Skip to content

V
V8
Commit 01e3be50 authored by Ivica Bogosavljevic's avatar Ivica Bogosavljevic Committed by Commit Bot
Browse files
Options

MIPS[64]: Implement 3R MSA instructions in simulator 

Bug: 
Change-Id: Ia7bbbb2aba28ec36b470a1b23ebe38fc3a09e600
Reviewed-on: https://chromium-review.googlesource.com/657757
Commit-Queue: Ivica Bogosavljevic <ivica.bogosavljevic@imgtec.com>
Reviewed-by: 's avatarMiran Karić <Miran.Karic@imgtec.com>
Reviewed-by: 's avatarJakob Kummerow <jkummerow@chromium.org>
Cr-Commit-Position: refs/heads/master@{#48108}
parent a76d0a77
Expand all Show whitespace changes
Inline Side-by-side
Showing with 2197 additions and 150 deletions
src/mips/simulator-mips.cc
View file @ 01e3be50
......@@ -4897,53 +4897,213 @@ void Simulator::DecodeTypeMsaMI10() {
#undef MSA_MI10_STORE
}
void Simulator::DecodeTypeMsa3R() {
DCHECK(IsMipsArchVariant(kMips32r6));
DCHECK(CpuFeatures::IsSupported(MIPS_SIMD));
uint32_t opcode = instr_.InstructionBits() & kMsa3RMask;
template <typename T>
T Simulator::Msa3RInstrHelper(uint32_t opcode, T wd, T ws, T wt) {
typedef typename std::make_unsigned<T>::type uT;
T res;
T wt_modulo = wt % (sizeof(T) * 8);
switch (opcode) {
case SLL_MSA:
res = static_cast<T>(ws << wt_modulo);
break;
case SRA_MSA:
res = static_cast<T>(ArithmeticShiftRight(ws, wt_modulo));
break;
case SRL_MSA:
res = static_cast<T>(static_cast<uT>(ws) >> wt_modulo);
break;
case BCLR:
res = static_cast<T>(static_cast<T>(~(1ull << wt_modulo)) & ws);
break;
case BSET:
res = static_cast<T>(static_cast<T>(1ull << wt_modulo) | ws);
break;
case BNEG:
case BINSL:
case BINSR:
res = static_cast<T>(static_cast<T>(1ull << wt_modulo) ^ ws);
break;
case BINSL: {
int elem_size = 8 * sizeof(T);
int bits = wt_modulo + 1;
if (bits == elem_size) {
res = static_cast<T>(ws);
} else {
uint64_t mask = ((1ull << bits) - 1) << (elem_size - bits);
res = static_cast<T>((static_cast<T>(mask) & ws) |
(static_cast<T>(~mask) & wd));
}
} break;
case BINSR: {
int elem_size = 8 * sizeof(T);
int bits = wt_modulo + 1;
if (bits == elem_size) {
res = static_cast<T>(ws);
} else {
uint64_t mask = (1ull << bits) - 1;
res = static_cast<T>((static_cast<T>(mask) & ws) |
(static_cast<T>(~mask) & wd));
}
} break;
case ADDV:
res = ws + wt;
break;
case SUBV:
res = ws - wt;
break;
case MAX_S:
res = Max(ws, wt);
break;
case MAX_U:
res = static_cast<T>(Max(static_cast<uT>(ws), static_cast<uT>(wt)));
break;
case MIN_S:
res = Min(ws, wt);
break;
case MIN_U:
res = static_cast<T>(Min(static_cast<uT>(ws), static_cast<uT>(wt)));
break;
case MAX_A:
// We use negative abs in order to avoid problems
// with corner case for MIN_INT
res = Nabs(ws) < Nabs(wt) ? ws : wt;
break;
case MIN_A:
// We use negative abs in order to avoid problems
// with corner case for MIN_INT
res = Nabs(ws) > Nabs(wt) ? ws : wt;
break;
case CEQ:
res = static_cast<T>(!Compare(ws, wt) ? -1ull : 0ull);
break;
case CLT_S:
