fixing wsi crashes

build.zig

@@ -432,7 +432,8 @@ fn addMultithreadedCTS(b: *std.Build, target: std.Build.ResolvedTarget, impl: *c
     }
 
     run.addArg("run");
-    run.addArg("--verbose");
+    run.addArg("--timeout");
+    run.addArg("300");
     run.addArg("--deqp");
     run.addArg(cts_exe_path);
     run.addArg("--caselist");

build.zig.zon

@@ -32,8 +32,8 @@
 
         // Soft dependencies
         .SPIRV_Interpreter = .{
-            .url = "git+https://github.com/Kbz-8/SPIRV-Interpreter#18f910f38a6a9c8a00294947ac545bbffdc55477",
+            .url = "git+https://github.com/Kbz-8/SPIRV-Interpreter#9b6f409bffdbb9105d32d73fa3c3025134b0e7f0",
-            .hash = "SPIRV_Interpreter-0.0.1-ajmpn9wmCQAuq_IL_yvnqUJgHnWyGAtYomv0x8HjTaRh",
+            .hash = "SPIRV_Interpreter-0.0.1-ajmpn_U2CQArCKl6BQ5HxvEE9bslrF21gzZVufvTHkJe",
             .lazy = true,
         },
         //.SPIRV_Interpreter = .{

src/software/device/BoundedAllocator.zig

@@ -42,36 +42,75 @@ pub inline fn queryPeakFootprint(self: *Self) usize {
     return self.peak_concurrent_bytes_allocated.load(.monotonic);
 }
 
+fn reserve(self: *Self, len: usize) bool {
+    while (true) {
+        const current = self.current_bytes_allocated.load(.monotonic);
+        if (len > self.bound -| current)
+            return false;
+        const new_current = current + len;
+        if (self.current_bytes_allocated.cmpxchgWeak(current, new_current, .monotonic, .monotonic) == null) {
+            _ = self.total_bytes_allocated.fetchAdd(len, .monotonic);
+            while (true) {
+                const peak = self.peak_concurrent_bytes_allocated.load(.monotonic);
+                if (new_current <= peak)
+                    break;
+                if (self.peak_concurrent_bytes_allocated.cmpxchgWeak(peak, new_current, .monotonic, .monotonic) == null)
+                    break;
+            }
+            return true;
+        }
+    }
+}
+
+fn release(self: *Self, len: usize) void {
+    while (true) {
+        const current = self.current_bytes_allocated.load(.monotonic);
+        const new_current = current -| len;
+        if (self.current_bytes_allocated.cmpxchgWeak(current, new_current, .monotonic, .monotonic) == null)
+            return;
+    }
+}
+
 fn alloc(context: *anyopaque, len: usize, alignment: Alignment, ret_addr: usize) ?[*]u8 {
     const self: *Self = @ptrCast(@alignCast(context));
-    if (self.current_bytes_allocated.fetchAdd(len, .monotonic) >= self.bound)
+    if (!self.reserve(len))
         return null;
-    _ = self.total_bytes_allocated.fetchAdd(len, .monotonic);
+    return self.child_allocator.rawAlloc(len, alignment, ret_addr) orelse {
-    if (self.current_bytes_allocated.load(.monotonic) > self.peak_concurrent_bytes_allocated.load(.monotonic))
+        self.release(len);
-        self.peak_concurrent_bytes_allocated.store(self.current_bytes_allocated.load(.monotonic), .monotonic);
+        return null;
-    return self.child_allocator.rawAlloc(len, alignment, ret_addr);
+    };
 }
 
