dynarmic/src/backend/x64/emit_x64_vector_floating_point.cpp

1477 lines
54 KiB
C++

/* This file is part of the dynarmic project.
* Copyright (c) 2016 MerryMage
* This software may be used and distributed according to the terms of the GNU
* General Public License version 2 or any later version.
*/
#include <array>
#include <limits>
#include <tuple>
#include <type_traits>
#include <utility>
#include "backend/x64/abi.h"
#include "backend/x64/block_of_code.h"
#include "backend/x64/emit_x64.h"
#include "common/assert.h"
#include "common/bit_util.h"
#include "common/fp/fpcr.h"
#include "common/fp/info.h"
#include "common/fp/op.h"
#include "common/fp/util.h"
#include "common/mp/cartesian_product.h"
#include "common/mp/function_info.h"
#include "common/mp/integer.h"
#include "common/mp/list.h"
#include "common/mp/lut.h"
#include "common/mp/to_tuple.h"
#include "common/mp/vlift.h"
#include "common/mp/vllift.h"
#include "frontend/ir/basic_block.h"
#include "frontend/ir/microinstruction.h"
namespace Dynarmic::BackendX64 {
using namespace Xbyak::util;
namespace mp = Common::mp;
namespace {
template<size_t fsize, typename T>
T ChooseOnFsize([[maybe_unused]] T f32, [[maybe_unused]] T f64) {
static_assert(fsize == 32 || fsize == 64, "fsize must be either 32 or 64");
if constexpr (fsize == 32) {
return f32;
} else {
return f64;
}
}
#define FCODE(NAME) (code.*ChooseOnFsize<fsize>(&Xbyak::CodeGenerator::NAME##s, &Xbyak::CodeGenerator::NAME##d))
template<size_t fsize, template<typename> class Indexer, size_t narg>
struct NaNHandler {
public:
using FPT = mp::unsigned_integer_of_size<fsize>;
using function_type = void(*)(std::array<VectorArray<FPT>, narg>&, FP::FPCR);
static function_type GetDefault() {
return GetDefaultImpl(std::make_index_sequence<narg - 1>{});
}
private:
template<size_t... argi>
static function_type GetDefaultImpl(std::index_sequence<argi...>) {
const auto result = [](std::array<VectorArray<FPT>, narg>& values, FP::FPCR) {
VectorArray<FPT>& result = values[0];
for (size_t elementi = 0; elementi < result.size(); ++elementi) {
const auto current_values = Indexer<FPT>{}(elementi, values[argi + 1]...);
if (auto r = FP::ProcessNaNs(std::get<argi>(current_values)...)) {
result[elementi] = *r;
} else if (FP::IsNaN(result[elementi])) {
result[elementi] = FP::FPInfo<FPT>::DefaultNaN();
}
}
};
return static_cast<function_type>(result);
}
};
template<size_t fsize, size_t nargs, typename NaNHandler>
void HandleNaNs(BlockOfCode& code, EmitContext& ctx, std::array<Xbyak::Xmm, nargs + 1> xmms, const Xbyak::Xmm& nan_mask, NaNHandler nan_handler) {
static_assert(fsize == 32 || fsize == 64, "fsize must be either 32 or 64");
if (code.DoesCpuSupport(Xbyak::util::Cpu::tSSE41)) {
code.ptest(nan_mask, nan_mask);
} else {
const Xbyak::Reg32 bitmask = ctx.reg_alloc.ScratchGpr().cvt32();
code.movmskps(bitmask, nan_mask);
code.cmp(bitmask, 0);
}
Xbyak::Label end;
Xbyak::Label nan;
code.jnz(nan, code.T_NEAR);
code.L(end);
code.SwitchToFarCode();
code.L(nan);
const Xbyak::Xmm result = xmms[0];
code.sub(rsp, 8);
ABI_PushCallerSaveRegistersAndAdjustStackExcept(code, HostLocXmmIdx(result.getIdx()));
const size_t stack_space = xmms.size() * 16;
code.sub(rsp, stack_space + ABI_SHADOW_SPACE);
for (size_t i = 0; i < xmms.size(); ++i) {
code.movaps(xword[rsp + ABI_SHADOW_SPACE + i * 16], xmms[i]);
}
code.lea(code.ABI_PARAM1, ptr[rsp + ABI_SHADOW_SPACE + 0 * 16]);
code.mov(code.ABI_PARAM2, ctx.FPCR().Value());
code.CallFunction(nan_handler);
code.movaps(result, xword[rsp + ABI_SHADOW_SPACE + 0 * 16]);
code.add(rsp, stack_space + ABI_SHADOW_SPACE);
ABI_PopCallerSaveRegistersAndAdjustStackExcept(code, HostLocXmmIdx(result.getIdx()));
code.add(rsp, 8);
code.jmp(end, code.T_NEAR);
code.SwitchToNearCode();
}
template<size_t fsize>
Xbyak::Address GetVectorOf(BlockOfCode& code, u64 value) {
if constexpr (fsize == 32) {
return code.MConst(xword, (value << 32) | value, (value << 32) | value);
} else {
return code.MConst(xword, value, value);
}
}
template<size_t fsize, u64 value>
Xbyak::Address GetVectorOf(BlockOfCode& code) {
if constexpr (fsize == 32) {
return code.MConst(xword, (value << 32) | value, (value << 32) | value);
} else {
return code.MConst(xword, value, value);
}
}
template<size_t fsize>
Xbyak::Address GetNaNVector(BlockOfCode& code) {
using FPT = mp::unsigned_integer_of_size<fsize>;
return GetVectorOf<fsize, FP::FPInfo<FPT>::DefaultNaN()>(code);
}
template<size_t fsize>
Xbyak::Address GetNegativeZeroVector(BlockOfCode& code) {
using FPT = mp::unsigned_integer_of_size<fsize>;
return GetVectorOf<fsize, FP::FPInfo<FPT>::Zero(true)>(code);
}
template<size_t fsize>
Xbyak::Address GetSmallestNormalVector(BlockOfCode& code) {
using FPT = mp::unsigned_integer_of_size<fsize>;
constexpr FPT smallest_normal_number = FP::FPValue<FPT, false, FP::FPInfo<FPT>::exponent_min, 1>();
return GetVectorOf<fsize, smallest_normal_number>(code);
}
template<size_t fsize, bool sign, int exponent, mp::unsigned_integer_of_size<fsize> value>
Xbyak::Address GetVectorOf(BlockOfCode& code) {
using FPT = mp::unsigned_integer_of_size<fsize>;
return GetVectorOf<fsize, FP::FPValue<FPT, sign, exponent, value>()>(code);
}
template<size_t fsize>
void ForceToDefaultNaN(BlockOfCode& code, EmitContext& ctx, Xbyak::Xmm result) {
if (ctx.FPCR().DN()) {
const Xbyak::Xmm nan_mask = xmm0;
if (code.DoesCpuSupport(Xbyak::util::Cpu::tAVX)) {
FCODE(vcmpunordp)(nan_mask, result, result);
FCODE(blendvp)(result, GetNaNVector<fsize>(code));
} else {
code.movaps(nan_mask, result);
FCODE(cmpordp)(nan_mask, nan_mask);
code.andps(result, nan_mask);
code.andnps(nan_mask, GetNaNVector<fsize>(code));
code.orps(result, nan_mask);
}
}
}
template<size_t fsize>
void ZeroIfNaN(BlockOfCode& code, Xbyak::Xmm result) {
const Xbyak::Xmm nan_mask = xmm0;
