improving spirv kill support

build.zig.zon

@@ -6,8 +6,8 @@
 
     .dependencies = .{
         .vulkan_headers = .{
-            .url = "git+https://github.com/KhronosGroup/Vulkan-Headers?ref=v1.4.330#ee3b5caaa7e372715873c7b9c390ee1c3ca5db25",
+            .url = "git+https://github.com/KhronosGroup/Vulkan-Headers/#01393c3df0e5285b54ee6527466513f9e614be94",
-            .hash = "N-V-__8AAFXYAQKsK51AAGXB9HziPDFjS_DVUq6_QHVxHrBM",
+            .hash = "N-V-__8AAEu6eQLGuoPmVY5dM6IQxznJGorQMHhLn5xFnn_8",
         },
         .vulkan_zig = .{
             .url = "git+https://github.com/Snektron/vulkan-zig#be16fc99f7f794cba90334978c6c5aadeda8094f",
@@ -26,8 +26,8 @@
             .hash = "N-V-__8AAF9uOh0I4P_99za7N822J3JwsDaqONrFVrcEQo59",
         },
         .SPIRV_Interpreter = .{
-            .url = "git+https://git.kbz8.me/kbz_8/SPIRV-Interpreter#089a33981c449220fd74287560347e72e5ce53fa",
+            .url = "git+https://git.kbz8.me/kbz_8/SPIRV-Interpreter#c7320325fdf6b1e5e6371ff620ed08b926587eed",
-            .hash = "SPIRV_Interpreter-0.0.1-ajmpn8hBBwCp53qmU9TiIKGybBRsx-edOSeE44gnNgQR",
+            .hash = "SPIRV_Interpreter-0.0.1-ajmpn5F_BwBQHPDX6S9rSgLls4FYjpdxGD30B17paW7L",
             .lazy = true,
         },
         //.SPIRV_Interpreter = .{

src/soft/SoftPipeline.zig

@@ -104,6 +104,7 @@ pub fn createCompute(device: *base.Device, allocator: std.mem.Allocator, cache:
                     .writeImageInt4 = writeImageInt4,
                     .sampleImageFloat4 = sampleImageFloat4,
                     .sampleImageInt4 = sampleImageInt4,
+                    .sampleImageDref = sampleImageDref,
                     .queryImageSize = queryImageSize,
                 },
             ) catch |err| {
@@ -192,6 +193,7 @@ pub fn createGraphics(device: *base.Device, allocator: std.mem.Allocator, cache:
                         .writeImageInt4 = writeImageInt4,
                         .sampleImageFloat4 = sampleImageFloat4,
                         .sampleImageInt4 = sampleImageInt4,
+                        .sampleImageDref = sampleImageDref,
                         .queryImageSize = queryImageSize,
                     },
                 ) catch |err| {
@@ -382,7 +384,7 @@ fn writeImageInt4(context: *anyopaque, dim: spv.SpvDim, x: i32, y: i32, z: i32,
     }
 }
 
-fn sampleImageFloat4(context: *anyopaque, context2: *anyopaque, dim: spv.SpvDim, x: f32, y: f32, z: f32, lod: ?f32) SpvRuntimeError!spv.Runtime.Vec4(f32) {
+fn sampleImageFloat4(context: *anyopaque, context2: *anyopaque, dim: spv.SpvDim, x: f32, y: f32, z: f32, lod: ?f32, offset: spv.Runtime.ImageOffset) SpvRuntimeError!spv.Runtime.Vec4(f32) {
     var pixel = zm.f32x4s(0.0);
 
     if (dim == .Buffer) {
@@ -395,7 +397,7 @@ fn sampleImageFloat4(context: *anyopaque, context2: *anyopaque, dim: spv.SpvDim,
         const image: *SoftImage = @alignCast(@fieldParentPtr("interface", image_view.interface.image));
 
         const sampler: *SoftSampler = @ptrCast(@alignCast(context2));
-        pixel = SoftSampler.sampleImageFloat4(image, image_view, sampler, dim, x, y, z, lod) catch return SpvRuntimeError.Unknown;
+        pixel = SoftSampler.sampleImageFloat4(image, image_view, sampler, dim, x, y, z, lod, offset) catch return SpvRuntimeError.Unknown;
     }
 
     return .{
@@ -406,7 +408,7 @@ fn sampleImageFloat4(context: *anyopaque, context2: *anyopaque, dim: spv.SpvDim,
     };
 }
 
