Compile fixes for Windows-on-ARM64

Source/Core/Common/Arm64Emitter.cpp

@@ -364,6 +364,8 @@ void ARM64XEmitter::FlushIcacheSection(u8* start, u8* end)
 #if defined(IOS)
   // Header file says this is equivalent to: sys_icache_invalidate(start, end - start);
   sys_cache_control(kCacheFunctionPrepareForExecution, start, end - start);
+#elif defined(WIN32)
+  FlushInstructionCache(GetCurrentProcess(), start, end - start);
 #else
   // Don't rely on GCC's __clear_cache implementation, as it caches
   // icache/dcache cache line sizes, that can vary between cores on
@@ -2172,6 +2174,8 @@ void ARM64XEmitter::ABI_PopRegisters(BitSet32 registers, BitSet32 ignore_mask)
   ARM64Reg second;
   if (!(num_regs & 1))
     second = (ARM64Reg)(X0 + *it++);
+  else
+    second = {};
 
   // 8 byte per register, but 16 byte alignment, so we may have to padd one register.
   // Only update the SP on the last load to avoid the dependency between those loads.
@@ -4164,20 +4168,19 @@ void ARM64XEmitter::ANDSI2R(ARM64Reg Rd, ARM64Reg Rn, u64 imm, ARM64Reg scratch)
 void ARM64XEmitter::AddImmediate(ARM64Reg Rd, ARM64Reg Rn, u64 imm, bool shift, bool negative,
                                  bool flags)
 {
-  switch ((negative << 1) | flags)
+  if (!negative)
   {
-  case 0:
+    if (!flags)
       ADD(Rd, Rn, imm, shift);
-    break;
+    else
-  case 1:
       ADDS(Rd, Rn, imm, shift);
-    break;
+  }
-  case 2:
+  else
+  {
+    if (!flags)
       SUB(Rd, Rn, imm, shift);
-    break;
+    else
-  case 3:
       SUBS(Rd, Rn, imm, shift);
-    break;
   }
 }
 
@@ -4185,7 +4188,7 @@ void ARM64XEmitter::ADDI2R_internal(ARM64Reg Rd, ARM64Reg Rn, u64 imm, bool nega
                                     ARM64Reg scratch)
 {
   bool has_scratch = scratch != INVALID_REG;
-  u64 imm_neg = Is64Bit(Rd) ? -imm : -imm & 0xFFFFFFFFuLL;
+  u64 imm_neg = Is64Bit(Rd) ? u64(-s64(imm)) : u64(-s64(imm)) & 0xFFFFFFFFuLL;
   bool neg_neg = negative ? false : true;
 
   // Fast paths, aarch64 immediate instructions
@@ -4232,20 +4235,19 @@ void ARM64XEmitter::ADDI2R_internal(ARM64Reg Rd, ARM64Reg Rn, u64 imm, bool nega
              (u32)imm);
 
   negative ^= MOVI2R2(scratch, imm, imm_neg);
-  switch ((negative << 1) | flags)
+  if (!negative)
   {
-  case 0:
+    if (!flags)
       ADD(Rd, Rn, scratch);
-    break;
+    else
-  case 1:
       ADDS(Rd, Rn, scratch);
-    break;
+  }
-  case 2:
+  else
+  {
+    if (!flags)
       SUB(Rd, Rn, scratch);
-    break;
+    else
-  case 3:
       SUBS(Rd, Rn, scratch);
-    break;
   }
 }
 

Source/Core/Common/ArmCPUDetect.cpp

@@ -2,13 +2,17 @@
 // Licensed under GPLv2+
 // Refer to the license.txt file included.
 
