VERY experimental Nintendo Switch 2 emulator
C++
[dynarmic] fix pre-SSE4.1 having errors on CMHI/CMLO, fix extra nuisances and add INTERP testcase (#4025)
lizziead9af2502712
diff --git a/src/common/x64/xbyak.h b/src/common/x64/xbyak.hindex 5ee12576e9..e541215994 100644--- a/src/common/x64/xbyak.h+++ b/src/common/x64/xbyak.h@@ -262,20 +262,16 @@ enum {CMP_ORD = 7,};-constexpr bool IsWithin2G(uintptr_t ref, uintptr_t target) {- const u64 distance = target - (ref + 5);- return !(distance >= 0x8000'0000ULL && distance <= ~0x8000'0000ULL);-}--inline bool IsWithin2G(const Xbyak::CodeGenerator& code, uintptr_t target) {- return IsWithin2G(reinterpret_cast<uintptr_t>(code.getCurr()), target);+constexpr bool IsWithin2G(uintptr_t ref, uintptr_t target) noexcept {+ u64 const distance = target - (ref + 5);+ return (distance & 0xffff'ffff) == distance;}template <typename T>inline void CallFarFunction(Xbyak::CodeGenerator& code, const T f) {static_assert(std::is_pointer_v<T>, "Argument must be a (function) pointer.");- size_t addr = reinterpret_cast<size_t>(f);- if (IsWithin2G(code, addr)) {+ uintptr_t addr = uintptr_t(f);+ if (IsWithin2G(uintptr_t(code.getCurr()), addr)) {code.call(f);} else {// ABI_RETURN is a safe temp register to use before a calldiff --git a/src/dynarmic/src/dynarmic/backend/x64/block_of_code.h b/src/dynarmic/src/dynarmic/backend/x64/block_of_code.hindex 857b1a4484..d7502d2b83 100644--- a/src/dynarmic/src/dynarmic/backend/x64/block_of_code.h+++ b/src/dynarmic/src/dynarmic/backend/x64/block_of_code.h@@ -13,8 +13,9 @@#include <memory>#include <type_traits>-#include "dynarmic/mcl/bit.hpp"#include "common/common_types.h"+#include "common/x64/xbyak.h"+#include "dynarmic/mcl/bit.hpp"#include "dynarmic/backend/x64/xbyak.h"#include "dynarmic/backend/x64/abi.h"#include "dynarmic/backend/x64/callback.h"@@ -82,28 +83,17 @@ public:/// Code emitter: Load required flags for conditional cond from rax into host rflagsvoid LoadRequiredFlagsForCondFromRax(IR::Cond cond);- /// Code emitter: Calls the function- template<typename FunctionPointer>- void CallFunction(FunctionPointer fn) {- static_assert(std::is_pointer_v<FunctionPointer> && std::is_function_v<std::remove_pointer_t<FunctionPointer>>,- "Supplied type must be a pointer to a function");-- const u64 address = reinterpret_cast<u64>(fn);- const u64 distance = address - (getCurr<u64>() + 5);-- if (distance >= 0x0000000080000000ULL && distance < 0xFFFFFFFF80000000ULL) {- // Far call- mov(rax, address);- call(rax);- } else {- call(fn);- }+ /// @brief Code emitter: Calls the function+ template<typename F>+ void CallFunction(F fn) {+ static_assert(std::is_pointer_v<F> && std::is_function_v<std::remove_pointer_t<F>>, "Supplied type must be a pointer to a function");+ ::Common::X64::CallFarFunction(*this, fn);}- /// Code emitter: Calls the lambda. Lambda must not have any captures.