if (code.DoesCpuSupport(Xbyak::util::Cpu::tAVX)) {
FCODE(vcmpordp)(nan_mask, result, result);
FCODE(vandp)(result, result, nan_mask);
} else {
code.movaps(nan_mask, result);
FCODE(cmpordp)(nan_mask, nan_mask);
code.andps(result, nan_mask);
}
}
template<size_t fsize>
void DenormalsAreZero(BlockOfCode& code, EmitContext& ctx, std::initializer_list<Xbyak::Xmm> to_daz, Xbyak::Xmm tmp) {
if (ctx.FPCR().FZ()) {
if (ctx.FPCR().RMode() != FP::RoundingMode::TowardsMinusInfinity) {
code.movaps(tmp, GetNegativeZeroVector<fsize>(code));
} else {
code.xorps(tmp, tmp);
}
for (const Xbyak::Xmm& xmm : to_daz) {
FCODE(addp)(xmm, tmp);
}
}
}
template<typename T>
struct DefaultIndexer {
std::tuple<T> operator()(size_t i, const VectorArray<T>& a) {
return std::make_tuple(a[i]);
}
std::tuple<T, T> operator()(size_t i, const VectorArray<T>& a, const VectorArray<T>& b) {
return std::make_tuple(a[i], b[i]);
}
std::tuple<T, T, T> operator()(size_t i, const VectorArray<T>& a, const VectorArray<T>& b, const VectorArray<T>& c) {
return std::make_tuple(a[i], b[i], c[i]);
}
};
template<typename T>
struct PairedIndexer {
std::tuple<T, T> operator()(size_t i, const VectorArray<T>& a, const VectorArray<T>& b) {
constexpr size_t halfway = std::tuple_size_v<VectorArray<T>> / 2;
const size_t which_array = i / halfway;
i %= halfway;
switch (which_array) {
case 0:
return std::make_tuple(a[2 * i], a[2 * i + 1]);
case 1:
return std::make_tuple(b[2 * i], b[2 * i + 1]);
}
UNREACHABLE();
return {};
}
};
template<typename T>
struct PairedLowerIndexer {
std::tuple<T, T> operator()(size_t i, const VectorArray<T>& a, const VectorArray<T>& b) {
constexpr size_t array_size = std::tuple_size_v<VectorArray<T>>;
if constexpr (array_size == 4) {
switch (i) {
case 0:
return std::make_tuple(a[0], a[1]);
case 1:
return std::make_tuple(b[0], b[1]);
default:
return std::make_tuple(0, 0);
}
} else if constexpr (array_size == 2) {
if (i == 0) {
return std::make_tuple(a[0], b[0]);
}
return std::make_tuple(0, 0);
}
UNREACHABLE();
return {};
}
};
template<size_t fsize, template<typename> class Indexer, typename Function>
void EmitTwoOpVectorOperation(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst, Function fn, typename NaNHandler<fsize, Indexer, 2>::function_type nan_handler = NaNHandler<fsize, Indexer, 2>::GetDefault()) {
static_assert(fsize == 32 || fsize == 64, "fsize must be either 32 or 64");
if (!ctx.AccurateNaN() || ctx.FPCR().DN()) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
Xbyak::Xmm result;
if constexpr (std::is_member_function_pointer_v<Function>) {
result = ctx.reg_alloc.UseScratchXmm(args[0]);
(code.*fn)(result);
} else {
const Xbyak::Xmm xmm_a = ctx.reg_alloc.UseXmm(args[0]);
result = ctx.reg_alloc.ScratchXmm();
fn(result, xmm_a);
}
ForceToDefaultNaN<fsize>(code, ctx, result);
ctx.reg_alloc.DefineValue(inst, result);
return;
}
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm result = ctx.reg_alloc.ScratchXmm();
const Xbyak::Xmm xmm_a = ctx.reg_alloc.UseXmm(args[0]);
const Xbyak::Xmm nan_mask = ctx.reg_alloc.ScratchXmm();
if constexpr (std::is_member_function_pointer_v<Function>) {
code.movaps(result, xmm_a);
(code.*fn)(result);
} else {
fn(result, xmm_a);
}
if (code.DoesCpuSupport(Xbyak::util::Cpu::tAVX)) {
FCODE(vcmpunordp)(nan_mask, result, result);
} else {
code.movaps(nan_mask, result);
FCODE(cmpunordp)(nan_mask, nan_mask);
}
HandleNaNs<fsize, 1>(code, ctx, {result, xmm_a}, nan_mask, nan_handler);
ctx.reg_alloc.DefineValue(inst, result);
}
template<size_t fsize, template<typename> class Indexer, typename Function>
void EmitThreeOpVectorOperation(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst, Function fn, typename NaNHandler<fsize, Indexer, 3>::function_type nan_handler = NaNHandler<fsize, Indexer, 3>::GetDefault()) {
static_assert(fsize == 32 || fsize == 64, "fsize must be either 32 or 64");
if (!ctx.AccurateNaN() || ctx.FPCR().DN()) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm xmm_a = ctx.reg_alloc.UseScratchXmm(args[0]);
const Xbyak::Xmm xmm_b = ctx.reg_alloc.UseXmm(args[1]);
if constexpr (std::is_member_function_pointer_v<Function>) {
(code.*fn)(xmm_a, xmm_b);
} else {
fn(xmm_a, xmm_b);
}
ForceToDefaultNaN<fsize>(code, ctx, xmm_a);
ctx.reg_alloc.DefineValue(inst, xmm_a);
return;
}
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm result = ctx.reg_alloc.ScratchXmm();
const Xbyak::Xmm xmm_a = ctx.reg_alloc.UseXmm(args[0]);
const Xbyak::Xmm xmm_b = ctx.reg_alloc.UseXmm(args[1]);
const Xbyak::Xmm nan_mask = ctx.reg_alloc.ScratchXmm();
code.movaps(nan_mask, xmm_b);
code.movaps(result, xmm_a);
FCODE(cmpunordp)(nan_mask, xmm_a);
if constexpr (std::is_member_function_pointer_v<Function>) {
(code.*fn)(result, xmm_b);
} else {
fn(result, xmm_b);
}
FCODE(cmpunordp)(nan_mask, result);
HandleNaNs<fsize, 2>(code, ctx, {result, xmm_a, xmm_b}, nan_mask, nan_handler);
ctx.reg_alloc.DefineValue(inst, result);
}
template<typename Lambda>
void EmitTwoOpFallback(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst, Lambda lambda) {
const auto fn = static_cast<mp::equivalent_function_type_t<Lambda>*>(lambda);
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm arg1 = ctx.reg_alloc.UseXmm(args[0]);
const Xbyak::Xmm result = ctx.reg_alloc.ScratchXmm();
ctx.reg_alloc.EndOfAllocScope();
ctx.reg_alloc.HostCall(nullptr);
constexpr u32 stack_space = 2 * 16;
code.sub(rsp, stack_space + ABI_SHADOW_SPACE);
code.lea(code.ABI_PARAM1, ptr[rsp + ABI_SHADOW_SPACE + 0 * 16]);
code.lea(code.ABI_PARAM2, ptr[rsp + ABI_SHADOW_SPACE + 1 * 16]);
code.mov(code.ABI_PARAM3.cvt32(), ctx.FPCR().Value());
code.lea(code.ABI_PARAM4, code.ptr[code.r15 + code.GetJitStateInfo().offsetof_fpsr_exc]);
code.movaps(xword[code.ABI_PARAM2], arg1);
code.CallFunction(fn);
code.movaps(result, xword[rsp + ABI_SHADOW_SPACE + 0 * 16]);
