GPU: Remove C++ <17 protections from GPU code

Author: davidrohr

GPU/Common/GPUCommonAlgorithm.h

@@ -17,8 +17,7 @@
 
 #include "GPUCommonDef.h"
 
-#if !defined(GPUCA_GPUCODE)
-//&& (!defined __cplusplus || __cplusplus < 201402L) // This would enable to custom search also on the CPU if available by the compiler, but it is not always faster, so we stick to std::sort
+#if !defined(GPUCA_GPUCODE) // Could also enable custom search on the CPU, but it is not always faster, so we stick to std::sort
 #include <algorithm>
 #define GPUCA_ALGORITHM_STD
 #endif

GPU/Common/GPUCommonConstants.h

@@ -19,7 +19,7 @@
 
 namespace GPUCA_NAMESPACE::gpu::gpu_common_constants
 {
-static CONSTEXPR const float kCLight = 0.000299792458f; // TODO: Duplicate of MathConstants, fix this now that we use only OpenCL CPP
+static constexpr const float kCLight = 0.000299792458f; // TODO: Duplicate of MathConstants, fix this now that we use only OpenCL CPP
 }
 
 #endif

GPU/Common/GPUCommonDef.h

@@ -40,13 +40,6 @@
   #endif
 #endif
 
-// Definitions for C++11 features
-#if defined(__cplusplus) && __cplusplus >= 201703L
-  #define CONSTEXPR constexpr
-#else
-  #define CONSTEXPR
-#endif
-
 // Set AliRoot / O2 namespace
 #if defined(GPUCA_STANDALONE) || (defined(GPUCA_O2_LIB) && !defined(GPUCA_O2_INTERFACE)) || defined(GPUCA_ALIROOT_LIB) || defined (GPUCA_GPUCODE)
   #define GPUCA_ALIGPUCODE

GPU/Common/GPUCommonLogger.h

@@ -45,7 +45,7 @@ struct DummyLogger {
 #define LOGP(...)
 // #define LOGP(...) static_assert(false, "LOGP(...) unsupported in GPU code");
 
-#elif defined(GPUCA_STANDALONE) || defined(GPUCA_ALIROOT_LIB) || (!defined(__cplusplus) || __cplusplus < 201703L)
+#elif defined(GPUCA_STANDALONE) || defined(GPUCA_ALIROOT_LIB)
 #include <iostream>
 #include <cstdio>
 #define LOG(type) std::cout

GPU/Common/GPUCommonMath.h

@@ -73,8 +73,8 @@ class GPUCommonMath
   GPUd() static float Log(float x);
   GPUd() static float Exp(float x);
   GPUhdni() static float Copysign(float x, float y);
-  GPUd() static CONSTEXPR float TwoPi() { return 6.2831853f; }
-  GPUd() static CONSTEXPR float Pi() { return 3.1415927f; }
+  GPUd() static constexpr float TwoPi() { return 6.2831853f; }
+  GPUd() static constexpr float Pi() { return 3.1415927f; }
   GPUd() static float Round(float x);
   GPUd() static float Floor(float x);
   GPUd() static uint32_t Float2UIntReint(const float& x);
@@ -143,12 +143,12 @@ class GPUCommonMath
   GPUd() static float FMulRZ(float a, float b);
 
   template <int32_t I, class T>
-  GPUd() CONSTEXPR static T nextMultipleOf(T val);
+  GPUd() constexpr static T nextMultipleOf(T val);
 
   template <typename... Args>
   GPUdi() static float Sum2(float w, Args... args)
   {
-    if CONSTEXPR (sizeof...(Args) == 0) {
+    if constexpr (sizeof...(Args) == 0) {
       return w * w;
     } else {
       return w * w + Sum2(args...);
@@ -181,9 +181,9 @@ typedef GPUCommonMath CAMath;
 #endif // clang-format on
 
 template <int32_t I, class T>
-GPUdi() CONSTEXPR T GPUCommonMath::nextMultipleOf(T val)
+GPUdi() constexpr T GPUCommonMath::nextMultipleOf(T val)
 {
-  if CONSTEXPR (I & (I - 1)) {
+  if constexpr (I & (I - 1)) {
     T tmp = val % I;
     if (tmp) {
       val += I - tmp;