+ /// @brief Code emitter: Calls the lambda. Lambda must not have any captures.template<typename Lambda>void CallLambda(Lambda l) {- CallFunction(Common::FptrCast(l));+ ::Common::X64::CallFarFunction(*this, Common::FptrCast(l));}void ZeroExtendFrom(size_t bitsize, Xbyak::Reg64 reg) {diff --git a/src/dynarmic/src/dynarmic/backend/x64/emit_x64_vector.cpp b/src/dynarmic/src/dynarmic/backend/x64/emit_x64_vector.cppindex d2edea780c..4df09e4797 100644--- a/src/dynarmic/src/dynarmic/backend/x64/emit_x64_vector.cpp+++ b/src/dynarmic/src/dynarmic/backend/x64/emit_x64_vector.cpp@@ -2103,36 +2103,57 @@ void EmitX64::EmitVectorMaxU64(EmitContext& ctx, IR::Inst* inst) {} else if (code.HasHostFeature(HostFeature::AVX)) {auto const x = ctx.reg_alloc.UseScratchXmm(code, args[0]);auto const y = ctx.reg_alloc.UseXmm(code, args[1]);- auto const tmp = ctx.reg_alloc.ScratchXmm(code);- code.vmovdqa(xmm0, code.Const(xword, 0x8000000000000000, 0x8000000000000000));- code.vpsubq(tmp, y, xmm0);- code.vpsubq(xmm0, x, xmm0);- code.vpcmpgtq(xmm0, tmp, xmm0);- code.pblendvb(x, y);+ auto const tmp0 = ctx.reg_alloc.ScratchXmm(code);+ auto const tmp1 = ctx.reg_alloc.ScratchXmm(code, HostLoc::XMM0);+ code.vmovdqa(tmp1, code.Const(xword, 0x8000000000000000, 0x8000000000000000));+ code.vpsubq(tmp0, y, tmp1);+ code.vpsubq(tmp1, x, tmp1);+ code.vpcmpgtq(tmp1, tmp0, tmp1);+ code.pblendvb(x, y); // XMM0 is implicitctx.reg_alloc.DefineValue(code, inst, x);+ } else if (code.HasHostFeature(HostFeature::SSE41)) {+ auto const tmp0 = ctx.reg_alloc.UseScratchXmm(code, args[0]);+ auto const tmp1 = ctx.reg_alloc.UseXmm(code, args[1]);+ auto const tmp2 = ctx.reg_alloc.ScratchXmm(code);+ auto const tmp3 = ctx.reg_alloc.ScratchXmm(code);+ auto const tmp4 = ctx.reg_alloc.ScratchXmm(code);+ auto const tax = ctx.reg_alloc.ScratchGpr(code);+ auto const tdx = ctx.reg_alloc.ScratchGpr(code);+ code.movq(tdx, tmp1);+ code.movq(tax, tmp0);+ code.movhlps(tmp3, tmp0);+ code.cmp(tdx, tax);+ code.movhlps(tmp2, tmp1);+ code.cmovnb(tax, tdx);+ code.movq(tdx, tmp2);+ code.pinsrq(tmp4, tax, 0);+ code.movq(tax, tmp3);+ code.cmp(tdx, tax);+ code.cmovnb(tax, tdx);+ code.pinsrq(tmp4, tax, 1);+ ctx.reg_alloc.DefineValue(code, inst, tmp4);} else {auto const tmp0 = ctx.reg_alloc.UseScratchXmm(code, args[0]);auto const tmp1 = ctx.reg_alloc.UseXmm(code, args[1]);auto const tmp2 = ctx.reg_alloc.ScratchXmm(code);auto const tmp3 = ctx.reg_alloc.ScratchXmm(code);auto const tmp4 = ctx.reg_alloc.ScratchXmm(code);- auto const tmp5 = ctx.reg_alloc.ScratchXmm(code);- code.movdqa(tmp2, code.Const(xword, 0x8000000080000000, 