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

adds mask_mask logical instructions

Browse Source
This commit is contained in:
Eyck-Alexander Jentzsch 2025-02-19 19:46:33 +01:00
parent 63889b02e7
commit c1277b6528
3 changed files with 82 additions and 37 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
gen_input/templates/interp/CORENAME.cpp.gtl
View File

@ -494,6 +494,9 @@ if(vector != null) {%>
throw new std::runtime_error("Unsupported sew bit value"); throw new std::runtime_error("Unsupported sew bit value");
} }
} }
void mask_mask_op(uint8_t* V, unsigned funct6, unsigned funct3, uint64_t vl, uint64_t vstart, unsigned vd, unsigned vs2, unsigned vs1){
return softvector::mask_mask_op<${vlen}>(V, funct6, funct3, vl, vstart, vd, vs2, vs1);
}
<%}%> <%}%>
uint64_t fetch_count{0}; uint64_t fetch_count{0};
uint64_t tval{0}; uint64_t tval{0};

2
src/vm/vector_functions.h
View File

@ -100,6 +100,8 @@ bool sat_vector_imm_op(uint8_t* V, unsigned funct6, unsigned funct3, uint64_t vl
template <unsigned VLEN, typename dest_elem_t, typename src_elem_t = dest_elem_t> template <unsigned VLEN, typename dest_elem_t, typename src_elem_t = dest_elem_t>
void vector_red_op(uint8_t* V, unsigned funct6, unsigned funct3, uint64_t vl, uint64_t vstart, vtype_t vtype, bool vm, unsigned vd, void vector_red_op(uint8_t* V, unsigned funct6, unsigned funct3, uint64_t vl, uint64_t vstart, vtype_t vtype, bool vm, unsigned vd,
unsigned vs2, unsigned vs1); unsigned vs2, unsigned vs1);
template <unsigned VLEN>
void mask_mask_op(uint8_t* V, unsigned funct6, unsigned funct3, uint64_t vl, uint64_t vstart, unsigned vd, unsigned vs2, unsigned vs1);
} // namespace softvector } // namespace softvector
#include "vm/vector_functions.hpp" #include "vm/vector_functions.hpp"
#endif /* _VM_VECTOR_FUNCTIONS_H_ */ #endif /* _VM_VECTOR_FUNCTIONS_H_ */

