Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Added function decorations so that pcg32 can be called from Cuda kernels #53

Open
wants to merge 1 commit into
base: master
Choose a base branch
from
Open

Added function decorations so that pcg32 can be called from Cuda kernels #53

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
27 changes: 27 additions & 0 deletions include/pcg_extras.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -300,6 +300,9 @@ inline itype rotl(itype value, bitcount_t rot)
}

template <typename itype>
#ifdef __CUDACC__
__host__ __device__
#endif
inline itype rotr(itype value, bitcount_t rot)
{
constexpr bitcount_t bits = sizeof(itype) * 8;
Expand All @@ -320,25 +323,37 @@ inline itype rotr(itype value, bitcount_t rot)
*/
#if PCG_USE_INLINE_ASM && __GNUC__ && (__x86_64__ || __i386__)

#ifdef __CUDACC__
__host__ __device__
#endif
inline uint8_t rotr(uint8_t value, bitcount_t rot)
{
asm ("rorb %%cl, %0" : "=r" (value) : "0" (value), "c" (rot));
return value;
}

#ifdef __CUDACC__
__host__ __device__
#endif
inline uint16_t rotr(uint16_t value, bitcount_t rot)
{
asm ("rorw %%cl, %0" : "=r" (value) : "0" (value), "c" (rot));
return value;
}

#ifdef __CUDACC__
__host__ __device__
#endif
inline uint32_t rotr(uint32_t value, bitcount_t rot)
{
asm ("rorl %%cl, %0" : "=r" (value) : "0" (value), "c" (rot));
return value;
}

#if __x86_64__
#ifdef __CUDACC__
__host__ __device__
#endif
inline uint64_t rotr(uint64_t value, bitcount_t rot)
{
asm ("rorq %%cl, %0" : "=r" (value) : "0" (value), "c" (rot));
Expand All @@ -351,21 +366,33 @@ inline uint64_t rotr(uint64_t value, bitcount_t rot)

#pragma intrinsic(_rotr, _rotr64, _rotr8, _rotr16)

#ifdef __CUDACC__
__host__ __device__
#endif
inline uint8_t rotr(uint8_t value, bitcount_t rot)
{
return _rotr8(value, rot);
}

#ifdef __CUDACC__
__host__ __device__
#endif
inline uint16_t rotr(uint16_t value, bitcount_t rot)
{
return _rotr16(value, rot);
}

#ifdef __CUDACC__
__host__ __device__
#endif
inline uint32_t rotr(uint32_t value, bitcount_t rot)
{
return _rotr(value, rot);
}

#ifdef __CUDACC__
__host__ __device__
#endif
inline uint64_t rotr(uint64_t value, bitcount_t rot)
{
return _rotr64(value, rot);
Expand Down
76 changes: 75 additions & 1 deletion include/pcg_random.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -410,16 +410,26 @@ class engine : protected output_mixin,
}

protected:

#ifdef __CUDACC__
__host__ __device__
#endif
itype bump(itype state)
{
return state * multiplier() + increment();
}

#ifdef __CUDACC__
__host__ __device__
#endif
itype base_generate()
{
return state_ = bump(state_);
}

#ifdef __CUDACC__
__host__ __device__
#endif
itype base_generate0()
{
itype old_state = state_;
Expand All @@ -428,6 +438,9 @@ class engine : protected output_mixin,
}

public:
#ifdef __CUDACC__
__host__ __device__
#endif
result_type operator()()
{
if (output_previous)
Expand All @@ -436,12 +449,18 @@ class engine : protected output_mixin,
return this->output(base_generate());
}

#ifdef __CUDACC__
__host__ __device__
#endif
result_type operator()(result_type upper_bound)
{
return bounded_rand(*this, upper_bound);
}

protected:
#ifdef __CUDACC__
__host__ __device__
#endif
static itype advance(itype state, itype delta,
itype cur_mult, itype cur_plus);

Expand All @@ -454,6 +473,9 @@ class engine : protected output_mixin,
}

public:
#ifdef __CUDACC__
__host__ __device__
#endif
void advance(itype delta)
{
state_ = advance(state_, delta, this->multiplier(), this->increment());
Expand All @@ -464,6 +486,9 @@ class engine : protected output_mixin,
advance(-delta);
}

#ifdef __CUDACC__
__host__ __device__
#endif
void discard(itype delta)
{
advance(delta);
Expand All @@ -481,6 +506,9 @@ class engine : protected output_mixin,
}
}

#ifdef __CUDACC__
__host__ __device__
#endif
engine(itype state = itype(0xcafef00dd15ea5e5ULL))
: state_(this->is_mcg ? state|state_type(3U)
: bump(state + this->increment()))
Expand All @@ -490,8 +518,10 @@ class engine : protected output_mixin,

// This function may or may not exist. It thus has to be a template
// to use SFINAE; users don't have to worry about its template-ness.

