thirdpartymirrors/Zelda64Recomp
1
0
Fork 0
mirror of https://github.com/Zelda64Recomp/Zelda64Recomp.git synced 2025-10-30 08:03:03 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions
Zelda64Recomp/include/rsp_vu_impl.h

1537 lines
48 KiB
C++
Raw Blame History

// This file is modified from the Ares N64 emulator core. Ares can
// be found at https://github.com/ares-emulator/ares. The original license
// for this portion of Ares is as follows:
// ----------------------------------------------------------------------
// ares
//
// Copyright(c) 2004 - 2021 ares team, Near et al
//
// Permission to use, copy, modify, and /or distribute this software for any
// purpose with or without fee is hereby granted, provided that the above
// copyright noticeand this permission notice appear in all copies.
//
// THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
// WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
// MERCHANTABILITY AND FITNESS.IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
// ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
// WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
// ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
// OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
// ----------------------------------------------------------------------
#include <cstdint>
#include <algorithm>
using u32 = uint32_t;
#define ACCH vpu.acch
#define ACCM vpu.accm
#define ACCL vpu.accl
#define VCOH vpu.vcoh
#define VCOL vpu.vcol
#define VCCH vpu.vcch
#define VCCL vpu.vccl
#define VCE vpu.vce
#define DIVIN vpu.divin
#define DIVOUT vpu.divout
#define DIVDP vpu.divdp
auto RSP::r128::operator()(u32 index) const -> r128 {
if constexpr (Accuracy::RSP::SISD) {
r128 v{ *this };
switch (index) {
case 0: break;
case 1: break;
case 2: v.u16(1) = v.u16(0); v.u16(3) = v.u16(2); v.u16(5) = v.u16(4); v.u16(7) = v.u16(6); break;
case 3: v.u16(0) = v.u16(1); v.u16(2) = v.u16(3); v.u16(4) = v.u16(5); v.u16(6) = v.u16(7); break;
case 4: v.u16(1) = v.u16(2) = v.u16(3) = v.u16(0); v.u16(5) = v.u16(6) = v.u16(7) = v.u16(4); break;
case 5: v.u16(0) = v.u16(2) = v.u16(3) = v.u16(1); v.u16(4) = v.u16(6) = v.u16(7) = v.u16(5); break;
case 6: v.u16(0) = v.u16(1) = v.u16(3) = v.u16(2); v.u16(4) = v.u16(5) = v.u16(7) = v.u16(6); break;
case 7: v.u16(0) = v.u16(1) = v.u16(2) = v.u16(3); v.u16(4) = v.u16(5) = v.u16(6) = v.u16(7); break;
case 8: for (u32 n = 0; n < 8; n++) v.u16(n) = v.u16(0); break;
case 9: for (u32 n = 0; n < 8; n++) v.u16(n) = v.u16(1); break;
case 10: for (u32 n = 0; n < 8; n++) v.u16(n) = v.u16(2); break;
case 11: for (u32 n = 0; n < 8; n++) v.u16(n) = v.u16(3); break;
case 12: for (u32 n = 0; n < 8; n++) v.u16(n) = v.u16(4); break;
case 13: for (u32 n = 0; n < 8; n++) v.u16(n) = v.u16(5); break;
case 14: for (u32 n = 0; n < 8; n++) v.u16(n) = v.u16(6); break;
case 15: for (u32 n = 0; n < 8; n++) v.u16(n) = v.u16(7); break;
}
return v;
}
if constexpr (Accuracy::RSP::SIMD) {
#if ARCHITECTURE_SUPPORTS_SSE4_1
static const __m128i shuffle[16] = {
//vector
_mm_set_epi8(15,14,13,12,11,10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0), //01234567
_mm_set_epi8(15,14,13,12,11,10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0), //01234567
//scalar quarter
_mm_set_epi8(15,14,15,14,11,10,11,10, 7, 6, 7, 6, 3, 2, 3, 2), //00224466
_mm_set_epi8(13,12,13,12, 9, 8, 9, 8, 5, 4, 5, 4, 1, 0, 1, 0), //11335577
//scalar half
_mm_set_epi8(15,14,15,14,15,14,15,14, 7, 6, 7, 6, 7, 6, 7, 6), //00004444
_mm_set_epi8(13,12,13,12,13,12,13,12, 5, 4, 5, 4, 5, 4, 5, 4), //11115555
_mm_set_epi8(11,10,11,10,11,10,11,10, 3, 2, 3, 2, 3, 2, 3, 2), //22226666
_mm_set_epi8(9, 8, 9, 8, 9, 8, 9, 8, 1, 0, 1, 0, 1, 0, 1, 0), //33337777
//scalar whole
_mm_set_epi8(15,14,15,14,15,14,15,14,15,14,15,14,15,14,15,14), //00000000
_mm_set_epi8(13,12,13,12,13,12,13,12,13,12,13,12,13,12,13,12), //11111111
_mm_set_epi8(11,10,11,10,11,10,11,10,11,10,11,10,11,10,11,10), //22222222
_mm_set_epi8(9, 8, 9, 8, 9, 8, 9, 8, 9, 8, 9, 8, 9, 8, 9, 8), //33333333
_mm_set_epi8(7, 6, 7, 6, 7, 6, 7, 6, 7, 6, 7, 6, 7, 6, 7, 6), //44444444
_mm_set_epi8(5, 4, 5, 4, 5, 4, 5, 4, 5, 4, 5, 4, 5, 4, 5, 4), //55555555
_mm_set_epi8(3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2), //66666666
_mm_set_epi8(1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0), //77777777
};
//todo: benchmark to see if testing for cases 0&1 to return value directly is faster
r128 ret;
ret.v128 = _mm_shuffle_epi8(v128, shuffle[index]);
return ret;
#endif
}
}
auto RSP::accumulatorGet(u32 index) const -> u64 {
return (u64)ACCH.u16(index) << 32 | (u64)ACCM.u16(index) << 16 | (u64)ACCL.u16(index) << 0;
}
auto RSP::accumulatorSet(u32 index, u64 value) -> void {
ACCH.u16(index) = value >> 32;
ACCM.u16(index) = value >> 16;
ACCL.u16(index) = value >> 0;
}
auto RSP::accumulatorSaturate(u32 index, bool slice, u16 negative, u16 positive) const -> u16 {
if (ACCH.s16(index) < 0) {
if (ACCH.u16(index) != 0xffff) return negative;
if (ACCM.s16(index) >= 0) return negative;
} else {
if (ACCH.u16(index) != 0x0000) return positive;
if (ACCM.s16(index) < 0) return positive;
}
return !slice ? ACCL.u16(index) : ACCM.u16(index);
