From 86df10be88d26550fa3b9e72d9b62e68bcfeffcc Mon Sep 17 00:00:00 2001
From: Ingo <heckflosse@i-weyrich.de>
Date: Sun, 7 Jun 2015 12:18:38 +0200
Subject: [PATCH] Disable nested parallelism when compiled with clang, Issue
 2731

---
 rtengine/FTblockDN.cc         |  78 +++++++++++++------------
 rtengine/cplx_wavelet_level.h |  26 ++++-----
 rtengine/ipwavelet.cc         | 106 +++++++++++++++++++---------------
 rtengine/opthelper.h          |   3 +
 4 files changed, 117 insertions(+), 96 deletions(-)

diff --git a/rtengine/FTblockDN.cc b/rtengine/FTblockDN.cc
index e8c94b10f..2343ff865 100644
--- a/rtengine/FTblockDN.cc
+++ b/rtengine/FTblockDN.cc
@@ -645,7 +645,8 @@ do {
 	// Calculate number of tiles. If less than omp_get_max_threads(), then limit num_threads to number of tiles
 	int numthreads = MIN(numtiles,omp_get_max_threads());
 	if(options.rgbDenoiseThreadLimit > 0)
-		numthreads = MIN(numthreads,options.rgbDenoiseThreadLimit);
+		numthreads = MIN(numthreads,options.rgbDenoiseThreadLimit);
+#ifdef _RT_NESTED_OPENMP
 	denoiseNestedLevels = omp_get_max_threads() / numthreads;
 	bool oldNested = omp_get_nested();
 	if(denoiseNestedLevels < 2)
@@ -654,7 +655,8 @@ do {
 		omp_set_nested(true);
 	if(options.rgbDenoiseThreadLimit > 0)
 		while(denoiseNestedLevels*numthreads > options.rgbDenoiseThreadLimit)
-			denoiseNestedLevels--;
+			denoiseNestedLevels--;
+#endif
 	if(settings->verbose)
 		printf("RGB_denoise uses %d main thread(s) and up to %d nested thread(s) for each main thread\n",numthreads,denoiseNestedLevels);
 #endif
@@ -736,8 +738,8 @@ do {
 
 				if(!denoiseMethodRgb){//lab mode
 						//modification Jacques feb 2013 and july 2014
-#ifdef _OPENMP
-#pragma omp parallel for num_threads(denoiseNestedLevels) if(denoiseNestedLevels>1)
+#ifdef _RT_NESTED_OPENMP
+#pragma omp parallel for num_threads(denoiseNestedLevels) if(denoiseNestedLevels>1)
 #endif
 					for (int i=tiletop; i<tilebottom; i++) {
 						int i1 = i - tiletop;
@@ -781,8 +783,8 @@ do {
 						}
 					}
 				} else {//RGB mode
-#ifdef _OPENMP
-#pragma omp parallel for num_threads(denoiseNestedLevels) if(denoiseNestedLevels>1)
+#ifdef _RT_NESTED_OPENMP
+#pragma omp parallel for num_threads(denoiseNestedLevels) if(denoiseNestedLevels>1)
 #endif
 					for (int i=tiletop; i<tilebottom; i++) {
 						int i1 = i - tiletop;
@@ -814,8 +816,8 @@ do {
 					}
 				}
 			} else {//image is not raw; use Lab parametrization
-#ifdef _OPENMP
-#pragma omp parallel for num_threads(denoiseNestedLevels) if(denoiseNestedLevels>1)
+#ifdef _RT_NESTED_OPENMP
+#pragma omp parallel for num_threads(denoiseNestedLevels) if(denoiseNestedLevels>1)
 #endif
 				for (int i=tiletop; i<tilebottom; i++) {
 					int i1 = i - tiletop;
@@ -928,7 +930,7 @@ do {
 				if(!memoryAllocationFailed) {
 					// precalculate madL, because it's used in adecomp and bdecomp
 					int maxlvl = Ldecomp->maxlevel();
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel for schedule(dynamic) collapse(2) num_threads(denoiseNestedLevels) if(denoiseNestedLevels>1)
 #endif
 					for (int lvl=0; lvl<maxlvl; lvl++) {
@@ -1023,7 +1025,7 @@ do {
 							if(!memoryAllocationFailed) {
 								// copy labdn->L to Lin before it gets modified by reconstruction
 								Lin = new array2D<float>(width,height);
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel for num_threads(denoiseNestedLevels) if(denoiseNestedLevels>1)
 #endif
 								for(int i=0;i<height;i++)
@@ -1134,14 +1136,14 @@ do {
 						fLbloxArray[i] = (float*) fftwf_malloc(max_numblox_W*TS*TS*sizeof(float));
 					}
 
