adding face culling, better varying management and more primitive topologies

src/soft/device/Renderer.zig

@@ -40,7 +40,10 @@ pub const DynamicState = struct {
 
 pub const Vertex = struct {
     position: F32x4,
-    outputs: [spv.SPIRV_MAX_OUTPUT_LOCATIONS]?[]u8,
+    outputs: [spv.SPIRV_MAX_OUTPUT_LOCATIONS]?struct {
+        interpolation_type: enum { smooth, flat, noperspective },
+        blob: []u8,
+    },
 };
 
 pub const Fragment = struct {
@@ -99,7 +102,7 @@ pub fn draw(self: *Self, vertex_count: usize, instance_count: usize, first_verte
     };
 
     for (draw_call.vertices) |*vertex| {
-        vertex.outputs = [_]?[]u8{null} ** spv.SPIRV_MAX_OUTPUT_LOCATIONS;
+        @memset(vertex.outputs[0..], null);
     }
 
     self.vertexShaderStage(allocator, &draw_call, vertex_count, instance_count) catch |err| {
@@ -215,21 +218,34 @@ fn rasterizationStage(self: *Self, allocator: std.mem.Allocator, draw_call: *Dra
     const pipeline_data = (self.state.pipeline orelse return VkError.InvalidHandleDrv).interface.mode.graphics;
     const topology = pipeline_data.input_assembly.topology;
     switch (topology) {
-        .triangle_list => for (0..@divExact(draw_call.vertices.len, 3)) |triangle_index| {
+        .triangle_list => for (0..@divTrunc(draw_call.vertices.len, 3)) |triangle_index| {
             const first_vertex = triangle_index * 3;
             const v0 = &draw_call.vertices[first_vertex + 0];
             const v1 = &draw_call.vertices[first_vertex + 1];
             const v2 = &draw_call.vertices[first_vertex + 2];
 
-            switch (pipeline_data.rasterization.polygon_mode) {
+            try self.rasterizeTriangle(allocator, &fragments, v0, v1, v2, v0, v1, v2);
-                .fill => try rasterizer.drawTriangleFilled(allocator, &fragments, v0, v1, v2),
+        },
-                .line => {
+        .triangle_fan => if (draw_call.vertices.len >= 3) {
-                    try rasterizer.drawLineBresenham(allocator, &fragments, v0, v1);
+            const v0 = &draw_call.vertices[0];
-                    try rasterizer.drawLineBresenham(allocator, &fragments, v1, v2);
+            for (1..(draw_call.vertices.len - 1)) |vertex_index| {
-                    try rasterizer.drawLineBresenham(allocator, &fragments, v2, v0);
+                const v1 = &draw_call.vertices[vertex_index];
-                },
+                const v2 = &draw_call.vertices[vertex_index + 1];
-                .point => {},
+
-                else => base.unsupported("polygon mode {any}", .{pipeline_data.rasterization.polygon_mode}),
+                try self.rasterizeTriangle(allocator, &fragments, v0, v1, v2, v0, v1, v2);
+            }
+        },
+        .triangle_strip => if (draw_call.vertices.len >= 3) {
+            for (0..(draw_call.vertices.len - 2)) |vertex_index| {
+                const v0 = &draw_call.vertices[vertex_index + 0];
+                const v1 = &draw_call.vertices[vertex_index + 1];
+                const v2 = &draw_call.vertices[vertex_index + 2];
+
+                if ((vertex_index & 1) == 0) {
+                    try self.rasterizeTriangle(allocator, &fragments, v0, v1, v2, v0, v1, v2);
+                } else {
+                    try self.rasterizeTriangle(allocator, &fragments, v0, v1, v2, v1, v0, v2);
+                }
             }
         },
         else => base.unsupported("primitive topology {any}", .{topology}),
@@ -238,6 +254,68 @@ fn rasterizationStage(self: *Self, allocator: std.mem.Allocator, draw_call: *Dra
     draw_call.fragments = fragments.toOwnedSlice(allocator) catch return VkError.OutOfDeviceMemory;
 }
 
