fixing CI

.gitea/workflows/build.yml

@@ -6,10 +6,6 @@ on:
   pull_request:
     branches: [ "master" ]
 
-permissions:
-  contents: read
-  deployments: write
-
 jobs:
   build:
     runs-on: ubuntu-latest
@@ -17,21 +13,30 @@ jobs:
 
     steps:
     - uses: actions/checkout@v4
-    - uses: https://codeberg.org/mlugg/setup-zig@v2
-    - uses: actions/setup-node@v6
       with:
-        node-version: 24
+        fetch-depth: 0
+
+    - uses: https://codeberg.org/mlugg/setup-zig@v2
 
     - name: Building
       run: zig build -Dno-example=true
 
+    - name: Building FFI
+      run: zig build ffi-c -Dno-example=true
+
     - name: Generating docs
       run: zig build docs -Dno-example=true
 
-    - name: Publish to Cloudflare Pages
-      uses: cloudflare/wrangler-action@v3
+    - name: Deploying docs
+      if: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
+      uses: milanmk/actions-file-deployer@master
       with:
-        apiToken: ${{ secrets.CLOUDFLARE_API_TOKEN }}
-        accountId: ${{ secrets.CLOUDFLARE_ACCOUNT_ID }}
-        command: pages deploy zig-out/docs --project-name=spirv-interpreter-docs
-        gitHubToken: ${{ secrets.GITHUB_TOKEN }}
+        remote-protocol: sftp
+        remote-host: ${{ secrets.SFTP_HOST_DOCS }}
+        remote-user: ${{ secrets.SFTP_USER_DOCS }}
+        remote-password: ${{ secrets.SFTP_PASSWORD_DOCS }}
+        remote-port: 6969
+        local-path: "./zig-out/docs"
+        remote-path: "/www"
+        sync: full
+

.gitea/workflows/test.yml

@@ -20,4 +20,4 @@ jobs:
     - uses: https://codeberg.org/mlugg/setup-zig@v2
 
     - name: Test
-      run: zig build test -Dno-example=true
+      run: zig build test -Dno-example=true --release=fast

README.md

@@ -84,3 +84,7 @@ int main(void)
     return 0;
 }
 ```
+
+## Documentation
+
+A full documentation on how to use this interpreter will come soon (when I'll stop being lazy).

build.zig

@@ -31,7 +31,7 @@ pub fn build(b: *std.Build) void {
 
     addSandbox(b, target, optimize, use_llvm, spv_mod, &install_spv_lib.step);
     addExample(b, target, optimize, use_llvm, spv_mod, &install_spv_lib.step);
-    addZigTests(b, target, optimize, spv_mod, zmath);
+    addZigTests(b, target, optimize, use_llvm, spv_mod, zmath);
     addCffi(b, target, optimize, use_llvm, spv_mod);
     addDocs(b, spv_mod);
 }
@@ -45,7 +45,7 @@ fn addExample(
     install_spv_lib_step: *std.Build.Step,
 ) void {
     const no_example = b.option(bool, "no-example", "Skip example build") orelse false;
-    if (!no_example and false) {
+    if (!no_example) {
         const sdl3 = b.lazyDependency("sdl3", .{
             .target = target,
             .optimize = optimize,
@@ -146,15 +146,22 @@ fn addZigTests(
     b: *std.Build,
     target: std.Build.ResolvedTarget,
     optimize: std.builtin.OptimizeMode,
+    use_llvm: bool,
     spv_mod: *std.Build.Module,
     zmath: *std.Build.Dependency,
 ) void {
     const no_test = b.option(bool, "no-test", "Skip unit test dependencies fetch") orelse false;
     if (no_test) return;
 
+    const test_filter = b.option(
+        []const u8,
+        "test-filter",
+        "Only run tests whose name contains this substring",
+    );
+
     const nzsl = b.lazyDependency("NZSL", .{
         .target = target,
-        .optimize = optimize,
+        .optimize = .ReleaseFast,
     }) orelse return;
 
     const tests = b.addTest(.{
@@ -172,6 +179,8 @@ fn addZigTests(
             .path = b.path("test/test_runner.zig"),
             .mode = .simple,
         },
+        .filters = if (test_filter) |filter| &.{filter} else &.{},
+        .use_llvm = use_llvm,
     });
 
     const run_tests = b.addRunArtifact(tests);

build.zig.zon

@@ -7,15 +7,15 @@
             .hash = "zmath-0.11.0-dev-wjwivdMsAwD-xaLj76YHUq3t9JDH-X16xuMTmnDzqbu2",
         },
         .NZSL = .{ // For unit tests
-            .url = "git+https://git.kbz8.me/kbz_8/NZigSL#ab95fc3734da46079fda2a4cd0f14143d92bf633",
-            .hash = "NZSL-1.1.2-N0xSVCR7AACeI_Wa6JPggJzy9_MPCpWC-2OHkMowwX-7",
+            .url = "git+https://git.kbz8.me/kbz_8/NZigSL#3ab28e423cdab27797f779d352094a69800c0c6b",
+            .hash = "NZSL-1.1.5-N0xSVHV7AAB168H2nqMDYY0hjUXtXJ9_eQGNHasSr9r8",
+            .lazy = true,
+        },
+        .sdl3 = .{
+            .url = "git+https://codeberg.org/7Games/zig-sdl3?ref=v0.2.0#40c2e4b579aa556db37a502c936426aa1c8b5c95",
+            .hash = "sdl3-0.2.0-NmT1Q0mFJwBi9kZmArzh2rfJ_mFshydV0zPGULVlpACc",
             .lazy = true,
         },
-        //.sdl3 = .{
-        //    .url = "git+https://codeberg.org/7Games/zig-sdl3?ref=v0.1.6#9c1842246c59f03f87ba59b160ca7e3d5e5ce972",
-        //    .hash = "sdl3-0.1.6-NmT1Q5sQJgCzT6hLj7WOSrwxE0Qsef1wIkDopbOOFru0",
-        //    .lazy = true,
-        //},
         .pretty = .{
             .url = "git+https://github.com/Kbz-8/pretty.git#f91d534d033277ca1ae7fcd598a070e8b3ddc532",
             .hash = "pretty-0.10.6-Tm65r7FTAQA5BEL8tcIcF-Wp4XRC7J7BhuRI0KUnFj2X",

example/main.zig

@@ -4,14 +4,22 @@ const spv = @import("spv");
 
 const shader_source = @embedFile("shader.spv");
 
-const screen_width = 300;
-const screen_height = 300;
+const screen_width = 400;
+const screen_height = 240;
 
 pub fn main() !void {
     {
         //var gpa: std.heap.DebugAllocator(.{}) = .init;
         //defer _ = gpa.deinit();
 
+        const allocator = std.heap.smp_allocator;
+
+        var threaded: std.Io.Threaded = .init(allocator, .{
+            .async_limit = @enumFromInt(screen_height),
+        });
+        defer threaded.deinit();
+        const io = threaded.io();
+
         defer sdl3.shutdown();
         const init_flags = sdl3.InitFlags{ .video = true, .events = true };
         try sdl3.init(init_flags);
@@ -22,26 +30,28 @@ pub fn main() !void {
 
         const surface = try window.getSurface();
 
-        const allocator = std.heap.smp_allocator;
-
         var module = try spv.Module.init(allocator, @ptrCast(@alignCast(shader_source)), .{});
         defer module.deinit(allocator);
 
-        var runner_cache: std.ArrayList(Runner) = try .initCapacity(allocator, screen_height);
+        const fragment_count = screen_height * screen_width;
+
+        const batch_size = switch (threaded.async_limit) {
+            .nothing => 1,
+            .unlimited => std.Thread.getCpuCount() catch 1, // If we cannot get the CPU count, fallback on single runtime
+            else => |count| @intFromEnum(count),
+        };
+
+        const runners: []Runner = try allocator.alloc(Runner, batch_size);
         defer {
-            for (runner_cache.items) |*runner| {
-                allocator.free(runner.heap);
+            for (runners) |*runner| {
+                runner.rt.deinit(allocator);
             }
-            runner_cache.deinit(allocator);
+            allocator.free(runners);
         }
 
-        for (0..screen_height) |_| {
-            const heap = try allocator.alloc(u8, module.needed_runtime_bytes);
-            errdefer allocator.free(heap);
-
-            var buffer_allocator: std.heap.FixedBufferAllocator = .init(heap);
-            var rt = try spv.Runtime.init(buffer_allocator.allocator(), &module);
-            (try runner_cache.addOne(allocator)).* = .{
+        for (runners) |*runner| {
+            var rt = try spv.Runtime.init(allocator, &module, undefined);
+            runner.* = .{
                 .allocator = allocator,
                 .surface = surface,
                 .rt = rt,
@@ -50,15 +60,13 @@ pub fn main() !void {
                 .time = try rt.getResultByName("time"),
                 .pos = try rt.getResultByName("pos"),
                 .res = try rt.getResultByName("res"),
-                .heap = heap,
+                .invocation_count = fragment_count,
+                .batch_size = batch_size,
             };
         }
+        const timer = std.Io.Timestamp.now(io, .real);
 
-        var thread_pool: std.Thread.Pool = undefined;
-        try thread_pool.init(.{ .allocator = allocator });
-
-        var timer = try std.time.Timer.start();
-
+        var wg: std.Io.Group = .init;
         var quit = false;
         while (!quit) {
             try surface.clear(.{ .r = 0.0, .g = 0.0, .b = 0.0, .a = 1.0 });
@@ -76,21 +84,13 @@ pub fn main() !void {
 
                 const pixel_map: [*]u32 = @as([*]u32, @ptrCast(@alignCast((surface.getPixels() orelse return).ptr)));
 
-                const delta: f32 = @as(f32, @floatFromInt(timer.read())) / std.time.ns_per_s;
+                const duration = timer.untilNow(io, .real);
+                const delta: f32 = @as(f32, @floatFromInt(duration.toNanoseconds())) / std.time.ns_per_s;
 
-                var frame_timer = try std.time.Timer.start();
-                defer {
-                    const ns = frame_timer.lap();
-                    const ms = @as(f32, @floatFromInt(ns)) / std.time.ns_per_s;
-                    std.log.info("Took {d:.3}s - {d:.3}fps to render", .{ ms, 1.0 / ms });
+                for (0..@min(batch_size, fragment_count)) |batch_id| {
+                    wg.async(io, Runner.runWrapper, .{ &runners[batch_id], batch_id, pixel_map, delta });
                 }
-
-                var wait_group: std.Thread.WaitGroup = .{};
-                for (0..screen_height) |y| {
-                    const runner = &runner_cache.items[y];
-                    thread_pool.spawnWg(&wait_group, Runner.runWrapper, .{ runner, y, pixel_map, delta });
-                }
-                thread_pool.waitAndWork(&wait_group);
+                try wg.await(io);
             }
 
             try window.updateSurface();
@@ -110,24 +110,29 @@ const Runner = struct {
     time: spv.SpvWord,
     pos: spv.SpvWord,
     res: spv.SpvWord,
-    heap: []u8,
+    invocation_count: usize,
+    batch_size: usize,
 
-    fn runWrapper(self: *Self, y: usize, pixel_map: [*]u32, timer: f32) void {
-        @call(.always_inline, Self.run, .{ self, y, pixel_map, timer }) catch |err| {
+    fn runWrapper(self: *Self, batch_id: usize, pixel_map: [*]u32, timer: f32) void {
+        @call(.always_inline, Self.run, .{ self, batch_id, pixel_map, timer }) catch |err| {
             std.log.err("{s}", .{@errorName(err)});
             if (@errorReturnTrace()) |trace| {
-                std.debug.dumpStackTrace(trace.*);
+                std.debug.dumpErrorReturnTrace(trace);
             }
             std.process.abort();
         };
     }
 
-    fn run(self: *Self, y: usize, pixel_map: [*]u32, timer: f32) !void {
+    fn run(self: *Self, batch_id: usize, pixel_map: [*]u32, timer: f32) !void {
         var rt = self.rt; // Copy to avoid pointer access of `self` at runtime. Okay as Runtime contains only pointers and trivially copyable fields
 
         var output: [4]f32 = undefined;
 
-        for (0..screen_width) |x| {
+        var invocation_index: usize = batch_id;
+        while (invocation_index < self.invocation_count) : (invocation_index += self.batch_size) {
+            const y = @divTrunc(invocation_index, screen_width);
+            const x = @mod(invocation_index, screen_width);
+
             try rt.writeInput(std.mem.asBytes(&timer), self.time);
             try rt.writeInput(std.mem.asBytes(&[_]f32{ @floatFromInt(screen_width), @floatFromInt(screen_height) }), self.res);
             try rt.writeInput(std.mem.asBytes(&[_]f32{ @floatFromInt(x), @floatFromInt(y) }), self.pos);