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 				int masterThread = omp_get_thread_num();
 #endif
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel num_threads(denoiseNestedLevels) if(denoiseNestedLevels>1)
 #endif
 {
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 				int subThread = masterThread * denoiseNestedLevels + omp_get_thread_num();
 #else
 				int subThread = 0;
@@ -1151,7 +1153,7 @@ do {
 				float *fLblox = fLbloxArray[subThread];
 				float pBuf[width + TS + 2*blkrad*offset] ALIGNED16;
 				float nbrwt[TS*TS] ALIGNED64;
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp for
 #endif
 				for (int vblk=0; vblk<numblox_H; vblk++) {
@@ -1239,7 +1241,7 @@ do {
 	}
 				//%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
 
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel for num_threads(denoiseNestedLevels) if(denoiseNestedLevels>1)
 #endif
 				for (int i=0; i<height; i++) {
@@ -1283,8 +1285,8 @@ do {
 				realred /= 100.f;
 				realblue /= 100.f;
 
-#ifdef _OPENMP
-#pragma omp parallel for schedule(dynamic,16) num_threads(denoiseNestedLevels) if(denoiseNestedLevels>1)
+#ifdef _RT_NESTED_OPENMP
+#pragma omp parallel for schedule(dynamic,16) num_threads(denoiseNestedLevels)
 #endif
 				for (int i=tiletop; i<tilebottom; i++){
 					int i1 = i-tiletop;
@@ -1329,8 +1331,8 @@ do {
 					}
 				}
 				} else {//RGB mode
-#ifdef _OPENMP
-#pragma omp parallel for num_threads(denoiseNestedLevels) if(denoiseNestedLevels>1)
+#ifdef _RT_NESTED_OPENMP
+#pragma omp parallel for num_threads(denoiseNestedLevels)
 #endif
 				for (int i=tiletop; i<tilebottom; i++){
 					int i1 = i-tiletop;
@@ -1365,8 +1367,8 @@ do {
 
 				}
 			} else {
-#ifdef _OPENMP
-#pragma omp parallel for num_threads(denoiseNestedLevels) if(denoiseNestedLevels>1)
+#ifdef _RT_NESTED_OPENMP
+#pragma omp parallel for num_threads(denoiseNestedLevels)
 #endif
 				for (int i=tiletop; i<tilebottom; i++){
 					int i1 = i-tiletop;
@@ -1427,7 +1429,7 @@ do {
 		}
 	}
 	
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 omp_set_nested(oldNested);
 #endif
 	//copy denoised image to output
@@ -1854,7 +1856,7 @@ SSEFUNCTION	bool ImProcFunctions::WaveletDenoiseAll_BiShrinkL(wavelet_decomposit
 			maxHL = WaveletCoeffs_L.level_H(lvl);
 	}
 	bool memoryAllocationFailed = false;
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel num_threads(denoiseNestedLevels) if(denoiseNestedLevels>1)
 #endif
 {
@@ -1868,7 +1870,7 @@ SSEFUNCTION	bool ImProcFunctions::WaveletDenoiseAll_BiShrinkL(wavelet_decomposit
 	
 	if(!memoryAllocationFailed) {
 	
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp for schedule(dynamic) collapse(2)
 #endif
 		for (int lvl=maxlvl-1; lvl>=0; lvl--) {//for levels less than max, use level diff to make edge mask
@@ -1968,7 +1970,7 @@ SSEFUNCTION	bool ImProcFunctions::WaveletDenoiseAll_BiShrinkAB(wavelet_decomposi
 				maxHL = WaveletCoeffs_L.level_H(lvl);
 		}
 		bool memoryAllocationFailed = false;
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel num_threads(denoiseNestedLevels) if(denoiseNestedLevels>1)
 #endif
 {
@@ -1983,7 +1985,7 @@ SSEFUNCTION	bool ImProcFunctions::WaveletDenoiseAll_BiShrinkAB(wavelet_decomposi
 		if(!memoryAllocationFailed) {
 		
