early-access version 1546

README.md

@@ -1,7 +1,7 @@
 yuzu emulator early access
 =============
 
-This is the source code for early-access 1543.
+This is the source code for early-access 1546.
 
 ## Legal Notice
 

externals/dynarmic/src/CMakeLists.txt

@@ -163,6 +163,7 @@ if ("A32" IN_LIST DYNARMIC_FRONTENDS)
         frontend/A32/translate/impl/thumb32_data_processing_shifted_register.cpp
         frontend/A32/translate/impl/thumb32_load_byte.cpp
         frontend/A32/translate/impl/thumb32_load_halfword.cpp
+        frontend/A32/translate/impl/thumb32_load_store_dual.cpp
         frontend/A32/translate/impl/thumb32_load_store_multiple.cpp
         frontend/A32/translate/impl/thumb32_load_word.cpp
         frontend/A32/translate/impl/thumb32_long_multiply.cpp

externals/dynarmic/src/backend/x64/block_of_code.cpp

@@ -13,6 +13,7 @@
 #include "backend/x64/block_of_code.h"
 #include "backend/x64/perf_map.h"
 #include "common/assert.h"
+#include "common/bit_util.h"
 
 #ifdef _WIN32
     #include <windows.h>
@@ -43,8 +44,8 @@ const std::array<Xbyak::Reg64, 6> BlockOfCode::ABI_PARAMS = {BlockOfCode::ABI_PA
 
 namespace {
 
-constexpr size_t TOTAL_CODE_SIZE = 128 * 1024 * 1024;
+constexpr size_t TOTAL_CODE_SIZE = 256 * 1024 * 1024;
-constexpr size_t FAR_CODE_OFFSET = 100 * 1024 * 1024;
+constexpr size_t FAR_CODE_OFFSET = 200 * 1024 * 1024;
 constexpr size_t CONSTANT_POOL_SIZE = 2 * 1024 * 1024;
 
 class CustomXbyakAllocator : public Xbyak::Allocator {
@@ -364,7 +365,21 @@ bool BlockOfCode::HasBMI2() const {
 }
 
 bool BlockOfCode::HasFastBMI2() const {
-    return DoesCpuSupport(Xbyak::util::Cpu::tBMI2) && !DoesCpuSupport(Xbyak::util::Cpu::tAMD);
+    if (DoesCpuSupport(Xbyak::util::Cpu::tBMI2)) {
+        // BMI2 instructions such as pdep and pext have been very slow up until Zen 3.
+        // Check for Zen 3 or newer by its family (0x19).
+        // See also: https://en.wikichip.org/wiki/amd/cpuid
+        if (DoesCpuSupport(Xbyak::util::Cpu::tAMD)) {
+            std::array<u32, 4> data{};
+            cpu_info.getCpuid(1, data.data());
+            const u32 family_base     = Common::Bits< 8, 11>(data[0]);
+            const u32 family_extended = Common::Bits<20, 27>(data[0]);
+            const u32 family = family_base + family_extended;
+            return family >= 0x19;
+        }
+        return true;
+    }
+    return false;
 }
 
 bool BlockOfCode::HasFMA() const {

externals/dynarmic/src/backend/x64/emit_x64_vector_floating_point.cpp

@@ -337,8 +337,12 @@ void EmitTwoOpVectorOperation(BlockOfCode& code, EmitContext& ctx, IR::Inst* ins
     ctx.reg_alloc.DefineValue(inst, result);
 }
 
+enum CheckInputNaN {
+    Yes, No,
+};
+
 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()) {
+void EmitThreeOpVectorOperation(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst, Function fn, CheckInputNaN check_input_nan = CheckInputNaN::No, 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");
 
     auto args = ctx.reg_alloc.GetArgumentInfo(inst);
@@ -371,15 +375,31 @@ void EmitThreeOpVectorOperation(BlockOfCode& code, EmitContext& ctx, IR::Inst* i
     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 (check_input_nan == CheckInputNaN::Yes) {
+        if (code.HasAVX()) {
+            FCODE(vcmpunordp)(nan_mask, xmm_a, xmm_b);
+        } else {
+            code.movaps(nan_mask, xmm_b);
+            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);
+
+    if (check_input_nan == CheckInputNaN::Yes) {
+        FCODE(cmpunordp)(nan_mask, result);
+    } else if (code.HasAVX()) {
+        FCODE(vcmpunordp)(nan_mask, result, result);
+    } else {
+        code.movaps(nan_mask, result);
+        FCODE(cmpunordp)(nan_mask, nan_mask);
+    }
 