code.add(rsp, stack_space + ABI_SHADOW_SPACE);
ctx.reg_alloc.DefineValue(inst, result);
}
template<typename Lambda>
void EmitThreeOpFallbackWithoutRegAlloc(BlockOfCode& code, EmitContext& ctx, Xbyak::Xmm result, Xbyak::Xmm arg1, Xbyak::Xmm arg2, Lambda lambda) {
const auto fn = static_cast<mp::equivalent_function_type_t<Lambda>*>(lambda);
#ifdef _WIN32
constexpr u32 stack_space = 4 * 16;
code.sub(rsp, stack_space + ABI_SHADOW_SPACE);
code.lea(code.ABI_PARAM1, ptr[rsp + ABI_SHADOW_SPACE + 1 * 16]);
code.lea(code.ABI_PARAM2, ptr[rsp + ABI_SHADOW_SPACE + 2 * 16]);
code.lea(code.ABI_PARAM3, ptr[rsp + ABI_SHADOW_SPACE + 3 * 16]);
code.mov(code.ABI_PARAM4.cvt32(), ctx.FPCR().Value());
code.lea(rax, code.ptr[code.r15 + code.GetJitStateInfo().offsetof_fpsr_exc]);
code.mov(qword[rsp + ABI_SHADOW_SPACE + 0], rax);
#else
constexpr u32 stack_space = 3 * 16;
code.sub(rsp, stack_space + ABI_SHADOW_SPACE);
code.lea(code.ABI_PARAM1, ptr[rsp + ABI_SHADOW_SPACE + 0 * 16]);
code.lea(code.ABI_PARAM2, ptr[rsp + ABI_SHADOW_SPACE + 1 * 16]);
code.lea(code.ABI_PARAM3, ptr[rsp + ABI_SHADOW_SPACE + 2 * 16]);
code.mov(code.ABI_PARAM4.cvt32(), ctx.FPCR().Value());
code.lea(code.ABI_PARAM5, code.ptr[code.r15 + code.GetJitStateInfo().offsetof_fpsr_exc]);
#endif
code.movaps(xword[code.ABI_PARAM2], arg1);
code.movaps(xword[code.ABI_PARAM3], arg2);
code.CallFunction(fn);
#ifdef _WIN32
code.movaps(result, xword[rsp + ABI_SHADOW_SPACE + 1 * 16]);
#else
code.movaps(result, xword[rsp + ABI_SHADOW_SPACE + 0 * 16]);
#endif
code.add(rsp, stack_space + ABI_SHADOW_SPACE);
}
template<typename Lambda>
void EmitThreeOpFallback(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst, Lambda lambda) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm arg1 = ctx.reg_alloc.UseXmm(args[0]);
const Xbyak::Xmm arg2 = ctx.reg_alloc.UseXmm(args[1]);
const Xbyak::Xmm result = ctx.reg_alloc.ScratchXmm();
ctx.reg_alloc.EndOfAllocScope();
ctx.reg_alloc.HostCall(nullptr);
EmitThreeOpFallbackWithoutRegAlloc(code, ctx, result, arg1, arg2, lambda);
ctx.reg_alloc.DefineValue(inst, result);
}
template<typename Lambda>
void EmitFourOpFallbackWithoutRegAlloc(BlockOfCode& code, EmitContext& ctx, Xbyak::Xmm result, Xbyak::Xmm arg1, Xbyak::Xmm arg2, Xbyak::Xmm arg3, Lambda lambda) {
const auto fn = static_cast<mp::equivalent_function_type_t<Lambda>*>(lambda);
#ifdef _WIN32
constexpr u32 stack_space = 5 * 16;
code.sub(rsp, stack_space + ABI_SHADOW_SPACE);
code.lea(code.ABI_PARAM1, ptr[rsp + ABI_SHADOW_SPACE + 1 * 16]);
code.lea(code.ABI_PARAM2, ptr[rsp + ABI_SHADOW_SPACE + 2 * 16]);
code.lea(code.ABI_PARAM3, ptr[rsp + ABI_SHADOW_SPACE + 3 * 16]);
code.lea(code.ABI_PARAM4, ptr[rsp + ABI_SHADOW_SPACE + 4 * 16]);
code.mov(qword[rsp + ABI_SHADOW_SPACE + 0], ctx.FPCR().Value());
code.lea(rax, code.ptr[code.r15 + code.GetJitStateInfo().offsetof_fpsr_exc]);
code.mov(qword[rsp + ABI_SHADOW_SPACE + 8], rax);
#else
constexpr u32 stack_space = 4 * 16;
code.sub(rsp, stack_space + ABI_SHADOW_SPACE);
code.lea(code.ABI_PARAM1, ptr[rsp + ABI_SHADOW_SPACE + 0 * 16]);
code.lea(code.ABI_PARAM2, ptr[rsp + ABI_SHADOW_SPACE + 1 * 16]);
code.lea(code.ABI_PARAM3, ptr[rsp + ABI_SHADOW_SPACE + 2 * 16]);
code.lea(code.ABI_PARAM4, ptr[rsp + ABI_SHADOW_SPACE + 3 * 16]);
code.mov(code.ABI_PARAM5.cvt32(), ctx.FPCR().Value());
code.lea(code.ABI_PARAM6, code.ptr[code.r15 + code.GetJitStateInfo().offsetof_fpsr_exc]);
#endif
code.movaps(xword[code.ABI_PARAM2], arg1);
code.movaps(xword[code.ABI_PARAM3], arg2);
code.movaps(xword[code.ABI_PARAM4], arg3);
code.CallFunction(fn);
#ifdef _WIN32
code.movaps(result, xword[rsp + ABI_SHADOW_SPACE + 1 * 16]);
#else
code.movaps(result, xword[rsp + ABI_SHADOW_SPACE + 0 * 16]);
#endif
code.add(rsp, stack_space + ABI_SHADOW_SPACE);
}
template<typename Lambda>
void EmitFourOpFallback(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst, Lambda lambda) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm arg1 = ctx.reg_alloc.UseXmm(args[0]);
const Xbyak::Xmm arg2 = ctx.reg_alloc.UseXmm(args[1]);
const Xbyak::Xmm arg3 = ctx.reg_alloc.UseXmm(args[2]);
const Xbyak::Xmm result = ctx.reg_alloc.ScratchXmm();
ctx.reg_alloc.EndOfAllocScope();
ctx.reg_alloc.HostCall(nullptr);
EmitFourOpFallbackWithoutRegAlloc(code, ctx, result, arg1, arg2, arg3, lambda);
ctx.reg_alloc.DefineValue(inst, result);
}
} // anonymous namespace
void EmitX64::EmitFPVectorAbs16(EmitContext& ctx, IR::Inst* inst) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);
const Xbyak::Address mask = code.MConst(xword, 0x7FFF7FFF7FFF7FFF, 0x7FFF7FFF7FFF7FFF);
code.pand(a, mask);
ctx.reg_alloc.DefineValue(inst, a);
}
void EmitX64::EmitFPVectorAbs32(EmitContext& ctx, IR::Inst* inst) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);
const Xbyak::Address mask = code.MConst(xword, 0x7FFFFFFF7FFFFFFF, 0x7FFFFFFF7FFFFFFF);
code.andps(a, mask);
ctx.reg_alloc.DefineValue(inst, a);
}
void EmitX64::EmitFPVectorAbs64(EmitContext& ctx, IR::Inst* inst) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);
const Xbyak::Address mask = code.MConst(xword, 0x7FFFFFFFFFFFFFFF, 0x7FFFFFFFFFFFFFFF);
code.andpd(a, mask);
ctx.reg_alloc.DefineValue(inst, a);
}
void EmitX64::EmitFPVectorAdd32(EmitContext& ctx, IR::Inst* inst) {
EmitThreeOpVectorOperation<32, DefaultIndexer>(code, ctx, inst, &Xbyak::CodeGenerator::addps);
}
void EmitX64::EmitFPVectorAdd64(EmitContext& ctx, IR::Inst* inst) {
EmitThreeOpVectorOperation<64, DefaultIndexer>(code, ctx, inst, &Xbyak::CodeGenerator::addpd);
}
void EmitX64::EmitFPVectorDiv32(EmitContext& ctx, IR::Inst* inst) {
EmitThreeOpVectorOperation<32, DefaultIndexer>(code, ctx, inst, &Xbyak::CodeGenerator::divps);
}
void EmitX64::EmitFPVectorDiv64(EmitContext& ctx, IR::Inst* inst) {
EmitThreeOpVectorOperation<64, DefaultIndexer>(code, ctx, inst, &Xbyak::CodeGenerator::divpd);
