Windows: Added Windows ARM support to JUCE

modules/juce_audio_basics/buffers/juce_FloatVectorOperations.cpp

@@ -1430,34 +1430,44 @@ void JUCE_CALLTYPE FloatVectorOperations::convertFixedToFloat (float* dest, cons
 intptr_t JUCE_CALLTYPE FloatVectorOperations::getFpStatusRegister() noexcept
 {
     intptr_t fpsr = 0;
-  #if JUCE_INTEL && JUCE_USE_SSE_INTRINSICS
+ #if JUCE_INTEL && JUCE_USE_SSE_INTRINSICS
     fpsr = static_cast<intptr_t> (_mm_getcsr());
-  #elif defined(__arm64__) || defined(__aarch64__) || JUCE_USE_ARM_NEON
-   #if defined(__arm64__) || defined(__aarch64__)
+ #elif (JUCE_64BIT && JUCE_ARM) || JUCE_USE_ARM_NEON
+  #if _MSC_VER
+    // _control87 returns static values for x86 bits that don't exist on arm
+    // to emulate x86 behaviour. We are only ever interested in de-normal bits
+    // so mask out only those.
+    fpsr = (intptr_t) (_control87 (0, 0) & _MCW_DN);
+  #else
+   #if JUCE_64BIT
     asm volatile("mrs %0, fpcr"
                  : "=r"(fpsr));
    #elif JUCE_USE_ARM_NEON
     asm volatile("vmrs %0, fpscr"
                  : "=r"(fpsr));
    #endif
-  #else
-   #if ! (defined(JUCE_INTEL) || defined(JUCE_ARM))
-    jassertfalse; // No support for getting the floating point status register for your platform
-   #endif
   #endif
+ #else
+  #if ! (defined (JUCE_INTEL) || defined (JUCE_ARM))
+    jassertfalse; // No support for getting the floating point status register for your platform
+  #endif
+ #endif
 
     return fpsr;
 }
 
 void JUCE_CALLTYPE FloatVectorOperations::setFpStatusRegister (intptr_t fpsr) noexcept
 {
-  #if JUCE_INTEL && JUCE_USE_SSE_INTRINSICS
+ #if JUCE_INTEL && JUCE_USE_SSE_INTRINSICS
     // the volatile keyword here is needed to workaround a bug in AppleClang 13.0
     // which aggressively optimises away the variable otherwise
     volatile auto fpsr_w = static_cast<uint32_t> (fpsr);
     _mm_setcsr (fpsr_w);
-  #elif defined(__arm64__) || defined(__aarch64__) || JUCE_USE_ARM_NEON
-   #if defined(__arm64__) || defined(__aarch64__)
+ #elif (JUCE_64BIT && JUCE_ARM) || JUCE_USE_ARM_NEON
+  #if _MSC_VER
+    _control87 ((unsigned int) fpsr, _MCW_DN);
+  #else
+   #if JUCE_64BIT
     asm volatile("msr fpcr, %0"
                  :
                  : "ri"(fpsr));
@@ -1466,17 +1476,18 @@ void JUCE_CALLTYPE FloatVectorOperations::setFpStatusRegister (intptr_t fpsr) no
                  :
                  : "ri"(fpsr));
    #endif
-  #else
-   #if ! (defined(JUCE_INTEL) || defined(JUCE_ARM))
-    jassertfalse; // No support for getting the floating point status register for your platform
-   #endif
-    ignoreUnused (fpsr);
   #endif
+ #else
+  #if ! (defined (JUCE_INTEL) || defined (JUCE_ARM))
+    jassertfalse; // No support for getting the floating point status register for your platform
+  #endif
+    ignoreUnused (fpsr);
+ #endif
 }
 