res = static_cast<T>((Compare(ws, wt) == -1) ? -1ull : 0ull);
break;
case CLT_U:
res = static_cast<T>(
(Compare(static_cast<uT>(ws), static_cast<uT>(wt)) == -1) ? -1ull
: 0ull);
break;
case CLE_S:
res = static_cast<T>((Compare(ws, wt) != 1) ? -1ull : 0ull);
break;
case CLE_U:
res = static_cast<T>(
(Compare(static_cast<uT>(ws), static_cast<uT>(wt)) != 1) ? -1ull
: 0ull);
break;
case ADD_A:
case ADDS_A:
res = static_cast<T>(Abs(ws) + Abs(wt));
break;
case ADDS_A: {
T ws_nabs = Nabs(ws);
T wt_nabs = Nabs(wt);
if (ws_nabs < -std::numeric_limits<T>::max() - wt_nabs) {
res = std::numeric_limits<T>::max();
} else {
res = -(ws_nabs + wt_nabs);
}
} break;
case ADDS_S:
case ADDS_U:
res = SaturateAdd(ws, wt);
break;
case ADDS_U: {
uT ws_u = static_cast<uT>(ws);
uT wt_u = static_cast<uT>(wt);
res = static_cast<T>(SaturateAdd(ws_u, wt_u));
} break;
case AVE_S:
case AVE_U:
res = static_cast<T>((wt & ws) + ((wt ^ ws) >> 1));
break;
case AVE_U: {
uT ws_u = static_cast<uT>(ws);
uT wt_u = static_cast<uT>(wt);
res = static_cast<T>((wt_u & ws_u) + ((wt_u ^ ws_u) >> 1));
} break;
case AVER_S:
case AVER_U:
res = static_cast<T>((wt | ws) - ((wt ^ ws) >> 1));
break;
case AVER_U: {
uT ws_u = static_cast<uT>(ws);
uT wt_u = static_cast<uT>(wt);
res = static_cast<T>((wt_u | ws_u) - ((wt_u ^ ws_u) >> 1));
} break;
case SUBS_S:
case SUBS_U:
case SUBSUS_U:
case SUBSUU_S:
res = SaturateSub(ws, wt);
break;
case SUBS_U: {
uT ws_u = static_cast<uT>(ws);
uT wt_u = static_cast<uT>(wt);
res = static_cast<T>(SaturateSub(ws_u, wt_u));
} break;
case SUBSUS_U: {
uT wsu = static_cast<uT>(ws);
if (wt > 0) {
uT wtu = static_cast<uT>(wt);
if (wtu > wsu) {
res = 0;
} else {
res = static_cast<T>(wsu - wtu);
}
} else {
if (wsu > std::numeric_limits<uT>::max() + wt) {
res = static_cast<T>(std::numeric_limits<uT>::max());
} else {
res = static_cast<T>(wsu - wt);
}
}
} break;
case SUBSUU_S: {
uT wsu = static_cast<uT>(ws);
uT wtu = static_cast<uT>(wt);
uT wdu;
if (wsu > wtu) {
wdu = wsu - wtu;
if (wdu > std::numeric_limits<T>::max()) {
res = std::numeric_limits<T>::max();
} else {
res = static_cast<T>(wdu);
}
} else {
wdu = wtu - wsu;
CHECK(-std::numeric_limits<T>::max() ==
std::numeric_limits<T>::min() + 1);
if (wdu <= std::numeric_limits<T>::max()) {
res = -static_cast<T>(wdu);
} else {
res = std::numeric_limits<T>::min();
}
}
} break;
case ASUB_S:
case ASUB_U:
res = static_cast<T>(Abs(ws - wt));
break;
case ASUB_U: {
uT wsu = static_cast<uT>(ws);
uT wtu = static_cast<uT>(wt);
res = static_cast<T>(wsu > wtu ? wsu - wtu : wtu - wsu);
} break;
case MULV:
res = ws * wt;
break;
case MADDV:
res = wd + ws * wt;
break;
case MSUBV:
res = wd - ws * wt;
break;
case DIV_S_MSA:
res = wt != 0 ? ws / wt : static_cast<T>(Unpredictable);
break;
case DIV_U:
res = wt != 0 ? static_cast<T>(static_cast<uT>(ws) / static_cast<uT>(wt))
: static_cast<T>(Unpredictable);
break;
case MOD_S:
res = wt != 0 ? ws % wt : static_cast<T>(Unpredictable);
break;
case MOD_U:
