improving fragment outputs management

src/soft/device/blitter.zig

@@ -673,8 +673,11 @@ pub fn readFloat4(map: []const u8, src_format: vk.Format) F32x4 {
         },
 
         .r16g16b16a16_sfloat => c = std.mem.bytesToValue(@Vector(4, f16), map),
+
         .r32g32b32a32_sfloat => c = std.mem.bytesToValue(F32x4, map),
 
+        .r32g32b32a32_uint => c = @floatFromInt(std.mem.bytesToValue(U32x4, map)),
+
         .s8_uint => c[0] = @floatFromInt(map[0]),
 
         .b8g8r8a8_srgb,

src/soft/device/fragment.zig

@@ -18,13 +18,13 @@ pub fn shaderInvocation(
     batch_id: usize,
     position: zm.F32x4,
     inputs: [spv.SPIRV_MAX_OUTPUT_LOCATIONS]VertexInterpolation,
-) SpvRuntimeError!zm.F32x4 {
+) SpvRuntimeError![spv.SPIRV_MAX_OUTPUT_LOCATIONS][@sizeOf(zm.F32x4)]u8 {
     const io = draw_call.renderer.device.interface.io();
 
     _ = position;
-    const pipeline = draw_call.renderer.state.pipeline orelse return zm.f32x4s(0.0);
+    const pipeline = draw_call.renderer.state.pipeline orelse return undefined;
 
-    const shader = pipeline.stages.getPtrAssertContains(.fragment);
+    const shader = pipeline.stages.getPtr(.fragment) orelse return undefined;
     const runtime = &shader.runtimes[batch_id];
     const mutex = &runtime.mutex;
     const rt = &runtime.rt;
@@ -53,10 +53,11 @@ pub fn shaderInvocation(
         else => return err,
     };
 
-    var color = zm.f32x4s(0.0);
-    try rt.readOutput(std.mem.asBytes(&color), output_result);
+    var outputs: [spv.SPIRV_MAX_OUTPUT_LOCATIONS][@sizeOf(zm.F32x4)]u8 = undefined;
+
+    try rt.readOutput(std.mem.asBytes(&outputs), output_result);
 
     try rt.flushDescriptorSets(allocator);
 
-    return std.math.clamp(color, zm.f32x4s(0.0), zm.f32x4s(1.0));
+    return outputs;
 }

src/soft/device/rasterizer/bresenham.zig

@@ -157,19 +157,22 @@ inline fn run(data: RunData) !void {
             return;
         };
 
-        try render_target.writeFloat4(
-            .{
-                .x = pixel_x,
-                .y = pixel_y,
-                .z = 0, // FIXME
-            },
-            .{
-                .aspect_mask = render_target_view.subresource_range.aspect_mask,
-                .mip_level = render_target_view.subresource_range.base_mip_level,
-                .array_layer = render_target_view.subresource_range.base_array_layer,
-            },
-            render_target_view.format,
-            pixel,
-        );
+        _ = pixel;
+        _ = render_target;
+
+        //try render_target.writeFloat4(
+        //    .{
+        //        .x = pixel_x,
+        //        .y = pixel_y,
+        //        .z = 0, // FIXME
+        //    },
+        //    .{
+        //        .aspect_mask = render_target_view.subresource_range.aspect_mask,
+        //        .mip_level = render_target_view.subresource_range.base_mip_level,
+        //        .array_layer = render_target_view.subresource_range.base_array_layer,
+        //    },
+        //    render_target_view.format,
+        //    pixel,
+        //);
     }
 }

src/soft/device/rasterizer/edge_function.zig

@@ -162,7 +162,7 @@ inline fn run(data: RunData) !void {
                     continue;
             }
 
-            const pixel = fragment.shaderInvocation(
+            const outputs = fragment.shaderInvocation(
                 data.allocator,
                 data.draw_call,
                 data.batch_id,
@@ -175,7 +175,6 @@ inline fn run(data: RunData) !void {
                         std.debug.dumpErrorReturnTrace(trace);
                     }
                 }
-
                 return;
             };
 
@@ -192,9 +191,17 @@ inline fn run(data: RunData) !void {
                 if (z >= depth_value[0])
                     continue;
                 blitter.writeFloat4(zm.f32x4s(z), depth.base[depth_offset..], depth.format);
-                blitter.writeFloat4(pixel, data.color_attachment_access.base[color_offset..], data.color_attachment_access.format);
+
+                // Doubled line to stay inside the critical section
+                if (base.format.isUnnormalizedInteger(data.color_attachment_access.format))
+                    blitter.writeInt4(std.mem.bytesToValue(@Vector(4, u32), &outputs[0]), data.color_attachment_access.base[color_offset..], data.color_attachment_access.format)
+                else
+                    blitter.writeFloat4(std.mem.bytesToValue(@Vector(4, f32), &outputs[0]), data.color_attachment_access.base[color_offset..], data.color_attachment_access.format);
             } else {
-                blitter.writeFloat4(pixel, data.color_attachment_access.base[color_offset..], data.color_attachment_access.format);
+                if (base.format.isUnnormalizedInteger(data.color_attachment_access.format))
+                    blitter.writeInt4(std.mem.bytesToValue(@Vector(4, u32), &outputs[0]), data.color_attachment_access.base[color_offset..], data.color_attachment_access.format)
+                else
+                    blitter.writeFloat4(std.mem.bytesToValue(@Vector(4, f32), &outputs[0]), data.color_attachment_access.base[color_offset..], data.color_attachment_access.format);
             }
         }
     }