 fn resize(context: *anyopaque, ptr: []u8, alignment: Alignment, new_len: usize, ret_addr: usize) bool {
     const self: *Self = @ptrCast(@alignCast(context));
-    _ = self.current_bytes_allocated.fetchSub(ptr.len, .monotonic);
+    if (new_len > ptr.len) {
-    if (self.current_bytes_allocated.fetchAdd(new_len, .monotonic) >= self.bound)
+        const delta = new_len - ptr.len;
+        if (!self.reserve(delta))
+            return false;
+        if (self.child_allocator.rawResize(ptr, alignment, new_len, ret_addr))
+            return true;
+        self.release(delta);
         return false;
-    _ = self.total_bytes_allocated.fetchAdd(new_len, .monotonic);
+    }
     return self.child_allocator.rawResize(ptr, alignment, new_len, ret_addr);
 }
 
 fn remap(context: *anyopaque, ptr: []u8, alignment: Alignment, new_len: usize, ret_addr: usize) ?[*]u8 {
     const self: *Self = @ptrCast(@alignCast(context));
-    _ = self.current_bytes_allocated.fetchSub(ptr.len, .monotonic);
+    if (new_len > ptr.len) {
-    if (self.current_bytes_allocated.fetchAdd(new_len, .monotonic) >= self.bound)
+        const delta = new_len - ptr.len;
-        return null;
+        if (!self.reserve(delta))
-    _ = self.total_bytes_allocated.fetchAdd(new_len, .monotonic);
+            return null;
-    return self.child_allocator.rawRemap(ptr, alignment, new_len, ret_addr);
+        return self.child_allocator.rawRemap(ptr, alignment, new_len, ret_addr) orelse {
+            self.release(delta);
+            return null;
+        };
+    }
+    return self.child_allocator.rawRemap(ptr, alignment, new_len, ret_addr) orelse null;
 }
 
 fn free(context: *anyopaque, ptr: []u8, alignment: Alignment, ret_addr: usize) void {
     const self: *Self = @ptrCast(@alignCast(context));
-    _ = self.current_bytes_allocated.fetchSub(ptr.len, .monotonic);
+    self.release(ptr.len);
     return self.child_allocator.rawFree(ptr, alignment, ret_addr);
 }

src/software/device/Device.zig

@@ -118,11 +118,10 @@ fn writeDescriptorSet(rt: *spv.Runtime, payload: DescriptorPayload, set: u32, bi
     for (rt.mod.bindings.items) |entry| {
         if (entry.set != set or entry.binding != binding)
             continue;
-
+        const variable = if (rt.results[entry.result].variant) |*variant| switch (variant.*) {
-        const variable = switch (rt.results[entry.result].variant orelse continue) {
             .Variable => |*variable| variable,
             else => continue,
-        };
+        } else continue;
 
         writeDescriptorValue(&variable.value, payload, descriptor_index) catch |err| switch (err) {
             spv.Runtime.RuntimeError.InvalidValueType,

src/software/device/fragment.zig

@@ -82,6 +82,11 @@ pub fn shaderInvocation(
         SpvRuntimeError.NotFound => {},
         else => return err,
     };
+    const helper_invocation = false;
+    rt.writeBuiltIn(allocator, std.mem.asBytes(&helper_invocation), .HelperInvocation) catch |err| switch (err) {
+        SpvRuntimeError.NotFound => {},
+        else => return err,
+    };
 
     const SoftPipeline = @import("../SoftPipeline.zig");
     const previous_fragment_coord = SoftPipeline.current_fragment_coord;
@@ -157,6 +162,10 @@ pub fn shaderInvocation(
         },
         else => return err,
     };
+    if (rt.helper_invocation) {
+        try rt.flushDescriptorSets(allocator);
+        return SpvRuntimeError.Killed;
+    }
 
     var outputs = std.mem.zeroes([spv.SPIRV_MAX_OUTPUT_LOCATIONS][@sizeOf(zm.F32x4)]u8);
 

src/software/device/rasterizer/common.zig

@@ -216,20 +216,13 @@ pub fn depthTestSampleAndUpdate(
     z: f32,
     state: ?vk.PipelineDepthStencilStateCreateInfo,
 ) VkError!bool {
+    const depth_state = state orelse return true;
     const depth_offset = targetSampleOffset(depth.*, x, y, sample_index) orelse return false;
 
     depth.mutex.lock(io) catch return VkError.DeviceLost;
     defer depth.mutex.unlock(io);
 