res = wt != 0 ? static_cast<T>(static_cast<uT>(ws) % static_cast<uT>(wt))
: static_cast<T>(Unpredictable);
break;
case DOTP_S:
case DOTP_U:
case DPADD_S:
......@@ -4959,8 +5119,17 @@ void Simulator::DecodeTypeMsa3R() {
case ILVEV:
case ILVOD:
case VSHF:
case SRAR:
case SRLR:
UNIMPLEMENTED();
break;
case SRAR: {
int bit = wt_modulo == 0 ? 0 : (ws >> (wt_modulo - 1)) & 1;
res = static_cast<T>(ArithmeticShiftRight(ws, wt_modulo) + bit);
} break;
case SRLR: {
uT wsu = static_cast<uT>(ws);
int bit = wt_modulo == 0 ? 0 : (wsu >> (wt_modulo - 1)) & 1;
res = static_cast<T>((wsu >> wt_modulo) + bit);
} break;
case HADD_S:
case HADD_U:
case HSUB_S:
......@@ -4970,9 +5139,45 @@ void Simulator::DecodeTypeMsa3R() {
default:
UNREACHABLE();
}
}
return res;
}
void Simulator::DecodeTypeMsa3RF() {
void Simulator::DecodeTypeMsa3R() {
DCHECK(IsMipsArchVariant(kMips32r6));
DCHECK(CpuFeatures::IsSupported(MIPS_SIMD));
uint32_t opcode = instr_.InstructionBits() & kMsa3RMask;
msa_reg_t ws, wd, wt;
#define MSA_3R_DF(elem, num_of_lanes) \
get_msa_register(instr_.WdValue(), wd.elem); \
get_msa_register(instr_.WsValue(), ws.elem); \
get_msa_register(instr_.WtValue(), wt.elem); \
for (int i = 0; i < num_of_lanes; i++) { \
wd.elem[i] = Msa3RInstrHelper(opcode, wd.elem[i], ws.elem[i], wt.elem[i]); \
} \
set_msa_register(instr_.WdValue(), wd.elem); \
TraceMSARegWr(wd.elem);
switch (DecodeMsaDataFormat()) {
case MSA_BYTE:
MSA_3R_DF(b, kMSALanesByte);
break;
case MSA_HALF:
MSA_3R_DF(h, kMSALanesHalf);
break;
case MSA_WORD:
MSA_3R_DF(w, kMSALanesWord);
break;
case MSA_DWORD:
MSA_3R_DF(d, kMSALanesDword);
break;
default:
UNREACHABLE();
}
#undef MSA_3R_DF
}
void Simulator::DecodeTypeMsa3RF() {
DCHECK(IsMipsArchVariant(kMips32r6));
DCHECK(CpuFeatures::IsSupported(MIPS_SIMD));
uint32_t opcode = instr_.InstructionBits() & kMsa3RFMask;
......@@ -5023,9 +5228,9 @@ void Simulator::DecodeTypeMsa3RF() {
default:
UNREACHABLE();
}
}
}
void Simulator::DecodeTypeMsaVec() {
void Simulator::DecodeTypeMsaVec() {
DCHECK(IsMipsArchVariant(kMips32r6));
DCHECK(CpuFeatures::IsSupported(MIPS_SIMD));
uint32_t opcode = instr_.InstructionBits() & kMsaVECMask;
......@@ -5066,9 +5271,9 @@ void Simulator::DecodeTypeMsaVec() {
}
set_msa_register(instr_.WdValue(), wd.w);
TraceMSARegWr(wd.d);
}
}
void Simulator::DecodeTypeMsa2R() {
void Simulator::DecodeTypeMsa2R() {
DCHECK(IsMipsArchVariant(kMips32r6));
DCHECK(CpuFeatures::IsSupported(MIPS_SIMD));
uint32_t opcode = instr_.InstructionBits() & kMsa2RMask;
......
src/mips/simulator-mips.h
View file @ 01e3be50
......@@ -436,6 +436,8 @@ class Simulator {
T MsaI5InstrHelper(uint32_t opcode, T ws, int32_t i5);
template <typename T>
T MsaBitInstrHelper(uint32_t opcode, T wd, T ws, int32_t m);
template <typename T>
T Msa3RInstrHelper(uint32_t opcode, T wd, T ws, T wt);
inline int32_t rs_reg() const { return instr_.RsValue(); }
inline int32_t rs() const { return get_register(rs_reg()); }
......