}
void EmitX64::EmitFPVectorEqual32(EmitContext& ctx, IR::Inst* inst) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);
const Xbyak::Xmm b = ctx.reg_alloc.UseXmm(args[1]);
code.cmpeqps(a, b);
ctx.reg_alloc.DefineValue(inst, a);
}
void EmitX64::EmitFPVectorEqual64(EmitContext& ctx, IR::Inst* inst) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);
const Xbyak::Xmm b = ctx.reg_alloc.UseXmm(args[1]);
code.cmpeqpd(a, b);
ctx.reg_alloc.DefineValue(inst, a);
}
void EmitX64::EmitFPVectorFromSignedFixed32(EmitContext& ctx, IR::Inst* inst) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm xmm = ctx.reg_alloc.UseScratchXmm(args[0]);
const int fbits = args[1].GetImmediateU8();
const FP::RoundingMode rounding_mode = static_cast<FP::RoundingMode>(args[2].GetImmediateU8());
ASSERT(rounding_mode == ctx.FPCR().RMode());
code.cvtdq2ps(xmm, xmm);
if (fbits != 0) {
code.mulps(xmm, GetVectorOf<32>(code, static_cast<u32>(127 - fbits) << 23));
}
ctx.reg_alloc.DefineValue(inst, xmm);
}
void EmitX64::EmitFPVectorFromSignedFixed64(EmitContext& ctx, IR::Inst* inst) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm xmm = ctx.reg_alloc.UseScratchXmm(args[0]);
const int fbits = args[1].GetImmediateU8();
const FP::RoundingMode rounding_mode = static_cast<FP::RoundingMode>(args[2].GetImmediateU8());
ASSERT(rounding_mode == ctx.FPCR().RMode());
if (code.DoesCpuSupport(Xbyak::util::Cpu::tAVX512VL) && code.DoesCpuSupport(Xbyak::util::Cpu::tAVX512DQ)) {
code.vcvtqq2pd(xmm, xmm);
} else if (code.DoesCpuSupport(Xbyak::util::Cpu::tSSE41)) {
const Xbyak::Xmm xmm_tmp = ctx.reg_alloc.ScratchXmm();
const Xbyak::Reg64 tmp = ctx.reg_alloc.ScratchGpr();
// First quadword
code.movq(tmp, xmm);
code.cvtsi2sd(xmm, tmp);
// Second quadword
code.pextrq(tmp, xmm, 1);
code.cvtsi2sd(xmm_tmp, tmp);
// Combine
code.unpcklpd(xmm, xmm_tmp);
} else {
const Xbyak::Xmm high_xmm = ctx.reg_alloc.ScratchXmm();
const Xbyak::Xmm xmm_tmp = ctx.reg_alloc.ScratchXmm();
const Xbyak::Reg64 tmp = ctx.reg_alloc.ScratchGpr();
// First quadword
code.movhlps(high_xmm, xmm);
code.movq(tmp, xmm);
code.cvtsi2sd(xmm, tmp);
// Second quadword
code.movq(tmp, high_xmm);
code.cvtsi2sd(xmm_tmp, tmp);
// Combine
code.unpcklpd(xmm, xmm_tmp);
}
if (fbits != 0) {
code.mulpd(xmm, GetVectorOf<64>(code, static_cast<u64>(1023 - fbits) << 52));
}
ctx.reg_alloc.DefineValue(inst, xmm);
}
void EmitX64::EmitFPVectorFromUnsignedFixed32(EmitContext& ctx, IR::Inst* inst) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm xmm = ctx.reg_alloc.UseScratchXmm(args[0]);
const int fbits = args[1].GetImmediateU8();
const FP::RoundingMode rounding_mode = static_cast<FP::RoundingMode>(args[2].GetImmediateU8());
ASSERT(rounding_mode == ctx.FPCR().RMode());
if (code.DoesCpuSupport(Xbyak::util::Cpu::tAVX512DQ) && code.DoesCpuSupport(Xbyak::util::Cpu::tAVX512VL)) {
code.vcvtudq2ps(xmm, xmm);
} else {
const Xbyak::Address mem_4B000000 = code.MConst(xword, 0x4B0000004B000000, 0x4B0000004B000000);
const Xbyak::Address mem_53000000 = code.MConst(xword, 0x5300000053000000, 0x5300000053000000);
const Xbyak::Address mem_D3000080 = code.MConst(xword, 0xD3000080D3000080, 0xD3000080D3000080);
const Xbyak::Xmm tmp = ctx.reg_alloc.ScratchXmm();
if (code.DoesCpuSupport(Xbyak::util::Cpu::tAVX)) {
code.vpblendw(tmp, xmm, mem_4B000000, 0b10101010);
code.vpsrld(xmm, xmm, 16);
code.vpblendw(xmm, xmm, mem_53000000, 0b10101010);
code.vaddps(xmm, xmm, mem_D3000080);
code.vaddps(xmm, tmp, xmm);
} else {
const Xbyak::Address mem_0xFFFF = code.MConst(xword, 0x0000FFFF0000FFFF, 0x0000FFFF0000FFFF);
code.movdqa(tmp, mem_0xFFFF);
code.pand(tmp, xmm);
code.por(tmp, mem_4B000000);
code.psrld(xmm, 16);
code.por(xmm, mem_53000000);
code.addps(xmm, mem_D3000080);
code.addps(xmm, tmp);
}
}
if (fbits != 0) {
code.mulps(xmm, GetVectorOf<32>(code, static_cast<u32>(127 - fbits) << 23));
}
if (ctx.FPCR().RMode() == FP::RoundingMode::TowardsMinusInfinity) {
code.pand(xmm, code.MConst(xword, 0x7FFFFFFF7FFFFFFF, 0x7FFFFFFF7FFFFFFF));
}
ctx.reg_alloc.DefineValue(inst, xmm);
}
void EmitX64::EmitFPVectorFromUnsignedFixed64(EmitContext& ctx, IR::Inst* inst) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm xmm = ctx.reg_alloc.UseScratchXmm(args[0]);
const int fbits = args[1].GetImmediateU8();
const FP::RoundingMode rounding_mode = static_cast<FP::RoundingMode>(args[2].GetImmediateU8());
ASSERT(rounding_mode == ctx.FPCR().RMode());
if (code.DoesCpuSupport(Xbyak::util::Cpu::tAVX512DQ) && code.DoesCpuSupport(Xbyak::util::Cpu::tAVX512VL)) {
code.vcvtuqq2pd(xmm, xmm);
} else {
const Xbyak::Address unpack = code.MConst(xword, 0x4530000043300000, 0);
const Xbyak::Address subtrahend = code.MConst(xword, 0x4330000000000000, 0x4530000000000000);
const Xbyak::Xmm unpack_reg = ctx.reg_alloc.ScratchXmm();
const Xbyak::Xmm subtrahend_reg = ctx.reg_alloc.ScratchXmm();
const Xbyak::Xmm tmp1 = ctx.reg_alloc.ScratchXmm();
if (code.DoesCpuSupport(Xbyak::util::Cpu::tAVX)) {
code.vmovapd(unpack_reg, unpack);
code.vmovapd(subtrahend_reg, subtrahend);
code.vunpcklps(tmp1, xmm, unpack_reg);
code.vsubpd(tmp1, tmp1, subtrahend_reg);
code.vpermilps(xmm, xmm, 0b01001110);
code.vunpcklps(xmm, xmm, unpack_reg);
code.vsubpd(xmm, xmm, subtrahend_reg);
code.vhaddpd(xmm, tmp1, xmm);
} else {
const Xbyak::Xmm tmp2 = ctx.reg_alloc.ScratchXmm();
code.movapd(unpack_reg, unpack);
code.movapd(subtrahend_reg, subtrahend);
code.pshufd(tmp1, xmm, 0b01001110);
code.punpckldq(xmm, unpack_reg);
code.subpd(xmm, subtrahend_reg);
code.pshufd(tmp2, xmm, 0b01001110);
code.addpd(xmm, tmp2);
code.punpckldq(tmp1, unpack_reg);
code.subpd(tmp1, subtrahend_reg);
code.pshufd(unpack_reg, tmp1, 0b01001110);
code.addpd(unpack_reg, tmp1);
code.unpcklpd(xmm, unpack_reg);
}
}
if (fbits != 0) {