 
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp for schedule(dynamic) collapse(2)
 #endif
 		for (int lvl=0; lvl<maxlvl; lvl++) {
@@ -2001,7 +2003,7 @@ SSEFUNCTION	bool ImProcFunctions::WaveletDenoiseAll_BiShrinkAB(wavelet_decomposi
 			}	
 		}
 		
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp for schedule(dynamic) collapse(2)
 #endif
 		for (int lvl=maxlvl-1; lvl>=0; lvl--) {//for levels less than max, use level diff to make edge mask
@@ -2086,7 +2088,7 @@ SSEFUNCTION	bool ImProcFunctions::WaveletDenoiseAll_BiShrinkAB(wavelet_decomposi
 				maxHL = WaveletCoeffs_L.level_H(lvl);
 		}
 		bool memoryAllocationFailed = false;
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel num_threads(denoiseNestedLevels) if(denoiseNestedLevels>1)
 #endif
 {
@@ -2099,7 +2101,7 @@ SSEFUNCTION	bool ImProcFunctions::WaveletDenoiseAll_BiShrinkAB(wavelet_decomposi
 		}
 		
 		if(!memoryAllocationFailed) {
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp for schedule(dynamic) collapse(2)
 #endif
 			for (int lvl=0; lvl<maxlvl; lvl++) {
@@ -2130,7 +2132,7 @@ SSEFUNCTION	bool ImProcFunctions::WaveletDenoiseAll_BiShrinkAB(wavelet_decomposi
 				maxHL = WaveletCoeffs_L.level_H(lvl);
 		}
 		bool memoryAllocationFailed = false;
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel num_threads(denoiseNestedLevels) if(denoiseNestedLevels>1)
 #endif
 {
@@ -2143,7 +2145,7 @@ SSEFUNCTION	bool ImProcFunctions::WaveletDenoiseAll_BiShrinkAB(wavelet_decomposi
 		}
 		
 		if(!memoryAllocationFailed) {
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp for schedule(dynamic) collapse(2)
 #endif
 			for (int lvl=0; lvl<maxlvl; lvl++) {
@@ -2589,7 +2591,7 @@ SSEFUNCTION void ImProcFunctions::RGB_denoise_info(Imagefloat * src, Imagefloat
 	for (int i=0; i<hei; i++)
 		bcalc[i] = new float[wid];
 
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel for if(multiThread)
 #endif
 	for(int ii=0;ii<hei;ii++){
@@ -2707,7 +2709,7 @@ SSEFUNCTION void ImProcFunctions::RGB_denoise_info(Imagefloat * src, Imagefloat
 			//fill tile from image; convert RGB to "luma/chroma"
 
 			if (isRAW) {//image is raw; use channel differences for chroma channels
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel for if(multiThread)
 #endif
 					for (int i=tiletop; i<tilebottom; i+=2) {
@@ -2747,7 +2749,7 @@ SSEFUNCTION void ImProcFunctions::RGB_denoise_info(Imagefloat * src, Imagefloat
 						}
 #endif
 					}
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel for if(multiThread)
 #endif
 					for (int i=tiletop; i<tilebottom; i+=2) {
@@ -2762,7 +2764,7 @@ SSEFUNCTION void ImProcFunctions::RGB_denoise_info(Imagefloat * src, Imagefloat
 					}
 				if (!denoiseMethodRgb){//lab mode, modification Jacques feb 2013 and july 2014
 
