[compiler-rt][X86] Match the detection of cpu's for __cpu_model to the latest version of gcc

[craig.topper]

Summary

We were missing many feature flags that newer gcc supports and we had our own set of feature flags that gcc didnt' support that were overlapping. Clang's implementation assumes gcc's features list so a mismatch here is problematic.

I've also matched the cpu type/subtype lists with gcc and removed all the cpus that gcc doesn't support. I've also removed the fallback autodetection logic that was taken from Host.cpp. It was the main reason we had extra feature flags relative to gcc. I don't think gcc does this in libgcc.

Unfortunately, libgcc seems to have ignored their own source code comment and has added new cpu types/subtypes in the middle of their enums at various times so I can't guarantee any compatibility with older libgcc here. But maybe once we have a matching implementation we can treat that as a reason to file bugs against libgcc if they do it again going forward.

Once this support is in place we can consider implementing __builtin_cpu_is in clang. This could also be needed for function dispatching that Erich Keane is working on.

[craig.topper]

Add back INTEL_CORE2 to the switch. Guess I got carried away with deleting.

[asbirlea]

Does it make sense to update this together with Host.cpp?

[craig.topper]

I can't remove the older processors from Host.cpp. Host.cpp also contains a newer processor called Goldmont that's not in libgcc yet. I think these files have slightly different goals.

[echristo]

In D35214#804024, @craig.topper wrote:

I can't remove the older processors from Host.cpp. Host.cpp also contains a newer processor called Goldmont that's not in libgcc yet. I think these files have slightly different goals.

Gah.

They're not supposed to fwiw.

I'll trust you here, but can you make sure that the code in Host.cpp is as up to date with all of this as you can make it? If nothing else Host.cpp should be a strict superset of this.

-eric

[craig.topper]

I've synced what I can of this into Host.cpp now. Host.cpp now has two feature variables now because just in matching libgcc we used 31 bits. The fall back CPU detection for Intel family 6 required 3 more bits. So I gave Host.cpp a second variable. Now getAvailableFeatures in Host.cpp returns both feature variables through outparams. I've changed getAvailableFeatures here to return its single feature variable as an outparam as well to keep them somewhat consistent.

[craig.topper]

Found a mistake in my previous patch around getAvailableFeatures where I forgot to still update the Features variable that gets passed to the AMD and Intel CPU detection

[asbirlea]

LGTM.
Thank you for the updates in Host.cpp too; the comments there are great for pointing future changes to libgcc/compiler-rt.

lib/builtins/cpu_model.c

 enum ProcessorVendors {
   VENDOR_INTEL = 1,
   VENDOR_AMD,
   VENDOR_OTHER,
   VENDOR_MAX
 };
 
 enum ProcessorTypes {
-  INTEL_ATOM = 1,
+  INTEL_BONNELL = 1,
   INTEL_CORE2,
   INTEL_COREI7,
   AMDFAM10H,
   AMDFAM15H,
-  INTEL_i386,
-  INTEL_i486,
-  INTEL_PENTIUM,
-  INTEL_PENTIUM_PRO,
-  INTEL_PENTIUM_II,
-  INTEL_PENTIUM_III,
-  INTEL_PENTIUM_IV,
-  INTEL_PENTIUM_M,
-  INTEL_CORE_DUO,
-  INTEL_XEONPHI,
-  INTEL_X86_64,
-  INTEL_NOCONA,
-  INTEL_PRESCOTT,
-  AMD_i486,
-  AMDPENTIUM,
-  AMDATHLON,
-  AMDFAM14H,
-  AMDFAM16H,
+  INTEL_SILVERMONT,
+  INTEL_KNL,
+  AMD_BTVER1,
+  AMD_BTVER2,
   AMDFAM17H,
   CPU_TYPE_MAX
 };
 