code.mulpd(xmm, GetVectorOf<64>(code, static_cast<u64>(1023 - fbits) << 52));
}
if (ctx.FPCR().RMode() == FP::RoundingMode::TowardsMinusInfinity) {
code.pand(xmm, code.MConst(xword, 0x7FFFFFFFFFFFFFFF, 0x7FFFFFFFFFFFFFFF));
}
ctx.reg_alloc.DefineValue(inst, xmm);
}
void EmitX64::EmitFPVectorGreater32(EmitContext& ctx, IR::Inst* inst) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm a = ctx.reg_alloc.UseXmm(args[0]);
const Xbyak::Xmm b = ctx.reg_alloc.UseScratchXmm(args[1]);
code.cmpltps(b, a);
ctx.reg_alloc.DefineValue(inst, b);
}
void EmitX64::EmitFPVectorGreater64(EmitContext& ctx, IR::Inst* inst) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm a = ctx.reg_alloc.UseXmm(args[0]);
const Xbyak::Xmm b = ctx.reg_alloc.UseScratchXmm(args[1]);
code.cmpltpd(b, a);
ctx.reg_alloc.DefineValue(inst, b);
}
void EmitX64::EmitFPVectorGreaterEqual32(EmitContext& ctx, IR::Inst* inst) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm a = ctx.reg_alloc.UseXmm(args[0]);
const Xbyak::Xmm b = ctx.reg_alloc.UseScratchXmm(args[1]);
code.cmpleps(b, a);
ctx.reg_alloc.DefineValue(inst, b);
}
void EmitX64::EmitFPVectorGreaterEqual64(EmitContext& ctx, IR::Inst* inst) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm a = ctx.reg_alloc.UseXmm(args[0]);
const Xbyak::Xmm b = ctx.reg_alloc.UseScratchXmm(args[1]);
code.cmplepd(b, a);
ctx.reg_alloc.DefineValue(inst, b);
}
template<size_t fsize, bool is_max>
static void EmitFPVectorMinMax(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst) {
if (ctx.FPCR().DN()) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm result = ctx.reg_alloc.UseScratchXmm(args[0]);
const Xbyak::Xmm xmm_b = ctx.FPCR().FZ() ? ctx.reg_alloc.UseScratchXmm(args[1]) : ctx.reg_alloc.UseXmm(args[1]);
const Xbyak::Xmm mask = xmm0;
const Xbyak::Xmm eq = ctx.reg_alloc.ScratchXmm();
const Xbyak::Xmm nan_mask = ctx.reg_alloc.ScratchXmm();
DenormalsAreZero<fsize>(code, ctx, {result, xmm_b}, mask);
if (code.DoesCpuSupport(Xbyak::util::Cpu::tAVX)) {
FCODE(vcmpeqp)(mask, result, xmm_b);
FCODE(vcmpunordp)(nan_mask, result, xmm_b);
if constexpr (is_max) {
FCODE(vandp)(eq, result, xmm_b);
FCODE(vmaxp)(result, result, xmm_b);
} else {
FCODE(vorp)(eq, result, xmm_b);
FCODE(vminp)(result, result, xmm_b);
}
FCODE(blendvp)(result, eq);
FCODE(vblendvp)(result, result, GetNaNVector<fsize>(code), nan_mask);
} else {
code.movaps(mask, result);
code.movaps(eq, result);
code.movaps(nan_mask, result);
FCODE(cmpneqp)(mask, xmm_b);
FCODE(cmpordp)(nan_mask, xmm_b);
if constexpr (is_max) {
code.andps(eq, xmm_b);
FCODE(maxp)(result, xmm_b);
} else {
code.orps(eq, xmm_b);
FCODE(minp)(result, xmm_b);
}
code.andps(result, mask);
code.andnps(mask, eq);
code.orps(result, mask);
code.andps(result, nan_mask);
code.andnps(nan_mask, GetNaNVector<fsize>(code));
code.orps(result, nan_mask);
}
ctx.reg_alloc.DefineValue(inst, result);
return;
}
EmitThreeOpVectorOperation<fsize, DefaultIndexer>(code, ctx, inst, [&](const Xbyak::Xmm& result, Xbyak::Xmm xmm_b){
const Xbyak::Xmm mask = xmm0;
const Xbyak::Xmm eq = ctx.reg_alloc.ScratchXmm();
if (ctx.FPCR().FZ()) {
const Xbyak::Xmm prev_xmm_b = xmm_b;
xmm_b = ctx.reg_alloc.ScratchXmm();
code.movaps(xmm_b, prev_xmm_b);
DenormalsAreZero<fsize>(code, ctx, {result, xmm_b}, mask);
}
// What we are doing here is handling the case when the inputs are differently signed zeros.
// x86-64 treats differently signed zeros as equal while ARM does not.
// Thus if we AND together things that x86-64 thinks are equal we'll get the positive zero.
if (code.DoesCpuSupport(Xbyak::util::Cpu::tAVX)) {
FCODE(vcmpeqp)(mask, result, xmm_b);
if constexpr (is_max) {
FCODE(vandp)(eq, result, xmm_b);
FCODE(vmaxp)(result, result, xmm_b);
} else {
FCODE(vorp)(eq, result, xmm_b);
FCODE(vminp)(result, result, xmm_b);
}
FCODE(blendvp)(result, eq);
} else {
code.movaps(mask, result);
code.movaps(eq, result);
FCODE(cmpneqp)(mask, xmm_b);
if constexpr (is_max) {
code.andps(eq, xmm_b);
FCODE(maxp)(result, xmm_b);
} else {
code.orps(eq, xmm_b);
FCODE(minp)(result, xmm_b);
}
code.andps(result, mask);
code.andnps(mask, eq);
code.orps(result, mask);
}
});
}
void EmitX64::EmitFPVectorMax32(EmitContext& ctx, IR::Inst* inst) {
EmitFPVectorMinMax<32, true>(code, ctx, inst);
}
void EmitX64::EmitFPVectorMax64(EmitContext& ctx, IR::Inst* inst) {
EmitFPVectorMinMax<64, true>(code, ctx, inst);
}
void EmitX64::EmitFPVectorMin32(EmitContext& ctx, IR::Inst* inst) {
EmitFPVectorMinMax<32, false>(code, ctx, inst);
}
void EmitX64::EmitFPVectorMin64(EmitContext& ctx, IR::Inst* inst) {
EmitFPVectorMinMax<64, false>(code, ctx, inst);
}
void EmitX64::EmitFPVectorMul32(EmitContext& ctx, IR::Inst* inst) {
EmitThreeOpVectorOperation<32, DefaultIndexer>(code, ctx, inst, &Xbyak::CodeGenerator::mulps);
}
void EmitX64::EmitFPVectorMul64(EmitContext& ctx, IR::Inst* inst) {
EmitThreeOpVectorOperation<64, DefaultIndexer>(code, ctx, inst, &Xbyak::CodeGenerator::mulpd);
}
template<size_t fsize>
void EmitFPVectorMulAdd(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst) {
using FPT = mp::unsigned_integer_of_size<fsize>;
const auto fallback_fn = [](VectorArray<FPT>& result, const VectorArray<FPT>& addend, const VectorArray<FPT>& op1, const VectorArray<FPT>& op2, FP::FPCR fpcr, FP::FPSR& fpsr) {
for (size_t i = 0; i < result.size(); i++) {
result[i] = FP::FPMulAdd<FPT>(addend[i], op1[i], op2[i], fpcr, fpsr);
}
};
if (code.DoesCpuSupport(Xbyak::util::Cpu::tFMA) && code.DoesCpuSupport(Xbyak::util::Cpu::tAVX)) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm result = ctx.reg_alloc.ScratchXmm();
const Xbyak::Xmm xmm_a = ctx.reg_alloc.UseXmm(args[0]);
const Xbyak::Xmm xmm_b = ctx.reg_alloc.UseXmm(args[1]);
const Xbyak::Xmm xmm_c = ctx.reg_alloc.UseXmm(args[2]);
const Xbyak::Xmm tmp = ctx.reg_alloc.ScratchXmm();
Xbyak::Label end, fallback;
