mirrors/dynarmic
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

TranlateArm: implement remaining multiplies

Browse source 
SMLALxy, SMLAxy, SMULxy SMLAWy, SMULWy, SMLAD, SMLALD, SMLSD, SMLSLD,
SMUAD, SMUSD
This commit is contained in:
Tillmann Karras 2016-08-17 14:35:26 +01:00 committed by MerryMage
parent f12578b9ab
commit dad7724b86
2 changed files with 232 additions and 25 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

206
src/frontend/translate/translate_arm/multiply.cpp
View file

@ -147,25 +147,94 @@ bool ArmTranslatorVisitor::arm_UMULL(Cond cond, bool S, Reg dHi, Reg dLo, Reg m,
// Multiply (Halfword) instructions // Multiply (Halfword) instructions
bool ArmTranslatorVisitor::arm_SMLALxy(Cond cond, Reg dHi, Reg dLo, Reg m, bool M, bool N, Reg n) { bool ArmTranslatorVisitor::arm_SMLALxy(Cond cond, Reg dHi, Reg dLo, Reg m, bool M, bool N, Reg n) {
return InterpretThisInstruction(); if (dLo == Reg::PC || dHi == Reg::PC || n == Reg::PC || m == Reg::PC)
return UnpredictableInstruction();
if (dLo == dHi)
return UnpredictableInstruction();
if (ConditionPassed(cond)) {
auto n32 = ir.GetRegister(n);
auto m32 = ir.GetRegister(m);
auto n16 = N ? ir.ArithmeticShiftRight(n32, ir.Imm8(16), ir.Imm1(0)).result
: ir.SignExtendHalfToWord(ir.LeastSignificantHalf(n32));
auto m16 = M ? ir.ArithmeticShiftRight(m32, ir.Imm8(16), ir.Imm1(0)).result
: ir.SignExtendHalfToWord(ir.LeastSignificantHalf(m32));
auto product = ir.SignExtendWordToLong(ir.Mul(n16, m16));
auto addend = ir.Pack2x32To1x64(ir.GetRegister(dLo), ir.GetRegister(dHi));
auto result = ir.Add64(product, addend);
ir.SetRegister(dLo, ir.LeastSignificantWord(result));
ir.SetRegister(dHi, ir.MostSignificantWord(result).result);
}
return true;
} }
bool ArmTranslatorVisitor::arm_SMLAxy(Cond cond, Reg d, Reg a, Reg m, bool M, bool N, Reg n) { bool ArmTranslatorVisitor::arm_SMLAxy(Cond cond, Reg d, Reg a, Reg m, bool M, bool N, Reg n) {
return InterpretThisInstruction(); if (d == Reg::PC || n == Reg::PC || m == Reg::PC || a == Reg::PC)
return UnpredictableInstruction();
if (ConditionPassed(cond)) {
auto n32 = ir.GetRegister(n);
auto m32 = ir.GetRegister(m);
auto n16 = N ? ir.ArithmeticShiftRight(n32, ir.Imm8(16), ir.Imm1(0)).result
: ir.SignExtendHalfToWord(ir.LeastSignificantHalf(n32));
auto m16 = M ? ir.ArithmeticShiftRight(m32, ir.Imm8(16), ir.Imm1(0)).result
: ir.SignExtendHalfToWord(ir.LeastSignificantHalf(m32));
auto product = ir.Mul(n16, m16);
auto result_overflow = ir.AddWithCarry(product, ir.GetRegister(a), ir.Imm1(0));
ir.SetRegister(d, result_overflow.result);
ir.OrQFlag(result_overflow.overflow);
}
return true;
} }