 enum ProcessorSubtypes {
   INTEL_COREI7_NEHALEM = 1,
   INTEL_COREI7_WESTMERE,
   INTEL_COREI7_SANDYBRIDGE,
   AMDFAM10H_BARCELONA,
   AMDFAM10H_SHANGHAI,
   AMDFAM10H_ISTANBUL,
   AMDFAM15H_BDVER1,
   AMDFAM15H_BDVER2,
-  INTEL_PENTIUM_MMX,
-  INTEL_CORE2_65,
-  INTEL_CORE2_45,
+  AMDFAM15H_BDVER3,
+  AMDFAM15H_BDVER4,
+  AMDFAM17H_ZNVER1,
   INTEL_COREI7_IVYBRIDGE,
   INTEL_COREI7_HASWELL,
   INTEL_COREI7_BROADWELL,
   INTEL_COREI7_SKYLAKE,
   INTEL_COREI7_SKYLAKE_AVX512,
-  INTEL_ATOM_BONNELL,
-  INTEL_ATOM_SILVERMONT,
-  INTEL_KNIGHTS_LANDING,
-  AMDPENTIUM_K6,
-  AMDPENTIUM_K62,
-  AMDPENTIUM_K63,
-  AMDPENTIUM_GEODE,
-  AMDATHLON_TBIRD,
-  AMDATHLON_MP,
-  AMDATHLON_XP,
-  AMDATHLON_K8SSE3,
-  AMDATHLON_OPTERON,
-  AMDATHLON_FX,
-  AMDATHLON_64,
-  AMD_BTVER1,
-  AMD_BTVER2,
-  AMDFAM15H_BDVER3,
-  AMDFAM15H_BDVER4,
-  AMDFAM17H_ZNVER1,
   CPU_SUBTYPE_MAX
 };
 
 enum ProcessorFeatures {
   FEATURE_CMOV = 0,
   FEATURE_MMX,
   FEATURE_POPCNT,
   FEATURE_SSE,
   FEATURE_SSE2,
   FEATURE_SSE3,
   FEATURE_SSSE3,
   FEATURE_SSE4_1,
   FEATURE_SSE4_2,
   FEATURE_AVX,
   FEATURE_AVX2,
-  FEATURE_AVX512,
-  FEATURE_AVX512SAVE,
-  FEATURE_MOVBE,
-  FEATURE_ADX,
-  FEATURE_EM64T
+  FEATURE_SSE4_A,
+  FEATURE_FMA4,
+  FEATURE_XOP,
+  FEATURE_FMA,
+  FEATURE_AVX512F,
+  FEATURE_BMI,
+  FEATURE_BMI2,
+  FEATURE_AES,
+  FEATURE_PCLMUL,
+  FEATURE_AVX512VL,
+  FEATURE_AVX512BW,
+  FEATURE_AVX512DQ,
+  FEATURE_AVX512CD,
+  FEATURE_AVX512ER,
+  FEATURE_AVX512PF,
+  FEATURE_AVX512VBMI,
+  FEATURE_AVX512IFMA,
+  FEATURE_AVX5124VNNIW,
+  FEATURE_AVX5124FMAPS,
+  FEATURE_AVX512VPOPCNTDQ
 };
 
 // The check below for i386 was copied from clang's cpuid.h (__get_cpuid_max).
 // Check motivated by bug reports for OpenSSL crashing on CPUs without CPUID
 // support. Consequently, for i386, the presence of CPUID is checked first
 // via the corresponding eflags bit.
 static bool isCpuIdSupported() {
 #if defined(__GNUC__) || defined(__clang__)