-fn sampleImageInt4(context: *anyopaque, context2: *anyopaque, dim: spv.SpvDim, x: f32, y: f32, z: f32, lod: ?f32) SpvRuntimeError!spv.Runtime.Vec4(u32) {
+fn sampleImageInt4(context: *anyopaque, context2: *anyopaque, dim: spv.SpvDim, x: f32, y: f32, z: f32, lod: ?f32, offset: spv.Runtime.ImageOffset) SpvRuntimeError!spv.Runtime.Vec4(u32) {
     var pixel = @Vector(4, u32){ 0, 0, 0, 0 };
 
     if (dim == .Buffer) {
@@ -419,7 +421,7 @@ fn sampleImageInt4(context: *anyopaque, context2: *anyopaque, dim: spv.SpvDim, x
         const image: *SoftImage = @alignCast(@fieldParentPtr("interface", image_view.interface.image));
 
         const sampler: *SoftSampler = @ptrCast(@alignCast(context2));
-        pixel = SoftSampler.sampleImageInt4(image, image_view, sampler, dim, x, y, z, lod) catch return SpvRuntimeError.Unknown;
+        pixel = SoftSampler.sampleImageInt4(image, image_view, sampler, dim, x, y, z, lod, offset) catch return SpvRuntimeError.Unknown;
     }
 
     return .{
@@ -430,6 +432,16 @@ fn sampleImageInt4(context: *anyopaque, context2: *anyopaque, dim: spv.SpvDim, x
     };
 }
 
+fn sampleImageDref(context: *anyopaque, context2: *anyopaque, dim: spv.SpvDim, x: f32, y: f32, z: f32, dref: f32, lod: ?f32, offset: spv.Runtime.ImageOffset) SpvRuntimeError!f32 {
+    if (dim == .Buffer)
+        return SpvRuntimeError.UnsupportedSpirV;
+
+    const image_view: *SoftImageView = @ptrCast(@alignCast(context));
+    const image: *SoftImage = @alignCast(@fieldParentPtr("interface", image_view.interface.image));
+    const sampler: *SoftSampler = @ptrCast(@alignCast(context2));
+    return SoftSampler.sampleImageDref(image, image_view, sampler, dim, x, y, z, dref, lod, offset) catch return SpvRuntimeError.Unknown;
+}
+
 fn queryImageSize(context: *anyopaque, dim: spv.SpvDim, arrayed: bool) SpvRuntimeError!spv.Runtime.Vec4(u32) {
     if (dim == .Buffer) {
         const buffer_view: *SoftBufferView = @ptrCast(@alignCast(context));

src/soft/SoftSampler.zig

@@ -15,6 +15,7 @@ const SoftImageView = @import("SoftImageView.zig");
 
 const Self = @This();
 pub const Interface = base.Sampler;
+pub const ImageOffset = spv.Runtime.ImageOffset;
 
 const CubeCoordinate = struct {
     face: u32,
@@ -276,6 +277,20 @@ fn swizzleInt4(color: U32x4, components: vk.ComponentMapping) U32x4 {
     };
 }
 
+fn compareDepth(op: vk.CompareOp, reference: f32, value: f32) bool {
+    return switch (op) {
+        .never => false,
+        .less => reference < value,
+        .equal => reference == value,
+        .less_or_equal => reference <= value,
+        .greater => reference > value,
+        .not_equal => reference != value,
+        .greater_or_equal => reference >= value,
+        .always => true,
+        else => false,
+    };
+}
+
 fn readSampledFloat4(
     image: *SoftImage,
     image_view: *SoftImageView,
@@ -410,7 +425,7 @@ fn readSampledInt4(
     );
 }
 