GPU/GPUTracking/Base/GPUGeneralKernels.h

@@ -81,7 +81,7 @@ class GPUKernelTemplate
   };
 
   typedef GPUconstantref() GPUConstantMem processorType;
-  GPUhdi() CONSTEXPR static GPUDataTypes::RecoStep GetRecoStep() { return GPUCA_RECO_STEP::NoRecoStep; }
+  GPUhdi() constexpr static GPUDataTypes::RecoStep GetRecoStep() { return GPUCA_RECO_STEP::NoRecoStep; }
   GPUhdi() static processorType* Processor(GPUConstantMem& processors)
   {
     return &processors;
@@ -96,7 +96,7 @@ class GPUKernelTemplate
 class GPUMemClean16 : public GPUKernelTemplate
 {
  public:
-  GPUhdi() CONSTEXPR static GPUDataTypes::RecoStep GetRecoStep() { return GPUCA_RECO_STEP::NoRecoStep; }
+  GPUhdi() constexpr static GPUDataTypes::RecoStep GetRecoStep() { return GPUCA_RECO_STEP::NoRecoStep; }
   template <int32_t iKernel = defaultKernel>
   GPUd() static void Thread(int32_t nBlocks, int32_t nThreads, int32_t iBlock, int32_t iThread, GPUsharedref() GPUSharedMemory& smem, processorType& processors, GPUglobalref() void* ptr, uint64_t size);
 };
@@ -105,7 +105,7 @@ class GPUMemClean16 : public GPUKernelTemplate
 class GPUitoa : public GPUKernelTemplate
 {
  public:
-  GPUhdi() CONSTEXPR static GPUDataTypes::RecoStep GetRecoStep() { return GPUCA_RECO_STEP::NoRecoStep; }
+  GPUhdi() constexpr static GPUDataTypes::RecoStep GetRecoStep() { return GPUCA_RECO_STEP::NoRecoStep; }
   template <int32_t iKernel = defaultKernel>
   GPUd() static void Thread(int32_t nBlocks, int32_t nThreads, int32_t iBlock, int32_t iThread, GPUsharedref() GPUSharedMemory& smem, processorType& processors, GPUglobalref() int32_t* ptr, uint64_t size);
 };

GPU/GPUTracking/Base/GPUParam.inc

@@ -116,7 +116,7 @@ GPUdi() float GPUParam::GetSystematicClusterErrorC122(float x, float y, uint8_t
   if (dx > occupancyTotal * rec.tpc.sysClusErrorC12Box) {
     return 0.f;
   }
-  CONSTEXPR float dEdgeInv = 18.f / CAMath::Pi();
+  constexpr float dEdgeInv = 18.f / CAMath::Pi();
   const float dy = (sector == (GPUCA_NSLICES / 2 + 1) ? 0.5f : -0.5f) * (y / x) * dEdgeInv + 0.5f;
   const float errC12 = rec.tpc.sysClusErrorC12Norm * occupancyTotal * dy;
   return errC12 * errC12;

GPU/GPUTracking/Base/GPUProcessor.h

@@ -108,7 +108,7 @@ class GPUProcessor
     if (basePtr == 0) {
       basePtr = 1;
     }
-    CONSTEXPR const size_t maxAlign = (alignof(S) > alignment) ? alignof(S) : alignment;
+    constexpr const size_t maxAlign = (alignof(S) > alignment) ? alignof(S) : alignment;
     basePtr += getAlignment<maxAlign>(basePtr);
     S* retVal = (S*)(basePtr);
     basePtr += nEntries * sizeof(S);

GPU/GPUTracking/Base/opencl/GPUReconstructionOCL.cl

@@ -14,9 +14,6 @@
 
 // clang-format off
 #define __OPENCL__
-#if defined(__cplusplus) && __cplusplus >= 201703L
-  #define __OPENCL__
-#endif
 #define GPUCA_GPUTYPE_OPENCL
 