src/mips64/simulator-mips64.cc
View file @ 01e3be50
......@@ -5122,53 +5122,213 @@ void Simulator::DecodeTypeMsaMI10() {
#undef MSA_MI10_STORE
}
void Simulator::DecodeTypeMsa3R() {
DCHECK(kArchVariant == kMips64r6);
DCHECK(CpuFeatures::IsSupported(MIPS_SIMD));
uint32_t opcode = instr_.InstructionBits() & kMsa3RMask;
template <typename T>
T Simulator::Msa3RInstrHelper(uint32_t opcode, T wd, T ws, T wt) {
typedef typename std::make_unsigned<T>::type uT;
T res;
int wt_modulo = wt % (sizeof(T) * 8);
switch (opcode) {
case SLL_MSA:
res = static_cast<T>(ws << wt_modulo);
break;
case SRA_MSA:
res = static_cast<T>(ArithmeticShiftRight(ws, wt_modulo));
break;
case SRL_MSA:
res = static_cast<T>(static_cast<uT>(ws) >> wt_modulo);
break;
case BCLR:
res = static_cast<T>(static_cast<T>(~(1ull << wt_modulo)) & ws);
break;
case BSET:
res = static_cast<T>(static_cast<T>(1ull << wt_modulo) | ws);
break;
case BNEG:
case BINSL:
case BINSR:
res = static_cast<T>(static_cast<T>(1ull << wt_modulo) ^ ws);
break;
case BINSL: {
int elem_size = 8 * sizeof(T);
int bits = wt_modulo + 1;
if (bits == elem_size) {
res = static_cast<T>(ws);
} else {
uint64_t mask = ((1ull << bits) - 1) << (elem_size - bits);
res = static_cast<T>((static_cast<T>(mask) & ws) |
(static_cast<T>(~mask) & wd));
}
} break;
case BINSR: {
int elem_size = 8 * sizeof(T);
int bits = wt_modulo + 1;
if (bits == elem_size) {
res = static_cast<T>(ws);
} else {
uint64_t mask = (1ull << bits) - 1;
res = static_cast<T>((static_cast<T>(mask) & ws) |
(static_cast<T>(~mask) & wd));
}
} break;
case ADDV:
res = ws + wt;
break;
case SUBV:
res = ws - wt;
break;
case MAX_S:
res = Max(ws, wt);
break;
case MAX_U:
res = static_cast<T>(Max(static_cast<uT>(ws), static_cast<uT>(wt)));
break;
case MIN_S:
res = Min(ws, wt);
break;
case MIN_U:
res = static_cast<T>(Min(static_cast<uT>(ws), static_cast<uT>(wt)));
break;
case MAX_A:
// We use negative abs in order to avoid problems
// with corner case for MIN_INT
res = Nabs(ws) < Nabs(wt) ? ws : wt;
break;
case MIN_A:
// We use negative abs in order to avoid problems
// with corner case for MIN_INT
res = Nabs(ws) > Nabs(wt) ? ws : wt;
break;
case CEQ:
res = static_cast<T>(!Compare(ws, wt) ? -1ull : 0ull);
break;
case CLT_S:
res = static_cast<T>((Compare(ws, wt) == -1) ? -1ull : 0ull);
break;
case CLT_U:
res = static_cast<T>(
(Compare(static_cast<uT>(ws), static_cast<uT>(wt)) == -1) ? -1ull
: 0ull);
break;
case CLE_S:
res = static_cast<T>((Compare(ws, wt) != 1) ? -1ull : 0ull);
break;
case CLE_U:
res = static_cast<T>(
(Compare(static_cast<uT>(ws), static_cast<uT>(wt)) != 1) ? -1ull
: 0ull);
break;
case ADD_A:
case ADDS_A:
res = static_cast<T>(Abs(ws) + Abs(wt));
break;
case ADDS_A: {
T ws_nabs = Nabs(ws);
T wt_nabs = Nabs(wt);
if (ws_nabs < -std::numeric_limits<T>::max() - wt_nabs) {
res = std::numeric_limits<T>::max();
} else {
res = -(ws_nabs + wt_nabs);
}
} break;
case ADDS_S:
case ADDS_U:
res = SaturateAdd(ws, wt);
break;
case ADDS_U: {
uT ws_u = static_cast<uT>(ws);