-pub fn sampleImageFloat4(image: *SoftImage, image_view: *SoftImageView, sampler: *Self, dim: spv.SpvDim, x: f32, y: f32, z: f32, lod: ?f32) VkError!F32x4 {
+pub fn sampleImageFloat4(image: *SoftImage, image_view: *SoftImageView, sampler: *Self, dim: spv.SpvDim, x: f32, y: f32, z: f32, lod: ?f32, offset: ImageOffset) VkError!F32x4 {
     const mip_level = sampleMipLevel(image, image_view, sampler, lod);
     const extent = image.getMipLevelExtent(mip_level);
     const coord: CubeCoordinate = switch (image_view.interface.view_type) {
@@ -454,9 +469,9 @@ pub fn sampleImageFloat4(image: *SoftImage, image_view: *SoftImageView, sampler:
         *const ImageSamplingContext,
         &context,
         zm.f32x4(
-            coord.u * scale_u,
+            coord.u * scale_u + @as(f32, @floatFromInt(offset.x)),
-            coord.v * scale_v,
+            coord.v * scale_v + @as(f32, @floatFromInt(offset.y)),
-            coord.w * scale_w,
+            coord.w * scale_w + @as(f32, @floatFromInt(offset.z)),
             0.0,
         ),
         switch (sampler.interface.mag_filter) {
@@ -468,7 +483,7 @@ pub fn sampleImageFloat4(image: *SoftImage, image_view: *SoftImageView, sampler:
     );
 }
 
-pub fn sampleImageInt4(image: *SoftImage, image_view: *SoftImageView, sampler: *Self, dim: spv.SpvDim, x: f32, y: f32, z: f32, lod: ?f32) VkError!U32x4 {
+pub fn sampleImageInt4(image: *SoftImage, image_view: *SoftImageView, sampler: *Self, dim: spv.SpvDim, x: f32, y: f32, z: f32, lod: ?f32, offset: ImageOffset) VkError!U32x4 {
     const mip_level = sampleMipLevel(image, image_view, sampler, lod);
     const extent = image.getMipLevelExtent(mip_level);
     const coord: CubeCoordinate = switch (image_view.interface.view_type) {
@@ -507,13 +522,20 @@ pub fn sampleImageInt4(image: *SoftImage, image_view: *SoftImageView, sampler: *
         dim,
         coord,
         mip_level,
-        @intFromFloat(@floor(coord.u * scale_u)),
+        @as(i32, @intFromFloat(@floor(coord.u * scale_u))) + offset.x,
-        @intFromFloat(@floor(coord.v * scale_v)),
+        @as(i32, @intFromFloat(@floor(coord.v * scale_v))) + offset.y,
-        @intFromFloat(@floor(coord.w * scale_w)),
+        @as(i32, @intFromFloat(@floor(coord.w * scale_w))) + offset.z,
     );
     return swizzleInt4(color, image_view.interface.components);
 }
 
+pub fn sampleImageDref(image: *SoftImage, image_view: *SoftImageView, sampler: *Self, dim: spv.SpvDim, x: f32, y: f32, z: f32, dref: f32, lod: ?f32, offset: ImageOffset) VkError!f32 {
+    const color = try sampleImageFloat4(image, image_view, sampler, dim, x, y, z, lod, offset);
+    if (sampler.interface.compare_enable == .false)
+        return color[0];
+    return if (compareDepth(sampler.interface.compare_op, dref, color[0])) 1.0 else 0.0;
+}
+
 pub fn sampleFloat4(
     comptime Context: type,
     context: Context,