0x8000000080000000));- code.movdqa(tmp3, tmp1);- code.pxor(tmp3, tmp2);- code.pxor(tmp2, tmp0);- code.movdqa(tmp4, tmp2);- code.pcmpeqd(tmp2, tmp3);- code.pcmpgtd(tmp4, tmp3);- code.pshufd(tmp2, tmp2, 245);- code.pshufd(tmp5, tmp4, 160);- code.pshufd(tmp3, tmp4, 245);- code.pand(tmp2, tmp5);- code.por(tmp3, tmp2);- code.pand(tmp0, tmp3);- code.pandn(tmp3, tmp1);- code.por(tmp0, tmp3);- ctx.reg_alloc.DefineValue(code, inst, tmp0);+ auto const tax = ctx.reg_alloc.ScratchGpr(code);+ auto const tdx = ctx.reg_alloc.ScratchGpr(code);+ code.movq(tdx, tmp1);+ code.movq(tax, tmp0);+ code.movhlps(tmp3, tmp0);+ code.cmp(tdx, tax);+ code.movhlps(tmp2, tmp1);+ code.cmovnb(tax, tdx);+ code.movq(tdx, tmp2);+ code.movq(tmp4, tax);+ code.movq(tax, tmp3);+ code.cmp(tdx, tax);+ code.cmovnb(tax, tdx);+ code.movq(tmp3, tax);+ code.punpcklqdq(tmp4, tmp3);+ ctx.reg_alloc.DefineValue(code, inst, tmp4);}}@@ -2247,37 +2268,57 @@ void EmitX64::EmitVectorMinU64(EmitContext& ctx, IR::Inst* inst) {} else if (code.HasHostFeature(HostFeature::AVX)) {auto const x = ctx.reg_alloc.UseXmm(code, args[0]);auto const y = ctx.reg_alloc.UseScratchXmm(code, args[1]);- auto const tmp = ctx.reg_alloc.ScratchXmm(code);- code.vmovdqa(xmm0, code.Const(xword, 0x8000000000000000, 0x8000000000000000));- code.vpsubq(tmp, y, xmm0);- code.vpsubq(xmm0, x, xmm0);- code.vpcmpgtq(xmm0, tmp, xmm0);+ auto const tmp0 = ctx.reg_alloc.ScratchXmm(code);+ auto const tmp1 = ctx.reg_alloc.ScratchXmm(code, HostLoc::XMM0);+ code.vmovdqa(tmp1, code.Const(xword, 0x8000000000000000, 0x8000000000000000));+ code.vpsubq(tmp0, y, tmp1);+ code.vpsubq(tmp1, x, tmp1);+ code.vpcmpgtq(tmp1, tmp0, tmp1);code.pblendvb(y, x);ctx.reg_alloc.DefineValue(code, inst, y);+ } else if (code.HasHostFeature(HostFeature::SSE41)) {+ auto const tmp0 = ctx.reg_alloc.UseScratchXmm(code, args[0]);+ auto const tmp1 = ctx.reg_alloc.UseXmm(code, args[1]);+ auto const tmp2 = ctx.reg_alloc.ScratchXmm(code);+ auto const tmp3 = ctx.reg_alloc.ScratchXmm(code);+ auto const tmp4 = ctx.reg_alloc.ScratchXmm(code);+ auto const tax = ctx.reg_alloc.ScratchGpr(code);+ auto const tdx = ctx.reg_alloc.ScratchGpr(code);+ code.movq(tdx, tmp1);+ code.movq(tax, tmp0);+ code.movhlps(tmp3, tmp0);+ code.cmp(tdx, tax);+ code.movhlps(tmp2, tmp1);+ code.cmovbe(tax, tdx);+ code.movq(tdx, tmp2);+ code.pinsrq(tmp4, tax, 0);+ code.movq(tax, tmp3);+ code.cmp(tdx, tax);+ code.cmovbe(tax, tdx);+ code.pinsrq(tmp4, tax, 1);+ ctx.reg_alloc.DefineValue(code, inst, tmp4);} else {auto const tmp0 = ctx.reg_alloc.UseScratchXmm(code, args[0]);auto const tmp1 = ctx.reg_alloc.UseXmm(code, args[1]);auto const tmp2 = ctx.reg_alloc.ScratchXmm(code);auto const tmp3 = ctx.reg_alloc.ScratchXmm(code);auto const tmp4 = ctx.reg_alloc.ScratchXmm(code);- auto const tmp5 = ctx.reg_alloc.ScratchXmm(code);- code.movdqa(tmp2, code.Const(xword, 