 void JUCE_CALLTYPE FloatVectorOperations::enableFlushToZeroMode (bool shouldEnable) noexcept
 {
-  #if JUCE_USE_SSE_INTRINSICS || (JUCE_USE_ARM_NEON || defined(__arm64__) || defined(__aarch64__))
+  #if JUCE_USE_SSE_INTRINSICS || (JUCE_USE_ARM_NEON || (JUCE_64BIT && JUCE_ARM))
    #if JUCE_USE_SSE_INTRINSICS
     intptr_t mask = _MM_FLUSH_ZERO_MASK;
    #else /*JUCE_USE_ARM_NEON*/
@@ -1484,7 +1495,7 @@ void JUCE_CALLTYPE FloatVectorOperations::enableFlushToZeroMode (bool shouldEnab
    #endif
     setFpStatusRegister ((getFpStatusRegister() & (~mask)) | (shouldEnable ? mask : 0));
   #else
-   #if ! (defined(JUCE_INTEL) || defined(JUCE_ARM))
+   #if ! (defined (JUCE_INTEL) || defined (JUCE_ARM))
     jassertfalse; // No support for flush to zero mode on your platform
    #endif
     ignoreUnused (shouldEnable);
@@ -1493,7 +1504,7 @@ void JUCE_CALLTYPE FloatVectorOperations::enableFlushToZeroMode (bool shouldEnab
 
 void JUCE_CALLTYPE FloatVectorOperations::disableDenormalisedNumberSupport (bool shouldDisable) noexcept
 {
-  #if JUCE_USE_SSE_INTRINSICS || (JUCE_USE_ARM_NEON || defined(__arm64__) || defined(__aarch64__))
+  #if JUCE_USE_SSE_INTRINSICS || (JUCE_USE_ARM_NEON || (JUCE_64BIT && JUCE_ARM))
    #if JUCE_USE_SSE_INTRINSICS
     intptr_t mask = 0x8040;
    #else /*JUCE_USE_ARM_NEON*/
@@ -1504,7 +1515,7 @@ void JUCE_CALLTYPE FloatVectorOperations::disableDenormalisedNumberSupport (bool
   #else
     ignoreUnused (shouldDisable);
 
-   #if ! (defined(JUCE_INTEL) || defined(JUCE_ARM))
+   #if ! (defined (JUCE_INTEL) || defined (JUCE_ARM))
     jassertfalse; // No support for disable denormals mode on your platform
    #endif
   #endif
@@ -1512,7 +1523,7 @@ void JUCE_CALLTYPE FloatVectorOperations::disableDenormalisedNumberSupport (bool
 
 bool JUCE_CALLTYPE FloatVectorOperations::areDenormalsDisabled() noexcept
 {
-  #if JUCE_USE_SSE_INTRINSICS || (JUCE_USE_ARM_NEON || defined(__arm64__) || defined(__aarch64__))
+  #if JUCE_USE_SSE_INTRINSICS || (JUCE_USE_ARM_NEON || (JUCE_64BIT && JUCE_ARM))
    #if JUCE_USE_SSE_INTRINSICS
     intptr_t mask = 0x8040;
    #else /*JUCE_USE_ARM_NEON*/
@@ -1527,7 +1538,7 @@ bool JUCE_CALLTYPE FloatVectorOperations::areDenormalsDisabled() noexcept
 
 ScopedNoDenormals::ScopedNoDenormals() noexcept
 {
-  #if JUCE_USE_SSE_INTRINSICS || (JUCE_USE_ARM_NEON || defined(__arm64__) || defined(__aarch64__))
+  #if JUCE_USE_SSE_INTRINSICS || (JUCE_USE_ARM_NEON || (JUCE_64BIT && JUCE_ARM))
    #if JUCE_USE_SSE_INTRINSICS
     intptr_t mask = 0x8040;
    #else /*JUCE_USE_ARM_NEON*/
@@ -1541,7 +1552,7 @@ ScopedNoDenormals::ScopedNoDenormals() noexcept
 
