implementing proper graphics pipeline creations

build.zig

@@ -155,12 +155,14 @@ fn customSoft(b: *std.Build, lib: *std.Build.Step.Compile, options: *std.Build.S
     }) orelse return error.UnresolvedDependency;
     lib.root_module.addImport("spv", spv.module("spv"));
 
+    const single_threaded_option = b.option(bool, "single-threaded", "Single threaded runtime mode") orelse false;
     const debug_allocator_option = b.option(bool, "debug-allocator", "Debug device allocator") orelse false;
     const shaders_simd_option = b.option(bool, "shader-simd", "Shaders SIMD acceleration") orelse true;
     const single_threaded_compute_option = b.option(bool, "single-threaded-compute", "Single threaded compute shaders execution") orelse true;
     const compute_dump_early_results_table_option = b.option(u32, "compute-dump-early-results-table", "Dump compute shaders results table before invocation");
     const compute_dump_final_results_table_option = b.option(u32, "compute-dump-final-results-table", "Dump compute shaders results table after invocation");
 
+    options.addOption(bool, "single_threaded", single_threaded_option);
     options.addOption(bool, "debug_allocator", debug_allocator_option);
     options.addOption(bool, "shaders_simd", shaders_simd_option);
     options.addOption(bool, "single_threaded_compute", single_threaded_compute_option);

src/soft/SoftCommandBuffer.zig

@@ -218,7 +218,7 @@ pub fn bindVertexBuffer(interface: *Interface, index: usize, buffer: *base.Buffe
 
         pub fn execute(context: *anyopaque, device: *ExecutionDevice) VkError!void {
             const impl: *Impl = @ptrCast(@alignCast(context));
-            device.renderer.dynamic_state.vertex_buffers[impl.index] = .{
+            device.pipeline_states[ExecutionDevice.GRAPHICS_PIPELINE_STATE].data.graphics.vertex_buffers[impl.index] = .{
                 .buffer = impl.buffer,
                 .offset = impl.offset,
                 .size = 0,
@@ -482,8 +482,7 @@ pub fn draw(interface: *Interface, vertex_count: usize, instance_count: usize, f
 
         pub fn execute(context: *anyopaque, device: *ExecutionDevice) VkError!void {
             const impl: *Impl = @ptrCast(@alignCast(context));
-            _ = impl;
+            device.renderer.drawPrimitive(impl.vertex_count, impl.instance_count, impl.first_vertex, impl.first_instance);
-            _ = device;
         }
     };
 

src/soft/SoftInstance.zig

@@ -33,7 +33,7 @@ pub fn create(allocator: std.mem.Allocator, infos: *const vk.InstanceCreateInfo)
     errdefer allocator.destroy(self);
 
     self.allocator = std.heap.smp_allocator;
-    self.threaded = std.Io.Threaded.init(self.allocator, .{});
+    self.threaded = if (comptime base.config.single_threaded) .init_single_threaded else std.Io.Threaded.init(self.allocator, .{});
     self.io_impl = self.threaded.io();
 
     self.interface = try base.Instance.init(allocator, infos);

src/soft/SoftPipeline.zig

@@ -32,6 +32,7 @@ const Stages = enum {
 };
 
 interface: Interface,
+runtimes_allocator: std.heap.ArenaAllocator,
 stages: std.EnumMap(Stages, Shader),
 
 pub fn createCompute(device: *base.Device, allocator: std.mem.Allocator, cache: ?*base.PipelineCache, info: *const vk.ComputePipelineCreateInfo) VkError!*Self {
@@ -50,6 +51,10 @@ pub fn createCompute(device: *base.Device, allocator: std.mem.Allocator, cache:
 
     const device_allocator = soft_device.device_allocator.allocator();
 