0x8000000080000000, 0x8000000080000000));- code.movdqa(tmp3, tmp1);- code.pxor(tmp3, tmp2);- code.pxor(tmp2, tmp0);- code.movdqa(tmp4, tmp2);- code.pcmpeqd(tmp2, tmp3);- code.pcmpgtd(tmp4, tmp3);- code.pshufd(tmp3, tmp2, 245);- code.pshufd(tmp5, tmp4, 160);- code.pshufd(tmp2, tmp4, 245);- code.pand(tmp3, tmp5);- code.por(tmp2, tmp3);- code.pand(tmp1, tmp2);- code.pandn(tmp2, tmp0);- code.por(tmp2, tmp1);- //code.movdqa(tmp0, tmp2);- ctx.reg_alloc.DefineValue(code, inst, tmp2);+ auto const tax = ctx.reg_alloc.ScratchGpr(code);+ auto const tdx = ctx.reg_alloc.ScratchGpr(code);+ code.movq(tdx, tmp1);+ code.movq(tax, tmp0);+ code.movhlps(tmp3, tmp0);+ code.cmp(tdx, tax);+ code.movhlps(tmp2, tmp1);+ code.cmovbe(tax, tdx);+ code.movq(tdx, tmp2);+ code.movq(tmp4, tax);+ code.movq(tax, tmp3);+ code.cmp(tdx, tax);+ code.cmovbe(tax, tdx);+ code.movq(tmp3, tax);+ code.punpcklqdq(tmp4, tmp3);+ ctx.reg_alloc.DefineValue(code, inst, tmp4);}}@@ -2471,13 +2512,11 @@ void EmitX64::EmitVectorNarrow64(EmitContext& ctx, IR::Inst* inst) {ctx.reg_alloc.DefineValue(code, inst, result);} else {- auto const a = ctx.reg_alloc.UseScratchXmm(code, args[0]);- auto const zeros = ctx.reg_alloc.ScratchXmm(code);-- code.pxor(zeros, zeros);- code.shufps(a, zeros, 0b00001000);-- ctx.reg_alloc.DefineValue(code, inst, a);+ auto const a = ctx.reg_alloc.UseScratchXmm(code, args[0]);+ auto const zeros = ctx.reg_alloc.ScratchXmm(code);+ code.pxor(zeros, zeros);+ code.shufps(a, zeros, 0b00001000);+ ctx.reg_alloc.DefineValue(code, inst, a);}}@@ -2490,11 +2529,11 @@ void EmitX64::EmitVectorNot(EmitContext& ctx, IR::Inst* inst) {code.vpternlogq(result, operand, operand, u8(~Tern::c));ctx.reg_alloc.DefineValue(code, inst, result);} else {- auto const xmm_a = ctx.reg_alloc.UseScratchXmm(code, args[0]);- auto const xmm_b = ctx.reg_alloc.ScratchXmm(code);- code.pcmpeqw(xmm_b, xmm_b);- code.pxor(xmm_a, xmm_b);- ctx.reg_alloc.DefineValue(code, inst, xmm_a);+ auto const xmm_a = ctx.reg_alloc.UseScratchXmm(code, args[0]);+ auto const xmm_b = ctx.reg_alloc.ScratchXmm(code);+ code.pcmpeqw(xmm_b, xmm_b);+ code.pxor(xmm_a, xmm_b);+ ctx.reg_alloc.DefineValue(code, inst, xmm_a);}}@@ -2504,11 +2543,9 @@ void EmitX64::EmitVectorOr(EmitContext& ctx, IR::Inst* inst) {void EmitX64::EmitVectorPairedAddLower8(EmitContext& ctx, IR::Inst* inst) {auto args = ctx.reg_alloc.GetArgumentInfo(inst);-auto const xmm_a = ctx.reg_alloc.UseScratchXmm(code, args[0]);auto const xmm_b = ctx.reg_alloc.UseXmm(code, args[1]);auto const tmp = ctx.reg_alloc.ScratchXmm(code);-code.punpcklqdq(xmm_a, xmm_b);code.movdqa(tmp, xmm_a);code.psllw(xmm_a, 8);@@ -2516,7 +2553,6 @@ void EmitX64::EmitVectorPairedAddLower8(EmitContext& ctx, IR::Inst* inst) {code.pxor(tmp, tmp);code.psrlw(xmm_a, 