 ScopedNoDenormals::~ScopedNoDenormals() noexcept
 {
-   #if JUCE_USE_SSE_INTRINSICS || (JUCE_USE_ARM_NEON || defined(__arm64__) || defined(__aarch64__))
+   #if JUCE_USE_SSE_INTRINSICS || (JUCE_USE_ARM_NEON || (JUCE_64BIT && JUCE_ARM))
     FloatVectorOperations::setFpStatusRegister (fpsr);
    #endif
 }

modules/juce_audio_basics/buffers/juce_FloatVectorOperations.h

@@ -261,7 +261,7 @@ public:
     ~ScopedNoDenormals() noexcept;
 
 private:
-  #if JUCE_USE_SSE_INTRINSICS || (JUCE_USE_ARM_NEON || defined (__arm64__) || defined (__aarch64__))
+  #if JUCE_USE_SSE_INTRINSICS || (JUCE_USE_ARM_NEON || (JUCE_64BIT && JUCE_ARM))
     intptr_t fpsr;
   #endif
 };

modules/juce_audio_formats/codecs/oggvorbis/libvorbis-1.3.7/lib/os.h

@@ -119,7 +119,7 @@ static inline int vorbis_ftoi(double f){  /* yes, double!  Otherwise,
 
 /* MSVC inline assembly. 32 bit only; inline ASM isn't implemented in the
  * 64 bit compiler and doesn't work on arm. */
-#if defined(_MSC_VER) && defined(_M_IX86) && !defined(_WIN32_WCE)
+#if (defined(_MSC_VER) && defined(_M_IX86) && !defined(_WIN32_WCE)) && (! JUCE_ARM)
 #  define VORBIS_FPU_CONTROL
 
 typedef ogg_int16_t vorbis_fpu_control;
@@ -146,7 +146,7 @@ static __inline void vorbis_fpu_restore(vorbis_fpu_control fpu){
 
 /* Optimized code path for x86_64 builds. Uses SSE2 intrinsics. This can be
    done safely because all x86_64 CPUs supports SSE2. */
-#if ((JUCE_MSVC && JUCE_64BIT) || (JUCE_GCC && defined (__x86_64__)))
+#if (((JUCE_MSVC && JUCE_64BIT) || (JUCE_GCC && defined (__x86_64__)))) && (! JUCE_ARM)
 #  define VORBIS_FPU_CONTROL
 
 typedef ogg_int16_t vorbis_fpu_control;

modules/juce_core/native/juce_win32_SystemStats.cpp

@@ -23,11 +23,6 @@
 namespace juce
 {
 
-#if JUCE_MSVC && ! defined (__INTEL_COMPILER)
- #pragma intrinsic (__cpuid)
- #pragma intrinsic (__rdtsc)
-#endif
-
 void Logger::outputDebugString (const String& text)
 {
     OutputDebugString ((text + "\n").toWideCharPointer());
@@ -39,9 +34,50 @@ void Logger::outputDebugString (const String& text)
  JUCE_API void  juceDLL_free (void* block)    { std::free (block); }
 #endif
 
-//==============================================================================
+static int findNumberOfPhysicalCores() noexcept
+{
+   #if JUCE_MINGW
+    // Not implemented in MinGW
+    jassertfalse;
 
-#if JUCE_MINGW || JUCE_CLANG
+    return 1;
+   #else
+
+    DWORD bufferSize = 0;
+    GetLogicalProcessorInformation (nullptr, &bufferSize);
+
+    const auto numBuffers = (size_t) (bufferSize / sizeof (SYSTEM_LOGICAL_PROCESSOR_INFORMATION));
+
+    if (numBuffers == 0)
+    {
+        jassertfalse;
+        return 0;
+    };
+
+    HeapBlock<SYSTEM_LOGICAL_PROCESSOR_INFORMATION> buffer (numBuffers);
+
+    if (! GetLogicalProcessorInformation (buffer, &bufferSize))
+    {
+        jassertfalse;
+        return 0;
+    }
+
+    return (int) std::count_if (buffer.get(), buffer.get() + numBuffers, [] (const auto& info)
+    {
+        return info.Relationship == RelationProcessorCore;
+    });
+
+   #endif // JUCE_MINGW
+}
+
+//==============================================================================
+#if JUCE_INTEL
+ #if JUCE_MSVC && ! defined (__INTEL_COMPILER)
+  #pragma intrinsic (__cpuid)
+  #pragma intrinsic (__rdtsc)
+ #endif
+
+ #if JUCE_MINGW || JUCE_CLANG
 static void callCPUID (int result[4], uint32 type)
 {
   uint32 la = (uint32) result[0], lb = (uint32) result[1],
@@ -59,12 +95,12 @@ static void callCPUID (int result[4], uint32 type)
   result[0] = (int) la; result[1] = (int) lb;
   result[2] = (int) lc; result[3] = (int) ld;
 }
-#else
+ #else
 static void callCPUID (int result[4], int infoType)
 {
     __cpuid (result, infoType);
 }
-#endif
+ #endif
 
