optimize main plugin loop

this makes it faster to have the main loop focus on running the actual
plugin while pre- and post- parts of the loop become their own thing
with bigger buffers.

src/image.zig

@@ -7,6 +7,9 @@ const plugins = @import("plugin.zig");
 /// Approximate size of the BMP header.
 pub const BMPHeaderSize: usize = 82000;
 
+/// Buffer size for main copying
+pub const BufferSize: usize = 60000;
+
 pub const ImageError = error{
     OpenFail,
     InvalidPlugin,
@@ -155,6 +158,34 @@ pub const Image = struct {
         return n_read == 1;
     }
 
+    fn copyBytes(
+        self: *Image,
+        out_file: *c.SNDFILE,
+        buf: []f32,
+        start: usize,
+        end: usize,
+    ) !void {
+        const total_bytes = end - start;
+
+        var i: usize = start;
+        while (i < end) : (i += buf.len) {
+            const read_bytes = c.sf_readf_float(self.sndfile, buf.ptr, @intCast(i64, buf.len));
+
+            var view: []f32 = buf[0..buf.len];
+            if (i + buf.len > end) {
+                const excess = i + buf.len - end;
+                view = buf[0..excess];
+                _ = c.sf_seek(
+                    self.sndfile,
+                    -@intCast(i64, excess),
+                    c.SEEK_CUR,
+                );
+            }
+
+            try swrite(out_file, view.ptr, @intCast(i64, view.len));
+        }
+    }
+
     /// Run a plugin over the image.
     /// This setups a new lilv world/plugin among other things.
     /// The internal SNDFILE pointer is modified to point to the output of the
@@ -228,28 +259,52 @@ pub const Image = struct {
         // make sure we start from 0
         _ = c.sf_seek(self.sndfile, 0, c.SEEK_SET);
 
-        var i: usize = 0;
+        // there are four main stages:
+        // - the bmp header copy
+        // - pre-plugin
+        // - plugin
+        // - post-plugin
 
-        while (i < file_end) : (i += 1) {
+        var header_buf = try self.allocator.alloc(f32, BMPHeaderSize);
-            // if we're still on the bmp header phase, copy bytes
+        defer self.allocator.free(header_buf);
-            // if we aren't in the range from seek_pos, copy bytes
-            // if we are in the range, run lilv plugin
 
-            // TODO speed this up by not reading byte-by-byte.
+        var file_copy_buf = try self.allocator.alloc(f32, BufferSize);
+        defer self.allocator.free(file_copy_buf);
+
+        // bmp header copy
+        const header_len = @intCast(i64, header_buf.len);
+        _ = c.sf_readf_float(self.sndfile, header_buf.ptr, header_len);
+        try swrite(out_file, header_buf.ptr, header_len);
+
+        // pre-plugin copy
+        try self.copyBytes(
+            out_file,
+            file_copy_buf,
+            BMPHeaderSize,
+            seek_pos.start,
+        );
+
+        var i: usize = seek_pos.start;
+
+        while (i < seek_pos.end) : (i += 1) {
             const read_bytes = c.sf_readf_float(self.sndfile, rctx.in_buf.ptr, 1);
             if (read_bytes == 0) {
                 std.debug.warn("WARN! reached EOF at idx={}\n", i);
                 break;
             }
 
-            if (i < BMPHeaderSize or !seek_pos.contains(i)) {
+            lv2.lilv_instance_run(rctx.instance, 1);
-                try swrite(out_file, rctx.in_buf.ptr, 1);
+            try swrite(out_file, rctx.out_buf.ptr, 1);
-            } else {
-                lv2.lilv_instance_run(rctx.instance, 1);
-                try swrite(out_file, rctx.out_buf.ptr, 1);
-            }
         }
 
+        // post-plugin copy
+        try self.copyBytes(
+            out_file,
+            file_copy_buf,
+            seek_pos.end,
+            file_end,
+        );
+
         c.sf_write_sync(out_file);
         _ = c.sf_close(out_file);
         _ = c.sf_close(self.sndfile);