 
 static void
 getIntelProcessorTypeAndSubtype(unsigned Family, unsigned Model,
                                 unsigned Brand_id, unsigned Features,
                                 unsigned *Type, unsigned *Subtype) {
   if (Brand_id != 0)
     return;
   switch (Family) {
-  case 3:
-    *Type = INTEL_i386;
-    break;
-  case 4:
-    switch (Model) {
-    case 0: // Intel486 DX processors
-    case 1: // Intel486 DX processors
-    case 2: // Intel486 SX processors
-    case 3: // Intel487 processors, IntelDX2 OverDrive processors,
-            // IntelDX2 processors
-    case 4: // Intel486 SL processor
-    case 5: // IntelSX2 processors
-    case 7: // Write-Back Enhanced IntelDX2 processors
-    case 8: // IntelDX4 OverDrive processors, IntelDX4 processors
-    default:
-      *Type = INTEL_i486;
-      break;
-    }
-    break;
-  case 5:
-    switch (Model) {
-    case 1: // Pentium OverDrive processor for Pentium processor (60, 66),
-            // Pentium processors (60, 66)
-    case 2: // Pentium OverDrive processor for Pentium processor (75, 90,
-            // 100, 120, 133), Pentium processors (75, 90, 100, 120, 133,
-            // 150, 166, 200)
-    case 3: // Pentium OverDrive processors for Intel486 processor-based
-            // systems
-      *Type = INTEL_PENTIUM;
-      break;
-    case 4: // Pentium OverDrive processor with MMX technology for Pentium
-            // processor (75, 90, 100, 120, 133), Pentium processor with
-            // MMX technology (166, 200)
-      *Type = INTEL_PENTIUM;
-      *Subtype = INTEL_PENTIUM_MMX;
-      break;
-    default:
-      *Type = INTEL_PENTIUM;
-      break;
-    }
-    break;
   case 6:
     switch (Model) {
-    case 0x01: // Pentium Pro processor
-      *Type = INTEL_PENTIUM_PRO;
-      break;
-    case 0x03: // Intel Pentium II OverDrive processor, Pentium II processor,
-               // model 03
-    case 0x05: // Pentium II processor, model 05, Pentium II Xeon processor,
-               // model 05, and Intel Celeron processor, model 05
-    case 0x06: // Celeron processor, model 06
-      *Type = INTEL_PENTIUM_II;
-      break;
-    case 0x07: // Pentium III processor, model 07, and Pentium III Xeon
-               // processor, model 07
-    case 0x08: // Pentium III processor, model 08, Pentium III Xeon processor,
-               // model 08, and Celeron processor, model 08
-    case 0x0a: // Pentium III Xeon processor, model 0Ah
-    case 0x0b: // Pentium III processor, model 0Bh
-      *Type = INTEL_PENTIUM_III;
-      break;
-    case 0x09: // Intel Pentium M processor, Intel Celeron M processor model 09.
-    case 0x0d: // Intel Pentium M processor, Intel Celeron M processor, model
-               // 0Dh. All processors are manufactured using the 90 nm process.
-    case 0x15: // Intel EP80579 Integrated Processor and Intel EP80579
-               // Integrated Processor with Intel QuickAssist Technology
-      *Type = INTEL_PENTIUM_M;
-      break;
-    case 0x0e: // Intel Core Duo processor, Intel Core Solo processor, model
-               // 0Eh. All processors are manufactured using the 65 nm process.
-      *Type = INTEL_CORE_DUO;
-      break;   // yonah
     case 0x0f: // Intel Core 2 Duo processor, Intel Core 2 Duo mobile
                // processor, Intel Core 2 Quad processor, Intel Core 2 Quad
                // mobile processor, Intel Core 2 Extreme processor, Intel
                // Pentium Dual-Core processor, Intel Xeon processor, model
                // 0Fh. All processors are manufactured using the 65 nm process.
     case 0x16: // Intel Celeron processor model 16h. All processors are
                // manufactured using the 65 nm process
-      *Type = INTEL_CORE2; // "core2"
-      *Subtype = INTEL_CORE2_65;
-      break;
     case 0x17: // Intel Core 2 Extreme processor, Intel Xeon processor, model
                // 17h. All processors are manufactured using the 45 nm process.
                //
                // 45nm: Penryn , Wolfdale, Yorkfield (XE)
     case 0x1d: // Intel Xeon processor MP. All processors are manufactured using
                // the 45 nm process.
       *Type = INTEL_CORE2; // "penryn"
-      *Subtype = INTEL_CORE2_45;
       break;
     case 0x1a: // Intel Core i7 processor and Intel Xeon processor. All
                // processors are manufactured using the 45 nm process.
     case 0x1e: // Intel(R) Core(TM) i7 CPU         870  @ 2.93GHz.
                // As found in a Summer 2010 model iMac.
     case 0x1f:
     case 0x2e:             // Nehalem EX
       *Type = INTEL_COREI7; // "nehalem"
case 0x55:
       *Subtype = INTEL_COREI7_SKYLAKE_AVX512; // "skylake-avx512"
       break;
 
     case 0x1c: // Most 45 nm Intel Atom processors
     case 0x26: // 45 nm Atom Lincroft
     case 0x27: // 32 nm Atom Medfield
     case 0x35: // 32 nm Atom Midview
     case 0x36: // 32 nm Atom Midview
-      *Type = INTEL_ATOM;
-      *Subtype = INTEL_ATOM_BONNELL;
+      *Type = INTEL_BONNELL;
       break; // "bonnell"
 
