mirrors/ZLUDA
0
0
Fork 0
mirror of https://github.com/vosen/ZLUDA.git synced 2025-10-04 23:30:18 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

Implement vote instruction and add support for %laneid (#484)

Browse source
This commit is contained in:
Andrzej Janik 2025-08-29 03:23:09 +02:00 • committed by GitHub
commit ea99dcc0b0
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
19 changed files with 538 additions and 15 deletions
Download patch file Download diff file Expand all files Collapse all files

43
ptx/lib/zluda_ptx_impl.cpp
View file

@ -48,6 +48,11 @@ extern "C"
return (uint32_t)__ockl_get_num_groups(member);
}
uint32_t FUNC(sreg_laneid)()
{
return __lane_id();
}
uint32_t __ockl_bfe_u32(uint32_t, uint32_t, uint32_t) __device__;
uint32_t FUNC(bfe_u32)(uint32_t base, uint32_t pos_32, uint32_t len_32)
{
@ -519,4 +524,42 @@ typedef uint32_t ShflSyncResult __attribute__((ext_vector_type(2)));
{
return in;
}
__device__ static inline uint32_t ballot(bool value, bool negate)
{
__builtin_amdgcn_wave_barrier();
return __builtin_amdgcn_ballot_w32(negate ? !value : value);
}
bool FUNC(vote_sync_any_pred)(bool value, uint32_t membermask __attribute__((unused)))
{
return ballot(value, false) != 0;
}
bool FUNC(vote_sync_any_pred_negate)(bool value, uint32_t membermask __attribute__((unused)))
{
return ballot(value, true) != 0;
}
// IMPORTANT: exec mask must be a subset of membermask, the behavior is undefined otherwise
bool FUNC(vote_sync_all_pred)(bool value, uint32_t membermask __attribute__((unused)))
{
return ballot(value, false) == __builtin_amdgcn_read_exec_lo();
}
// also known as "none"
bool FUNC(vote_sync_all_pred_negate)(bool value, uint32_t membermask __attribute__((unused)))
{
return ballot(value, false) == 0;
}
uint32_t FUNC(vote_sync_ballot_b32)(bool value, uint32_t membermask __attribute__((unused)))
{
return ballot(value, false);
}
uint32_t FUNC(vote_sync_ballot_b32_negate)(bool value, uint32_t membermask __attribute__((unused)))
{
return ballot(value, true);
}
}