template <typename sm = stream_mixin>
#ifdef __CUDACC__
__host__ __device__
#endif
engine(itype state, typename sm::stream_state stream_seed)
: stream_mixin(stream_seed),
state_(this->is_mcg ? state|state_type(3U)
Expand All @@ -501,6 +531,9 @@ class engine : protected output_mixin,
}

template<typename SeedSeq>
#ifdef __CUDACC__
__host__ __device__
#endif
engine(SeedSeq&& seedSeq, typename std::enable_if<
!stream_mixin::can_specify_stream
&& !std::is_convertible<SeedSeq, itype>::value
Expand All @@ -512,6 +545,9 @@ class engine : protected output_mixin,
}

template<typename SeedSeq>
#ifdef __CUDACC__
__host__ __device__
#endif
engine(SeedSeq&& seedSeq, typename std::enable_if<
stream_mixin::can_specify_stream
&& !std::is_convertible<SeedSeq, itype>::value
Expand Down Expand Up @@ -640,6 +676,9 @@ operator>>(std::basic_istream<CharT,Traits>& in,
template <typename xtype, typename itype,
typename output_mixin, bool output_previous,
typename stream_mixin, typename multiplier_mixin>
#ifdef __CUDACC__
__host__ __device__
#endif
itype engine<xtype,itype,output_mixin,output_previous,stream_mixin,
multiplier_mixin>::advance(
itype state, itype delta, itype cur_mult, itype cur_plus)
Expand Down Expand Up @@ -810,6 +849,9 @@ using mcg_base = engine<xtype, itype,

template <typename xtype, typename itype>
struct xsh_rs_mixin {
#ifdef __CUDACC__
__host__ __device__
#endif
static xtype output(itype internal)
{
constexpr bitcount_t bits = bitcount_t(sizeof(itype) * 8);
Expand Down Expand Up @@ -843,6 +885,9 @@ struct xsh_rs_mixin {

template <typename xtype, typename itype>
struct xsh_rr_mixin {
#ifdef __CUDACC__
__host__ __device__
#endif
static xtype output(itype internal)
{
constexpr bitcount_t bits = bitcount_t(sizeof(itype) * 8);
Expand Down Expand Up @@ -878,6 +923,9 @@ struct xsh_rr_mixin {

template <typename xtype, typename itype>
struct rxs_mixin {
#ifdef __CUDACC__
__host__ __device__
#endif
static xtype output_rxs(itype internal)
{
constexpr bitcount_t bits = bitcount_t(sizeof(itype) * 8);
Expand Down Expand Up @@ -944,6 +992,9 @@ PCG_DEFINE_CONSTANT(pcg128_t, mcg, unmultiplier,

template <typename xtype, typename itype>
struct rxs_m_xs_mixin {
#ifdef __CUDACC__
__host__ __device__
#endif
static xtype output(itype internal)
{
constexpr bitcount_t xtypebits = bitcount_t(sizeof(xtype) * 8);
Expand Down Expand Up @@ -992,6 +1043,9 @@ struct rxs_m_xs_mixin {

template <typename xtype, typename itype>
struct rxs_m_mixin {
#ifdef __CUDACC__
__host__ __device__
#endif
static xtype output(itype internal)
{
constexpr bitcount_t xtypebits = bitcount_t(sizeof(xtype) * 8);
Expand Down Expand Up @@ -1030,6 +1084,9 @@ struct rxs_m_mixin {

template <typename xtype, typename itype>
struct dxsm_mixin {
#ifdef __CUDACC__
__host__ __device__
#endif
inline xtype output(itype internal)
{
constexpr bitcount_t xtypebits = bitcount_t(sizeof(xtype) * 8);
Expand Down Expand Up @@ -1058,6 +1115,9 @@ struct dxsm_mixin {

template <typename xtype, typename itype>
struct xsl_rr_mixin {
#ifdef __CUDACC__
__host__ __device__
#endif
static xtype output(itype internal)
{
constexpr bitcount_t xtypebits = bitcount_t(sizeof(xtype) * 8);
Expand Down Expand Up @@ -1104,6 +1164,9 @@ template <typename xtype, typename itype>
struct xsl_rr_rr_mixin {
typedef typename halfsize_trait<itype>::type htype;

#ifdef __CUDACC__
__host__ __device__
#endif
static itype output(itype internal)
{
constexpr bitcount_t htypebits = bitcount_t(sizeof(htype) * 8);
Expand Down Expand Up @@ -1144,6 +1207,10 @@ struct xsl_rr_rr_mixin {

template <typename xtype, typename itype>
struct xsh_mixin {

#ifdef __CUDACC__
__host__ __device__
#endif
static xtype output(itype internal)
{
constexpr bitcount_t xtypebits = bitcount_t(sizeof(xtype) * 8);
Expand All @@ -1167,6 +1234,10 @@ struct xsh_mixin {

template <typename xtype, typename itype>
struct xsl_mixin {

#ifdef __CUDACC__
__host__ __device__
#endif
inline xtype output(itype internal)
{
constexpr bitcount_t xtypebits = bitcount_t(sizeof(xtype) * 8);
Expand Down Expand Up @@ -1293,6 +1364,9 @@ class extended : public baseclass {
return baseclass::period_pow2() + table_size*extvalclass::period_pow2();
}

#ifdef __CUDACC__
__host__ __device__
#endif
PCG_ALWAYS_INLINE result_type operator()()
{
result_type rhs = get_extended_value();
Expand Down