}
auto RSP::CFC2(r32& rt, u8 rd) -> void {
r128 hi, lo;
switch (rd & 3) {
case 0x00: hi = VCOH; lo = VCOL; break;
case 0x01: hi = VCCH; lo = VCCL; break;
case 0x02: hi = zero; lo = VCE; break;
case 0x03: hi = zero; lo = VCE; break; //unverified
}
if constexpr (Accuracy::RSP::SISD) {
rt = 0;
for (u32 n = 0; n < 8; n++) {
rt |= lo.get(n) << (0 + n);
rt |= hi.get(n) << (8 + n);
}
rt = s16(rt);
}
if constexpr (Accuracy::RSP::SIMD) {
#if ARCHITECTURE_SUPPORTS_SSE4_1
static const v128 reverse = _mm_set_epi8(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15);
rt = s16(_mm_movemask_epi8(_mm_shuffle_epi8(_mm_packs_epi16(hi, lo), reverse)));
#endif
}
}
auto RSP::CTC2(cr32& rt, u8 rd) -> void {
r128* hi; r128* lo;
r128 null;
switch (rd & 3) {
case 0x00: hi = &VCOH; lo = &VCOL; break;
case 0x01: hi = &VCCH; lo = &VCCL; break;
case 0x02: hi = &null; lo = &VCE; break;
case 0x03: hi = &null; lo = &VCE; break; //unverified
}
if constexpr (Accuracy::RSP::SISD) {
for (u32 n = 0; n < 8; n++) {
lo->set(n, rt & 1 << (0 + n));
hi->set(n, rt & 1 << (8 + n));
}
}
if constexpr (Accuracy::RSP::SIMD) {
#if ARCHITECTURE_SUPPORTS_SSE4_1
static const v128 mask = _mm_set_epi16(0x0101, 0x0202, 0x0404, 0x0808, 0x1010, 0x2020, 0x4040, 0x8080);
lo->v128 = _mm_cmpeq_epi8(_mm_and_si128(_mm_shuffle_epi8(r128{ ~rt >> 0 }, zero), mask), zero);
hi->v128 = _mm_cmpeq_epi8(_mm_and_si128(_mm_shuffle_epi8(r128{ ~rt >> 8 }, zero), mask), zero);
#endif
}
}
template<u8 e>
auto RSP::LBV(r128& vt, cr32& rs, s8 imm) -> void {
auto address = rs + imm;
vt.byte(e) = RSP_MEM_B(0, address);
}
template<u8 e>
auto RSP::LDV(r128& vt, cr32& rs, s8 imm) -> void {
auto address = rs + imm * 8;
auto start = e;
auto end = std::min(start + 8, 16);
for (u32 offset = start; offset < end; offset++) {
vt.byte(offset & 15) = RSP_MEM_B(0, address++);
}
}
template<u8 e>
auto RSP::LFV(r128& vt, cr32& rs, s8 imm) -> void {
auto address = rs + imm * 16;
auto index = (address & 7) - e;
address &= ~7;
auto start = e;
auto end = std::min(start + 8, 16);
r128 tmp;
for (u32 offset = 0; offset < 4; offset++) {
tmp.element(offset + 0) = RSP_MEM_B(0, address + (index + offset * 4 + 0 & 15)) << 7;
tmp.element(offset + 4) = RSP_MEM_B(0, address + (index + offset * 4 + 8 & 15)) << 7;
}
for (u32 offset = start; offset < end; offset++) {
vt.byte(offset) = tmp.byte(offset);
}
}
template<u8 e>
auto RSP::LHV(r128& vt, cr32& rs, s8 imm) -> void {
auto address = rs + imm * 16;
auto index = (address & 7) - e;
address &= ~7;
for (u32 offset = 0; offset < 8; offset++) {
vt.element(offset) = RSP_MEM_B(0, address + (index + offset * 2 & 15)) << 7;
}
}
template<u8 e>
auto RSP::LLV(r128& vt, cr32& rs, s8 imm) -> void {
auto address = rs + imm * 4;
auto start = e;
auto end = std::min(start + 4, 16);
for (u32 offset = start; offset < end; offset++) {
vt.byte(offset & 15) = RSP_MEM_B(0, address++);
}
}
template<u8 e>
auto RSP::LPV(r128& vt, cr32& rs, s8 imm) -> void {
auto address = rs + imm * 8;
auto index = (address & 7) - e;
address &= ~7;
for (u32 offset = 0; offset < 8; offset++) {
vt.element(offset) = RSP_MEM_B(0, address + (index + offset & 15)) << 8;
}
}
template<u8 e>
auto RSP::LQV(r128& vt, cr32& rs, s8 imm) -> void {
auto address = rs + imm * 16;
auto start = e;
auto end = std::min((u32)(16 + e - (address & 15)), (u32)16);
for (u32 offset = start; offset < end; offset++) {
vt.byte(offset & 15) = RSP_MEM_B(0, address++);
}
}
template<u8 e>
auto RSP::LRV(r128& vt, cr32& rs, s8 imm) -> void {
auto address = rs + imm * 16;
auto index = e;
auto start = 16 - ((address & 15) - index);
address &= ~15;
for (u32 offset = start; offset < 16; offset++) {
vt.byte(offset & 15) = RSP_MEM_B(0, address++);
}
}
template<u8 e>
auto RSP::LSV(r128& vt, cr32& rs, s8 imm) -> void {
auto address = rs + imm * 2;
auto start = e;
auto end = std::min(start + 2, 16);
for (u32 offset = start; offset < end; offset++) {
vt.byte(offset & 15) = RSP_MEM_B(0, address++);
}
}
template<u8 e>
auto RSP::LTV(u8 vt, cr32& rs, s8 imm) -> void {
auto address = rs + imm * 16;
auto begin = address & ~7;
address = begin + ((e + (address & 8)) & 15);
auto vtbase = vt & ~7;
auto vtoff = e >> 1;
for (u32 i = 0; i < 8; i++) {
vpu.r[vtbase + vtoff].byte(i * 2 + 0) = RSP_MEM_B(0, address++);
if (address == begin + 16) address = begin;
vpu.r[vtbase + vtoff].byte(i * 2 + 1) = RSP_MEM_B(0, address++);
if (address == begin + 16) address = begin;
vtoff = vtoff + 1 & 7;
}
}
template<u8 e>
auto RSP::LUV(r128& vt, cr32& rs, s8 imm) -> void {
auto address = rs + imm * 8;
auto index = (address & 7) - e;
address &= ~7;
for (u32 offset = 0; offset < 8; offset++) {
vt.element(offset) = RSP_MEM_B(0, address + (index + offset & 15)) << 7;
}
}
template<u8 e>
auto RSP::LWV(r128& vt, cr32& rs, s8 imm) -> void {
auto address = rs + imm * 16;
auto start = 16 - e;
auto end = e + 16;
for (u32 offset = start; offset < end; offset++) {
vt.byte(offset & 15) = RSP_MEM_B(0, address);
address += 4;
}
}
template<u8 e>
auto RSP::MFC2(r32& rt, cr128& vs) -> void {
auto hi = vs.byte(e + 0 & 15);
auto lo = vs.byte(e + 1 & 15);
rt = s16(hi << 8 | lo << 0);
}
template<u8 e>
auto RSP::MTC2(cr32& rt, r128& vs) -> void {