@@ -135,13 +140,13 @@ const Runner = struct {
             try rt.readOutput(std.mem.asBytes(output[0..]), self.color);
 
             const rgba = self.surface.mapRgba(
-                @intCast(@max(@min(@as(i32, @intFromFloat(output[0] * 255.0)), 255), 0)),
-                @intCast(@max(@min(@as(i32, @intFromFloat(output[1] * 255.0)), 255), 0)),
-                @intCast(@max(@min(@as(i32, @intFromFloat(output[2] * 255.0)), 255), 0)),
-                @intCast(@max(@min(@as(i32, @intFromFloat(output[3] * 255.0)), 255), 0)),
+                @intCast(std.math.clamp(@as(i32, @intFromFloat(output[0] * 255.0)), 0, 255)),
+                @intCast(std.math.clamp(@as(i32, @intFromFloat(output[1] * 255.0)), 0, 255)),
+                @intCast(std.math.clamp(@as(i32, @intFromFloat(output[2] * 255.0)), 0, 255)),
+                @intCast(std.math.clamp(@as(i32, @intFromFloat(output[3] * 255.0)), 0, 255)),
             );
 
-            pixel_map[(y * self.surface.getWidth()) + x] = rgba.value;
+            pixel_map[invocation_index] = rgba.value;
         }
     }
 };

ffi/SpirvInterpreter.h

@@ -38,112 +38,7 @@ extern "C"
 #endif
 
 #ifndef spirv_H
-	typedef enum SpvBuiltIn_ SpvBuiltIn;
-	typedef enum SpvDecoration_ SpvDecoration;
-#endif /* spirv_H */
-
-typedef int SpvBool;
-typedef unsigned char SpvByte;
-typedef unsigned long SpvWord;
-typedef unsigned long SpvSize;
-
-typedef enum
-{
-	SPV_RESULT_SUCCESS = 0,
-	SPV_RESULT_DIVISION_BY_ZERO = -1,
-	SPV_RESULT_INVALID_ENTRY_POINT = -2,
-	SPV_RESULT_INVALID_SPIRV = -3,
-	SPV_RESULT_INVALID_VALUE_TYPE = -4,
-	SPV_RESULT_KILLED = -5,
-	SPV_RESULT_NOT_FOUND = -6,
-	SPV_RESULT_OUT_OF_MEMORY = -7,
-	SPV_RESULT_OUT_OF_BOUNDS = -8,
-	SPV_RESULT_TODO = -9,
-	SPV_RESULT_UNREACHABLE = -10,
-	SPV_RESULT_UNSUPPORTED_SPIRV = -11,
-	SPV_RESULT_UNSUPPORTED_EXTENSION = -12,
-	SPV_RESULT_UNSUPPORTED_ENDIANNESS = -13,
-	SPV_RESULT_INVALID_MAGIC = -14,
-	SPV_RESULT_UNKNOWN = -15,
-} SpvResult;
-
-typedef struct
-{
-	SpvBool use_simd_vectors_specializations;
-} SpvModuleOptions;
-
-typedef struct
-{
-	SpvWord id;
-	SpvSize offset;
-	SpvSize size;
-} SpvRuntimeSpecializationEntry;
-
-typedef enum
-{
-	SPV_LOCATION_INPUT = 0,
-	SPV_LOCATION_OUTPUT = 1,
-} SpvLocationType;
-
-typedef struct
-{
-	float x;
-	float y;
-	float z;
-	float w;
-} SpvVec4f;
-
-typedef struct
-{
-	unsigned int x;
-	unsigned int y;
-	unsigned int z;
-	unsigned int w;
-} SpvVec4u;
-
-typedef SpvResult (*SpvReadImageFloat4_PFN)(void* driver_image, int x, int y, int z, SpvVec4f* dst);
-typedef SpvResult (*SpvReadImageInt4_PFN)(void* driver_image, int x, int y, int z, SpvVec4u* dst);
-typedef SpvResult (*SpvWriteImageFloat4_PFN)(void* driver_image, int x, int y, int z, SpvVec4f src);
-typedef SpvResult (*SpvWriteImageInt4_PFN)(void* driver_image, int x, int y, int z, SpvVec4u src);
-
-typedef struct
-{
-	SpvReadImageFloat4_PFN SpvReadImageFloat4;
-	SpvReadImageInt4_PFN  SpvReadImageInt4;
-	SpvWriteImageFloat4_PFN SpvWriteImageFloat4;
-	SpvWriteImageInt4_PFN SpvWriteImageInt4;
-} SpvImageAPI;
-
-typedef void* SpvModule;
-typedef void* SpvRuntime;
-
-SPV_API SpvResult SpvInitModule(SpvModule* module, const SpvWord* source, SpvSize source_len, SpvModuleOptions options);
-SPV_API void SpvDeinitModule(SpvModule module);
-
-SPV_API SpvResult SpvInitRuntime(SpvRuntime* runtime, SpvModule module, SpvImageAPI image_api);
-SPV_API void SpvDeinitRuntime(SpvRuntime runtime);
-
-SPV_API SpvResult SpvFlushDescriptorSets(SpvRuntime runtime);
-
-SPV_API SpvResult SpvAddSpecializationInfo(SpvRuntime runtime, SpvRuntimeSpecializationEntry entry, const SpvByte* data, SpvSize data_size);
-
-SPV_API SpvResult SpvGetResultByName(SpvRuntime runtime, const char* name, SpvWord* result);
-SPV_API SpvResult SpvGetResultLocation(SpvRuntime runtime, SpvWord location, SpvLocationType type, SpvWord* result);
-SPV_API SpvResult SpvGetEntryPointByName(SpvRuntime runtime, const char* name, SpvWord* result);
-SPV_API SpvResult SpvGetResultMemorySize(SpvRuntime runtime, SpvWord result, SpvSize* size);
-SPV_API SpvBool SpvHasResultDecoration(SpvRuntime runtime, SpvWord result, SpvDecoration decoration);
-
-SPV_API SpvResult SpvCallEntryPoint(SpvRuntime runtime, SpvWord entry_point_index);
-
-SPV_API SpvResult SpvReadOutput(SpvRuntime runtime, SpvByte* output, SpvSize output_size, SpvWord result);
-SPV_API SpvResult SpvReadBuiltIn(SpvRuntime runtime, SpvByte* output, SpvSize output_size, SpvBuiltIn builtin);
-
-SPV_API SpvResult SpvWriteInput(SpvRuntime runtime, const SpvByte* input, SpvSize input_size, SpvWord result);
-SPV_API SpvResult SpvWriteBuiltIn(SpvRuntime runtime, const SpvByte* input, SpvSize input_size, SpvBuiltIn builtin);
-SPV_API SpvResult SpvWriteDescriptorSet(SpvRuntime runtime, const SpvByte* input, SpvSize input_size, SpvWord set, SpvWord binding, SpvWord descriptor_index);
-
-#ifndef spirv_H
-	enum SpvBuiltIn_
+	typedef enum SpvBuiltIn_
 	{
 		SpvBuiltInPosition = 0,
 		SpvBuiltInPointSize = 1,
@@ -295,8 +190,8 @@ SPV_API SpvResult SpvWriteDescriptorSet(SpvRuntime runtime, const SpvByte* input
 		SpvBuiltInHitLSSRadiiNV = 5421,
 		SpvBuiltInClusterIDNV = 5436,
 		SpvBuiltInCullMaskKHR = 6021,
-		SpvBuiltInMax = 0x7fffffff,
-	};
+		SpvBuiltInMax = 0x7fffffff
+	} SpvBuiltIn;
 
 	typedef enum SpvDecoration_ {
 		SpvDecorationRelaxedPrecision = 0,
@@ -497,10 +392,127 @@ SPV_API SpvResult SpvWriteDescriptorSet(SpvRuntime runtime, const SpvByte* input
 		SpvDecorationConditionalINTEL = 6247,
 		SpvDecorationCacheControlLoadINTEL = 6442,
 		SpvDecorationCacheControlStoreINTEL = 6443,
-		SpvDecorationMax = 0x7fffffff,
+		SpvDecorationMax = 0x7fffffff
 	} SpvDecoration;
+
+	typedef enum SpvDim_ {
+		_1D = 0,
+		_2D = 1,
+		_3D = 2,
+		Cube = 3,
+		Rect = 4,
+		Buffer = 5,
+		SubpassData = 6,
+		TileImageDataEXT = 4173,
+		Max = 0x7fffffff
+	} SpvDim;
 #endif /* spirv_H */
 
+typedef int SpvBool;
+typedef unsigned char SpvByte;
+typedef unsigned long SpvWord;
+typedef unsigned long SpvSize;
+
+typedef enum
+{
+	SPV_RESULT_SUCCESS = 0,
+	SPV_RESULT_BARRIER = 1,
+	SPV_RESULT_KILLED = 2,
+	SPV_RESULT_DIVISION_BY_ZERO = -1,
+	SPV_RESULT_INVALID_ENTRY_POINT = -2,
+	SPV_RESULT_INVALID_SPIRV = -3,
+	SPV_RESULT_INVALID_VALUE_TYPE = -4,
+	SPV_RESULT_NOT_FOUND = -5,
+	SPV_RESULT_OUT_OF_MEMORY = -6,
+	SPV_RESULT_OUT_OF_BOUNDS = -7,
+	SPV_RESULT_TODO = -8,
+	SPV_RESULT_UNREACHABLE = -9,
+	SPV_RESULT_UNSUPPORTED_SPIRV = -10,
+	SPV_RESULT_UNSUPPORTED_EXTENSION = -11,
+	SPV_RESULT_UNSUPPORTED_ENDIANNESS = -12,
+	SPV_RESULT_INVALID_MAGIC = -13,
+	SPV_RESULT_UNKNOWN = -14
+} SpvResult;
+
+typedef struct
+{
+	SpvBool use_simd_vectors_specializations;
+} SpvModuleOptions;
+
+typedef struct
+{
+	SpvWord id;
+	SpvSize offset;
+	SpvSize size;
+} SpvRuntimeSpecializationEntry;
+
+typedef enum
+{
+	SPV_LOCATION_INPUT = 0,
+	SPV_LOCATION_OUTPUT = 1
+} SpvLocationType;
+
+typedef struct
+{
+	float x;
+	float y;
+	float z;
+	float w;
+} SpvVec4f;
+
+typedef struct
+{
+	unsigned int x;
+	unsigned int y;
+	unsigned int z;
+	unsigned int w;
+} SpvVec4u;
+
+typedef SpvResult (*SpvReadImageFloat4_PFN)(void* driver_image, SpvDim dim, int x, int y, int z, SpvVec4f* dst);
+typedef SpvResult (*SpvReadImageInt4_PFN)(void* driver_image, SpvDim dim, int x, int y, int z, SpvVec4u* dst);
+typedef SpvResult (*SpvWriteImageFloat4_PFN)(void* driver_image, SpvDim dim, int x, int y, int z, SpvVec4f src);
+typedef SpvResult (*SpvWriteImageInt4_PFN)(void* driver_image, SpvDim dim, int x, int y, int z, SpvVec4u src);
+typedef SpvResult (*SpvSampleImageFloat4_PFN)(void* driver_image, void* driver_sampler, SpvDim dim, float x, float y, float z, SpvVec4f* dst);
+typedef SpvResult (*SpvQueryImageSize_PFN)(void* driver_image, SpvDim dim, SpvBool arrayed, SpvVec4u* dst);
+
+typedef struct
+{
+	SpvReadImageFloat4_PFN SpvReadImageFloat4;
+	SpvReadImageInt4_PFN  SpvReadImageInt4;
+	SpvWriteImageFloat4_PFN SpvWriteImageFloat4;
+	SpvWriteImageInt4_PFN SpvWriteImageInt4;
+	SpvSampleImageFloat4_PFN SpvSampleImageFloat4;
+	SpvQueryImageSize_PFN SpvQueryImageSize;
+} SpvImageAPI;
+
+typedef void* SpvModule;
+typedef void* SpvRuntime;
+
+SPV_API SpvResult SpvInitModule(SpvModule* module, const SpvWord* source, SpvSize source_len, SpvModuleOptions options);
+SPV_API void SpvDeinitModule(SpvModule module);
+
+SPV_API SpvResult SpvInitRuntime(SpvRuntime* runtime, SpvModule module, SpvImageAPI image_api);
+SPV_API void SpvDeinitRuntime(SpvRuntime runtime);
+
+SPV_API SpvResult SpvFlushDescriptorSets(SpvRuntime runtime);
+
+SPV_API SpvResult SpvAddSpecializationInfo(SpvRuntime runtime, SpvRuntimeSpecializationEntry entry, const SpvByte* data, SpvSize data_size);
+
+SPV_API SpvResult SpvGetResultByName(SpvRuntime runtime, const char* name, SpvWord* result);
+SPV_API SpvResult SpvGetResultLocation(SpvRuntime runtime, SpvWord location, SpvLocationType type, SpvWord* result);
+SPV_API SpvResult SpvGetEntryPointByName(SpvRuntime runtime, const char* name, SpvWord* result);
+SPV_API SpvResult SpvGetResultMemorySize(SpvRuntime runtime, SpvWord result, SpvSize* size);
+SPV_API SpvBool SpvHasResultDecoration(SpvRuntime runtime, SpvWord result, SpvDecoration decoration);
+
+SPV_API SpvResult SpvCallEntryPoint(SpvRuntime runtime, SpvWord entry_point_index);
+
+SPV_API SpvResult SpvReadOutput(SpvRuntime runtime, SpvByte* output, SpvSize output_size, SpvWord result);
+SPV_API SpvResult SpvReadBuiltIn(SpvRuntime runtime, SpvByte* output, SpvSize output_size, SpvBuiltIn builtin);
+
+SPV_API SpvResult SpvWriteInput(SpvRuntime runtime, const SpvByte* input, SpvSize input_size, SpvWord result);
+SPV_API SpvResult SpvWriteBuiltIn(SpvRuntime runtime, const SpvByte* input, SpvSize input_size, SpvBuiltIn builtin);
+SPV_API SpvResult SpvWriteDescriptorSet(SpvRuntime runtime, const SpvByte* input, SpvSize input_size, SpvWord set, SpvWord binding, SpvWord descriptor_index);
+
 #ifdef __cplusplus
 }
 #endif

ffi/ffi.zig

@@ -6,21 +6,22 @@ pub const SpvCSize = c_ulong;
 
 pub const Result = enum(c_int) {
     Success = 0,
+    Barrier = 1,
+    Killed = 2,
     DivisionByZero = -1,
     InvalidEntryPoint = -2,
     InvalidSpirV = -3,
     InvalidValueType = -4,
-    Killed = -5,
-    NotFound = -6,
-    OutOfMemory = -7,
-    OutOfBounds = -8,
-    ToDo = -9,
-    Unreachable = -10,
-    UnsupportedSpirV = -11,
-    UnsupportedExtension = -12,
-    UnsupportedEndianness = -13,
-    InvalidMagic = -14,
-    Unknown = -15,
+    NotFound = -5,
+    OutOfMemory = -6,
+    OutOfBounds = -7,
+    ToDo = -8,
+    Unreachable = -9,
+    UnsupportedSpirV = -10,
+    UnsupportedExtension = -11,
+    UnsupportedEndianness = -12,
+    InvalidMagic = -13,
+    Unknown = -14,
 };
 
 comptime {

ffi/runtime.zig

@@ -27,67 +27,74 @@ const Vec4u = extern struct {
     w: c_uint,
 };
 
-const readImageFloat4_PFN = *const fn (driver_image: ?*anyopaque, x: c_int, y: c_int, z: c_int, dst: *Vec4f) callconv(.c) ffi.Result;
-const readImageInt4_PFN = *const fn (driver_image: ?*anyopaque, x: c_int, y: c_int, z: c_int, dst: *Vec4u) callconv(.c) ffi.Result;
-const writeImageFloat4_PFN = *const fn (driver_image: ?*anyopaque, x: c_int, y: c_int, z: c_int, src: Vec4f) callconv(.c) ffi.Result;
-const writeImageInt4_PFN = *const fn (driver_image: ?*anyopaque, x: c_int, y: c_int, z: c_int, src: Vec4u) callconv(.c) ffi.Result;
+const readImageFloat4_PFN = *const fn (driver_image: ?*anyopaque, dim: spv.spv.SpvDim, x: c_int, y: c_int, z: c_int, dst: *Vec4f) callconv(.c) ffi.Result;