src/soft/device/fragment.zig

@@ -99,8 +99,7 @@ pub fn shaderInvocation(
         SpvRuntimeError.OutOfBounds => {},
         SpvRuntimeError.Killed => {
             try rt.flushDescriptorSets(allocator);
-            freeOwnedInputs(allocator, fragment_inputs);
+            return SpvRuntimeError.Killed;
-            return undefined; // FIXME
         },
         else => return err,
     };
@@ -154,13 +153,32 @@ fn readFragmentOutput(
     component: usize,
     result_word: spv.SpvWord,
 ) SpvRuntimeError!void {
-    const memory_size = try rt.getResultMemorySize(result_word);
+    const value = try rt.results[result_word].getConstValue();
-    const offset = component * @sizeOf(f32);
+    switch (value.*) {
-    const location_size = @sizeOf(zm.F32x4);
+        .Array => |array| {
-    const direct_capacity = location_size - offset;
+            for (array.values, 0..) |element, element_index| {
+                const target_location = location + element_index;
+                if (target_location >= outputs.len)
+                    return SpvRuntimeError.OutOfBounds;
 
-    if (memory_size <= direct_capacity) {
+                const memory_size = try element.getPlainMemorySize();
-        try rt.readOutput(outputs[location][offset .. offset + memory_size], result_word);
+                const output = allocator.alloc(u8, memory_size + INTERFACE_BLOB_PADDING) catch return SpvRuntimeError.OutOfMemory;
+                defer allocator.free(output);
+                @memset(output, 0);
+
+                _ = try element.read(output);
+                try copyFragmentOutputBytes(outputs, output[0..memory_size], target_location, component);
+            }
+            return;
+        },
+        else => {},
+    }
+
+    const memory_size = try rt.getResultMemorySize(result_word);
+    if (memory_size <= INTERFACE_BLOB_PADDING) {
+        var output = std.mem.zeroes([INTERFACE_BLOB_PADDING]u8);
+        try rt.readOutput(output[0..memory_size], result_word);
+        try copyFragmentOutputBytes(outputs, output[0..memory_size], location, component);
         return;
     }
 
@@ -168,7 +186,25 @@ fn readFragmentOutput(
     defer allocator.free(output);
     @memset(output, 0);
 
-    try rt.readOutput(output, result_word);
+    try rt.readOutput(output[0..memory_size], result_word);
+    try copyFragmentOutputBytes(outputs, output[0..memory_size], location, component);
+}
+
+fn copyFragmentOutputBytes(
+    outputs: *[spv.SPIRV_MAX_OUTPUT_LOCATIONS][@sizeOf(zm.F32x4)]u8,
+    output: []const u8,
+    location: usize,
+    component: usize,
+) SpvRuntimeError!void {
+    const memory_size = output.len;
+    const offset = component * @sizeOf(f32);
+    const location_size = @sizeOf(zm.F32x4);
+    const direct_capacity = location_size - offset;
+
+    if (memory_size <= direct_capacity) {
+        @memcpy(outputs[location][offset .. offset + memory_size], output);
+        return;
+    }
 
     var source_offset: usize = 0;
     var target_location = location;

src/soft/device/rasterizer.zig

@@ -17,6 +17,7 @@ const DrawCall = Renderer.DrawCall;
 const SoftImage = @import("../SoftImage.zig");
 
 const VkError = base.VkError;
+const SpvRuntimeError = spv.Runtime.RuntimeError;
 
 pub fn processThenFragmentStage(renderer: *Renderer, allocator: std.mem.Allocator, draw_call: *DrawCall) VkError!void {
     const io = draw_call.renderer.device.interface.io();
@@ -315,6 +316,9 @@ fn clipTransformAndRasterizePoint(
                     try common.interpolateVertexOutputs(allocator, &transformed, &transformed, &transformed, 1.0, 0.0, 0.0),
                     null,
                 ) catch |err| {
+                    if (err == SpvRuntimeError.Killed)
+                        continue;
+
                     std.log.scoped(.@"Fragment stage").err("catched a '{s}'", .{@errorName(err)});
                     if (comptime base.config.logs == .verbose) {
                         if (@errorReturnTrace()) |trace| {

src/soft/device/rasterizer/bresenham.zig

@@ -163,6 +163,9 @@ inline fn run(data: RunData) !void {
                 try common.interpolateLineOutputs(data.allocator, data.start_vertex, data.end_vertex, t),
                 null,
             ) catch |err| {
+                if (err == SpvRuntimeError.Killed)
+                    continue;
+
                 std.log.scoped(.@"Fragment stage").err("catched a '{s}'", .{@errorName(err)});
                 if (comptime base.config.logs == .verbose) {
                     if (@errorReturnTrace()) |trace| {