+    var runtimes_allocator_arena: std.heap.ArenaAllocator = .init(device_allocator);
+    errdefer runtimes_allocator_arena.deinit();
+    const runtimes_allocator = runtimes_allocator_arena.allocator();
+
     const instance: *SoftInstance = @alignCast(@fieldParentPtr("interface", device.instance));
     const runtimes_count = switch (instance.threaded.async_limit) {
         .nothing => 1,
@@ -62,22 +67,17 @@ pub fn createCompute(device: *base.Device, allocator: std.mem.Allocator, cache:
 
     self.* = .{
         .interface = interface,
+        .runtimes_allocator = runtimes_allocator_arena,
         .stages = std.EnumMap(Stages, Shader).init(.{
             .compute = blk: {
                 var shader: Shader = undefined;
                 soft_module.ref();
                 shader.module = soft_module;
 
-                const runtimes = device_allocator.alloc(spv.Runtime, runtimes_count) catch return VkError.OutOfHostMemory;
+                const runtimes = runtimes_allocator.alloc(spv.Runtime, runtimes_count) catch return VkError.OutOfHostMemory;
-                errdefer {
-                    for (runtimes) |*runtime| {
-                        runtime.deinit(device_allocator);
-                    }
-                    device_allocator.free(runtimes);
-                }
 
                 for (runtimes) |*runtime| {
-                    runtime.* = spv.Runtime.init(device_allocator, &soft_module.module) catch |err| {
+                    runtime.* = spv.Runtime.init(runtimes_allocator, &soft_module.module) catch |err| {
                         std.log.scoped(.SpvRuntimeInit).err("SPIR-V Runtime failed to initialize, {s}", .{@errorName(err)});
                         return VkError.Unknown;
                     };
@@ -85,7 +85,7 @@ pub fn createCompute(device: *base.Device, allocator: std.mem.Allocator, cache:
                         if (specialization.p_map_entries) |map| {
                             const data: []const u8 = @as([*]const u8, @ptrCast(@alignCast(specialization.p_data)))[0..specialization.data_size];
                             for (map[0..], 0..specialization.map_entry_count) |entry, _| {
-                                runtime.addSpecializationInfo(device_allocator, .{
+                                runtime.addSpecializationInfo(runtimes_allocator, .{
                                     .id = @intCast(entry.constant_id),
                                     .offset = @intCast(entry.offset),
                                     .size = @intCast(entry.size),
@@ -96,7 +96,9 @@ pub fn createCompute(device: *base.Device, allocator: std.mem.Allocator, cache:
                 }
 
                 shader.runtimes = runtimes;
-                shader.entry = device_allocator.dupe(u8, std.mem.span(info.stage.p_name)) catch return VkError.OutOfHostMemory;
+                shader.entry = runtimes_allocator.dupe(u8, std.mem.span(info.stage.p_name)) catch return VkError.OutOfHostMemory;
+
+                std.log.scoped(.ComputePipeline).debug("Created {d} runtimes for compute stage", .{runtimes_count});
                 break :blk shader;
             },
         }),
@@ -114,6 +116,13 @@ pub fn createGraphics(device: *base.Device, allocator: std.mem.Allocator, cache:
         .destroy = destroy,
     };
 
+    const soft_device: *SoftDevice = @alignCast(@fieldParentPtr("interface", device));
+    const device_allocator = soft_device.device_allocator.allocator();
+
+    var runtimes_allocator_arena: std.heap.ArenaAllocator = .init(device_allocator);
+    errdefer runtimes_allocator_arena.deinit();
+    const runtimes_allocator = runtimes_allocator_arena.allocator();
+
     const instance: *SoftInstance = @alignCast(@fieldParentPtr("interface", device.instance));
     const runtimes_count = switch (instance.threaded.async_limit) {
         .nothing => 1,
@@ -124,36 +133,76 @@ pub fn createGraphics(device: *base.Device, allocator: std.mem.Allocator, cache:
         },
     };
 