vs.byte(e + 0) = rt >> 8;
if (e != 15) vs.byte(e + 1) = rt >> 0;
}
template<u8 e>
auto RSP::SBV(cr128& vt, cr32& rs, s8 imm) -> void {
auto address = rs + imm;
RSP_MEM_B(0, address) = vt.byte(e);
}
template<u8 e>
auto RSP::SDV(cr128& vt, cr32& rs, s8 imm) -> void {
auto address = rs + imm * 8;
auto start = e;
auto end = start + 8;
for (u32 offset = start; offset < end; offset++) {
RSP_MEM_B(0, address++) = vt.byte(offset & 15);
}
}
template<u8 e>
auto RSP::SFV(cr128& vt, cr32& rs, s8 imm) -> void {
auto address = rs + imm * 16;
auto base = address & 7;
address &= ~7;
switch (e) {
case 0: case 15:
RSP_MEM_B(0, address + (base + 0 & 15)) = vt.element(0) >> 7;
RSP_MEM_B(0, address + (base + 4 & 15)) = vt.element(1) >> 7;
RSP_MEM_B(0, address + (base + 8 & 15)) = vt.element(2) >> 7;
RSP_MEM_B(0, address + (base + 12 & 15)) = vt.element(3) >> 7;
break;
case 1:
RSP_MEM_B(0, address + (base + 0 & 15)) = vt.element(6) >> 7;
RSP_MEM_B(0, address + (base + 4 & 15)) = vt.element(7) >> 7;
RSP_MEM_B(0, address + (base + 8 & 15)) = vt.element(4) >> 7;
RSP_MEM_B(0, address + (base + 12 & 15)) = vt.element(5) >> 7;
break;
case 4:
RSP_MEM_B(0, address + (base + 0 & 15)) = vt.element(1) >> 7;
RSP_MEM_B(0, address + (base + 4 & 15)) = vt.element(2) >> 7;
RSP_MEM_B(0, address + (base + 8 & 15)) = vt.element(3) >> 7;
RSP_MEM_B(0, address + (base + 12 & 15)) = vt.element(0) >> 7;
break;
case 5:
RSP_MEM_B(0, address + (base + 0 & 15)) = vt.element(7) >> 7;
RSP_MEM_B(0, address + (base + 4 & 15)) = vt.element(4) >> 7;
RSP_MEM_B(0, address + (base + 8 & 15)) = vt.element(5) >> 7;
RSP_MEM_B(0, address + (base + 12 & 15)) = vt.element(6) >> 7;
break;
case 8:
RSP_MEM_B(0, address + (base + 0 & 15)) = vt.element(4) >> 7;
RSP_MEM_B(0, address + (base + 4 & 15)) = vt.element(5) >> 7;
RSP_MEM_B(0, address + (base + 8 & 15)) = vt.element(6) >> 7;
RSP_MEM_B(0, address + (base + 12 & 15)) = vt.element(7) >> 7;
break;
case 11:
RSP_MEM_B(0, address + (base + 0 & 15)) = vt.element(3) >> 7;
RSP_MEM_B(0, address + (base + 4 & 15)) = vt.element(0) >> 7;
RSP_MEM_B(0, address + (base + 8 & 15)) = vt.element(1) >> 7;
RSP_MEM_B(0, address + (base + 12 & 15)) = vt.element(2) >> 7;
break;
case 12:
RSP_MEM_B(0, address + (base + 0 & 15)) = vt.element(5) >> 7;
RSP_MEM_B(0, address + (base + 4 & 15)) = vt.element(6) >> 7;
RSP_MEM_B(0, address + (base + 8 & 15)) = vt.element(7) >> 7;
RSP_MEM_B(0, address + (base + 12 & 15)) = vt.element(4) >> 7;
break;
default:
RSP_MEM_B(0, address + (base + 0 & 15)) = 0;
RSP_MEM_B(0, address + (base + 4 & 15)) = 0;
RSP_MEM_B(0, address + (base + 8 & 15)) = 0;
RSP_MEM_B(0, address + (base + 12 & 15)) = 0;
break;
}
}
template<u8 e>
auto RSP::SHV(cr128& vt, cr32& rs, s8 imm) -> void {
auto address = rs + imm * 16;
auto index = address & 7;
address &= ~7;
for (u32 offset = 0; offset < 8; offset++) {
auto byte = e + offset * 2;
auto value = vt.byte(byte + 0 & 15) << 1 | vt.byte(byte + 1 & 15) >> 7;
RSP_MEM_B(0, address + (index + offset * 2 & 15)) = value;
}
}
template<u8 e>
auto RSP::SLV(cr128& vt, cr32& rs, s8 imm) -> void {
auto address = rs + imm * 4;
auto start = e;
auto end = start + 4;
for (u32 offset = start; offset < end; offset++) {
RSP_MEM_B(0, address++) = vt.byte(offset & 15);
}
}
template<u8 e>
auto RSP::SPV(cr128& vt, cr32& rs, s8 imm) -> void {
auto address = rs + imm * 8;
auto start = e;
auto end = start + 8;
for (u32 offset = start; offset < end; offset++) {
if ((offset & 15) < 8) {
RSP_MEM_B(0, address++) = vt.byte((offset & 7) << 1);
} else {
RSP_MEM_B(0, address++) = vt.element(offset & 7) >> 7;
}
}
}
template<u8 e>
auto RSP::SQV(cr128& vt, cr32& rs, s8 imm) -> void {
auto address = rs + imm * 16;
auto start = e;
auto end = start + (16 - (address & 15));
for (u32 offset = start; offset < end; offset++) {
RSP_MEM_B(0, address++) = vt.byte(offset & 15);
}
}
template<u8 e>
auto RSP::SRV(cr128& vt, cr32& rs, s8 imm) -> void {
auto address = rs + imm * 16;
auto start = e;
auto end = start + (address & 15);
auto base = 16 - (address & 15);
address &= ~15;
for (u32 offset = start; offset < end; offset++) {
RSP_MEM_B(0, address++) = vt.byte(offset + base & 15);
}
}
template<u8 e>
auto RSP::SSV(cr128& vt, cr32& rs, s8 imm) -> void {
auto address = rs + imm * 2;
auto start = e;
auto end = start + 2;
for (u32 offset = start; offset < end; offset++) {
RSP_MEM_B(0, address++) = vt.byte(offset & 15);
}
}
template<u8 e>
auto RSP::STV(u8 vt, cr32& rs, s8 imm) -> void {
auto address = rs + imm * 16;
auto start = vt & ~7;
auto end = start + 8;
auto element = 16 - (e & ~1);
auto base = (address & 7) - (e & ~1);
address &= ~7;
for (u32 offset = start; offset < end; offset++) {
RSP_MEM_B(0, address + (base++ & 15)) = vpu.r[offset].byte(element++ & 15);
RSP_MEM_B(0, address + (base++ & 15)) = vpu.r[offset].byte(element++ & 15);
}
}
template<u8 e>
auto RSP::SUV(cr128& vt, cr32& rs, s8 imm) -> void {
auto address = rs + imm * 8;
auto start = e;
auto end = start + 8;
for (u32 offset = start; offset < end; offset++) {
if ((offset & 15) < 8) {
RSP_MEM_B(0, address++) = vt.element(offset & 7) >> 7;
} else {
RSP_MEM_B(0, address++) = vt.byte((offset & 7) << 1);
}
