError!*Self {
     const self = allocator.create(Self) catch return VkError.OutOfHostMemory;
     errdefer allocator.destroy(self);
@@ -24,27 +26,48 @@ pub fn create(device: *base.Device, allocator: std.mem.Allocator, info: *const v
         .allocateDescriptorSet = allocateDescriptorSet,
         .destroy = destroy,
         .freeDescriptorSet = freeDescriptorSet,
+        .reset = reset,
     };
 
     self.* = .{
         .interface = interface,
+        .list = std.ArrayList(*SoftDescriptorSet).initCapacity(allocator, info.max_sets) catch return VkError.OutOfHostMemory,
     };
     return self;
 }
 
 pub fn allocateDescriptorSet(interface: *Interface, layout: *base.DescriptorSetLayout) VkError!*base.DescriptorSet {
+    const self: *Self = @alignCast(@fieldParentPtr("interface", interface));
     const allocator = VulkanAllocator.init(null, .object).allocator();
     const set = try SoftDescriptorSet.create(interface.owner, allocator, layout);
+    self.list.appendAssumeCapacity(set);
     return &set.interface;
 }
 
 pub fn destroy(interface: *Interface, allocator: std.mem.Allocator) void {
     const self: *Self = @alignCast(@fieldParentPtr("interface", interface));
+    self.list.deinit(allocator);
     allocator.destroy(self);
 }
 
 pub fn freeDescriptorSet(interface: *Interface, set: *base.DescriptorSet) VkError!void {
-    _ = interface;
+    const self: *Self = @alignCast(@fieldParentPtr("interface", interface));
+    const soft_set: *SoftDescriptorSet = @alignCast(@fieldParentPtr("interface", set));
+
+    if (std.mem.indexOfScalar(*SoftDescriptorSet, self.list.items, soft_set)) |pos| {
+        _ = self.list.orderedRemove(pos);
+    }
+
     const allocator = VulkanAllocator.init(null, .object).allocator();
-    allocator.destroy(set);
+    allocator.destroy(soft_set);
+}
+
+pub fn reset(interface: *Interface, _: vk.DescriptorPoolResetFlags) VkError!void {
+    const self: *Self = @alignCast(@fieldParentPtr("interface", interface));
+    const allocator = VulkanAllocator.init(null, .object).allocator();
+
+    for (self.list.items) |set| {
+        allocator.destroy(set);
+    }
+    self.list.clearRetainingCapacity();
 }

src/soft/SoftDescriptorSet.zig

@@ -5,8 +5,12 @@ const base = @import("base");
 const VkError = base.VkError;
 const Device = base.Device;
 const Buffer = base.Buffer;
+const ImageView = base.ImageView;
 
 const SoftBuffer = @import("SoftBuffer.zig");
+const SoftImage = @import("SoftImage.zig");
+const SoftImageView = @import("SoftImageView.zig");
+const SoftSampler = @import("SoftSampler.zig");
 
 const NonDispatchable = base.NonDispatchable;
 
@@ -19,9 +23,20 @@ const DescriptorBuffer = struct {
     size: vk.DeviceSize,
 };
 
+const DescriptorTexture = struct {
+    sampler: ?*SoftSampler,
+    view: ?*SoftImageView,
+};
+
+const DescriptorImage = struct {
+    object: ?*SoftImage,
+};
+
 const Descriptor = union(enum) {
     buffer: []DescriptorBuffer,
-    image: struct {},
+    texture: []DescriptorTexture,
+    image: []DescriptorImage,
+    unsupported: struct {},
 };
 
 interface: Interface,
@@ -48,6 +63,7 @@ pub fn create(device: *base.Device, allocator: std.mem.Allocator, layout: *base.
         for (layout.bindings) |binding| {
             const struct_size: usize = switch (binding.descriptor_type) {
                 .storage_buffer, .storage_buffer_dynamic => @sizeOf(DescriptorBuffer),
+                .storage_image, .input_attachment => @sizeOf(DescriptorImage),
                 else => 0,
             };
 
@@ -68,6 +84,9 @@ pub fn create(device: *base.Device, allocator: std.mem.Allocator, layout: *base.
             .storage_buffer, .storage_buffer_dynamic => descriptor.* = .{
                 .buffer = local_allocator.alloc(DescriptorBuffer, binding.array_size) catch return VkError.OutOfHostMemory,
             },
+            .storage_image, .input_attachment => descriptor.* = .{
+                .image = local_allocator.alloc(DescriptorImage, binding.array_size) catch return VkError.OutOfHostMemory,
+            },
             else => {},
         }
     }
@@ -113,6 +132,19 @@ pub fn write(interface: *Interface, write_data: vk.WriteDescriptorSet) VkError!v
                 }
             }
         },
-        else => base.unsupported("descriptor type {s} for writting", .{@tagName(write_data.descriptor_type)}),
+        .storage_image, .input_attachment => {
+            for (write_data.p_image_info, 0..write_data.descriptor_count) |image_info, i| {
+                const desc = &self.descriptors[write_data.dst_binding].image[i];
+                desc.* = .{ .object = null };
+                if (image_info.image_view != .null_handle) {
+                    const image_view = try NonDispatchable(ImageView).fromHandleObject(image_info.image_view);
+                    desc.object = @as(*SoftImage, @alignCast(@fieldParentPtr("interface", image_view.image)));
+                }
+            }
+        },
+        else => {
+            self.descriptors[write_data.dst_binding] = .{ .unsupported = .{} };
+            base.unsupported("descriptor type {s} for writting", .{@tagName(write_data.descriptor_type)});
+        },
     }
 }

src/soft/SoftDevice.zig

@@ -2,10 +2,9 @@ const std = @import("std");
 const vk = @import("vulkan");
 const base = @import("base");
 const builtin = @import("builtin");
-const config = @import("config");
+const config = base.config;
 
 const SoftQueue = @import("SoftQueue.zig");
-const Blitter = @import("device/Blitter.zig");
 
 pub const SoftBinarySemaphore = @import("SoftBinarySemaphore.zig");
 pub const SoftBuffer = @import("SoftBuffer.zig");
@@ -19,6 +18,7 @@ pub const SoftEvent = @import("SoftEvent.zig");
 pub const SoftFence = @import("SoftFence.zig");
 pub const SoftFramebuffer = @import("SoftFramebuffer.zig");
 pub const SoftImage = @import("SoftImage.zig");
+pub const SoftInstance = @import("SoftInstance.zig");
 pub const SoftImageView = @import("SoftImageView.zig");
 pub const SoftPipeline = @import("SoftPipeline.zig");
 pub const SoftPipelineCache = @import("SoftPipelineCache.zig");
@@ -35,16 +35,20 @@ pub const Interface = base.Device;
 
 const SpawnError = std.Thread.SpawnError;
 
-interface: Interface,
+const DeviceAllocator = struct {
-device_allocator: if (config.debug_allocator) std.heap.DebugAllocator(.{}) else std.heap.ThreadSafeAllocator,
+    pub inline fn allocator(_: @This()) std.mem.Allocator {
-workers: std.Thread.Pool,
+        return std.heap.smp_allocator;
-blitter: Blitter,
+    }
+};
 
-pub fn create(physical_device: *base.PhysicalDevice, allocator: std.mem.Allocator, info: *const vk.DeviceCreateInfo) VkError!*Self {
+interface: Interface,
+device_allocator: if (config.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;
     errdefer allocator.destroy(self);
 
-    var interface = try Interface.init(allocator, physical_device, info);
+    var interface = try Interface.init(allocator, instance, physical_device, info);
 
     interface.vtable = &.{
         .createQueue = SoftQueue.create,
@@ -77,14 +81,7 @@ pub fn create(physical_device: *base.PhysicalDevice, allocator: std.mem.Allocato
 
     self.* = .{
         .interface = interface,
-        .device_allocator = if (config.debug_allocator) .init else .{ .child_allocator = std.heap.c_allocator }, // TODO: better device allocator
+        .device_allocator = if (config.debug_allocator) .init else .{},
-        .workers = undefined,
-        .blitter = .init,
-    };
-
-    self.workers.init(.{ .allocator = self.device_allocator.allocator() }) catch |err| return switch (err) {
-        SpawnError.OutOfMemory, SpawnError.LockedMemoryLimitExceeded => VkError.OutOfDeviceMemory,
-        else => VkError.Unknown,
     };
 
     try self.interface.createQueues(allocator, info);
@@ -93,16 +90,14 @@ pub fn create(physical_device: *base.PhysicalDevice, allocator: std.mem.Allocato
 
 pub fn destroy(interface: *Interface, allocator: std.mem.Allocator) VkError!void {
     const self: *Self = @alignCast(@fieldParentPtr("interface", interface));
-    self.workers.deinit();
 
     if (config.debug_allocator) {
         // All device memory allocations should've been freed by now
         if (!self.device_allocator.detectLeaks()) {
             std.log.scoped(.vkDestroyDevice).debug("No device memory leaks detected", .{});
         }
+        allocator.destroy(self);
     }
-
-    allocator.destroy(self);
 }
 
 pub fn allocateMemory(interface: *Interface, allocator: std.mem.Allocator, info: *const vk.MemoryAllocateInfo) VkError!*base.DeviceMemory {

src/soft/SoftEvent.zig

@@ -9,8 +9,8 @@ const Self = @This();
 pub const Interface = base.Event;
 
 interface: Interface,
-mutex: std.Thread.Mutex,
+mutex: std.Io.Mutex,
-condition: std.Thread.Condition,
+condition: std.Io.Condition,
 is_signaled: bool,
 
 pub fn create(device: *base.Device, allocator: std.mem.Allocator, info: *const vk.EventCreateInfo) VkError!*Self {
@@ -29,8 +29,8 @@ pub fn create(device: *base.Device, allocator: std.mem.Allocator, info: *const v
 
     self.* = .{
         .interface = interface,
-        .mutex = std.Thread.Mutex{},
+        .mutex = .init,
-        .condition = std.Thread.Condition{},
+        .condition = .init,
         .is_signaled = false,
     };
     return self;
@@ -43,9 +43,10 @@ pub fn destroy(interface: *Interface, allocator: std.mem.Allocator) void {
 
 pub fn getStatus(interface: *Interface) VkError!void {
     const self: *Self = @alignCast(@fieldParentPtr("interface", interface));
+    const io = interface.owner.io();
 
-    self.mutex.lock();
+    self.mutex.lock(io) catch return VkError.DeviceLost;
-    defer self.mutex.unlock();
+    defer self.mutex.unlock(io);
 
     if (!self.is_signaled) {
         return VkError.EventReset;
@@ -54,35 +55,41 @@ pub fn getStatus(interface: *Interface) VkError!void {
 
 pub fn reset(interface: *Interface) VkError!void {
     const self: *Self = @alignCast(@fieldParentPtr("interface", interface));
+    const io = interface.owner.io();
 
-    self.mutex.lock();
+    self.mutex.lock(io) catch return VkError.DeviceLost;
-    defer self.mutex.unlock();
+    defer self.mutex.unlock(io);
 
     self.is_signaled = false;
 }
 
 pub fn signal(interface: *Interface) VkError!void {
     const self: *Self = @alignCast(@fieldParentPtr("interface", interface));
+    const io = interface.owner.io();
 
-    self.mutex.lock();
+    self.mutex.lock(io) catch return VkError.DeviceLost;
-    defer self.mutex.unlock();
+    defer self.mutex.unlock(io);
 
     self.is_signaled = true;
-    self.condition.broadcast();
+    self.condition.broadcast(io);
 }
 
 pub fn wait(interface: *Interface, timeout: u64) VkError!void {
     const self: *Self = @alignCast(@fieldParentPtr("interface", interface));
+    const io = interface.owner.io();
 
-    self.mutex.lock();
+    self.mutex.lock(io) catch return VkError.DeviceLost;
-    defer self.mutex.unlock();
+    defer self.mutex.unlock(io);
 
     if (self.is_signaled) return;
     if (timeout == 0) return VkError.Timeout;
 
-    if (timeout == std.math.maxInt(@TypeOf(timeout))) {
+    if (timeout != std.math.maxInt(@TypeOf(timeout))) {
-        self.condition.wait(&self.mutex);
+        const duration: std.Io.Clock.Duration = .{
-    } else {
+            .raw = .fromNanoseconds(@intCast(timeout)),
-        self.condition.timedWait(&self.mutex, timeout) catch return VkError.Timeout;
+            .clock = .cpu_process,
+        };
+        duration.sleep(io) catch return VkError.DeviceLost;
     }
+    self.condition.wait(io, &self.mutex) catch return VkError.DeviceLost;
 }

src/soft/SoftFence.zig

@@ -9,8 +9,8 @@ const Self = @This();
 pub const Interface = base.Fence;
 
 interface: Interface,
-mutex: std.Thread.Mutex,
+mutex: std.Io.Mutex,
-condition: std.Thread.Condition,
+condition: std.Io.Condition,
 is_signaled: bool,
 
 pub fn create(device: *Device, allocator: std.mem.Allocator, info: *const vk.FenceCreateInfo) VkError!*Self {
@@ -29,8 +29,8 @@ pub fn create(device: *Device, allocator: std.mem.Allocator, info: *const vk.Fen
 
     self.* = .{
         .interface = interface,
-        .mutex = std.Thread.Mutex{},
+        .mutex = .init,
-        .condition = std.Thread.Condition{},
+        .condition = .init,
         .is_signaled = info.flags.signaled_bit,
     };
     return self;
@@ -55,26 +55,31 @@ pub fn reset(interface: *Interface) VkError!void {
 
 pub fn signal(interface: *Interface) VkError!void {
     const self: *Self = @alignCast(@fieldParentPtr("interface", interface));
+    const io = interface.owner.io();
 
-    self.mutex.lock();
+    self.mutex.lock(io) catch return VkError.DeviceLost;
-    defer self.mutex.unlock();
+    defer self.mutex.unlock(io);
 
     self.is_signaled = true;
-    self.condition.broadcast();
+    self.condition.broadcast(io);
 }
 
 pub fn wait(interface: *Interface, timeout: u64) VkError!void {
     const self: *Self = @alignCast(@fieldParentPtr("interface", interface));
+    const io = interface.owner.io();
 
-    self.mutex.lock();
+    self.mutex.lock(io) catch return VkError.DeviceLost;
-    defer self.mutex.unlock();
+    defer self.mutex.unlock(io);
 
     if (self.is_signaled) return;
     if (timeout == 0) return VkError.Timeout;
 
-    if (timeout == std.math.maxInt(@TypeOf(timeout))) {
+    if (timeout != std.math.maxInt(@TypeOf(timeout))) {
-        self.condition.wait(&self.mutex);
+        const duration: std.Io.Clock.Duration = .{
-    } else {
+            .raw = .fromNanoseconds(@intCast(timeout)),
-        self.condition.timedWait(&self.mutex, timeout) catch return VkError.Timeout;
+            .clock = .cpu_process,
+        };
+        duration.sleep(io) catch return VkError.DeviceLost;
     }
+    self.condition.wait(io, &self.mutex) catch return VkError.DeviceLost;
 }

src/soft/SoftImage.zig

@@ -24,6 +24,7 @@ pub fn create(device: *base.Device, allocator: std.mem.Allocator, info: *const v
     interface.vtable = &.{
         .destroy = destroy,
         .getMemoryRequirements = getMemoryRequirements,
+        .getTotalSizeForAspect = getTotalSizeForAspect,
     };
 
     self.* = .{
@@ -37,46 +38,172 @@ pub fn destroy(interface: *Interface, allocator: std.mem.Allocator) void {
     allocator.destroy(self);
 }
 
-pub fn getMemoryRequirements(interface: *Interface, requirements: *vk.MemoryRequirements) void {
+pub fn getMemoryRequirements(_: *Interface, requirements: *vk.MemoryRequirements) VkError!void {
-    _ = interface;
     requirements.alignment = lib.MEMORY_REQUIREMENTS_IMAGE_ALIGNMENT;
 }
 
-inline fn clear(self: *Self, pixel: vk.ClearValue, format: vk.Format, view_format: vk.Format, range: vk.ImageSubresourceRange, area: ?vk.Rect2D) void {
+pub fn getClearFormat(self: *Self) VkError!vk.Format {
-    const soft_device: *SoftDevice = @alignCast(@fieldParentPtr("interface", self.interface.owner));
+    return if (base.vku.vkuFormatIsSINT(@intCast(@intFromEnum(self.interface.format))))
-    soft_device.blitter.clear(pixel, format, self, view_format, range, area);
-}
-
-pub fn clearRange(self: *Self, color: vk.ClearColorValue, range: vk.ImageSubresourceRange) void {
-    std.debug.assert(range.aspect_mask == vk.ImageAspectFlags{ .color_bit = true });
-
-    const clear_format: vk.Format = if (base.vku.vkuFormatIsSINT(@intCast(@intFromEnum(self.interface.format))))
         .r32g32b32a32_sint
     else if (base.vku.vkuFormatIsUINT(@intCast(@intFromEnum(self.interface.format))))
         .r32g32b32a32_uint
     else
         .r32g32b32a32_sfloat;
-    self.clear(.{ .color = color }, clear_format, self.interface.format, range, null);
 }
 