 String SystemStats::getCpuVendor()
 {
@@ -103,34 +139,6 @@ String SystemStats::getCpuModel()
     return String (name).trim();
 }
 
-static int findNumberOfPhysicalCores() noexcept
-{
-   #if JUCE_MINGW
-    // Not implemented in MinGW
-    jassertfalse;
-
-    return 1;
-   #else
-
-    int numPhysicalCores = 0;
-    DWORD bufferSize = 0;
-    GetLogicalProcessorInformation (nullptr, &bufferSize);
-
-    if (auto numBuffers = (size_t) (bufferSize / sizeof (SYSTEM_LOGICAL_PROCESSOR_INFORMATION)))
-    {
-        HeapBlock<SYSTEM_LOGICAL_PROCESSOR_INFORMATION> buffer (numBuffers);
-
-        if (GetLogicalProcessorInformation (buffer, &bufferSize))
-            for (size_t i = 0; i < numBuffers; ++i)
-                if (buffer[i].Relationship == RelationProcessorCore)
-                    ++numPhysicalCores;
-    }
-
-    return numPhysicalCores;
-   #endif // JUCE_MINGW
-}
-
-//==============================================================================
 void CPUInformation::initialise() noexcept
 {
     int info[4] = { 0 };
@@ -176,6 +184,49 @@ void CPUInformation::initialise() noexcept
     if (numPhysicalCPUs <= 0)
         numPhysicalCPUs = numLogicalCPUs;
 }
+#elif JUCE_ARM
+String SystemStats::getCpuVendor()
+{
+    static const auto cpuVendor = []
+    {
+        static constexpr auto* path = "HKEY_LOCAL_MACHINE\\HARDWARE\\DESCRIPTION\\System\\CentralProcessor\\0\\VendorIdentifier";
+        auto vendor = RegistryKeyWrapper::getValue (path, {}, 0).trim();
+
+        return vendor.isEmpty() ? String ("Unknown Vendor") : vendor;
+    }();
+
+    return cpuVendor;
+}
+
+String SystemStats::getCpuModel()
+{
+    static const auto cpuModel = []
+    {
+        static constexpr auto* path = "HKEY_LOCAL_MACHINE\\HARDWARE\\DESCRIPTION\\System\\CentralProcessor\\0\\ProcessorNameString";
+        auto model = RegistryKeyWrapper::getValue (path, {}, 0).trim();
+
+        return model.isEmpty() ? String ("Unknown Model") : model;
+    }();
+
+    return cpuModel;
+}
+
+void CPUInformation::initialise() noexcept
+{
+    // Windows for arm requires at least armv7 which has neon support
+    hasNeon = true;
+
+    SYSTEM_INFO systemInfo;
+    GetNativeSystemInfo (&systemInfo);
+    numLogicalCPUs  = (int) systemInfo.dwNumberOfProcessors;
+    numPhysicalCPUs = findNumberOfPhysicalCores();
+
+    if (numPhysicalCPUs <= 0)
+        numPhysicalCPUs = numLogicalCPUs;
+}
+#else
+ #error Unknown CPU architecture type
+#endif
 
