.diff

diff --git a/libctru/include/3ds/services/gspgpu.h b/libctru/include/3ds/services/gspgpu.h
index e03de9c9..2cf9693d 100644
--- a/libctru/include/3ds/services/gspgpu.h
+++ b/libctru/include/3ds/services/gspgpu.h
@@ -62,7 +62,17 @@ typedef enum
 	GSPGPU_EVENT_MAX,      ///< Used to know how many events there are.
 } GSPGPU_Event;
 
-/// GSPGPU performance log entry
+/**
+ * GSPGPU performance log entry.
+ *
+ * Use the lastDurationUs field when benchmarking single GPU operations, this is usally meant
+ * for 3D library writers.
+ *
+ * Use the difference between two totalDurationUs when using a GPU library (e.g. citro3d), as
+ * there can be multiple GPU operations (e.g. P3D, PPF) per render pass, or per frame, and so on.
+ * Don't use totalDurationUs as-is (rather, take the difference as just described), because it
+ * can overflow.
+ */
 typedef struct
 {
 	u32 lastDurationUs;    ///< Duration of the last corresponding PICA200 operation (time between op is started and IRQ is received).
@@ -294,7 +304,11 @@ Result GSPGPU_SetPerfLogMode(bool enabled);
 
 /**
  * @brief Retrieves the performance log.
- * @param outPerfLog Pointer to output the performance log to.
+ * @param[out] outPerfLog Pointer to output the performance log to.
+ * @note  Use the difference between two totalDurationUs when using a GPU library (e.g. citro3d), as
+ *        there can be multiple GPU operations (e.g. P3D, PPF) per render pass, or per frame, and so on.
+ *        Don't use totalDurationUs as-is (rather, take the difference as just described), because it
+ *        can overflow.
  * @note  For a MemoryFill operation that uses both PSC0 and PSC1, take the maximum
  *        of the two "last duration" entries.
  * @note  For PDC0/PDC1 (VBlank0/1), the "last duration" entry corresponds to the time between
@@ -304,7 +318,9 @@ Result GSPGPU_SetPerfLogMode(bool enabled);
  *        in the PDC0 handler.
  * @bug   The official sysmodule doesn't handle the PDC0/1 entries correctly after init. On the first
  *        frame \ref GSPGPU_SetPerfLogMode is enabled, "last duration" will have a nonsensical
- *        value; and "total duration" stays nonsensical.
+ *        value; and "total duration" stays nonsensical. This isn't much of a problem, except for the
+ *        first frame, because "total duration" is not supposed to be used as-is (you are supposed
+ *        to take the difference of this field between two time points of your choosing, instead).
  * @bug   Since it is running at approx. 3.25 GiB/s per bank, some small PSC operations might
  *        complete before the official GSP has time to record the start time.
  * @bug   The official sysmodule doesn't properly handle data synchronization for the perflog,