-pub fn copyImage(self: *const Self, self_layout: vk.ImageLayout, dst: *Self, dst_layout: vk.ImageLayout, regions: []const vk.ImageCopy) VkError!void {
+/// Based on SwiftShader vk::Image::copyTo
-    _ = self;
+pub fn copyToImage(self: *const Self, dst: *Self, region: vk.ImageCopy) VkError!void {
-    _ = self_layout;
+    const combined_depth_stencil_aspect: vk.ImageAspectFlags = .{
-    _ = dst;
+        .depth_bit = true,
-    _ = dst_layout;
+        .stencil_bit = true,
-    _ = regions;
+    };
-    std.log.scoped(.commandExecutor).warn("FIXME: implement image to image copy", .{});
+
+    if (region.src_subresource.aspect_mask == combined_depth_stencil_aspect and
+        region.dst_subresource.aspect_mask == combined_depth_stencil_aspect)
+    {
+        var single_aspect_region = region;
+
+        single_aspect_region.src_subresource.aspect_mask = .{ .depth_bit = true };
+        single_aspect_region.dst_subresource.aspect_mask = .{ .depth_bit = true };
+        try self.copyToImageSingleAspect(dst, single_aspect_region);
+
+        single_aspect_region.src_subresource.aspect_mask = .{ .stencil_bit = true };
+        single_aspect_region.dst_subresource.aspect_mask = .{ .stencil_bit = true };
+        try self.copyToImageSingleAspect(dst, single_aspect_region);
+    } else {
+        try self.copyToImageSingleAspect(dst, region);
+    }
+}
+
+/// Based on SwiftShader vk::Image::copySingleAspectTo
+pub fn copyToImageSingleAspect(self: *const Self, dst: *Self, region: vk.ImageCopy) VkError!void {
+    if (!(region.src_subresource.aspect_mask == vk.ImageAspectFlags{ .color_bit = true } or
+        region.src_subresource.aspect_mask == vk.ImageAspectFlags{ .depth_bit = true } or
+        region.src_subresource.aspect_mask == vk.ImageAspectFlags{ .stencil_bit = true }))
+    {
+        base.unsupported("src subresource aspectMask {f}", .{region.src_subresource.aspect_mask});
+        return VkError.ValidationFailed;
+    }
+
+    if (!(region.dst_subresource.aspect_mask == vk.ImageAspectFlags{ .color_bit = true } or
+        region.dst_subresource.aspect_mask == vk.ImageAspectFlags{ .depth_bit = true } or
+        region.dst_subresource.aspect_mask == vk.ImageAspectFlags{ .stencil_bit = true }))
+    {
+        base.unsupported("dst subresource aspectMask {f}", .{region.dst_subresource.aspect_mask});
+        return VkError.ValidationFailed;
+    }
+
+    const src_format = self.interface.formatFromAspect(region.src_subresource.aspect_mask);
+    const bytes_per_block = base.format.texelSize(src_format);
+
+    const src_extent = self.getMipLevelExtent(region.src_subresource.mip_level);
+    const dst_extent = dst.getMipLevelExtent(region.src_subresource.mip_level);
+
+    const one_is_3D = (self.interface.image_type == .@"3d") != (dst.interface.image_type == .@"3d");
+    const both_are_3D = (self.interface.image_type == .@"3d") and (dst.interface.image_type == .@"3d");
+
+    const src_row_pitch_bytes = self.getRowPitchMemSizeForMipLevel(region.src_subresource.aspect_mask, region.src_subresource.mip_level);
+    const src_depth_pitch_bytes = self.getSliceMemSizeForMipLevel(region.src_subresource.aspect_mask, region.src_subresource.mip_level);
+    const dst_row_pitch_bytes = dst.getRowPitchMemSizeForMipLevel(region.dst_subresource.aspect_mask, region.dst_subresource.mip_level);
+    const dst_depth_pitch_bytes = dst.getSliceMemSizeForMipLevel(region.dst_subresource.aspect_mask, region.dst_subresource.mip_level);
+
+    const src_array_pitch = self.getLayerSize(region.src_subresource.aspect_mask);
+    const dst_array_pitch = dst.getLayerSize(region.dst_subresource.aspect_mask);
+
+    const src_layer_pitch = if (self.interface.image_type == .@"3d") src_depth_pitch_bytes else src_array_pitch;
+    const dst_layer_pitch = if (dst.interface.image_type == .@"3d") dst_depth_pitch_bytes else dst_array_pitch;
+
+    // If one image is 3D, extent.depth must match the layer count. If both images are 2D,
+    // depth is 1 but the source and destination subresource layer count must match.
+    const layer_count = if (one_is_3D) region.extent.depth else region.src_subresource.layer_count;
+
+    // Copies between 2D and 3D images are treated as layers, so only use depth as the slice count when both images are 3D.
+    const slice_count = if (both_are_3D) region.extent.depth else self.interface.samples.toInt();
+
+    const is_single_slice = (slice_count == 1);
+    const is_single_row = (region.extent.height == 1) and is_single_slice;
+
+    // In order to copy multiple rows using a single memcpy call, we
+    // have to make sure that we need to copy the entire row and that
+    // both source and destination rows have the same size in bytes
+    const is_entire_row = (region.extent.width == src_extent.width) and (region.extent.width == dst_extent.width);
+
+    // In order to copy multiple slices using a single memcpy call, we
+    // have to make sure that we need to copy the entire slice and that
+    // both source and destination slices have the same size in bytes
+    const is_entire_slice = is_entire_row and
+        (region.extent.height == src_extent.height) and
+        (region.extent.height == dst_extent.height) and
+        (src_depth_pitch_bytes == dst_depth_pitch_bytes);
+
+    const src_texel_offset = try self.getTexelMemoryOffset(region.src_offset, .{
+        .aspect_mask = region.src_subresource.aspect_mask,
+        .mip_level = region.src_subresource.mip_level,
+        .array_layer = region.src_subresource.base_array_layer,
+    });
+    const src_size = try self.interface.getTotalSizeForAspect(region.src_subresource.aspect_mask) - src_texel_offset;
+    const src_memory = if (self.interface.memory) |memory| memory else return VkError.InvalidDeviceMemoryDrv;
+    var src_map: []u8 = @as([*]u8, @ptrCast(try src_memory.map(self.interface.memory_offset + src_texel_offset, src_size)))[0..src_size];
+
+    const dst_texel_offset = try self.getTexelMemoryOffset(region.dst_offset, .{
+        .aspect_mask = region.dst_subresource.aspect_mask,
+        .mip_level = region.dst_subresource.mip_level,
+        .array_layer = region.dst_subresource.base_array_layer,
+    });
+    const dst_size = try dst.interface.getTotalSizeForAspect(region.dst_subresource.aspect_mask) - dst_texel_offset;
+    const dst_memory = if (dst.interface.memory) |memory| memory else return VkError.InvalidDeviceMemoryDrv;
+    var dst_map: []u8 = @as([*]u8, @ptrCast(try dst_memory.map(self.interface.memory_offset + dst_texel_offset, dst_size)))[0..dst_size];
+
+    for (0..layer_count) |_| {
+        if (is_single_row) {
+            const copy_size = region.extent.width * bytes_per_block;
+            @memcpy(dst_map[0..copy_size], src_map[0..copy_size]);
+        } else if (is_entire_row and is_single_slice) {
+            const copy_size = region.extent.height * src_row_pitch_bytes;
+            @memcpy(dst_map[0..copy_size], src_map[0..copy_size]);
+        } else if (is_entire_slice) {
+            const copy_size = slice_count * src_depth_pitch_bytes;
+            @memcpy(dst_map[0..copy_size], src_map[0..copy_size]);
+        } else if (is_entire_row) {
+            const slice_size = region.extent.height * src_row_pitch_bytes;
+            var src_slice_memory = src_map[0..];
+            var dst_slice_memory = dst_map[0..];
+
+            for (0..slice_count) |_| {
+                @memcpy(dst_slice_memory[0..slice_size], src_slice_memory[0..slice_size]);
+                src_slice_memory = src_slice_memory[src_depth_pitch_bytes..];
+                dst_slice_memory = dst_slice_memory[dst_depth_pitch_bytes..];
+            }
+        } else {
+            const row_size = region.extent.width * bytes_per_block;
+            var src_slice_memory = src_map[0..];
+            var dst_slice_memory = dst_map[0..];
+
+            for (0..slice_count) |_| {
+                var src_row_memory = src_slice_memory[0..];
+                var dst_row_memory = dst_slice_memory[0..];
+
+                for (0..region.extent.height) |_| {
+                    @memcpy(dst_row_memory[0..row_size], src_row_memory[0..row_size]);
+                    src_row_memory = src_row_memory[src_row_pitch_bytes..];
+                    dst_row_memory = dst_row_memory[dst_row_pitch_bytes..];
+                }
+            }
+        }
+
+        src_map = src_map[src_layer_pitch..];
+        dst_map = dst_map[dst_layer_pitch..];
+    }
 }
 
 pub fn copyToBuffer(self: *const Self, dst: *SoftBuffer, region: vk.BufferImageCopy) VkError!void {
     const dst_size = dst.interface.size - region.buffer_offset;
+    const dst_offset = dst.interface.offset + region.buffer_offset;
     const dst_memory = if (dst.interface.memory) |memory| memory else return VkError.InvalidDeviceMemoryDrv;
-    const dst_map: []u8 = @as([*]u8, @ptrCast(try dst_memory.map(region.buffer_offset, dst_size)))[0..dst_size];
+    const dst_map: []u8 = @as([*]u8, @ptrCast(try dst_memory.map(dst_offset, dst_size)))[0..dst_size];
     try self.copy(
         null,
         dst_map,
-        @intCast(region.buffer_row_length),
-        @intCast(region.buffer_image_height),
         region.image_subresource,
         region.image_offset,
         region.image_extent,
@@ -85,35 +212,196 @@ pub fn copyToBuffer(self: *const Self, dst: *SoftBuffer, region: vk.BufferImageC
 
 pub fn copyFromBuffer(self: *const Self, src: *const SoftBuffer, region: vk.BufferImageCopy) VkError!void {
     const src_size = src.interface.size - region.buffer_offset;
+    const src_offset = src.interface.offset + region.buffer_offset;
     const src_memory = if (src.interface.memory) |memory| memory else return VkError.InvalidDeviceMemoryDrv;
-    const src_map: []u8 = @as([*]u8, @ptrCast(try src_memory.map(region.buffer_offset, src_size)))[0..src_size];
+    const src_map: []u8 = @as([*]u8, @ptrCast(try src_memory.map(src_offset, src_size)))[0..src_size];
     try self.copy(
         src_map,
         null,
-        @intCast(region.buffer_row_length),
-        @intCast(region.buffer_image_height),
         region.image_subresource,
         region.image_offset,
         region.image_extent,
     );
 }
 
+/// Based on SwiftShader vk::Image::copy
 pub fn copy(
     self: *const Self,
-    src_memory: ?[]const u8,
+    base_src_memory: ?[]const u8,
-    dst_memory: ?[]u8,
+    base_dst_memory: ?[]u8,
-    row_len: usize,
-    image_height: usize,
     image_subresource: vk.ImageSubresourceLayers,
-    image_copy_offset: vk.Offset3D,
+    image_offset: vk.Offset3D,
-    image_copy_extent: vk.Extent3D,
+    image_extent: vk.Extent3D,
 ) VkError!void {
-    _ = self;
+    std.debug.assert((base_src_memory == null) != (base_dst_memory == null));
-    _ = src_memory;
+
-    _ = dst_memory;
+    const is_source: bool = base_src_memory != null;
-    _ = row_len;
+
-    _ = image_height;
+    if (image_subresource.aspect_mask.subtract(.{
-    _ = image_subresource;
+        .color_bit = true,
-    _ = image_copy_offset;
+        .depth_bit = true,
-    _ = image_copy_extent;
+        .stencil_bit = true,
+    }).toInt() != 0) {
+        base.unsupported("aspectMask {f}", .{image_subresource.aspect_mask});
+        return VkError.ValidationFailed;
+    }
+
+    const format = self.interface.formatFromAspect(image_subresource.aspect_mask);
+
+    // TODO: handle extent of compressed formats
+
+    if (image_extent.width == 0 or image_extent.height == 0 or image_extent.depth == 0) {
+        return;
+    }
+
+    const bytes_per_block = base.format.texelSize(format);
+    const memory_row_pitch_bytes = image_extent.width * bytes_per_block;
+    const memory_slice_pitch_bytes = image_extent.height * memory_row_pitch_bytes;
+
+    const image_texel_offset = try self.getTexelMemoryOffset(image_offset, .{
+        .aspect_mask = image_subresource.aspect_mask,
+        .mip_level = image_subresource.mip_level,
+        .array_layer = image_subresource.base_array_layer,
+    });
+    const image_size = try self.interface.getTotalSizeForAspect(image_subresource.aspect_mask) - image_texel_offset;
+    const image_memory = if (self.interface.memory) |memory| memory else return VkError.InvalidDeviceMemoryDrv;
+    const image_map: []u8 = @as([*]u8, @ptrCast(try image_memory.map(self.interface.memory_offset + image_texel_offset, image_size)))[0..image_size];
+
+    var src_memory = if (is_source) base_src_memory orelse return VkError.InvalidDeviceMemoryDrv else image_map;
+    var dst_memory = if (is_source) image_map else base_dst_memory orelse return VkError.InvalidDeviceMemoryDrv;
+
+    const src_slice_pitch_bytes = if (is_source) memory_slice_pitch_bytes else self.getSliceMemSizeForMipLevel(image_subresource.aspect_mask, image_subresource.mip_level);
+    const dst_slice_pitch_bytes = if (is_source) self.getSliceMemSizeForMipLevel(image_subresource.aspect_mask, image_subresource.mip_level) else memory_slice_pitch_bytes;
+    const src_row_pitch_bytes = if (is_source) memory_row_pitch_bytes else self.getRowPitchMemSizeForMipLevel(image_subresource.aspect_mask, image_subresource.mip_level);
+    const dst_row_pitch_bytes = if (is_source) self.getRowPitchMemSizeForMipLevel(image_subresource.aspect_mask, image_subresource.mip_level) else memory_row_pitch_bytes;
+
+    const src_layer_size = if (is_source) memory_slice_pitch_bytes else self.getLayerSize(image_subresource.aspect_mask);
+    const dst_layer_size = if (is_source) self.getLayerSize(image_subresource.aspect_mask) else memory_slice_pitch_bytes;
+
+    const layer_count = if (image_subresource.layer_count == vk.REMAINING_ARRAY_LAYERS) self.interface.array_layers - image_subresource.base_array_layer else image_subresource.layer_count;
+
+    const copy_size = image_extent.width * bytes_per_block;
+
+    for (0..layer_count) |_| {
+        var src_layer_memory = src_memory[0..];
+        var dst_layer_memory = dst_memory[0..];
+
+        for (0..image_extent.depth) |_| {
+            var src_slice_memory = src_layer_memory[0..];
+            var dst_slice_memory = dst_layer_memory[0..];
+
+            for (0..image_extent.height) |_| {
+                @memcpy(dst_slice_memory[0..copy_size], src_slice_memory[0..copy_size]);
+                src_slice_memory = src_slice_memory[src_row_pitch_bytes..];
+                dst_slice_memory = dst_slice_memory[dst_row_pitch_bytes..];
+            }
+            src_layer_memory = src_layer_memory[src_slice_pitch_bytes..];
+            dst_layer_memory = dst_layer_memory[dst_slice_pitch_bytes..];
+        }
+        src_memory = src_memory[src_layer_size..];
+        dst_memory = dst_memory[dst_layer_size..];
+    }
+}
+
+pub fn getTexelMemoryOffsetInSubresource(self: *const Self, offset: vk.Offset3D, subresource: vk.ImageSubresource) usize {
+    return @as(usize, @intCast(offset.z)) * self.getSliceMemSizeForMipLevel(subresource.aspect_mask, subresource.mip_level) +
+        @as(usize, @intCast(offset.y)) * self.getRowPitchMemSizeForMipLevel(subresource.aspect_mask, subresource.mip_level) +
+        @as(usize, @intCast(offset.x)) * base.format.texelSize(base.format.fromAspect(self.interface.format, subresource.aspect_mask));
+}
+
+pub fn getTexelMemoryOffset(self: *const Self, offset: vk.Offset3D, subresource: vk.ImageSubresource) VkError!usize {
+    return self.getTexelMemoryOffsetInSubresource(offset, subresource) + try self.getSubresourceOffset(subresource.aspect_mask, subresource.mip_level, subresource.array_layer);
+}
+
+fn getSubresourceOffset(self: *const Self, aspect_mask: vk.ImageAspectFlags, mip_level: u32, layer: u32) VkError!usize {
+    var offset = try self.getAspectOffset(aspect_mask);
+    for (0..mip_level) |mip| {
+        offset += self.getMultiSampledLevelSize(aspect_mask, @intCast(mip));
+    }
+
+    const is_3D = (self.interface.image_type == .@"3d") and self.interface.flags.@"2d_array_compatible_bit";
+    const layer_offset = if (is_3D)
+        self.getSliceMemSizeForMipLevel(aspect_mask, mip_level)
+    else
+        self.getLayerSize(aspect_mask);
+    return offset + layer * layer_offset;
+}
+
+fn getAspectOffset(self: *const Self, aspect_mask: vk.ImageAspectFlags) VkError!usize {
+    return switch (self.interface.format) {
+        .d16_unorm_s8_uint,
+        .d24_unorm_s8_uint,
+        .d32_sfloat_s8_uint,
+        => if (aspect_mask.stencil_bit)
+            try self.interface.getTotalSizeForAspect(.{ .depth_bit = true })
+        else
+            0,
+        else => 0,
+    };
+}
+
+fn getTotalSizeForAspect(interface: *const Interface, aspect_mask: vk.ImageAspectFlags) VkError!usize {
+    const self: *const Self = @alignCast(@fieldParentPtr("interface", interface));
+
+    if (aspect_mask.subtract(.{
+        .color_bit = true,
+        .depth_bit = true,
+        .stencil_bit = true,
+    }).toInt() != 0) {
+        base.unsupported("aspectMask {f}", .{aspect_mask});
+        return VkError.ValidationFailed;
+    }
+
+    var size: usize = 0;
+
+    if (aspect_mask.color_bit)
+        size += self.getLayerSize(.{ .color_bit = true });
+    if (aspect_mask.depth_bit)
+        size += self.getLayerSize(.{ .depth_bit = true });
+    if (aspect_mask.stencil_bit)
+        size += self.getLayerSize(.{ .stencil_bit = true });
+
+    return size * self.interface.array_layers;
+}
+
+pub fn getLayerSize(self: *const Self, aspect_mask: vk.ImageAspectFlags) usize {
+    var size: usize = 0;
+    for (0..self.interface.mip_levels) |mip_level| {
+        size += self.getMultiSampledLevelSize(aspect_mask, @intCast(mip_level));
+    }
+    return size;
+}
+
+pub inline fn getMultiSampledLevelSize(self: *const Self, aspect_mask: vk.ImageAspectFlags, mip_level: u32) usize {
+    return self.getMipLevelSize(aspect_mask, mip_level) * self.interface.samples.toInt();
+}
+
+pub inline fn getMipLevelSize(self: *const Self, aspect_mask: vk.ImageAspectFlags, mip_level: u32) usize {
+    return self.getSliceMemSizeForMipLevel(aspect_mask, mip_level) * self.getMipLevelExtent(mip_level).depth;
+}
+
+pub fn getMipLevelExtent(self: *const Self, mip_level: u32) vk.Extent3D {
+    var extent: vk.Extent3D = .{
+        .width = self.interface.extent.width >> @intCast(mip_level),
+        .height = self.interface.extent.height >> @intCast(mip_level),
+        .depth = self.interface.extent.depth >> @intCast(mip_level),
+    };
+
+    if (extent.width == 0) extent.width = 1;
+    if (extent.height == 0) extent.height = 1;
+    if (extent.depth == 0) extent.depth = 1;
+
+    return extent;
+}
+
+pub fn getSliceMemSizeForMipLevel(self: *const Self, aspect_mask: vk.ImageAspectFlags, mip_level: u32) usize {
+    const mip_extent = self.getMipLevelExtent(mip_level);
+    const format = self.interface.formatFromAspect(aspect_mask);
+    return base.format.sliceMemSize(format, mip_extent.width, mip_extent.height);
+}
+
+pub fn getRowPitchMemSizeForMipLevel(self: *const Self, aspect_mask: vk.ImageAspectFlags, mip_level: u32) usize {
+    const mip_extent = self.getMipLevelExtent(mip_level);
+    const format = self.interface.formatFromAspect(aspect_mask);
+    return base.format.pitchMemSize(format, mip_extent.width);
 }