bool ArmTranslatorVisitor::arm_SMULxy(Cond cond, Reg d, Reg m, bool M, bool N, Reg n) { bool ArmTranslatorVisitor::arm_SMULxy(Cond cond, Reg d, Reg m, bool M, bool N, Reg n) {
return InterpretThisInstruction(); if (d == Reg::PC || n == Reg::PC || m == Reg::PC)
return UnpredictableInstruction();
if (ConditionPassed(cond)) {
auto n32 = ir.GetRegister(n);
auto m32 = ir.GetRegister(m);
auto n16 = N ? ir.ArithmeticShiftRight(n32, ir.Imm8(16), ir.Imm1(0)).result
: ir.SignExtendHalfToWord(ir.LeastSignificantHalf(n32));
auto m16 = M ? ir.ArithmeticShiftRight(m32, ir.Imm8(16), ir.Imm1(0)).result
: ir.SignExtendHalfToWord(ir.LeastSignificantHalf(m32));
auto result = ir.Mul(n16, m16);
ir.SetRegister(d, result);
}
return true;
} }
// Multiply (word by halfword) instructions // Multiply (word by halfword) instructions
bool ArmTranslatorVisitor::arm_SMLAWy(Cond cond, Reg d, Reg a, Reg m, bool M, Reg n) { bool ArmTranslatorVisitor::arm_SMLAWy(Cond cond, Reg d, Reg a, Reg m, bool M, Reg n) {
return InterpretThisInstruction(); if (d == Reg::PC || n == Reg::PC || m == Reg::PC || a == Reg::PC)
return UnpredictableInstruction();
if (ConditionPassed(cond)) {
auto n32 = ir.SignExtendWordToLong(ir.GetRegister(n));
auto m32 = ir.GetRegister(m);
if (M)
m32 = ir.LogicalShiftRight(m32, ir.Imm8(16), ir.Imm1(0)).result;
auto m16 = ir.LeastSignificantHalf(m32);
m16 = ir.SignExtendWordToLong(ir.SignExtendHalfToWord(m16));
auto product = ir.LeastSignificantWord(ir.LogicalShiftRight64(ir.Mul64(n32, m16), ir.Imm8(16)));
auto result_overflow = ir.AddWithCarry(product, ir.GetRegister(a), ir.Imm1(0));
ir.SetRegister(d, result_overflow.result);
ir.OrQFlag(result_overflow.overflow);
}
return true;
} }
bool ArmTranslatorVisitor::arm_SMULWy(Cond cond, Reg d, Reg m, bool M, Reg n) { bool ArmTranslatorVisitor::arm_SMULWy(Cond cond, Reg d, Reg m, bool M, Reg n) {
return InterpretThisInstruction(); if (d == Reg::PC || n == Reg::PC || m == Reg::PC)
return UnpredictableInstruction();
if (ConditionPassed(cond)) {
auto n32 = ir.SignExtendWordToLong(ir.GetRegister(n));
auto m32 = ir.GetRegister(m);
if (M)
m32 = ir.LogicalShiftRight(m32, ir.Imm8(16), ir.Imm1(0)).result;
auto m16 = ir.LeastSignificantHalf(m32);
m16 = ir.SignExtendWordToLong(ir.SignExtendHalfToWord(m16));
auto result = ir.LogicalShiftRight64(ir.Mul64(n32, m16), ir.Imm8(16));
ir.SetRegister(d, ir.LeastSignificantWord(result));
}
return true;
} }
@ -223,27 +292,142 @@ bool ArmTranslatorVisitor::arm_SMMUL(Cond cond, Reg d, Reg m, bool R, Reg n) {
// Multiply (Dual) instructions // Multiply (Dual) instructions
bool ArmTranslatorVisitor::arm_SMLAD(Cond cond, Reg d, Reg a, Reg m, bool M, Reg n) { bool ArmTranslatorVisitor::arm_SMLAD(Cond cond, Reg d, Reg a, Reg m, bool M, Reg n) {
return InterpretThisInstruction(); if (a == Reg::PC)