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel for if(multiThread)
 #endif
 				for (int i=tiletop; i<tilebottom; i++) {
@@ -2883,17 +2885,17 @@ SSEFUNCTION void ImProcFunctions::RGB_denoise_info(Imagefloat * src, Imagefloat
 				schoice=2;
 
 			const int levwav=5;
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel sections if(multiThread)
 #endif
 {
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp section
 #endif
 {
 			adecomp = new wavelet_decomposition (labdn->data+datalen, labdn->W, labdn->H, levwav, 1 );
 }
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp section
 #endif
 {
diff --git a/rtengine/cplx_wavelet_level.h b/rtengine/cplx_wavelet_level.h
index 3eef39397..1c1efde27 100644
--- a/rtengine/cplx_wavelet_level.h
+++ b/rtengine/cplx_wavelet_level.h
@@ -215,7 +215,7 @@ namespace rtengine {
 		 * Applies a Haar filter 
 		 *
 		 */
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel for num_threads(numThreads) if(numThreads>1)
 #endif
 		for (int k=0; k<height; k++) {
@@ -234,18 +234,18 @@ namespace rtengine {
 		 * Applies a Haar filter 
 		 *
 		 */
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel num_threads(numThreads) if(numThreads>1)
 #endif
 {
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp for nowait
 #endif
 		for(int i = 0; i < skip; i++) {
 			for(int j=0;j<width;j++)
 				dst[width*i+j] = (srcLo[i*width+j] + srcHi[i*width+j]);			
 		}
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp for
 #endif
 		for(int i = skip; i < height; i++) {
@@ -394,7 +394,7 @@ namespace rtengine {
 
 		// calculate coefficients
 		int shift = skip*(taps-offset-1);//align filter with data
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel for num_threads(numThreads) if(numThreads>1)
 #endif
 		for (int k=0; k<height; k++) {
@@ -450,7 +450,7 @@ namespace rtengine {
 		__m128 fourv = _mm_set1_ps(4.f);
 		__m128 srcFactorv = _mm_set1_ps(srcFactor);
 		__m128 dstFactorv = _mm_set1_ps(blend);
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel for num_threads(numThreads) if(numThreads>1)
 #endif
 		for(int i = 0; i < dstheight; i++) {
@@ -509,7 +509,7 @@ namespace rtengine {
 		// calculate coefficients
 		int shift=skip*(taps-offset-1);//align filter with data
 
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel for num_threads(numThreads) if(numThreads>1)
 #endif
 		for(int i = 0; i < dstheight; i++) {
@@ -550,14 +550,14 @@ namespace rtengine {
 				}
 			}
 		}
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel num_threads(numThreads) if(numThreads>1)
 #endif
 {
 		T tmpLo[m_w] ALIGNED64;
 		T tmpHi[m_w] ALIGNED64;
 		if(subsamp_out) {
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp for
 #endif
 			for(int row=0;row<m_h;row+=2) {
@@ -566,7 +566,7 @@ namespace rtengine {
 				AnalysisFilterSubsampHorizontal (tmpHi, wavcoeffs[2], wavcoeffs[3], filterH, filterH+taps, taps, offset, m_w, m_w2, row/2);
 			}
 		} else {
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp for
 #endif
 			for(int row=0;row<m_h;row++) {
@@ -580,7 +580,7 @@ namespace rtengine {
 #else
 	template<typename T> template<typename E> void wavelet_level<T>::decompose_level(E *src, E *dst, float *filterV, float *filterH, int taps, int offset) { 
 
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel num_threads(numThreads) if(numThreads>1)
 #endif
 {
@@ -588,7 +588,7 @@ namespace rtengine {
 		T tmpHi[m_w] ALIGNED64;
 		/* filter along rows and columns */
 		if(subsamp_out) {
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp for
 #endif
 			for(int row=0;row<m_h;row+=2) {
@@ -597,7 +597,7 @@ namespace rtengine {
 				AnalysisFilterSubsampHorizontal (tmpHi, wavcoeffs[2], wavcoeffs[3], filterH, filterH+taps, taps, offset, m_w, m_w2, row/2);
 			}
 		} else {
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp for
 #endif
 			for(int row=0;row<m_h;row++) {
diff --git a/rtengine/ipwavelet.cc b/rtengine/ipwavelet.cc
index d52c9ea3b..55b6d8b09 100644
--- a/rtengine/ipwavelet.cc
+++ b/rtengine/ipwavelet.cc
@@ -438,7 +438,8 @@ SSEFUNCTION void ImProcFunctions::ip_wavelet(LabImage * lab, LabImage * dst, int
 	