src/soft/SoftInstance.zig

@@ -11,6 +11,8 @@ const Self = @This();
 pub const Interface = base.Instance;
 
 interface: Interface,
+threaded: std.Io.Threaded,
+allocator: std.mem.Allocator,
 
 fn castExtension(comptime ext: vk.ApiInfo) vk.ExtensionProperties {
     var props: vk.ExtensionProperties = .{
@@ -29,6 +31,9 @@ pub fn create(allocator: std.mem.Allocator, infos: *const vk.InstanceCreateInfo)
     const self = allocator.create(Self) catch return VkError.OutOfHostMemory;
     errdefer allocator.destroy(self);
 
+    self.allocator = std.heap.smp_allocator;
+    self.threaded = std.Io.Threaded.init(self.allocator, .{});
+
     self.interface = try base.Instance.init(allocator, infos);
     self.interface.dispatch_table = &.{
         .destroy = destroy,
@@ -36,12 +41,14 @@ pub fn create(allocator: std.mem.Allocator, infos: *const vk.InstanceCreateInfo)
     self.interface.vtable = &.{
         .requestPhysicalDevices = requestPhysicalDevices,
         .releasePhysicalDevices = releasePhysicalDevices,
+        .io = io,
     };
     return &self.interface;
 }
 
 fn destroy(interface: *Interface, allocator: std.mem.Allocator) VkError!void {
     const self: *Self = @alignCast(@fieldParentPtr("interface", interface));
+    self.threaded.deinit();
     allocator.destroy(self);
 }
 
@@ -60,3 +67,8 @@ fn releasePhysicalDevices(interface: *Interface, allocator: std.mem.Allocator) V
     interface.physical_devices.deinit(allocator);
     interface.physical_devices = .empty;
 }
+
+fn io(interface: *Interface) std.Io {
+    const self: *Self = @alignCast(@fieldParentPtr("interface", interface));
+    return self.threaded.io();
+}

src/soft/SoftPhysicalDevice.zig

@@ -17,7 +17,7 @@ var device_name = [_]u8{0} ** vk.MAX_PHYSICAL_DEVICE_NAME_SIZE;
 
 interface: Interface,
 
-pub fn create(allocator: std.mem.Allocator, instance: *const base.Instance) VkError!*Self {
+pub fn create(allocator: std.mem.Allocator, instance: *base.Instance) VkError!*Self {
     const command_allocator = VulkanAllocator.from(allocator).cloneWithScope(.command).allocator();
 
     const self = allocator.create(Self) catch return VkError.OutOfHostMemory;
@@ -224,7 +224,7 @@ pub fn destroy(interface: *Interface, allocator: std.mem.Allocator) VkError!void
 }
 
 pub fn createDevice(interface: *Interface, allocator: std.mem.Allocator, infos: *const vk.DeviceCreateInfo) VkError!*base.Device {
-    const device = try SoftDevice.create(interface, allocator, infos);
+    const device = try SoftDevice.create(interface.instance, interface, allocator, infos);
     return &device.interface;
 }
 
@@ -515,7 +515,7 @@ pub fn getFormatProperties(interface: *Interface, format: vk.Format) VkError!vk.
         else => {},
     }
 
-    if (base.Image.formatSupportsColorAttachemendBlend(format)) {
+    if (base.format.supportsColorAttachemendBlend(format)) {
         properties.optimal_tiling_features.color_attachment_blend_bit = true;
     }
 

src/soft/SoftPipeline.zig

@@ -10,6 +10,7 @@ const NonDispatchable = base.NonDispatchable;
 const ShaderModule = base.ShaderModule;
 
 const SoftDevice = @import("SoftDevice.zig");
+const SoftInstance = @import("SoftInstance.zig");
 const SoftShaderModule = @import("SoftShaderModule.zig");
 
 const Self = @This();
@@ -49,27 +50,54 @@ pub fn createCompute(device: *base.Device, allocator: std.mem.Allocator, cache:
 
     const device_allocator = soft_device.device_allocator.allocator();
 
+    const instance: *SoftInstance = @alignCast(@fieldParentPtr("interface", device.instance));
+    const runtimes_count = switch (instance.threaded.async_limit) {
+        .nothing => 1,
+        .unlimited => std.Thread.getCpuCount() catch 1, // If we cannot get the CPU count, fallback on single runtime
+        else => |count| blk: {
+            const cpu_count: usize = std.Thread.getCpuCount() catch break :blk @intFromEnum(count);
+            break :blk if (@intFromEnum(count) >= cpu_count) cpu_count else @intFromEnum(count);
+        },
+    };
+
     self.* = .{
         .interface = interface,
         .stages = std.EnumMap(Stages, Shader).init(.{
-            .compute = .{
+            .compute = blk: {
-                .module = blk: {
+                var shader: Shader = undefined;
-                    soft_module.ref();
+                soft_module.ref();
-                    break :blk soft_module;
+                shader.module = soft_module;
-                },
-                .runtimes = blk: {
-                    const runtimes = device_allocator.alloc(spv.Runtime, soft_device.workers.getIdCount()) catch return VkError.OutOfHostMemory;
-                    errdefer device_allocator.free(runtimes);
 
+                const runtimes = device_allocator.alloc(spv.Runtime, runtimes_count) catch return VkError.OutOfHostMemory;
+                errdefer {
                     for (runtimes) |*runtime| {
-                        runtime.* = spv.Runtime.init(device_allocator, &soft_module.module) catch |err| {
+                        runtime.deinit(device_allocator);
-                            std.log.scoped(.SpvRuntimeInit).err("SPIR-V Runtime failed to initialize, {s}", .{@errorName(err)});
-                            return VkError.Unknown;
-                        };
                     }
-                    break :blk runtimes;
+                    device_allocator.free(runtimes);
-                },
+                }
-                .entry = allocator.dupe(u8, std.mem.span(info.stage.p_name)) catch return VkError.OutOfHostMemory,
+
+                for (runtimes) |*runtime| {
+                    runtime.* = spv.Runtime.init(device_allocator, &soft_module.module) catch |err| {
+                        std.log.scoped(.SpvRuntimeInit).err("SPIR-V Runtime failed to initialize, {s}", .{@errorName(err)});
+                        return VkError.Unknown;
+                    };
+                    if (info.stage.p_specialization_info) |specialization| {
+                        if (specialization.p_map_entries) |map| {
+                            const data: []const u8 = @as([*]const u8, @ptrCast(@alignCast(specialization.p_data)))[0..specialization.data_size];
+                            for (map[0..], 0..specialization.map_entry_count) |entry, _| {
+                                runtime.addSpecializationInfo(device_allocator, .{
+                                    .id = @intCast(entry.constant_id),
+                                    .offset = @intCast(entry.offset),
+                                    .size = @intCast(entry.size),
+                                }, data) catch return VkError.OutOfHostMemory;
+                            }
+                        }
+                    }
+                }
+
+                shader.runtimes = runtimes;
+                shader.entry = device_allocator.dupe(u8, std.mem.span(info.stage.p_name)) catch return VkError.OutOfHostMemory;
+                break :blk shader;
             },
         }),
     };
@@ -86,9 +114,17 @@ pub fn createGraphics(device: *base.Device, allocator: std.mem.Allocator, cache:
         .destroy = destroy,
     };
 
-    const soft_device: *SoftDevice = @alignCast(@fieldParentPtr("interface", device));
+    const instance: *SoftInstance = @alignCast(@fieldParentPtr("interface", device.instance));
+    const runtimes_count = switch (instance.threaded.async_limit) {
+        .nothing => 1,
+        .unlimited => std.Thread.getCpuCount() catch 1, // If we cannot get the CPU count, fallback on single runtime
+        else => |count| blk: {
+            const cpu_count: usize = std.Thread.getCpuCount() catch break :blk @intFromEnum(count);
+            break :blk if (@intFromEnum(count) >= cpu_count) cpu_count else @intFromEnum(count);
+        },
+    };
 
-    const runtimes = allocator.alloc(spv.Runtime, soft_device.workers.getIdCount()) catch return VkError.OutOfHostMemory;
+    const runtimes = allocator.alloc(spv.Runtime, runtimes_count) catch return VkError.OutOfHostMemory;
     errdefer allocator.free(runtimes);
 
     //for (runtimes) |*runtime| {
@@ -112,12 +148,12 @@ pub fn destroy(interface: *Interface, allocator: std.mem.Allocator) void {
 
     var it = self.stages.iterator();
     while (it.next()) |stage| {
-        stage.value.module.unref(allocator);
         for (stage.value.runtimes) |*runtime| {
             runtime.deinit(device_allocator);
         }
         device_allocator.free(stage.value.runtimes);
-        allocator.free(stage.value.entry);
+        device_allocator.free(stage.value.entry);
+        stage.value.module.unref(allocator);
     }
     allocator.destroy(self);
 }

src/soft/SoftQueue.zig

@@ -17,7 +17,7 @@ const Self = @This();
 pub const Interface = base.Queue;
 
 interface: Interface,
-lock: std.Thread.RwLock,
+lock: std.Io.RwLock,
 
 pub fn create(allocator: std.mem.Allocator, device: *base.Device, index: u32, family_index: u32, flags: vk.DeviceQueueCreateFlags) VkError!*Interface {
     const self = allocator.create(Self) catch return VkError.OutOfHostMemory;
@@ -33,7 +33,7 @@ pub fn create(allocator: std.mem.Allocator, device: *base.Device, index: u32, fa
 
     self.* = .{
         .interface = interface,
-        .lock = .{},
+        .lock = .init,
     };
     return &self.interface;
 }
@@ -56,10 +56,11 @@ pub fn submit(interface: *Interface, infos: []Interface.SubmitInfo, p_fence: ?*b
     const soft_device: *SoftDevice = @alignCast(@fieldParentPtr("interface", interface.owner));
 
     const allocator = soft_device.device_allocator.allocator();
+    const io = soft_device.interface.io();
 
     // Lock here to avoid acquiring it in `waitIdle` before runners start
-    self.lock.lockShared();
+    self.lock.lockShared(io) catch return VkError.DeviceLost;
-    defer self.lock.unlockShared();
+    defer self.lock.unlockShared(io);
 
     for (infos) |info| {
         // Cloning info to keep them alive until command execution ends
@@ -68,19 +69,23 @@ pub fn submit(interface: *Interface, infos: []Interface.SubmitInfo, p_fence: ?*b
         };
         const runners_counter = allocator.create(RefCounter) catch return VkError.OutOfDeviceMemory;
         runners_counter.* = .init;
-        soft_device.workers.spawn(Self.taskRunner, .{ self, cloned_info, p_fence, runners_counter }) catch return VkError.Unknown;
+        _ = soft_device.interface.io().async(Self.taskRunner, .{ self, cloned_info, p_fence, runners_counter });
     }
 }
 
 pub fn waitIdle(interface: *Interface) VkError!void {
     const self: *Self = @alignCast(@fieldParentPtr("interface", interface));
-    self.lock.lock();
+    const io = interface.owner.io();
-    defer self.lock.unlock();
+
+    self.lock.lock(io) catch return VkError.DeviceLost;
+    defer self.lock.unlock(io);
 }
 
 fn taskRunner(self: *Self, info: Interface.SubmitInfo, p_fence: ?*base.Fence, runners_counter: *RefCounter) void {
-    self.lock.lockShared();
+    const io = self.interface.owner.io();
-    defer self.lock.unlockShared();
+
+    self.lock.lockShared(io) catch return;
+    defer self.lock.unlockShared(io);
 
     runners_counter.ref();
     defer {

src/soft/SoftShaderModule.zig

@@ -32,13 +32,13 @@ pub fn create(device: *base.Device, allocator: std.mem.Allocator, info: *const v
     self.* = .{
         .interface = interface,
         .module = spv.Module.init(allocator, code, .{
-            .use_simd_vectors_specializations = !std.process.hasEnvVarConstant(lib.NO_SHADER_SIMD_ENV_NAME),
+            .use_simd_vectors_specializations = base.config.shaders_simd,
         }) catch |err| switch (err) {
             spv.Module.ModuleError.OutOfMemory => return VkError.OutOfHostMemory,
             else => {
                 std.log.scoped(.@"SPIR-V module").err("module creation catched a '{s}'", .{@errorName(err)});
                 if (@errorReturnTrace()) |trace| {
-                    std.debug.dumpStackTrace(trace.*);
+                    std.debug.dumpErrorReturnTrace(trace);
                 }
                 return VkError.ValidationFailed;
             },

src/soft/device/Blitter.zig

@@ -4,19 +4,17 @@ const std = @import("std");
 const vk = @import("vulkan");
 const base = @import("base");
 
+const VkError = base.VkError;
+
 pub const SoftImage = @import("../SoftImage.zig");
 pub const SoftImageView = @import("../SoftImageView.zig");
 
 const Self = @This();
 
-blit_mutex: std.Thread.Mutex,
+pub const init: Self = .{};
 
-pub const init: Self = .{
+pub fn clear(self: *Self, pixel: vk.ClearValue, format: vk.Format, dest: *SoftImage, view_format: vk.Format, range: vk.ImageSubresourceRange, area: ?vk.Rect2D) VkError!void {
-    .blit_mutex = .{},
+    const dst_format = base.format.fromAspect(view_format, range.aspect_mask);
-};
-
-pub fn clear(self: *Self, pixel: vk.ClearValue, format: vk.Format, dest: *SoftImage, view_format: vk.Format, range: vk.ImageSubresourceRange, area: ?vk.Rect2D) void {
-    const dst_format = base.Image.formatFromAspect(view_format, range.aspect_mask);
     if (dst_format == .undefined) {
         return;
     }
@@ -40,13 +38,13 @@ pub fn clear(self: *Self, pixel: vk.ClearValue, format: vk.Format, dest: *SoftIm
         }
     }
 
-    if (self.fastClear(clamped_pixel, format, dest, dst_format, range, area)) {
+    if (try self.fastClear(clamped_pixel, format, dest, dst_format, range, area)) {
         return;
     }
     base.logger.fixme("implement slow clear", .{});
 }
 