+const readImageInt4_PFN = *const fn (driver_image: ?*anyopaque, dim: spv.spv.SpvDim, x: c_int, y: c_int, z: c_int, dst: *Vec4u) callconv(.c) ffi.Result;
+const writeImageFloat4_PFN = *const fn (driver_image: ?*anyopaque, dim: spv.spv.SpvDim, x: c_int, y: c_int, z: c_int, src: Vec4f) callconv(.c) ffi.Result;
+const writeImageInt4_PFN = *const fn (driver_image: ?*anyopaque, dim: spv.spv.SpvDim, x: c_int, y: c_int, z: c_int, src: Vec4u) callconv(.c) ffi.Result;
+const sampleImageFloat4_PFN = *const fn (driver_image: ?*anyopaque, driver_sampler: ?*anyopaque, dim: spv.spv.SpvDim, x: f32, y: f32, z: f32, dst: *Vec4f) callconv(.c) ffi.Result;
+const queryImageSize_PFN = *const fn (driver_image: ?*anyopaque, dim: spv.spv.SpvDim, arrayed: ffi.SpvCBool, dst: *Vec4u) callconv(.c) ffi.Result;
 
 const ImageAPI = extern struct {
     readImageFloat4: readImageFloat4_PFN,
     readImageInt4: readImageInt4_PFN,
     writeImageFloat4: writeImageFloat4_PFN,
     writeImageInt4: writeImageInt4_PFN,
+    sampleImageFloat4: sampleImageFloat4_PFN,
+    queryImageSize: queryImageSize_PFN,
 };
 
 fn toCResult(err: spv.Runtime.RuntimeError) ffi.Result {
     return switch (err) {
+        spv.Runtime.RuntimeError.Barrier => ffi.Result.Barrier,
         spv.Runtime.RuntimeError.DivisionByZero => ffi.Result.DivisionByZero,
         spv.Runtime.RuntimeError.InvalidEntryPoint => ffi.Result.InvalidEntryPoint,
         spv.Runtime.RuntimeError.InvalidSpirV => ffi.Result.InvalidSpirV,
         spv.Runtime.RuntimeError.InvalidValueType => ffi.Result.InvalidValueType,
         spv.Runtime.RuntimeError.Killed => ffi.Result.Killed,
         spv.Runtime.RuntimeError.NotFound => ffi.Result.NotFound,
-        spv.Runtime.RuntimeError.OutOfMemory => ffi.Result.OutOfMemory,
         spv.Runtime.RuntimeError.OutOfBounds => ffi.Result.OutOfBounds,
+        spv.Runtime.RuntimeError.OutOfMemory => ffi.Result.OutOfMemory,
         spv.Runtime.RuntimeError.ToDo => ffi.Result.ToDo,
-        spv.Runtime.RuntimeError.Unreachable => ffi.Result.Unreachable,
-        spv.Runtime.RuntimeError.UnsupportedSpirV => ffi.Result.UnsupportedSpirV,
-        spv.Runtime.RuntimeError.UnsupportedExtension => ffi.Result.UnsupportedExtension,
         spv.Runtime.RuntimeError.Unknown => ffi.Result.Unknown,
+        spv.Runtime.RuntimeError.Unreachable => ffi.Result.Unreachable,
+        spv.Runtime.RuntimeError.UnsupportedExtension => ffi.Result.UnsupportedExtension,
+        spv.Runtime.RuntimeError.UnsupportedSpirV => ffi.Result.UnsupportedSpirV,
     };
 }
 
 fn fromCResult(res: ffi.Result) spv.Runtime.RuntimeError!void {
     return switch (res) {
+        ffi.Result.Barrier => spv.Runtime.RuntimeError.Barrier,
         ffi.Result.DivisionByZero => spv.Runtime.RuntimeError.DivisionByZero,
         ffi.Result.InvalidEntryPoint => spv.Runtime.RuntimeError.InvalidEntryPoint,
         ffi.Result.InvalidSpirV => spv.Runtime.RuntimeError.InvalidSpirV,
         ffi.Result.InvalidValueType => spv.Runtime.RuntimeError.InvalidValueType,
         ffi.Result.Killed => spv.Runtime.RuntimeError.Killed,
         ffi.Result.NotFound => spv.Runtime.RuntimeError.NotFound,
-        ffi.Result.OutOfMemory => spv.Runtime.RuntimeError.OutOfMemory,
         ffi.Result.OutOfBounds => spv.Runtime.RuntimeError.OutOfBounds,
+        ffi.Result.OutOfMemory => spv.Runtime.RuntimeError.OutOfMemory,
         ffi.Result.ToDo => spv.Runtime.RuntimeError.ToDo,
-        ffi.Result.Unreachable => spv.Runtime.RuntimeError.Unreachable,
-        ffi.Result.UnsupportedSpirV => spv.Runtime.RuntimeError.UnsupportedSpirV,
-        ffi.Result.UnsupportedExtension => spv.Runtime.RuntimeError.UnsupportedExtension,
         ffi.Result.Unknown => spv.Runtime.RuntimeError.Unknown,
+        ffi.Result.Unreachable => spv.Runtime.RuntimeError.Unreachable,
+        ffi.Result.UnsupportedExtension => spv.Runtime.RuntimeError.UnsupportedExtension,
+        ffi.Result.UnsupportedSpirV => spv.Runtime.RuntimeError.UnsupportedSpirV,
         else => {},
     };
 }
 
+/// Hacky wrapper
 const ImageAPIBridge = struct {
-    threadlocal var current_image_api: ?*const ImageAPI = null; // Hacky
+    threadlocal var current_image_api: ?*const ImageAPI = null;
 
     fn getImageAPI() spv.Runtime.RuntimeError!*const ImageAPI {
         return current_image_api orelse spv.Runtime.RuntimeError.Unknown;
     }
 
-    fn readImageFloat4(driver_image: *anyopaque, x: i32, y: i32, z: i32) spv.Runtime.RuntimeError!spv.Runtime.Vec4(f32) {
+    fn readImageFloat4(driver_image: *anyopaque, dim: spv.spv.SpvDim, x: i32, y: i32, z: i32) spv.Runtime.RuntimeError!spv.Runtime.Vec4(f32) {
         const image_api = try getImageAPI();
 
         var dst: Vec4f = undefined;
-        const result = image_api.readImageFloat4(driver_image, @intCast(x), @intCast(y), @intCast(z), &dst);
+        const result = image_api.readImageFloat4(driver_image, dim, @intCast(x), @intCast(y), @intCast(z), &dst);
 
         try fromCResult(result);
 
@@ -99,11 +106,11 @@ const ImageAPIBridge = struct {
         };
     }
 
-    fn readImageInt4(driver_image: *anyopaque, x: i32, y: i32, z: i32) spv.Runtime.RuntimeError!spv.Runtime.Vec4(u32) {
+    fn readImageInt4(driver_image: *anyopaque, dim: spv.spv.SpvDim, x: i32, y: i32, z: i32) spv.Runtime.RuntimeError!spv.Runtime.Vec4(u32) {
         const image_api = try getImageAPI();
 
         var dst: Vec4u = undefined;
-        const result = image_api.readImageInt4(driver_image, @intCast(x), @intCast(y), @intCast(z), &dst);
+        const result = image_api.readImageInt4(driver_image, dim, @intCast(x), @intCast(y), @intCast(z), &dst);
 
         try fromCResult(result);
 
@@ -115,21 +122,53 @@ const ImageAPIBridge = struct {
         };
     }
 
-    fn writeImageFloat4(driver_image: *anyopaque, x: i32, y: i32, z: i32, pixel: spv.Runtime.Vec4(f32)) spv.Runtime.RuntimeError!void {
+    fn writeImageFloat4(driver_image: *anyopaque, dim: spv.spv.SpvDim, x: i32, y: i32, z: i32, pixel: spv.Runtime.Vec4(f32)) spv.Runtime.RuntimeError!void {
         const image_api = try getImageAPI();
 
-        const result = image_api.writeImageFloat4(driver_image, @intCast(x), @intCast(y), @intCast(z), .{ .x = pixel.x, .y = pixel.y, .z = pixel.z, .w = pixel.w });
+        const result = image_api.writeImageFloat4(driver_image, dim, @intCast(x), @intCast(y), @intCast(z), .{ .x = pixel.x, .y = pixel.y, .z = pixel.z, .w = pixel.w });
 
         try fromCResult(result);
     }
 
-    fn writeImageInt4(driver_image: *anyopaque, x: i32, y: i32, z: i32, pixel: spv.Runtime.Vec4(u32)) spv.Runtime.RuntimeError!void {
+    fn writeImageInt4(driver_image: *anyopaque, dim: spv.spv.SpvDim, x: i32, y: i32, z: i32, pixel: spv.Runtime.Vec4(u32)) spv.Runtime.RuntimeError!void {
         const image_api = try getImageAPI();
 
-        const result = image_api.writeImageInt4(driver_image, @intCast(x), @intCast(y), @intCast(z), .{ .x = pixel.x, .y = pixel.y, .z = pixel.z, .w = pixel.w });
+        const result = image_api.writeImageInt4(driver_image, dim, @intCast(x), @intCast(y), @intCast(z), .{ .x = pixel.x, .y = pixel.y, .z = pixel.z, .w = pixel.w });
 
         try fromCResult(result);
     }
+
+    fn sampleImageFloat4(driver_image: *anyopaque, driver_sampler: *anyopaque, dim: spv.spv.SpvDim, x: f32, y: f32, z: f32) spv.Runtime.RuntimeError!spv.Runtime.Vec4(f32) {
+        const image_api = try getImageAPI();
+
+        var dst: Vec4f = undefined;
+        const result = image_api.sampleImageFloat4(driver_image, driver_sampler, dim, x, y, z, &dst);
+
+        try fromCResult(result);
+
+        return .{
+            .x = dst.x,
+            .y = dst.y,
+            .z = dst.z,
+            .w = dst.w,
+        };
+    }
+
+    fn queryImageSize(driver_image: *anyopaque, dim: spv.spv.SpvDim, arrayed: bool) spv.Runtime.RuntimeError!spv.Runtime.Vec4(u32) {
+        const image_api = try getImageAPI();
+
+        var dst: Vec4u = undefined;
+        const result = image_api.queryImageSize(driver_image, dim, if (arrayed) 1 else 0, &dst);
+
+        try fromCResult(result);
+
+        return .{
+            .x = dst.x,
+            .y = dst.y,
+            .z = dst.z,
+            .w = dst.w,
+        };
+    }
 };
 
 const RuntimeWrapper = struct {
@@ -151,6 +190,8 @@ export fn SpvInitRuntime(rt: **RuntimeWrapper, module: *spv.Module, image_api: I
             .readImageInt4 = ImageAPIBridge.readImageInt4,
             .writeImageFloat4 = ImageAPIBridge.writeImageFloat4,
             .writeImageInt4 = ImageAPIBridge.writeImageInt4,
+            .sampleImageFloat4 = ImageAPIBridge.sampleImageFloat4,
+            .queryImageSize = ImageAPIBridge.queryImageSize,
         },
     ) catch |err| {
         allocator.destroy(rt.*);

src/GLSL_std_450/opcodes.zig

@@ -84,6 +84,7 @@ pub fn initRuntimeDispatcher() void {
     runtime_dispatcher[@intFromEnum(ext.GLSLOp.Exp)]       = MathEngine(.Float, .Exp).opSingleOperator;
     runtime_dispatcher[@intFromEnum(ext.GLSLOp.Exp2)]      = MathEngine(.Float, .Exp2).opSingleOperator;
     runtime_dispatcher[@intFromEnum(ext.GLSLOp.FAbs)]      = MathEngine(.Float, .FAbs).opSingleOperator;
+    runtime_dispatcher[@intFromEnum(ext.GLSLOp.FClamp)]    = MathEngine(.Float, .FClamp).opTripleOperators;
     runtime_dispatcher[@intFromEnum(ext.GLSLOp.FMax)]      = MathEngine(.Float, .FMax).opDoubleOperators;
     runtime_dispatcher[@intFromEnum(ext.GLSLOp.FSign)]     = MathEngine(.Float, .FSign).opSingleOperator;
     runtime_dispatcher[@intFromEnum(ext.GLSLOp.Floor)]     = MathEngine(.Float, .Floor).opSingleOperator;
@@ -93,11 +94,13 @@ pub fn initRuntimeDispatcher() void {
     runtime_dispatcher[@intFromEnum(ext.GLSLOp.Normalize)] = opNormalize;
     runtime_dispatcher[@intFromEnum(ext.GLSLOp.Round)]     = MathEngine(.Float, .Round).opSingleOperator;
     runtime_dispatcher[@intFromEnum(ext.GLSLOp.SAbs)]      = MathEngine(.SInt,  .SAbs).opSingleOperator;
+    runtime_dispatcher[@intFromEnum(ext.GLSLOp.SClamp)]    = MathEngine(.SInt, .SClamp).opTripleOperators;
     runtime_dispatcher[@intFromEnum(ext.GLSLOp.SSign)]     = MathEngine(.SInt, .SSign).opSingleOperator;
     runtime_dispatcher[@intFromEnum(ext.GLSLOp.Sin)]       = MathEngine(.Float, .Sin).opSingleOperator;
     runtime_dispatcher[@intFromEnum(ext.GLSLOp.Sqrt)]      = MathEngine(.Float, .Sqrt).opSingleOperator;
     runtime_dispatcher[@intFromEnum(ext.GLSLOp.Tan)]       = MathEngine(.Float, .Tan).opSingleOperator;
     runtime_dispatcher[@intFromEnum(ext.GLSLOp.Trunc)]     = MathEngine(.Float, .Trunc).opSingleOperator;
+    runtime_dispatcher[@intFromEnum(ext.GLSLOp.UClamp)]    = MathEngine(.UInt, .UClamp).opTripleOperators;
     runtime_dispatcher[@intFromEnum(ext.GLSLOp.FindILsb)]  = IntBitEngine(.FindILsb).op;
     runtime_dispatcher[@intFromEnum(ext.GLSLOp.FindSMsb)]  = IntBitEngine(.FindSMsb).op;
     runtime_dispatcher[@intFromEnum(ext.GLSLOp.FindUMsb)]  = IntBitEngine(.FindUMsb).op;
@@ -242,6 +245,63 @@ fn MathEngine(comptime T: PrimitiveType, comptime Op: MathOp) type {
                 else => return RuntimeError.InvalidSpirV,
             }
         }
+
+        fn opTripleOperators(_: std.mem.Allocator, target_type_id: SpvWord, id: SpvWord, _: SpvWord, rt: *Runtime) RuntimeError!void {
+            const target_type = (try rt.results[target_type_id].getVariant()).Type;