 		numthreads = MIN(numtiles,omp_get_max_threads());
 		if(maxnumberofthreadsforwavelet > 0)
-			numthreads = MIN(numthreads,maxnumberofthreadsforwavelet);
+			numthreads = MIN(numthreads,maxnumberofthreadsforwavelet);
+#ifdef _RT_NESTED_OPENMP
 		wavNestedLevels = omp_get_max_threads() / numthreads;
 		bool oldNested = omp_get_nested();
 		if(wavNestedLevels < 2)
@@ -447,7 +448,8 @@ SSEFUNCTION void ImProcFunctions::ip_wavelet(LabImage * lab, LabImage * dst, int
 			omp_set_nested(true);
 		if(maxnumberofthreadsforwavelet > 0)
 			while(wavNestedLevels*numthreads > maxnumberofthreadsforwavelet)
-				wavNestedLevels--;
+				wavNestedLevels--;
+#endif
 		if(settings->verbose)
 			printf("Ip Wavelet uses %d main thread(s) and up to %d nested thread(s) for each main thread\n",numthreads,wavNestedLevels);
 
@@ -493,16 +495,16 @@ SSEFUNCTION void ImProcFunctions::ip_wavelet(LabImage * lab, LabImage * dst, int
 						for (int i=0; i<tileheight; i++)
 							Lold[i] = LoldBuffer + i*tilewidth;
 					}
-						
+
 				} else {
 					labco = new LabImage(width,height);
 					Lold = lab->L;
 				}
-				
-#ifdef _OPENMP
+
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel for num_threads(wavNestedLevels) if(wavNestedLevels>1)
 #endif
-				
+
 					for (int i=tiletop; i<tilebottom; i++) {
 						int i1 = i - tiletop;
 						int j;
@@ -554,7 +556,7 @@ SSEFUNCTION void ImProcFunctions::ip_wavelet(LabImage * lab, LabImage * dst, int
 						}
 					}
 	
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel for num_threads(wavNestedLevels) if(wavNestedLevels>1)
 #endif
 					for (int i=1; i<hei-1; i++) {
@@ -618,7 +620,7 @@ SSEFUNCTION void ImProcFunctions::ip_wavelet(LabImage * lab, LabImage * dst, int
 						
 					float madL[8][3];
 					bool memoryAllocationFailed = false;
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel for schedule(dynamic) collapse(2) num_threads(wavNestedLevels) if(wavNestedLevels>1)
 #endif				
 					for (int lvl=0; lvl<3; lvl++) {
@@ -777,9 +779,9 @@ SSEFUNCTION void ImProcFunctions::ip_wavelet(LabImage * lab, LabImage * dst, int
 					}
 					bool highlight = params->toneCurve.hrenabled;
 
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel for schedule(dynamic,16) num_threads(wavNestedLevels) if(wavNestedLevels>1)
-#endif				
+#endif
 					for (int i=tiletop; i<tilebottom; i++){
 							int i1 = i-tiletop;
 							float L,a,b;
@@ -928,7 +930,7 @@ SSEFUNCTION void ImProcFunctions::ip_wavelet(LabImage * lab, LabImage * dst, int
 		delete [] sigmaN;
 		
 }
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 omp_set_nested(oldNested);
 #endif	
 		if(numtiles > 1) {
@@ -936,10 +938,10 @@ omp_set_nested(oldNested);
 			delete dsttmp;
 		}
 		
-//	if (settings->verbose) {
+	if (settings->verbose) {
 		t2e.set();
 		printf("Wavelet performed in %d usec:\n", t2e.etime(t1e));
-//	}
+	}
 		