-    const runtimes = allocator.alloc(spv.Runtime, runtimes_count) catch return VkError.OutOfHostMemory;
-    errdefer allocator.free(runtimes);
-
-    //for (runtimes) |*runtime| {
-    //    runtime.* = spv.Runtime.init() catch |err| {
-    //        std.log.scoped(.SpvRuntimeInit).err("SPIR-V Runtime failed to initialize, {s}", .{@errorName(err)});
-    //        return VkError.Unknown;
-    //    };
-    //}
-
     self.* = .{
         .interface = interface,
-        .stages = std.enums.EnumMap(Stages, Shader).init(.{}),
+        .runtimes_allocator = runtimes_allocator_arena,
+        .stages = std.EnumMap(Stages, Shader).init(.{}),
     };
+
+    if (info.p_stages) |stages| {
+        for (stages[0..], 0..info.stage_count) |stage, _| {
+            var shader: Shader = undefined;
+
+            const module = try NonDispatchable(ShaderModule).fromHandleObject(stage.module);
+            const soft_module: *SoftShaderModule = @alignCast(@fieldParentPtr("interface", module));
+            soft_module.ref();
+            shader.module = soft_module;
+
+            const runtimes = runtimes_allocator.alloc(spv.Runtime, runtimes_count) catch return VkError.OutOfHostMemory;
+
+            for (runtimes) |*runtime| {
+                runtime.* = spv.Runtime.init(runtimes_allocator, &soft_module.module) catch |err| {
+                    std.log.scoped(.SpvRuntimeInit).err("SPIR-V Runtime failed to initialize, {s}", .{@errorName(err)});
+                    return VkError.Unknown;
+                };
+                if (stage.p_specialization_info) |specialization| {
+                    if (specialization.p_map_entries) |map| {
+                        const data: []const u8 = @as([*]const u8, @ptrCast(@alignCast(specialization.p_data)))[0..specialization.data_size];
+                        for (map[0..], 0..specialization.map_entry_count) |entry, _| {
+                            runtime.addSpecializationInfo(runtimes_allocator, .{
+                                .id = @intCast(entry.constant_id),
+                                .offset = @intCast(entry.offset),
+                                .size = @intCast(entry.size),
+                            }, data) catch return VkError.OutOfHostMemory;
+                        }
+                    }
+                }
+            }
+
+            shader.runtimes = runtimes;
+            shader.entry = runtimes_allocator.dupe(u8, std.mem.span(stage.p_name)) catch return VkError.OutOfHostMemory;
+
+            std.log.scoped(.GraphicsPipeline).debug("Created {d} runtimes for:", .{runtimes_count});
+
+            if (stage.stage.contains(.{ .vertex_bit = true })) {
+                std.log.scoped(.GraphicsPipeline).debug(">   Vertex stage", .{});
+                self.stages.put(.vertex, shader);
+            } else if (stage.stage.contains(.{ .fragment_bit = true })) {
+                std.log.scoped(.GraphicsPipeline).debug(">   Fragment stage", .{});
+                self.stages.put(.fragment, shader);
+            } else if (stage.stage.contains(.{ .tessellation_control_bit = true })) {
+                std.log.scoped(.GraphicsPipeline).debug(">   Tessellation control stage", .{});
+                self.stages.put(.tessellation_control, shader);
+            } else if (stage.stage.contains(.{ .tessellation_evaluation_bit = true })) {
+                std.log.scoped(.GraphicsPipeline).debug(">   Tessellation evaluation stage", .{});
+                self.stages.put(.tessellation_evaluation, shader);
+            } else if (stage.stage.contains(.{ .geometry_bit = true })) {
+                std.log.scoped(.GraphicsPipeline).debug(">   Geometry stage", .{});
+                self.stages.put(.geometry, shader);
+            } else {
+                std.log.scoped(.GraphicsPipeline).err(">   invalid stage", .{});
+                return VkError.Unknown;
+            }
+        }
+    } else {
+        return VkError.ValidationFailed;
+    }
+
     return self;
 }
 
 pub fn destroy(interface: *Interface, allocator: std.mem.Allocator) void {
     const self: *Self = @alignCast(@fieldParentPtr("interface", interface));
-    const soft_device: *SoftDevice = @alignCast(@fieldParentPtr("interface", interface.owner));
+    self.runtimes_allocator.deinit();
-    const device_allocator = soft_device.device_allocator.allocator();
-
-    var it = self.stages.iterator();
-    while (it.next()) |stage| {
-        for (stage.value.runtimes) |*runtime| {
-            runtime.deinit(device_allocator);
-        }
-        device_allocator.free(stage.value.runtimes);
-        device_allocator.free(stage.value.entry);
-        stage.value.module.unref(allocator);
-    }
     allocator.destroy(self);
 }