+            const dst = try rt.results[id].getValue();
+            const x = try rt.results[try rt.it.next()].getValue();
+            const min_val = try rt.results[try rt.it.next()].getValue();
+            const max_val = try rt.results[try rt.it.next()].getValue();
+
+            const lane_bits = try Result.resolveLaneBitWidth(target_type, rt);
+
+            const operator = struct {
+                fn operation(comptime TT: type, v: TT, lo: TT, hi: TT) RuntimeError!TT {
+                    return switch (Op) {
+                        .FClamp, .SClamp, .UClamp => @min(@max(v, lo), hi),
+                        else => RuntimeError.InvalidSpirV,
+                    };
+                }
+
+                fn applyScalar(bit_count: SpvWord, d: *Value, v: *const Value, lo: *const Value, hi: *const Value) RuntimeError!void {
+                    switch (bit_count) {
+                        inline 8, 16, 32, 64 => |bits| {
+                            if (bits == 8 and T == .Float) return RuntimeError.InvalidSpirV;
+
+                            const ScalarT = Value.getPrimitiveFieldType(T, bits);
+                            const d_field = try Value.getPrimitiveField(T, bits, d);
+                            const v_field = try Value.getPrimitiveField(T, bits, @constCast(v));
+                            const lo_field = try Value.getPrimitiveField(T, bits, @constCast(lo));
+                            const hi_field = try Value.getPrimitiveField(T, bits, @constCast(hi));
+                            d_field.* = try operation(ScalarT, v_field.*, lo_field.*, hi_field.*);
+                        },
+                        else => return RuntimeError.InvalidSpirV,
+                    }
+                }
+            };
+
+            switch (dst.*) {
+                .Int, .Float => try operator.applyScalar(lane_bits, dst, x, min_val, max_val),
+
+                .Vector => |dst_vec| for (dst_vec, x.Vector, min_val.Vector, max_val.Vector) |*d_lane, x_lane, min_lane, max_lane| {
+                    try operator.applyScalar(lane_bits, d_lane, &x_lane, &min_lane, &max_lane);
+                },
+
+                .Vector4f32 => |*d| d.* = try operator.operation(@Vector(4, f32), x.Vector4f32, min_val.Vector4f32, max_val.Vector4f32),
+                .Vector3f32 => |*d| d.* = try operator.operation(@Vector(3, f32), x.Vector3f32, min_val.Vector3f32, max_val.Vector3f32),
+                .Vector2f32 => |*d| d.* = try operator.operation(@Vector(2, f32), x.Vector2f32, min_val.Vector2f32, max_val.Vector2f32),
+
+                .Vector4i32 => |*d| d.* = try operator.operation(@Vector(4, i32), x.Vector4i32, min_val.Vector4i32, max_val.Vector4i32),
+                .Vector3i32 => |*d| d.* = try operator.operation(@Vector(3, i32), x.Vector3i32, min_val.Vector3i32, max_val.Vector3i32),
+                .Vector2i32 => |*d| d.* = try operator.operation(@Vector(2, i32), x.Vector2i32, min_val.Vector2i32, max_val.Vector2i32),
+
+                .Vector4u32 => |*d| d.* = try operator.operation(@Vector(4, u32), x.Vector4u32, min_val.Vector4u32, max_val.Vector4u32),
+                .Vector3u32 => |*d| d.* = try operator.operation(@Vector(3, u32), x.Vector3u32, min_val.Vector3u32, max_val.Vector3u32),
+                .Vector2u32 => |*d| d.* = try operator.operation(@Vector(2, u32), x.Vector2u32, min_val.Vector2u32, max_val.Vector2u32),
+
+                else => return RuntimeError.InvalidSpirV,
+            }
+        }
     };
 }
 

src/Module.zig

@@ -9,8 +9,6 @@ const SpvByte = spv.SpvByte;
 const SpvWord = spv.SpvWord;
 const SpvBool = spv.SpvBool;
 
-const SpvBinding = spv.SpvBinding;
-
 const Result = @import("Result.zig");
 const Runtime = @import("Runtime.zig");
 const Value = @import("Value.zig").Value;
@@ -19,6 +17,7 @@ const WordIterator = @import("WordIterator.zig");
 const Self = @This();
 
 pub const ModuleOptions = struct {
+    /// Also affects matrices
     use_simd_vectors_specializations: bool = true,
 };
 
@@ -29,6 +28,12 @@ const SpvEntryPoint = struct {
     globals: []SpvWord,
 };
 
+const BindingEntry = struct {
+    set: SpvWord,
+    binding: SpvWord,
+    result: SpvWord,
+};
+
 pub const ModuleError = error{
     InvalidSpirV,
     InvalidMagic,
@@ -72,9 +77,8 @@ geometry_output: SpvWord,
 
 input_locations: [lib.SPIRV_MAX_INPUT_LOCATIONS]SpvWord,
 output_locations: [lib.SPIRV_MAX_OUTPUT_LOCATIONS]SpvWord,
-bindings: [lib.SPIRV_MAX_SET][lib.SPIRV_MAX_SET_BINDINGS]SpvWord,
+bindings: std.ArrayList(BindingEntry),
 builtins: std.EnumMap(spv.SpvBuiltIn, SpvWord),
-push_constants: []Value,
 
 pub fn init(allocator: std.mem.Allocator, source: []const SpvWord, options: ModuleOptions) ModuleError!Self {
     var self: Self = std.mem.zeroInit(Self, .{
@@ -82,12 +86,14 @@ pub fn init(allocator: std.mem.Allocator, source: []const SpvWord, options: Modu
         .code = allocator.dupe(SpvWord, source) catch return ModuleError.OutOfMemory,
         .extensions = std.ArrayList([]const u8).empty,
         .entry_points = std.ArrayList(SpvEntryPoint).empty,
+        .bindings = std.ArrayList(BindingEntry).empty,
         .capabilities = std.EnumSet(spv.SpvCapability).initEmpty(),
         .local_size_x = 1,
         .local_size_y = 1,
         .local_size_z = 1,
     });
     errdefer allocator.free(self.code);
+    errdefer self.bindings.deinit(allocator);
 
     op.initRuntimeDispatcher();
 
@@ -125,7 +131,7 @@ pub fn init(allocator: std.mem.Allocator, source: []const SpvWord, options: Modu
     _ = self.it.skip(); // Skip schema
 
     try self.pass(allocator); // Setup pass
-    try self.applyDecorations();
+    try self.applyDecorations(allocator);
 
     return self;
 }
@@ -180,7 +186,7 @@ fn resolveConstantWord(self: *const Self, id: SpvWord) ?SpvWord {
 }
 
 fn findAccessChainToMember(self: *const Self, base_id: SpvWord, member_index: SpvWord) ?SpvWord {
-    for (self.results, 0..) |result, id| {
+    for (self.results, 0..) |*result, id| {
         const variant = result.variant orelse continue;
 
         switch (variant) {
@@ -197,12 +203,7 @@ fn findAccessChainToMember(self: *const Self, base_id: SpvWord, member_index: Sp
     return null;
 }
 
-fn applyInterfaceDecoration(
-    self: *Self,
-    storage_class: spv.SpvStorageClass,
-    decoration: Result.Decoration,
-    id: SpvWord,
-) ModuleError!void {
+fn applyInterfaceDecoration(self: *Self, storage_class: spv.SpvStorageClass, decoration: Result.Decoration, id: SpvWord) ModuleError!void {
     switch (storage_class) {
         .Input => switch (decoration.rtype) {
             .BuiltIn => self.builtins.put(
@@ -224,12 +225,7 @@ fn applyInterfaceDecoration(
     }
 }
 
-fn applyStructMemberInterfaceDecorations(
-    self: *Self,
-    storage_class: spv.SpvStorageClass,
-    type_word: SpvWord,
-    id: SpvWord,
-) ModuleError!void {
+fn applyStructMemberInterfaceDecorations(self: *Self, storage_class: spv.SpvStorageClass, type_word: SpvWord, id: SpvWord) ModuleError!void {
     switch (storage_class) {
         .Input, .Output => {},
         else => return,
@@ -266,21 +262,24 @@ fn applyStructMemberInterfaceDecorations(
     }
 }
 
-fn applyDecorations(self: *Self) ModuleError!void {
+fn applyDecorations(self: *Self, allocator: std.mem.Allocator) ModuleError!void {
     for (self.results, 0..) |result, id| {
         if (result.variant == null)
             continue;
 
-        var set: ?usize = null;
-        var binding: ?usize = null;
+        var set: ?SpvWord = null;
+        var binding: ?SpvWord = null;
 
         for (result.decorations.items) |decoration| {
-            switch (result.variant.?) {
+            if (result.variant) |*variant| switch (variant.*) {
                 .Variable => |v| {
                     try self.applyInterfaceDecoration(v.storage_class, decoration, @intCast(id));
 
                     switch (v.storage_class) {
-                        .StorageBuffer, .Uniform, .UniformConstant => {
+                        .StorageBuffer,
+                        .Uniform,
+                        .UniformConstant,
+                        => {
                             switch (decoration.rtype) {
                                 .Binding => binding = decoration.literal_1,
                                 .DescriptorSet => set = decoration.literal_1,
@@ -301,20 +300,50 @@ fn applyDecorations(self: *Self) ModuleError!void {
                     }
                 },
                 else => {},
-            }
+            };
         }
 
-        switch (result.variant.?) {
-            .Variable => |v| try self.applyStructMemberInterfaceDecorations(v.storage_class, v.type_word, @intCast(id)),
+        if (result.variant) |*variant| switch (variant.*) {
+            .Variable => |*v| {
+                try self.applyStructMemberInterfaceDecorations(v.storage_class, v.type_word, @intCast(id));
+                switch (v.storage_class) {
+                    .StorageBuffer,
+                    .Uniform,
+                    .PushConstant,
+                    => if (v.value == .Structure) {
+                        if (self.results[v.type_word].variant) |type_variant| switch (type_variant) {
+                            .Type => |type_data| switch (type_data) {
+                                .Structure => |s| @memcpy(@constCast(v.value.Structure.offsets), s.members_offsets),
+                                else => {},
+                            },
+                            else => {},
+                        };
+                    },
                     else => {},
                 }
+            },
+            else => {},
+        };
 
         if (set != null and binding != null) {
-            self.bindings[set.?][binding.?] = @intCast(id);
+            self.bindings.append(allocator, .{
+                .set = set.?,
+                .binding = binding.?,
+                .result = @intCast(id),
+            }) catch return ModuleError.OutOfMemory;
         }
     }
 }
 
+pub fn getBindingResult(self: *const Self, set: SpvWord, binding: SpvWord) ?SpvWord {
+    for (self.bindings.items) |entry| {
+        if (entry.set == set and entry.binding == binding) {
+            return entry.result;
+        }
+    }
+    return null;
+}
+
 pub fn deinit(self: *Self, allocator: std.mem.Allocator) void {
     allocator.free(self.code);
     for (self.entry_points.items) |entry| {
@@ -327,6 +356,7 @@ pub fn deinit(self: *Self, allocator: std.mem.Allocator) void {
         allocator.free(ext);
     }
     self.extensions.deinit(allocator);
+    self.bindings.deinit(allocator);
 
     for (self.results) |*result| {
         result.deinit(allocator);

src/Result.zig

@@ -139,7 +139,9 @@ pub const TypeData = union(Type) {
         access: ?spv.SpvAccessQualifier,
     },
     Sampler: struct {},
-    SampledImage: struct {},
+    SampledImage: struct {
+        image_type: SpvWord,
+    },
     Pointer: struct {
         storage_class: spv.SpvStorageClass,
         target: SpvWord,
@@ -390,6 +392,7 @@ pub fn resolveLaneBitWidth(target_type: TypeData, rt: *const Runtime) RuntimeErr
         .Float => |f| f.bit_length,
         .Int => |i| i.bit_length,
         .Vector => |v| continue :sw (try rt.results[v.components_type_word].getVariant()).Type,
+        .Matrix => |m| continue :sw (try rt.results[m.column_type_word].getVariant()).Type,
         .Vector4f32,
         .Vector3f32,
