move OS-related #defines to .h file; speed improvements

Combine two loops in stereo CIC computation into one for-loop.
Add compiler optimization "unroll-loops" to isr if OS enabled. Speed
improved from 113µs to 65µs per isr call for CIC interpolation filtering
(Measured on Teensy 3.2 96MHz optimized).

output_pt8211.cpp

@@ -29,12 +29,6 @@
 #include "output_pt8211.h"
 #include "memcpy_audio.h"
 
-	//uncomment to enable oversampling:
-#define OVERSAMPLING
-	//uncomment ONE of these to define interpolation type for oversampling:
-// #define INTERPOLATION_LINEAR
-#define INTERPOLATION_CIC
-
 audio_block_t * AudioOutputPT8211::block_left_1st = NULL;
 audio_block_t * AudioOutputPT8211::block_right_1st = NULL;
 audio_block_t * AudioOutputPT8211::block_left_2nd = NULL;
@@ -81,7 +75,6 @@ void AudioOutputPT8211::begin(void)
 	dma.attachInterrupt(isr);
 }
 
-
 void AudioOutputPT8211::isr(void)
 {
 	int16_t *dest;
@@ -145,46 +138,42 @@ void AudioOutputPT8211::isr(void)
 					static int32_t combROld[2] = {0};
 					
 					combL[0] = valL - oldL;
-					combL[1] = combL[0] - combLOld[0];
-					combL[2] = combL[1] - combLOld[1];
-					// combL[2] now holds input val
-					combLOld[0] = combL[0];
-					combLOld[1] = combL[1];
-					for (int j = 0; j < 4; j++) {
-						int32_t integrateL[3];
-						static int32_t integrateLOld[3] = {0};
-						integrateL[0] = ( (j==0) ? (combL[2]) : (0) ) + integrateLOld[0];
-						integrateL[1] = integrateL[0] + integrateLOld[1];
-						integrateL[2] = integrateL[1] + integrateLOld[2];
-						// integrateL[2] now holds j'th upsampled value
-						*(dest+j*2) = integrateL[2] >> 4;
-						integrateLOld[0] = integrateL[0];
-						integrateLOld[1] = integrateL[1];
-						integrateLOld[2] = integrateL[2];
-					}
 					combR[0] = valR - oldR;
+					combL[1] = combL[0] - combLOld[0];
 					combR[1] = combR[0] - combROld[0];
+					combL[2] = combL[1] - combLOld[1];
 					combR[2] = combR[1] - combROld[1];
+					// combL[2] now holds input val
 					// combR[2] now holds input val
+					oldL = valL;
+					oldR = valR;
+					combLOld[0] = combL[0];
 					combROld[0] = combR[0];
+					combLOld[1] = combL[1];
 					combROld[1] = combR[1];
-					
 					for (int j = 0; j < 4; j++) {
+						int32_t integrateL[3];
 						int32_t integrateR[3];
+						static int32_t integrateLOld[3] = {0};
 						static int32_t integrateROld[3] = {0};
+						integrateL[0] = ( (j==0) ? (combL[2]) : (0) ) + integrateLOld[0];
 						integrateR[0] = ( (j==0) ? (combR[2]) : (0) ) + integrateROld[0];
+						integrateL[1] = integrateL[0] + integrateLOld[1];
 						integrateR[1] = integrateR[0] + integrateROld[1];
+						integrateL[2] = integrateL[1] + integrateLOld[2];
 						integrateR[2] = integrateR[1] + integrateROld[2];
+						// integrateL[2] now holds j'th upsampled value
 						// integrateR[2] now holds j'th upsampled value
+						*(dest+j*2) = integrateL[2] >> 4;
 						*(dest+j*2+1) = integrateR[2] >> 4;
+						integrateLOld[0] = integrateL[0];
 						integrateROld[0] = integrateR[0];
+						integrateLOld[1] = integrateL[1];
 						integrateROld[1] = integrateR[1];
+						integrateLOld[2] = integrateL[2];
 						integrateROld[2] = integrateR[2];
 					}
-
 					dest+=8;
-					oldL = valL;
-					oldR = valR;
 				}
 			#else
 				#error no interpolation method defined for oversampling.
@@ -338,7 +327,6 @@ void AudioOutputPT8211::isr(void)
 
 
 
-
 void AudioOutputPT8211::update(void)
 {
 

output_pt8211.h

@@ -29,6 +29,12 @@
 #ifndef output_pt8211_h_
 #define output_pt8211_h_
 
+	//uncomment to enable oversampling:
+#define OVERSAMPLING
+	//uncomment ONE of these to define interpolation type for oversampling:
+// #define INTERPOLATION_LINEAR
+#define INTERPOLATION_CIC
+
 #include "Arduino.h"
 #include "AudioStream.h"
 #include "DMAChannel.h"
@@ -47,7 +53,11 @@ protected:
 	static audio_block_t *block_right_1st;
 	static bool update_responsibility;
 	static DMAChannel dma;
-	static void isr(void);
+	static void isr(void)
+	#if defined(OVERSAMPLING)
+		__attribute__((optimize("unroll-loops")))
+	#endif
+	;
 private:
 	static audio_block_t *block_left_2nd;
 	static audio_block_t *block_right_2nd;