     HandleNaNs<fsize, 2>(code, ctx, fpcr_controlled, {result, xmm_a, xmm_b}, nan_mask, nan_handler);
 
@@ -951,7 +971,7 @@ static void EmitFPVectorMinMax(BlockOfCode& code, EmitContext& ctx, IR::Inst* in
             code.andnps(mask, eq);
             code.orps(result, mask);
         }
-    });
+    }, CheckInputNaN::Yes);
 }
 
 void EmitX64::EmitFPVectorMax32(EmitContext& ctx, IR::Inst* inst) {

externals/dynarmic/src/frontend/A32/decoder/thumb32.inc

@@ -13,17 +13,17 @@ INST(thumb32_LDMDB,          "LDMDB/LDMEA",              "1110100100W1nnnniiiiii
 // Load/Store Dual, Load/Store Exclusive, Table Branch
 //INST(thumb32_STREX,          "STREX",                    "111010000100--------------------")
 //INST(thumb32_LDREX,          "LDREX",                    "111010000101--------------------")
-//INST(thumb32_STRD_imm_1,     "STRD (imm)",               "11101000-110--------------------")
+INST(thumb32_STRD_imm_1,     "STRD (imm)",               "11101000U110nnnnttttssssiiiiiiii")
-//INST(thumb32_STRD_imm_2,     "STRD (imm)",               "11101001-1-0--------------------")
+INST(thumb32_STRD_imm_2,     "STRD (imm)",               "11101001U1W0nnnnttttssssiiiiiiii")
-//INST(thumb32_LDRD_imm_1,     "LDRD (lit)",               "11101000-1111111----------------")
+INST(thumb32_LDRD_lit_1,     "LDRD (lit)",               "11101000U1111111ttttssssiiiiiiii")
-//INST(thumb32_LDRD_imm_2,     "LDRD (lit)",               "11101001-1-11111----------------")
+INST(thumb32_LDRD_lit_2,     "LDRD (lit)",               "11101001U1W11111ttttssssiiiiiiii")
-//INST(thumb32_LDRD_imm_1,     "LDRD (imm)",               "11101000-111--------------------")
+INST(thumb32_LDRD_imm_1,     "LDRD (imm)",               "11101000U111nnnnttttssssiiiiiiii")
-//INST(thumb32_LDRD_imm_2,     "LDRD (imm)",               "11101001-1-1--------------------")
+INST(thumb32_LDRD_imm_2,     "LDRD (imm)",               "11101001U1W1nnnnttttssssiiiiiiii")
 //INST(thumb32_STREXB,         "STREXB",                   "111010001100------------0100----")
 //INST(thumb32_STREXH,         "STREXH",                   "111010001100------------0101----")
 //INST(thumb32_STREXD,         "STREXD",                   "111010001100------------0111----")
-//INST(thumb32_TBB,            "TBB",                      "111010001101------------0000----")
+INST(thumb32_TBB,            "TBB",                      "111010001101nnnn111100000000mmmm")
-//INST(thumb32_TBH,            "TBH",                      "111010001101------------0001----")
+INST(thumb32_TBH,            "TBH",                      "111010001101nnnn111100000001mmmm")
 //INST(thumb32_LDREXB,         "LDREXB",                   "111010001101------------0100----")
 //INST(thumb32_LDREXH,         "LDREXH",                   "111010001101------------0101----")
 //INST(thumb32_LDREXD,         "LDREXD",                   "111010001101------------0111----")