}
}
template<u8 e>
auto RSP::SWV(cr128& vt, cr32& rs, s8 imm) -> void {
auto address = rs + imm * 16;
auto start = e;
auto end = start + 16;
auto base = address & 7;
address &= ~7;
for (u32 offset = start; offset < end; offset++) {
RSP_MEM_B(0, address + (base++ & 15)) = vt.byte(offset & 15);
}
}
template<u8 e>
auto RSP::VABS(r128& vd, cr128& vs, cr128& vt) -> void {
if constexpr (Accuracy::RSP::SISD) {
r128 vte = vt(e);
for (u32 n = 0; n < 8; n++) {
if (vs.s16(n) < 0) {
if (vte.s16(n) == -32768) {
ACCL.s16(n) = -32768;
vd.s16(n) = 32767;
} else {
ACCL.s16(n) = -vte.s16(n);
vd.s16(n) = -vte.s16(n);
}
} else if (vs.s16(n) > 0) {
ACCL.s16(n) = +vte.s16(n);
vd.s16(n) = +vte.s16(n);
} else {
ACCL.s16(n) = 0;
vd.s16(n) = 0;
}
}
}
if constexpr (Accuracy::RSP::SIMD) {
#if ARCHITECTURE_SUPPORTS_SSE4_1
r128 vs0, slt;
vs0 = _mm_cmpeq_epi16(vs, zero);
slt = _mm_srai_epi16(vs, 15);
vd = _mm_andnot_si128(vs0, vt(e));
vd = _mm_xor_si128(vd, slt);
ACCL = _mm_sub_epi16(vd, slt);
vd = _mm_subs_epi16(vd, slt);
#endif
}
}
template<u8 e>
auto RSP::VADD(r128& vd, cr128& vs, cr128& vt) -> void {
if constexpr (Accuracy::RSP::SISD) {
cr128 vte = vt(e);
for (u32 n = 0; n < 8; n++) {
s32 result = vs.s16(n) + vte.s16(n) + VCOL.get(n);
ACCL.s16(n) = result;
vd.s16(n) = sclamp<16>(result);
}
VCOL = zero;
VCOH = zero;
}
if constexpr (Accuracy::RSP::SIMD) {
#if ARCHITECTURE_SUPPORTS_SSE4_1
r128 vte = vt(e), sum, min, max;
sum = _mm_add_epi16(vs, vte);
ACCL = _mm_sub_epi16(sum, VCOL);
min = _mm_min_epi16(vs, vte);
max = _mm_max_epi16(vs, vte);
min = _mm_subs_epi16(min, VCOL);
vd = _mm_adds_epi16(min, max);
VCOL = zero;
VCOH = zero;
#endif
}
}
template<u8 e>
auto RSP::VADDC(r128& vd, cr128& vs, cr128& vt) -> void {
if constexpr (Accuracy::RSP::SISD) {
cr128 vte = vt(e);
for (u32 n = 0; n < 8; n++) {
u32 result = vs.u16(n) + vte.u16(n);
ACCL.u16(n) = result;
VCOL.set(n, result >> 16);
}
VCOH = zero;
vd = ACCL;
}
if constexpr (Accuracy::RSP::SIMD) {
#if ARCHITECTURE_SUPPORTS_SSE4_1
r128 vte = vt(e), sum;
sum = _mm_adds_epu16(vs, vte);
ACCL = _mm_add_epi16(vs, vte);
VCOL = _mm_cmpeq_epi16(sum, ACCL);
VCOL = _mm_cmpeq_epi16(VCOL, zero);
VCOH = zero;
vd = ACCL;
#endif
}
}
template<u8 e>
auto RSP::VAND(r128& vd, cr128& vs, cr128& vt) -> void {
if constexpr (Accuracy::RSP::SISD) {
r128 vte = vt(e);
for (u32 n = 0; n < 8; n++) {
ACCL.u16(n) = vs.u16(n) & vte.u16(n);
}
vd = ACCL;
}
if constexpr (Accuracy::RSP::SIMD) {
#if ARCHITECTURE_SUPPORTS_SSE4_1
ACCL = _mm_and_si128(vs, vt(e));
vd = ACCL;
#endif
}
}
template<u8 e>
auto RSP::VCH(r128& vd, cr128& vs, cr128& vt) -> void {
if constexpr (Accuracy::RSP::SISD) {
cr128 vte = vt(e);
for (u32 n = 0; n < 8; n++) {
if ((vs.s16(n) ^ vte.s16(n)) < 0) {
s16 result = vs.s16(n) + vte.s16(n);
ACCL.s16(n) = (result <= 0 ? -vte.s16(n) : vs.s16(n));
VCCL.set(n, result <= 0);
VCCH.set(n, vte.s16(n) < 0);
VCOL.set(n, 1);
VCOH.set(n, result != 0 && vs.u16(n) != (vte.u16(n) ^ 0xffff));
VCE.set(n, result == -1);
} else {
s16 result = vs.s16(n) - vte.s16(n);
ACCL.s16(n) = (result >= 0 ? vte.s16(n) : vs.s16(n));
VCCL.set(n, vte.s16(n) < 0);
VCCH.set(n, result >= 0);
VCOL.set(n, 0);
VCOH.set(n, result != 0 && vs.u16(n) != (vte.u16(n) ^ 0xffff));
VCE.set(n, 0);
}
}
vd = ACCL;
}
if constexpr (Accuracy::RSP::SIMD) {
#if ARCHITECTURE_SUPPORTS_SSE4_1
r128 vte = vt(e), nvt, diff, diff0, vtn, dlez, dgez, mask;
VCOL = _mm_xor_si128(vs, vte);
VCOL = _mm_cmplt_epi16(VCOL, zero);
nvt = _mm_xor_si128(vte, VCOL);
nvt = _mm_sub_epi16(nvt, VCOL);
diff = _mm_sub_epi16(vs, nvt);
diff0 = _mm_cmpeq_epi16(diff, zero);
vtn = _mm_cmplt_epi16(vte, zero);
dlez = _mm_cmpgt_epi16(diff, zero);
dgez = _mm_or_si128(dlez, diff0);
dlez = _mm_cmpeq_epi16(zero, dlez);
VCCH = _mm_blendv_epi8(dgez, vtn, VCOL);
VCCL = _mm_blendv_epi8(vtn, dlez, VCOL);
VCE = _mm_cmpeq_epi16(diff, VCOL);
VCE = _mm_and_si128(VCE, VCOL);
VCOH = _mm_or_si128(diff0, VCE);
VCOH = _mm_cmpeq_epi16(VCOH, zero);
mask = _mm_blendv_epi8(VCCH, VCCL, VCOL);
ACCL = _mm_blendv_epi8(vs, nvt, mask);
vd = ACCL;
#endif
}
}
template<u8 e>
auto RSP::VCL(r128& vd, cr128& vs, cr128& vt) -> void {
if constexpr (Accuracy::RSP::SISD) {
cr128 vte = vt(e);
for (u32 n = 0; n < 8; n++) {
if (VCOL.get(n)) {
if (VCOH.get(n)) {
ACCL.u16(n) = VCCL.get(n) ? -vte.u16(n) : vs.u16(n);
} else {
u16 sum = vs.u16(n) + vte.u16(n);
bool carry = (vs.u16(n) + vte.u16(n)) != sum;
if (VCE.get(n)) {
ACCL.u16(n) = VCCL.set(n, (!sum || !carry)) ? -vte.u16(n) : vs.u16(n);
} else {
ACCL.u16(n) = VCCL.set(n, (!sum && !carry)) ? -vte.u16(n) : vs.u16(n);
}
}
} else {
if (VCOH.get(n)) {
ACCL.u16(n) = VCCH.get(n) ? vte.u16(n) : vs.u16(n);
} else {
ACCL.u16(n) = VCCH.set(n, (s32)vs.u16(n) - (s32)vte.u16(n) >= 0) ? vte.u16(n) : vs.u16(n);
}
}
}
VCOL = zero;
VCOH = zero;
VCE = zero;
vd = ACCL;
}
if constexpr (Accuracy::RSP::SIMD) {
#if ARCHITECTURE_SUPPORTS_SSE4_1
r128 vte = vt(e), nvt, diff, ncarry, nvce, diff0, lec1, lec2, leeq, geeq, le, ge, mask;
nvt = _mm_xor_si128(vte, VCOL);
nvt = _mm_sub_epi16(nvt, VCOL);
diff = _mm_sub_epi16(vs, nvt);
ncarry = _mm_adds_epu16(vs, vte);
ncarry = _mm_cmpeq_epi16(diff, ncarry);
nvce = _mm_cmpeq_epi16(VCE, zero);
diff0 = _mm_cmpeq_epi16(diff, zero);