8);code.packuswb(xmm_a, tmp);-ctx.reg_alloc.DefineValue(code, inst, xmm_a);}diff --git a/src/dynarmic/src/dynarmic/common/llvm_disassemble.cpp b/src/dynarmic/src/dynarmic/common/llvm_disassemble.cppindex 6aff6d6cc7..6896b43d62 100644--- a/src/dynarmic/src/dynarmic/common/llvm_disassemble.cpp+++ b/src/dynarmic/src/dynarmic/common/llvm_disassemble.cpp@@ -7,7 +7,6 @@*/#include <string>-#include <fmt/format.h>#ifdef DYNARMIC_USE_LLVM@@ -16,7 +15,6 @@#endif#include "common/assert.h"-#include <bit>#include "common/common_types.h"#include "dynarmic/common/llvm_disassemble.h"diff --git a/src/dynarmic/tests/A64/a64.cpp b/src/dynarmic/tests/A64/a64.cppindex 9bd9c36a86..99cc9d3d4d 100644--- a/src/dynarmic/tests/A64/a64.cpp+++ b/src/dynarmic/tests/A64/a64.cpp@@ -2587,11 +2587,11 @@ TEST_CASE("A64: Manual Vector Min/Max U64 (Optimizer Test)", "[a64]") {// MaxU64 pattern: (a > b) ? a : bcode.CMHI(V2.D2(), V0.D2(), V1.D2()); // V2 = Mask (A > B)- code.BSL(V2.B16(), V0.B16(), V1.B16()); // V2 = Resul+ code.BSL(V2.B16(), V0.B16(), V1.B16()); // V2 = Result// MinU64 pattern: (a > b) ? b : acode.CMHI(V3.D2(), V0.D2(), V1.D2()); // V3 = Mask (A > B)- code.BSL(V3.B16(), V1.B16(), V0.B16()); // V3 = Resul+ code.BSL(V3.B16(), V1.B16(), V0.B16()); // V3 = Resultjit.SetPC(0);jit.SetVector(0, {100, 20});@@ -2603,3 +2603,32 @@ TEST_CASE("A64: Manual Vector Min/Max U64 (Optimizer Test)", "[a64]") {CHECK(jit.GetVector(2) == Vector{100, 200});CHECK(jit.GetVector(3) == Vector{50, 20});}++TEST_CASE("A64: Rounding", "[a64]") {+ A64TestEnv env;+ A64::UserConfig jit_user_config{};+ jit_user_config.callbacks = &env;+ A64::Jit jit{jit_user_config};++ oaknut::VectorCodeGenerator code{env.code_mem, nullptr};++ code.FRINTN(V1.S4(), V0.S4()); // ToNearest_TieEven+ code.FRINTM(V2.S4(), V0.S4()); // TowardsMinusInfinity+ code.FRINTP(V3.S4(), V0.S4()); // TowardsPlusInfinity+ code.FRINTZ(V4.S4(), V0.S4()); // TowardsZero+ code.FRINTA(V5.S4(), V0.S4()); // ToNearest_TieAwayFromZero+ code.FRINTX(V6.S4(), V0.S4()); // ToNearest_TieAwayFromZero++ jit.SetPC(0);+ jit.SetVector(0, {0x4001e17c4001e17c, 0x4001e17c4001e17c});+ env.ticks_left = env.code_mem.size();+ CheckedRun([&]() { jit.Run(); });++ CHECK(jit.GetVector(0) == Vector{0x4001e17c4001e17c, 0x4001e17c4001e17c});+ CHECK(jit.GetVector(1) == Vector{0x4000000040000000, 0x4000000040000000});+ CHECK(jit.GetVector(2) == Vector{0x4000000040000000, 0x4000000040000000});+ CHECK(jit.GetVector(3) == Vector{0x4040000040400000, 0x4040000040400000});+ CHECK(jit.GetVector(4) == Vector{0x4000000040000000, 0x4000000040000000});+ CHECK(jit.GetVector(5) == Vector{0x4000000040000000, 0x4000000040000000});+ CHECK(jit.GetVector(6) == Vector{0x4000000040000000, 0x4000000040000000});+}diff --git a/src/dynarmic/tests/A64/testenv.h