Update some documentation.

miniaudio.h

@@ -460,6 +460,8 @@ is at the end, use `ma_sound_at_end()`. Looping of a sound can be controlled wit
 miniaudio should work cleanly out of the box without the need to download or install any
 dependencies. See below for platform-specific details.
 
+Note that GCC and Clang require `-msse2`, `-mavx2`, etc. for SIMD optimizations.
+
 
 2.1. Windows
 ------------
@@ -3601,6 +3603,12 @@ Some backends have some nuance details you may want to be aware of.
   miniaudio's built-in resampler is to take advantage of any potential device-specific
   optimizations the driver may implement.
 
+BSD
+---
+- The sndio backend is currently only enabled on OpenBSD builds.
+- The audio(4) backend is supported on OpenBSD, but you may need to disable sndiod before you can
+  use it.
+
 15.4. UWP
 ---------
 - UWP only supports default playback and capture devices.
@@ -3631,14 +3639,28 @@ Some backends have some nuance details you may want to be aware of.
 
 16. Optimization Tips
 =====================
+See below for some tips on improving performance.
 
-16.1. High Level API
+16.1. Low Level API
+-------------------
+- In the data callback, if your data is already clipped prior to copying it into the output buffer,
+  set the `noClip` config option in the device config to true. This will disable miniaudio's built
+  in clipping function.
+- By default, miniaudio will pre-silence the data callback's output buffer. If you know that you
+  will always write valid data to the output buffer you can disable pre-silencing by setting the
+  `noPreSilence` config option in the device config to true.
+
+16.2. High Level API
 --------------------
 - If a sound does not require doppler or pitch shifting, consider disabling pitching by
   initializing the sound with the `MA_SOUND_FLAG_NO_PITCH` flag.
 - If a sound does not require spatialization, disable it by initializing the sound with the
   `MA_SOUND_FLAG_NO_SPATIALIZATION` flag. It can be re-enabled again post-initialization with
   `ma_sound_set_spatialization_enabled()`.
+- If you know all of your sounds will always be the same sample rate, set the engine's sample
+  rate to match that of the sounds. Likewise, if you're using a self-managed resource manager,
+  consider setting the decoded sample rate to match your sounds. By configuring everything to
+  use a consistent sample rate, sample rate conversion can be avoided.
 
 
 
@@ -3647,17 +3669,6 @@ Some backends have some nuance details you may want to be aware of.
 - Automatic stream routing is enabled on a per-backend basis. Support is explicitly enabled for
   WASAPI and Core Audio, however other backends such as PulseAudio may naturally support it, though
   not all have been tested.
-- The contents of the output buffer passed into the data callback will always be pre-initialized to
-  silence unless the `noPreSilencedOutputBuffer` config variable in `ma_device_config` is set to
-  true, in which case it'll be undefined which will require you to write something to the entire
-  buffer.
-- By default miniaudio will automatically clip samples. This only applies when the playback sample
-  format is configured as `ma_format_f32`. If you are doing clipping yourself, you can disable this
-  overhead by setting `noClip` to true in the device config.
-- Note that GCC and Clang requires `-msse2`, `-mavx2`, etc. for SIMD optimizations.
-- The sndio backend is currently only enabled on OpenBSD builds.
-- The audio(4) backend is supported on OpenBSD, but you may need to disable sndiod before you can
-  use it.
 - When compiling with VC6 and earlier, decoding is restricted to files less than 2GB in size. This
   is due to 64-bit file APIs not being available.
 */