externals/dynarmic/src/frontend/A32/translate/impl/thumb32_load_store_dual.cpp

@@ -0,0 +1,146 @@
+/* This file is part of the dynarmic project.
+ * Copyright (c) 2021 MerryMage
+ * SPDX-License-Identifier: 0BSD
+ */
+
+#include "common/bit_util.h"
+#include "frontend/A32/translate/impl/translate_thumb.h"
+
+namespace Dynarmic::A32 {
+static bool ITBlockCheck(const A32::IREmitter& ir) {
+    return ir.current_location.IT().IsInITBlock() && !ir.current_location.IT().IsLastInITBlock();
+}
+
+static bool TableBranch(ThumbTranslatorVisitor& v, Reg n, Reg m, bool half) {
+    if (m == Reg::PC) {
+        return v.UnpredictableInstruction();
+    }
+    if (ITBlockCheck(v.ir)) {
+        return v.UnpredictableInstruction();
+    }
+
+    const auto reg_m = v.ir.GetRegister(m);
+    const auto reg_n = v.ir.GetRegister(n);
+
+    IR::U32 halfwords;
+    if (half) {
+        const auto data = v.ir.ReadMemory16(v.ir.Add(reg_n, v.ir.LogicalShiftLeft(reg_m, v.ir.Imm8(1))));
+        halfwords = v.ir.ZeroExtendToWord(data);
+    } else {
+        halfwords = v.ir.ZeroExtendToWord(v.ir.ReadMemory8(v.ir.Add(reg_n, reg_m)));
+    }
+
+    const auto current_pc = v.ir.Imm32(v.ir.PC());
+    const auto branch_value = v.ir.Add(current_pc, v.ir.Add(halfwords, halfwords));
+
+    v.ir.UpdateUpperLocationDescriptor();
+    v.ir.BranchWritePC(branch_value);
+    v.ir.SetTerm(IR::Term::FastDispatchHint{});
+    return false;
+}
+
+static bool LoadDualImmediate(ThumbTranslatorVisitor& v, bool P, bool U, bool W,
+                              Reg n, Reg t, Reg t2, Imm<8> imm8) {
+    if (W && (n == t || n == t2)) {
+        return v.UnpredictableInstruction();
+    }
+    if (t == Reg::PC || t2 == Reg::PC || t == t2) {
+        return v.UnpredictableInstruction();
+    }
+
+    const u32 imm = imm8.ZeroExtend() << 2;
+    const IR::U32 reg_n = v.ir.GetRegister(n);
+    const IR::U32 offset_address = U ? v.ir.Add(reg_n, v.ir.Imm32(imm))
+                                     : v.ir.Sub(reg_n, v.ir.Imm32(imm));
+    const IR::U32 address_1 = P ? offset_address
+                                : reg_n;
+    const IR::U32 address_2 = v.ir.Add(address_1, v.ir.Imm32(4));
+
+    v.ir.SetRegister(t, v.ir.ReadMemory32(address_1));
+    v.ir.SetRegister(t2, v.ir.ReadMemory32(address_2));
+
+    if (W) {
+        v.ir.SetRegister(n, offset_address);
+    }
+    return true;
+}
+
+static bool LoadDualLiteral(ThumbTranslatorVisitor& v, bool U, bool W, Reg t, Reg t2, Imm<8> imm8) {
+    if (t == Reg::PC || t2 == Reg::PC || t == t2) {
+        return v.UnpredictableInstruction();
+    }
+    if (W) {
+        return v.UnpredictableInstruction();
+    }
+
+    const auto imm = imm8.ZeroExtend() << 2;
+    const auto address_1 = U ? v.ir.Add(v.ir.Imm32(v.ir.AlignPC(4)), v.ir.Imm32(imm))
+                             : v.ir.Sub(v.ir.Imm32(v.ir.AlignPC(4)), v.ir.Imm32(imm));
+    const auto address_2 = v.ir.Add(address_1, v.ir.Imm32(4));
+
+    v.ir.SetRegister(t, v.ir.ReadMemory32(address_1));
+    v.ir.SetRegister(t2, v.ir.ReadMemory32(address_2));
+    return true;
+}
+
+static bool StoreDual(ThumbTranslatorVisitor& v, bool P, bool U, bool W, Reg n, Reg t, Reg t2, Imm<8> imm8) {
+    if (W && (n == t || n == t2)) {
+        return v.UnpredictableInstruction();
+    }
+    if (n == Reg::PC || t == Reg::PC || t2 == Reg::PC) {
+        return v.UnpredictableInstruction();
+    }
+
+    const u32 imm = imm8.ZeroExtend() << 2;
+    const IR::U32 reg_n = v.ir.GetRegister(n);
+    const IR::U32 reg_t = v.ir.GetRegister(t);
+    const IR::U32 reg_t2 = v.ir.GetRegister(t2);
+
+    const IR::U32 offset_address = U ? v.ir.Add(reg_n, v.ir.Imm32(imm))
+                                     : v.ir.Sub(reg_n, v.ir.Imm32(imm));
+    const IR::U32 address_1 = P ? offset_address
+                                : reg_n;
+    const IR::U32 address_2 = v.ir.Add(address_1, v.ir.Imm32(4));
+
+    v.ir.WriteMemory32(address_1, reg_t);
+    v.ir.WriteMemory32(address_2, reg_t2);
+
+    if (W) {
+        v.ir.SetRegister(n, offset_address);
+    }
+    return true;
+}
+
+bool ThumbTranslatorVisitor::thumb32_LDRD_imm_1(bool U, Reg n, Reg t, Reg t2, Imm<8> imm8) {
+    return LoadDualImmediate(*this, false, U, true, n, t, t2, imm8);
+}
+
+bool ThumbTranslatorVisitor::thumb32_LDRD_imm_2(bool U, bool W, Reg n, Reg t, Reg t2, Imm<8> imm8) {
+    return LoadDualImmediate(*this, true, U, W, n, t, t2, imm8);
+}
+
+bool ThumbTranslatorVisitor::thumb32_LDRD_lit_1(bool U, Reg t, Reg t2, Imm<8> imm8) {
+    return LoadDualLiteral(*this, U, true, t, t2, imm8);
+}
+
+bool ThumbTranslatorVisitor::thumb32_LDRD_lit_2(bool U, bool W, Reg t, Reg t2, Imm<8> imm8) {
+    return LoadDualLiteral(*this, U, W, t, t2, imm8);
+}
+
+bool ThumbTranslatorVisitor::thumb32_STRD_imm_1(bool U, Reg n, Reg t, Reg t2, Imm<8> imm8) {
+    return StoreDual(*this, false, U, true, n, t, t2, imm8);
+}
+
+bool ThumbTranslatorVisitor::thumb32_STRD_imm_2(bool U, bool W, Reg n, Reg t, Reg t2, Imm<8> imm8) {
+    return StoreDual(*this, true, U, W, n, t, t2, imm8);
+}
+
+bool ThumbTranslatorVisitor::thumb32_TBB(Reg n, Reg m) {
+    return TableBranch(*this, n, m, false);
+}
+
+bool ThumbTranslatorVisitor::thumb32_TBH(Reg n, Reg m) {
+    return TableBranch(*this, n, m, true);
+}
+
+} // namespace Dynarmic::A32