         .Vector2f32,
@@ -408,6 +411,7 @@ pub fn resolveLaneCount(target_type: TypeData) RuntimeError!SpvWord {
     return switch (target_type) {
         .Bool, .Float, .Int => 1,
         .Vector => |v| v.member_count,
+        .Matrix => |m| m.member_count,
         .Vector4f32, .Vector4i32, .Vector4u32 => 4,
         .Vector3f32, .Vector3i32, .Vector3u32 => 3,
         .Vector2f32, .Vector2i32, .Vector2u32 => 2,
@@ -419,6 +423,7 @@ pub fn resolveSign(target_type: TypeData, rt: *const Runtime) RuntimeError!enum
     return sw: switch (target_type) {
         .Int => |i| if (i.is_signed) .signed else .unsigned,
         .Vector => |v| continue :sw (try rt.results[v.components_type_word].getVariant()).Type,
+        .Matrix => |m| continue :sw (try rt.results[m.column_type_word].getVariant()).Type,
         .Vector4i32 => .signed,
         .Vector3i32 => .signed,
         .Vector2i32 => .signed,
@@ -457,7 +462,6 @@ pub fn getMemberCounts(self: *const Self) usize {
                 .Vector2f32, .Vector2i32, .Vector2u32 => return 2,
                 .Matrix => |m| return m.member_count,
                 .Array => |a| return a.member_count,
-                .SampledImage => return 2,
                 .Structure => |s| return s.members_type_word.len,
                 .Function => |f| return f.params.len,
                 else => {},

src/Runtime.zig

@@ -18,6 +18,7 @@ const WordIterator = @import("WordIterator.zig");
 const Self = @This();
 
 pub const RuntimeError = error{
+    Barrier,
     DivisionByZero,
     InvalidEntryPoint,
     InvalidSpirV,
@@ -33,6 +34,11 @@ pub const RuntimeError = error{
     Unknown,
 };
 
+pub const EntryPointStatus = enum {
+    completed,
+    barrier,
+};
+
 pub const SpecializationEntry = struct {
     id: SpvWord,
     offset: usize,
@@ -55,10 +61,12 @@ pub fn Vec4(comptime T: type) type {
 }
 
 pub const ImageAPI = struct {
-    readImageFloat4: *const fn (driver_image: *anyopaque, x: i32, y: i32, z: i32) RuntimeError!Vec4(f32),
-    readImageInt4: *const fn (driver_image: *anyopaque, x: i32, y: i32, z: i32) RuntimeError!Vec4(u32),
-    writeImageFloat4: *const fn (driver_image: *anyopaque, x: i32, y: i32, z: i32, pixel: Vec4(f32)) RuntimeError!void,
-    writeImageInt4: *const fn (driver_image: *anyopaque, x: i32, y: i32, z: i32, pixel: Vec4(u32)) RuntimeError!void,
+    readImageFloat4: *const fn (driver_image: *anyopaque, dim: spv.SpvDim, x: i32, y: i32, z: i32) RuntimeError!Vec4(f32),
+    readImageInt4: *const fn (driver_image: *anyopaque, dim: spv.SpvDim, x: i32, y: i32, z: i32) RuntimeError!Vec4(u32),
+    writeImageFloat4: *const fn (driver_image: *anyopaque, dim: spv.SpvDim, x: i32, y: i32, z: i32, pixel: Vec4(f32)) RuntimeError!void,
+    writeImageInt4: *const fn (driver_image: *anyopaque, dim: spv.SpvDim, x: i32, y: i32, z: i32, pixel: Vec4(u32)) RuntimeError!void,
+    sampleImageFloat4: *const fn (driver_image: *anyopaque, driver_sampler: *anyopaque, dim: spv.SpvDim, x: f32, y: f32, z: f32) RuntimeError!Vec4(f32),
+    queryImageSize: *const fn (driver_image: *anyopaque, dim: spv.SpvDim, arrayed: bool) RuntimeError!Vec4(u32),
 };
 
 mod: *Module,
@@ -117,6 +125,17 @@ pub fn addSpecializationInfo(self: *Self, allocator: std.mem.Allocator, entry: S
     self.specialization_constants.put(allocator, entry.id, slice) catch return RuntimeError.OutOfMemory;
 }
 
+pub fn copySpecializationConstantsFrom(self: *Self, allocator: std.mem.Allocator, other: *const Self) RuntimeError!void {
+    var it = other.specialization_constants.iterator();
+    while (it.next()) |entry| {
+        const slice = allocator.dupe(u8, entry.value_ptr.*) catch return RuntimeError.OutOfMemory;
+        self.specialization_constants.put(allocator, entry.key_ptr.*, slice) catch {
+            allocator.free(slice);
+            return RuntimeError.OutOfMemory;
+        };
+    }
+}
+
 pub fn getEntryPointByName(self: *const Self, name: []const u8) RuntimeError!SpvWord {
     for (self.mod.entry_points.items, 0..) |entry_point, i| {
         if (blk: {
@@ -174,8 +193,11 @@ pub fn dumpResultsTable(self: *Self, allocator: std.mem.Allocator, writer: *std.
 
 /// Calls an entry point, `entry_point_index` being the index of the entry point ordered by declaration in the bytecode
 pub fn callEntryPoint(self: *Self, allocator: std.mem.Allocator, entry_point_index: SpvWord) RuntimeError!void {
-    self.reset();
+    _ = try self.beginEntryPoint(allocator, entry_point_index);
+}
 
+pub fn beginEntryPoint(self: *Self, allocator: std.mem.Allocator, entry_point_index: SpvWord) RuntimeError!EntryPointStatus {
+    self.reset();
     if (entry_point_index > self.mod.entry_points.items.len)
         return RuntimeError.InvalidEntryPoint;
 
@@ -210,13 +232,23 @@ pub fn callEntryPoint(self: *Self, allocator: std.mem.Allocator, entry_point_ind
     }
 
     // Execution pass
-    try self.pass(allocator, null);
+    return self.continueEntryPoint(allocator);
+}
+
+pub fn continueEntryPoint(self: *Self, allocator: std.mem.Allocator) RuntimeError!EntryPointStatus {
+    self.pass(allocator, null) catch |err| switch (err) {
+        RuntimeError.Barrier => return .barrier,
+        else => return err,
+    };
+    return .completed;
 }
 
 fn pass(self: *Self, allocator: std.mem.Allocator, op_set: ?std.EnumSet(spv.SpvOp)) RuntimeError!void {
     self.it.did_jump = false; // To reset function jump
     while (self.it.nextOrNull()) |opcode_data| {
-        const word_count = ((opcode_data & (~spv.SpvOpCodeMask)) >> spv.SpvWordCountShift) - 1;
+        const word_count_with_header = (opcode_data & (~spv.SpvOpCodeMask)) >> spv.SpvWordCountShift;
+        if (word_count_with_header == 0) return RuntimeError.InvalidSpirV;
+        const word_count = word_count_with_header - 1;
         const opcode = (opcode_data & spv.SpvOpCodeMask);
 
         if (op_set) |set| {
@@ -240,24 +272,32 @@ fn pass(self: *Self, allocator: std.mem.Allocator, op_set: ?std.EnumSet(spv.SpvO
     }
 }
 
+pub fn populatePushConstants(self: *Self, blob: []const u8) RuntimeError!void {
+    for (self.results) |*result| {
+        if (result.variant == null or std.meta.activeTag(result.variant.?) != .Variable)
+            continue;
+        const variable = &result.variant.?.Variable;
+        if (variable.storage_class != .PushConstant)
+            continue;
+        _ = try variable.value.write(blob);
+    }
+}
+
 pub fn writeDescriptorSet(self: *const Self, input: []const u8, set: SpvWord, binding: SpvWord, descriptor_index: SpvWord) RuntimeError!void {
-    if (set < lib.SPIRV_MAX_SET and binding < lib.SPIRV_MAX_SET_BINDINGS) {
-        const value = &self.results[self.mod.bindings[set][binding]].variant.?.Variable.value;
+    const result = self.mod.getBindingResult(set, binding) orelse return RuntimeError.NotFound;
+    const value = &(self.results[result].variant orelse return).Variable.value;
     switch (value.*) {
         .Array => |arr| {
             if (descriptor_index >= arr.values.len)
                 return RuntimeError.NotFound;
-                _ = try arr.values[descriptor_index].writeConst(input);
+            _ = try arr.values[descriptor_index].write(input);
         },
         else => {
             if (descriptor_index != 0)
                 return RuntimeError.NotFound;
-                _ = try value.writeConst(input);
+            _ = try value.write(input);
         },
     }
-    } else {
-        return RuntimeError.NotFound;
-    }
 }
 
 fn readResultValue(self: *const Self, output: []u8, result: SpvWord) RuntimeError!void {
@@ -280,15 +320,15 @@ fn readResultValue(self: *const Self, output: []u8, result: SpvWord) RuntimeErro
 fn writeResultValue(self: *const Self, input: []const u8, result: SpvWord) RuntimeError!void {
     if (self.results[result].variant) |*variant| {
         switch (variant.*) {
-            .Variable => |*v| _ = try v.value.writeConst(input),
+            .Variable => |*v| _ = try v.value.write(input),
             .AccessChain => |*a| switch (a.value) {
                 .Pointer => |ptr| switch (ptr.ptr) {
-                    .common => |value_ptr| _ = try value_ptr.writeConst(input),
+                    .common => |value_ptr| _ = try value_ptr.write(input),
                     .f32_ptr => |value_ptr| std.mem.copyForwards(u8, std.mem.asBytes(value_ptr), input[0..@sizeOf(f32)]),
                     .i32_ptr => |value_ptr| std.mem.copyForwards(u8, std.mem.asBytes(value_ptr), input[0..@sizeOf(i32)]),
                     .u32_ptr => |value_ptr| std.mem.copyForwards(u8, std.mem.asBytes(value_ptr), input[0..@sizeOf(u32)]),
                 },
-                else => _ = try a.value.writeConst(input),
+                else => _ = try a.value.write(input),
             },
             else => return RuntimeError.InvalidSpirV,
         }

src/Value.zig

@@ -1,5 +1,6 @@
 const std = @import("std");
 const lib = @import("lib.zig");
+const spv = @import("spv.zig");
 
 const Result = @import("Result.zig");
 const Runtime = @import("Runtime.zig");
@@ -83,14 +84,14 @@ pub const Value = union(Type) {
             errdefer allocator.destroy(value);
 
             value.* = try Value.init(allocator, results, self.type_word, false);
-            _ = try value.writeConst(self.data[self.getOffsetOfIndex(index)..]);
+            _ = try value.write(self.data[self.getOffsetOfIndex(index)..]);
 
             return value;
         }
 
         pub inline fn createLocalValueFromIndex(self: *const @This(), allocator: std.mem.Allocator, results: []const Result, index: usize) RuntimeError!Value {
             var value = try Value.init(allocator, results, self.type_word, false);
-            _ = try value.writeConst(self.data[self.getOffsetOfIndex(index)..]);
+            _ = try value.write(self.data[self.getOffsetOfIndex(index)..]);
             return value;
         }
 
@@ -103,6 +104,7 @@ pub const Value = union(Type) {
         }
     },
     Structure: struct {
+        external_data: ?[]u8,
         offsets: []const ?SpvWord,
         values: []Self,
     },
@@ -111,8 +113,14 @@ pub const Value = union(Type) {
         type_word: SpvWord,
         driver_image: *anyopaque,
     },
-    Sampler: struct {},
-    SampledImage: struct {},
+    Sampler: struct {
+        driver_sampler: *anyopaque,
+    },
+    SampledImage: struct {
+        type_word: SpvWord,
+        driver_image: *anyopaque,
+        driver_sampler: *anyopaque,
+    },
     Pointer: struct {
         ptr: union(enum) {
             common: *Self,
@@ -120,24 +128,43 @@ pub const Value = union(Type) {
             i32_ptr: *i32, //< For vector specializations
             u32_ptr: *u32,
         },
-        is_owner_of_uniform_slice: bool = false,
+        image_texel: ?struct {
+            driver_image: *anyopaque,
+            dim: spv.SpvDim,
+            x: i32,
+            y: i32,
+            z: i32,
+        } = null,
+
+        /// Exact byte window in externally visible descriptor storage that
+        /// corresponds to this pointer. For a pointer to struct member N this
+        /// starts at the member offset, not at the containing struct.
         uniform_slice_window: ?[]u8 = null,
+
+        /// Heap-owned value that backs a pointer into a materialized runtime
+        /// array element. This may differ from ptr.common when the pointer is
+        /// to a child/member of that materialized value.