 }//end o
 
@@ -960,12 +962,12 @@ omp_set_nested(oldNested);
 		float thres = 5.f;//different fom zero to take into account only data large enough
 		max=0.f;
 		min=0.f;
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel num_threads(wavNestedLevels) if(wavNestedLevels>1)
 #endif
 {
 		float lmax = 0.f, lmin = 0.f;
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp for reduction(+:averaP,averaN,countP,countN) nowait
 #endif
 		for(int i=0;i<datalen;i++) {
@@ -982,7 +984,7 @@ omp_set_nested(oldNested);
 				countN++;
 			}
 		}
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp critical
 #endif
 {
@@ -1002,7 +1004,7 @@ omp_set_nested(oldNested);
 		float variP = 0.f, variN = 0.f;
 		float thres = 5.f;//different fom zero to take into account only data large enough
 
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel for reduction(+:variP,variN,countP,countN) num_threads(wavNestedLevels) if(wavNestedLevels>1)
 #endif
 		for(int i=0;i<datalen;i++) {
@@ -1087,7 +1089,7 @@ float *ImProcFunctions::ContrastDR(float *Source, int skip, struct cont_params c
 	int n=W_L*H_L;
 	if(Contrast == NULL) Contrast = new float[n];
 	memcpy(Contrast, Source, n*sizeof(float));
-#ifdef _OPENMP	
+#ifdef _RT_NESTED_OPENMP	
 #pragma omp parallel for
 #endif	
 				for (int i=0; i<W_L*H_L; i++) {//contrast
@@ -1122,12 +1124,12 @@ SSEFUNCTION float *ImProcFunctions::CompressDR(float *Source, int skip, struct c
 	int n=W_L*H_L;
 	
 #ifdef __SSE2__
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel
 #endif
 {
 	__m128 epsv = _mm_set1_ps( eps );
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp for
 #endif
 	for(int ii = 0; ii < n-3; ii+=4)
@@ -1137,7 +1139,7 @@ SSEFUNCTION float *ImProcFunctions::CompressDR(float *Source, int skip, struct c
 		Source[ii] = xlogf(Source[ii] + eps);
 
 #else
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel for
 #endif
 	for(int ii = 0; ii < n; ii++)
@@ -1168,7 +1170,7 @@ SSEFUNCTION float *ImProcFunctions::CompressDR(float *Source, int skip, struct c
 	
 	
 #ifdef __SSE2__
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel
 #endif
 {
@@ -1176,7 +1178,7 @@ SSEFUNCTION float *ImProcFunctions::CompressDR(float *Source, int skip, struct c
 	__m128 epsv = _mm_set1_ps( eps );
 	__m128 DetailBoostv = _mm_set1_ps( DetailBoost );
 	__m128 tempv = _mm_set1_ps( temp );
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp for
 #endif
 	for(int i = 0; i < n-3; i+=4){
@@ -1195,7 +1197,7 @@ SSEFUNCTION float *ImProcFunctions::CompressDR(float *Source, int skip, struct c
 	}
 	
 #else
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel for
 #endif
 	for(int i = 0; i < n; i++){
@@ -1218,7 +1220,9 @@ void ImProcFunctions::ContrastResid(float * WavCoeffs_L0,  unsigned int Iterates
 		cp.TMmeth=2;//default after testing
 		if(cp.TMmeth ==1) {min0 = 0.0f;max0=32768.f;}
 		else if (cp.TMmeth ==2) {min0 = 0.0f;}
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel for
+#endif
 			for(int i = 0; i < W_L*H_L; i++)
 				{ WavCoeffs_L0[i]= (WavCoeffs_L0[i] - min0)/max0;
 					WavCoeffs_L0[i]*=gamm;
@@ -1232,9 +1236,9 @@ void ImProcFunctions::ContrastResid(float * WavCoeffs_L0,  unsigned int Iterates
 			CompressDR(WavCoeffs_L0, skip, cp, W_L, H_L, Compression,DetailBoost,max0, min0, ave, ah, bh, al, bl, factorx, WavCoeffs_L0);
 