externals/dynarmic/src/frontend/A32/translate/impl/translate_thumb.h

@@ -179,6 +179,16 @@ struct ThumbTranslatorVisitor final {
     bool thumb32_STMIA(bool W, Reg n, Imm<15> reg_list);
     bool thumb32_STMDB(bool W, Reg n, Imm<15> reg_list);
 
+    // thumb32 load/store dual, load/store exclusive, table branch instructions
+    bool thumb32_LDRD_imm_1(bool U, Reg n, Reg t, Reg t2, Imm<8> imm8);
+    bool thumb32_LDRD_imm_2(bool U, bool W, Reg n, Reg t, Reg t2, Imm<8> imm8);
+    bool thumb32_LDRD_lit_1(bool U, Reg t, Reg t2, Imm<8> imm8);
+    bool thumb32_LDRD_lit_2(bool U, bool W, Reg t, Reg t2, Imm<8> imm8);
+    bool thumb32_STRD_imm_1(bool U, Reg n, Reg t, Reg t2, Imm<8> imm8);
+    bool thumb32_STRD_imm_2(bool U, bool W, Reg n, Reg t, Reg t2, Imm<8> imm8);
+    bool thumb32_TBB(Reg n, Reg m);
+    bool thumb32_TBH(Reg n, Reg m);
+
     // thumb32 data processing (shifted register) instructions
     bool thumb32_TST_reg(Reg n, Imm<3> imm3, Imm<2> imm2, ShiftType type, Reg m);
     bool thumb32_AND_reg(bool S, Reg n, Imm<3> imm3, Reg d, Imm<2> imm2, ShiftType type, Reg m);