src/soft/device/rasterizer/edge_function.zig

@@ -223,6 +223,9 @@ inline fn run(data: RunData) !void {
                     inputs,
                     derivative_inputs,
                 ) catch |err| {
+                    if (err == SpvRuntimeError.Killed)
+                        continue;
+
                     std.log.scoped(.@"Fragment stage").err("catched a '{s}'", .{@errorName(err)});
                     if (comptime base.config.logs == .verbose) {
                         if (@errorReturnTrace()) |trace| {

src/soft/device/vertex_dispatcher.zig

@@ -107,9 +107,16 @@ inline fn run(data: RunData) !void {
                     SpvRuntimeError.NotFound => continue,
                     else => return err,
                 };
+
                 const memory_size = try rt.getResultMemorySize(result_word);
+
+                const result_is_integer = blk: {
+                    const result_type = rt.getResultPrimitiveType(result_word) catch break :blk false;
+                    break :blk result_type == .SInt or result_type == .UInt;
+                };
+
                 output.outputs[location][component] = .{
-                    .interpolation_type = if (rt.hasResultDecoration(result_word, .Flat) or rt.resultIsInteger(result_word)) .flat else .smooth, // TODO : handle noperspective
+                    .interpolation_type = if (rt.hasResultDecoration(result_word, .Flat) or result_is_integer) .flat else .smooth, // TODO : handle noperspective
                     .blob = data.allocator.alloc(u8, memory_size + INTERFACE_BLOB_PADDING) catch return VkError.OutOfDeviceMemory,
                     .size = memory_size,
                 };
@@ -181,7 +188,10 @@ fn writeVertexInput(rt: *spv.Runtime, allocator: std.mem.Allocator, raw_input: [
         return;
     }
 
-    const input_memory_size = try rt.getInputLocationMemorySize(location);
+    const input_memory_size = rt.getInputLocationMemorySize(location) catch |err| switch (err) {
+        SpvRuntimeError.NotFound => return,
+        else => return err,
+    };
 
     if (expanded_slice.len >= input_memory_size) {
         try rt.writeInputLocation(expanded_slice[0..input_memory_size], location);

src/vulkan/Sampler.zig

@@ -18,6 +18,8 @@ address_mode_u: vk.SamplerAddressMode,
 address_mode_v: vk.SamplerAddressMode,
 address_mode_w: vk.SamplerAddressMode,
 mip_lod_bias: f32,
+compare_enable: vk.Bool32,
+compare_op: vk.CompareOp,
 min_lod: f32,
 max_lod: f32,
 border_color: vk.BorderColor,
@@ -40,6 +42,8 @@ pub fn init(device: *Device, allocator: std.mem.Allocator, info: *const vk.Sampl
         .address_mode_v = info.address_mode_v,
         .address_mode_w = info.address_mode_w,
         .mip_lod_bias = info.mip_lod_bias,
+        .compare_enable = info.compare_enable,
+        .compare_op = info.compare_op,
         .min_lod = info.min_lod,
         .max_lod = info.max_lod,
         .border_color = info.border_color,

src/vulkan/lib.zig

@@ -52,9 +52,7 @@ pub const ShaderModule = @import("ShaderModule.zig");
 pub const SurfaceKHR = @import("wsi/SurfaceKHR.zig");
 pub const SwapchainKHR = @import("wsi/SwapchainKHR.zig");
 
-// To be commented out when spec 1.4.354 is released
+pub const VULKAN_VENDOR_ID: i32 = @intFromEnum(vk.VendorId.ape);
-// pub const VULKAN_VENDOR_ID: i32 = @intFromEnum(vk.VendorId.ape);
-pub const VULKAN_VENDOR_ID: i32 = 0x10008;
 
 /// Default driver name
 pub const DRIVER_NAME = "Unnamed Ape Driver";