 #ifdef __OPENCL__

GPU/GPUTracking/DataCompression/GPUTPCCompressionKernels.cxx

@@ -332,7 +332,7 @@ GPUdi() GPUTPCCompressionGatherKernels::Vec128* GPUTPCCompressionGatherKernels::
 template <typename T, typename S>
 GPUdi() bool GPUTPCCompressionGatherKernels::isAlignedTo(const S* ptr)
 {
-  if CONSTEXPR (alignof(S) >= alignof(T)) {
+  if constexpr (alignof(S) >= alignof(T)) {
     static_cast<void>(ptr);
     return true;
   } else {
@@ -343,10 +343,10 @@ GPUdi() bool GPUTPCCompressionGatherKernels::isAlignedTo(const S* ptr)
 template <>
 GPUdi() void GPUTPCCompressionGatherKernels::compressorMemcpy<uint8_t>(uint8_t* GPUrestrict() dst, const uint8_t* GPUrestrict() src, uint32_t size, int32_t nThreads, int32_t iThread)
 {
-  CONSTEXPR const int32_t vec128Elems = CpyVector<uint8_t, Vec128>::Size;
-  CONSTEXPR const int32_t vec64Elems = CpyVector<uint8_t, Vec64>::Size;
-  CONSTEXPR const int32_t vec32Elems = CpyVector<uint8_t, Vec32>::Size;
-  CONSTEXPR const int32_t vec16Elems = CpyVector<uint8_t, Vec16>::Size;
+  constexpr const int32_t vec128Elems = CpyVector<uint8_t, Vec128>::Size;
+  constexpr const int32_t vec64Elems = CpyVector<uint8_t, Vec64>::Size;
+  constexpr const int32_t vec32Elems = CpyVector<uint8_t, Vec32>::Size;
+  constexpr const int32_t vec16Elems = CpyVector<uint8_t, Vec16>::Size;
 
   if (size >= uint32_t(nThreads * vec128Elems)) {
     compressorMemcpyVectorised<uint8_t, Vec128>(dst, src, size, nThreads, iThread);
@@ -364,9 +364,9 @@ GPUdi() void GPUTPCCompressionGatherKernels::compressorMemcpy<uint8_t>(uint8_t*
 template <>
 GPUdi() void GPUTPCCompressionGatherKernels::compressorMemcpy<uint16_t>(uint16_t* GPUrestrict() dst, const uint16_t* GPUrestrict() src, uint32_t size, int32_t nThreads, int32_t iThread)
 {
-  CONSTEXPR const int32_t vec128Elems = CpyVector<uint16_t, Vec128>::Size;
-  CONSTEXPR const int32_t vec64Elems = CpyVector<uint16_t, Vec64>::Size;
-  CONSTEXPR const int32_t vec32Elems = CpyVector<uint16_t, Vec32>::Size;
+  constexpr const int32_t vec128Elems = CpyVector<uint16_t, Vec128>::Size;
+  constexpr const int32_t vec64Elems = CpyVector<uint16_t, Vec64>::Size;
+  constexpr const int32_t vec32Elems = CpyVector<uint16_t, Vec32>::Size;
 
   if (size >= uint32_t(nThreads * vec128Elems)) {
     compressorMemcpyVectorised<uint16_t, Vec128>(dst, src, size, nThreads, iThread);
@@ -382,8 +382,8 @@ GPUdi() void GPUTPCCompressionGatherKernels::compressorMemcpy<uint16_t>(uint16_t
 template <>
 GPUdi() void GPUTPCCompressionGatherKernels::compressorMemcpy<uint32_t>(uint32_t* GPUrestrict() dst, const uint32_t* GPUrestrict() src, uint32_t size, int32_t nThreads, int32_t iThread)
 {
-  CONSTEXPR const int32_t vec128Elems = CpyVector<uint32_t, Vec128>::Size;
-  CONSTEXPR const int32_t vec64Elems = CpyVector<uint32_t, Vec64>::Size;
+  constexpr const int32_t vec128Elems = CpyVector<uint32_t, Vec128>::Size;
+  constexpr const int32_t vec64Elems = CpyVector<uint32_t, Vec64>::Size;
 
   if (size >= uint32_t(nThreads * vec128Elems)) {
     compressorMemcpyVectorised<uint32_t, Vec128>(dst, src, size, nThreads, iThread);
@@ -446,8 +446,8 @@ GPUdi() void GPUTPCCompressionGatherKernels::compressorMemcpyBuffered(V* buf, T*
   V* GPUrestrict() dstAligned = nullptr;
 
   T* bufT = reinterpret_cast<T*>(buf);
-  CONSTEXPR const int32_t bufSize = GPUCA_WARP_SIZE;
-  CONSTEXPR const int32_t bufTSize = bufSize * sizeof(V) / sizeof(T);
+  constexpr const int32_t bufSize = GPUCA_WARP_SIZE;
+  constexpr const int32_t bufTSize = bufSize * sizeof(V) / sizeof(T);
 
   for (uint32_t i = 0; i < nEntries; i++) {
     uint32_t srcPos = 0;