+        uniform_backing_value: ?*Self = null,
     },
 
     pub inline fn getCompositeDataOrNull(self: *const Self) ?[]Self {
         return switch (self.*) {
             .Structure => |*s| s.values,
             .Array => |*a| a.values,
-            .Vector, .Matrix => |v| v,
+            .Vector => |v| v,
             else => null,
         };
     }
 
     pub fn init(allocator: std.mem.Allocator, results: []const Result, target_type: SpvWord, is_externally_visible: bool) RuntimeError!Self {
         const resolved = results[target_type].resolveTypeWordOrNull() orelse target_type;
-        const member_count = results[resolved].getMemberCounts();
+        return initUnresolved(allocator, results, resolved, is_externally_visible);
+    }
 
-        return switch (results[resolved].variant.?) {
+    pub fn initUnresolved(allocator: std.mem.Allocator, results: []const Result, target_type: SpvWord, is_externally_visible: bool) RuntimeError!Self {
+        const member_count = results[target_type].getMemberCounts();
+
+        return switch (results[target_type].variant.?) {
             .Type => |t| switch (t) {
                 .Void => .{ .Void = .{} },
                 .Bool => .{ .Bool = false },
@@ -159,13 +186,13 @@ pub const Value = union(Type) {
                     }
                     break :blk self;
                 },
-                .Vector4f32 => .{ .Vector4f32 = Vec4f32{ 0.0, 0.0, 0.0, 0.0 } },
+                .Vector4f32 => .{ .Vector4f32 = Vec4f32{ 0.0, 0.0, 0.0, 1.0 } },
                 .Vector3f32 => .{ .Vector3f32 = Vec3f32{ 0.0, 0.0, 0.0 } },
                 .Vector2f32 => .{ .Vector2f32 = Vec2f32{ 0.0, 0.0 } },
-                .Vector4i32 => .{ .Vector4i32 = Vec4i32{ 0, 0, 0, 0 } },
+                .Vector4i32 => .{ .Vector4i32 = Vec4i32{ 0, 0, 0, 1 } },
                 .Vector3i32 => .{ .Vector3i32 = Vec3i32{ 0, 0, 0 } },
                 .Vector2i32 => .{ .Vector2i32 = Vec2i32{ 0, 0 } },
-                .Vector4u32 => .{ .Vector4u32 = Vec4u32{ 0, 0, 0, 0 } },
+                .Vector4u32 => .{ .Vector4u32 = Vec4u32{ 0, 0, 0, 1 } },
                 .Vector3u32 => .{ .Vector3u32 = Vec3u32{ 0, 0, 0 } },
                 .Vector2u32 => .{ .Vector2u32 = Vec2u32{ 0, 0 } },
                 .Matrix => |m| blk: {
@@ -205,6 +232,7 @@ pub const Value = union(Type) {
                 .Structure => |s| blk: {
                     const self: Self = .{
                         .Structure = .{
+                            .external_data = null,
                             .offsets = allocator.dupe(?SpvWord, s.members_offsets) catch return RuntimeError.OutOfMemory,
                             .values = allocator.alloc(Self, member_count) catch return RuntimeError.OutOfMemory,
                         },
@@ -225,12 +253,27 @@ pub const Value = union(Type) {
                 },
                 .Image => .{
                     .Image = .{
-                        .type_word = resolved,
+                        .type_word = target_type,
                         .driver_image = undefined,
                     },
                 },
-                .Sampler => RuntimeError.ToDo,
-                .SampledImage => RuntimeError.ToDo,
+                .Sampler => .{
+                    .Sampler = .{
+                        .driver_sampler = undefined,
+                    },
+                },
+                .SampledImage => .{
+                    .SampledImage = .{
+                        .type_word = target_type,
+                        .driver_image = undefined,
+                        .driver_sampler = undefined,
+                    },
+                },
+                .Pointer => .{
+                    .Pointer = .{
+                        .ptr = undefined,
+                    },
+                },
                 else => RuntimeError.InvalidSpirV,
             },
             else => RuntimeError.InvalidSpirV,
@@ -273,6 +316,7 @@ pub const Value = union(Type) {
                     errdefer allocator.free(values);
                     for (values, s.values) |*new_value, value| new_value.* = try value.dupe(allocator);
                     break :blk .{
+                        .external_data = s.external_data,
                         .offsets = allocator.dupe(?SpvWord, s.offsets) catch return RuntimeError.OutOfMemory,
                         .values = values,
                     };
@@ -283,6 +327,23 @@ pub const Value = union(Type) {
     }
 
     pub fn read(self: *const Self, output: []u8) RuntimeError!usize {
+        const vecRoutine = struct {
+            inline fn routine(comptime T: type, vec: T, out: []u8) RuntimeError!usize {
+                const size = @typeInfo(T).vector.len * 4;
+                if (size >= out.len) {
+                    const range = @typeInfo(T).vector.len;
+                    inline for (0..range) |i| {
+                        const start = i * 4;
+                        const end = (i + 1) * 4;
+                        if (start >= out.len or end > out.len) return i * 4;
+                        @memcpy(out[start..end], std.mem.asBytes(&vec[i]));
+                    }
+                }
+                std.mem.bytesAsValue(T, out[0..]).* = vec;
+                return size;
+            }
+        }.routine;
+
         switch (self.*) {
             .Bool => |b| {
                 output[0] = if (b == true) 1 else 0;
@@ -291,76 +352,35 @@ pub const Value = union(Type) {
             .Int => |i| {
                 switch (i.bit_count) {
                     8 => output[0] = @bitCast(i.value.uint8),
-                    16 => std.mem.copyForwards(u8, output[0..], std.mem.asBytes(&i.value.uint16)),
-                    32 => std.mem.copyForwards(u8, output[0..], std.mem.asBytes(&i.value.uint32)),
-                    64 => std.mem.copyForwards(u8, output[0..], std.mem.asBytes(&i.value.uint64)),
+                    16 => @memcpy(output[0..2], std.mem.asBytes(&i.value.uint16)),
+                    32 => @memcpy(output[0..4], std.mem.asBytes(&i.value.uint32)),
+                    64 => @memcpy(output[0..8], std.mem.asBytes(&i.value.uint64)),
                     else => return RuntimeError.InvalidValueType,
                 }
                 return @divExact(i.bit_count, 8);
             },
             .Float => |f| {
                 switch (f.bit_count) {
-                    16 => std.mem.copyForwards(u8, output[0..], std.mem.asBytes(&f.value.float16)),
-                    32 => std.mem.copyForwards(u8, output[0..], std.mem.asBytes(&f.value.float32)),
-                    64 => std.mem.copyForwards(u8, output[0..], std.mem.asBytes(&f.value.float64)),
+                    16 => @memcpy(output[0..2], std.mem.asBytes(&f.value.float16)),
+                    32 => @memcpy(output[0..4], std.mem.asBytes(&f.value.float32)),
+                    64 => @memcpy(output[0..8], std.mem.asBytes(&f.value.float64)),
                     else => return RuntimeError.InvalidValueType,
                 }
                 return @divExact(f.bit_count, 8);
             },
-            .Vector4f32 => |vec| {
-                inline for (0..4) |i| {
-                    std.mem.copyForwards(u8, output[(i * 4)..], std.mem.asBytes(&vec[i]));
-                }
-                return 4 * 4;
-            },
-            .Vector3f32 => |vec| {
-                inline for (0..3) |i| {
-                    std.mem.copyForwards(u8, output[(i * 4)..], std.mem.asBytes(&vec[i]));
-                }
-                return 3 * 4;
-            },
-            .Vector2f32 => |vec| {
-                inline for (0..2) |i| {
-                    std.mem.copyForwards(u8, output[(i * 4)..], std.mem.asBytes(&vec[i]));
-                }
-                return 2 * 4;
-            },
-            .Vector4i32 => |vec| {
-                inline for (0..4) |i| {
-                    std.mem.copyForwards(u8, output[(i * 4)..], std.mem.asBytes(&vec[i]));
-                }
-                return 4 * 4;
-            },
-            .Vector3i32 => |vec| {
-                inline for (0..3) |i| {
-                    std.mem.copyForwards(u8, output[(i * 4)..], std.mem.asBytes(&vec[i]));
-                }
-                return 3 * 4;
-            },
-            .Vector2i32 => |vec| {
-                inline for (0..2) |i| {
-                    std.mem.copyForwards(u8, output[(i * 4)..], std.mem.asBytes(&vec[i]));
-                }
-                return 2 * 4;
-            },
-            .Vector4u32 => |vec| {
-                inline for (0..4) |i| {
-                    std.mem.copyForwards(u8, output[(i * 4)..], std.mem.asBytes(&vec[i]));
-                }
-                return 4 * 4;
-            },
-            .Vector3u32 => |vec| {
-                inline for (0..3) |i| {
-                    std.mem.copyForwards(u8, output[(i * 4)..], std.mem.asBytes(&vec[i]));
-                }
-                return 3 * 4;
-            },
-            .Vector2u32 => |vec| {
-                inline for (0..2) |i| {
-                    std.mem.copyForwards(u8, output[(i * 4)..], std.mem.asBytes(&vec[i]));
-                }
-                return 2 * 4;
-            },
+
+            .Vector4f32 => |vec| return vecRoutine(@TypeOf(vec), vec, output),
+            .Vector3f32 => |vec| return vecRoutine(@TypeOf(vec), vec, output),
+            .Vector2f32 => |vec| return vecRoutine(@TypeOf(vec), vec, output),
+
+            .Vector4i32 => |vec| return vecRoutine(@TypeOf(vec), vec, output),
+            .Vector3i32 => |vec| return vecRoutine(@TypeOf(vec), vec, output),
+            .Vector2i32 => |vec| return vecRoutine(@TypeOf(vec), vec, output),
+
+            .Vector4u32 => |vec| return vecRoutine(@TypeOf(vec), vec, output),
+            .Vector3u32 => |vec| return vecRoutine(@TypeOf(vec), vec, output),
+            .Vector2u32 => |vec| return vecRoutine(@TypeOf(vec), vec, output),
+
             .Vector, .Matrix => |values| {
                 var offset: usize = 0;
                 for (values) |v| {
@@ -377,29 +397,38 @@ pub const Value = union(Type) {
                 return offset;
             },
             .Structure => |s| {
-                var offset: usize = 0;
+                var end_offset: usize = 0;
                 for (s.values, 0..) |v, i| {
-                    const read_size = try v.read(output[offset..]);
-                    if (i + 1 < s.offsets.len) {
-                        if (s.offsets[i + 1]) |o| {
-                            offset = o;
-                            continue;
+                    const member_offset: usize = @intCast(s.offsets[i] orelse end_offset);
+                    const read_size = try v.read(output[member_offset..]);
+                    end_offset = @max(end_offset, member_offset + read_size);
                 }
-                    }
-                    offset += read_size;
-                }
-                return offset;
+                return end_offset;
             },
+            .RuntimeArray => {},
             else => return RuntimeError.InvalidValueType,
         }
         return 0;
     }
 
-    pub fn writeConst(self: *Self, input: []const u8) RuntimeError!usize {
-        return self.write(@constCast(input));
+    pub fn write(self: *Self, input: []const u8) RuntimeError!usize {
+        const vecRoutine = struct {
+            inline fn routine(comptime T: type, vec: *T, in: []const u8) RuntimeError!usize {
+                const size = @typeInfo(T).vector.len * 4;
+                if (size >= in.len) {
+                    const range = @typeInfo(T).vector.len;
+                    inline for (0..range) |i| {
+                        const start = i * 4;
+                        const end = (i + 1) * 4;
+                        if (start >= in.len or end > in.len) return i * 4;
+                        @memcpy(std.mem.asBytes(&vec[i]), in[start..end]);
                     }
+                }
+                vec.* = std.mem.bytesToValue(T, in[0..]);
+                return size;
+            }
+        }.routine;
 
-    pub fn write(self: *Self, input: []u8) RuntimeError!usize {
         switch (self.*) {
             .Bool => |*b| {
                 b.* = if (input[0] != 0) true else false;
@@ -408,103 +437,35 @@ pub const Value = union(Type) {
             .Int => |*i| {
                 switch (i.bit_count) {
                     8 => i.value.uint8 = @bitCast(input[0]),
-                    16 => std.mem.copyForwards(u8, std.mem.asBytes(&i.value.uint16), input[0..2]),
-                    32 => std.mem.copyForwards(u8, std.mem.asBytes(&i.value.uint32), input[0..4]),
-                    64 => std.mem.copyForwards(u8, std.mem.asBytes(&i.value.uint64), input[0..8]),
+                    16 => @memcpy(std.mem.asBytes(&i.value.uint16), input[0..2]),
+                    32 => @memcpy(std.mem.asBytes(&i.value.uint32), input[0..4]),
+                    64 => @memcpy(std.mem.asBytes(&i.value.uint64), input[0..8]),
                     else => return RuntimeError.InvalidValueType,
                 }
                 return @divExact(i.bit_count, 8);
             },
             .Float => |*f| {
                 switch (f.bit_count) {
-                    16 => std.mem.copyForwards(u8, std.mem.asBytes(&f.value.float16), input[0..2]),
-                    32 => std.mem.copyForwards(u8, std.mem.asBytes(&f.value.float32), input[0..4]),
-                    64 => std.mem.copyForwards(u8, std.mem.asBytes(&f.value.float64), input[0..8]),
+                    16 => @memcpy(std.mem.asBytes(&f.value.float16), input[0..2]),
+                    32 => @memcpy(std.mem.asBytes(&f.value.float32), input[0..4]),
+                    64 => @memcpy(std.mem.asBytes(&f.value.float64), input[0..8]),
                     else => return RuntimeError.InvalidValueType,
                 }
                 return @divExact(f.bit_count, 8);
