relaxing the cmake requirements and build improvements

CMakeLists.txt

@@ -1,4 +1,5 @@
-cmake_minimum_required(VERSION 3.23 FATAL_ERROR)
+#cmake_minimum_required(VERSION 3.23 FATAL_ERROR)
+cmake_minimum_required(VERSION 3.18 FATAL_ERROR)
 
 project(leapct)
 

setup_AMD.py

@@ -79,15 +79,18 @@
     if rocm: # AMD ROCM GPU
         extra_compile_args={'cxx': ['-D__USE_GPU'], 
                             'nvcc': ['-D__USE_GPU', '-O3']}
+        libraries = []
     else: # CUDA GPU
         #extra_compile_args={'cxx': ['-D__USE_GPU'], 
         #                    'nvcc': ['-D__USE_GPU', '-O3']}
         extra_compile_args={'cxx': ['-D__USE_GPU', '-lcufft', '-D__INCLUDE_CUFFT'], 
                             'nvcc': ['-D__USE_GPU', '-O3', '-lcufft', '-D__INCLUDE_CUFFT']}
+        libraries = ['cufft']
     ext_mod = CUDAExtension(
         name='leapct',
         sources=source_files,
         extra_compile_args=extra_compile_args,
+        libraries = libraries,
         #extra_link_args=["-lcufft"], 
         extra_cflags=['-O3'])
 else:

src/CMakeLists.txt

@@ -3,8 +3,8 @@
 # You could use cmake as old as version 3.18 if you replaced
 # the line below that specifies all-major with the line below
 # it which specifies the list of CUDA architectures 
-cmake_minimum_required(VERSION 3.23 FATAL_ERROR)
-#cmake_minimum_required(VERSION 3.18 FATAL_ERROR)
+#cmake_minimum_required(VERSION 3.23 FATAL_ERROR)
+cmake_minimum_required(VERSION 3.18 FATAL_ERROR)
 
 project(leapct CXX CUDA)
 
@@ -157,7 +157,11 @@ target_compile_options(leapct PRIVATE $<$<COMPILE_LANGUAGE:CUDA>:
 
 #set_property(TARGET leapct PROPERTY CXX_STANDARD 14)
 #set_property(TARGET leapct PROPERTY CUDA_ARCHITECTURES native)
-set_property(TARGET leapct PROPERTY CUDA_ARCHITECTURES all-major)
+if(CMAKE_VERSION VERSION_GREATER "3.23")
+    message("Building for all major cuda architectures")
+    set_property(TARGET leapct PROPERTY CUDA_ARCHITECTURES all-major)
+endif()
+#set_property(TARGET leapct PROPERTY CUDA_ARCHITECTURES all-major)
 #set_property(TARGET ${PROJECT_NAME} PROPERTY CUDA_ARCHITECTURES 60 61 62 70 72 75 80 86 87 89)
 #set_property(TARGET ${PROJECT_NAME} PROPERTY CUDA_ARCHITECTURES 75 86 87 89)
 #set_property(TARGET leapct PROPERTY CUDA_ARCHITECTURES OFF)

src/find_center_cpu.cpp

@@ -14,6 +14,7 @@
 #include "find_center_cpu.h"
 
 using namespace std;
+#define USE_MEAN_DIFFERENCE_METRIC
 
 bool findCenter_cpu(float* g, parameters* params, int iRow)
 {
@@ -68,7 +69,7 @@ bool findCenter_parallel_cpu(float* g, parameters* params, int iRow)
 
 	int centerCol_low, centerCol_high;
 
-	setDefaultRange_centerCol(params->numCols, centerCol_low, centerCol_high);
+	setDefaultRange_centerCol(params, centerCol_low, centerCol_high);
 
 	double* shiftCosts = (double*)malloc(sizeof(double) * params->numCols);
 
@@ -90,6 +91,7 @@ bool findCenter_parallel_cpu(float* g, parameters* params, int iRow)
 		float u_0 = -(float(n) + params->colShiftFromFilter) * params->pixelWidth;
 
 		double num = 0.0;
+		double count = 0.0;
 		for (int i = 0; i <= conj_ind - 1; i++)
 		{
 			float phi = params->phis[i];
@@ -122,13 +124,21 @@ bool findCenter_parallel_cpu(float* g, parameters* params, int iRow)
 							//	printf("%f and %f\n", val, val_conj);
 
 							num += (val - val_conj) * (val - val_conj);
+							count += 1.0;
 						}
 					}
 				}
 			}
 		}
 		//printf("%f ", num);
+#ifdef USE_MEAN_DIFFERENCE_METRIC
+		if (count > 0.0)
+			shiftCosts[n] = num / count;
+		else
+			shiftCosts[n] = 0.0;
+#else
 		shiftCosts[n] = num;
+#endif
 	}
 
 	for (int i = centerCol_low; i <= centerCol_high; i++)
@@ -162,7 +172,7 @@ bool findCenter_cone_cpu(float* g, parameters* params, int iRow)
 
 	int centerCol_low, centerCol_high;
 