-    if (state) |depth_state| {
+    return depthTestAndUpdateAtOffset(depth, depth_offset, z, depth_state);
-        return depthTestAndUpdateAtOffset(depth, depth_offset, z, depth_state);
-    }
-
-    const depth_value = blitter.readFloat4(depth.base[depth_offset..], depth.format);
-    if (z >= depth_value[0])
-        return false;
-    blitter.writeFloat4(zm.f32x4s(z), depth.base[depth_offset..], depth.format);
-    return true;
 }
 
 fn depthTestAndUpdateAtOffset(depth: *RenderTargetAccess, offset: usize, z: f32, state: vk.PipelineDepthStencilStateCreateInfo) bool {
@@ -589,24 +582,16 @@ pub fn writeToTargets(
 
         // After work depth test to avoid overwritten depth pixels during fragment invocations.
         var depth_passed: ?bool = null;
-        if (!depth_already_applied and depth_attachment_access != null) {
+        if (!depth_already_applied and depth_attachment_access != null and depth_stencil_state != null) {
             const depth = depth_attachment_access.?;
             const depth_offset = targetSampleOffset(depth.*, x, y, sample_index) orelse continue;
 
             depth.mutex.lock(io) catch return VkError.DeviceLost;
             defer depth.mutex.unlock(io);
 
-            if (depth_stencil_state) |state| {
+            depth_passed = depthTestAndUpdateAtOffset(depth, depth_offset, z, depth_stencil_state.?);
-                depth_passed = depthTestAndUpdateAtOffset(depth, depth_offset, z, state);
+            if (!depth_passed.? and stencil_state == null)
-                if (!depth_passed.? and stencil_state == null)
+                continue;
-                    continue;
-            } else {
-                const depth_value = blitter.readFloat4(depth.base[depth_offset..], depth.format);
-                if (z >= depth_value[0])
-                    continue;
-                blitter.writeFloat4(zm.f32x4s(z), depth.base[depth_offset..], depth.format);
-                depth_passed = true;
-            }
         }
 
         if (stencil_attachment_access) |stencil| {

src/vulkan/PhysicalDevice.zig

@@ -164,21 +164,29 @@ pub fn getSurfaceCapabilitiesKHR(_: *Self, surface: *SurfaceKHR, capabilities: *
 
 pub fn getSurfaceFormatsKHR(_: *Self, _: *SurfaceKHR, count: *u32, p_formats: ?[*]vk.SurfaceFormatKHR) VkError!void {
     const surface_formats = SurfaceKHR.getFormats();
-    count.* = surface_formats.len;
     if (p_formats) |formats| {
-        for (formats[0..], surface_formats[0..]) |*format, surface_format| {
+        const write_count = @min(count.*, surface_formats.len);
+        for (formats[0..write_count], surface_formats[0..write_count]) |*format, surface_format| {
             format.* = surface_format;
         }
+        count.* = @intCast(write_count);
+        if (write_count < surface_formats.len) return VkError.Incomplete;
+    } else {
+        count.* = @intCast(surface_formats.len);
     }
 }
 
 pub fn getSurfacePresentModesKHR(_: *Self, _: *SurfaceKHR, count: *u32, p_modes: ?[*]vk.PresentModeKHR) VkError!void {
     const surface_modes = SurfaceKHR.getPresentModes();
-    count.* = surface_modes.len;
     if (p_modes) |modes| {
-        for (modes[0..], surface_modes[0..]) |*mode, surface_mode| {
+        const write_count = @min(count.*, surface_modes.len);
+        for (modes[0..write_count], surface_modes[0..write_count]) |*mode, surface_mode| {
             mode.* = surface_mode;
         }
+        count.* = @intCast(write_count);
+        if (write_count < surface_modes.len) return VkError.Incomplete;
+    } else {
+        count.* = @intCast(surface_modes.len);
     }
 }
 