 
-			#ifdef _OPENMP
-			#pragma omp parallel for            // removed schedule(dynamic,10)
-			#endif
+#ifdef _RT_NESTED_OPENMP
+#pragma omp parallel for            // removed schedule(dynamic,10)
+#endif
 			for(int ii = 0; ii < W_L*H_L; ii++)
 				WavCoeffs_L0[ii] = WavCoeffs_L0[ii]*max0*(1.f/gamm) + min0;	
 	}	
@@ -1255,7 +1259,9 @@ void ImProcFunctions::ContrastResid(float * WavCoeffs_L0,  unsigned int Iterates
 		if(cp.TMmeth ==1) {min0 = 0.0f;max0=32768.f;}
 		else if (cp.TMmeth ==2) {min0 = 0.0f;}
 		//	max0=32768.f;
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel for
+#endif
 			for(int i = 0; i < W_L*H_L; i++)
 				{ WavCoeffs_L0[i]= (WavCoeffs_L0[i] - min0)/max0;
 					WavCoeffs_L0[i]*=gamm;
@@ -1272,9 +1278,9 @@ void ImProcFunctions::ContrastResid(float * WavCoeffs_L0,  unsigned int Iterates
 		    epd.CompressDynamicRange(WavCoeffs_L0, sca/float(skip), edgest, Compression, DetailBoost, Iterates, rew, WavCoeffs_L0);
 
 	//Restore past range, also desaturate a bit per Mantiuk's Color correction for tone mapping.
-			#ifdef _OPENMP
-			#pragma omp parallel for            // removed schedule(dynamic,10)
-			#endif
+#ifdef _RT_NESTED_OPENMP
+#pragma omp parallel for            // removed schedule(dynamic,10)
+#endif
 			for(int ii = 0; ii < W_L*H_L; ii++)
 				WavCoeffs_L0[ii] = WavCoeffs_L0[ii]*max0*(1.f/gamm) + min0;	
 	}	
@@ -1288,7 +1294,7 @@ void ImProcFunctions::WaveletcontAllLfinal(LabImage * labco, float ** varhue, fl
 		float * WavCoeffs_L0 = WaveletCoeffs_L.coeff0;
 		
 		
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp for schedule(dynamic) collapse(2)
 #endif		
 		for (int dir=1; dir<4; dir++) {
@@ -1326,24 +1332,31 @@ void ImProcFunctions::WaveletcontAllLfinal(LabImage * labco, float ** varhue, fl
 		float min0 = FLT_MAX;
 		
 		if(contrast != 0.f || cp.tonemap) { // contrast = 0.f means that all will be multiplied by 1.f, so we can skip this step
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel for reduction(+:avedbl) num_threads(wavNestedLevels) if(wavNestedLevels>1)
 #endif
 			for (int i=0; i<W_L*H_L; i++) {
 				avedbl += WavCoeffs_L0[i];
 			}
 		
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel num_threads(wavNestedLevels) if(wavNestedLevels>1)
+#endif
 {
 	float lminL = FLT_MAX;
 	float lmaxL = 0.f;
+
+#ifdef _RT_NESTED_OPENMP
 #pragma omp for
+#endif
 	for(int i = 0; i < W_L*H_L; i++) {
 		if(WavCoeffs_L0[i] < lminL) lminL = WavCoeffs_L0[i];
 		if(WavCoeffs_L0[i] > lmaxL) lmaxL = WavCoeffs_L0[i];
 		
 	}
+#ifdef _RT_NESTED_OPENMP
 #pragma omp critical
+#endif
 	{ 	if(lminL < min0) min0 = lminL;
 		if(lmaxL > max0) max0 = lmaxL;
 	}
@@ -1389,13 +1402,13 @@ if(cp.tonemap && cp.contmet==2) {
 			for (int i=0; i<W_L*H_L; i++) 
 				koeLi[j][i]=0.f;
 		}
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel num_threads(wavNestedLevels) if(wavNestedLevels>1)
 #endif
 {
 		if(contrast != 0.f) { // contrast = 0.f means that all will be multiplied by 1.f, so we can skip this step
   {
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp for
 #endif						
 			for (int i=0; i<W_L*H_L; i++) {//contrast
@@ -1421,14 +1434,17 @@ if(cp.tonemap && cp.contmet==2) {
 if(cp.tonemap && cp.contmet==1) {
 	float maxp=max0*256.f;
 	float minp=min0*256.f;
-	#pragma omp single
+#ifdef _RT_NESTED_OPENMP
+#pragma omp single
+#endif
 		ContrastResid(WavCoeffs_L0, 5, skip, cp, W_L, H_L, maxp, minp, ave, ah, bh, al, bl, factorx );
 }
+#ifdef _RT_NESTED_OPENMP
 #pragma omp barrier
-		
+#endif		
 		