-#include <asm/hwcap.h>
 #include <cstring>
 #include <fstream>
 #include <sstream>
 #include <string>
+#include <thread>
+
+#ifndef _WIN32
+#include <asm/hwcap.h>
 #include <sys/auxv.h>
 #include <unistd.h>
+#endif
 
 #include <fmt/format.h>
 
@@ -16,6 +20,8 @@
 #include "Common/CommonTypes.h"
 #include "Common/FileUtil.h"
 
+#ifndef WIN32
+
 const char procfile[] = "/proc/cpuinfo";
 
 static std::string GetCPUString()
@@ -42,6 +48,8 @@ static std::string GetCPUString()
   return cpu_string;
 }
 
+#endif
+
 CPUInfo cpu_info;
 
 CPUInfo::CPUInfo()
@@ -60,6 +68,21 @@ void CPUInfo::Detect()
   Mode64bit = true;
   vendor = CPUVendor::ARM;
 
+#ifdef _WIN32
+  num_cores = std::thread::hardware_concurrency();
+
+  // Windows does not provide any mechanism for querying the system registers on ARMv8, unlike Linux
+  // which traps the register reads and emulates them in the kernel. There are environment variables
+  // containing some of the CPU-specific values, which we could use for a lookup table in the
+  // future. For now, assume all features are present as all known devices which are Windows-on-ARM
+  // compatible also support these extensions.
+  bFP = true;
+  bASIMD = true;
+  bAES = true;
+  bCRC32 = true;
+  bSHA1 = true;
+  bSHA2 = true;
+#else
   // Get the information about the CPU
   num_cores = sysconf(_SC_NPROCESSORS_CONF);
   strncpy(cpu_string, GetCPUString().c_str(), sizeof(cpu_string));
@@ -71,6 +94,7 @@ void CPUInfo::Detect()
   bCRC32 = hwcaps & HWCAP_CRC32;
   bSHA1 = hwcaps & HWCAP_SHA1;
   bSHA2 = hwcaps & HWCAP_SHA2;
+#endif
 }
 
 // Turn the CPU info into a string we can show

Source/Core/Common/Hash.cpp

@@ -11,7 +11,7 @@
 #include "Common/CommonFuncs.h"
 #include "Common/Intrinsics.h"
 
-#ifdef _M_ARM_64
+#if defined(_M_ARM_64) && !defined(_MSC_VER)
 #include <arm_acle.h>
 #endif
 

Source/Core/Core/MachineContext.h

@@ -33,6 +33,10 @@ typedef CONTEXT SContext;
 #define CTX_R14 R14
 #define CTX_R15 R15
 #define CTX_RIP Rip
+#elif _M_ARM64
+#define CTX_REG(x) X[x]
+#define CTX_SP Sp
+#define CTX_PC Pc
 #else
 #error No context definition for architecture
 #endif

Source/Core/Core/PowerPC/JitArm64/Jit.cpp

@@ -754,9 +754,9 @@ void JitArm64::DoJit(u32 em_address, JitBlock* b, u32 nextPC)
         LDR(INDEX_UNSIGNED, WA, PPC_REG, PPCSTATE_OFF(msr));
         FixupBranch b1 = TBNZ(WA, 13);  // Test FP enabled bit
 
-        FixupBranch far = B();
+        FixupBranch far_addr = B();
         SwitchToFarCode();
-        SetJumpTarget(far);
+        SetJumpTarget(far_addr);
 
         gpr.Flush(FLUSH_MAINTAIN_STATE);
         fpr.Flush(FLUSH_MAINTAIN_STATE);

Source/Core/Core/PowerPC/JitArm64/JitArm64_Branch.cpp

@@ -143,9 +143,9 @@ void JitArm64::bcx(UGeckoInstruction inst)
         JumpIfCRFieldBit(inst.BI >> 2, 3 - (inst.BI & 3), !(inst.BO_2 & BO_BRANCH_IF_TRUE));
   }
 
-  FixupBranch far = B();
+  FixupBranch far_addr = B();
   SwitchToFarCode();
-  SetJumpTarget(far);
+  SetJumpTarget(far_addr);
 
   if (inst.LK)
   {
@@ -160,12 +160,12 @@ void JitArm64::bcx(UGeckoInstruction inst)
   if (js.op->branchIsIdleLoop)
   {
     // make idle loops go faster
-    ARM64Reg WA = gpr.GetReg();
+    ARM64Reg WA2 = gpr.GetReg();
-    ARM64Reg XA = EncodeRegTo64(WA);
+    ARM64Reg XA2 = EncodeRegTo64(WA2);
 