-fn fastClear(self: *Self, clear_value: vk.ClearValue, clear_format: vk.Format, dest: *SoftImage, view_format: vk.Format, range: vk.ImageSubresourceRange, render_area: ?vk.Rect2D) bool {
+fn fastClear(self: *Self, clear_value: vk.ClearValue, clear_format: vk.Format, dest: *SoftImage, view_format: vk.Format, range: vk.ImageSubresourceRange, render_area: ?vk.Rect2D) VkError!bool {
     _ = self;
     _ = render_area;
     _ = range;
@@ -90,7 +88,7 @@ fn fastClear(self: *Self, clear_value: vk.ClearValue, clear_format: vk.Format, d
     }
 
     if (dest.interface.memory) |memory| {
-        const image_size = dest.interface.getTotalSize();
+        const image_size = try dest.interface.getTotalSize();
         const memory_map = memory.map(dest.interface.memory_offset, image_size) catch return false;
         defer memory.unmap();
 
@@ -102,3 +100,126 @@ fn fastClear(self: *Self, clear_value: vk.ClearValue, clear_format: vk.Format, d
     }
     return false;
 }
+
+pub fn blitRegion(_: *Self, src: *const SoftImage, dst: *SoftImage, region: vk.ImageBlit, filter: vk.Filter) VkError!void {
+    var dst_offset_0 = region.dst_offsets[0];
+    var dst_offset_1 = region.dst_offsets[1];
+    var src_offset_0 = region.src_offsets[0];
+    var src_offset_1 = region.src_offsets[1];
+
+    if (dst_offset_0.x > dst_offset_1.x) {
+        std.mem.swap(i32, &src_offset_0.x, &src_offset_1.x);
+        std.mem.swap(i32, &dst_offset_0.x, &dst_offset_1.x);
+    }
+
+    if (dst_offset_0.y > dst_offset_1.y) {
+        std.mem.swap(i32, &src_offset_0.y, &src_offset_1.y);
+        std.mem.swap(i32, &dst_offset_0.y, &dst_offset_1.y);
+    }
+
+    if (dst_offset_0.z > dst_offset_1.z) {
+        std.mem.swap(i32, &src_offset_0.z, &src_offset_1.z);
+        std.mem.swap(i32, &dst_offset_0.z, &dst_offset_1.z);
+    }
+
+    const src_extent = src.getMipLevelExtent(region.src_subresource.mip_level);
+
+    _ = src_extent;
+
+    const width_ratio = @as(f32, @floatFromInt(src_offset_1.x - src_offset_0.x)) / @as(f32, @floatFromInt(dst_offset_1.x - dst_offset_0.x));
+    const height_ratio = @as(f32, @floatFromInt(src_offset_1.y - src_offset_0.y)) / @as(f32, @floatFromInt(dst_offset_1.y - dst_offset_0.y));
+    const depth_ratio = @as(f32, @floatFromInt(src_offset_1.z - src_offset_0.z)) / @as(f32, @floatFromInt(dst_offset_1.z - dst_offset_0.z));
+    const x0 = @as(f32, @floatFromInt(src_offset_0.x)) + (0.5 - @as(f32, @floatFromInt(dst_offset_0.x))) * width_ratio;
+    const y0 = @as(f32, @floatFromInt(src_offset_0.y)) + (0.5 - @as(f32, @floatFromInt(dst_offset_0.y))) * height_ratio;
+    const z0 = @as(f32, @floatFromInt(src_offset_0.z)) + (0.5 - @as(f32, @floatFromInt(dst_offset_0.z))) * depth_ratio;
+
+    _ = x0;
+    _ = y0;
+    _ = z0;
+
+    const src_format = base.format.fromAspect(src.interface.format, region.src_subresource.aspect_mask);
+    const dst_format = base.format.fromAspect(dst.interface.format, region.dst_subresource.aspect_mask);
+
+    const apply_filter = (filter != .nearest);
+    const allow_srgb_conversion = apply_filter or base.format.isSrgb(src_format) != base.format.isSrgb(dst_format);
+
+    _ = allow_srgb_conversion;
+}
+
+// State state(srcFormat, dstFormat, src->getSampleCount(), dst->getSampleCount(),
+//             Options{ doFilter, allowSRGBConversion });
+// state.clampToEdge = (region.srcOffsets[0].x < 0) ||
+//                     (region.srcOffsets[0].y < 0) ||
+//                     (static_cast<uint32_t>(region.srcOffsets[1].x) > srcExtent.width) ||
+//                     (static_cast<uint32_t>(region.srcOffsets[1].y) > srcExtent.height) ||
+//                     (doFilter && ((x0 < 0.5f) || (y0 < 0.5f)));
+// state.filter3D = (region.srcOffsets[1].z - region.srcOffsets[0].z) !=
+//                  (region.dstOffsets[1].z - region.dstOffsets[0].z);
+//
+// auto blitRoutine = getBlitRoutine(state);
+// if(!blitRoutine)
+// {
+//     return;
+// }
+//
+// BlitData data = {
+//     nullptr,                                                                                 // source
+//     nullptr,                                                                                 // dest
+//     assert_cast<uint32_t>(src->rowPitchBytes(srcAspect, region.srcSubresource.mipLevel)),    // sPitchB
+//     assert_cast<uint32_t>(dst->rowPitchBytes(dstAspect, region.dstSubresource.mipLevel)),    // dPitchB
+//     assert_cast<uint32_t>(src->slicePitchBytes(srcAspect, region.srcSubresource.mipLevel)),  // sSliceB
+//     assert_cast<uint32_t>(dst->slicePitchBytes(dstAspect, region.dstSubresource.mipLevel)),  // dSliceB
+//
+//     x0,
+//     y0,
+//     z0,
+//     widthRatio,
+//     heightRatio,
+//     depthRatio,
+//
+//     region.dstOffsets[0].x,  // x0d
+//     region.dstOffsets[1].x,  // x1d
+//     region.dstOffsets[0].y,  // y0d
+//     region.dstOffsets[1].y,  // y1d
+//     region.dstOffsets[0].z,  // z0d
+//     region.dstOffsets[1].z,  // z1d
+//
+//     static_cast<int>(srcExtent.width),   // sWidth
+//     static_cast<int>(srcExtent.height),  // sHeight
+//     static_cast<int>(srcExtent.depth),   // sDepth
+//
+//     false,  // filter3D
+// };
+//
+// VkImageSubresource srcSubres = {
+//     region.srcSubresource.aspectMask,
+//     region.srcSubresource.mipLevel,
+//     region.srcSubresource.baseArrayLayer
+// };
+//
+// VkImageSubresource dstSubres = {
+//     region.dstSubresource.aspectMask,
+//     region.dstSubresource.mipLevel,
+//     region.dstSubresource.baseArrayLayer
+// };
+//
+// VkImageSubresourceRange dstSubresRange = {
+//     region.dstSubresource.aspectMask,
+//     region.dstSubresource.mipLevel,
+//     1,  // levelCount
+//     region.dstSubresource.baseArrayLayer,
+//     region.dstSubresource.layerCount
+// };
+//
+// uint32_t lastLayer = src->getLastLayerIndex(dstSubresRange);
+//
+// for(; dstSubres.arrayLayer <= lastLayer; srcSubres.arrayLayer++, dstSubres.arrayLayer++)
+// {
+//     data.source = src->getTexelPointer({ 0, 0, 0 }, srcSubres);
+//     data.dest = dst->getTexelPointer({ 0, 0, 0 }, dstSubres);
+//
+//     ASSERT(data.source < src->end());
+//     ASSERT(data.dest < dst->end());
+//
+//     blitRoutine(&data);
+// }

src/soft/device/ComputeRoutines.zig

@@ -29,11 +29,19 @@ device: *SoftDevice,
 state: *PipelineState,
 batch_size: usize,
 
+invocation_index: std.atomic.Value(usize),
+
+early_dump: ?u32,
+final_dump: ?u32,
+
 pub fn init(device: *SoftDevice, state: *PipelineState) Self {
     return .{
         .device = device,
         .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,
     };
 }
 
@@ -51,9 +59,11 @@ pub fn dispatch(self: *Self, group_count_x: u32, group_count_y: u32, group_count
 
     const invocations_per_workgroup = spv_module.local_size_x * spv_module.local_size_y * spv_module.local_size_z;
 
-    var wg: std.Thread.WaitGroup = .{};
+    self.invocation_index.store(0, .monotonic);
+
+    var wg: std.Io.Group = .init;
     for (0..@min(self.batch_size, group_count)) |batch_id| {
-        if (std.process.hasEnvVarConstant(lib.SINGLE_THREAD_COMPUTE_EXECUTION_ENV_NAME)) {
+        if (base.config.single_threaded_compute) {
             @branchHint(.cold); // Should only be reached for debugging
 
             runWrapper(
@@ -69,7 +79,7 @@ pub fn dispatch(self: *Self, group_count_x: u32, group_count_y: u32, group_count
                 },
             );
         } else {
-            self.device.workers.spawnWg(&wg, runWrapper, .{
+            wg.async(self.device.interface.io(), runWrapper, .{
                 RunData{
                     .self = self,
                     .batch_id = batch_id,
@@ -83,20 +93,21 @@ pub fn dispatch(self: *Self, group_count_x: u32, group_count_y: u32, group_count
             });
         }
     }
-    self.device.workers.waitAndWork(&wg);
+    wg.await(self.device.interface.io()) catch return VkError.DeviceLost;
 }
 
 fn runWrapper(data: RunData) void {
     @call(.always_inline, run, .{data}) catch |err| {
         std.log.scoped(.@"SPIR-V runtime").err("SPIR-V runtime catched a '{s}'", .{@errorName(err)});
         if (@errorReturnTrace()) |trace| {
-            std.debug.dumpStackTrace(trace.*);
+            std.debug.dumpErrorReturnTrace(trace);
         }
     };
 }
 
 inline fn run(data: RunData) !void {
     const allocator = data.self.device.device_allocator.allocator();
+    const io = data.self.device.interface.io();
 
     const shader = data.pipeline.stages.getPtrAssertContains(.compute);
     const rt = &shader.runtimes[data.batch_id];
@@ -128,12 +139,19 @@ inline fn run(data: RunData) !void {
         });
 
         for (0..data.invocations_per_workgroup) |i| {
+            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)),
             }, i);
 
+            if (data.self.early_dump != null and data.self.early_dump.? == invocation_index) {
+                @branchHint(.cold);
+                try dumpResultsTable(allocator, io, rt, true);
+            }
+
             rt.callEntryPoint(allocator, entry) catch |err| switch (err) {
                 // Some errors can be ignored
                 SpvRuntimeError.OutOfBounds,
@@ -141,11 +159,30 @@ inline fn run(data: RunData) !void {
                 => {},
                 else => return err,
             };
+
+            if (data.self.final_dump != null and data.self.final_dump.? == invocation_index) {
+                @branchHint(.cold);
+                try dumpResultsTable(allocator, io, rt, false);
+            }
+
             try rt.flushDescriptorSets(allocator);
         }
     }
 }
 
+inline fn dumpResultsTable(allocator: std.mem.Allocator, io: std.Io, rt: *spv.Runtime, is_early: bool) !void {
+    @branchHint(.cold);
+    const file = try std.Io.Dir.cwd().createFile(
+        io,
+        std.fmt.comptimePrint("{s}_compute_result_table_dump.txt", .{if (is_early) "early" else "final"}),
+        .{ .truncate = true },
+    );
+    defer file.close(io);
+    var buffer = [_]u8{0} ** 1024;
+    var writer = file.writer(io, buffer[0..]);
+    try rt.dumpResultsTable(allocator, &writer.interface);
+}
+
 fn writeDescriptorSets(self: *Self, rt: *spv.Runtime) !void {
     sets: for (self.state.sets[0..], 0..) |set, set_index| {
         if (set == null)
@@ -165,6 +202,18 @@ fn writeDescriptorSets(self: *Self, rt: *spv.Runtime) !void {
                         );
                     }
                 },
+                .image => |image_data_array| for (image_data_array, 0..) |image_data, descriptor_index| {
+                    if (image_data.object) |image| {
+                        const memory = if (image.interface.memory) |memory| memory else continue :bindings;
+                        const map: []u8 = @as([*]u8, @ptrCast(try memory.map(image.interface.memory_offset, try image.interface.getTotalSize())))[0..try image.interface.getTotalSize()];
+                        try rt.writeDescriptorSet(
+                            map,
+                            @as(u32, @intCast(set_index)),
+                            @as(u32, @intCast(binding_index)),
+                            @as(u32, @intCast(descriptor_index)),
+                        );
+                    }
+                },
                 else => {},
             }
         }

src/soft/device/Device.zig

@@ -6,6 +6,7 @@ const SoftDescriptorSet = @import("../SoftDescriptorSet.zig");
 const SoftDevice = @import("../SoftDevice.zig");
 const SoftPipeline = @import("../SoftPipeline.zig");
 
+const Blitter = @import("Blitter.zig");
 const ComputeRoutines = @import("ComputeRoutines.zig");
 const PipelineState = @import("PipelineState.zig");
 
@@ -13,6 +14,8 @@ const VkError = base.VkError;
 
 const Self = @This();
 
+blitter: Blitter,
+
 compute_routines: ComputeRoutines,
 
 /// .graphics = 0
@@ -20,6 +23,7 @@ compute_routines: ComputeRoutines,
 pipeline_states: [2]PipelineState,
 
 pub const init: Self = .{
+    .blitter = .init,
     .compute_routines = undefined,
     .pipeline_states = undefined,
 };

src/soft/lib.zig

@@ -40,9 +40,6 @@ pub const VULKAN_VERSION = vk.makeApiVersion(0, 1, 0, 0);
 pub const DRIVER_VERSION = vk.makeApiVersion(0, 0, 0, 1);
 pub const DEVICE_ID = 0x600DCAFE;
 
-pub const NO_SHADER_SIMD_ENV_NAME = "STROLL_SOFT_NO_SIMD";
-pub const SINGLE_THREAD_COMPUTE_EXECUTION_ENV_NAME = "STROLL_SINGLE_THREAD_COMPUTE_EXECUTION";
-
 /// Generic system memory.
 pub const MEMORY_TYPE_GENERIC_BIT = 0;
 

src/vulkan/CommandBuffer.zig

@@ -31,7 +31,7 @@ pool: *CommandPool,
 state: State,
 begin_info: ?vk.CommandBufferBeginInfo,
 host_allocator: VulkanAllocator,
-state_mutex: std.Thread.Mutex,
+state_mutex: std.Io.Mutex,
 
 vtable: *const VTable,
 dispatch_table: *const DispatchTable,
@@ -40,6 +40,7 @@ pub const DispatchTable = struct {
     begin: *const fn (*Self, *const vk.CommandBufferBeginInfo) VkError!void,
     bindDescriptorSets: *const fn (*Self, vk.PipelineBindPoint, u32, [lib.VULKAN_MAX_DESCRIPTOR_SETS]?*DescriptorSet, []const u32) VkError!void,
     bindPipeline: *const fn (*Self, vk.PipelineBindPoint, *Pipeline) VkError!void,
+    blitImage: *const fn (*Self, *Image, vk.ImageLayout, *Image, vk.ImageLayout, []const vk.ImageBlit, vk.Filter) VkError!void,
     clearColorImage: *const fn (*Self, *Image, vk.ImageLayout, *const vk.ClearColorValue, vk.ImageSubresourceRange) VkError!void,
     copyBuffer: *const fn (*Self, *Buffer, *Buffer, []const vk.BufferCopy) VkError!void,
     copyBufferToImage: *const fn (*Self, *Buffer, *Image, vk.ImageLayout, []const vk.BufferImageCopy) VkError!void,
@@ -48,6 +49,7 @@ pub const DispatchTable = struct {
     dispatch: *const fn (*Self, u32, u32, u32) VkError!void,
     dispatchIndirect: *const fn (*Self, *Buffer, vk.DeviceSize) VkError!void,
     end: *const fn (*Self) VkError!void,
+    executeCommands: *const fn (*Self, *Self) VkError!void,
     fillBuffer: *const fn (*Self, *Buffer, vk.DeviceSize, vk.DeviceSize, u32) VkError!void,
     pipelineBarrier: *const fn (*Self, vk.PipelineStageFlags, vk.PipelineStageFlags, vk.DependencyFlags, []const vk.MemoryBarrier, []const vk.BufferMemoryBarrier, []const vk.ImageMemoryBarrier) VkError!void,
     reset: *const fn (*Self, vk.CommandBufferResetFlags) VkError!void,
@@ -67,7 +69,7 @@ pub fn init(device: *Device, allocator: std.mem.Allocator, info: *const vk.Comma
         .state = .Initial,
         .begin_info = null,
         .host_allocator = VulkanAllocator.from(allocator).cloneWithScope(.object),
-        .state_mutex = .{},
+        .state_mutex = .init,
         .vtable = undefined,
         .dispatch_table = undefined,
     };
@@ -77,8 +79,11 @@ inline fn transitionState(self: *Self, target: State, from_allowed: []const Stat
     if (!std.EnumSet(State).initMany(from_allowed).contains(self.state)) {
         return error.NotAllowed;
     }
-    self.state_mutex.lock();
+    const io = self.owner.io();
-    defer self.state_mutex.unlock();
+
+    self.state_mutex.lockUncancelable(io);
+    defer self.state_mutex.unlock(io);
+
     self.state = target;
 }
 