lec1 = _mm_and_si128(diff0, ncarry);
lec1 = _mm_and_si128(nvce, lec1);
lec2 = _mm_or_si128(diff0, ncarry);
lec2 = _mm_and_si128(VCE, lec2);
leeq = _mm_or_si128(lec1, lec2);
geeq = _mm_subs_epu16(vte, vs);
geeq = _mm_cmpeq_epi16(geeq, zero);
le = _mm_andnot_si128(VCOH, VCOL);
le = _mm_blendv_epi8(VCCL, leeq, le);
ge = _mm_or_si128(VCOL, VCOH);
ge = _mm_blendv_epi8(geeq, VCCH, ge);
mask = _mm_blendv_epi8(ge, le, VCOL);
ACCL = _mm_blendv_epi8(vs, nvt, mask);
VCCH = ge;
VCCL = le;
VCOH = zero;
VCOL = zero;
VCE = zero;
vd = ACCL;
#endif
}
}
template<u8 e>
auto RSP::VCR(r128& vd, cr128& vs, cr128& vt) -> void {
if constexpr (Accuracy::RSP::SISD) {
cr128 vte = vt(e);
for (u32 n = 0; n < 8; n++) {
if ((vs.s16(n) ^ vte.s16(n)) < 0) {
VCCH.set(n, vte.s16(n) < 0);
ACCL.u16(n) = VCCL.set(n, vs.s16(n) + vte.s16(n) + 1 <= 0) ? ~vte.u16(n) : vs.u16(n);
} else {
VCCL.set(n, vte.s16(n) < 0);
ACCL.u16(n) = VCCH.set(n, vs.s16(n) - vte.s16(n) >= 0) ? vte.u16(n) : vs.u16(n);
}
}
VCOL = zero;
VCOH = zero;
VCE = zero;
vd = ACCL;
}
if constexpr (Accuracy::RSP::SIMD) {
#if ARCHITECTURE_SUPPORTS_SSE4_1
r128 vte = vt(e), sign, dlez, dgez, nvt, mask;
sign = _mm_xor_si128(vs, vte);
sign = _mm_srai_epi16(sign, 15);
dlez = _mm_and_si128(vs, sign);
dlez = _mm_add_epi16(dlez, vte);
VCCL = _mm_srai_epi16(dlez, 15);
dgez = _mm_or_si128(vs, sign);
dgez = _mm_min_epi16(dgez, vte);
VCCH = _mm_cmpeq_epi16(dgez, vte);
nvt = _mm_xor_si128(vte, sign);
mask = _mm_blendv_epi8(VCCH, VCCL, sign);
ACCL = _mm_blendv_epi8(vs, nvt, mask);
vd = ACCL;
VCOL = zero;
VCOH = zero;
VCE = zero;
#endif
}
}
template<u8 e>
auto RSP::VEQ(r128& vd, cr128& vs, cr128& vt) -> void {
if constexpr (Accuracy::RSP::SISD) {
cr128 vte = vt(e);
for (u32 n = 0; n < 8; n++) {
ACCL.u16(n) = VCCL.set(n, !VCOH.get(n) && vs.u16(n) == vte.u16(n)) ? vs.u16(n) : vte.u16(n);
}
VCCH = zero; //unverified
VCOL = zero;
VCOH = zero;
vd = ACCL;
}
if constexpr (Accuracy::RSP::SIMD) {
#if ARCHITECTURE_SUPPORTS_SSE4_1
r128 vte = vt(e), eq;
eq = _mm_cmpeq_epi16(vs, vte);
VCCL = _mm_andnot_si128(VCOH, eq);
ACCL = _mm_blendv_epi8(vte, vs, VCCL);
VCCH = zero; //unverified
VCOH = zero;
VCOL = zero;
vd = ACCL;
#endif
}
}
template<u8 e>
auto RSP::VGE(r128& vd, cr128& vs, cr128& vt) -> void {
if constexpr (Accuracy::RSP::SISD) {
cr128 vte = vt(e);
for (u32 n = 0; n < 8; n++) {
ACCL.u16(n) = VCCL.set(n, vs.s16(n) > vte.s16(n) || (vs.s16(n) == vte.s16(n) && (!VCOL.get(n) || !VCOH.get(n)))) ? vs.u16(n) : vte.u16(n);
}
VCCH = zero; //unverified
VCOL = zero;
VCOH = zero;
vd = ACCL;
}
if constexpr (Accuracy::RSP::SIMD) {
#if ARCHITECTURE_SUPPORTS_SSE4_1
r128 vte = vt(e), eq, gt, es;
eq = _mm_cmpeq_epi16(vs, vte);
gt = _mm_cmpgt_epi16(vs, vte);
es = _mm_and_si128(VCOH, VCOL);
eq = _mm_andnot_si128(es, eq);
VCCL = _mm_or_si128(gt, eq);
ACCL = _mm_blendv_epi8(vte, vs, VCCL);
VCCH = zero;
VCOH = zero;
VCOL = zero;
vd = ACCL;
#endif
}
}
template<u8 e>
auto RSP::VLT(r128& vd, cr128& vs, cr128& vt) -> void {
if constexpr (Accuracy::RSP::SISD) {
cr128 vte = vt(e);
for (u32 n = 0; n < 8; n++) {
ACCL.u16(n) = VCCL.set(n, vs.s16(n) < vte.s16(n) || (vs.s16(n) == vte.s16(n) && VCOL.get(n) && VCOH.get(n))) ? vs.u16(n) : vte.u16(n);
}
VCCH = zero;
VCOL = zero;
VCOH = zero;
vd = ACCL;
}
if constexpr (Accuracy::RSP::SIMD) {
#if ARCHITECTURE_SUPPORTS_SSE4_1
r128 vte = vt(e), eq, lt;
eq = _mm_cmpeq_epi16(vs, vte);
lt = _mm_cmplt_epi16(vs, vte);
eq = _mm_and_si128(VCOH, eq);
eq = _mm_and_si128(VCOL, eq);
VCCL = _mm_or_si128(lt, eq);
ACCL = _mm_blendv_epi8(vte, vs, VCCL);
VCCH = zero;
VCOH = zero;
VCOL = zero;
vd = ACCL;
#endif
}
}
template<bool U, u8 e>
auto RSP::VMACF(r128& vd, cr128& vs, cr128& vt) -> void {
if constexpr (Accuracy::RSP::SISD) {
cr128 vte = vt(e);
for (u32 n = 0; n < 8; n++) {
accumulatorSet(n, accumulatorGet(n) + (s64)vs.s16(n) * (s64)vte.s16(n) * 2);
if constexpr (U == 0) {
vd.u16(n) = accumulatorSaturate(n, 1, 0x8000, 0x7fff);
}
if constexpr (U == 1) {
vd.u16(n) = ACCH.s16(n) < 0 ? 0x0000 : ACCH.s16(n) || ACCM.s16(n) < 0 ? 0xffff : ACCM.u16(n);
}
}
}
if constexpr (Accuracy::RSP::SIMD) {
#if ARCHITECTURE_SUPPORTS_SSE4_1
r128 vte = vt(e), lo, md, hi, carry, omask;
lo = _mm_mullo_epi16(vs, vte);
hi = _mm_mulhi_epi16(vs, vte);
md = _mm_slli_epi16(hi, 1);
carry = _mm_srli_epi16(lo, 15);
hi = _mm_srai_epi16(hi, 15);
md = _mm_or_si128(md, carry);
lo = _mm_slli_epi16(lo, 1);
omask = _mm_adds_epu16(ACCL, lo);
ACCL = _mm_add_epi16(ACCL, lo);
omask = _mm_cmpeq_epi16(ACCL, omask);
omask = _mm_cmpeq_epi16(omask, zero);
md = _mm_sub_epi16(md, omask);
carry = _mm_cmpeq_epi16(md, zero);
carry = _mm_and_si128(carry, omask);
hi = _mm_sub_epi16(hi, carry);
omask = _mm_adds_epu16(ACCM, md);
ACCM = _mm_add_epi16(ACCM, md);
omask = _mm_cmpeq_epi16(ACCM, omask);
omask = _mm_cmpeq_epi16(omask, zero);
ACCH = _mm_add_epi16(ACCH, hi);
ACCH = _mm_sub_epi16(ACCH, omask);
if constexpr (!U) {
lo = _mm_unpacklo_epi16(ACCM, ACCH);
hi = _mm_unpackhi_epi16(ACCM, ACCH);
vd = _mm_packs_epi32(lo, hi);
} else {
r128 mmask, hmask;
mmask = _mm_srai_epi16(ACCM, 15);
hmask = _mm_srai_epi16(ACCH, 15);
md = _mm_or_si128(mmask, ACCM);
omask = _mm_cmpgt_epi16(ACCH, zero);