return arm_SMUAD(cond, d, m, M, n);
if (d == Reg::PC || n == Reg::PC || m == Reg::PC)
return UnpredictableInstruction();
if (ConditionPassed(cond)) {
auto n32 = ir.GetRegister(n);
auto m32 = ir.GetRegister(m);
auto n_lo = ir.SignExtendHalfToWord(ir.LeastSignificantHalf(n32));
auto m_lo = ir.SignExtendHalfToWord(ir.LeastSignificantHalf(m32));
auto n_hi = ir.ArithmeticShiftRight(n32, ir.Imm8(16), ir.Imm1(0)).result;
auto m_hi = ir.ArithmeticShiftRight(m32, ir.Imm8(16), ir.Imm1(0)).result;
if (M)
std::swap(m_lo, m_hi);
auto product_lo = ir.Mul(n_lo, m_lo);
auto product_hi = ir.Mul(n_hi, m_hi);
auto addend = ir.GetRegister(a);
auto result_overflow = ir.AddWithCarry(product_lo, product_hi, ir.Imm1(0));
ir.OrQFlag(result_overflow.overflow);
result_overflow = ir.AddWithCarry(result_overflow.result, addend, ir.Imm1(0));
ir.SetRegister(d, result_overflow.result);
ir.OrQFlag(result_overflow.overflow);
}
return true;
} }
bool ArmTranslatorVisitor::arm_SMLALD(Cond cond, Reg dHi, Reg dLo, Reg m, bool M, Reg n) { bool ArmTranslatorVisitor::arm_SMLALD(Cond cond, Reg dHi, Reg dLo, Reg m, bool M, Reg n) {
return InterpretThisInstruction(); if (dLo == Reg::PC || dHi == Reg::PC || n == Reg::PC || m == Reg::PC)
return UnpredictableInstruction();
if (dLo == dHi)
return UnpredictableInstruction();
if (ConditionPassed(cond)) {
auto n32 = ir.GetRegister(n);
auto m32 = ir.GetRegister(m);
auto n_lo = ir.SignExtendHalfToWord(ir.LeastSignificantHalf(n32));
auto m_lo = ir.SignExtendHalfToWord(ir.LeastSignificantHalf(m32));
auto n_hi = ir.ArithmeticShiftRight(n32, ir.Imm8(16), ir.Imm1(0)).result;
auto m_hi = ir.ArithmeticShiftRight(m32, ir.Imm8(16), ir.Imm1(0)).result;
if (M)
std::swap(m_lo, m_hi);
auto product_lo = ir.SignExtendWordToLong(ir.Mul(n_lo, m_lo));
auto product_hi = ir.SignExtendWordToLong(ir.Mul(n_hi, m_hi));
auto addend = ir.Pack2x32To1x64(ir.GetRegister(dLo), ir.GetRegister(dHi));
auto result = ir.Add64(ir.Add64(product_lo, product_hi), addend);
ir.SetRegister(dLo, ir.LeastSignificantWord(result));
ir.SetRegister(dHi, ir.MostSignificantWord(result).result);
}
return true;
} }
bool ArmTranslatorVisitor::arm_SMLSD(Cond cond, Reg d, Reg a, Reg m, bool M, Reg n) { bool ArmTranslatorVisitor::arm_SMLSD(Cond cond, Reg d, Reg a, Reg m, bool M, Reg n) {
return InterpretThisInstruction(); if (a == Reg::PC)
return arm_SMUSD(cond, d, m, M, n);
if (d == Reg::PC || n == Reg::PC || m == Reg::PC)
return UnpredictableInstruction();
if (ConditionPassed(cond)) {
auto n32 = ir.GetRegister(n);
auto m32 = ir.GetRegister(m);
auto n_lo = ir.SignExtendHalfToWord(ir.LeastSignificantHalf(n32));
auto m_lo = ir.SignExtendHalfToWord(ir.LeastSignificantHalf(m32));