@@ -90,7 +95,8 @@ pub inline fn begin(self: *Self, info: *const vk.CommandBufferBeginInfo) VkError
     if (!self.pool.flags.reset_command_buffer_bit) {
         self.transitionState(.Recording, &.{.Initial}) catch return VkError.ValidationFailed;
     } else {
-        self.transitionState(.Recording, &.{ .Initial, .Executable, .Invalid }) catch return VkError.ValidationFailed;
+        try self.reset(.{});
+        self.transitionState(.Recording, &.{ .Initial, .Recording, .Executable, .Invalid }) catch return VkError.ValidationFailed;
     }
     try self.dispatch_table.begin(self, info);
     self.begin_info = info.*;
@@ -99,6 +105,7 @@ pub inline fn begin(self: *Self, info: *const vk.CommandBufferBeginInfo) VkError
 pub inline fn end(self: *Self) VkError!void {
     self.transitionState(.Executable, &.{.Recording}) catch return VkError.ValidationFailed;
     try self.dispatch_table.end(self);
+    self.begin_info = null;
 }
 
 pub inline fn reset(self: *Self, flags: vk.CommandBufferResetFlags) VkError!void {
@@ -133,6 +140,7 @@ pub inline fn finish(self: *Self) VkError!void {
 // Commands ====================================================================================================
 
 pub inline fn bindDescriptorSets(self: *Self, bind_point: vk.PipelineBindPoint, first_set: u32, sets: []const vk.DescriptorSet, dynamic_offsets: []const u32) VkError!void {
+    std.debug.assert(sets.len < lib.VULKAN_MAX_DESCRIPTOR_SETS);
     var inner_sets = [_]?*DescriptorSet{null} ** lib.VULKAN_MAX_DESCRIPTOR_SETS;
     for (sets, inner_sets[0..sets.len]) |set, *inner_set| {
         inner_set.* = try NonDispatchable(DescriptorSet).fromHandleObject(set);
@@ -144,6 +152,10 @@ pub inline fn bindPipeline(self: *Self, bind_point: vk.PipelineBindPoint, pipeli
     try self.dispatch_table.bindPipeline(self, bind_point, pipeline);
 }
 
+pub inline fn blitImage(self: *Self, src: *Image, src_layout: vk.ImageLayout, dst: *Image, dst_layout: vk.ImageLayout, regions: []const vk.ImageBlit, filter: vk.Filter) VkError!void {
+    try self.dispatch_table.blitImage(self, src, src_layout, dst, dst_layout, regions, filter);
+}
+
 pub inline fn clearColorImage(self: *Self, image: *Image, layout: vk.ImageLayout, color: *const vk.ClearColorValue, ranges: []const vk.ImageSubresourceRange) VkError!void {
     for (ranges) |range| {
         try self.dispatch_table.clearColorImage(self, image, layout, color, range);
@@ -174,6 +186,10 @@ pub inline fn dispatchIndirect(self: *Self, buffer: *Buffer, offset: vk.DeviceSi
     try self.dispatch_table.dispatchIndirect(self, buffer, offset);
 }
 
+pub inline fn executeCommands(self: *Self, commands: *Self) VkError!void {
+    try self.dispatch_table.executeCommands(self, commands);
+}
+
 pub inline fn fillBuffer(self: *Self, buffer: *Buffer, offset: vk.DeviceSize, size: vk.DeviceSize, data: u32) VkError!void {
     try self.dispatch_table.fillBuffer(self, buffer, offset, size, data);
 }

src/vulkan/CommandPool.zig

@@ -72,9 +72,8 @@ pub fn freeCommandBuffers(self: *Self, cmds: []*Dispatchable(CommandBuffer)) VkE
     // Ugly method but it works well
     var len: usize = 0;
     for (cmds) |cmd| {
-        if (std.mem.indexOf(*Dispatchable(CommandBuffer), self.buffers.items, &[_]*Dispatchable(CommandBuffer){cmd})) |i| {
+        if (std.mem.indexOfScalar(*Dispatchable(CommandBuffer), self.buffers.items, cmd)) |i| {
             const save = self.buffers.orderedRemove(i);
-            // Append the now free command buffer at the end of the pool
             self.buffers.appendAssumeCapacity(save);
             len += 1;
         }
@@ -82,8 +81,6 @@ pub fn freeCommandBuffers(self: *Self, cmds: []*Dispatchable(CommandBuffer)) VkE
     const new_first_free_buffer_index, const has_overflown = @subWithOverflow(self.first_free_buffer_index, len);
     if (has_overflown == 0) {
         self.first_free_buffer_index = new_first_free_buffer_index;
-    } else {
-        std.log.scoped(.CommandPool).warn("Avoided an underflow. This should not happen, please fill an issue.", .{});
     }
 }
 

src/vulkan/DescriptorPool.zig

@@ -21,6 +21,7 @@ pub const VTable = struct {
     allocateDescriptorSet: *const fn (*Self, *DescriptorSetLayout) VkError!*DescriptorSet,
     destroy: *const fn (*Self, std.mem.Allocator) void,
     freeDescriptorSet: *const fn (*Self, *DescriptorSet) VkError!void,
+    reset: *const fn (*Self, vk.DescriptorPoolResetFlags) VkError!void,
 };
 
 pub fn init(device: *Device, allocator: std.mem.Allocator, info: *const vk.DescriptorPoolCreateInfo) VkError!Self {
@@ -43,3 +44,7 @@ pub inline fn destroy(self: *Self, allocator: std.mem.Allocator) void {
 pub inline fn freeDescriptorSet(self: *Self, set: *DescriptorSet) VkError!void {
     try self.vtable.freeDescriptorSet(self, set);
 }
+
+pub inline fn reset(self: *Self, flags: vk.DescriptorPoolResetFlags) VkError!void {
+    try self.vtable.reset(self, flags);
+}

src/vulkan/Device.zig

@@ -1,8 +1,6 @@
 const std = @import("std");
 const vk = @import("vulkan");
 
-const logger = @import("lib.zig").logger;
-
 const Dispatchable = @import("Dispatchable.zig").Dispatchable;
 const NonDispatchable = @import("NonDispatchable.zig").NonDispatchable;
 const VulkanAllocator = @import("VulkanAllocator.zig");
@@ -23,6 +21,7 @@ const Event = @import("Event.zig");
 const Fence = @import("Fence.zig");
 const Framebuffer = @import("Framebuffer.zig");
 const Image = @import("Image.zig");
+const Instance = @import("Instance.zig");
 const ImageView = @import("ImageView.zig");
 const Pipeline = @import("Pipeline.zig");
 const PipelineCache = @import("PipelineCache.zig");
@@ -35,6 +34,7 @@ const ShaderModule = @import("ShaderModule.zig");
 const Self = @This();
 pub const ObjectType: vk.ObjectType = .device;
 
+instance: *Instance,
 physical_device: *const PhysicalDevice,
 queues: std.AutoArrayHashMapUnmanaged(u32, std.ArrayList(*Dispatchable(Queue))),
 host_allocator: VulkanAllocator,
@@ -71,9 +71,10 @@ pub const DispatchTable = struct {
     destroy: *const fn (*Self, std.mem.Allocator) VkError!void,
 };
 
-pub fn init(allocator: std.mem.Allocator, physical_device: *const PhysicalDevice, info: *const vk.DeviceCreateInfo) VkError!Self {
+pub fn init(allocator: std.mem.Allocator, instance: *Instance, physical_device: *const PhysicalDevice, info: *const vk.DeviceCreateInfo) VkError!Self {
     _ = info;
     return .{
+        .instance = instance,
         .physical_device = physical_device,
         .queues = .empty,
         .host_allocator = VulkanAllocator.from(allocator).clone(),
@@ -99,14 +100,15 @@ pub fn createQueues(self: *Self, allocator: std.mem.Allocator, info: *const vk.D
 
         const queue = try self.vtable.createQueue(allocator, self, queue_info.queue_family_index, @intCast(family_ptr.items.len), queue_info.flags);
 
-        logger.getManager().get().indent();
-        defer logger.getManager().get().unindent();
-
         const dispatchable_queue = try Dispatchable(Queue).wrap(allocator, queue);
         family_ptr.append(allocator, dispatchable_queue) catch return VkError.OutOfHostMemory;
     }
 }
 
+pub fn io(self: *const Self) std.Io {
+    return self.instance.io();
+}
+
 pub inline fn destroy(self: *Self, allocator: std.mem.Allocator) VkError!void {
     var it = self.queues.iterator();
     while (it.next()) |entry| {

src/vulkan/Image.zig

@@ -18,6 +18,7 @@ extent: vk.Extent3D,
 mip_levels: u32,
 array_layers: u32,
 samples: vk.SampleCountFlags,
+flags: vk.ImageCreateFlags,
 tiling: vk.ImageTiling,
 usage: vk.ImageUsageFlags,
 memory: ?*DeviceMemory,
@@ -28,7 +29,8 @@ vtable: *const VTable,
 
 pub const VTable = struct {
     destroy: *const fn (*Self, std.mem.Allocator) void,
-    getMemoryRequirements: *const fn (*Self, *vk.MemoryRequirements) void,
+    getMemoryRequirements: *const fn (*Self, *vk.MemoryRequirements) VkError!void,
+    getTotalSizeForAspect: *const fn (*const Self, vk.ImageAspectFlags) VkError!usize,
 };
 
 pub fn init(device: *Device, allocator: std.mem.Allocator, info: *const vk.ImageCreateInfo) VkError!Self {
@@ -41,6 +43,7 @@ pub fn init(device: *Device, allocator: std.mem.Allocator, info: *const vk.Image
         .mip_levels = info.mip_levels,
         .array_layers = info.array_layers,
         .samples = info.samples,
+        .flags = info.flags,
         .tiling = info.tiling,
         .usage = info.usage,
         .memory = null,
@@ -55,7 +58,7 @@ pub inline fn destroy(self: *Self, allocator: std.mem.Allocator) void {
 }
 
 pub inline fn bindMemory(self: *Self, memory: *DeviceMemory, offset: vk.DeviceSize) VkError!void {
-    const image_size = self.getTotalSize();
+    const image_size = try self.getTotalSize();
     if (offset >= image_size or !self.allowed_memory_types.isSet(memory.memory_type_index)) {
         return VkError.ValidationFailed;
     }
@@ -63,94 +66,28 @@ pub inline fn bindMemory(self: *Self, memory: *DeviceMemory, offset: vk.DeviceSi
     self.memory_offset = offset;
 }
 
-pub inline fn getMemoryRequirements(self: *Self, requirements: *vk.MemoryRequirements) void {
+pub inline fn getMemoryRequirements(self: *Self, requirements: *vk.MemoryRequirements) VkError!void {
-    const image_size = self.getTotalSize();
+    requirements.size = try self.getTotalSize();
-    requirements.size = image_size;
     requirements.memory_type_bits = self.allowed_memory_types.mask;
-    self.vtable.getMemoryRequirements(self, requirements);
+    try self.vtable.getMemoryRequirements(self, requirements);
 }
 
-pub inline fn getClearFormat(self: *Self) vk.Format {
+pub inline fn getTexelSize(self: *const Self) usize {
-    return if (lib.vku.vkuFormatIsSINT(@intCast(@intFromEnum(self.format))))
+    return lib.format.texelSize(self.format);
-        .r32g32b32a32_sint
-    else if (lib.vku.vkuFormatIsUINT(@intCast(@intFromEnum(self.format))))
-        .r32g32b32a32_uint
-    else
-        .r32g32b32a32_sfloat;
 }
 
-pub inline fn getPixelSize(self: *const Self) usize {
+pub inline fn getTotalSizeForAspect(self: *const Self, aspect: vk.ImageAspectFlags) VkError!usize {
-    return lib.vku.vkuFormatTexelBlockSize(@intCast(@intFromEnum(self.format)));
+    return self.vtable.getTotalSizeForAspect(self, aspect);
 }
 
-pub inline fn getTotalSize(self: *const Self) usize {
+pub inline fn getTotalSize(self: *const Self) VkError!usize {
-    const pixel_size = self.getPixelSize();
+    return self.vtable.getTotalSizeForAspect(self, lib.format.toAspect(self.format));
-    return self.extent.width * self.extent.height * self.extent.depth * pixel_size;
 }
 
-pub inline fn getFormatPixelSize(format: vk.Format) usize {
+pub inline fn formatFromAspect(self: *const Self, aspect_mask: vk.ImageAspectFlags) vk.Format {
-    return lib.vku.vkuFormatTexelBlockSize(@intCast(@intFromEnum(format)));
+    return lib.format.fromAspect(self.format, aspect_mask);
 }
 
-pub inline fn getFormatTotalSize(self: *const Self, format: vk.Format) usize {
+pub inline fn formatToAspect(self: *const Self, aspect_mask: vk.ImageAspectFlags) vk.ImageAspectFlags {
-    const pixel_size = self.getFormatPixelSize(format);
+    return lib.format.toAspect(self.format, aspect_mask);
-    return self.extent.width * self.extent.height * self.extent.depth * pixel_size;
-}
-
-pub fn formatSupportsColorAttachemendBlend(format: vk.Format) bool {
-    return switch (format) {
-        // Vulkan 1.1 mandatory
-        .r5g6b5_unorm_pack16,
-        .a1r5g5b5_unorm_pack16,
-        .r8_unorm,
-        .r8g8_unorm,
-        .r8g8b8a8_unorm,
-        .r8g8b8a8_srgb,
-        .b8g8r8a8_unorm,
-        .b8g8r8a8_srgb,
-        .a8b8g8r8_unorm_pack32,
-        .a8b8g8r8_srgb_pack32,
-        .a2b10g10r10_unorm_pack32,
-        .r16_sfloat,
-        .r16g16_sfloat,
-        .r16g16b16a16_sfloat,
-        // optional
-        .r4g4b4a4_unorm_pack16,
-        .b4g4r4a4_unorm_pack16,
-        .b5g6r5_unorm_pack16,
-        .r5g5b5a1_unorm_pack16,
-        .b5g5r5a1_unorm_pack16,
-        .a2r10g10b10_unorm_pack32,
-        .r16_unorm,
-        .r16g16_unorm,
-        .r16g16b16a16_unorm,
-        .r32_sfloat,
-        .r32g32_sfloat,
-        .r32g32b32a32_sfloat,
-        .b10g11r11_ufloat_pack32,
-        .a4r4g4b4_unorm_pack16,
-        .a4b4g4r4_unorm_pack16,
-        => true,
-        else => false,
-    };
-}
-
-pub fn formatFromAspect(base_format: vk.Format, aspect: vk.ImageAspectFlags) vk.Format {
-    if (aspect.color_bit or (aspect.color_bit and aspect.stencil_bit)) {
-        return base_format;
-    } else if (aspect.depth_bit) {
-        if (base_format == .d16_unorm or base_format == .d16_unorm_s8_uint) {
-            return .d16_unorm;
-        } else if (base_format == .d24_unorm_s8_uint) {
-            return .x8_d24_unorm_pack32;
-        } else if (base_format == .d32_sfloat or base_format == .d32_sfloat_s8_uint) {
-            return .d32_sfloat;
-        }
-    } else if (aspect.stencil_bit) {
-        if (base_format == .s8_uint or base_format == .d16_unorm_s8_uint or base_format == .d24_unorm_s8_uint or base_format == .d32_sfloat_s8_uint) {
-            return .s8_uint;
-        }
-    }
-    lib.unsupported("format {d}", .{@intFromEnum(base_format)});
-    return base_format;
 }

src/vulkan/Instance.zig

@@ -1,8 +1,7 @@
 const std = @import("std");
 const builtin = @import("builtin");
 const vk = @import("vulkan");
-
+const config = @import("lib.zig").config;
-const logger = @import("lib.zig").logger;
 
 const VkError = @import("error_set.zig").VkError;
 const Dispatchable = @import("Dispatchable.zig").Dispatchable;
@@ -21,6 +20,12 @@ comptime {
 const Self = @This();
 pub const ObjectType: vk.ObjectType = .instance;
 
+const DeviceAllocator = struct {
+    pub inline fn allocator(_: @This()) std.mem.Allocator {
+        return std.heap.smp_allocator;
+    }
+};
+
 physical_devices: std.ArrayList(*Dispatchable(PhysicalDevice)),
 dispatch_table: *const DispatchTable,
 vtable: *const VTable,
@@ -28,6 +33,7 @@ vtable: *const VTable,
 pub const VTable = struct {
     releasePhysicalDevices: *const fn (*Self, std.mem.Allocator) VkError!void,
     requestPhysicalDevices: *const fn (*Self, std.mem.Allocator) VkError!void,
+    io: *const fn (*Self) std.Io,
 };
 
 pub const DispatchTable = struct {
@@ -85,9 +91,6 @@ pub fn releasePhysicalDevices(self: *Self, allocator: std.mem.Allocator) VkError
 }
 
 pub fn requestPhysicalDevices(self: *Self, allocator: std.mem.Allocator) VkError!void {
-    logger.getManager().get().indent();
-    defer logger.getManager().get().unindent();
-
     try self.vtable.requestPhysicalDevices(self, allocator);
     if (self.physical_devices.items.len == 0) {
         std.log.scoped(.vkCreateInstance).err("No VkPhysicalDevice found", .{});
@@ -97,3 +100,7 @@ pub fn requestPhysicalDevices(self: *Self, allocator: std.mem.Allocator) VkError
         std.log.scoped(.vkCreateInstance).debug("Found VkPhysicalDevice named {s}", .{physical_device.object.props.device_name});
     }
 }
+
+pub fn io(self: *Self) std.Io {
+    return self.vtable.io(self);
+}

src/vulkan/PhysicalDevice.zig

@@ -13,7 +13,7 @@ props: vk.PhysicalDeviceProperties,
 mem_props: vk.PhysicalDeviceMemoryProperties,
 features: vk.PhysicalDeviceFeatures,
 queue_family_props: std.ArrayList(vk.QueueFamilyProperties),
-instance: *const Instance,
+instance: *Instance,
 dispatch_table: *const DispatchTable,
 
 pub const DispatchTable = struct {
@@ -25,7 +25,7 @@ pub const DispatchTable = struct {
     release: *const fn (*Self, std.mem.Allocator) VkError!void,
 };
 
