From 4a4442ad91d34964b1d695381fafbdef07bb14ce Mon Sep 17 00:00:00 2001
From: Chris Robinson <chris.kcat@gmail.com>
Date: Mon, 8 May 2017 16:23:16 -0700
Subject: Store the output limiter values as fixed-point integers

This helps keep the squared sum stable over larger updates, also avoiding the
need to keep recalculating it.
---
 Alc/ALu.c              | 17 ++++++-----------
 OpenAL32/Include/alu.h |  8 ++++++--
 2 files changed, 12 insertions(+), 13 deletions(-)

diff --git a/Alc/ALu.c b/Alc/ALu.c
index 8d9356c5..b8c93c69 100644
--- a/Alc/ALu.c
+++ b/Alc/ALu.c
@@ -1455,24 +1455,18 @@ static void ApplyLimiter(struct OutputLimiter *Limiter,
     {
         ALfloat lastgain = Limiter->Gain;
         ALsizei wpos = Limiter->Pos;
-        ALfloat sum = 0.0f;
+        ALuint sum = Limiter->SquaredSum;
         ALfloat gain;
 
-        /* Unfortunately we can't store the running sum due to fp inaccuracies
-         * causing it to drift over time. So we need to recalculate it every
-         * once in a while (i.e. every invocation).
-         */
-        for(i = 0;i < LIMITER_WINDOW_SIZE;i++)
-            sum += Limiter->Window[i];
-
         for(i = 0;i < SamplesToDo;i++)
         {
             sum -= Limiter->Window[wpos];
-            Limiter->Window[wpos] = Values[i];
-            sum += Values[i];
+            Limiter->Window[wpos] = fastf2u(minf(Values[i]*65536.0f, LIMITER_VALUE_MAX));
+            sum += Limiter->Window[wpos];
 
             /* Clamp limiter range to 0dB...-80dB. */
-            gain = 1.0f / clampf(sqrtf(sum / (ALfloat)LIMITER_WINDOW_SIZE), 1.0f, 1000.0f);
+            gain = 1.0f / clampf(sqrtf((ALfloat)sum / ((ALfloat)LIMITER_WINDOW_SIZE*65536.0f)),
+                                 1.0f, 1000.0f);
             if(lastgain >= gain)
                 lastgain = maxf(lastgain*AttackRate, gain);
             else
@@ -1485,6 +1479,7 @@ static void ApplyLimiter(struct OutputLimiter *Limiter,
 
         Limiter->Gain = lastgain;
         Limiter->Pos = wpos;
+        Limiter->SquaredSum = sum;
     }
     if(do_limit)
     {
diff --git a/OpenAL32/Include/alu.h b/OpenAL32/Include/alu.h
index 6fdbac6f..8ac25e35 100644
--- a/OpenAL32/Include/alu.h
+++ b/OpenAL32/Include/alu.h
@@ -299,9 +299,13 @@ void DeinitVoice(ALvoice *voice);
 
 #define LIMITER_WINDOW_SIZE (1<<7) /* 128 */
 #define LIMITER_WINDOW_MASK (LIMITER_WINDOW_SIZE-1)
+#define LIMITER_VALUE_MAX  (UINT_MAX / LIMITER_WINDOW_SIZE)
 struct OutputLimiter {
-    /* RMS detection window and the next write pos. */
-    alignas(16) ALfloat Window[LIMITER_WINDOW_SIZE];
+    /* RMS detection window, sum of values in the window, and the next write
+     * pos. Values are 16.16 fixed-point.
+     */
+    ALuint Window[LIMITER_WINDOW_SIZE];
+    ALuint SquaredSum;
     ALsizei Pos;
 
     /* In milliseconds. */
-- 
cgit v1.2.3