code.movaps(result, xmm_a);
FCODE(vfmadd231p)(result, xmm_b, xmm_c);
code.movaps(tmp, GetNegativeZeroVector<fsize>(code));
code.andnps(tmp, result);
FCODE(vcmpeq_uqp)(tmp, tmp, GetSmallestNormalVector<fsize>(code));
code.vptest(tmp, tmp);
code.jnz(fallback, code.T_NEAR);
code.L(end);
code.SwitchToFarCode();
code.L(fallback);
code.sub(rsp, 8);
ABI_PushCallerSaveRegistersAndAdjustStackExcept(code, HostLocXmmIdx(result.getIdx()));
EmitFourOpFallbackWithoutRegAlloc(code, ctx, result, xmm_a, xmm_b, xmm_c, fallback_fn);
ABI_PopCallerSaveRegistersAndAdjustStackExcept(code, HostLocXmmIdx(result.getIdx()));
code.add(rsp, 8);
code.jmp(end, code.T_NEAR);
code.SwitchToNearCode();
ctx.reg_alloc.DefineValue(inst, result);
return;
}
EmitFourOpFallback(code, ctx, inst, fallback_fn);
}
void EmitX64::EmitFPVectorMulAdd32(EmitContext& ctx, IR::Inst* inst) {
EmitFPVectorMulAdd<32>(code, ctx, inst);
}
void EmitX64::EmitFPVectorMulAdd64(EmitContext& ctx, IR::Inst* inst) {
EmitFPVectorMulAdd<64>(code, ctx, inst);
}
template<size_t fsize>
static void EmitFPVectorMulX(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst) {
using FPT = mp::unsigned_integer_of_size<fsize>;
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
if (ctx.FPCR().DN() && code.DoesCpuSupport(Xbyak::util::Cpu::tAVX)) {
const Xbyak::Xmm result = ctx.reg_alloc.UseScratchXmm(args[0]);
const Xbyak::Xmm operand = ctx.reg_alloc.UseXmm(args[1]);
const Xbyak::Xmm tmp = ctx.reg_alloc.ScratchXmm();
const Xbyak::Xmm twos = ctx.reg_alloc.ScratchXmm();
FCODE(vcmpunordp)(xmm0, result, operand);
FCODE(vxorp)(twos, result, operand);
FCODE(mulp)(result, operand);
FCODE(andp)(twos, GetNegativeZeroVector<fsize>(code));
FCODE(vcmpunordp)(tmp, result, result);
FCODE(blendvp)(result, GetNaNVector<fsize>(code));
FCODE(orp)(twos, GetVectorOf<fsize, false, 0, 2>(code));
FCODE(andnp)(xmm0, tmp);
FCODE(blendvp)(result, twos);
ctx.reg_alloc.DefineValue(inst, result);
return;
}
const Xbyak::Xmm result = ctx.reg_alloc.ScratchXmm();
const Xbyak::Xmm xmm_a = ctx.reg_alloc.UseXmm(args[0]);
const Xbyak::Xmm xmm_b = ctx.reg_alloc.UseXmm(args[1]);
const Xbyak::Xmm nan_mask = ctx.reg_alloc.ScratchXmm();
code.movaps(nan_mask, xmm_b);
code.movaps(result, xmm_a);
FCODE(cmpunordp)(nan_mask, xmm_a);
FCODE(mulp)(result, xmm_b);
FCODE(cmpunordp)(nan_mask, result);
const auto nan_handler = static_cast<void(*)(std::array<VectorArray<FPT>, 3>&, FP::FPCR)>(
[](std::array<VectorArray<FPT>, 3>& values, FP::FPCR fpcr) {
VectorArray<FPT>& result = values[0];
for (size_t elementi = 0; elementi < result.size(); ++elementi) {
if (auto r = FP::ProcessNaNs(values[1][elementi], values[2][elementi])) {
result[elementi] = fpcr.DN() ? FP::FPInfo<FPT>::DefaultNaN() : *r;
} else if (FP::IsNaN(result[elementi])) {
const FPT sign = (values[1][elementi] ^ values[2][elementi]) & FP::FPInfo<FPT>::sign_mask;
result[elementi] = sign | FP::FPValue<FPT, false, 0, 2>();
}
}
}
);
HandleNaNs<fsize, 2>(code, ctx, {result, xmm_a, xmm_b}, nan_mask, nan_handler);
ctx.reg_alloc.DefineValue(inst, result);
}
void EmitX64::EmitFPVectorMulX32(EmitContext& ctx, IR::Inst* inst) {
EmitFPVectorMulX<32>(code, ctx, inst);
}
void EmitX64::EmitFPVectorMulX64(EmitContext& ctx, IR::Inst* inst) {
EmitFPVectorMulX<64>(code, ctx, inst);
}
void EmitX64::EmitFPVectorNeg16(EmitContext& ctx, IR::Inst* inst) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);
const Xbyak::Address mask = code.MConst(xword, 0x8000800080008000, 0x8000800080008000);
code.pxor(a, mask);
ctx.reg_alloc.DefineValue(inst, a);
}
void EmitX64::EmitFPVectorNeg32(EmitContext& ctx, IR::Inst* inst) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);
const Xbyak::Address mask = code.MConst(xword, 0x8000000080000000, 0x8000000080000000);
code.pxor(a, mask);
ctx.reg_alloc.DefineValue(inst, a);
}
void EmitX64::EmitFPVectorNeg64(EmitContext& ctx, IR::Inst* inst) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);
const Xbyak::Address mask = code.MConst(xword, 0x8000000000000000, 0x8000000000000000);
code.pxor(a, mask);
ctx.reg_alloc.DefineValue(inst, a);
}
void EmitX64::EmitFPVectorPairedAdd32(EmitContext& ctx, IR::Inst* inst) {
EmitThreeOpVectorOperation<32, PairedIndexer>(code, ctx, inst, &Xbyak::CodeGenerator::haddps);
}
void EmitX64::EmitFPVectorPairedAdd64(EmitContext& ctx, IR::Inst* inst) {
EmitThreeOpVectorOperation<64, PairedIndexer>(code, ctx, inst, &Xbyak::CodeGenerator::haddpd);
}
void EmitX64::EmitFPVectorPairedAddLower32(EmitContext& ctx, IR::Inst* inst) {
EmitThreeOpVectorOperation<32, PairedLowerIndexer>(code, ctx, inst, [&](Xbyak::Xmm result, Xbyak::Xmm xmm_b) {
const Xbyak::Xmm zero = ctx.reg_alloc.ScratchXmm();
code.xorps(zero, zero);
code.punpcklqdq(result, xmm_b);
code.haddps(result, zero);
});
}
void EmitX64::EmitFPVectorPairedAddLower64(EmitContext& ctx, IR::Inst* inst) {
EmitThreeOpVectorOperation<64, PairedLowerIndexer>(code, ctx, inst, [&](Xbyak::Xmm result, Xbyak::Xmm xmm_b) {
const Xbyak::Xmm zero = ctx.reg_alloc.ScratchXmm();
code.xorps(zero, zero);
code.punpcklqdq(result, xmm_b);
code.haddpd(result, zero);
});
}
template<typename FPT>
static void EmitRecipEstimate(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst) {
EmitTwoOpFallback(code, ctx, inst, [](VectorArray<FPT>& result, const VectorArray<FPT>& operand, FP::FPCR fpcr, FP::FPSR& fpsr) {
for (size_t i = 0; i < result.size(); i++) {
result[i] = FP::FPRecipEstimate<FPT>(operand[i], fpcr, fpsr);
}
});
}
void EmitX64::EmitFPVectorRecipEstimate32(EmitContext& ctx, IR::Inst* inst) {
EmitRecipEstimate<u32>(code, ctx, inst);
}
void EmitX64::EmitFPVectorRecipEstimate64(EmitContext& ctx, IR::Inst* inst) {
EmitRecipEstimate<u64>(code, ctx, inst);
}
template<size_t fsize>
static void EmitRecipStepFused(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst) {