 		if(cp.conres != 0.f || cp.conresH != 0.f) { // cp.conres = 0.f and cp.comresH = 0.f means that all will be multiplied by 1.f, so we can skip this step
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp for nowait
 #endif				
 			for (int i=0; i<W_L*H_L; i++) {
@@ -1481,7 +1497,7 @@ if(cp.detectedge && lipschitz==true) {	//enabled Lipschitz control...more memory
 		for (int i=0; i<H_L; i++){
 			tmC[i] = &tmCBuffer[i*W_L];
 		}
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp for schedule(dynamic) collapse(2)
 #endif		
 		for (int lvl=0; lvl<3; lvl++) {
@@ -1499,7 +1515,7 @@ if(cp.detectedge && lipschitz==true) {	//enabled Lipschitz control...more memory
 		float aamp=1.f+cp.eddetthrHi/100.f;
 
 		for (int lvl=0; lvl<3; lvl++) {
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp for schedule(dynamic,16)
 #endif		
 		for (int i=1; i<H_L-1; i++) {
@@ -1561,7 +1577,7 @@ if(cp.detectedge && lipschitz==true) {	//enabled Lipschitz control...more memory
 		// end 
 }	
 	
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp for schedule(dynamic) collapse(2)
 #endif		
 		for (int dir=1; dir<4; dir++) {
@@ -1593,7 +1609,7 @@ if(cp.detectedge && lipschitz==true) {	//enabled Lipschitz control...more memory
                
                         float * WavCoeffs_a0 = WaveletCoeffs_a.coeff0;
                         float * WavCoeffs_b0 = WaveletCoeffs_b.coeff0;
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel num_threads(wavNestedLevels) if(wavNestedLevels>1)
 #endif
 {
@@ -1601,7 +1617,7 @@ if(cp.detectedge && lipschitz==true) {	//enabled Lipschitz control...more memory
                         float huebuffer[W_L] ALIGNED64;
                         float chrbuffer[W_L] ALIGNED64;
 #endif // __SSE2__
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp for schedule(dynamic,16)
 #endif
                         for (int i=0; i<H_L; i++) {
@@ -1655,13 +1671,13 @@ if(cp.detectedge && lipschitz==true) {	//enabled Lipschitz control...more memory
 	
 		float * WavCoeffs_ab0 = WaveletCoeffs_ab.coeff0;
 		
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp parallel num_threads(wavNestedLevels) if(wavNestedLevels>1)
 #endif
 {
 		if(cp.chrores != 0.f) { // cp.chrores == 0.f means all will be multiplied by 1.f, so we can skip the processing of residual
 
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp for nowait
 #endif				
 			for (int i=0; i<W_L*H_L; i++) {
@@ -1707,7 +1723,7 @@ if(cp.detectedge && lipschitz==true) {	//enabled Lipschitz control...more memory
 			}
 		}
 
-#ifdef _OPENMP
+#ifdef _RT_NESTED_OPENMP
 #pragma omp for schedule(dynamic) collapse(2)
 #endif		
 		for (int dir=1; dir<4; dir++) {
diff --git a/rtengine/opthelper.h b/rtengine/opthelper.h
index cdb691ddd..c6d8a63b0 100644
--- a/rtengine/opthelper.h
+++ b/rtengine/opthelper.h
@@ -69,4 +69,7 @@
 		#define ALIGNED64
 		#define ALIGNED16
 	#endif
+	#ifndef __clang__
+	    #define _RT_NESTED_OPENMP _OPENMP
+	#endif
 #endif