-    MOVP2R(XA, &CoreTiming::Idle);
+    MOVP2R(XA2, &CoreTiming::Idle);
-    BLR(XA);
+    BLR(XA2);
-    gpr.Unlock(WA);
+    gpr.Unlock(WA2);
 
     WriteExceptionExit(js.op->branchTo);
   }
@@ -260,9 +260,9 @@ void JitArm64::bclrx(UGeckoInstruction inst)
 
   if (conditional)
   {
-    FixupBranch far = B();
+    FixupBranch far_addr = B();
     SwitchToFarCode();
-    SetJumpTarget(far);
+    SetJumpTarget(far_addr);
   }
 
   LDR(INDEX_UNSIGNED, WA, PPC_REG, PPCSTATE_OFF(spr[SPR_LR]));

Source/Core/Core/PowerPC/JitArm64/JitArm64_FloatingPoint.cpp

@@ -35,7 +35,7 @@ void JitArm64::fp_arith(UGeckoInstruction inst)
   bool inputs_are_singles = fpr.IsSingle(a, !packed) && (!use_b || fpr.IsSingle(b, !packed)) &&
                             (!use_c || fpr.IsSingle(c, !packed));
 
-  ARM64Reg VA, VB, VC, VD;
+  ARM64Reg VA{}, VB{}, VC{}, VD{};
 
   if (packed)
   {

Source/Core/Core/PowerPC/JitArm64/JitArm64_Integer.cpp

@@ -368,7 +368,11 @@ void JitArm64::cntlzwx(UGeckoInstruction inst)
 
   if (gpr.IsImm(s))
   {
+#ifdef _MSC_VER
+    gpr.SetImmediate(a, _CountLeadingZeros(gpr.GetImm(s)));
+#else
     gpr.SetImmediate(a, __builtin_clz(gpr.GetImm(s)));
+#endif
     if (inst.Rc)
       ComputeRC0(gpr.GetImm(a));
   }
@@ -931,7 +935,7 @@ void JitArm64::subfex(UGeckoInstruction inst)
 
     // d = ~a + b + carry;
     if (gpr.IsImm(a))
-      MOVI2R(WA, ~gpr.GetImm(a));
+      MOVI2R(WA, u32(~gpr.GetImm(a)));
     else
       MVN(WA, gpr.R(a));
     ADCS(gpr.R(d), WA, gpr.R(b));
@@ -1187,7 +1191,7 @@ void JitArm64::divwx(UGeckoInstruction inst)
     if (inst.Rc)
       ComputeRC0(imm_d);
   }
-  else if (gpr.IsImm(b) && gpr.GetImm(b) != 0 && gpr.GetImm(b) != -1u)
+  else if (gpr.IsImm(b) && gpr.GetImm(b) != 0 && gpr.GetImm(b) != UINT32_C(0xFFFFFFFF))
   {
     ARM64Reg WA = gpr.GetReg();
     MOVI2R(WA, gpr.GetImm(b));

Source/Core/Core/PowerPC/JitArm64/JitArm64_LoadStore.cpp

@@ -408,7 +408,7 @@ void JitArm64::stX(UGeckoInstruction inst)
     gpr.BindToRegister(a, false);
 
     ARM64Reg WA = gpr.GetReg();
-    ARM64Reg RB;
+    ARM64Reg RB = {};
     ARM64Reg RA = gpr.R(a);
     if (regOffset != -1)
       RB = gpr.R(regOffset);
@@ -549,9 +549,9 @@ void JitArm64::dcbx(UGeckoInstruction inst)
   LSR(value, value, addr);  // move current bit to bit 0
 
   FixupBranch bit_not_set = TBZ(value, 0);
-  FixupBranch far = B();
+  FixupBranch far_addr = B();
   SwitchToFarCode();
-  SetJumpTarget(far);
+  SetJumpTarget(far_addr);
 