using FPT = mp::unsigned_integer_of_size<fsize>;
const auto fallback_fn = [](VectorArray<FPT>& result, const VectorArray<FPT>& op1, const VectorArray<FPT>& op2, FP::FPCR fpcr, FP::FPSR& fpsr) {
for (size_t i = 0; i < result.size(); i++) {
result[i] = FP::FPRecipStepFused<FPT>(op1[i], op2[i], fpcr, fpsr);
}
};
if (code.DoesCpuSupport(Xbyak::util::Cpu::tFMA) && code.DoesCpuSupport(Xbyak::util::Cpu::tAVX)) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm result = ctx.reg_alloc.ScratchXmm();
const Xbyak::Xmm operand1 = ctx.reg_alloc.UseXmm(args[0]);
const Xbyak::Xmm operand2 = ctx.reg_alloc.UseXmm(args[1]);
const Xbyak::Xmm tmp = ctx.reg_alloc.ScratchXmm();
Xbyak::Label end, fallback;
code.movaps(result, GetVectorOf<fsize, false, 0, 2>(code));
FCODE(vfnmadd231p)(result, operand1, operand2);
FCODE(vcmpunordp)(tmp, result, result);
code.vptest(tmp, tmp);
code.jnz(fallback, code.T_NEAR);
code.L(end);
code.SwitchToFarCode();
code.L(fallback);
code.sub(rsp, 8);
ABI_PushCallerSaveRegistersAndAdjustStackExcept(code, HostLocXmmIdx(result.getIdx()));
EmitThreeOpFallbackWithoutRegAlloc(code, ctx, result, operand1, operand2, fallback_fn);
ABI_PopCallerSaveRegistersAndAdjustStackExcept(code, HostLocXmmIdx(result.getIdx()));
code.add(rsp, 8);
code.jmp(end, code.T_NEAR);
code.SwitchToNearCode();
ctx.reg_alloc.DefineValue(inst, result);
return;
}
EmitThreeOpFallback(code, ctx, inst, fallback_fn);
}
void EmitX64::EmitFPVectorRecipStepFused32(EmitContext& ctx, IR::Inst* inst) {
EmitRecipStepFused<32>(code, ctx, inst);
}
void EmitX64::EmitFPVectorRecipStepFused64(EmitContext& ctx, IR::Inst* inst) {
EmitRecipStepFused<64>(code, ctx, inst);
}
template<size_t fsize>
void EmitFPVectorRoundInt(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst) {
using FPT = mp::unsigned_integer_of_size<fsize>;
const auto rounding = static_cast<FP::RoundingMode>(inst->GetArg(1).GetU8());
const bool exact = inst->GetArg(2).GetU1();
if (code.DoesCpuSupport(Xbyak::util::Cpu::tSSE41) && rounding != FP::RoundingMode::ToNearest_TieAwayFromZero && !exact) {
const u8 round_imm = [&]() -> u8 {
switch (rounding) {
case FP::RoundingMode::ToNearest_TieEven:
return 0b00;
case FP::RoundingMode::TowardsPlusInfinity:
return 0b10;
case FP::RoundingMode::TowardsMinusInfinity:
return 0b01;
case FP::RoundingMode::TowardsZero:
return 0b11;
default:
UNREACHABLE();
}
return 0;
}();
EmitTwoOpVectorOperation<fsize, DefaultIndexer>(code, ctx, inst, [&](const Xbyak::Xmm& result, const Xbyak::Xmm& xmm_a){
FCODE(roundp)(result, xmm_a, round_imm);
});
return;
}
using rounding_list = mp::list<
std::integral_constant<FP::RoundingMode, FP::RoundingMode::ToNearest_TieEven>,
std::integral_constant<FP::RoundingMode, FP::RoundingMode::TowardsPlusInfinity>,
std::integral_constant<FP::RoundingMode, FP::RoundingMode::TowardsMinusInfinity>,
std::integral_constant<FP::RoundingMode, FP::RoundingMode::TowardsZero>,
std::integral_constant<FP::RoundingMode, FP::RoundingMode::ToNearest_TieAwayFromZero>
>;
using exact_list = mp::list<mp::vlift<true>, mp::vlift<false>>;
using key_type = std::tuple<FP::RoundingMode, bool>;
using value_type = void(*)(VectorArray<FPT>&, const VectorArray<FPT>&, FP::FPCR, FP::FPSR&);
static const auto lut = mp::GenerateLookupTableFromList<key_type, value_type>(
[](auto arg) {
return std::pair<key_type, value_type>{
mp::to_tuple<decltype(arg)>,
static_cast<value_type>(
[](VectorArray<FPT>& output, const VectorArray<FPT>& input, FP::FPCR fpcr, FP::FPSR& fpsr) {
constexpr FP::RoundingMode rounding_mode = std::get<0>(mp::to_tuple<decltype(arg)>);
constexpr bool exact = std::get<1>(mp::to_tuple<decltype(arg)>);
for (size_t i = 0; i < output.size(); ++i) {
output[i] = static_cast<FPT>(FP::FPRoundInt<FPT>(input[i], fpcr, rounding_mode, exact, fpsr));
}
}
)
};
},
mp::cartesian_product<rounding_list, exact_list>{}
);
EmitTwoOpFallback(code, ctx, inst, lut.at(std::make_tuple(rounding, exact)));
}
void EmitX64::EmitFPVectorRoundInt32(EmitContext& ctx, IR::Inst* inst) {
EmitFPVectorRoundInt<32>(code, ctx, inst);
}
void EmitX64::EmitFPVectorRoundInt64(EmitContext& ctx, IR::Inst* inst) {
EmitFPVectorRoundInt<64>(code, ctx, inst);
}
template<typename FPT>
static void EmitRSqrtEstimate(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst) {
EmitTwoOpFallback(code, ctx, inst, [](VectorArray<FPT>& result, const VectorArray<FPT>& operand, FP::FPCR fpcr, FP::FPSR& fpsr) {
for (size_t i = 0; i < result.size(); i++) {
result[i] = FP::FPRSqrtEstimate<FPT>(operand[i], fpcr, fpsr);
}
});
}
void EmitX64::EmitFPVectorRSqrtEstimate32(EmitContext& ctx, IR::Inst* inst) {
EmitRSqrtEstimate<u32>(code, ctx, inst);
}
void EmitX64::EmitFPVectorRSqrtEstimate64(EmitContext& ctx, IR::Inst* inst) {
EmitRSqrtEstimate<u64>(code, ctx, inst);
}
template<size_t fsize>
static void EmitRSqrtStepFused(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst) {
using FPT = mp::unsigned_integer_of_size<fsize>;
const auto fallback_fn = [](VectorArray<FPT>& result, const VectorArray<FPT>& op1, const VectorArray<FPT>& op2, FP::FPCR fpcr, FP::FPSR& fpsr) {
for (size_t i = 0; i < result.size(); i++) {
result[i] = FP::FPRSqrtStepFused<FPT>(op1[i], op2[i], fpcr, fpsr);
}
};
if (code.DoesCpuSupport(Xbyak::util::Cpu::tFMA) && code.DoesCpuSupport(Xbyak::util::Cpu::tAVX)) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm result = ctx.reg_alloc.ScratchXmm();
const Xbyak::Xmm operand1 = ctx.reg_alloc.UseXmm(args[0]);
const Xbyak::Xmm operand2 = ctx.reg_alloc.UseXmm(args[1]);
const Xbyak::Xmm tmp = ctx.reg_alloc.ScratchXmm();
const Xbyak::Xmm mask = ctx.reg_alloc.ScratchXmm();
Xbyak::Label end, fallback;
code.vmovaps(result, GetVectorOf<fsize, false, 0, 3>(code));
FCODE(vfnmadd231p)(result, operand1, operand2);
// An explanation for this is given in EmitFPRSqrtStepFused.
