KernelGen splitCK flux - fix sign (6517ec4f) · Commits · ExaHyPE / ExaHyPE-Engine

KernelGenerator/kernelgenerator/templates/aderdg/fusedSPTVI_linear_split_ck_cpp.template

+8 −5

Original line number	Diff line number	Diff line
		@@ -97,9 +97,11 @@ int {{codeNamespace}}::fusedSpaceTimePredictorVolumeIntegral{{nameSuffix}}(
		// local tmp array
		double tmpArray[{{nVarPad*nDof}}] __attribute__((aligned(ALIGNMENT))); //used by flux and ncp
		double dudxT_by_dx[{{nDof*nDofPad}}] __attribute__((aligned(ALIGNMENT)));
		double negativeDudxT_by_dx[{{nDof*nDofPad}}] __attribute__((aligned(ALIGNMENT)));
		{% if useLibxsmm %}
		#if defined(USE_IPO) && ! defined(UNSAFE_IPO)
		double doNotOptimizeAway = dudxT_by_dx[0]; // used to prevent the compiler from optimizing dudxT_by_dx away
		double doNotOptimizeAway1 = dudxT_by_dx[0]; // used to prevent the compiler from optimizing dudxT_by_dx away
		double doNotOptimizeAway2 = negativeDudxT_by_dx[0]; // used to prevent the compiler from optimizing dudxT_by_dx away
		#endif
		{% endif %}

		@@ -158,9 +160,10 @@ int {{codeNamespace}}::fusedSpaceTimePredictorVolumeIntegral{{nameSuffix}}(


		// 0. precompute 1/dx * dudx_T. Assume dx[0] == dx[1] == dx[2]
		#pragma omp simd aligned(dudxT_by_dx,dudx_T:ALIGNMENT)
		#pragma omp simd aligned(dudxT_by_dx,negativeDudxT_by_dx,dudx_T:ALIGNMENT)
		for (int it = 0; it < {{nDof*nDofPad}}; it++) {
		dudxT_by_dx[it] = invDx * dudx_T[it];
		negativeDudxT_by_dx[it] = -dudxT_by_dx[it];
		}

		//***************************
		@@ -204,7 +207,7 @@ int {{codeNamespace}}::fusedSpaceTimePredictorVolumeIntegral{{nameSuffix}}(
		for (int x = 0; x < {{nDof}} ; x++){
		solver.{{solverName}}::flux_x(lQi+{{idx(0,yz,x,0)}}, lPi+{{idxLPi(0,yz,x,0)}}, tmpArray+x*{{nVarPad}}); //nVar
		}
		{{ m.matmul('flux_x_sck', 'tmpArray', 'dudxT_by_dx', 'lQi_next', '0', '0', idx(0,yz,0,0)) \| indent(6) }}{##}
		{{ m.matmul('flux_x_sck', 'tmpArray', 'negativeDudxT_by_dx', 'lQi_next', '0', '0', idx(0,yz,0,0)) \| indent(6) }}{##}
		}
		{% endif %}

		@@ -230,7 +233,7 @@ int {{codeNamespace}}::fusedSpaceTimePredictorVolumeIntegral{{nameSuffix}}(
		for (int y = 0; y < {{nDof}} ; y++){
		solver.{{solverName}}::flux_y(lQi+{{idx(z,y,x,0)}}, lPi+{{idxLPi(z,y,x,0)}}, tmpArray+y*{{nVarPad}}); //nVar
		}
		{{ m.matmul('flux_y_sck', 'tmpArray', 'dudxT_by_dx', 'lQi_next', '0', '0', idx(z,0,x,0)) \| indent(8) }}{##}
		{{ m.matmul('flux_y_sck', 'tmpArray', 'negativeDudxT_by_dx', 'lQi_next', '0', '0', idx(z,0,x,0)) \| indent(8) }}{##}
		}
		}
		{% endif %}
		@@ -259,7 +262,7 @@ int {{codeNamespace}}::fusedSpaceTimePredictorVolumeIntegral{{nameSuffix}}(
		for (int z = 0; z < {{nDof}}; z++) {
		solver.{{solverName}}::flux_z(lQi+{{idx(z,0,xy,0)}}, lPi+{{idxLPi(z,0,xy,0)}}, tmpArray+z*{{nVarPad}}); //nVar
		}
		{{ m.matmul('flux_z_sck', 'tmpArray', 'dudxT_by_dx', 'lQi_next', '0', '0', idx(0,0,xy,0)) \| indent(6) }}{##}
		{{ m.matmul('flux_z_sck', 'tmpArray', 'negativeDudxT_by_dx', 'lQi_next', '0', '0', idx(0,0,xy,0)) \| indent(6) }}{##}
		}
		{% endif %}