-pub fn init(allocator: std.mem.Allocator, instance: *const Instance) VkError!Self {
+pub fn init(allocator: std.mem.Allocator, instance: *Instance) VkError!Self {
     _ = allocator;
     return .{
         .props = .{

src/vulkan/Queue.zig

@@ -53,9 +53,8 @@ pub const SubmitInfo = struct {
     }
 
     fn deinitBlob(allocator: std.mem.Allocator, self: *std.ArrayList(SubmitInfo)) void {
-        for (self.items) |submit_info| {
+        for (self.items) |*submit_info| {
-            const command_buffers = &submit_info.command_buffers;
+            submit_info.command_buffers.deinit(allocator);
-            @constCast(command_buffers).deinit(allocator);
         }
         self.deinit(allocator);
     }

src/vulkan/VulkanAllocator.zig

@@ -9,16 +9,22 @@ const Alignment = std.mem.Alignment;
 
 const Self = @This();
 
+const FallbackAllocator = struct {
+    pub inline fn allocator(_: @This()) std.mem.Allocator {
+        return std.heap.smp_allocator;
+    }
+};
+
 callbacks: ?vk.AllocationCallbacks,
 scope: vk.SystemAllocationScope,
-fallback_allocator: std.heap.ThreadSafeAllocator,
+fallback_allocator: FallbackAllocator,
 
 pub fn init(callbacks: ?*const vk.AllocationCallbacks, scope: vk.SystemAllocationScope) Self {
     const deref_callbacks = if (callbacks) |c| c.* else null;
     return .{
         .callbacks = deref_callbacks,
         .scope = scope,
-        .fallback_allocator = .{ .child_allocator = std.heap.c_allocator },
+        .fallback_allocator = .{},
     };
 }
 

src/vulkan/error_set.zig

@@ -50,7 +50,7 @@ pub const VkError = error{
     IncompatibleShaderBinaryExt,
     PipelineBinaryMissingKhr,
     NotEnoughSpaceKhr,
-    // ====== Internal errors
+    // == Set of internal errors for better debugging. They map to VK_UNKNOWN_ERROR
     InvalidHandleDrv,
     InvalidPipelineDrv,
     InvalidDeviceMemoryDrv,

src/vulkan/format.zig

@@ -0,0 +1,120 @@
+const std = @import("std");
+const vk = @import("vulkan");
+const lib = @import("lib.zig");
+
+pub fn fromAspect(format: vk.Format, aspect: vk.ImageAspectFlags) vk.Format {
+    if (aspect.color_bit or (aspect.color_bit and aspect.stencil_bit)) {
+        return format;
+    } else if (aspect.depth_bit) {
+        if (format == .d16_unorm or format == .d16_unorm_s8_uint) {
+            return .d16_unorm;
+        } else if (format == .d24_unorm_s8_uint) {
+            return .x8_d24_unorm_pack32;
+        } else if (format == .d32_sfloat or format == .d32_sfloat_s8_uint) {
+            return .d32_sfloat;
+        }
+    } else if (aspect.stencil_bit) {
+        if (format == .s8_uint or format == .d16_unorm_s8_uint or format == .d24_unorm_s8_uint or format == .d32_sfloat_s8_uint) {
+            return .s8_uint;
+        }
+    }
+    lib.unsupported("format {s}", .{@tagName(format)});
+    return format;
+}
+
+pub fn toAspect(format: vk.Format) vk.ImageAspectFlags {
+    var aspect: vk.ImageAspectFlags = .{};
+    if (lib.vku.vkuFormatHasDepth(@intCast(@intFromEnum(format))))
+        aspect.depth_bit = true;
+    if (lib.vku.vkuFormatHasStencil(@intCast(@intFromEnum(format))))
+        aspect.stencil_bit = true;
+
+    if (aspect.toInt() == 0)
+        aspect.color_bit = true;
+
+    return aspect;
+}
+
+pub inline fn texelSize(format: vk.Format) usize {
+    return lib.vku.vkuFormatTexelBlockSize(@intCast(@intFromEnum(format)));
+}
+
+pub inline fn supportsColorAttachemendBlend(format: vk.Format) bool {
+    return switch (format) {
+        // Vulkan 1.1 mandatory
+        .r5g6b5_unorm_pack16,
+        .a1r5g5b5_unorm_pack16,
+        .r8_unorm,
+        .r8g8_unorm,
+        .r8g8b8a8_unorm,
+        .r8g8b8a8_srgb,
+        .b8g8r8a8_unorm,
+        .b8g8r8a8_srgb,
+        .a8b8g8r8_unorm_pack32,
+        .a8b8g8r8_srgb_pack32,
+        .a2b10g10r10_unorm_pack32,
+        .r16_sfloat,
+        .r16g16_sfloat,
+        .r16g16b16a16_sfloat,
+        // optional
+        .r4g4b4a4_unorm_pack16,
+        .b4g4r4a4_unorm_pack16,
+        .b5g6r5_unorm_pack16,
+        .r5g5b5a1_unorm_pack16,
+        .b5g5r5a1_unorm_pack16,
+        .a2r10g10b10_unorm_pack32,
+        .r16_unorm,
+        .r16g16_unorm,
+        .r16g16b16a16_unorm,
+        .r32_sfloat,
+        .r32g32_sfloat,
+        .r32g32b32a32_sfloat,
+        .b10g11r11_ufloat_pack32,
+        .a4r4g4b4_unorm_pack16,
+        .a4b4g4r4_unorm_pack16,
+        => true,
+        else => false,
+    };
+}
+
+pub inline fn pitchMemSize(format: vk.Format, width: usize) usize {
+    // To be updated for compressed formats handling
+    return texelSize(format) * width;
+}
+
+pub inline fn sliceMemSize(format: vk.Format, width: usize, height: usize) usize {
+    // To be updated for compressed formats handling
+    return pitchMemSize(format, width) * height;
+}
+
+pub inline fn isDepthAndStencil(format: vk.Format) bool {
+    return lib.vku.vkuFormatIsDepthAndStencil(@intCast(@intFromEnum(format)));
+}
+
+pub inline fn isSrgb(format: vk.Format) bool {
+    return lib.vku.vkuFormatIsSRGB(@intCast(@intFromEnum(format)));
+}
+
+pub inline fn isSfloat(format: vk.Format) bool {
+    return lib.vku.vkuFormatIsSFLOAT(@intCast(@intFromEnum(format)));
+}
+
+pub inline fn isSint(format: vk.Format) bool {
+    return lib.vku.vkuFormatIsSINT(@intCast(@intFromEnum(format)));
+}
+
+pub inline fn isSnorm(format: vk.Format) bool {
+    return lib.vku.vkuFormatIsSNORM(@intCast(@intFromEnum(format)));
+}
+
+pub inline fn isUfloat(format: vk.Format) bool {
+    return lib.vku.vkuFormatIsUFLOAT(@intCast(@intFromEnum(format)));
+}
+
+pub inline fn isUint(format: vk.Format) bool {
+    return lib.vku.vkuFormatIsUINT(@intCast(@intFromEnum(format)));
+}
+
+pub inline fn isUnorm(format: vk.Format) bool {
+    return lib.vku.vkuFormatIsUNORM(@intCast(@intFromEnum(format)));
+}

src/vulkan/lib.zig

@@ -9,7 +9,9 @@ pub const vku = @cImport({
 
 pub const errors = @import("error_set.zig");
 pub const lib_vulkan = @import("lib_vulkan.zig");
-pub const logger = @import("logger/logger.zig");
+pub const logger = @import("logger.zig");
+pub const format = @import("format.zig");
+pub const config = @import("config");
 
 pub const Dispatchable = @import("Dispatchable.zig").Dispatchable;
 pub const NonDispatchable = @import("NonDispatchable.zig").NonDispatchable;
@@ -83,27 +85,6 @@ pub const std_options: std.Options = .{
     .logFn = logger.log,
 };
 
-pub const LogVerboseLevel = enum {
-    None,
-    Standard,
-    High,
-    TooMuch,
-};
-
-pub inline fn getLogVerboseLevel() LogVerboseLevel {
-    const allocator = std.heap.c_allocator;
-    const level = std.process.getEnvVarOwned(allocator, DRIVER_LOGS_ENV_NAME) catch return .None;
-    defer allocator.free(level);
-    return if (std.mem.eql(u8, level, "none"))
-        .None
-    else if (std.mem.eql(u8, level, "all"))
-        .High
-    else if (std.mem.eql(u8, level, "stupid"))
-        .TooMuch
-    else
-        .Standard;
-}
-
 pub inline fn unsupported(comptime fmt: []const u8, args: anytype) void {
     std.log.scoped(.UNSUPPORTED).warn(fmt, args);
 }

src/vulkan/lib_vulkan.zig

@@ -44,22 +44,14 @@ pub const RenderPass = @import("RenderPass.zig");
 pub const Sampler = @import("Sampler.zig");
 pub const ShaderModule = @import("ShaderModule.zig");
 
-fn entryPointBeginLogTrace(comptime scope: @Type(.enum_literal)) void {
+fn entryPointBeginLogTrace(comptime scope: @EnumLiteral()) void {
     std.log.scoped(scope).debug("Calling {s}...", .{@tagName(scope)});
-    logger.getManager().get().indent();
 }
 
-fn entryPointEndLogTrace() void {
+fn entryPointEndLogTrace() void {}
-    logger.getManager().get().unindent();
-}
-
-fn entryPointNotFoundErrorLog(comptime scope: @Type(.enum_literal), name: []const u8) void {
-    if (lib.getLogVerboseLevel() != .TooMuch) return;
-    std.log.scoped(scope).err("Could not find function {s}", .{name});
-}
 
 inline fn notImplementedWarning() void {
-    logger.nestedFixme("function not yet implemented", .{});
+    logger.fixme("function not yet implemented", .{});
 }
 
 fn functionMapEntryPoint(comptime name: []const u8) struct { []const u8, vk.PfnVoidFunction } {
@@ -251,9 +243,6 @@ pub export fn stroll_icdNegotiateLoaderICDInterfaceVersion(p_version: *u32) call
 }
 
 pub export fn vk_icdGetInstanceProcAddr(p_instance: vk.Instance, p_name: ?[*:0]const u8) callconv(vk.vulkan_call_conv) vk.PfnVoidFunction {
-    if (lib.getLogVerboseLevel() == .TooMuch) {
-        entryPointBeginLogTrace(.vk_icdGetInstanceProcAddr);
-    }
     defer entryPointEndLogTrace();
 
     if (p_name == null) return null;
@@ -264,9 +253,6 @@ pub export fn vk_icdGetInstanceProcAddr(p_instance: vk.Instance, p_name: ?[*:0]c
 }
 
 pub export fn stroll_icdGetPhysicalDeviceProcAddr(_: vk.Instance, p_name: ?[*:0]const u8) callconv(vk.vulkan_call_conv) vk.PfnVoidFunction {
-    if (lib.getLogVerboseLevel() == .TooMuch) {
-        entryPointBeginLogTrace(.vk_icdGetPhysicalDeviceProcAddr);
-    }
     defer entryPointEndLogTrace();
 
     if (p_name == null) return null;
@@ -274,16 +260,12 @@ pub export fn stroll_icdGetPhysicalDeviceProcAddr(_: vk.Instance, p_name: ?[*:0]
 
     if (physical_device_pfn_map.get(name)) |pfn| return pfn;
 
-    entryPointNotFoundErrorLog(.vk_icdGetPhysicalDeviceProcAddr, name);
     return null;
 }
 
 // Global functions ==========================================================================================================================================
 
 pub export fn vkGetInstanceProcAddr(p_instance: vk.Instance, p_name: ?[*:0]const u8) callconv(vk.vulkan_call_conv) vk.PfnVoidFunction {
-    if (lib.getLogVerboseLevel() == .TooMuch) {
-        entryPointBeginLogTrace(.vkGetInstanceProcAddr);
-    }
     defer entryPointEndLogTrace();
 
     if (p_name == null) return null;
@@ -295,7 +277,6 @@ pub export fn vkGetInstanceProcAddr(p_instance: vk.Instance, p_name: ?[*:0]const
         if (physical_device_pfn_map.get(name)) |pfn| return pfn;
         if (device_pfn_map.get(name)) |pfn| return pfn;
     }
-    entryPointNotFoundErrorLog(.vkGetInstanceProcAddr, name);
     return null;
 }
 
@@ -342,8 +323,6 @@ pub export fn strollEnumerateInstanceVersion(version: *u32) callconv(vk.vulkan_c
 // Instance functions ========================================================================================================================================
 
 pub export fn strollDestroyInstance(p_instance: vk.Instance, callbacks: ?*const vk.AllocationCallbacks) callconv(vk.vulkan_call_conv) void {
-    defer logger.getManager().deinit();
-
     entryPointBeginLogTrace(.vkDestroyInstance);
     defer entryPointEndLogTrace();
 
@@ -624,7 +603,7 @@ pub export fn strollAllocateCommandBuffers(p_device: vk.Device, info: *const vk.
 }
 
 pub export fn strollAllocateDescriptorSets(p_device: vk.Device, info: *const vk.DescriptorSetAllocateInfo, p_sets: [*]vk.DescriptorSet) callconv(vk.vulkan_call_conv) vk.Result {
-    entryPointBeginLogTrace(.vkAllocateCommandBuffers);
+    entryPointBeginLogTrace(.vkAllocateDescriptorSets);
     defer entryPointEndLogTrace();
 
     Dispatchable(Device).checkHandleValidity(p_device) catch |err| return toVkResult(err);
@@ -1322,9 +1301,6 @@ pub export fn strollGetDeviceMemoryCommitment(p_device: vk.Device, p_memory: vk.
 }
 
 pub export fn strollGetDeviceProcAddr(p_device: vk.Device, p_name: ?[*:0]const u8) callconv(vk.vulkan_call_conv) vk.PfnVoidFunction {
-    if (lib.getLogVerboseLevel() == .TooMuch) {
-        entryPointBeginLogTrace(.vkGetDeviceProcAddr);
-    }
     defer entryPointEndLogTrace();
 
     if (p_name == null) return null;
@@ -1333,7 +1309,6 @@ pub export fn strollGetDeviceProcAddr(p_device: vk.Device, p_name: ?[*:0]const u
     if (p_device == .null_handle) return null;
     if (device_pfn_map.get(name)) |pfn| return pfn;
 
-    entryPointNotFoundErrorLog(.vkGetDeviceProcAddr, name);
     return null;
 }
 
@@ -1385,7 +1360,7 @@ pub export fn strollGetImageMemoryRequirements(p_device: vk.Device, p_image: vk.
     Dispatchable(Device).checkHandleValidity(p_device) catch |err| return errorLogger(err);
 
     const image = NonDispatchable(Image).fromHandleObject(p_image) catch |err| return errorLogger(err);
-    image.getMemoryRequirements(requirements);
+    image.getMemoryRequirements(requirements) catch |err| return errorLogger(err);
 }
 
 pub export fn strollGetImageSparseMemoryRequirements(p_device: vk.Device, p_image: vk.Image, requirements: *vk.SparseImageMemoryRequirements) callconv(vk.vulkan_call_conv) void {
@@ -1517,28 +1492,18 @@ pub export fn strollResetCommandPool(p_device: vk.Device, p_pool: vk.CommandPool
 
     Dispatchable(Device).checkHandleValidity(p_device) catch |err| return toVkResult(err);
     const pool = NonDispatchable(CommandPool).fromHandleObject(p_pool) catch |err| return toVkResult(err);
-
+    pool.reset(flags) catch |err| return toVkResult(err);
-    notImplementedWarning();
+    return .success;
-
-    _ = pool;
-    _ = flags;
-
-    return .error_unknown;
 }
 
-pub export fn strollResetDescriptorPool(p_device: vk.Device, p_pool: vk.DescriptorPool, flags: vk.CommandPoolResetFlags) callconv(vk.vulkan_call_conv) vk.Result {
+pub export fn strollResetDescriptorPool(p_device: vk.Device, p_pool: vk.DescriptorPool, flags: vk.DescriptorPoolResetFlags) callconv(vk.vulkan_call_conv) vk.Result {
     entryPointBeginLogTrace(.vkResetDescriptorPool);
     defer entryPointEndLogTrace();
 