code.vmovaps(mask, GetVectorOf<fsize, fsize == 32 ? 0x7f000000 : 0x7fe0000000000000>(code));
FCODE(vandp)(tmp, result, mask);
if constexpr (fsize == 32) {
code.vpcmpeqd(tmp, tmp, mask);
} else {
code.vpcmpeqq(tmp, tmp, mask);
}
code.ptest(tmp, tmp);
code.jnz(fallback, code.T_NEAR);
FCODE(vmulp)(result, result, GetVectorOf<fsize, false, -1, 1>(code));
code.L(end);
code.SwitchToFarCode();
code.L(fallback);
code.sub(rsp, 8);
ABI_PushCallerSaveRegistersAndAdjustStackExcept(code, HostLocXmmIdx(result.getIdx()));
EmitThreeOpFallbackWithoutRegAlloc(code, ctx, result, operand1, operand2, fallback_fn);
ABI_PopCallerSaveRegistersAndAdjustStackExcept(code, HostLocXmmIdx(result.getIdx()));
code.add(rsp, 8);
code.jmp(end, code.T_NEAR);
code.SwitchToNearCode();
ctx.reg_alloc.DefineValue(inst, result);
return;
}
EmitThreeOpFallback(code, ctx, inst, fallback_fn);
}
void EmitX64::EmitFPVectorRSqrtStepFused32(EmitContext& ctx, IR::Inst* inst) {
EmitRSqrtStepFused<32>(code, ctx, inst);
}
void EmitX64::EmitFPVectorRSqrtStepFused64(EmitContext& ctx, IR::Inst* inst) {
EmitRSqrtStepFused<64>(code, ctx, inst);
}
void EmitX64::EmitFPVectorSqrt32(EmitContext& ctx, IR::Inst* inst) {
EmitTwoOpVectorOperation<32, DefaultIndexer>(code, ctx, inst, [this](const Xbyak::Xmm& result, const Xbyak::Xmm& operand) {
code.sqrtps(result, operand);
});
}
void EmitX64::EmitFPVectorSqrt64(EmitContext& ctx, IR::Inst* inst) {
EmitTwoOpVectorOperation<64, DefaultIndexer>(code, ctx, inst, [this](const Xbyak::Xmm& result, const Xbyak::Xmm& operand) {
code.sqrtpd(result, operand);
});
}
void EmitX64::EmitFPVectorSub32(EmitContext& ctx, IR::Inst* inst) {
EmitThreeOpVectorOperation<32, DefaultIndexer>(code, ctx, inst, &Xbyak::CodeGenerator::subps);
}
void EmitX64::EmitFPVectorSub64(EmitContext& ctx, IR::Inst* inst) {
EmitThreeOpVectorOperation<64, DefaultIndexer>(code, ctx, inst, &Xbyak::CodeGenerator::subpd);
}
template<size_t fsize, bool unsigned_>
void EmitFPVectorToFixed(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst) {
using FPT = mp::unsigned_integer_of_size<fsize>;
const size_t fbits = inst->GetArg(1).GetU8();
const auto rounding = static_cast<FP::RoundingMode>(inst->GetArg(2).GetU8());
// TODO: AVX512 implementation
if (code.DoesCpuSupport(Xbyak::util::Cpu::tSSE41) && rounding != FP::RoundingMode::ToNearest_TieAwayFromZero) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm src = ctx.reg_alloc.UseScratchXmm(args[0]);
const int round_imm = [&]{
switch (rounding) {
case FP::RoundingMode::ToNearest_TieEven:
default:
return 0b00;
case FP::RoundingMode::TowardsPlusInfinity:
return 0b10;
case FP::RoundingMode::TowardsMinusInfinity:
return 0b01;
case FP::RoundingMode::TowardsZero:
return 0b11;
}
}();
const auto perform_conversion = [&code, &ctx](const Xbyak::Xmm& src) {
// MSVC doesn't allow us to use a [&] capture, so we have to do this instead.
(void)ctx;
if constexpr (fsize == 32) {
code.cvttps2dq(src, src);
} else {
const Xbyak::Reg64 hi = ctx.reg_alloc.ScratchGpr();
const Xbyak::Reg64 lo = ctx.reg_alloc.ScratchGpr();
code.cvttsd2si(lo, src);
code.punpckhqdq(src, src);
code.cvttsd2si(hi, src);
code.movq(src, lo);
code.pinsrq(src, hi, 1);
ctx.reg_alloc.Release(hi);
ctx.reg_alloc.Release(lo);
}
};
if (fbits != 0) {
const u64 scale_factor = fsize == 32
? static_cast<u64>(fbits + 127) << 23
: static_cast<u64>(fbits + 1023) << 52;
FCODE(mulp)(src, GetVectorOf<fsize>(code, scale_factor));
}
FCODE(roundp)(src, src, static_cast<u8>(round_imm));
ZeroIfNaN<fsize>(code, src);
constexpr u64 float_upper_limit_signed = fsize == 32 ? 0x4f000000 : 0x43e0000000000000;
[[maybe_unused]] constexpr u64 float_upper_limit_unsigned = fsize == 32 ? 0x4f800000 : 0x43f0000000000000;
if constexpr (unsigned_) {
// Zero is minimum
code.xorps(xmm0, xmm0);
FCODE(cmplep)(xmm0, src);
FCODE(andp)(src, xmm0);
// Will we exceed unsigned range?
const Xbyak::Xmm exceed_unsigned = ctx.reg_alloc.ScratchXmm();
code.movaps(exceed_unsigned, GetVectorOf<fsize, float_upper_limit_unsigned>(code));
FCODE(cmplep)(exceed_unsigned, src);
// Will be exceed signed range?
const Xbyak::Xmm tmp = ctx.reg_alloc.ScratchXmm();
code.movaps(tmp, GetVectorOf<fsize, float_upper_limit_signed>(code));
code.movaps(xmm0, tmp);
FCODE(cmplep)(xmm0, src);
FCODE(andp)(tmp, xmm0);
FCODE(subp)(src, tmp);
perform_conversion(src);
if constexpr (fsize == 32) {
code.pslld(xmm0, 31);
} else {
code.psllq(xmm0, 63);
}
FCODE(orp)(src, xmm0);
// Saturate to max
FCODE(orp)(src, exceed_unsigned);
} else {
constexpr u64 integer_max = static_cast<FPT>(std::numeric_limits<std::conditional_t<unsigned_, FPT, std::make_signed_t<FPT>>>::max());
code.movaps(xmm0, GetVectorOf<fsize, float_upper_limit_signed>(code));
FCODE(cmplep)(xmm0, src);
perform_conversion(src);
FCODE(blendvp)(src, GetVectorOf<fsize, integer_max>(code));
}
ctx.reg_alloc.DefineValue(inst, src);
return;
}
using fbits_list = mp::vllift<std::make_index_sequence<fsize + 1>>;
using rounding_list = mp::list<
std::integral_constant<FP::RoundingMode, FP::RoundingMode::ToNearest_TieEven>,
std::integral_constant<FP::RoundingMode, FP::RoundingMode::TowardsPlusInfinity>,
std::integral_constant<FP::RoundingMode, FP::RoundingMode::TowardsMinusInfinity>,
std::integral_constant<FP::RoundingMode, FP::RoundingMode::TowardsZero>,
std::integral_constant<FP::RoundingMode, FP::RoundingMode::ToNearest_TieAwayFromZero>
>;
using key_type = std::tuple<size_t, FP::RoundingMode>;
using value_type = void(*)(VectorArray<FPT>&, const VectorArray<FPT>&, FP::FPCR, FP::FPSR&);
static const auto lut = mp::GenerateLookupTableFromList<key_type, value_type>(
[](auto arg) {
return std::pair<key_type, value_type>{
mp::to_tuple<decltype(arg)>,
static_cast<value_type>(
[](VectorArray<FPT>& output, const VectorArray<FPT>& input, FP::FPCR fpcr, FP::FPSR& fpsr) {
constexpr size_t fbits = std::get<0>(mp::to_tuple<decltype(arg)>);
constexpr FP::RoundingMode rounding_mode = std::get<1>(mp::to_tuple<decltype(arg)>);
for (size_t i = 0; i < output.size(); ++i) {
output[i] = static_cast<FPT>(FP::FPToFixed<FPT>(fsize, input[i], fbits, unsigned_, fpcr, rounding_mode, fpsr));
}
}
)
};
},
mp::cartesian_product<fbits_list, rounding_list>{}
);
EmitTwoOpFallback(code, ctx, inst, lut.at(std::make_tuple(fbits, rounding)));
}
void EmitX64::EmitFPVectorToSignedFixed32(EmitContext& ctx, IR::Inst* inst) {
EmitFPVectorToFixed<32, false>(code, ctx, inst);
}
void EmitX64::EmitFPVectorToSignedFixed64(EmitContext& ctx, IR::Inst* inst) {
EmitFPVectorToFixed<64, false>(code, ctx, inst);
}
void EmitX64::EmitFPVectorToUnsignedFixed32(EmitContext& ctx, IR::Inst* inst) {
EmitFPVectorToFixed<32, true>(code, ctx, inst);
}
void EmitX64::EmitFPVectorToUnsignedFixed64(EmitContext& ctx, IR::Inst* inst) {
EmitFPVectorToFixed<64, true>(code, ctx, inst);
}
} // namespace Dynarmic::BackendX64