auto n_hi = ir.ArithmeticShiftRight(n32, ir.Imm8(16), ir.Imm1(0)).result;
auto m_hi = ir.ArithmeticShiftRight(m32, ir.Imm8(16), ir.Imm1(0)).result;
if (M)
std::swap(m_lo, m_hi);
auto product_lo = ir.Mul(n_lo, m_lo);
auto product_hi = ir.Mul(n_hi, m_hi);
auto addend = ir.GetRegister(a);
auto result_overflow = ir.AddWithCarry(ir.Sub(product_lo, product_hi), addend, ir.Imm1(0));
ir.SetRegister(d, result_overflow.result);
ir.OrQFlag(result_overflow.overflow);
}
return true;
} }
bool ArmTranslatorVisitor::arm_SMLSLD(Cond cond, Reg dHi, Reg dLo, Reg m, bool M, Reg n) { bool ArmTranslatorVisitor::arm_SMLSLD(Cond cond, Reg dHi, Reg dLo, Reg m, bool M, Reg n) {
return InterpretThisInstruction(); if (dLo == Reg::PC || dHi == Reg::PC || n == Reg::PC || m == Reg::PC)
return UnpredictableInstruction();
if (dLo == dHi)
return UnpredictableInstruction();
if (ConditionPassed(cond)) {
auto n32 = ir.GetRegister(n);
auto m32 = ir.GetRegister(m);
auto n_lo = ir.SignExtendHalfToWord(ir.LeastSignificantHalf(n32));
auto m_lo = ir.SignExtendHalfToWord(ir.LeastSignificantHalf(m32));
auto n_hi = ir.ArithmeticShiftRight(n32, ir.Imm8(16), ir.Imm1(0)).result;
auto m_hi = ir.ArithmeticShiftRight(m32, ir.Imm8(16), ir.Imm1(0)).result;
if (M)
std::swap(m_lo, m_hi);
auto product_lo = ir.SignExtendWordToLong(ir.Mul(n_lo, m_lo));
auto product_hi = ir.SignExtendWordToLong(ir.Mul(n_hi, m_hi));
auto addend = ir.Pack2x32To1x64(ir.GetRegister(dLo), ir.GetRegister(dHi));
auto result = ir.Add64(ir.Sub64(product_lo, product_hi), addend);
ir.SetRegister(dLo, ir.LeastSignificantWord(result));
ir.SetRegister(dHi, ir.MostSignificantWord(result).result);
}
return true;
} }
bool ArmTranslatorVisitor::arm_SMUAD(Cond cond, Reg d, Reg m, bool M, Reg n) { bool ArmTranslatorVisitor::arm_SMUAD(Cond cond, Reg d, Reg m, bool M, Reg n) {
return InterpretThisInstruction(); if (d == Reg::PC || n == Reg::PC || m == Reg::PC)
return UnpredictableInstruction();
if (ConditionPassed(cond)) {
auto n32 = ir.GetRegister(n);
auto m32 = ir.GetRegister(m);
auto n_lo = ir.SignExtendHalfToWord(ir.LeastSignificantHalf(n32));
auto m_lo = ir.SignExtendHalfToWord(ir.LeastSignificantHalf(m32));
auto n_hi = ir.ArithmeticShiftRight(n32, ir.Imm8(16), ir.Imm1(0)).result;
auto m_hi = ir.ArithmeticShiftRight(m32, ir.Imm8(16), ir.Imm1(0)).result;
if (M)
std::swap(m_lo, m_hi);
auto product_lo = ir.Mul(n_lo, m_lo);
auto product_hi = ir.Mul(n_hi, m_hi);
auto result_overflow = ir.AddWithCarry(product_lo, product_hi, ir.Imm1(0));
ir.SetRegister(d, result_overflow.result);
ir.OrQFlag(result_overflow.overflow);
}
return true;
} }
bool ArmTranslatorVisitor::arm_SMUSD(Cond cond, Reg d, Reg m, bool M, Reg n) { bool ArmTranslatorVisitor::arm_SMUSD(Cond cond, Reg d, Reg m, bool M, Reg n) {
