diff --git a/README.md b/README.md
index 5fa245f..b964b6f 100644
--- a/README.md
+++ b/README.md
@@ -80,17 +80,16 @@ Notes:
   [.vag header](https://psx-spx.consoledev.net/cdromfileformats/#cdrom-file-audio-single-samples-vag-sony)
   at the beginning of the file. The header is always 48 bytes long for `vag`
   files, while in the case of `vagi` files it is padded to the size specified
-  using the `-a` option (2048 bytes by default).
-
-- The `vagi` format extends the standard .vag header by changing the signature
-  to `VAGi` and adding the following fields:
+  using the `-a` option (2048 bytes by default). The `vagi` format extends the
+  standard .vag header by adding the following fields:
   - the file's interleave size at offset `0x08-0x0B` (little endian);
   - the loop start offset in bytes-per-channel, if any, at offset `0x14-0x17`
     (little endian);
   - the file's channel count at offset `0x1E`.
 
-- The `spu`, `vag`, `spui` and `vagi` formats support setting a loop start
-  point. If the input file is either a .wav file with sampler metadata (`smpl`
+- The `spu` and `vag` formats support encoding a loop point as part of the ADPCM
+  data, while `vagi` supports storing one in the header for use by the stream
+  driver. If the input file is either a .wav file with sampler metadata (`smpl`
   chunk) or in a format FFmpeg supports parsing cue/chapter markers from, the
   first marker will be used as the loop point by default. The `-l` and `-n`
   options can be used to manually set a loop point or ignore the one present in
diff --git a/psxavenc/filefmt.c b/psxavenc/filefmt.c
index a4d96e1..b625590 100644
--- a/psxavenc/filefmt.c
+++ b/psxavenc/filefmt.c
@@ -307,6 +307,8 @@ void encode_file_spui(const args_t *args, decoder_t *decoder, FILE *output) {
 
 	if (args->format == FORMAT_VAGI)
 		fseek(output, header_size, SEEK_SET);
+	else if (args->audio_loop_point >= 0 && !(args->flags & FLAG_QUIET))
+		fprintf(stderr, "Warning: ignoring loop point as there is no header to store it in\n");
 
 	int audio_state_size = sizeof(psx_audio_encoder_channel_state_t) * args->audio_channels;
 	psx_audio_encoder_channel_state_t *audio_state = malloc(audio_state_size);
@@ -342,12 +344,15 @@ void encode_file_spui(const args_t *args, decoder_t *decoder, FILE *output) {
 			if (length > 0) {
 				uint8_t *last_block = chunk_ptr + length - PSX_AUDIO_SPU_BLOCK_SIZE;
 
-				if (args->flags & FLAG_SPU_ENABLE_LOOP) {
+				if (
+					(args->flags & FLAG_SPU_ENABLE_LOOP) ||
+					(decoder->end_of_input && args->audio_loop_point >= 0)
+				) {
 					last_block[1] = PSX_AUDIO_SPU_LOOP_REPEAT;
 				} else if (decoder->end_of_input) {
 					// HACK: the trailing block should in theory be appended to
 					// the existing data, but it's easier to just zerofill and
-					// repurpose the last encoded block
+					// repurpose the last encoded block.
 					memset(last_block, 0, PSX_AUDIO_SPU_BLOCK_SIZE);
 					last_block[1] = PSX_AUDIO_SPU_LOOP_TRAP;
 				}