             },
-            .Vector4f32 => |*vec| {
-                inline for (0..4) |i| {
-                    const start = i * 4;
-                    const end = (i + 1) * 4;
-                    if (start >= input.len or end > input.len) return RuntimeError.OutOfBounds;
-                    std.mem.copyForwards(u8, std.mem.asBytes(&vec[i]), input[start..end]);
-                }
-                return 4 * 4;
-            },
-            .Vector3f32 => |*vec| {
-                inline for (0..3) |i| {
-                    const start = i * 4;
-                    const end = (i + 1) * 4;
-                    if (start >= input.len or end > input.len) return RuntimeError.OutOfBounds;
-                    std.mem.copyForwards(u8, std.mem.asBytes(&vec[i]), input[start..end]);
-                }
-                return 3 * 4;
-            },
-            .Vector2f32 => |*vec| {
-                inline for (0..2) |i| {
-                    const start = i * 4;
-                    const end = (i + 1) * 4;
-                    if (start >= input.len or end > input.len) return RuntimeError.OutOfBounds;
-                    std.mem.copyForwards(u8, std.mem.asBytes(&vec[i]), input[start..end]);
-                }
-                return 2 * 4;
-            },
-            .Vector4i32 => |*vec| {
-                inline for (0..4) |i| {
-                    const start = i * 4;
-                    const end = (i + 1) * 4;
-                    if (start >= input.len or end > input.len) return RuntimeError.OutOfBounds;
-                    std.mem.copyForwards(u8, std.mem.asBytes(&vec[i]), input[start..end]);
-                }
-                return 4 * 4;
-            },
-            .Vector3i32 => |*vec| {
-                inline for (0..3) |i| {
-                    const start = i * 4;
-                    const end = (i + 1) * 4;
-                    if (start >= input.len or end > input.len) return RuntimeError.OutOfBounds;
-                    std.mem.copyForwards(u8, std.mem.asBytes(&vec[i]), input[start..end]);
-                }
-                return 3 * 4;
-            },
-            .Vector2i32 => |*vec| {
-                inline for (0..2) |i| {
-                    const start = i * 4;
-                    const end = (i + 1) * 4;
-                    if (start >= input.len or end > input.len) return RuntimeError.OutOfBounds;
-                    std.mem.copyForwards(u8, std.mem.asBytes(&vec[i]), input[start..end]);
-                }
-                return 2 * 4;
-            },
-            .Vector4u32 => |*vec| {
-                inline for (0..4) |i| {
-                    const start = i * 4;
-                    const end = (i + 1) * 4;
-                    if (start >= input.len or end > input.len) return RuntimeError.OutOfBounds;
-                    std.mem.copyForwards(u8, std.mem.asBytes(&vec[i]), input[start..end]);
-                }
-                return 4 * 4;
-            },
-            .Vector3u32 => |*vec| {
-                inline for (0..3) |i| {
-                    const start = i * 4;
-                    const end = (i + 1) * 4;
-                    if (start >= input.len or end > input.len) return RuntimeError.OutOfBounds;
-                    std.mem.copyForwards(u8, std.mem.asBytes(&vec[i]), input[start..end]);
-                }
-                return 3 * 4;
-            },
-            .Vector2u32 => |*vec| {
-                inline for (0..2) |i| {
-                    const start = i * 4;
-                    const end = (i + 1) * 4;
-                    if (start >= input.len or end > input.len) return RuntimeError.OutOfBounds;
-                    std.mem.copyForwards(u8, std.mem.asBytes(&vec[i]), input[start..end]);
-                }
-                return 2 * 4;
-            },
+
+            .Vector4f32 => |*vec| return vecRoutine(@TypeOf(vec.*), vec, input),
+            .Vector3f32 => |*vec| return vecRoutine(@TypeOf(vec.*), vec, input),
+            .Vector2f32 => |*vec| return vecRoutine(@TypeOf(vec.*), vec, input),
+
+            .Vector4i32 => |*vec| return vecRoutine(@TypeOf(vec.*), vec, input),
+            .Vector3i32 => |*vec| return vecRoutine(@TypeOf(vec.*), vec, input),
+            .Vector2i32 => |*vec| return vecRoutine(@TypeOf(vec.*), vec, input),
+
+            .Vector4u32 => |*vec| return vecRoutine(@TypeOf(vec.*), vec, input),
+            .Vector3u32 => |*vec| return vecRoutine(@TypeOf(vec.*), vec, input),
+            .Vector2u32 => |*vec| return vecRoutine(@TypeOf(vec.*), vec, input),
+
             .Vector, .Matrix => |*values| {
                 var offset: usize = 0;
                 for (values.*) |*v| {
@@ -520,22 +481,23 @@ pub const Value = union(Type) {
                 }
                 return offset;
             },
-            .Structure => |s| {
-                var offset: usize = 0;
+            .Structure => |*s| {
+                var end_offset: usize = 0;
                 for (s.values, 0..) |*v, i| {
-                    const write_size = try v.write(input[offset..]);
-                    if (i + 1 < s.offsets.len) {
-                        if (s.offsets[i + 1]) |o| {
-                            offset = o;
-                            continue;
+                    const member_offset: usize = @intCast(s.offsets[i] orelse end_offset);
+                    const write_size = try v.write(input[member_offset..]);
+                    end_offset = @max(end_offset, member_offset + write_size);
                 }
-                    }
-                    offset += write_size;
-                }
-                return offset;
+                s.external_data = @constCast(input[0..end_offset]);
+                return end_offset;
             },
-            .RuntimeArray => |*arr| arr.data = input[0..],
+            .RuntimeArray => |*arr| arr.data = @constCast(input[0..]),
             .Image => |*img| img.driver_image = @ptrFromInt(std.mem.bytesToValue(usize, input[0..])),
+            .Sampler => |*sampler| sampler.driver_sampler = @ptrFromInt(std.mem.bytesToValue(usize, input[0..])),
+            .SampledImage => |*img| {
+                img.driver_image = @ptrFromInt(std.mem.bytesToValue(usize, input[0..]));
+                img.driver_sampler = @ptrFromInt(std.mem.bytesToValue(usize, input[@sizeOf(usize)..]));
+            },
             else => return RuntimeError.InvalidValueType,
         }
         return 0;
@@ -560,13 +522,8 @@ pub const Value = union(Type) {
             .Structure => |s| blk: {
                 var size: usize = 0;
                 for (s.values, 0..) |v, i| {
-                    if (i + 1 < s.offsets.len) {
-                        if (s.offsets[i + 1]) |o| {
-                            size = o;
-                            continue;
-                        }
-                    }
-                    size += try v.getPlainMemorySize();
+                    const member_offset: usize = @intCast(s.offsets[i] orelse size);
+                    size = @max(size, member_offset + try v.getPlainMemorySize());
                 }
                 break :blk size;
             },
@@ -617,14 +574,35 @@ pub const Value = union(Type) {
                     switch (p.ptr) {
                         .common => |ptr| {
                             _ = try ptr.read(window);
-                            ptr.deinit(allocator);
-                            if (p.is_owner_of_uniform_slice)
-                                allocator.destroy(ptr);
                         },
-                        else => {},
+                        .f32_ptr => |ptr| {
+                            if (window.len < @sizeOf(f32)) return RuntimeError.OutOfBounds;
+                            @memcpy(window[0..@sizeOf(f32)], std.mem.asBytes(ptr));
+                        },
+                        .i32_ptr => |ptr| {
+                            if (window.len < @sizeOf(i32)) return RuntimeError.OutOfBounds;
+                            @memcpy(window[0..@sizeOf(i32)], std.mem.asBytes(ptr));
+                        },
+                        .u32_ptr => |ptr| {
+                            if (window.len < @sizeOf(u32)) return RuntimeError.OutOfBounds;
+                            @memcpy(window[0..@sizeOf(u32)], std.mem.asBytes(ptr));
+                        },
                     }
+
+                    if (p.uniform_backing_value) |backing| {
+                        backing.deinit(allocator);
+                        allocator.destroy(backing);
+                        p.uniform_backing_value = null;
+                    }
+
                     p.uniform_slice_window = null;
                 }
+
+                if (p.uniform_backing_value) |backing| {
+                    backing.deinit(allocator);
+                    allocator.destroy(backing);
+                    p.uniform_backing_value = null;
+                }
             },
             else => {},
         }
@@ -658,28 +636,64 @@ pub const Value = union(Type) {
             .Vector => |lanes| (try getPrimitiveField(T, bits, &lanes[lane_index])).*,
 
             .Vector2i32 => |*vec| switch (lane_index) {
-                inline 0...1 => |i| if (bits == 32) @as(TT, @bitCast(vec[i])) else return RuntimeError.InvalidSpirV,
+                inline 0...1 => |i| blk: {
+                    if (bits == 32) {
+                        break :blk @as(TT, @bitCast(vec[i]));
+                    } else {
+                        return RuntimeError.InvalidSpirV;
+                    }
+                },
                 else => return RuntimeError.InvalidSpirV,
             },
             .Vector3i32 => |*vec| switch (lane_index) {
-                inline 0...2 => |i| if (bits == 32) @as(TT, @bitCast(vec[i])) else return RuntimeError.InvalidSpirV,
+                inline 0...2 => |i| blk: {
+                    if (bits == 32) {
+                        break :blk @as(TT, @bitCast(vec[i]));
+                    } else {
+                        return RuntimeError.InvalidSpirV;
+                    }
+                },
                 else => return RuntimeError.InvalidSpirV,
             },
             .Vector4i32 => |*vec| switch (lane_index) {
-                inline 0...3 => |i| if (bits == 32) @as(TT, @bitCast(vec[i])) else return RuntimeError.InvalidSpirV,
+                inline 0...3 => |i| blk: {
+                    if (bits == 32) {
+                        break :blk @as(TT, @bitCast(vec[i]));
+                    } else {
+                        return RuntimeError.InvalidSpirV;
+                    }
+                },
                 else => return RuntimeError.InvalidSpirV,
             },
 
             .Vector2u32 => |*vec| switch (lane_index) {
-                inline 0...1 => |i| if (bits == 32) @as(TT, @bitCast(vec[i])) else return RuntimeError.InvalidSpirV,
+                inline 0...1 => |i| blk: {
+                    if (bits == 32) {
+                        break :blk @as(TT, @bitCast(vec[i]));
+                    } else {
+                        return RuntimeError.InvalidSpirV;
+                    }
+                },
                 else => return RuntimeError.InvalidSpirV,
             },
             .Vector3u32 => |*vec| switch (lane_index) {
-                inline 0...2 => |i| if (bits == 32) @as(TT, @bitCast(vec[i])) else return RuntimeError.InvalidSpirV,
+                inline 0...2 => |i| blk: {
+                    if (bits == 32) {
+                        break :blk @as(TT, @bitCast(vec[i]));
+                    } else {
+                        return RuntimeError.InvalidSpirV;
+                    }
+                },
                 else => return RuntimeError.InvalidSpirV,
             },
             .Vector4u32 => |*vec| switch (lane_index) {
-                inline 0...3 => |i| if (bits == 32) @as(TT, @bitCast(vec[i])) else return RuntimeError.InvalidSpirV,
+                inline 0...3 => |i| blk: {
+                    if (bits == 32) {
+                        break :blk @as(TT, @bitCast(vec[i]));
+                    } else {
+                        return RuntimeError.InvalidSpirV;
+                    }
+                },
                 else => return RuntimeError.InvalidSpirV,
             },
 
@@ -694,28 +708,52 @@ pub const Value = union(Type) {
             .Vector => |lanes| try setScalarLaneValue(T, bits, &lanes[lane_index], value),
 
             .Vector2i32 => |*vec| switch (lane_index) {
-                inline 0...1 => |i| vec[i] = if (bits == 32) @bitCast(value) else return RuntimeError.InvalidSpirV,
+                inline 0...1 => |i| if (bits == 32) {
+                    vec[i] = @bitCast(value);
+                } else {
+                    return RuntimeError.InvalidSpirV;
+                },
                 else => return RuntimeError.InvalidSpirV,
             },
             .Vector3i32 => |*vec| switch (lane_index) {
-                inline 0...2 => |i| vec[i] = if (bits == 32) @bitCast(value) else return RuntimeError.InvalidSpirV,
+                inline 0...2 => |i| if (bits == 32) {
+                    vec[i] = @bitCast(value);
+                } else {
+                    return RuntimeError.InvalidSpirV;
+                },
                 else => return RuntimeError.InvalidSpirV,
             },
             .Vector4i32 => |*vec| switch (lane_index) {
-                inline 0...3 => |i| vec[i] = if (bits == 32) @bitCast(value) else return RuntimeError.InvalidSpirV,
+                inline 0...3 => |i| if (bits == 32) {
+                    vec[i] = @bitCast(value);
+                } else {
+                    return RuntimeError.InvalidSpirV;
+                },
                 else => return RuntimeError.InvalidSpirV,
             },
 
             .Vector2u32 => |*vec| switch (lane_index) {