return InterpretThisInstruction(); if (d == Reg::PC || n == Reg::PC || m == Reg::PC)
return UnpredictableInstruction();
if (ConditionPassed(cond)) {
auto n32 = ir.GetRegister(n);
auto m32 = ir.GetRegister(m);
auto n_lo = ir.SignExtendHalfToWord(ir.LeastSignificantHalf(n32));
auto m_lo = ir.SignExtendHalfToWord(ir.LeastSignificantHalf(m32));
auto n_hi = ir.ArithmeticShiftRight(n32, ir.Imm8(16), ir.Imm1(0)).result;
auto m_hi = ir.ArithmeticShiftRight(m32, ir.Imm8(16), ir.Imm1(0)).result;
if (M)
std::swap(m_lo, m_hi);
auto product_lo = ir.Mul(n_lo, m_lo);
auto product_hi = ir.Mul(n_hi, m_hi);
auto result = ir.Sub(product_lo, product_hi);
ir.SetRegister(d, result);
}
return true;
} }
} // namespace Arm } // namespace Arm

51
tests/arm/fuzz_arm.cpp
View file

@ -792,7 +792,7 @@ TEST_CASE("Fuzz ARM multiply instructions", "[JitX64]") {
Bits<12, 15>(inst) != Bits<16, 19>(inst); Bits<12, 15>(inst) != Bits<16, 19>(inst);
}; };
const std::array<InstructionGenerator, 10> instructions = {{ const std::array<InstructionGenerator, 21> instructions = {{
InstructionGenerator("cccc0000001Sddddaaaammmm1001nnnn", validate_d_a_m_n), // MLA InstructionGenerator("cccc0000001Sddddaaaammmm1001nnnn", validate_d_a_m_n), // MLA
InstructionGenerator("cccc0000000Sdddd0000mmmm1001nnnn", validate_d_m_n), // MUL InstructionGenerator("cccc0000000Sdddd0000mmmm1001nnnn", validate_d_m_n), // MUL
@ -802,27 +802,26 @@ TEST_CASE("Fuzz ARM multiply instructions", "[JitX64]") {
InstructionGenerator("cccc0000101Sddddaaaammmm1001nnnn", validate_h_l_m_n), // UMLAL InstructionGenerator("cccc0000101Sddddaaaammmm1001nnnn", validate_h_l_m_n), // UMLAL
InstructionGenerator("cccc0000100Sddddaaaammmm1001nnnn", validate_h_l_m_n), // UMULL InstructionGenerator("cccc0000100Sddddaaaammmm1001nnnn", validate_h_l_m_n), // UMULL
//InstructionGenerator("cccc00010100ddddaaaammmm1xy0nnnn", validate_d_a_m_n), // SMLALxy InstructionGenerator("cccc00010100ddddaaaammmm1xy0nnnn", validate_h_l_m_n), // SMLALxy
//InstructionGenerator("cccc00010000ddddaaaammmm1xy0nnnn", validate_d_a_m_n), // SMLAxy InstructionGenerator("cccc00010000ddddaaaammmm1xy0nnnn", validate_d_a_m_n), // SMLAxy
//InstructionGenerator("cccc00010110dddd0000mmmm1xy0nnnn", validate_d_m_n), // SMULxy InstructionGenerator("cccc00010110dddd0000mmmm1xy0nnnn", validate_d_m_n), // SMULxy
//InstructionGenerator("cccc00010010ddddaaaammmm1y00nnnn", validate_d_a_m_n), // SMLAWy InstructionGenerator("cccc00010010ddddaaaammmm1y00nnnn", validate_d_a_m_n), // SMLAWy
//InstructionGenerator("cccc00010010dddd0000mmmm1y10nnnn", validate_d_m_n), // SMULWy InstructionGenerator("cccc00010010dddd0000mmmm1y10nnnn", validate_d_m_n), // SMULWy