b/src/dynarmic/tests/A64/testenv.hindex c25790e1c9..1d260f5460 100644--- a/src/dynarmic/tests/A64/testenv.h+++ b/src/dynarmic/tests/A64/testenv.h@@ -17,24 +17,19 @@ using Vector = Dynarmic::A64::Vector;class A64TestEnv : public Dynarmic::A64::UserCallbacks {public:+ ankerl::unordered_dense::map<u64, u8> modified_memory;+ std::vector<u32> code_mem;u64 ticks_left = 0;-- bool code_mem_modified_by_guest = false;u64 code_mem_start_address = 0;- std::vector<u32> code_mem;-- ankerl::unordered_dense::map<u64, u8> modified_memory;- std::vector<std::string> interrupts;+ bool code_mem_modified_by_guest = false;bool IsInCodeMem(u64 vaddr) const {return vaddr >= code_mem_start_address && vaddr < code_mem_start_address + code_mem.size() * 4;}std::optional<std::uint32_t> MemoryReadCode(u64 vaddr) override {- if (!IsInCodeMem(vaddr)) {+ if (!IsInCodeMem(vaddr))return 0x14000000; // B .- }-const size_t index = (vaddr - code_mem_start_address) / 4;return code_mem[index];}@@ -43,10 +38,9 @@ public:if (IsInCodeMem(vaddr)) {return reinterpret_cast<u8*>(code_mem.data())[vaddr - code_mem_start_address];}- if (auto iter = modified_memory.find(vaddr); iter != modified_memory.end()) {- return iter->second;- }- return static_cast<u8>(vaddr);+ if (auto const it = modified_memory.find(vaddr); it != modified_memory.end())+ return it->second;+ return u8(vaddr);}std::uint16_t MemoryRead16(u64 vaddr) override {return u16(MemoryRead8(vaddr)) | u16(MemoryRead8(vaddr + 1)) << 8;@@ -68,16 +62,16 @@ public:modified_memory[vaddr] = value;}void MemoryWrite16(u64 vaddr, std::uint16_t value) override {- MemoryWrite8(vaddr, static_cast<u8>(value));- MemoryWrite8(vaddr + 1, static_cast<u8>(value >> 8));+ MemoryWrite8(vaddr, u8(value));+ MemoryWrite8(vaddr + 1, u8(value >> 8));}void MemoryWrite32(u64 vaddr, std::uint32_t value) override {- MemoryWrite16(vaddr, static_cast<u16>(value));- MemoryWrite16(vaddr + 2, static_cast<u16>(value >> 16));+ MemoryWrite16(vaddr, u16(value));+ MemoryWrite16(vaddr + 2, u16(value >> 16));}void MemoryWrite64(u64 vaddr, std::uint64_t value) override {- MemoryWrite32(vaddr, static_cast<u32>(value));- MemoryWrite32(vaddr + 4, static_cast<u32>(value >> 32));+ MemoryWrite32(vaddr, u32(value));+ MemoryWrite32(vaddr + 4, u32(value >> 32));}void MemoryWrite128(u64 vaddr, Vector value) override {MemoryWrite64(vaddr, value[0]);@@ -139,12 +133,12 @@ public:template<typename T>T read(u64 vaddr) {T value;- memcpy(&value, backing_memory + vaddr, sizeof(T));+ std::memcpy(&value, backing_memory + vaddr, sizeof(T));return value;}template<typename T>void write(u64 vaddr, const T& value) {- memcpy(backing_memory + vaddr, &value, sizeof(T));+ std::memcpy(backing_memory + vaddr, &value, sizeof(T));}std::optional<std::uint32_t> MemoryReadCode(u64 vaddr) override {