   BitSet32 gprs_to_push = gpr.GetCallerSavedUsed();
   BitSet32 fprs_to_push = fpr.GetCallerSavedUsed();
@@ -568,10 +568,10 @@ void JitArm64::dcbx(UGeckoInstruction inst)
   m_float_emit.ABI_PopRegisters(fprs_to_push, X30);
   ABI_PopRegisters(gprs_to_push);
 
-  FixupBranch near = B();
+  FixupBranch near_addr = B();
   SwitchToNearCode();
   SetJumpTarget(bit_not_set);
-  SetJumpTarget(near);
+  SetJumpTarget(near_addr);
 
   gpr.Unlock(addr, value, W30);
 }

Source/Core/Core/PowerPC/JitArm64/JitArm64_RegCache.cpp

@@ -149,6 +149,7 @@ Arm64GPRCache::GuestRegInfo Arm64GPRCache::GetGuestByIndex(size_t index)
   if (index >= GUEST_CR_OFFSET && index < GUEST_CR_OFFSET + GUEST_CR_COUNT)
     return GetGuestCR(index - GUEST_CR_OFFSET);
   ASSERT_MSG(DYNA_REC, false, "Invalid index for guest register");
+  return GetGuestGPR(0);
 }
 
 void Arm64GPRCache::FlushRegister(size_t index, bool maintain_state)
@@ -161,7 +162,7 @@ void Arm64GPRCache::FlushRegister(size_t index, bool maintain_state)
   {
     ARM64Reg host_reg = reg.GetReg();
     if (reg.IsDirty())
-      m_emit->STR(INDEX_UNSIGNED, host_reg, PPC_REG, guest_reg.ppc_offset);
+      m_emit->STR(INDEX_UNSIGNED, host_reg, PPC_REG, u32(guest_reg.ppc_offset));
 
     if (!maintain_state)
     {
@@ -173,14 +174,14 @@ void Arm64GPRCache::FlushRegister(size_t index, bool maintain_state)
   {
     if (!reg.GetImm())
     {
-      m_emit->STR(INDEX_UNSIGNED, bitsize == 64 ? ZR : WZR, PPC_REG, guest_reg.ppc_offset);
+      m_emit->STR(INDEX_UNSIGNED, bitsize == 64 ? ZR : WZR, PPC_REG, u32(guest_reg.ppc_offset));
     }
     else
     {
       ARM64Reg host_reg = bitsize != 64 ? GetReg() : EncodeRegTo64(GetReg());
 
       m_emit->MOVI2R(host_reg, reg.GetImm());
-      m_emit->STR(INDEX_UNSIGNED, host_reg, PPC_REG, guest_reg.ppc_offset);
+      m_emit->STR(INDEX_UNSIGNED, host_reg, PPC_REG, u32(guest_reg.ppc_offset));
 
       UnlockRegister(DecodeReg(host_reg));
     }
@@ -207,7 +208,7 @@ void Arm64GPRCache::FlushRegisters(BitSet32 regs, bool maintain_state)
           size_t ppc_offset = GetGuestByIndex(i).ppc_offset;
           ARM64Reg RX1 = R(GetGuestByIndex(i));
           ARM64Reg RX2 = R(GetGuestByIndex(i + 1));
-          m_emit->STP(INDEX_SIGNED, RX1, RX2, PPC_REG, ppc_offset);
+          m_emit->STP(INDEX_SIGNED, RX1, RX2, PPC_REG, u32(ppc_offset));
           if (!maintain_state)
           {
             UnlockRegister(DecodeReg(RX1));
@@ -285,7 +286,7 @@ ARM64Reg Arm64GPRCache::R(const GuestRegInfo& guest_reg)
     ARM64Reg host_reg = bitsize != 64 ? GetReg() : EncodeRegTo64(GetReg());
     reg.Load(host_reg);
     reg.SetDirty(false);
-    m_emit->LDR(INDEX_UNSIGNED, host_reg, PPC_REG, guest_reg.ppc_offset);
+    m_emit->LDR(INDEX_UNSIGNED, host_reg, PPC_REG, u32(guest_reg.ppc_offset));
     return host_reg;
   }
   break;
@@ -318,7 +319,7 @@ void Arm64GPRCache::BindToRegister(const GuestRegInfo& guest_reg, bool do_load)
     ARM64Reg host_reg = bitsize != 64 ? GetReg() : EncodeRegTo64(GetReg());
     reg.Load(host_reg);
     if (do_load)
-      m_emit->LDR(INDEX_UNSIGNED, host_reg, PPC_REG, guest_reg.ppc_offset);
+      m_emit->LDR(INDEX_UNSIGNED, host_reg, PPC_REG, u32(guest_reg.ppc_offset));
   }
 }
 
