From 7a52e42ee0676f47e801dd348b478302de5d2e50 Mon Sep 17 00:00:00 2001
From: Eric <ewild@sysmocom.de>
Date: Fri, 14 Aug 2020 03:11:22 +0200
Subject: transceiver: optimize code if optimizations are enabled

There is no point in checking basic stuff ten thousand times per second
since the sizes never change, so it's enough to enable the
checks/assertions for unoptimized (debug) builds.

This significantly decreases branch mispredictions.

Change-Id: Iebd9e91b3c7f37f2dc646d3017c45139977e4d15
---
 CommonLibs/Vector.h                | 33 +++++++++++++++++++--------------
 Transceiver52M/Resampler.cpp       |  6 ++++--
 Transceiver52M/arch/x86/convolve.c |  6 ++++--
 3 files changed, 27 insertions(+), 18 deletions(-)

diff --git a/CommonLibs/Vector.h b/CommonLibs/Vector.h
index d55c5b3..6d1045d 100644
--- a/CommonLibs/Vector.h
+++ b/CommonLibs/Vector.h
@@ -36,6 +36,11 @@
 #include <assert.h>
 #include <stdlib.h>
 
+#ifndef __OPTIMIZE__
+#define assert_no_opt(x) assert(x)
+#else
+#define assert_no_opt(x)
+#endif
 // We can't use Logger.h in this file...
 extern int gVectorDebug;
 #define BVDEBUG(msg) if (gVectorDebug) {std::cout << msg;}
@@ -81,8 +86,8 @@ template <class T> class Vector {
 	/** Return the size of the Vector. */
 	size_t size() const
 	{
-		assert(mStart>=mData);
-		assert(mEnd>=mStart);
+		assert_no_opt(mStart>=mData);
+		assert_no_opt(mEnd>=mStart);
 		return mEnd - mStart;
 	}
 
@@ -112,7 +117,7 @@ template <class T> class Vector {
 	/** Reduce addressable size of the Vector, keeping content. */
 	void shrink(size_t newSize)
 	{
-		assert(newSize <= mEnd - mStart);
+		assert_no_opt(newSize <= mEnd - mStart);
 		mEnd = mStart + newSize;
 	}
 
@@ -199,7 +204,7 @@ template <class T> class Vector {
 	{
 		T* wStart = mStart + start;
 		T* wEnd = wStart + span;
-		assert(wEnd<=mEnd);
+		assert_no_opt(wEnd<=mEnd);
 		return Vector<T>(NULL,wStart,wEnd);
 	}
 
@@ -208,7 +213,7 @@ template <class T> class Vector {
 	{
 		T* wStart = mStart + start;
 		T* wEnd = wStart + span;
-		assert(wEnd<=mEnd);
+		assert_no_opt(wEnd<=mEnd);
 		return Vector<T>(NULL,wStart,wEnd);
 	}
 
@@ -228,8 +233,8 @@ template <class T> class Vector {
 		unsigned int i;
 		T* dst = other.mStart + start;
 		T* src = mStart;
-		assert(dst+span<=other.mEnd);
-		assert(mStart+span<=mEnd);
+		assert_no_opt(dst+span<=other.mEnd);
+		assert_no_opt(mStart+span<=mEnd);
 		for (i = 0; i < span; i++, src++, dst++)
 			*dst = *src;
 		/*TODO if not non-trivially copiable type class, optimize:
@@ -250,8 +255,8 @@ template <class T> class Vector {
 	void segmentCopyTo(Vector<T>& other, size_t start, size_t span) const
 	{
 		const T* base = mStart + start;
-		assert(base+span<=mEnd);
-		assert(other.mStart+span<=other.mEnd);
+		assert_no_opt(base+span<=mEnd);
+		assert_no_opt(other.mStart+span<=other.mEnd);
 		memcpy(other.mStart,base,span*sizeof(T));
 	}
 
@@ -265,8 +270,8 @@ template <class T> class Vector {
 	{
 		const T* baseFrom = mStart + from;
 		T* baseTo = mStart + to;
-		assert(baseFrom+span<=mEnd);
-		assert(baseTo+span<=mEnd);
+		assert_no_opt(baseFrom+span<=mEnd);
+		assert_no_opt(baseTo+span<=mEnd);
 		memmove(baseTo,baseFrom,span*sizeof(T));
 	}
 
@@ -280,7 +285,7 @@ template <class T> class Vector {
 	{
 		T* dp=mStart+start;
 		T* end=dp+length;
-		assert(end<=mEnd);
+		assert_no_opt(end<=mEnd);
 		while (dp<end) *dp++=val;
 	}
 
@@ -292,13 +297,13 @@ template <class T> class Vector {
 
 	T& operator[](size_t index)
 	{
-		assert(mStart+index<mEnd);
+		assert_no_opt(mStart+index<mEnd);
 		return mStart[index];
 	}
 
 	const T& operator[](size_t index) const
 	{
-		assert(mStart+index<mEnd);
+		assert_no_opt(mStart+index<mEnd);
 		return mStart[index];
 	}
 
diff --git a/Transceiver52M/Resampler.cpp b/Transceiver52M/Resampler.cpp
index 7ba0219..2ca6406 100644
--- a/Transceiver52M/Resampler.cpp
+++ b/Transceiver52M/Resampler.cpp
@@ -99,6 +99,7 @@ void Resampler::initFilters(float bw)
 		reverse(&part[0], &part[filt_len]);
 }
 
+#ifndef __OPTIMIZE__
 static bool check_vec_len(int in_len, int out_len, int p, int q)
 {
 	if (in_len % q) {
@@ -129,14 +130,15 @@ static bool check_vec_len(int in_len, int out_len, int p, int q)
 
 	return true;
 }
+#endif
 
 int Resampler::rotate(const float *in, size_t in_len, float *out, size_t out_len)
 {
 	int n, path;
-
+#ifndef __OPTIMIZE__
 	if (!check_vec_len(in_len, out_len, p, q))
 		return -1;
-
+#endif
 	/* Generate output from precomputed input/output paths */
 	for (size_t i = 0; i < out_len; i++) {
 		n = in_index[i];
diff --git a/Transceiver52M/arch/x86/convolve.c b/Transceiver52M/arch/x86/convolve.c
index 66fca74..81ea782 100644
--- a/Transceiver52M/arch/x86/convolve.c
+++ b/Transceiver52M/arch/x86/convolve.c
@@ -99,11 +99,12 @@ int convolve_real(const float *x, int x_len,
 		  const float *h, int h_len,
 		  float *y, int y_len, int start, int len)
 {
+#ifndef __OPTIMIZE__
 	if (bounds_check(x_len, h_len, y_len, start, len) < 0)
 		return -1;
 
 	memset(y, 0, len * 2 * sizeof(float));
-
+#endif
 	switch (h_len) {
 	case 4:
 		c.conv_real4(x, x_len, h, h_len, y, y_len, start, len);
@@ -138,11 +139,12 @@ int convolve_complex(const float *x, int x_len,
 		     float *y, int y_len,
 		     int start, int len)
 {
+#ifndef __OPTIMIZE__
 	if (bounds_check(x_len, h_len, y_len, start, len) < 0)
 		return -1;
 
 	memset(y, 0, len * 2 * sizeof(float));
-
+#endif
 	if (!(h_len % 8))
 		c.conv_cmplx_8n(x, x_len, h, h_len, y, y_len, start, len);
 	else if (!(h_len % 4))
-- 
cgit v1.2.3