 #if JUCE_MSVC && JUCE_CHECK_MEMORY_LEAKS
 struct DebugFlagsInitialiser
@@ -442,11 +493,21 @@ double Time::getMillisecondCounterHiRes() noexcept       { return hiResCounterHa
 //==============================================================================
 static int64 juce_getClockCycleCounter() noexcept
 {
-   #if JUCE_MSVC
+ #if JUCE_MSVC
+  #if JUCE_INTEL
     // MS intrinsics version...
     return (int64) __rdtsc();
-
-   #elif JUCE_GCC || JUCE_CLANG
+  #elif JUCE_ARM
+   #if defined (_M_ARM)
+    return __rdpmccntr64();
+   #elif defined (_M_ARM64)
+    return _ReadStatusReg (ARM64_PMCCNTR_EL0);
+   #else
+    #error Unknown arm architecture
+   #endif
+  #endif
+ #elif JUCE_GCC || JUCE_CLANG
+  #if JUCE_INTEL
     // GNU inline asm version...
     unsigned int hi = 0, lo = 0;
 
@@ -462,9 +523,15 @@ static int64 juce_getClockCycleCounter() noexcept
          : "cc", "eax", "ebx", "ecx", "edx", "memory");
 
     return (int64) ((((uint64) hi) << 32) | lo);
-   #else
-    #error "unknown compiler?"
-   #endif
+  #elif JUCE_ARM
+    int64 retval;
+
+    __asm__ __volatile__ ("mrs %0, cntvct_el0" : "=r"(retval));
+    return retval;
+  #endif
+ #else
+  #error "unknown compiler?"
+ #endif
 }
 
 int SystemStats::getCpuSpeedInMegahertz()

modules/juce_core/system/juce_TargetPlatform.h

@@ -109,7 +109,11 @@
   /** If defined, this indicates that the processor is little-endian. */
   #define JUCE_LITTLE_ENDIAN 1
 
-  #define JUCE_INTEL 1
+  #if defined (_M_ARM) || defined (_M_ARM64) || defined (__arm__) || defined (__aarch64__)
+    #define JUCE_ARM 1
+  #else
+    #define JUCE_INTEL 1
+  #endif
 #endif
 
 //==============================================================================

modules/juce_dsp/juce_dsp.cpp

@@ -79,13 +79,13 @@
 #include "widgets/juce_Chorus.cpp"
 
 #if JUCE_USE_SIMD
- #if defined(__i386__) || defined(__amd64__) || defined(_M_X64) || defined(_X86_) || defined(_M_IX86)
+ #if JUCE_INTEL
   #ifdef __AVX2__
    #include "native/juce_avx_SIMDNativeOps.cpp"
   #else
    #include "native/juce_sse_SIMDNativeOps.cpp"
   #endif
- #elif defined(__arm__) || defined(_M_ARM) || defined (__arm64__) || defined (__aarch64__)
+ #elif JUCE_ARM
   #include "native/juce_neon_SIMDNativeOps.cpp"
  #else
   #error "SIMD register support not implemented for this platform"

modules/juce_dsp/juce_dsp.h

@@ -74,7 +74,9 @@
   #include <immintrin.h>
  #endif
 
-#elif defined (__ARM_NEON__) || defined (__ARM_NEON) || defined (__arm64__) || defined (__aarch64__)
+// it's ok to check for _M_ARM below as this is only defined on Windows for Arm 32-bit
+// which has a minimum requirement of armv7, which supports neon.
+#elif defined (__ARM_NEON__) || defined (__ARM_NEON) || defined (__arm64__) || defined (__aarch64__) || defined (_M_ARM) || defined (_M_ARM64)
 
  #ifndef JUCE_USE_SIMD
   #define JUCE_USE_SIMD 1
@@ -227,7 +229,7 @@ namespace juce
   #else
    #include "native/juce_sse_SIMDNativeOps.h"
   #endif
- #elif defined(__arm__) || defined(_M_ARM) || defined (__arm64__) || defined (__aarch64__)
+ #elif JUCE_ARM
   #include "native/juce_neon_SIMDNativeOps.h"
  #else
   #error "SIMD register support not implemented for this platform"