-                inline 0...1 => |i| vec[i] = if (bits == 32) @bitCast(value) else return RuntimeError.InvalidSpirV,
+                inline 0...1 => |i| if (bits == 32) {
+                    vec[i] = @bitCast(value);
+                } else {
+                    return RuntimeError.InvalidSpirV;
+                },
                 else => return RuntimeError.InvalidSpirV,
             },
             .Vector3u32 => |*vec| switch (lane_index) {
-                inline 0...2 => |i| vec[i] = if (bits == 32) @bitCast(value) else return RuntimeError.InvalidSpirV,
+                inline 0...2 => |i| if (bits == 32) {
+                    vec[i] = @bitCast(value);
+                } else {
+                    return RuntimeError.InvalidSpirV;
+                },
                 else => return RuntimeError.InvalidSpirV,
             },
             .Vector4u32 => |*vec| switch (lane_index) {
-                inline 0...3 => |i| vec[i] = if (bits == 32) @bitCast(value) else return RuntimeError.InvalidSpirV,
+                inline 0...3 => |i| if (bits == 32) {
+                    vec[i] = @bitCast(value);
+                } else {
+                    return RuntimeError.InvalidSpirV;
+                },
                 else => return RuntimeError.InvalidSpirV,
             },
 
@@ -752,6 +790,10 @@ pub const Value = union(Type) {
         }
     }
 
+    pub fn getPrimitiveFieldConst(comptime T: PrimitiveType, comptime BitCount: SpvWord, v: *const Value) RuntimeError!*const getPrimitiveFieldType(T, BitCount) {
+        return getPrimitiveField(T, BitCount, @constCast(v));
+    }
+
     pub fn getPrimitiveField(comptime T: PrimitiveType, comptime BitCount: SpvWord, v: *Value) RuntimeError!*getPrimitiveFieldType(T, BitCount) {
         if (std.meta.activeTag(v.*) == .Pointer) {
             return switch (v.Pointer.ptr) {
@@ -790,8 +832,8 @@ pub const Value = union(Type) {
     pub fn resolveLaneBitWidth(self: *const Self) RuntimeError!SpvWord {
         return switch (self.*) {
             .Bool => 8,
-            .Float => |f| f.bit_length,
-            .Int => |i| i.bit_length,
+            .Float => |f| @intCast(f.bit_count),
+            .Int => |i| @intCast(i.bit_count),
             .Vector => |v| v[0].resolveLaneBitWidth(),
             .Vector4f32,
             .Vector3f32,
@@ -831,4 +873,28 @@ pub const Value = union(Type) {
             else => .unsigned,
         };
     }
+
+    pub inline fn getVectorSpecialization(self: *const Self, comptime N: usize, comptime T: type) @Vector(N, T) {
+        return switch (T) {
+            f32 => switch (N) {
+                inline 4 => self.Vector4f32,
+                inline 3 => self.Vector3f32,
+                inline 2 => self.Vector2f32,
+                else => unreachable,
+            },
+            i32 => switch (N) {
+                inline 4 => self.Vector4i32,
+                inline 3 => self.Vector3i32,
+                inline 2 => self.Vector2i32,
+                else => unreachable,
+            },
+            u32 => switch (N) {
+                inline 4 => self.Vector4u32,
+                inline 3 => self.Vector3u32,
+                inline 2 => self.Vector2u32,
+                else => unreachable,
+            },
+            else => unreachable,
+        };
+    }
 };

src/lib.zig

@@ -41,6 +41,8 @@ pub const SpvByte = spv.SpvByte;
 pub const SpvWord = spv.SpvWord;
 pub const SpvBool = spv.SpvBool;
 
+pub const SpvDim = spv.SpvDim;
+
 pub const Vec4f32 = @Vector(4, f32);
 pub const Vec3f32 = @Vector(3, f32);
 pub const Vec2f32 = @Vector(2, f32);

src/spv.zig

@@ -263,9 +263,9 @@ pub const SpvStorageClass = enum(u32) {
 };
 
 pub const SpvDim = enum(u32) {
-    _1D = 0,
-    _2D = 1,
-    _3D = 2,
+    @"1D" = 0,
+    @"2D" = 1,
+    @"3D" = 2,
     Cube = 3,
     Rect = 4,
     Buffer = 5,

test/maths.zig

@@ -1,5 +1,6 @@
 const std = @import("std");
 const root = @import("root.zig");
+const zm = @import("zmath");
 const compileNzsl = root.compileNzsl;
 const case = root.case;
 
@@ -154,3 +155,203 @@ test "Maths vectors" {
         }
     }
 }
+
+// Tests all mathematical operation on vec2/3/4 with scalars with all NZSL supported primitive types
+test "Maths vectors with scalars" {
+    const allocator = std.testing.allocator;
+    const types = [_]type{ f32, f64, i32, u32 };
+    var operations = std.EnumMap(Operations, u8).init(.{
+        .Mul = '*',
+        .Div = '/',
+        .Mod = '%',
+    });
+
+    var it = operations.iterator();
+    while (it.next()) |op| {
+        inline for (2..5) |L| {
+            inline for (types) |T| {
+                const base_color: case.Vec(L, T) = .{ .val = case.random(@Vector(L, T)) };
+                const ratio = case.random(T);
+                const splat_ratio = @as(@Vector(L, T), @splat(ratio));
+                const expected = switch (op.key) {
+                    .Mul => if (@typeInfo(T) == .int) @mulWithOverflow(base_color.val, splat_ratio)[0] else base_color.val * splat_ratio,
+                    .Div => if (@typeInfo(T) == .int) @divTrunc(base_color.val, splat_ratio) else base_color.val / splat_ratio,
+                    .Mod => @mod(base_color.val, splat_ratio),
+                    else => unreachable,
+                };
+
+                const shader = try std.fmt.allocPrint(
+                    allocator,
+                    \\ [nzsl_version("1.1")]
+                    \\ [feature(float64)]
+                    \\ module;
+                    \\ 
+                    \\ struct FragOut
+                    \\ {{
+                    \\     [location(0)] color: vec{d}[{s}]
+                    \\ }}
+                    \\
+                    \\ [entry(frag)]
+                    \\ fn main() -> FragOut
+                    \\ {{
+                    \\     let output: FragOut;
+                    \\     output.color = vec{d}[{s}]({f}) {c} {d};
+                    \\     return output;
+                    \\ }}
+                ,
+                    .{
+                        L,
+                        @typeName(T),
+                        L,
+                        @typeName(T),
+                        base_color,
+                        op.value.*,
+                        ratio,
+                    },
+                );
+                defer allocator.free(shader);
+                const code = try compileNzsl(allocator, shader);
+                defer allocator.free(code);
+                try case.expect(.{
+                    .source = code,
+                    .expected_outputs = &.{
+                        std.mem.asBytes(&@as([L]T, expected)),
+                    },
+                });
+            }
+        }
+    }
+}
+
+// Tests all mathematical operation on mat3/4 with all NZSL supported primitive types
+test "Maths matrices" {
+    const allocator = std.testing.allocator;
+    const types = [_]type{ f32, f64 };
+    var operations = std.EnumMap(Operations, u8).init(.{
+        .Add = '+',
+        .Sub = '-',
+        .Mul = '*',
+    });
+
+    var it = operations.iterator();
+    while (it.next()) |op| {
+        inline for (3..5) |L| {
+            inline for (types) |T| {
+                const base: case.Mat(L, T) = .{ .val = case.random([L][L]T) };
+                const ratio: case.Mat(L, T) = .{ .val = case.random([L][L]T) };
+                var expected: case.Mat(L, T) = undefined;
+                for (expected.val[0..], base.val[0..], ratio.val[0..]) |*ec, bc, rc| {
+                    for (ec[0..], bc[0..], rc[0..]) |*e, b, r| {
+                        e.* = switch (op.key) {
+                            .Add => b + r,
+                            .Sub => b - r,
+                            .Mul => b * r,
+                            else => unreachable,
+                        };
+                    }
+                }
+
+                const shader = try std.fmt.allocPrint(
+                    allocator,
+                    \\ [nzsl_version("1.1")]
+                    \\ [feature(float64)]
+                    \\ module;
+                    \\ 
+                    \\ struct FragOut
+                    \\ {{
+                    \\     [location(0)] value: mat{d}[{s}]
+                    \\ }}
+                    \\
+                    \\ [entry(frag)]
+                    \\ fn main() -> FragOut
+                    \\ {{
+                    \\     let output: FragOut;
+                    \\     output.value = mat{d}[{s}]({f}) {c} mat{d}[{s}]({f});
+                    \\     return output;
+                    \\ }}
+                ,
+                    .{
+                        L,
+                        @typeName(T),
+                        L,
+                        @typeName(T),
+                        base,
+                        op.value.*,
+                        L,
+                        @typeName(T),
+                        ratio,
+                    },
+                );
+                defer allocator.free(shader);
+                const code = try compileNzsl(allocator, shader);
+                defer allocator.free(code);
+                try case.expect(.{
+                    .source = code,
+                    .expected_outputs = &.{
+                        std.mem.asBytes(&expected),
+                    },
+                });
+            }
+        }
+    }
+}
+
+// Tests all mathematical operation on mat3/4 with all NZSL supported vectors
+test "Maths matrices with vectors" {
+    const allocator = std.testing.allocator;
+    const types = [_]type{ f32, f64 };
+
+    inline for (3..5) |L| {
+        inline for (types) |T| {
+            const base: case.Mat(L, T) = .{ .val = case.random([L][L]T) };
+            const ratio: case.Vec(L, T) = .{ .val = case.random(@Vector(L, T)) };
+            var expected: @Vector(L, T) = undefined;
+
+            expected[0] = (base.val[0][0] * ratio.val[0]) + (base.val[0][1] * ratio.val[1]) + (base.val[0][2] * ratio.val[2]) + if (L == 4) (base.val[0][3] * ratio.val[3]) else 0.0;
+            expected[1] = (base.val[1][0] * ratio.val[0]) + (base.val[1][1] * ratio.val[1]) + (base.val[1][2] * ratio.val[2]) + if (L == 4) (base.val[1][3] * ratio.val[3]) else 0.0;
+            expected[2] = (base.val[2][0] * ratio.val[0]) + (base.val[2][1] * ratio.val[1]) + (base.val[2][2] * ratio.val[2]) + if (L == 4) (base.val[2][3] * ratio.val[3]) else 0.0;
+            if (L == 4)
+                expected[3] = (base.val[3][0] * ratio.val[0]) + (base.val[3][1] * ratio.val[1]) + (base.val[3][2] * ratio.val[2]) + (base.val[3][3] * ratio.val[3]);
+
+            const shader = try std.fmt.allocPrint(
+                allocator,
+                \\ [nzsl_version("1.1")]
+                \\ [feature(float64)]
+                \\ module;
+                \\ 
+                \\ struct FragOut
+                \\ {{
+                \\     [location(0)] value: vec{d}[{s}]
+                \\ }}
+                \\
+                \\ [entry(frag)]
+                \\ fn main() -> FragOut
+                \\ {{
+                \\     let output: FragOut;
+                \\     output.value = mat{d}[{s}]({f}) * vec{d}[{s}]({f});
+                \\     return output;
+                \\ }}
+            ,
+                .{
+                    L,
+                    @typeName(T),
+                    L,
+                    @typeName(T),
+                    base,
+                    L,
+                    @typeName(T),
+                    ratio,
+                },
+            );
+            defer allocator.free(shader);
+            const code = try compileNzsl(allocator, shader);
+            defer allocator.free(code);
+            try case.expect(.{
+                .source = code,
+                .expected_outputs = &.{
+                    std.mem.asBytes(&@as([L]T, expected)),
+                },
+            });
+        }
+    }
+}

test/root.zig

@@ -34,10 +34,10 @@ pub const case = struct {
         // To test with all important module options
         const module_options = [_]spv.Module.ModuleOptions{
             .{
-                .use_simd_vectors_specializations = true,
+                .use_simd_vectors_specializations = false,
             },
             .{
-                .use_simd_vectors_specializations = false,
+                .use_simd_vectors_specializations = true,
             },
         };
 
@@ -78,7 +78,7 @@ pub const case = struct {
     }
 
     pub fn random(comptime T: type) T {
-        var prng: std.Random.DefaultPrng = .init(@intCast(std.Io.Timestamp.now(std.testing.io, .real).toMicroseconds()));
+        var prng: std.Random.DefaultPrng = .init(@intCast(std.Io.Timestamp.now(std.testing.io, .real).toNanoseconds()));
         const rand = prng.random();
 
         return switch (@typeInfo(T)) {
@@ -91,6 +91,13 @@ pub const case = struct {
                 }
                 break :blk vec;
             },
+            .array => |a| blk: {
+                var arr: [a.len]a.child = undefined;
+                inline for (0..a.len) |i| {
+                    arr[i] = random(a.child);
+                }
+                break :blk arr;
+            },
             inline else => unreachable,
         };
     }
@@ -107,6 +114,21 @@ pub const case = struct {
             }
         };
     }
+
+    pub fn Mat(comptime len: usize, comptime T: type) type {
+        return struct {
+            const Self = @This();
+            val: [len][len]T,
+            pub fn format(self: *const Self, w: *std.Io.Writer) std.Io.Writer.Error!void {
+                inline for (0..len) |i| {
+                    inline for (0..len) |j| {
+                        try w.print("{d}", .{self.val[i][j]});
+                        if (i < len - 1 or j < len - 1) try w.writeAll(", ");
+                    }
+                }
+            }
+        };
+    }
 };
 
 test {