md = _mm_andnot_si128(hmask, md);
vd = _mm_or_si128(omask, md);
}
#endif
}
}
auto RSP::VMACQ(r128& vd) -> void {
for (u32 n = 0; n < 8; n++) {
s32 product = ACCH.element(n) << 16 | ACCM.element(n) << 0;
if (product < 0 && !(product & 1 << 5)) product += 32;
else if (product >= 32 && !(product & 1 << 5)) product -= 32;
ACCH.element(n) = product >> 16;
ACCM.element(n) = product >> 0;
vd.element(n) = sclamp<16>(product >> 1) & ~15;
}
}
template<u8 e>
auto RSP::VMADH(r128& vd, cr128& vs, cr128& vt) -> void {
if constexpr (Accuracy::RSP::SISD) {
cr128 vte = vt(e);
for (u32 n = 0; n < 8; n++) {
s32 result = (accumulatorGet(n) >> 16) + vs.s16(n) * vte.s16(n);
ACCH.u16(n) = result >> 16;
ACCM.u16(n) = result >> 0;
vd.u16(n) = accumulatorSaturate(n, 1, 0x8000, 0x7fff);
}
}
if constexpr (Accuracy::RSP::SIMD) {
#if ARCHITECTURE_SUPPORTS_SSE4_1
r128 vte = vt(e), lo, hi, omask;
lo = _mm_mullo_epi16(vs, vte);
hi = _mm_mulhi_epi16(vs, vte);
omask = _mm_adds_epu16(ACCM, lo);
ACCM = _mm_add_epi16(ACCM, lo);
omask = _mm_cmpeq_epi16(ACCM, omask);
omask = _mm_cmpeq_epi16(omask, zero);
hi = _mm_sub_epi16(hi, omask);
ACCH = _mm_add_epi16(ACCH, hi);
lo = _mm_unpacklo_epi16(ACCM, ACCH);
hi = _mm_unpackhi_epi16(ACCM, ACCH);
vd = _mm_packs_epi32(lo, hi);
#endif
}
}
template<u8 e>
auto RSP::VMADL(r128& vd, cr128& vs, cr128& vt) -> void {
if constexpr (Accuracy::RSP::SISD) {
cr128 vte = vt(e);
for (u32 n = 0; n < 8; n++) {
accumulatorSet(n, accumulatorGet(n) + (u32(vs.u16(n) * vte.u16(n)) >> 16));
vd.u16(n) = accumulatorSaturate(n, 0, 0x0000, 0xffff);
}
}
if constexpr (Accuracy::RSP::SIMD) {
#if ARCHITECTURE_SUPPORTS_SSE4_1
r128 vte = vt(e), hi, omask, nhi, nmd, shi, smd, cmask, cval;
hi = _mm_mulhi_epu16(vs, vte);
omask = _mm_adds_epu16(ACCL, hi);
ACCL = _mm_add_epi16(ACCL, hi);
omask = _mm_cmpeq_epi16(ACCL, omask);
omask = _mm_cmpeq_epi16(omask, zero);
hi = _mm_sub_epi16(zero, omask);
omask = _mm_adds_epu16(ACCM, hi);
ACCM = _mm_add_epi16(ACCM, hi);
omask = _mm_cmpeq_epi16(ACCM, omask);
omask = _mm_cmpeq_epi16(omask, zero);
ACCH = _mm_sub_epi16(ACCH, omask);
nhi = _mm_srai_epi16(ACCH, 15);
nmd = _mm_srai_epi16(ACCM, 15);
shi = _mm_cmpeq_epi16(nhi, ACCH);
smd = _mm_cmpeq_epi16(nhi, nmd);
cmask = _mm_and_si128(smd, shi);
cval = _mm_cmpeq_epi16(nhi, zero);
vd = _mm_blendv_epi8(cval, ACCL, cmask);
#endif
}
}
template<u8 e>
auto RSP::VMADM(r128& vd, cr128& vs, cr128& vt) -> void {
if constexpr (Accuracy::RSP::SISD) {
cr128 vte = vt(e);
for (u32 n = 0; n < 8; n++) {
accumulatorSet(n, accumulatorGet(n) + vs.s16(n) * vte.u16(n));
vd.u16(n) = accumulatorSaturate(n, 1, 0x8000, 0x7fff);
}
}
if constexpr (Accuracy::RSP::SIMD) {
#if ARCHITECTURE_SUPPORTS_SSE4_1
r128 vte = vt(e), lo, hi, sign, vta, omask;
lo = _mm_mullo_epi16(vs, vte);
hi = _mm_mulhi_epu16(vs, vte);
sign = _mm_srai_epi16(vs, 15);
vta = _mm_and_si128(vte, sign);
hi = _mm_sub_epi16(hi, vta);
omask = _mm_adds_epu16(ACCL, lo);
ACCL = _mm_add_epi16(ACCL, lo);
omask = _mm_cmpeq_epi16(ACCL, omask);
omask = _mm_cmpeq_epi16(omask, zero);
hi = _mm_sub_epi16(hi, omask);
omask = _mm_adds_epu16(ACCM, hi);
ACCM = _mm_add_epi16(ACCM, hi);
omask = _mm_cmpeq_epi16(ACCM, omask);
omask = _mm_cmpeq_epi16(omask, zero);
hi = _mm_srai_epi16(hi, 15);
ACCH = _mm_add_epi16(ACCH, hi);
ACCH = _mm_sub_epi16(ACCH, omask);
lo = _mm_unpacklo_epi16(ACCM, ACCH);
hi = _mm_unpackhi_epi16(ACCM, ACCH);
vd = _mm_packs_epi32(lo, hi);
#endif
}
}
template<u8 e>
auto RSP::VMADN(r128& vd, cr128& vs, cr128& vt) -> void {
if constexpr (Accuracy::RSP::SISD) {
cr128 vte = vt(e);
for (u32 n = 0; n < 8; n++) {
accumulatorSet(n, accumulatorGet(n) + s64(vs.u16(n) * vte.s16(n)));
vd.u16(n) = accumulatorSaturate(n, 0, 0x0000, 0xffff);
}
}
if constexpr (Accuracy::RSP::SIMD) {
#if ARCHITECTURE_SUPPORTS_SSE4_1
r128 vte = vt(e), lo, hi, sign, vsa, omask, nhi, nmd, shi, smd, cmask, cval;
lo = _mm_mullo_epi16(vs, vte);
hi = _mm_mulhi_epu16(vs, vte);
sign = _mm_srai_epi16(vte, 15);
vsa = _mm_and_si128(vs, sign);
hi = _mm_sub_epi16(hi, vsa);
omask = _mm_adds_epu16(ACCL, lo);
ACCL = _mm_add_epi16(ACCL, lo);
omask = _mm_cmpeq_epi16(ACCL, omask);
omask = _mm_cmpeq_epi16(omask, zero);
hi = _mm_sub_epi16(hi, omask);
omask = _mm_adds_epu16(ACCM, hi);
ACCM = _mm_add_epi16(ACCM, hi);
omask = _mm_cmpeq_epi16(ACCM, omask);
omask = _mm_cmpeq_epi16(omask, zero);
hi = _mm_srai_epi16(hi, 15);
ACCH = _mm_add_epi16(ACCH, hi);
ACCH = _mm_sub_epi16(ACCH, omask);
nhi = _mm_srai_epi16(ACCH, 15);
nmd = _mm_srai_epi16(ACCM, 15);
shi = _mm_cmpeq_epi16(nhi, ACCH);
smd = _mm_cmpeq_epi16(nhi, nmd);
cmask = _mm_and_si128(smd, shi);
cval = _mm_cmpeq_epi16(nhi, zero);
vd = _mm_blendv_epi8(cval, ACCL, cmask);
#endif
}
}
template<u8 e>
auto RSP::VMOV(r128& vd, u8 de, cr128& vt) -> void {
cr128 vte = vt(e);
vd.u16(de) = vte.u16(de);
ACCL = vte;
}
template<u8 e>
auto RSP::VMRG(r128& vd, cr128& vs, cr128& vt) -> void {
if constexpr (Accuracy::RSP::SISD) {
cr128 vte = vt(e);
for (u32 n = 0; n < 8; n++) {
ACCL.u16(n) = VCCL.get(n) ? vs.u16(n) : vte.u16(n);
}
VCOH = zero;
VCOL = zero;
vd = ACCL;
}
if constexpr (Accuracy::RSP::SIMD) {
#if ARCHITECTURE_SUPPORTS_SSE4_1
ACCL = _mm_blendv_epi8(vt(e), vs, VCCL);
VCOH = zero;
VCOL = zero;