src/soft/device/ComputeDispatcher.zig

@@ -63,33 +63,21 @@ pub fn dispatch(self: *Self, group_count_x: u32, group_count_y: u32, group_count
 
     var wg: std.Io.Group = .init;
     for (0..@min(self.batch_size, group_count)) |batch_id| {
-        if (comptime base.config.single_threaded_compute) {
+        const run_data: RunData = .{
-            runWrapper(
+            .self = self,
-                RunData{
+            .batch_id = batch_id,
-                    .self = self,
+            .group_count = group_count,
-                    .batch_id = batch_id,
+            .group_count_x = @as(usize, @intCast(group_count_x)),
-                    .group_count = group_count,
+            .group_count_y = @as(usize, @intCast(group_count_y)),
-                    .group_count_x = @as(usize, @intCast(group_count_x)),
+            .group_count_z = @as(usize, @intCast(group_count_z)),
-                    .group_count_y = @as(usize, @intCast(group_count_y)),
+            .invocations_per_workgroup = invocations_per_workgroup,
-                    .group_count_z = @as(usize, @intCast(group_count_z)),
+            .pipeline = pipeline,
-                    .invocations_per_workgroup = invocations_per_workgroup,
+        };
-                    .pipeline = pipeline,
+
-                },
+        if (comptime base.config.single_threaded_compute)
-            );
+            runWrapper(run_data)
-        } else {
+        else
-            wg.async(self.device.interface.io(), runWrapper, .{
+            wg.async(self.device.interface.io(), runWrapper, .{run_data});
-                RunData{
-                    .self = self,
-                    .batch_id = batch_id,
-                    .group_count = group_count,
-                    .group_count_x = @as(usize, @intCast(group_count_x)),
-                    .group_count_y = @as(usize, @intCast(group_count_y)),
-                    .group_count_z = @as(usize, @intCast(group_count_z)),
-                    .invocations_per_workgroup = invocations_per_workgroup,
-                    .pipeline = pipeline,
-                },
-            });
-        }
     }
     wg.await(self.device.interface.io()) catch return VkError.DeviceLost;
 }

src/soft/device/Device.zig

@@ -1,6 +1,7 @@
 const std = @import("std");
 const vk = @import("vulkan");
 const base = @import("base");
+const lib = @import("../lib.zig");
 
 const SoftDescriptorSet = @import("../SoftDescriptorSet.zig");
 const SoftDevice = @import("../SoftDevice.zig");
@@ -15,29 +16,45 @@ const VkError = base.VkError;
 
 const Self = @This();
 
+pub const GRAPHICS_PIPELINE_STATE = 0;
+pub const COMPUTE_PIPELINE_STATE = 1;
+
 pub const PipelineState = struct {
     pipeline: ?*SoftPipeline,
     sets: [base.VULKAN_MAX_DESCRIPTOR_SETS]?*SoftDescriptorSet,
+    data: union {
+        compute: struct {},
+        graphics: struct {
+            vertex_buffers: [lib.MAX_VERTEX_INPUT_BINDINGS]Renderer.VertexBuffer,
+        },
+    },
 };
 
 compute: ComputeDispatcher,
 renderer: Renderer,
 
-/// .graphics = 0
-/// .compute = 1
 pipeline_states: [2]PipelineState,
 
 /// Initializating an execution device and
 /// not creating one to avoid dangling pointers
 pub fn init(self: *Self, device: *SoftDevice) void {
-    for (self.pipeline_states[0..]) |*state| {
+    for (self.pipeline_states[0..], 0..) |*state, i| {
         state.* = .{
             .pipeline = null,
             .sets = [_]?*SoftDescriptorSet{null} ** base.VULKAN_MAX_DESCRIPTOR_SETS,
+            .data = switch (i) {
+                GRAPHICS_PIPELINE_STATE => .{
+                    .graphics = .{
+                        .vertex_buffers = undefined,
+                    },
+                },
+                COMPUTE_PIPELINE_STATE => .{ .compute = .{} },
+                else => unreachable,
+            },
         };
     }
     self.compute = .init(device, &self.pipeline_states[@intFromEnum(vk.PipelineBindPoint.compute)]);
-    self.renderer = .init();
+    self.renderer = .init(device, &self.pipeline_states[@intFromEnum(vk.PipelineBindPoint.compute)]);
 }
 
