GPUCommonMath: double functions, 64-bit bit counts, atomic load/store, host-device calls

Named double functions for GPUCommonMath::Sqrt,Sin,Cos,Tan,ASin,ACos,ATan,ATan2,Log,Copysign.

Clz64, Ctz64, Popcount64: 64-bit counterparts of Clz, Ctz and Popcount on
the same builtins and CUDA/HIP intrinsics. Named rather than uint64_t
overloads, which would change existing calls with 64-bit arguments and
make calls with int arguments ambiguous.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Felix Schlepper <felix.schlepper@cern.ch>

Author: f3sch

Common/MathUtils/include/MathUtils/detail/basicMath.h

@@ -91,17 +91,25 @@ GPUdi() T log(T x)
   return o2::gpu::GPUCommonMath::Log(x);
 };
 
-#ifndef GPUCA_GPUCODE_DEVICE // Double overloads using std namespace not visible on GPU
+#if !defined(GPUCA_GPUCODE_DEVICE) || defined(__CUDACC__) || defined(__HIPCC__) // Double specialisations, not for OpenCL and Metal
 template <>
 GPUhdi() double copysign(double x, double y)
 {
-  return std::copysign(x, y);
+  return o2::gpu::GPUCommonMath::Copysignd(x, y);
 }
 template <>
 GPUhdi() double sqrt(double x)
 {
-  return std::sqrt(x);
+  return o2::gpu::GPUCommonMath::Sqrtd(x);
+};
+template <>
+GPUdi() double log(double x)
+{
+  return o2::gpu::GPUCommonMath::Logd(x);
 };
+#endif
+
+#ifndef GPUCA_GPUCODE_DEVICE // Double overloads using std namespace not visible on GPU
 GPUhdi() double abs(double x)
 {
   return std::abs(x);
@@ -120,11 +128,6 @@ GPUdi() bool finite(double x)
   return std::isfinite(x);
 #endif
 }
-template <>
-GPUdi() double log(double x)
-{
-  return std::log(x);
-};
 #endif
 
 } // namespace detail

Common/MathUtils/include/MathUtils/detail/trigonometric.h

@@ -299,7 +299,7 @@ GPUhdi() constexpr T fastATan2(T y, T x)
     if (xx < 0) { // p1 is in the range [Pi/4, 3*Pi/4]
       phi0 = Pi075;
       tan = y1 > T(0) ? -x1 / y1 : T(0); // yy is always >=0, hence y1>=0
-    } else { // p1 is in the range [-Pi/4, Pi/4]
+    } else {                             // p1 is in the range [-Pi/4, Pi/4]
       phi0 = Pi025;
       tan = x1 > T(0) ? y1 / x1 : T(0);
     }
@@ -376,41 +376,41 @@ GPUdi() double pi()
   return o2::constants::math::PI;
 }
 
-#ifndef GPUCA_GPUCODE_DEVICE
+#if !defined(GPUCA_GPUCODE_DEVICE) || defined(__CUDACC__) || defined(__HIPCC__) // Double specialisations, not for OpenCL and Metal
 template <>
 GPUdi() double tan(double x)
 {
-  return std::tan(x);
+  return o2::gpu::GPUCommonMath::Tand(x);
 };
 template <>
 GPUdi() double sin(double x)
 {
-  return std::sin(x);
+  return o2::gpu::GPUCommonMath::Sind(x);
 };
 template <>
 GPUdi() double atan2(double y, double x)
 {
-  return std::atan2(y, x);
+  return o2::gpu::GPUCommonMath::ATan2d(y, x);
 };
 template <>
 GPUdi() double atan(double x)
 {
-  return std::atan(x);
+  return o2::gpu::GPUCommonMath::ATand(x);
 };
 template <>
 GPUdi() double asin(double x)
 {
-  return std::asin(x);
+  return o2::gpu::GPUCommonMath::ASind(x);
 };
 template <>
 GPUdi() double acos(double x)
 {
-  return std::acos(x);
+  return o2::gpu::GPUCommonMath::ACosd(x);
 };
 template <>
 GPUdi() double cos(double x)
 {
-  return std::cos(x);
+  return o2::gpu::GPUCommonMath::Cosd(x);
 };
 #endif