uT wt_u = static_cast<uT>(wt);
res = static_cast<T>(SaturateAdd(ws_u, wt_u));
} break;
case AVE_S:
case AVE_U:
res = static_cast<T>((wt & ws) + ((wt ^ ws) >> 1));
break;
case AVE_U: {
uT ws_u = static_cast<uT>(ws);
uT wt_u = static_cast<uT>(wt);
res = static_cast<T>((wt_u & ws_u) + ((wt_u ^ ws_u) >> 1));
} break;
case AVER_S:
case AVER_U:
res = static_cast<T>((wt | ws) - ((wt ^ ws) >> 1));
break;
case AVER_U: {
uT ws_u = static_cast<uT>(ws);
uT wt_u = static_cast<uT>(wt);
res = static_cast<T>((wt_u | ws_u) - ((wt_u ^ ws_u) >> 1));
} break;
case SUBS_S:
case SUBS_U:
case SUBSUS_U:
case SUBSUU_S:
res = SaturateSub(ws, wt);
break;
case SUBS_U: {
uT ws_u = static_cast<uT>(ws);
uT wt_u = static_cast<uT>(wt);
res = static_cast<T>(SaturateSub(ws_u, wt_u));
} break;
case SUBSUS_U: {
uT wsu = static_cast<uT>(ws);
if (wt > 0) {
uT wtu = static_cast<uT>(wt);
if (wtu > wsu) {
res = 0;
} else {
res = static_cast<T>(wsu - wtu);
}
} else {
if (wsu > std::numeric_limits<uT>::max() + wt) {
res = static_cast<T>(std::numeric_limits<uT>::max());
} else {
res = static_cast<T>(wsu - wt);
}
}
} break;
case SUBSUU_S: {
uT wsu = static_cast<uT>(ws);
uT wtu = static_cast<uT>(wt);
uT wdu;
if (wsu > wtu) {
wdu = wsu - wtu;
if (wdu > std::numeric_limits<T>::max()) {
res = std::numeric_limits<T>::max();
} else {
res = static_cast<T>(wdu);
}
} else {
wdu = wtu - wsu;
CHECK(-std::numeric_limits<T>::max() ==
std::numeric_limits<T>::min() + 1);
if (wdu <= std::numeric_limits<T>::max()) {
res = -static_cast<T>(wdu);
} else {
res = std::numeric_limits<T>::min();
}
}
} break;
case ASUB_S:
case ASUB_U:
res = static_cast<T>(Abs(ws - wt));
break;
case ASUB_U: {
uT wsu = static_cast<uT>(ws);
uT wtu = static_cast<uT>(wt);
res = static_cast<T>(wsu > wtu ? wsu - wtu : wtu - wsu);
} break;
case MULV:
res = ws * wt;
break;
case MADDV:
res = wd + ws * wt;
break;
case MSUBV:
res = wd - ws * wt;
break;
case DIV_S_MSA:
res = wt != 0 ? ws / wt : static_cast<T>(Unpredictable);
break;
case DIV_U:
res = wt != 0 ? static_cast<T>(static_cast<uT>(ws) / static_cast<uT>(wt))
: static_cast<T>(Unpredictable);
break;
case MOD_S:
res = wt != 0 ? ws % wt : static_cast<T>(Unpredictable);
break;
case MOD_U:
res = wt != 0 ? static_cast<T>(static_cast<uT>(ws) % static_cast<uT>(wt))
: static_cast<T>(Unpredictable);
break;
case DOTP_S:
case DOTP_U:
case DPADD_S:
......@@ -5184,8 +5344,17 @@ void Simulator::DecodeTypeMsa3R() {
case ILVEV:
case ILVOD:
case VSHF:
case SRAR:
case SRLR:
UNIMPLEMENTED();
break;
case SRAR: {
int bit = wt_modulo == 0 ? 0 : (ws >> (wt_modulo - 1)) & 1;
res = static_cast<T>(ArithmeticShiftRight(ws, wt_modulo) + bit);
} break;
case SRLR: {
uT wsu = static_cast<uT>(ws);
int bit = wt_modulo == 0 ? 0 : (wsu >> (wt_modulo - 1)) & 1;
res = static_cast<T>((wsu >> wt_modulo) + bit);
} break;
case HADD_S:
case HADD_U:
case HSUB_S:
......@@ -5195,6 +5364,42 @@ void Simulator::DecodeTypeMsa3R() {
default:
UNREACHABLE();
}
return res;
}
void Simulator::DecodeTypeMsa3R() {
