Enable -Xptas compile flag

Author: horizon-blue

CMakeLists.txt

@@ -10,6 +10,7 @@ set(CMAKE_CXX_STANDARD 20)
 set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
 
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
+set(CMAKE_CUDA_FLAGS_RELEASE "${CMAKE_CUDA_FLAGS_RELEASE} -Xptxas -O3")
 
 ################
 # Core Library #

genmetaballs/src/cuda/core/camera.cu

@@ -26,9 +26,9 @@ CUDA_CALLABLE PixelCoordRange::Iterator& PixelCoordRange::Iterator::operator++()
     return *this;
 }
 
-CUDA_CALLABLE bool PixelCoordRange::Sentinel::operator==(const Iterator& it) const {
+CUDA_CALLABLE bool operator!=(const PixelCoordRange::Iterator& it, const PixelCoordRange::Sentinel& sentinel) {
     // stop if we reach the end of rows, or if the range is empty
-    return it.py >= py_end || it.px_start >= it.px_end || it.py_start >= py_end;
+    return it.py < sentinel.py_end && it.px_start < it.px_end && it.py_start < sentinel.py_end;
 }
 
 CUDA_CALLABLE PixelCoordRange::Iterator PixelCoordRange::begin() const {

genmetaballs/src/cuda/core/camera.cuh

@@ -49,11 +49,11 @@ struct PixelCoordRange {
     // the Sentinel class only needs to hold the stop value (i.e. final row)
     struct Sentinel {
         uint32_t py_end;
-
-        // stopping criterion: true if current row (py) reaches py_end
-        CUDA_CALLABLE bool operator==(const Iterator& it) const;
     };
 
+    // stopping criterion: true if current row (py) reaches py_end
+    friend CUDA_CALLABLE bool operator!=(const Iterator& it, const Sentinel& sentinel);
+
     // range methods
     CUDA_CALLABLE Iterator begin() const;
     CUDA_CALLABLE Sentinel end() const;

tests/cpp_tests/test_confidence.cu

@@ -101,7 +101,7 @@ TEST(GpuConfidenceTest, ConfidenceMultipleValuesGPU_AllTypes) {
 
                 std::vector<float> actual;
                 if (conf_case.is_two_param) {
-                    TwoParameterConfidence conf(conf_case.beta4, conf_case.beta5);
+                    TwoParameterConfidence conf{conf_case.beta4, conf_case.beta5};
                     actual = gpu_get_confidence(sumexpd_vec, conf);
                 } else {
                     ZeroParameterConfidence conf;