GPU Compile Flags: Set denormals to zero unconditionally

Before I kept them in NO_FAST_MATH mode, but this yields warning by nvcc,
which for some stupid reason cannot be disabled. And in principle,
with denormals globally disabled, it is also deterministic again,
and disabling it everywhere is as simple as enabling them everywhere.

Author: davidrohr

GPU/GPUTracking/Base/opencl/CMakeLists.txt

@@ -27,7 +27,7 @@ set(OCL_FLAGS -Dcl_clang_storage_class_specifiers -cl-std=CLC++2021)
 if(NOT GPUCA_DETERMINISTIC_MODE GREATER_EQUAL ${GPUCA_DETERMINISTIC_MODE_MAP_NO_FAST_MATH})
   set(OCL_FLAGS ${OCL_FLAGS} -cl-denorms-are-zero -cl-mad-enable -cl-no-signed-zeros -cl-fast-relaxed-math)
 else()
-  set(OCL_FLAGS ${OCL_FLAGS} -cl-fp32-correctly-rounded-divide-sqrt)
+  set(OCL_FLAGS ${OCL_FLAGS} ${GPUCA_OCL_NO_FAST_MATH_FLAGS})
 endif()
 set(OCL_DEFINECL "-D$<JOIN:$<TARGET_PROPERTY:O2::GPUTracking,COMPILE_DEFINITIONS>,$<SEMICOLON>-D>"
             "-I$<JOIN:$<FILTER:$<TARGET_PROPERTY:O2::GPUTracking,INCLUDE_DIRECTORIES>,EXCLUDE,^/usr/include/?>,$<SEMICOLON>-I>"

GPU/GPUTracking/CMakeLists.txt

@@ -17,10 +17,10 @@ set(MODULE GPUTracking)
 if(GPUCA_DETERMINISTIC_MODE GREATER_EQUAL ${GPUCA_DETERMINISTIC_MODE_MAP_NO_FAST_MATH})
   set(CMAKE_CXX_FLAGS_${CMAKE_BUILD_TYPE_UPPER} "${CMAKE_CXX_FLAGS_${CMAKE_BUILD_TYPE_UPPER}} ${GPUCA_CXX_NO_FAST_MATH_FLAGS}")
   if(GPUCA_DETERMINISTIC_MODE GREATER_EQUAL ${GPUCA_DETERMINISTIC_MODE_MAP_OPTO2})
-    set(CMAKE_CXX_FLAGS_${CMAKE_BUILD_TYPE_UPPER} "${CMAKE_CXX_FLAGS_${CMAKE_BUILD_TYPE_UPPER}} -O2")
+    set(CMAKE_CXX_FLAGS_${CMAKE_BUILD_TYPE_UPPER} "${CMAKE_CXX_FLAGS_${CMAKE_BUILD_TYPE_UPPER}} -O2 -mdaz-ftz")
   endif()
 elseif(NOT CMAKE_BUILD_TYPE_UPPER STREQUAL "DEBUG")
-  set(CMAKE_CXX_FLAGS_${CMAKE_BUILD_TYPE_UPPER} "${CMAKE_CXX_FLAGS_${CMAKE_BUILD_TYPE_UPPER}} -O3 -ffast-math")
+  set(CMAKE_CXX_FLAGS_${CMAKE_BUILD_TYPE_UPPER} "${CMAKE_CXX_FLAGS_${CMAKE_BUILD_TYPE_UPPER}} -O3 -ffast-math -mdaz-ftz")
 endif()
 
 include(cmake/helpers.cmake)

dependencies/FindO2GPU.cmake

@@ -84,8 +84,9 @@ elseif(NOT GPUCA_DETERMINISTIC_MODE MATCHES "^[0-9]+$")
   set(GPUCA_DETERMINISTIC_MODE ${GPUCA_DETERMINISTIC_MODE_MAP_${GPUCA_DETERMINISTIC_MODE}})
   message(STATUS "Set to ${GPUCA_DETERMINISTIC_MODE}")
 endif()
-set(GPUCA_CXX_NO_FAST_MATH_FLAGS "-fno-fast-math -ffp-contract=off")
-set(GPUCA_CUDA_NO_FAST_MATH_FLAGS "--ftz=false --prec-div=true --prec-sqrt=true --fmad false")
+set(GPUCA_CXX_NO_FAST_MATH_FLAGS "-fno-fast-math -ffp-contract=off -mdaz-ftz")
+set(GPUCA_CUDA_NO_FAST_MATH_FLAGS "--ftz=true --prec-div=true --prec-sqrt=true --fmad false")
+set(GPUCA_OCL_NO_FAST_MATH_FLAGS -cl-fp32-correctly-rounded-divide-sqrt -cl-denorms-are-zero)
 if(GPUCA_DETERMINISTIC_MODE GREATER_EQUAL ${GPUCA_DETERMINISTIC_MODE_MAP_WHOLEO2})
   add_definitions(-DGPUCA_DETERMINISTIC_MODE)
   set(CMAKE_CXX_FLAGS_${CMAKE_BUILD_TYPE_UPPER} "${CMAKE_CXX_FLAGS_${CMAKE_BUILD_TYPE_UPPER}} ${GPUCA_CXX_NO_FAST_MATH_FLAGS}")
@@ -172,7 +173,7 @@ if(ENABLE_CUDA)
   endif()
 endif()
 
-# ---------------------------------- HIP ----------------------------------
+# ---------------------------------- OpenCL ----------------------------------
 if(ENABLE_OPENCL)
   find_package(OpenCL)
   if(ENABLE_OPENCL AND NOT ENABLE_OPENCL STREQUAL "AUTO")