DCHECK(kArchVariant == kMips64r6);
DCHECK(CpuFeatures::IsSupported(MIPS_SIMD));
uint32_t opcode = instr_.InstructionBits() & kMsa3RMask;
msa_reg_t ws, wd, wt;
#define MSA_3R_DF(elem, num_of_lanes) \
get_msa_register(instr_.WdValue(), wd.elem); \
get_msa_register(instr_.WsValue(), ws.elem); \
get_msa_register(instr_.WtValue(), wt.elem); \
for (int i = 0; i < num_of_lanes; i++) { \
wd.elem[i] = Msa3RInstrHelper(opcode, wd.elem[i], ws.elem[i], wt.elem[i]); \
} \
set_msa_register(instr_.WdValue(), wd.elem); \
TraceMSARegWr(wd.elem);
switch (DecodeMsaDataFormat()) {
case MSA_BYTE:
MSA_3R_DF(b, kMSALanesByte);
break;
case MSA_HALF:
MSA_3R_DF(h, kMSALanesHalf);
break;
case MSA_WORD:
MSA_3R_DF(w, kMSALanesWord);
break;
case MSA_DWORD:
MSA_3R_DF(d, kMSALanesDword);
break;
default:
UNREACHABLE();
}
#undef MSA_3R_DF
}
void Simulator::DecodeTypeMsa3RF() {
......
src/mips64/simulator-mips64.h
View file @ 01e3be50
......@@ -452,6 +452,8 @@ class Simulator {
T MsaI5InstrHelper(uint32_t opcode, T ws, int32_t i5);
template <typename T>
T MsaBitInstrHelper(uint32_t opcode, T wd, T ws, int32_t m);
template <typename T>
T Msa3RInstrHelper(uint32_t opcode, T wd, T ws, T wt);
// Executing is handled based on the instruction type.
void DecodeTypeRegister();
......
src/utils.h
View file @ 01e3be50
......@@ -196,6 +196,13 @@ typename std::make_unsigned<T>::type Abs(T a) {
return (x ^ y) - y;
}
// Returns the negative absolute value of its argument.
template <typename T,
typename = typename std::enable_if<std::is_signed<T>::value>::type>
T Nabs(T a) {
return a < 0 ? a : -a;
}
// Floor(-0.0) == 0.0
inline double Floor(double x) {
#if V8_CC_MSVC
......@@ -233,6 +240,47 @@ inline double Pow(double x, double y) {
return std::pow(x, y);
}
template <typename T>
T SaturateAdd(T a, T b) {
if (std::is_signed<T>::value) {
if (a > 0 && b > 0) {
if (a > std::numeric_limits<T>::max() - b) {
return std::numeric_limits<T>::max();
}
} else if (a < 0 && b < 0) {
if (a < std::numeric_limits<T>::min() - b) {
return std::numeric_limits<T>::min();
}
}
} else {
CHECK(std::is_unsigned<T>::value);
if (a > std::numeric_limits<T>::max() - b) {
return std::numeric_limits<T>::max();
}
}
return a + b;
}
template <typename T>
T SaturateSub(T a, T b) {
if (std::is_signed<T>::value) {
if (a > 0 && b < 0) {
if (a > std::numeric_limits<T>::max() + b) {
return std::numeric_limits<T>::max();
}
} else if (a < 0 && b > 0) {
if (a < std::numeric_limits<T>::min() + b) {
return std::numeric_limits<T>::min();
}
}
} else {
CHECK(std::is_unsigned<T>::value);
if (a < b) {
return static_cast<T>(0);
}
}
return a - b;
}
// ----------------------------------------------------------------------------
// BitField is a help template for encoding and decode bitfield with
......
test/cctest/test-assembler-mips.cc
View file @ 01e3be50
This diff is collapsed.
test/cctest/test-assembler-mips64.cc
View file @ 01e3be50
This diff is collapsed.
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment