introduce nthreads_max for amdgpu

Author: omlins

src/shared.jl

@@ -15,22 +15,23 @@ Return an expression that evaluates to `true` if the indices generated by @paral
     This macro is not intended for explicit manual usage. Calls to it are automatically added by @parallel where required.
 """
 
-const SUPPORTED_NDIMS           = [1, 2, 3]
-const NDIMS_NONE                = 0
-const ERRMSG_KERNEL_UNSUPPORTED = "unsupported kernel statements in @parallel kernel definition: @parallel is only applicable to kernels that contain exclusively array assignments using macros from FiniteDifferences{1|2|3}D or from another compatible computation submodule. @parallel_indices supports any kind of statements in the kernels."
-const ERRMSG_CHECK_NDIMS        = "ndims must be evaluatable at parse time (e.g. literal or constant) and has to be one of the following Integers: $(join(SUPPORTED_NDIMS,", "))"
-const ERRMSG_CHECK_MEMOPT       = "memopt must be evaluatable at parse time (e.g. literal or constant) and has to be of type Bool."
-const PSNumber                  = PKNumber
-const LOOPSIZE                  = 16
-const LOOPDIM_NONE              = 0
-const NTHREADS_MAX_LOOPOPT      = 128
-const USE_SHMEMHALO_DEFAULT     = true
-const USE_SHMEMHALO_1D_DEFAULT  = true
-const USE_FULLRANGE_DEFAULT     = (false, false, true)
-const FULLRANGE_THRESHOLD       = 1
-const NOEXPR                    = :(begin end)
-const MOD_METADATA              = :__metadata__ # gensym_world("__metadata__", @__MODULE__) # # TODO: name mangling should be used here later, or if there is any sense to leave it like that then at check whether it's available must be done before creating it
-const META_FUNCTION_PREFIX      = string(gensym_world("META", @__MODULE__))
+const SUPPORTED_NDIMS            = [1, 2, 3]
+const NDIMS_NONE                 = 0
+const ERRMSG_KERNEL_UNSUPPORTED  = "unsupported kernel statements in @parallel kernel definition: @parallel is only applicable to kernels that contain exclusively array assignments using macros from FiniteDifferences{1|2|3}D or from another compatible computation submodule. @parallel_indices supports any kind of statements in the kernels."
+const ERRMSG_CHECK_NDIMS         = "ndims must be evaluatable at parse time (e.g. literal or constant) and has to be one of the following Integers: $(join(SUPPORTED_NDIMS,", "))"
+const ERRMSG_CHECK_MEMOPT        = "memopt must be evaluatable at parse time (e.g. literal or constant) and has to be of type Bool."
+const PSNumber                   = PKNumber
+const LOOPSIZE                   = 16
+const LOOPDIM_NONE               = 0
+const NTHREADS_MAX_MEMOPT_CUDA   = 128
+const NTHREADS_MAX_MEMOPT_AMDGPU = 256
+const USE_SHMEMHALO_DEFAULT      = true
+const USE_SHMEMHALO_1D_DEFAULT   = true
+const USE_FULLRANGE_DEFAULT      = (false, false, true)
+const FULLRANGE_THRESHOLD        = 1
+const NOEXPR                     = :(begin end)
+const MOD_METADATA               = :__metadata__ # gensym_world("__metadata__", @__MODULE__) # # TODO: name mangling should be used here later, or if there is any sense to leave it like that then at check whether it's available must be done before creating it
+const META_FUNCTION_PREFIX       = string(gensym_world("META", @__MODULE__))
 
 
 ## FUNCTIONS TO DEAL WITH KERNEL DEFINITIONS