@@ -450,7 +451,7 @@ ARM64Reg Arm64FPRCache::R(size_t preg, RegType type)
       // Load the high 64bits from the file and insert them in to the high 64bits of the host
       // register
       ARM64Reg tmp_reg = GetReg();
-      m_float_emit->LDR(64, INDEX_UNSIGNED, tmp_reg, PPC_REG, PPCSTATE_OFF(ps[preg].ps1));
+      m_float_emit->LDR(64, INDEX_UNSIGNED, tmp_reg, PPC_REG, u32(PPCSTATE_OFF(ps[preg].ps1)));
       m_float_emit->INS(64, host_reg, 1, tmp_reg, 0);
       UnlockRegister(tmp_reg);
 
@@ -503,7 +504,8 @@ ARM64Reg Arm64FPRCache::R(size_t preg, RegType type)
       reg.Load(host_reg, REG_LOWER_PAIR);
     }
     reg.SetDirty(false);
-    m_float_emit->LDR(load_size, INDEX_UNSIGNED, host_reg, PPC_REG, PPCSTATE_OFF(ps[preg].ps0));
+    m_float_emit->LDR(load_size, INDEX_UNSIGNED, host_reg, PPC_REG,
+                      u32(PPCSTATE_OFF(ps[preg].ps0)));
     return host_reg;
   }
   default:
@@ -551,7 +553,7 @@ ARM64Reg Arm64FPRCache::RW(size_t preg, RegType type)
       // We are doing a full 128bit store because it takes 2 cycles on a Cortex-A57 to do a 128bit
       // store.
       // It would take longer to do an insert to a temporary and a 64bit store than to just do this.
-      m_float_emit->STR(128, INDEX_UNSIGNED, flush_reg, PPC_REG, PPCSTATE_OFF(ps[preg].ps0));
+      m_float_emit->STR(128, INDEX_UNSIGNED, flush_reg, PPC_REG, u32(PPCSTATE_OFF(ps[preg].ps0)));
       break;
     case REG_DUP_SINGLE:
       flush_reg = GetReg();
@@ -559,7 +561,7 @@ ARM64Reg Arm64FPRCache::RW(size_t preg, RegType type)
       [[fallthrough]];
     case REG_DUP:
       // Store PSR1 (which is equal to PSR0) in memory.
-      m_float_emit->STR(64, INDEX_UNSIGNED, flush_reg, PPC_REG, PPCSTATE_OFF(ps[preg].ps1));
+      m_float_emit->STR(64, INDEX_UNSIGNED, flush_reg, PPC_REG, u32(PPCSTATE_OFF(ps[preg].ps1)));
       break;
     default:
       // All other types doesn't store anything in PSR1.
@@ -684,7 +686,10 @@ void Arm64FPRCache::FlushRegister(size_t preg, bool maintain_state)
       store_size = 64;
 
     if (dirty)
-      m_float_emit->STR(store_size, INDEX_UNSIGNED, host_reg, PPC_REG, PPCSTATE_OFF(ps[preg].ps0));
+    {
+      m_float_emit->STR(store_size, INDEX_UNSIGNED, host_reg, PPC_REG,
+                        u32(PPCSTATE_OFF(ps[preg].ps0)));
+    }
 