-	setDefaultRange_centerCol(params->numCols, centerCol_low, centerCol_high);
+	setDefaultRange_centerCol(params, centerCol_low, centerCol_high);
 
 	double* shiftCosts = (double*)malloc(sizeof(double) * params->numCols);
 
@@ -193,6 +203,7 @@ bool findCenter_cone_cpu(float* g, parameters* params, int iRow)
 			u_0 = -(float(n) + params->colShiftFromFilter) * atanTu;
 
 		double num = 0.0;
+		double count = 0.0;
 		for (int i = 0; i <= conj_ind - 1; i++)
 		{
 			float phi = params->phis[i];
@@ -225,12 +236,20 @@ bool findCenter_cone_cpu(float* g, parameters* params, int iRow)
 						//	printf("%f and %f\n", val, val_conj);
 
 						num += (val - val_conj) * (val - val_conj);
+						count += 1.0;
 					}
 				}
 			}
 		}
 		//printf("%f ", num);
+#ifdef USE_MEAN_DIFFERENCE_METRIC
+		if (count > 0.0)
+			shiftCosts[n] = num / count;
+		else
+			shiftCosts[n] = 0.0;
+#else
 		shiftCosts[n] = num;
+#endif
 	}
 
 	for (int i = centerCol_low; i <= centerCol_high; i++)
@@ -285,15 +304,17 @@ float findMinimum(double* costVec, int startInd, int endInd, bool findOnlyLocalM
 	return retVal;
 }
 
-bool setDefaultRange_centerCol(int numCols, int& centerCol_low, int& centerCol_high)
+bool setDefaultRange_centerCol(parameters* params, int& centerCol_low, int& centerCol_high)
 {
 	double c = 0.23;
+	if (params->offsetScan == true)
+		c = 0.1;
 	int N_trim = 50;
-	if (numCols < 200)
+	if (params->numCols < 200)
 		N_trim = 5;
 
-	centerCol_low = int(floor(c * numCols));
-	centerCol_high = int(ceil(numCols - c * numCols));
+	centerCol_low = int(floor(c * params->numCols));
+	centerCol_high = int(ceil(params->numCols - c * params->numCols));
 
 	/*
 	if (left_center_right == -1)
@@ -306,12 +327,12 @@ bool setDefaultRange_centerCol(int numCols, int& centerCol_low, int& centerCol_h
 		centerCol_high = numCols - 1 - N_trim;
 	//*/
 
-	centerCol_low = max(0, min(numCols - 1, centerCol_low));
-	centerCol_high = max(0, min(numCols - 1, centerCol_high));
+	centerCol_low = max(0, min(params->numCols - 1, centerCol_low));
+	centerCol_high = max(0, min(params->numCols - 1, centerCol_high));
 	if (centerCol_low > centerCol_high)
 	{
 		centerCol_low = 5;
-		centerCol_high = numCols - 1 - 5;
+		centerCol_high = params->numCols - 1 - 5;
 	}
 	return true;
 }

src/find_center_cpu.h

@@ -28,7 +28,7 @@ bool findCenter_parallel_cpu(float* g, parameters* params, int iRow = -1);
 bool findCenter_fan_cpu(float* g, parameters* params, int iRow = -1);
 bool findCenter_cone_cpu(float* g, parameters* params, int iRow = -1);
 
-bool setDefaultRange_centerCol(int numCols, int& centerCol_low, int& centerCol_high);
+bool setDefaultRange_centerCol(parameters* params, int& centerCol_low, int& centerCol_high);
 float findMinimum(double* costVec, int startInd, int endInd, bool findOnlyLocalMin = true);
 
 #endif

src/tomographic_models_c_interface.cpp

@@ -931,7 +931,7 @@ PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) {
     m.def("shift_detector", &shift_detector, "");
     m.def("set_flatDetector", &set_flatDetector, "");
     m.def("set_curvedDetector", &set_curvedDetector, "");
-    m.def("int get_detectorType", &int get_detectorType, "");
+    m.def("get_detectorType", &get_detectorType, "");
     m.def("set_numCols", &set_numCols, "");
     m.def("set_numRows", &set_numRows, "");
     m.def("set_pixelHeight", &set_pixelHeight, "");
@@ -940,7 +940,7 @@ PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) {
     m.def("set_centerRow", &set_centerRow, "");
     m.def("set_volume", &set_volume, "");
     m.def("set_volumeDimensionOrder", &set_volumeDimensionOrder, "");
-    m.def("int get_volumeDimensionOrder", &int get_volumeDimensionOrder, "");
+    m.def("get_volumeDimensionOrder", &get_volumeDimensionOrder, "");
     m.def("set_default_volume", &set_default_volume, "");
     m.def("set_numZ", &set_numZ, "");
     m.def("set_numY", &set_numY, "");