src/vulkan/Queue.zig

@@ -149,7 +149,11 @@ pub fn presentKHR(_: *Self, info: *const vk.PresentInfoKHR) VkError!void {
                 }
             else
                 cmd_err = err;
+            continue;
         };
+        if (info.p_results) |results| {
+            results[i] = .success;
+        }
     }
 
     if (cmd_err) |err|

src/vulkan/lib_vulkan.zig

@@ -2458,11 +2458,15 @@ pub export fn apeGetSwapchainImagesKHR(p_device: vk.Device, p_swapchain: vk.Swap
     Dispatchable(Device).checkHandleValidity(p_device) catch |err| return toVkResult(err);
 
     const swapchain = NonDispatchable(SwapchainKHR).fromHandleObject(p_swapchain) catch |err| return toVkResult(err);
-    count.* = @intCast(swapchain.images.len);
     if (p_images) |images| {
-        for (images[0..], swapchain.images[0..]) |*image, *swapchain_image| {
+        const write_count = @min(count.*, swapchain.images.len);
+        for (images[0..write_count], swapchain.images[0..write_count]) |*image, *swapchain_image| {
             image.* = swapchain_image.non_dispatchable_image.toVkHandle(vk.Image);
         }
+        count.* = @intCast(write_count);
+        if (write_count < swapchain.images.len) return .incomplete;
+    } else {
+        count.* = @intCast(swapchain.images.len);
     }
 
     return .success;

src/vulkan/wsi/SurfaceKHR.zig

@@ -17,6 +17,7 @@ const formats = [_]vk.SurfaceFormatKHR{
 };
 
 const present_modes = [_]vk.PresentModeKHR{
+    .fifo_khr,
     .immediate_khr,
 };
 

src/vulkan/wsi/SwapchainKHR.zig

@@ -18,7 +18,19 @@ owner: *Device,
 surface: ?*SurfaceKHR,
 images: []PresentImage,
 
+fn validateImageAllocationSize(info: *const vk.SwapchainCreateInfoKHR) VkError!void {
+    if (info.image_extent.width == 0 or info.image_extent.height == 0 or info.image_array_layers == 0)
+        return VkError.OutOfDeviceMemory;
+
+    const texel_size = lib.format.texelSize(info.image_format);
+    var size = std.math.mul(usize, texel_size, info.image_extent.width) catch return VkError.OutOfDeviceMemory;
+    size = std.math.mul(usize, size, info.image_extent.height) catch return VkError.OutOfDeviceMemory;
+    _ = std.math.mul(usize, size, info.image_array_layers) catch return VkError.OutOfDeviceMemory;
+}
+
 pub fn create(device: *Device, allocator: std.mem.Allocator, surface: *SurfaceKHR, info: *const vk.SwapchainCreateInfoKHR) VkError!*Self {
+    try validateImageAllocationSize(info);
+
     const self = allocator.create(Self) catch return VkError.OutOfHostMemory;
     errdefer allocator.destroy(self);
 
@@ -27,6 +39,14 @@ pub fn create(device: *Device, allocator: std.mem.Allocator, surface: *SurfaceKH
         allocator.free(images);
     }
 
+    var initialized_image_count: usize = 0;
+    errdefer {
+        for (images[0..initialized_image_count]) |*image| {
+            surface.detachImage(allocator, image) catch {};
+            image.deinit(allocator);
+        }
+    }
+
     for (images) |*image| {
         image.* = try .init(device, allocator, &.{
             .format = info.image_format,
@@ -46,6 +66,7 @@ pub fn create(device: *Device, allocator: std.mem.Allocator, surface: *SurfaceKH
             .queue_family_index_count = info.queue_family_index_count,
             .initial_layout = .general,
         });
+        initialized_image_count += 1;
 
         try surface.attachImage(allocator, image);
     }