     if (!maintain_state)
     {
@@ -700,8 +705,8 @@ void Arm64FPRCache::FlushRegister(size_t preg, bool maintain_state)
       // Too bad moving them would break savestate compatibility between x86_64 and AArch64
       // m_float_emit->STP(64, INDEX_SIGNED, host_reg, host_reg, PPC_REG,
       // PPCSTATE_OFF(ps[preg].ps0));
-      m_float_emit->STR(64, INDEX_UNSIGNED, host_reg, PPC_REG, PPCSTATE_OFF(ps[preg].ps0));
+      m_float_emit->STR(64, INDEX_UNSIGNED, host_reg, PPC_REG, u32(PPCSTATE_OFF(ps[preg].ps0)));
-      m_float_emit->STR(64, INDEX_UNSIGNED, host_reg, PPC_REG, PPCSTATE_OFF(ps[preg].ps1));
+      m_float_emit->STR(64, INDEX_UNSIGNED, host_reg, PPC_REG, u32(PPCSTATE_OFF(ps[preg].ps1)));
     }
 
     if (!maintain_state)

Source/Core/Core/PowerPC/JitArm64/JitArm64_SystemRegisters.cpp

@@ -32,6 +32,7 @@ FixupBranch JitArm64::JumpIfCRFieldBit(int field, int bit, bool jump_if_set)
     return jump_if_set ? TBNZ(XA, 62) : TBZ(XA, 62);
   default:
     ASSERT_MSG(DYNA_REC, false, "Invalid CR bit");
+    return {};
   }
 }
 
@@ -196,9 +197,9 @@ void JitArm64::twx(UGeckoInstruction inst)
     SetJumpTarget(fixup);
   }
 
-  FixupBranch far = B();
+  FixupBranch far_addr = B();
   SwitchToFarCode();
-  SetJumpTarget(far);
+  SetJumpTarget(far_addr);
 
   gpr.Flush(FlushMode::FLUSH_MAINTAIN_STATE);
   fpr.Flush(FlushMode::FLUSH_MAINTAIN_STATE);

Source/Core/VideoCommon/VertexLoaderARM64.cpp

@@ -379,7 +379,7 @@ void VertexLoaderARM64::GenerateVertexLoader()
   bool has_tc_scale = false;
   for (int i = 0; i < 8; i++)
   {
-    has_tc |= tc[i];
+    has_tc |= tc[i] != 0;
     has_tc_scale |= !!m_VtxAttr.texCoord[i].Frac;
   }
 

Source/Core/VideoCommon/VideoBackendBase.cpp

@@ -17,14 +17,21 @@
 #include "Core/ConfigManager.h"
 #include "Core/Core.h"
 
+// OpenGL is not available on Windows-on-ARM64
+#if !defined(_WIN32) || !defined(_M_ARM64)
+#define HAS_OPENGL 1
+#endif
+
 // TODO: ugly
 #ifdef _WIN32
 #include "VideoBackends/D3D/VideoBackend.h"
 #include "VideoBackends/D3D12/VideoBackend.h"
 #endif
 #include "VideoBackends/Null/VideoBackend.h"
+#ifdef HAS_OPENGL
 #include "VideoBackends/OGL/VideoBackend.h"
 #include "VideoBackends/Software/VideoBackend.h"
+#endif
 #include "VideoBackends/Vulkan/VideoBackend.h"
 
 #include "VideoCommon/AsyncRequests.h"
@@ -182,13 +189,17 @@ u16 VideoBackendBase::Video_GetBoundingBox(int index)
 void VideoBackendBase::PopulateList()
 {
   // OGL > D3D11 > Vulkan > SW > Null
+#ifdef HAS_OPENGL
   g_available_video_backends.push_back(std::make_unique<OGL::VideoBackend>());
+#endif
 #ifdef _WIN32
   g_available_video_backends.push_back(std::make_unique<DX11::VideoBackend>());
   g_available_video_backends.push_back(std::make_unique<DX12::VideoBackend>());
 #endif
   g_available_video_backends.push_back(std::make_unique<Vulkan::VideoBackend>());
+#ifdef HAS_OPENGL
   g_available_video_backends.push_back(std::make_unique<SW::VideoSoftware>());
+#endif
   g_available_video_backends.push_back(std::make_unique<Null::VideoBackend>());
 
   const auto iter =