+fn triangleArea2(v0: *const Vertex, v1: *const Vertex, v2: *const Vertex) f32 {
+    const x0 = v0.position[0];
+    const y0 = v0.position[1];
+    const x1 = v1.position[0];
+    const y1 = v1.position[1];
+    const x2 = v2.position[0];
+    const y2 = v2.position[1];
+
+    return ((x1 - x0) * (y2 - y0)) - ((y1 - y0) * (x2 - x0));
+}
+
+fn triangleIsCulled(self: *Self, v0: *const Vertex, v1: *const Vertex, v2: *const Vertex) VkError!bool {
+    const pipeline_data = (self.state.pipeline orelse return VkError.InvalidHandleDrv).interface.mode.graphics;
+    const rasterization = pipeline_data.rasterization;
+    const cull_mode = rasterization.cull_mode;
+
+    if (!cull_mode.front_bit and !cull_mode.back_bit)
+        return false;
+
+    if (cull_mode.front_bit and cull_mode.back_bit)
+        return true;
+
+    const area = triangleArea2(v0, v1, v2);
+    if (area == 0.0)
+        return true;
+
+    const front_face = switch (rasterization.front_face) {
+        .counter_clockwise => area < 0.0,
+        .clockwise => area > 0.0,
+        else => return false,
+    };
+
+    return (cull_mode.front_bit and front_face) or (cull_mode.back_bit and !front_face);
+}
+
+fn rasterizeTriangle(
+    self: *Self,
+    allocator: std.mem.Allocator,
+    fragments: *std.ArrayList(Fragment),
+    v0: *Vertex,
+    v1: *Vertex,
+    v2: *Vertex,
+    cull_v0: *const Vertex,
+    cull_v1: *const Vertex,
+    cull_v2: *const Vertex,
+) VkError!void {
+    if (try self.triangleIsCulled(cull_v0, cull_v1, cull_v2))
+        return;
+
+    const pipeline_data = (self.state.pipeline orelse return VkError.InvalidHandleDrv).interface.mode.graphics;
+    switch (pipeline_data.rasterization.polygon_mode) {
+        .fill => try rasterizer.drawTriangleFilled(allocator, fragments, v0, v1, v2),
+        .line => {
+            try rasterizer.drawLineBresenham(allocator, fragments, v0, v1);
+            try rasterizer.drawLineBresenham(allocator, fragments, v1, v2);
+            try rasterizer.drawLineBresenham(allocator, fragments, v2, v0);
+        },
+        .point => {},
+        else => base.unsupported("polygon mode {any}", .{pipeline_data.rasterization.polygon_mode}),
+    }
+}
+
 fn fragmentShaderStage(self: *Self, draw_call: *DrawCall) !void {
     const pipeline = self.state.pipeline orelse return;
     const batch_size = (pipeline.stages.getPtr(.fragment) orelse return).runtimes.len;

src/soft/device/fragment_dispatcher.zig

@@ -48,9 +48,7 @@ inline fn run(data: RunData) !void {
                 SpvRuntimeError.NotFound => continue,
                 else => return err,
             };
-            if (result_word != 0) {
+            try rt.writeInput(fragment.inputs[location], result_word);
-                try rt.writeInput(fragment.inputs[location], result_word);
-            }
         }
 
         rt.callEntryPoint(allocator, entry) catch |err| switch (err) {

src/soft/device/rasterizer.zig

@@ -37,31 +37,31 @@ fn interpolateVertexOutputs(
         const out1 = v1.outputs[location] orelse continue;
         const out2 = v2.outputs[location] orelse continue;
 
-        if (out0.len == 0) {
+        if (out0.interpolation_type == .flat or out0.blob.len == 0) {
-            inputs[location] = out0;
+            inputs[location] = out0.blob;
             continue;
         }
 
-        const len = @min(out0.len, out1.len, out2.len);
+        const len = @min(out0.blob.len, out1.blob.len, out2.blob.len);
         const input = allocator.alloc(u8, len) catch return VkError.OutOfDeviceMemory;
 
         var byte_index: usize = 0;
         while (byte_index + @sizeOf(F32x4) <= len) : (byte_index += @sizeOf(F32x4)) {
-            const value0 = std.mem.bytesToValue(F32x4, out0[byte_index..]);
+            const value0 = std.mem.bytesToValue(F32x4, out0.blob[byte_index..]);
-            const value1 = std.mem.bytesToValue(F32x4, out1[byte_index..]);
+            const value1 = std.mem.bytesToValue(F32x4, out1.blob[byte_index..]);
-            const value2 = std.mem.bytesToValue(F32x4, out2[byte_index..]);
+            const value2 = std.mem.bytesToValue(F32x4, out2.blob[byte_index..]);
             writePacked(F32x4, input[byte_index..], interpolateF32x4(value0, value1, value2, b0, b1, b2));
         }
 
         while (byte_index + @sizeOf(f32) <= len) : (byte_index += @sizeOf(f32)) {
-            const value0 = std.mem.bytesToValue(f32, out0[byte_index..]);
+            const value0 = std.mem.bytesToValue(f32, out0.blob[byte_index..]);
-            const value1 = std.mem.bytesToValue(f32, out1[byte_index..]);
+            const value1 = std.mem.bytesToValue(f32, out1.blob[byte_index..]);
-            const value2 = std.mem.bytesToValue(f32, out2[byte_index..]);
+            const value2 = std.mem.bytesToValue(f32, out2.blob[byte_index..]);
             writePacked(f32, input[byte_index..], (value0 * b0) + (value1 * b1) + (value2 * b2));
         }
 
         if (byte_index < len)
-            @memcpy(input[byte_index..], out0[byte_index..len]);
+            @memcpy(input[byte_index..], out0.blob[byte_index..len]);
 
         inputs[location] = input;
     }

src/soft/device/vertex_dispatcher.zig

@@ -80,12 +80,11 @@ inline fn run(data: RunData) !void {
                 SpvRuntimeError.NotFound => continue,
                 else => return err,
             };
-            if (result_word == 0)
+            output.outputs[location] = .{
-                continue;
+                .interpolation_type = if (rt.hasResultDecoration(result_word, .Flat)) .flat else .smooth, // TODO : handle noperspective
-            const value = rt.results[result_word].getConstValue() catch continue;
+                .blob = data.allocator.alloc(u8, try rt.getResultMemorySize(result_word)) catch return VkError.OutOfDeviceMemory,
-            const needed_size = try value.getPlainMemorySize();
+            };
-            output.outputs[location] = data.allocator.alloc(u8, needed_size) catch return VkError.OutOfDeviceMemory;
+            try rt.readOutput(output.outputs[location].?.blob, result_word);
-            try rt.readOutput(output.outputs[location].?, result_word);
         }
     }
 }