vd = ACCL;
#endif
}
}
template<u8 e>
auto RSP::VMUDH(r128& vd, cr128& vs, cr128& vt) -> void {
if constexpr (Accuracy::RSP::SISD) {
cr128 vte = vt(e);
for (u32 n = 0; n < 8; n++) {
accumulatorSet(n, s64(vs.s16(n) * vte.s16(n)) << 16);
vd.u16(n) = accumulatorSaturate(n, 1, 0x8000, 0x7fff);
}
}
if constexpr (Accuracy::RSP::SIMD) {
#if ARCHITECTURE_SUPPORTS_SSE4_1
r128 vte = vt(e), lo, hi;
ACCL = zero;
ACCM = _mm_mullo_epi16(vs, vte);
ACCH = _mm_mulhi_epi16(vs, vte);
lo = _mm_unpacklo_epi16(ACCM, ACCH);
hi = _mm_unpackhi_epi16(ACCM, ACCH);
vd = _mm_packs_epi32(lo, hi);
#endif
}
}
template<u8 e>
auto RSP::VMUDL(r128& vd, cr128& vs, cr128& vt) -> void {
if constexpr (Accuracy::RSP::SISD) {
cr128 vte = vt(e);
for (u32 n = 0; n < 8; n++) {
accumulatorSet(n, u16(vs.u16(n) * vte.u16(n) >> 16));
}
vd = ACCL;
}
if constexpr (Accuracy::RSP::SIMD) {
#if ARCHITECTURE_SUPPORTS_SSE4_1
ACCL = _mm_mulhi_epu16(vs, vt(e));
ACCM = zero;
ACCH = zero;
vd = ACCL;
#endif
}
}
template<u8 e>
auto RSP::VMUDM(r128& vd, cr128& vs, cr128& vt) -> void {
if constexpr (Accuracy::RSP::SISD) {
cr128 vte = vt(e);
for (u32 n = 0; n < 8; n++) {
accumulatorSet(n, s32(vs.s16(n) * vte.u16(n)));
}
vd = ACCM;
}
if constexpr (Accuracy::RSP::SIMD) {
#if ARCHITECTURE_SUPPORTS_SSE4_1
r128 vte = vt(e), sign, vta;
ACCL = _mm_mullo_epi16(vs, vte);
ACCM = _mm_mulhi_epu16(vs, vte);
sign = _mm_srai_epi16(vs, 15);
vta = _mm_and_si128(vte, sign);
ACCM = _mm_sub_epi16(ACCM, vta);
ACCH = _mm_srai_epi16(ACCM, 15);
vd = ACCM;
#endif
}
}
template<u8 e>
auto RSP::VMUDN(r128& vd, cr128& vs, cr128& vt) -> void {
if constexpr (Accuracy::RSP::SISD) {
cr128 vte = vt(e);
for (u32 n = 0; n < 8; n++) {
accumulatorSet(n, s32(vs.u16(n) * vte.s16(n)));
}
vd = ACCL;
}
if constexpr (Accuracy::RSP::SIMD) {
#if ARCHITECTURE_SUPPORTS_SSE4_1
r128 vte = vt(e), sign, vsa;
ACCL = _mm_mullo_epi16(vs, vte);
ACCM = _mm_mulhi_epu16(vs, vte);
sign = _mm_srai_epi16(vte, 15);
vsa = _mm_and_si128(vs, sign);
ACCM = _mm_sub_epi16(ACCM, vsa);
ACCH = _mm_srai_epi16(ACCM, 15);
vd = ACCL;
#endif
}
}
template<bool U, u8 e>
auto RSP::VMULF(r128& vd, cr128& vs, cr128& vt) -> void {
if constexpr (Accuracy::RSP::SISD) {
cr128 vte = vt(e);
for (u32 n = 0; n < 8; n++) {
accumulatorSet(n, (s64)vs.s16(n) * (s64)vte.s16(n) * 2 + 0x8000);
if constexpr (U == 0) {
vd.u16(n) = accumulatorSaturate(n, 1, 0x8000, 0x7fff);
}
if constexpr (U == 1) {
vd.u16(n) = ACCH.s16(n) < 0 ? 0x0000 : (ACCH.s16(n) ^ ACCM.s16(n)) < 0 ? 0xffff : ACCM.u16(n);
}
}
}
if constexpr (Accuracy::RSP::SIMD) {
#if ARCHITECTURE_SUPPORTS_SSE4_1
r128 vte = vt(e), lo, hi, round, sign1, sign2, neq, eq, neg;
lo = _mm_mullo_epi16(vs, vte);
round = _mm_cmpeq_epi16(zero, zero);
sign1 = _mm_srli_epi16(lo, 15);
lo = _mm_add_epi16(lo, lo);
round = _mm_slli_epi16(round, 15);
hi = _mm_mulhi_epi16(vs, vte);
sign2 = _mm_srli_epi16(lo, 15);
ACCL = _mm_add_epi16(round, lo);
sign1 = _mm_add_epi16(sign1, sign2);
hi = _mm_slli_epi16(hi, 1);
neq = _mm_cmpeq_epi16(vs, vte);
ACCM = _mm_add_epi16(hi, sign1);
neg = _mm_srai_epi16(ACCM, 15);
if constexpr (!U) {
eq = _mm_and_si128(neq, neg);
ACCH = _mm_andnot_si128(neq, neg);
vd = _mm_add_epi16(ACCM, eq);
} else {
ACCH = _mm_andnot_si128(neq, neg);
hi = _mm_or_si128(ACCM, neg);
vd = _mm_andnot_si128(ACCH, hi);
}
#endif
}
}
template<u8 e>
auto RSP::VMULQ(r128& vd, cr128& vs, cr128& vt) -> void {
cr128 vte = vt(e);
for (u32 n = 0; n < 8; n++) {
s32 product = (s16)vs.element(n) * (s16)vte.element(n);
if (product < 0) product += 31; //round
ACCH.element(n) = product >> 16;
ACCM.element(n) = product >> 0;
ACCL.element(n) = 0;
vd.element(n) = sclamp<16>(product >> 1) & ~15;
}
}
template<u8 e>
auto RSP::VNAND(r128& vd, cr128& vs, cr128& vt) -> void {
if constexpr (Accuracy::RSP::SISD) {
cr128 vte = vt(e);
for (u32 n = 0; n < 8; n++) {
ACCL.u16(n) = ~(vs.u16(n) & vte.u16(n));
}
vd = ACCL;
}
if constexpr (Accuracy::RSP::SIMD) {
#if ARCHITECTURE_SUPPORTS_SSE4_1
ACCL = _mm_and_si128(vs, vt(e));
ACCL = _mm_xor_si128(ACCL, invert);
vd = ACCL;
#endif
}
}
template<u8 e>
auto RSP::VNE(r128& vd, cr128& vs, cr128& vt) -> void {
if constexpr (Accuracy::RSP::SISD) {
cr128 vte = vt(e);
for (u32 n = 0; n < 8; n++) {
ACCL.u16(n) = VCCL.set(n, vs.u16(n) != vte.u16(n) || VCOH.get(n)) ? vs.u16(n) : vte.u16(n);
}
VCCH = zero; //unverified
VCOL = zero;
VCOH = zero;
vd = ACCL;
}
if constexpr (Accuracy::RSP::SIMD) {
#if ARCHITECTURE_SUPPORTS_SSE4_1
r128 vte = vt(e), eq, ne;
eq = _mm_cmpeq_epi16(vs, vte);
ne = _mm_cmpeq_epi16(eq, zero);
VCCL = _mm_and_si128(VCOH, eq);
VCCL = _mm_or_si128(VCCL, ne);
ACCL = _mm_blendv_epi8(vte, vs, VCCL);
VCCH = zero;
VCOH = zero;
VCOL = zero;
vd = ACCL;
#endif
}
}
auto RSP::VNOP() -> void {
}
template<u8 e>
auto RSP::VNOR(r128& vd, cr128& vs, cr128& vt) -> void {
if constexpr (Accuracy::RSP::SISD) {
cr128 vte = vt(e);
for (u32 n = 0; n < 8; n++) {
ACCL.u16(n) = ~(vs.u16(n) | vte.u16(n));
}
vd = ACCL;
}
if constexpr (Accuracy::RSP::SIMD) {
#if ARCHITECTURE_SUPPORTS_SSE4_1
ACCL = _mm_or_si128(vs, vt(e));
ACCL = _mm_xor_si128(ACCL, invert);
vd = ACCL;
#endif
}
}
template<u8 e>
auto RSP::VNXOR(r128& vd, cr128& vs, cr128& vt) -> void {
if constexpr (Accuracy::RSP::SISD) {