114
src/vm/vector_functions.hpp
View File

@ -155,14 +155,6 @@ std::function<dest_elem_t(dest_elem_t, src2_elem_t, src1_elem_t)> get_funct(unsi
// case 0b001110: // VSLID1EUP // case 0b001110: // VSLID1EUP
// case 0b001111: // VSLIDE1DOWN // case 0b001111: // VSLIDE1DOWN
// case 0b010111: // VCOMPRESS // case 0b010111: // VCOMPRESS
// case 0b011000: // VMANDN
// case 0b011001: // VMAND
// case 0b011010: // VMOR
// case 0b011011: // VMXOR
// case 0b011100: // VMORN
// case 0b011101: // VMNAND
// case 0b011110: // VMNOR
// case 0b011111: // VMXNOR
case 0b100000: // VDIVU case 0b100000: // VDIVU
return [](dest_elem_t vd, src2_elem_t vs2, src1_elem_t vs1) -> dest_elem_t { return [](dest_elem_t vd, src2_elem_t vs2, src1_elem_t vs1) -> dest_elem_t {
if(vs1 == 0) if(vs1 == 0)
@ -380,34 +372,58 @@ void vector_imm_op(uint8_t* V, unsigned funct6, unsigned funct3, uint64_t vl, ui
} }
return; return;
} }
template <typename elem_t> std::function<bool(elem_t, elem_t)> get_mask_funct(unsigned funct) { template <typename elem_t> std::function<bool(elem_t, elem_t)> get_mask_funct(unsigned funct6, unsigned funct3) {
switch(funct) { if(funct3 == OPIVV || funct3 == OPIVX || funct3 == OPIVI)
case 0b011000: // VMSEQ switch(funct6) {
return [](elem_t vs2, elem_t vs1) { return vs2 == vs1; }; case 0b011000: // VMSEQ
case 0b011001: // VMSNE return [](elem_t vs2, elem_t vs1) { return vs2 == vs1; };
return [](elem_t vs2, elem_t vs1) { return vs2 != vs1; }; case 0b011001: // VMSNE
case 0b011010: // VMSLTU return [](elem_t vs2, elem_t vs1) { return vs2 != vs1; };
return [](elem_t vs2, elem_t vs1) { return vs2 < vs1; }; case 0b011010: // VMSLTU
case 0b011011: // VMSLT return [](elem_t vs2, elem_t vs1) { return vs2 < vs1; };
return [](elem_t vs2, elem_t vs1) { case 0b011011: // VMSLT
return static_cast<std::make_signed_t<elem_t>>(vs2) < static_cast<std::make_signed_t<elem_t>>(vs1); return [](elem_t vs2, elem_t vs1) {
}; return static_cast<std::make_signed_t<elem_t>>(vs2) < static_cast<std::make_signed_t<elem_t>>(vs1);
case 0b011100: // VMSLEU };
return [](elem_t vs2, elem_t vs1) { return vs2 <= vs1; }; case 0b011100: // VMSLEU
case 0b011101: // VMSLE return [](elem_t vs2, elem_t vs1) { return vs2 <= vs1; };
return [](elem_t vs2, elem_t vs1) { case 0b011101: // VMSLE
return static_cast<std::make_signed_t<elem_t>>(vs2) <= static_cast<std::make_signed_t<elem_t>>(vs1); return [](elem_t vs2, elem_t vs1) {
}; return static_cast<std::make_signed_t<elem_t>>(vs2) <= static_cast<std::make_signed_t<elem_t>>(vs1);
case 0b011110: // VMSGTU };
return [](elem_t vs2, elem_t vs1) { return vs2 > vs1; }; case 0b011110: // VMSGTU
case 0b011111: // VMSGT return [](elem_t vs2, elem_t vs1) { return vs2 > vs1; };
return [](elem_t vs2, elem_t vs1) { case 0b011111: // VMSGT
return static_cast<std::make_signed_t<elem_t>>(vs2) > static_cast<std::make_signed_t<elem_t>>(vs1); return [](elem_t vs2, elem_t vs1) {
}; return static_cast<std::make_signed_t<elem_t>>(vs2) > static_cast<std::make_signed_t<elem_t>>(vs1);
};
default: default:
throw new std::runtime_error("Unknown funct in get_mask_funct"); throw new std::runtime_error("Unknown funct6 in get_mask_funct");
} }
else if(funct3 == OPMVV || funct3 == OPMVX)
switch(funct6) {
case 0b011000: // VMANDN
return [](elem_t vs2, elem_t vs1) { return vs2 & !vs1; };
case 0b011001: // VMAND
return [](elem_t vs2, elem_t vs1) { return vs2 & vs1; };
case 0b011010: // VMOR
return [](elem_t vs2, elem_t vs1) { return vs2 | vs1; };
case 0b011011: // VMXOR
return [](elem_t vs2, elem_t vs1) { return vs2 ^ vs1; };
case 0b011100: // VMORN
return [](elem_t vs2, elem_t vs1) { return vs2 | !vs1; };
case 0b011101: // VMNAND
return [](elem_t vs2, elem_t vs1) { return !(vs2 & vs1); };
case 0b011110: // VMNOR
return [](elem_t vs2, elem_t vs1) { return !(vs2 | vs1); };
case 0b011111: // VMXNOR
