tor-grad in parallel

Makefile

@@ -201,14 +201,15 @@ odock: check-env-all stringify $(SRC)
 # 7cpa: for testing gradients of torsion genes (15 torsions) 
 # 3tmn: for testing gradients of torsion genes (1 torsion)
 
-PDB     := 3ce3
-NRUN    := 100
-POPSIZE := 150
-TESTNAME:= test
-TESTLS  := sd
+PDB      := 3ce3
+NRUN     := 100
+POPSIZE  := 150
+TESTNAME := test
+TESTLS   := sw
+NUM_LSIT := 10
 
 test: odock
-	$(BIN_DIR)/$(TARGET) -ffile ./input/$(PDB)/derived/$(PDB)_protein.maps.fld -lfile ./input/$(PDB)/derived/$(PDB)_ligand.pdbqt -nrun $(NRUN) -psize $(POPSIZE) -resnam $(TESTNAME) -gfpop 1 -lsmet $(TESTLS)
+	$(BIN_DIR)/$(TARGET) -ffile ./input/$(PDB)/derived/$(PDB)_protein.maps.fld -lfile ./input/$(PDB)/derived/$(PDB)_ligand.pdbqt -nrun $(NRUN) -psize $(POPSIZE) -resnam $(TESTNAME) -gfpop 1 -lsmet $(TESTLS) -lsit $(NUM_LSIT)
 
 ASTEX_PDB := 2bsm
 ASTEX_NRUN:= 10

device/calcenergy.cl

@@ -172,7 +172,14 @@ void gpu_calc_energy(
 				atom_to_rotate[2] -= rotation_movingvec[2];
 
 				// Transforming torsion angles into quaternions
+				//----------------------------------
+				// fastergrad
+				//----------------------------------
+				/*
 				rotation_angle  = native_divide(rotation_angle, 2.0f);
+				*/
+				rotation_angle  = rotation_angle * 0.5f;
+				//----------------------------------
 				float sin_angle = native_sin(rotation_angle);
 				quatrot_left_q  = native_cos(rotation_angle);
 				quatrot_left_x  = sin_angle*rotation_unitvec[0];

device/calcgradient.cl

@@ -218,7 +218,15 @@ void gpu_calc_gradient(
 				atom_to_rotate[2] -= rotation_movingvec[2];
 
 				// Transforming torsion angles into quaternions
+				//----------------------------------
+				// fastergrad
+				//----------------------------------
+				/*
 				rotation_angle  = native_divide(rotation_angle, 2.0f);
+				*/
+				rotation_angle  = rotation_angle * 0.5f;
+				//----------------------------------
+
 				float sin_angle = native_sin(rotation_angle);
 				quatrot_left_q  = native_cos(rotation_angle);
 				quatrot_left_x  = sin_angle*rotation_unitvec[0];
@@ -955,11 +963,22 @@ void gpu_calc_gradient(
 	// ------------------------------------------
 	// Obtaining torsion-related gradients
 	// ------------------------------------------
+
+	//----------------------------------
+	// fastergrad
+	//----------------------------------
+/*
 	if (get_local_id(0) == 2) {
 
 		for (uint rotbond_id = 0;
 			  rotbond_id < dockpars_num_of_genes-6;
 			  rotbond_id ++) {
+*/
+
+		for (uint rotbond_id = get_local_id(0);
+			  rotbond_id < dockpars_num_of_genes-6;
+			  rotbond_id +=NUM_OF_THREADS_PER_BLOCK) {
+	//----------------------------------
 
 			// Querying ids of atoms belonging to the rotatable bond in question
 			int atom1_id = rotbonds_const[2*rotbond_id];
@@ -1052,6 +1071,12 @@ void gpu_calc_gradient(
 			#endif
 			
 		} // End of iterations over rotatable bonds
-	}
 
+	//----------------------------------
+	// fastergrad
+	//----------------------------------
+/*
+	}
+*/
+	//----------------------------------
 }

device/kernel_gradient.cl

@@ -249,6 +249,24 @@ gradient_minimizer(
 	}
 	barrier(CLK_LOCAL_MEM_FENCE);
 
+
+	//----------------------------------
+	// fastergrad
+	//----------------------------------
+ 		
+	// Calculating maximum possible stepsize (alpha)
+	__local float max_trans_gene, max_rota_gene, max_tors_gene;
+	__local float max_trans_stepsize, max_rota_stepsize, max_tors_stepsize;
+	__local float max_stepsize;
+	//----------------------------------
+
+	//----------------------------------
+	// fastergrad
+	//----------------------------------
+
+	__local float torsions_genotype[ACTUAL_GENOTYPE_LENGTH];
+	//----------------------------------
+
 	// The termination criteria is based on 
 	// a maximum number of iterations, and
 	// the minimum step size allowed for single-floating point numbers 
@@ -297,11 +315,16 @@ gradient_minimizer(
 		#endif
 
 
-				
+		//----------------------------------
+		// fastergrad
+		//----------------------------------
+/* 		
 		// Calculating maximum possible stepsize (alpha)
 		__local float max_trans_gene, max_rota_gene, max_tors_gene;
 		__local float max_trans_stepsize, max_rota_stepsize, max_tors_stepsize;
 		__local float max_stepsize;
+*/
+		//----------------------------------
 
 		if (get_local_id(0) == 0) {
 			// Finding maximum of the absolute value 
@@ -324,7 +347,15 @@ gradient_minimizer(
 		}
 
 		// Copying torsions genes
+
+		//----------------------------------
+		// fastergrad
+		//----------------------------------
+/*
 		__local float torsions_genotype[ACTUAL_GENOTYPE_LENGTH];
+*/
+		//----------------------------------
+
 		for(uint i = get_local_id(0); 
 			 i < dockpars_num_of_genes-6; 
 			 i+= NUM_OF_THREADS_PER_BLOCK) {
@@ -441,6 +472,18 @@ gradient_minimizer(
 
 		barrier(CLK_LOCAL_MEM_FENCE);
 
+		//----------------------------------
+		// fastergrad
+		//----------------------------------
+/*		
+		if ((get_group_id(0) == 0) && (get_local_id(0) == 0)) {
+			for(uint i = 0; i < dockpars_num_of_genes; i++) {
+				printf("gradient[%u]=%f \n", i, gradient[i]);
+			}
+		}
+*/
+		//----------------------------------
+
 		
 		for(uint i = get_local_id(0); i < dockpars_num_of_genes; i+= NUM_OF_THREADS_PER_BLOCK) {
 

host/src/performdocking.cpp

@@ -423,7 +423,18 @@ filled with clock() */
 	// Setup here (temporarily?) the gradient and associated parameters.
 	// This should be ultimately configurable by the user as program exec. flags.
 	Gradientparameters gradientpars;
+
+	//----------------------------------
+	// fastergrad
+	//----------------------------------
+/*
 	gradientpars.max_num_of_iters = 300;
+*/
+/*
+	gradientpars.max_num_of_iters = 10;
+*/
+	gradientpars.max_num_of_iters =	(unsigned int) mypars->max_num_of_iters;
+	//----------------------------------
 
 /*
 	gradientpars.max_num_of_consec_fails = (unsigned int)(0.05 * gradientpars.max_num_of_iters);