cr128 vte = vt(e);
for (u32 n = 0; n < 8; n++) {
ACCL.u16(n) = ~(vs.u16(n) ^ vte.u16(n));
}
vd = ACCL;
}
if constexpr (Accuracy::RSP::SIMD) {
#if ARCHITECTURE_SUPPORTS_SSE4_1
ACCL = _mm_xor_si128(vs, vt(e));
ACCL = _mm_xor_si128(ACCL, invert);
vd = ACCL;
#endif
}
}
template<u8 e>
auto RSP::VOR(r128& vd, cr128& vs, cr128& vt) -> void {
if constexpr (Accuracy::RSP::SISD) {
cr128 vte = vt(e);
for (u32 n = 0; n < 8; n++) {
ACCL.u16(n) = vs.u16(n) | vte.u16(n);
}
vd = ACCL;
}
if constexpr (Accuracy::RSP::SIMD) {
#if ARCHITECTURE_SUPPORTS_SSE4_1
ACCL = _mm_or_si128(vs, vt(e));
vd = ACCL;
#endif
}
}
template<bool L, u8 e>
auto RSP::VRCP(r128& vd, u8 de, cr128& vt) -> void {
s32 result = 0;
s32 input = L && DIVDP ? DIVIN << 16 | vt.element(e & 7) : s16(vt.element(e & 7));
s32 mask = input >> 31;
s32 data = input ^ mask;
if (input > -32768) data -= mask;
if (data == 0) {
result = 0x7fff'ffff;
} else if (input == -32768) {
result = 0xffff'0000;
} else {
u32 shift = __builtin_clz(data);
u32 index = (u64(data) << shift & 0x7fc0'0000) >> 22;
result = rspReciprocals[index];
result = (0x10000 | result) << 14;
result = result >> 31 - shift ^ mask;
}
DIVDP = 0;
DIVOUT = result >> 16;
ACCL = vt(e);
vd.element(de) = result;
}
template<u8 e>
auto RSP::VRCPH(r128& vd, u8 de, cr128& vt) -> void {
ACCL = vt(e);
DIVDP = 1;
DIVIN = vt.element(e & 7);
vd.element(de) = DIVOUT;
}
template<bool D, u8 e>
auto RSP::VRND(r128& vd, u8 vs, cr128& vt) -> void {
cr128 vte = vt(e);
for (u32 n = 0; n < 8; n++) {
s32 product = (s16)vte.element(n);
if (vs & 1) product <<= 16;
s64 acc = 0;
acc |= ACCH.element(n); acc <<= 16;
acc |= ACCM.element(n); acc <<= 16;
acc |= ACCL.element(n); acc <<= 16;
acc >>= 16;
if (D == 0 && acc < 0) acc = sclip<48>(acc + product);
if (D == 1 && acc >= 0) acc = sclip<48>(acc + product);
ACCH.element(n) = acc >> 32;
ACCM.element(n) = acc >> 16;
ACCL.element(n) = acc >> 0;
vd.element(n) = sclamp<16>(acc >> 16);
}
}
template<bool L, u8 e>
auto RSP::VRSQ(r128& vd, u8 de, cr128& vt) -> void {
s32 result = 0;
s32 input = L && DIVDP ? DIVIN << 16 | vt.element(e & 7) : s16(vt.element(e & 7));
s32 mask = input >> 31;
s32 data = input ^ mask;
if (input > -32768) data -= mask;
if (data == 0) {
result = 0x7fff'ffff;
} else if (input == -32768) {
result = 0xffff'0000;
} else {
u32 shift = __builtin_clz(data);
u32 index = (u64(data) << shift & 0x7fc0'0000) >> 22;
result = rspInverseSquareRoots[index & 0x1fe | shift & 1];
result = (0x10000 | result) << 14;
result = result >> (31 - shift >> 1) ^ mask;
}
DIVDP = 0;
DIVOUT = result >> 16;
ACCL = vt(e);
vd.element(de) = result;
}
template<u8 e>
auto RSP::VRSQH(r128& vd, u8 de, cr128& vt) -> void {
ACCL = vt(e);
DIVDP = 1;
DIVIN = vt.element(e & 7);
vd.element(de) = DIVOUT;
}
template<u8 e>
auto RSP::VSAR(r128& vd, cr128& vs) -> void {
switch (e) {
case 0x8: vd = ACCH; break;
case 0x9: vd = ACCM; break;
case 0xa: vd = ACCL; break;
default: vd = zero; break;
}
}
template<u8 e>
auto RSP::VSUB(r128& vd, cr128& vs, cr128& vt) -> void {
if constexpr (Accuracy::RSP::SISD) {
cr128 vte = vt(e);
for (u32 n = 0; n < 8; n++) {
s32 result = vs.s16(n) - vte.s16(n) - VCOL.get(n);
ACCL.s16(n) = result;
vd.s16(n) = sclamp<16>(result);
}
VCOL = zero;
VCOH = zero;
}
if constexpr (Accuracy::RSP::SIMD) {
#if ARCHITECTURE_SUPPORTS_SSE4_1
r128 vte = vt(e), udiff, sdiff, ov;
udiff = _mm_sub_epi16(vte, VCOL);
sdiff = _mm_subs_epi16(vte, VCOL);
ACCL = _mm_sub_epi16(vs, udiff);
ov = _mm_cmpgt_epi16(sdiff, udiff);
vd = _mm_subs_epi16(vs, sdiff);
vd = _mm_adds_epi16(vd, ov);
VCOL = zero;
VCOH = zero;
#endif
}
}
template<u8 e>
auto RSP::VSUBC(r128& vd, cr128& vs, cr128& vt) -> void {
if constexpr (Accuracy::RSP::SISD) {
cr128 vte = vt(e);
for (u32 n = 0; n < 8; n++) {
u32 result = vs.u16(n) - vte.u16(n);
ACCL.u16(n) = result;
VCOL.set(n, result >> 16);
VCOH.set(n, result != 0);
}
vd = ACCL;
}
if constexpr (Accuracy::RSP::SIMD) {
#if ARCHITECTURE_SUPPORTS_SSE4_1
r128 vte = vt(e), equal, udiff, diff0;
udiff = _mm_subs_epu16(vs, vte);
equal = _mm_cmpeq_epi16(vs, vte);
diff0 = _mm_cmpeq_epi16(udiff, zero);
VCOH = _mm_cmpeq_epi16(equal, zero);
VCOL = _mm_andnot_si128(equal, diff0);
ACCL = _mm_sub_epi16(vs, vte);
vd = ACCL;
#endif
}
}
template<u8 e>
auto RSP::VXOR(r128& vd, cr128& vs, cr128& vt) -> void {
if constexpr (Accuracy::RSP::SISD) {
cr128 vte = vt(e);
for (u32 n = 0; n < 8; n++) {
ACCL.u16(n) = vs.u16(n) ^ vte.u16(n);
}
vd = ACCL;
}
if constexpr (Accuracy::RSP::SIMD) {
#if ARCHITECTURE_SUPPORTS_SSE4_1
ACCL = _mm_xor_si128(vs, vt(e));
vd = ACCL;
#endif
}
}
template<u8 e>
auto RSP::VZERO(r128& vd, cr128& vs, cr128& vt) -> void {
if constexpr (Accuracy::RSP::SISD) {
cr128 vte = vt(e);
for (u32 n = 0; n < 8; n++) {
s32 result = vs.s16(n) + vte.s16(n);
ACCL.s16(n) = result;
vd.s16(n) = 0;
}
}
if constexpr (Accuracy::RSP::SIMD) {
#if ARCHITECTURE_SUPPORTS_SSE4_1
r128 vte = vt(e), sum, min, max;
ACCL = _mm_add_epi16(vs, vte);
vd = _mm_xor_si128(vd, vd);
#endif
}
}
#undef ACCH
#undef ACCM
#undef ACCL
#undef VCOH
#undef VCOL
#undef VCCH
#undef VCCL
#undef VCE
#undef DIVIN
#undef DIVOUT
#undef DIVDP
Reference in a new issue View git blame Copy permalink