     Dispatchable(Device).checkHandleValidity(p_device) catch |err| return toVkResult(err);
     const pool = NonDispatchable(DescriptorPool).fromHandleObject(p_pool) catch |err| return toVkResult(err);
-
+    pool.reset(flags) catch |err| return toVkResult(err);
-    notImplementedWarning();
+    return .success;
-
-    _ = pool;
-    _ = flags;
-
-    return .error_unknown;
 }
 
 pub export fn strollResetEvent(p_device: vk.Device, p_event: vk.Fence) callconv(vk.vulkan_call_conv) vk.Result {
@@ -1735,16 +1700,7 @@ pub export fn strollCmdBlitImage(
     const src = NonDispatchable(Image).fromHandleObject(p_src_image) catch |err| return errorLogger(err);
     const dst = NonDispatchable(Image).fromHandleObject(p_dst_image) catch |err| return errorLogger(err);
 
-    notImplementedWarning();
+    cmd.blitImage(src, src_layout, dst, dst_layout, regions[0..count], filter) catch |err| return errorLogger(err);
-
-    _ = cmd;
-    _ = src;
-    _ = src_layout;
-    _ = dst;
-    _ = dst_layout;
-    _ = count;
-    _ = regions;
-    _ = filter;
 }
 
 pub export fn strollCmdClearAttachments(p_cmd: vk.CommandBuffer, attachment_count: u32, attachments: [*]const vk.ClearAttachment, rect_count: u32, rects: [*]const vk.ClearRect) callconv(vk.vulkan_call_conv) void {
@@ -1957,12 +1913,10 @@ pub export fn strollCmdExecuteCommands(p_cmd: vk.CommandBuffer, count: u32, p_cm
     defer entryPointEndLogTrace();
 
     const cmd = Dispatchable(CommandBuffer).fromHandleObject(p_cmd) catch |err| return errorLogger(err);
-
+    for (p_cmds, 0..count) |p_sec_cmd, _| {
-    notImplementedWarning();
+        const sec_cmd = Dispatchable(CommandBuffer).fromHandleObject(p_sec_cmd) catch |err| return errorLogger(err);
-
+        cmd.executeCommands(sec_cmd) catch |err| return errorLogger(err);
-    _ = cmd;
+    }
-    _ = count;
-    _ = p_cmds;
 }
 
 pub export fn strollCmdFillBuffer(p_cmd: vk.CommandBuffer, p_buffer: vk.Buffer, offset: vk.DeviceSize, size: vk.DeviceSize, data: u32) callconv(vk.vulkan_call_conv) void {

src/vulkan/logger.zig

@@ -0,0 +1,118 @@
+const std = @import("std");
+const builtin = @import("builtin");
+const root = @import("root");
+const lib = @import("lib.zig");
+
+comptime {
+    if (!builtin.is_test) {
+        if (!@hasDecl(root, "DRIVER_NAME")) {
+            @compileError("Missing DRIVER_NAME in module root");
+        }
+    }
+}
+
+var mutex: std.Io.Mutex = .init;
+
+pub inline fn fixme(comptime format: []const u8, args: anytype) void {
+    if (!lib.config.logs) {
+        return;
+    }
+    std.log.scoped(.FIXME).warn("FIXME: " ++ format, args);
+}
+
+pub fn log(comptime level: std.log.Level, comptime scope: @EnumLiteral(), comptime format: []const u8, args: anytype) void {
+    if (!lib.config.logs) {
+        return;
+    }
+
+    const scope_name = @tagName(scope);
+    const scope_prefix = comptime blk: {
+        const limit = 30 - 4;
+        break :blk if (scope_name.len >= limit)
+            std.fmt.comptimePrint("({s}...): ", .{scope_name[0..(limit - 3)]})
+        else
+            std.fmt.comptimePrint("({s}): ", .{scope_name});
+    };
+
+    const prefix = std.fmt.comptimePrint("{s: <10}", .{"[" ++ comptime level.asText() ++ "] "});
+
+    const level_color: std.Io.Terminal.Color = switch (level) {
+        .info, .debug => .blue,
+        .warn => .magenta,
+        .err => .red,
+    };
+
+    const allocator = std.heap.smp_allocator;
+
+    var threaded: std.Io.Threaded = .init(allocator, .{});
+    defer threaded.deinit();
+
+    const io = threaded.io();
+
+    const stderr_file = std.Io.File.stderr();
+    const stdout_file = std.Io.File.stdout();
+
+    const file = switch (level) {
+        .info, .debug => stdout_file,
+        .warn, .err => stderr_file,
+    };
+
+    file.lock(io, .exclusive) catch {};
+    defer file.unlock(io);
+
+    const now = std.Io.Timestamp.now(io, .cpu_process).toMilliseconds();
+
+    const now_ms: u16 = @intCast(@mod(now, std.time.ms_per_s));
+    const now_sec: u8 = @intCast(@mod(@divTrunc(now, std.time.ms_per_s), std.time.s_per_min));
+    const now_min: u8 = @intCast(@mod(@divTrunc(now, std.time.ms_per_min), 60));
+    const now_hour: u8 = @intCast(@mod(@divTrunc(now, std.time.ms_per_hour), 24));
+
+    var fmt_buffer = std.mem.zeroes([4096]u8);
+    var fmt_writer = std.Io.Writer.fixed(&fmt_buffer);
+    fmt_writer.print(format ++ "\n", args) catch {};
+    fmt_writer.flush() catch return;
+
+    mutex.lock(io) catch return;
+    defer mutex.unlock(io);
+
+    var last_pos: usize = 0;
+    while (std.mem.indexOfScalarPos(u8, &fmt_buffer, last_pos, '\n')) |pos| : (last_pos = pos + 1) {
+        var buffer = std.mem.zeroes([512]u8);
+        var file_writer = file.writer(io, &buffer);
+        var writer = &file_writer.interface;
+
+        const term: std.Io.Terminal = .{
+            .writer = writer,
+            .mode = std.Io.Terminal.Mode.detect(io, file, false, false) catch return,
+        };
+
+        term.setColor(.magenta) catch {};
+        writer.print("[StrollDriver ", .{}) catch continue;
+        if (!builtin.is_test) {
+            term.setColor(.cyan) catch {};
+            writer.print(root.DRIVER_NAME, .{}) catch continue;
+        }
+        term.setColor(.yellow) catch {};
+        writer.print(" {d}:{d}:{d}.{d:0>3}", .{ now_hour, now_min, now_sec, now_ms }) catch continue;
+        term.setColor(.magenta) catch {};
+        writer.print("]", .{}) catch continue;
+
+        term.setColor(.cyan) catch {};
+        writer.print("[Thread {d: >8}]", .{std.Thread.getCurrentId()}) catch continue;
+
+        term.setColor(level_color) catch {};
+        writer.print(prefix, .{}) catch continue;
+
+        term.setColor(switch (level) {
+            .err => .red,
+            .warn => .magenta,
+            else => .green,
+        }) catch {};
+        writer.print("{s: >30}", .{scope_prefix}) catch continue;
+
+        term.setColor(.reset) catch {};
+
+        writer.print("{s}\n", .{fmt_buffer[last_pos..pos]}) catch continue;
+        writer.flush() catch continue;
+    }
+}

src/vulkan/logger/DebugStack.zig

@@ -1,41 +0,0 @@
-const std = @import("std");
-
-const Self = @This();
-
-pub const Element = struct {
-    log: [512]u8,
-    indent_level: usize,
-    log_level: std.log.Level,
-};
-
-stack: std.ArrayList(Element),
-allocator: std.mem.Allocator = std.heap.c_allocator,
-
-pub const empty: Self = .{
-    .stack = .empty,
-};
-
-pub fn pushBack(self: *Self, element: Element) !void {
-    try self.stack.append(self.allocator, element);
-}
-
-pub fn popBack(self: *Self) ?Element {
-    return self.stack.pop();
-}
-
-pub fn popFront(self: *Self) Element {
-    return self.stack.orderedRemove(0);
-}
-
-pub fn getLastOrNull(self: *Self) ?Element {
-    return self.stack.getLastOrNull();
-}
-
-pub inline fn len(self: *Self) usize {
-    return self.stack.items.len;
-}
-
-pub fn deinit(self: *Self) void {
-    self.stack.deinit(self.allocator);
-    self.* = .empty;
-}

src/vulkan/logger/Manager.zig

@@ -1,54 +0,0 @@
-const std = @import("std");
-const DebugStack = @import("DebugStack.zig");
-const lib = @import("../lib.zig");
-
-const Self = @This();
-
-indent_enabled: bool,
-indent_level: usize,
-debug_stack: DebugStack,
-
-pub const init: Self = .{
-    .indent_enabled = true,
-    .indent_level = 0,
-    .debug_stack = .empty,
-};
-
-pub fn indent(self: *Self) void {
-    if (lib.getLogVerboseLevel() == .None) {
-        return;
-    }
-    const new_indent_level, const has_overflown = @addWithOverflow(self.indent_level, 1);
-    if (has_overflown == 0) {
-        self.indent_level = new_indent_level;
-    }
-}
-
-pub fn unindent(self: *Self) void {
-    if (lib.getLogVerboseLevel() == .None) {
-        return;
-    }
-    const new_indent_level, const has_overflown = @subWithOverflow(self.indent_level, 1);
-    if (has_overflown == 0) {
-        self.indent_level = new_indent_level;
-    }
-    loop: while (self.debug_stack.getLastOrNull()) |last| {
-        if (last.indent_level >= self.indent_level) {
-            _ = self.debug_stack.popBack();
-        } else {
-            break :loop;
-        }
-    }
-}
-
-pub inline fn enableIndent(self: *Self) void {
-    self.indent_enabled = true;
-}
-
-pub inline fn disableIndent(self: *Self) void {
-    self.indent_enabled = false;
-}
-
-pub inline fn deinit(self: *Self) void {
-    self.debug_stack.deinit();
-}

src/vulkan/logger/ThreadSafeManager.zig

@@ -1,39 +0,0 @@
-const std = @import("std");
-const Manager = @import("Manager.zig");
-
-const Self = @This();
-
-managers: std.AutoArrayHashMapUnmanaged(std.Thread.Id, Manager),
-allocator: std.heap.ThreadSafeAllocator,
-mutex: std.Thread.Mutex,
-
-pub const init: Self = .{
-    .managers = .empty,
-    .allocator = .{ .child_allocator = std.heap.c_allocator },
-    .mutex = .{},
-};
-
-pub fn get(self: *Self) *Manager {
-    const allocator = self.allocator.allocator();
-
-    self.mutex.lock();
-    defer self.mutex.unlock();
-
-    return (self.managers.getOrPutValue(allocator, std.Thread.getCurrentId(), .init) catch @panic("Out of memory")).value_ptr;
-}
-
-pub fn deinit(self: *Self) void {
-    {
-        self.mutex.lock();
-        defer self.mutex.unlock();
-
-        if (self.managers.getPtr(std.Thread.getCurrentId())) |manager| {
-            manager.deinit();
-            _ = self.managers.orderedRemove(std.Thread.getCurrentId());
-        }
-    }
-    if (self.managers.count() == 0) {
-        self.managers.deinit(self.allocator.allocator());
-        self.* = .init;
-    }
-}

src/vulkan/logger/logger.zig

@@ -1,146 +0,0 @@
-//! A driver-global logger that stack in memory all same-indent `debug` logs
-//! and only displays them in reverse order if a non-debug log is requested
-
-const std = @import("std");
-const builtin = @import("builtin");
-const zdt = @import("zdt");
-const root = @import("root");
-const lib = @import("../lib.zig");
-
-const ThreadSafeManager = @import("ThreadSafeManager.zig");
-
-comptime {
-    if (!builtin.is_test) {
-        if (!@hasDecl(root, "DRIVER_NAME")) {
-            @compileError("Missing DRIVER_NAME in module root");
-        }
-    }
-}
-
-var manager: ThreadSafeManager = .init;
-
-pub inline fn getManager() *ThreadSafeManager {
-    return &manager;
-}
-
-pub inline fn fixme(comptime format: []const u8, args: anytype) void {
-    if (lib.getLogVerboseLevel() == .None) {
-        return;
-    }
-    getManager().get().disableIndent();
-    defer getManager().get().enableIndent();
-    nestedFixme(format, args);
-}
-
-pub inline fn nestedFixme(comptime format: []const u8, args: anytype) void {
-    if (lib.getLogVerboseLevel() == .None) {
-        return;
-    }
-    std.log.scoped(.FIXME).warn("FIXME: " ++ format, args);
-}
-
-pub fn log(comptime level: std.log.Level, comptime scope: @Type(.enum_literal), comptime format: []const u8, args: anytype) void {
-    if (lib.getLogVerboseLevel() == .None) {
-        return;
-    }
-
-    const scope_name = @tagName(scope);
-    const scope_prefix = comptime blk: {
-        const limit = 30 - 4;
-        break :blk if (scope_name.len >= limit)
-            std.fmt.comptimePrint("({s}...): ", .{scope_name[0..(limit - 3)]})
-        else
-            std.fmt.comptimePrint("({s}): ", .{scope_name});
-    };
-
-    const prefix = std.fmt.comptimePrint("{s: <10}", .{"[" ++ comptime level.asText() ++ "] "});
-
-    const level_color: std.Io.tty.Color = switch (level) {
-        .info, .debug => .blue,
-        .warn => .magenta,
-        .err => .red,
-    };
-
-    std.debug.lockStdErr();
-    defer std.debug.unlockStdErr();
-
-    var stderr_file = std.fs.File.stderr();
-    var stdout_file = std.fs.File.stdout();
-
-    const file = switch (level) {
-        .info, .debug => stdout_file,
-        .warn, .err => stderr_file,
-    };
-
-    var timezone = zdt.Timezone.tzLocal(std.heap.page_allocator) catch zdt.Timezone.UTC;
-    defer timezone.deinit();
-    const now = zdt.Datetime.now(.{ .tz = &timezone }) catch zdt.Datetime{};
-
-    var fmt_buffer = std.mem.zeroes([4096]u8);
-    var fmt_writer = std.Io.Writer.fixed(&fmt_buffer);
-    fmt_writer.print(format ++ "\n", args) catch {};
-    fmt_writer.flush() catch return;
-
-    var last_pos: usize = 0;
-    while (std.mem.indexOfScalarPos(u8, &fmt_buffer, last_pos, '\n')) |pos| {
-        var buffer = std.mem.zeroes([512]u8);
-        var out_config = std.Io.tty.Config.detect(file);
-        var writer = std.Io.Writer.fixed(&buffer);
-
-        out_config.setColor(&writer, .magenta) catch {};
-        writer.print("[StrollDriver ", .{}) catch {};
-        if (!builtin.is_test) {
-            out_config.setColor(&writer, .cyan) catch {};
-            writer.print(root.DRIVER_NAME, .{}) catch {};
-        }
-        out_config.setColor(&writer, .yellow) catch {};
-        writer.print(" {d:02}:{d:02}:{d:02}.{d:03}", .{ now.hour, now.minute, now.second, @divFloor(now.nanosecond, std.time.ns_per_ms) }) catch {};
-        out_config.setColor(&writer, .magenta) catch {};
-        writer.print("]", .{}) catch {};
-
-        out_config.setColor(&writer, level_color) catch {};
-        writer.print(prefix, .{}) catch {};
-
-        out_config.setColor(&writer, switch (level) {
-            .err => .red,
-            .warn => .magenta,
-            else => .green,
-        }) catch {};
-        writer.print("{s: >30}", .{scope_prefix}) catch {};
-
-        out_config.setColor(&writer, .reset) catch {};
-
-        if (getManager().get().indent_enabled) {
-            for (0..getManager().get().indent_level) |_| {
-                writer.print(">   ", .{}) catch {};
-            }
-        }
-
-        writer.print("{s}\n", .{fmt_buffer[last_pos..pos]}) catch {};
-        writer.flush() catch return;
-
-        if (level == .debug and lib.getLogVerboseLevel() == .Standard) {
-            getManager().get().debug_stack.pushBack(.{
-                .log = buffer,
-                .indent_level = getManager().get().indent_level,
-                .log_level = level,
-            }) catch return;
-            return;
-        }
-
-        if (getManager().get().indent_enabled) {
-            while (getManager().get().debug_stack.len() != 0) {
-                const elem = getManager().get().debug_stack.popFront();
-                switch (elem.log_level) {
-                    .info, .debug => _ = stdout_file.write(&elem.log) catch {},
-                    .warn, .err => _ = stderr_file.write(&elem.log) catch {},
-                }
-            }
-        }
-        switch (level) {
-            .info, .debug => _ = stdout_file.write(&buffer) catch {},
-            .warn, .err => _ = stderr_file.write(&buffer) catch {},
-        }
-        last_pos = pos + 1;
-    }
-}