 pub fn deinit(self: *Self) void {

src/soft/device/Renderer.zig

@@ -1,8 +1,13 @@
 const std = @import("std");
 const vk = @import("vulkan");
 const base = @import("base");
+const zm = @import("zmath");
 const lib = @import("../lib.zig");
 
+const F32x4 = zm.F32x4;
+
+const PipelineState = @import("Device.zig").PipelineState;
+
 const SoftBuffer = @import("../SoftBuffer.zig");
 const SoftDescriptorSet = @import("../SoftDescriptorSet.zig");
 const SoftDevice = @import("../SoftDevice.zig");
@@ -42,22 +47,48 @@ pub const DynamicState = struct {
 
     vertex_input_bindings: [lib.MAX_VERTEX_INPUT_BINDINGS]VertexInputBindingState,
     vertex_input_attributes: [lib.MAX_VERTEX_INPUT_ATTRIBUTES]VertexInputAttributeState,
-
-    vertex_buffers: [lib.MAX_VERTEX_INPUT_BINDINGS]VertexBuffer,
 };
 
+const Vertex = struct {
+    position: F32x4,
+    point_size: f32,
+    index: usize,
+};
+
+device: *SoftDevice,
+state: *PipelineState,
+
 render_pass: ?*SoftRenderPass,
 framebuffer: ?*SoftFramebuffer,
 dynamic_state: DynamicState,
 
-pub fn init() Self {
+pub fn init(device: *SoftDevice, state: *PipelineState) Self {
     return .{
+        .device = device,
+        .state = state,
         .render_pass = null,
         .framebuffer = null,
         .dynamic_state = undefined,
     };
 }
 
+pub fn drawPrimitive(self: *Self, vertex_count: usize, instance_count: usize, first_vertex: usize, first_instance: usize) void {
+    const allocator = self.device.device_allocator.allocator();
+
+    const vertices = self.fetchVertexInput(allocator, vertex_count, instance_count, first_vertex, first_instance);
+    _ = vertices;
+}
+
 pub fn deinit(self: *Self) void {
     _ = self;
 }
+
+fn fetchVertexInput(self: *const Self, allocator: std.mem.Allocator, vertex_count: usize, instance_count: usize, first_vertex: usize, first_instance: usize) []Vertex {
+    _ = self;
+    _ = allocator;
+    _ = vertex_count;
+    _ = instance_count;
+    _ = first_vertex;
+    _ = first_instance;
+    return undefined;
+}

src/vulkan/logger.zig

@@ -14,17 +14,19 @@ comptime {
 var mutex: std.Io.Mutex = .init;
 
 pub inline fn fixme(comptime format: []const u8, args: anytype) void {
-    if (!lib.config.logs) {
+    if (comptime !lib.config.logs) {
         return;
     }
     std.log.scoped(.FIXME).warn("FIXME: " ++ format, args);
 }
 
 pub fn log(comptime level: std.log.Level, comptime scope: @EnumLiteral(), comptime format: []const u8, args: anytype) void {
-    if (!lib.config.logs) {
+    if (comptime !lib.config.logs) {
         return;
     }
 
+    const io = std.Options.debug_io;
+
     const scope_name = @tagName(scope);
     const scope_prefix = comptime blk: {
         const limit = 30 - 4;
@@ -42,13 +44,6 @@ pub fn log(comptime level: std.log.Level, comptime scope: @EnumLiteral(), compti
         .err => .red,
     };
 
-    const allocator = std.heap.smp_allocator;
-
-    var threaded: std.Io.Threaded = .init(allocator, .{});
-    defer threaded.deinit();
-
-    const io = threaded.io();
-
     const stderr_file = std.Io.File.stderr();
     const stdout_file = std.Io.File.stdout();