InstructionGenerator("cccc01110101dddd1111mmmm00R1nnnn", validate_d_m_n), // SMMUL InstructionGenerator("cccc01110101dddd1111mmmm00R1nnnn", validate_d_m_n), // SMMUL
InstructionGenerator("cccc01110101ddddaaaammmm00R1nnnn", validate_d_a_m_n), // SMMLA InstructionGenerator("cccc01110101ddddaaaammmm00R1nnnn", validate_d_a_m_n), // SMMLA
InstructionGenerator("cccc01110101ddddaaaammmm11R1nnnn", validate_d_a_m_n), // SMMLS InstructionGenerator("cccc01110101ddddaaaammmm11R1nnnn", validate_d_a_m_n), // SMMLS
InstructionGenerator("cccc01110000ddddaaaammmm00M1nnnn", validate_d_a_m_n), // SMLAD
//InstructionGenerator("cccc01110000ddddaaaammmm00M1nnnn", validate_d_a_m_n), // SMLAD InstructionGenerator("cccc01110100ddddaaaammmm00M1nnnn", validate_h_l_m_n), // SMLALD
//InstructionGenerator("cccc01110100ddddaaaammmm00M1nnnn", validate_d_a_m_n), // SMLALD InstructionGenerator("cccc01110000ddddaaaammmm01M1nnnn", validate_d_a_m_n), // SMLSD
//InstructionGenerator("cccc01110000ddddaaaammmm01M1nnnn", validate_d_a_m_n), // SMLSD InstructionGenerator("cccc01110100ddddaaaammmm01M1nnnn", validate_h_l_m_n), // SMLSLD
//InstructionGenerator("cccc01110100ddddaaaammmm01M1nnnn", validate_d_a_m_n), // SMLSLD InstructionGenerator("cccc01110000dddd1111mmmm00M1nnnn", validate_d_m_n), // SMUAD
//InstructionGenerator("cccc01110000dddd1111mmmm00M1nnnn", validate_d_m_n), // SMUAD InstructionGenerator("cccc01110000dddd1111mmmm01M1nnnn", validate_d_m_n), // SMUSD
//InstructionGenerator("cccc01110000dddd1111mmmm01M1nnnn", validate_d_m_n), // SMUSD
}}; }};
SECTION("Multiply") { SECTION("Multiply") {
FuzzJitArm(2, 2, 10000, [&]() -> u32 { FuzzJitArm(1, 1, 10000, [&]() -> u32 {
return instructions[RandInt<size_t>(0, instructions.size() - 1)].Generate(); return instructions[RandInt<size_t>(0, instructions.size() - 1)].Generate();
}); });
} }
@ -852,6 +851,30 @@ TEST_CASE("Fuzz ARM parallel instructions", "[JitX64]") {
} }
} }
TEST_CASE( "SMUAD", "[JitX64]" ) {
Dynarmic::Jit jit{GetUserCallbacks()};
code_mem.fill({});
code_mem[0] = 0xE700F211; // smuad r0, r1, r2
jit.Regs() = {
0, // Rd
0x80008000, // Rn
0x80008000, // Rm
0,
0, 0, 0, 0,
0, 0, 0, 0,
0, 0, 0, 0,
};
jit.Cpsr() = 0x000001d0; // User-mode
jit.Run(6);
REQUIRE(jit.Regs()[0] == 0x80000000);
REQUIRE(jit.Regs()[1] == 0x80008000);
REQUIRE(jit.Regs()[2] == 0x80008000);
REQUIRE(jit.Cpsr() == 0x080001d0);
}
TEST_CASE("VFP: VPUSH, VPOP", "[JitX64][vfp]") { TEST_CASE("VFP: VPUSH, VPOP", "[JitX64][vfp]") {
const auto is_valid = [](u32 instr) -> bool { const auto is_valid = [](u32 instr) -> bool {
auto regs = (instr & 0x100) ? (Bits<0, 7>(instr) >> 1) : Bits<0, 7>(instr); auto regs = (instr & 0x100) ? (Bits<0, 7>(instr) >> 1) : Bits<0, 7>(instr);