test/c/main.c

@@ -1,215 +0,0 @@
-#include <stdio.h>
-#include <stdlib.h>
-
-#define VK_NO_PROTOTYPES
-#include <vulkan/vulkan_core.h>
-
-#include <unistd.h>
-
-#include <dlfcn.h>
-
-#ifndef LIBVK
-	#define LIBVK "vulkan"
-#endif
-
-#define VOLK_IMPLEMENTATION
-#include <volk.h>
-
-#define KVF_IMPLEMENTATION
-#define KVF_ENABLE_VALIDATION_LAYERS
-#define KVF_NO_KHR
-#include <kvf.h>
-
-#define STB_IMAGE_WRITE_IMPLEMENTATION
-#include <stb_image_write.h>
-
-static const uint32_t vertex_shader[] = {
-	0x07230203,0x00010000,0x000d000b,0x00000036,0x00000000,0x00020011,0x00000001,0x0006000b,0x00000001,0x4c534c47,0x6474732e,0x3035342e,
-	0x00000000,0x0003000e,0x00000000,0x00000001,0x0008000f,0x00000000,0x00000004,0x6e69616d,0x00000000,0x00000022,0x00000026,0x00000031,
-	0x00030003,0x00000002,0x000001c2,0x000a0004,0x475f4c47,0x4c474f4f,0x70635f45,0x74735f70,0x5f656c79,0x656e696c,0x7269645f,0x69746365,
-	0x00006576,0x00080004,0x475f4c47,0x4c474f4f,0x6e695f45,0x64756c63,0x69645f65,0x74636572,0x00657669,0x00040005,0x00000004,0x6e69616d,
-	0x00000000,0x00050005,0x0000000c,0x69736f70,0x6e6f6974,0x00000073,0x00040005,0x00000017,0x6f6c6f63,0x00007372,0x00060005,0x00000020,
-	0x505f6c67,0x65567265,0x78657472,0x00000000,0x00060006,0x00000020,0x00000000,0x505f6c67,0x7469736f,0x006e6f69,0x00070006,0x00000020,
-	0x00000001,0x505f6c67,0x746e696f,0x657a6953,0x00000000,0x00070006,0x00000020,0x00000002,0x435f6c67,0x4470696c,0x61747369,0x0065636e,
-	0x00070006,0x00000020,0x00000003,0x435f6c67,0x446c6c75,0x61747369,0x0065636e,0x00030005,0x00000022,0x00000000,0x00060005,0x00000026,
-	0x565f6c67,0x65747265,0x646e4978,0x00007865,0x00050005,0x00000031,0x67617266,0x6f6c6f43,0x00000072,0x00050048,0x00000020,0x00000000,
-	0x0000000b,0x00000000,0x00050048,0x00000020,0x00000001,0x0000000b,0x00000001,0x00050048,0x00000020,0x00000002,0x0000000b,0x00000003,
-	0x00050048,0x00000020,0x00000003,0x0000000b,0x00000004,0x00030047,0x00000020,0x00000002,0x00040047,0x00000026,0x0000000b,0x0000002a,
-	0x00040047,0x00000031,0x0000001e,0x00000000,0x00020013,0x00000002,0x00030021,0x00000003,0x00000002,0x00030016,0x00000006,0x00000020,
-	0x00040017,0x00000007,0x00000006,0x00000002,0x00040015,0x00000008,0x00000020,0x00000000,0x0004002b,0x00000008,0x00000009,0x00000003,
-	0x0004001c,0x0000000a,0x00000007,0x00000009,0x00040020,0x0000000b,0x00000006,0x0000000a,0x0004003b,0x0000000b,0x0000000c,0x00000006,
-	0x0004002b,0x00000006,0x0000000d,0x00000000,0x0004002b,0x00000006,0x0000000e,0xbf000000,0x0005002c,0x00000007,0x0000000f,0x0000000d,
-	0x0000000e,0x0004002b,0x00000006,0x00000010,0x3f000000,0x0005002c,0x00000007,0x00000011,0x00000010,0x00000010,0x0005002c,0x00000007,
-	0x00000012,0x0000000e,0x00000010,0x0006002c,0x0000000a,0x00000013,0x0000000f,0x00000011,0x00000012,0x00040017,0x00000014,0x00000006,
-	0x00000003,0x0004001c,0x00000015,0x00000014,0x00000009,0x00040020,0x00000016,0x00000006,0x00000015,0x0004003b,0x00000016,0x00000017,
-	0x00000006,0x0004002b,0x00000006,0x00000018,0x3f800000,0x0006002c,0x00000014,0x00000019,0x00000018,0x0000000d,0x0000000d,0x0006002c,
-	0x00000014,0x0000001a,0x0000000d,0x00000018,0x0000000d,0x0006002c,0x00000014,0x0000001b,0x0000000d,0x0000000d,0x00000018,0x0006002c,
-	0x00000015,0x0000001c,0x00000019,0x0000001a,0x0000001b,0x00040017,0x0000001d,0x00000006,0x00000004,0x0004002b,0x00000008,0x0000001e,
-	0x00000001,0x0004001c,0x0000001f,0x00000006,0x0000001e,0x0006001e,0x00000020,0x0000001d,0x00000006,0x0000001f,0x0000001f,0x00040020,
-	0x00000021,0x00000003,0x00000020,0x0004003b,0x00000021,0x00000022,0x00000003,0x00040015,0x00000023,0x00000020,0x00000001,0x0004002b,
-	0x00000023,0x00000024,0x00000000,0x00040020,0x00000025,0x00000001,0x00000023,0x0004003b,0x00000025,0x00000026,0x00000001,0x00040020,
-	0x00000028,0x00000006,0x00000007,0x00040020,0x0000002e,0x00000003,0x0000001d,0x00040020,0x00000030,0x00000003,0x00000014,0x0004003b,
-	0x00000030,0x00000031,0x00000003,0x00040020,0x00000033,0x00000006,0x00000014,0x00050036,0x00000002,0x00000004,0x00000000,0x00000003,
-	0x000200f8,0x00000005,0x0003003e,0x0000000c,0x00000013,0x0003003e,0x00000017,0x0000001c,0x0004003d,0x00000023,0x00000027,0x00000026,
-	0x00050041,0x00000028,0x00000029,0x0000000c,0x00000027,0x0004003d,0x00000007,0x0000002a,0x00000029,0x00050051,0x00000006,0x0000002b,
-	0x0000002a,0x00000000,0x00050051,0x00000006,0x0000002c,0x0000002a,0x00000001,0x00070050,0x0000001d,0x0000002d,0x0000002b,0x0000002c,
-	0x0000000d,0x00000018,0x00050041,0x0000002e,0x0000002f,0x00000022,0x00000024,0x0003003e,0x0000002f,0x0000002d,0x0004003d,0x00000023,
-	0x00000032,0x00000026,0x00050041,0x00000033,0x00000034,0x00000017,0x00000032,0x0004003d,0x00000014,0x00000035,0x00000034,0x0003003e,
-	0x00000031,0x00000035,0x000100fd,0x00010038
-};
-
-static const uint32_t fragment_shader[] = {
-	0x07230203,0x00010000,0x000d000b,0x00000013,0x00000000,0x00020011,0x00000001,0x0006000b,0x00000001,0x4c534c47,0x6474732e,0x3035342e,
-	0x00000000,0x0003000e,0x00000000,0x00000001,0x0007000f,0x00000004,0x00000004,0x6e69616d,0x00000000,0x00000009,0x0000000c,0x00030010,
-	0x00000004,0x00000007,0x00030003,0x00000002,0x000001c2,0x000a0004,0x475f4c47,0x4c474f4f,0x70635f45,0x74735f70,0x5f656c79,0x656e696c,
-	0x7269645f,0x69746365,0x00006576,0x00080004,0x475f4c47,0x4c474f4f,0x6e695f45,0x64756c63,0x69645f65,0x74636572,0x00657669,0x00040005,
-	0x00000004,0x6e69616d,0x00000000,0x00050005,0x00000009,0x4374756f,0x726f6c6f,0x00000000,0x00050005,0x0000000c,0x67617266,0x6f6c6f43,
-	0x00000072,0x00040047,0x00000009,0x0000001e,0x00000000,0x00040047,0x0000000c,0x0000001e,0x00000000,0x00020013,0x00000002,0x00030021,
-	0x00000003,0x00000002,0x00030016,0x00000006,0x00000020,0x00040017,0x00000007,0x00000006,0x00000004,0x00040020,0x00000008,0x00000003,
-	0x00000007,0x0004003b,0x00000008,0x00000009,0x00000003,0x00040017,0x0000000a,0x00000006,0x00000003,0x00040020,0x0000000b,0x00000001,
-	0x0000000a,0x0004003b,0x0000000b,0x0000000c,0x00000001,0x0004002b,0x00000006,0x0000000e,0x3f800000,0x00050036,0x00000002,0x00000004,
-	0x00000000,0x00000003,0x000200f8,0x00000005,0x0004003d,0x0000000a,0x0000000d,0x0000000c,0x00050051,0x00000006,0x0000000f,0x0000000d,
-	0x00000000,0x00050051,0x00000006,0x00000010,0x0000000d,0x00000001,0x00050051,0x00000006,0x00000011,0x0000000d,0x00000002,0x00070050,
-	0x00000007,0x00000012,0x0000000f,0x00000010,0x00000011,0x0000000e,0x0003003e,0x00000009,0x00000012,0x000100fd,0x00010038
-};
-
-VkDeviceMemory CreateAndBindMemoryToBuffer(VkPhysicalDevice physical_device, VkDevice device, VkBuffer buffer, VkMemoryPropertyFlags props)
-{
-	VkMemoryRequirements requirements;
-	vkGetBufferMemoryRequirements(device, buffer, &requirements);
-
-	VkMemoryAllocateInfo alloc_info = {0};
-	alloc_info.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO;
-	alloc_info.allocationSize = requirements.size;
-	alloc_info.memoryTypeIndex = kvfFindMemoryType(physical_device, requirements.memoryTypeBits, props);
-
-	VkDeviceMemory memory;
-	kvfCheckVk(vkAllocateMemory(device, &alloc_info, NULL, &memory));
-	kvfCheckVk(vkBindBufferMemory(device, buffer, memory, 0));
-	return memory;
-}
-
-VkDeviceMemory CreateAndBindMemoryToImage(VkPhysicalDevice physical_device, VkDevice device, VkImage image, VkMemoryPropertyFlags props)
-{
-	VkMemoryRequirements requirements;
-	vkGetImageMemoryRequirements(device, image, &requirements);
-
-	VkMemoryAllocateInfo alloc_info = {0};
-	alloc_info.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO;
-	alloc_info.allocationSize = requirements.size;
-	alloc_info.memoryTypeIndex = kvfFindMemoryType(physical_device, requirements.memoryTypeBits, props);
-
-	VkDeviceMemory memory;
-	kvfCheckVk(vkAllocateMemory(device, &alloc_info, NULL, &memory));
-	kvfCheckVk(vkBindImageMemory(device, image, memory, 0));
-	return memory;
-}
-
-int main(void)
-{
-	volkInitialize();
-
-	void* lib = dlopen("./zig-out/lib/lib" LIBVK ".so", RTLD_NOW | RTLD_LOCAL);
-	if(!lib)
-	{
-		fprintf(stderr, "Could not open driver lib: %s\n", dlerror());
-		exit(EXIT_FAILURE);
-	}
-	puts("openned ./zig-out/lib/lib" LIBVK ".so");
-
-	VkDirectDriverLoadingInfoLUNARG direct_loading_info = {};
-	direct_loading_info.sType = VK_STRUCTURE_TYPE_DIRECT_DRIVER_LOADING_INFO_LUNARG;
-	direct_loading_info.pfnGetInstanceProcAddr = (PFN_vkGetInstanceProcAddrLUNARG)(dlsym(lib, "vk_icdGetInstanceProcAddr"));
-
-	VkDirectDriverLoadingListLUNARG direct_driver_list = {};
-	direct_driver_list.sType = VK_STRUCTURE_TYPE_DIRECT_DRIVER_LOADING_LIST_LUNARG;
-	direct_driver_list.mode = VK_DIRECT_DRIVER_LOADING_MODE_EXCLUSIVE_LUNARG;
-	direct_driver_list.driverCount = 1;
-	direct_driver_list.pDrivers = &direct_loading_info;
-
-	const char* extensions[] = { VK_LUNARG_DIRECT_DRIVER_LOADING_EXTENSION_NAME };
-
-	VkInstance instance = kvfCreateInstanceNext(extensions, 1, &direct_driver_list);
-	volkLoadInstance(instance);
-
-	VkPhysicalDevice physical_device = kvfPickGoodPhysicalDevice(instance, VK_NULL_HANDLE, NULL, 0);
-	VkDevice device = kvfCreateDevice(physical_device, NULL, 0, NULL);
-	volkLoadDevice(device);
-
-	VkImage image = kvfCreateImage(device, 600, 400, VK_FORMAT_R8G8B8A8_UNORM, VK_IMAGE_TILING_LINEAR, VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT | VK_IMAGE_USAGE_TRANSFER_DST_BIT, KVF_IMAGE_COLOR);
-	VkDeviceMemory memory = CreateAndBindMemoryToImage(physical_device, device, image, VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT);
-	VkImageView image_view = kvfCreateImageView(device, image, VK_FORMAT_R8G8B8A8_UNORM, VK_IMAGE_VIEW_TYPE_2D, VK_IMAGE_ASPECT_COLOR_BIT, 1);
-
-	VkAttachmentDescription attachment = kvfBuildAttachmentDescription(KVF_IMAGE_COLOR, VK_FORMAT_R8G8B8A8_UNORM, VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL, VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL, true, VK_SAMPLE_COUNT_1_BIT);
-	VkRenderPass renderpass = kvfCreateRenderPass(device, &attachment, 1, VK_PIPELINE_BIND_POINT_GRAPHICS);
-
-	VkFramebuffer framebuffer = kvfCreateFramebuffer(device, renderpass, &image_view, 1, (VkExtent2D){ .width = 600, .height = 400 });
-
-	VkShaderModule vertex_shader_module = kvfCreateShaderModule(device, (uint32_t*)vertex_shader, sizeof(vertex_shader) / sizeof(uint32_t));
-	VkShaderModule fragment_shader_module = kvfCreateShaderModule(device, (uint32_t*)fragment_shader, sizeof(fragment_shader) / sizeof(uint32_t));
-
-	VkPipelineLayout pipeline_layout = kvfCreatePipelineLayout(device, NULL, 0, NULL, 0);
-
-	KvfGraphicsPipelineBuilder* builder = kvfCreateGPipelineBuilder();
-	kvfGPipelineBuilderSetInputTopology(builder, VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST);
-	kvfGPipelineBuilderSetPolygonMode(builder, VK_POLYGON_MODE_FILL, 1.0f);
-	kvfGPipelineBuilderSetCullMode(builder, VK_CULL_MODE_NONE, VK_FRONT_FACE_CLOCKWISE);
-	kvfGPipelineBuilderSetMultisampling(builder, VK_SAMPLE_COUNT_1_BIT);
-	kvfGPipelineBuilderAddShaderStage(builder, VK_SHADER_STAGE_VERTEX_BIT, vertex_shader_module, "main");
-	kvfGPipelineBuilderAddShaderStage(builder, VK_SHADER_STAGE_FRAGMENT_BIT, fragment_shader_module, "main");
-	kvfGPipelineBuilderDisableDepthTest(builder);
-	kvfGPipelineBuilderDisableBlending(builder);
-
-	VkPipeline pipeline = kvfCreateGraphicsPipeline(device, VK_NULL_HANDLE, pipeline_layout, builder, renderpass);
-
-	kvfDestroyGPipelineBuilder(builder);
-	kvfDestroyShaderModule(device, vertex_shader_module);
-	kvfDestroyShaderModule(device, fragment_shader_module);
-
-	VkCommandBuffer cmd = kvfCreateCommandBuffer(device);
-
-	kvfCheckVk(vkResetCommandBuffer(cmd, 0));
-
-	kvfBeginCommandBuffer(cmd, 0);
-	{
-		vkCmdBindPipeline(cmd, VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
-		VkClearValue clear_color = {{{0.0f, 0.0f, 0.0f, 1.0f}}};
-		kvfBeginRenderPass(renderpass, cmd, framebuffer, (VkExtent2D){ .width = 600, .height = 400 }, &clear_color, 1);
-		VkViewport viewport = { 0 };
-		viewport.width = 600;
-		viewport.height = 400;
-		viewport.maxDepth = 1.0f;
-		vkCmdSetViewport(cmd, 0, 1, &viewport);
-		VkRect2D scissor = { 0 };
-		scissor.extent = (VkExtent2D){ .width = 600, .height = 400 };
-		vkCmdSetScissor(cmd, 0, 1, &scissor);
-		vkCmdDraw(cmd, 3, 1, 0, 0);
-		vkCmdEndRenderPass(cmd);
-	}
-	kvfEndCommandBuffer(cmd);
-
-	VkPipelineStageFlags wait_stages[] = { VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT };
-	kvfSubmitCommandBuffer(device, cmd, KVF_GRAPHICS_QUEUE, VK_NULL_HANDLE, VK_NULL_HANDLE, VK_NULL_HANDLE, wait_stages);
-	vkDeviceWaitIdle(device);
-
-	void* map = NULL;
-	kvfCheckVk(vkMapMemory(device, memory, 0, VK_WHOLE_SIZE, 0, &map));
-	if(!stbi_write_png("res.png", 256, 256, 4, map, 256 * 4))
-		fprintf(stderr, "Failed to write result image to file\n");
-	vkUnmapMemory(device, memory);
-
-	vkDeviceWaitIdle(device);
-	kvfDestroyPipelineLayout(device, pipeline_layout);
-	kvfDestroyPipeline(device, pipeline);
-	kvfDestroyRenderPass(device, renderpass);
-	kvfDestroyImageView(device, image_view);
-	kvfDestroyImage(device, image);
-	vkFreeMemory(device, memory, NULL);
-	kvfDestroyFramebuffer(device, framebuffer);
-
-	kvfDestroyDevice(device);
-	kvfDestroyInstance(instance);
-
-	dlclose(lib);
-	return 0;
-}