return [](elem_t vs2, elem_t vs1) { return !(vs2 ^ vs1); };
default:
throw new std::runtime_error("Unknown funct6 in get_mask_funct");
}
else
throw new std::runtime_error("Unknown funct3 in get_mask_funct");
} }
template <unsigned VLEN, typename elem_t> template <unsigned VLEN, typename elem_t>
void mask_vector_vector_op(uint8_t* V, unsigned funct6, unsigned funct3, uint64_t vl, uint64_t vstart, vtype_t vtype, bool vm, unsigned vd, void mask_vector_vector_op(uint8_t* V, unsigned funct6, unsigned funct3, uint64_t vl, uint64_t vstart, vtype_t vtype, bool vm, unsigned vd,
@ -417,7 +433,7 @@ void mask_vector_vector_op(uint8_t* V, unsigned funct6, unsigned funct3, uint64_
auto vs1_view = get_vreg<VLEN, elem_t>(V, vs1, elem_count); auto vs1_view = get_vreg<VLEN, elem_t>(V, vs1, elem_count);
auto vs2_view = get_vreg<VLEN, elem_t>(V, vs2, elem_count); auto vs2_view = get_vreg<VLEN, elem_t>(V, vs2, elem_count);
vmask_view vd_mask_view = read_vmask<VLEN>(V, elem_count, vd); vmask_view vd_mask_view = read_vmask<VLEN>(V, elem_count, vd);
auto fn = get_mask_funct<elem_t>(funct6); auto fn = get_mask_funct<elem_t>(funct6, funct3);
// elements w/ index smaller than vstart are in the prestart and get skipped // elements w/ index smaller than vstart are in the prestart and get skipped
// body is from vstart to min(elem_count, vl) // body is from vstart to min(elem_count, vl)
for(unsigned idx = vstart; idx < std::min(elem_count, vl); idx++) { for(unsigned idx = vstart; idx < std::min(elem_count, vl); idx++) {
@ -449,7 +465,7 @@ void mask_vector_imm_op(uint8_t* V, unsigned funct6, unsigned funct3, uint64_t v
vmask_view mask_reg = read_vmask<VLEN>(V, elem_count); vmask_view mask_reg = read_vmask<VLEN>(V, elem_count);
auto vs2_view = get_vreg<VLEN, elem_t>(V, vs2, elem_count); auto vs2_view = get_vreg<VLEN, elem_t>(V, vs2, elem_count);
vmask_view vd_mask_view = read_vmask<VLEN>(V, elem_count, vd); vmask_view vd_mask_view = read_vmask<VLEN>(V, elem_count, vd);
auto fn = get_mask_funct<elem_t>(funct6); auto fn = get_mask_funct<elem_t>(funct6, funct3);
// elements w/ index smaller than vstart are in the prestart and get skipped // elements w/ index smaller than vstart are in the prestart and get skipped
// body is from vstart to min(elem_count, vl) // body is from vstart to min(elem_count, vl)
for(unsigned idx = vstart; idx < std::min(elem_count, vl); idx++) { for(unsigned idx = vstart; idx < std::min(elem_count, vl); idx++) {
@ -866,4 +882,28 @@ void vector_red_op(uint8_t* V, unsigned funct6, unsigned funct3, uint64_t vl, ui
} }
return; return;
} }
template <unsigned VLEN>
void mask_mask_op(uint8_t* V, unsigned funct6, unsigned funct3, uint64_t vl, uint64_t vstart, unsigned vd, unsigned vs2, unsigned vs1) {
uint64_t elem_count = VLEN;
auto vs1_view = read_vmask<VLEN>(V, elem_count, vs1);
auto vs2_view = read_vmask<VLEN>(V, elem_count, vs2);
auto vd_view = read_vmask<VLEN>(V, elem_count, vd);
auto fn = get_mask_funct<unsigned>(funct6, funct3); // could be bool, but would break the make_signed_t in get_mask_funct
for(unsigned idx = vstart; idx < std::min(vl, elem_count); idx++) {
unsigned new_bit_value = fn(vs2_view[idx], vs1_view[idx]);
uint8_t* cur_mask_byte_addr = vd_view.start + idx / 8;
unsigned cur_bit = idx % 8;
*cur_mask_byte_addr = *cur_mask_byte_addr & ~(1U << cur_bit) | static_cast<unsigned>(new_bit_value) << cur_bit;
}
// the tail is all elements of the destination register beyond the first one
for(unsigned idx = 1; idx < VLEN; idx++) {
// always tail agnostic
// this is a nop, placeholder for vta behavior
unsigned new_bit_value = vd_view[idx];
uint8_t* cur_mask_byte_addr = vd_view.start + idx / 8;
unsigned cur_bit = idx % 8;
*cur_mask_byte_addr = *cur_mask_byte_addr & ~(1U << cur_bit) | static_cast<unsigned>(new_bit_value) << cur_bit;
}
return;
}
} // namespace softvector } // namespace softvector