DBT-RISE/DBT-RISE-TGC
12
0
Fork 0
Code Issues 1 Pull Requests 1 Projects Releases Wiki Activity

adds gather instructions

Browse Source
This commit is contained in:
Eyck-Alexander Jentzsch
2025-02-21 17:29:57 +01:00
parent a26505cb5c
commit 877cad27ba
3 changed files with 85 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
src/vm/vector_functions.h
View File

@ -115,7 +115,10 @@ template <unsigned VLEN, typename src_elem_t>
void vector_slide1up(uint8_t* V, uint64_t vl, uint64_t vstart, vtype_t vtype, bool vm, unsigned vd, unsigned vs2, int64_t imm);
template <unsigned VLEN, typename src_elem_t>
void vector_slide1down(uint8_t* V, uint64_t vl, uint64_t vstart, vtype_t vtype, bool vm, unsigned vd, unsigned vs2, int64_t imm);
template <unsigned VLEN, typename dest_elem_t, typename scr_elem_t = dest_elem_t>
void vector_vector_gather(uint8_t* V, uint64_t vl, uint64_t vstart, vtype_t vtype, bool vm, unsigned vd, unsigned vs2, unsigned vs1);
template <unsigned VLEN, typename scr_elem_t>
void vector_imm_gather(uint8_t* V, uint64_t vl, uint64_t vstart, vtype_t vtype, bool vm, unsigned vd, unsigned vs2, uint64_t imm);
} // namespace softvector
#include "vm/vector_functions.hpp"
#endif /* _VM_VECTOR_FUNCTIONS_H_ */

39
src/vm/vector_functions.hpp
View File

@ -1068,4 +1068,43 @@ void vector_slide1down(uint8_t* V, uint64_t vl, uint64_t vstart, vtype_t vtype,
vd_view[0] = vtype.vma() ? vd_view[0] : vd_view[0];
}
}
template <unsigned VLEN, typename dest_elem_t, typename scr_elem_t>
void vector_vector_gather(uint8_t* V, uint64_t vl, uint64_t vstart, vtype_t vtype, bool vm, unsigned vd, unsigned vs2, unsigned vs1) {
uint64_t vlmax = VLEN * vtype.lmul() / vtype.sew();
vmask_view mask_reg = read_vmask<VLEN>(V, vlmax);
auto vs1_view = get_vreg<VLEN, scr_elem_t>(V, vs1, vlmax);
auto vs2_view = get_vreg<VLEN, dest_elem_t>(V, vs2, vlmax);
auto vd_view = get_vreg<VLEN, dest_elem_t>(V, vd, vlmax);
for(unsigned idx = vstart; idx < std::min(vlmax, vl); idx++) {
bool mask_active = vm ? 1 : mask_reg[idx];
if(mask_active) {
vd_view[idx] = (vs1_view[idx] >= vlmax) ? 0 : vs2_view[vs1_view[idx]];
} else {
vd_view[idx] = vtype.vma() ? vd_view[idx] : vd_view[idx];
}
}
for(unsigned idx = vl; idx < vlmax; idx++) {
vd_view[idx] = vtype.vta() ? vd_view[idx] : vd_view[idx];
}
return;
}
template <unsigned VLEN, typename scr_elem_t>
void vector_imm_gather(uint8_t* V, uint64_t vl, uint64_t vstart, vtype_t vtype, bool vm, unsigned vd, unsigned vs2, uint64_t imm) {
uint64_t vlmax = VLEN * vtype.lmul() / vtype.sew();
vmask_view mask_reg = read_vmask<VLEN>(V, vlmax);
auto vs2_view = get_vreg<VLEN, scr_elem_t>(V, vs2, vlmax);
auto vd_view = get_vreg<VLEN, scr_elem_t>(V, vd, vlmax);
for(unsigned idx = vstart; idx < std::min(vlmax, vl); idx++) {
bool mask_active = vm ? 1 : mask_reg[idx];
if(mask_active) {
vd_view[idx] = (imm >= vlmax) ? 0 : vs2_view[imm];
} else {
vd_view[idx] = vtype.vma() ? vd_view[idx] : vd_view[idx];
}
}
for(unsigned idx = vl; idx < vlmax; idx++) {
vd_view[idx] = vtype.vta() ? vd_view[idx] : vd_view[idx];
}
return;
}
} // namespace softvector