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

adds whole register moves

Browse Source
This commit is contained in:
Eyck-Alexander Jentzsch 2025-02-21 18:22:49 +01:00
parent 877cad27ba
commit 221d2ee38c
3 changed files with 46 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
gen_input/templates/interp/CORENAME.cpp.gtl
View File

@ -664,6 +664,23 @@ if(vector != null) {%>
throw new std::runtime_error("Unsupported sew bit value");
}
}
void vector_compress(uint8_t* V, uint64_t vl, uint64_t vstart, softvector::vtype_t vtype, uint8_t vd, uint8_t vs2, uint8_t vs1, uint8_t sew_val){
switch(sew_val){
case 0b000:
return softvector::vector_compress<${vlen}, uint8_t>(V, vl, vstart, vtype, vd, vs2, vs1);
case 0b001:
return softvector::vector_compress<${vlen}, uint16_t>(V, vl, vstart, vtype, vd, vs2, vs1);
case 0b010:
return softvector::vector_compress<${vlen}, uint32_t>(V, vl, vstart, vtype, vd, vs2, vs1);
case 0b011:
return softvector::vector_compress<${vlen}, uint64_t>(V, vl, vstart, vtype, vd, vs2, vs1);
default:
throw new std::runtime_error("Unsupported sew bit value");
}
}
void vector_whole_move(uint8_t* V, uint8_t vd, uint8_t vs2, uint8_t count){
return softvector::vector_whole_move<${vlen}>(V, vd, vs2, count);
}
<%}%>
uint64_t fetch_count{0};
uint64_t tval{0};

3
src/vm/vector_functions.h
View File

@ -119,6 +119,9 @@ 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);
template <unsigned VLEN, typename scr_elem_t>
void vector_compress(uint8_t* V, uint64_t vl, uint64_t vstart, vtype_t vtype, unsigned vd, unsigned vs2, unsigned vs1);
template <unsigned VLEN> void vector_whole_move(uint8_t* V, unsigned vd, unsigned vs2, unsigned count);
} // namespace softvector
#include "vm/vector_functions.hpp"
#endif /* _VM_VECTOR_FUNCTIONS_H_ */

26
src/vm/vector_functions.hpp
View File

@ -34,6 +34,7 @@
#pragma once
#include "vm/vector_functions.h"
#include <cstdint>
#include <cstring>
#include <functional>
#include <limits>
#include <stdexcept>
@ -1107,4 +1108,29 @@ void vector_imm_gather(uint8_t* V, uint64_t vl, uint64_t vstart, vtype_t vtype,
}
return;
}
template <unsigned VLEN, typename scr_elem_t>
void vector_compress(uint8_t* V, uint64_t vl, uint64_t vstart, vtype_t vtype, unsigned vd, unsigned vs2, unsigned vs1) {
uint64_t vlmax = VLEN * vtype.lmul() / vtype.sew();
vmask_view mask_reg = read_vmask<VLEN>(V, vlmax, vs1);
auto vs2_view = get_vreg<VLEN, scr_elem_t>(V, vs2, vlmax);
auto vd_view = get_vreg<VLEN, scr_elem_t>(V, vd, vlmax);
unsigned current_pos = 0;
for(unsigned idx = vstart; idx < std::min(vlmax, vl); idx++) {
if(mask_reg[idx]) {
vd_view[current_pos] = vs2_view[idx];
current_pos += 1;
}
}
for(unsigned idx = current_pos; idx < vlmax; idx++) {
vd_view[idx] = vtype.vta() ? vd_view[idx] : vd_view[idx];
}
return;
}
template <unsigned VLEN> void vector_whole_move(uint8_t* V, unsigned vd, unsigned vs2, unsigned count) {
auto vd_view = get_vreg<VLEN, uint8_t>(V, vd, 1);
auto vs2_view = get_vreg<VLEN, uint8_t>(V, vs2, 1);
memcpy(vd_view.start, vs2_view.start, VLEN / 8 * count);
return;
}
} // namespace softvector