     // Atom Silvermont codes from the Intel software optimization guide.
     case 0x37:
     case 0x4a:
     case 0x4d:
     case 0x5a:
     case 0x5d:
     case 0x4c: // really airmont
-      *Type = INTEL_ATOM;
-      *Subtype = INTEL_ATOM_SILVERMONT;
+      *Type = INTEL_SILVERMONT;
       break; // "silvermont"
 
     case 0x57:
-      *Type = INTEL_XEONPHI; // knl
-      *Subtype = INTEL_KNIGHTS_LANDING;
-      break;
-
-    default: // Unknown family 6 CPU, try to guess.
-      if (Features & (1 << FEATURE_AVX512)) {
-        *Type = INTEL_XEONPHI; // knl
-        *Subtype = INTEL_KNIGHTS_LANDING;
-        break;
-      }
-      if (Features & (1 << FEATURE_ADX)) {
-        *Type = INTEL_COREI7;
-        *Subtype = INTEL_COREI7_BROADWELL;
-        break;
-      }
-      if (Features & (1 << FEATURE_AVX2)) {
-        *Type = INTEL_COREI7;
-        *Subtype = INTEL_COREI7_HASWELL;
-        break;
-      }
-      if (Features & (1 << FEATURE_AVX)) {
-        *Type = INTEL_COREI7;
-        *Subtype = INTEL_COREI7_SANDYBRIDGE;
-        break;
-      }
-      if (Features & (1 << FEATURE_SSE4_2)) {
-        if (Features & (1 << FEATURE_MOVBE)) {
-          *Type = INTEL_ATOM;
-          *Subtype = INTEL_ATOM_SILVERMONT;
-        } else {
-          *Type = INTEL_COREI7;
-          *Subtype = INTEL_COREI7_NEHALEM;
-        }
-        break;
-      }
-      if (Features & (1 << FEATURE_SSE4_1)) {
-        *Type = INTEL_CORE2; // "penryn"
-        *Subtype = INTEL_CORE2_45;
-        break;
-      }
-      if (Features & (1 << FEATURE_SSSE3)) {
-        if (Features & (1 << FEATURE_MOVBE)) {
-          *Type = INTEL_ATOM;
-          *Subtype = INTEL_ATOM_BONNELL; // "bonnell"
-        } else {
-          *Type = INTEL_CORE2; // "core2"
-          *Subtype = INTEL_CORE2_65;
-        }
-        break;
-      }
-      if (Features & (1 << FEATURE_EM64T)) {
-        *Type = INTEL_X86_64;
-        break; // x86-64
-      }
-      if (Features & (1 << FEATURE_SSE2)) {
-        *Type = INTEL_PENTIUM_M;
-        break;
-      }
-      if (Features & (1 << FEATURE_SSE)) {
-        *Type = INTEL_PENTIUM_III;
-        break;
-      }
-      if (Features & (1 << FEATURE_MMX)) {
-        *Type = INTEL_PENTIUM_II;
-        break;
-      }
-      *Type = INTEL_PENTIUM_PRO;
-      break;
-    }
-    break;
-  case 15: {
-    switch (Model) {
-    case 0: // Pentium 4 processor, Intel Xeon processor. All processors are
-            // model 00h and manufactured using the 0.18 micron process.
-    case 1: // Pentium 4 processor, Intel Xeon processor, Intel Xeon
-            // processor MP, and Intel Celeron processor. All processors are
-            // model 01h and manufactured using the 0.18 micron process.
-    case 2: // Pentium 4 processor, Mobile Intel Pentium 4 processor - M,
-            // Intel Xeon processor, Intel Xeon processor MP, Intel Celeron
-            // processor, and Mobile Intel Celeron processor. All processors
-            // are model 02h and manufactured using the 0.13 micron process.
-      *Type =
-          ((Features & (1 << FEATURE_EM64T)) ? INTEL_X86_64 : INTEL_PENTIUM_IV);
-      break;
-
-    case 3: // Pentium 4 processor, Intel Xeon processor, Intel Celeron D
-            // processor. All processors are model 03h and manufactured using
-            // the 90 nm process.
-    case 4: // Pentium 4 processor, Pentium 4 processor Extreme Edition,
-            // Pentium D processor, Intel Xeon processor, Intel Xeon
-            // processor MP, Intel Celeron D processor. All processors are
-            // model 04h and manufactured using the 90 nm process.
-    case 6: // Pentium 4 processor, Pentium D processor, Pentium processor
-            // Extreme Edition, Intel Xeon processor, Intel Xeon processor
-            // MP, Intel Celeron D processor. All processors are model 06h
-            // and manufactured using the 65 nm process.
-      *Type =
-          ((Features & (1 << FEATURE_EM64T)) ? INTEL_NOCONA : INTEL_PRESCOTT);
+      *Type = INTEL_KNL; // knl
       break;
 
-    default:
-      *Type =
-          ((Features & (1 << FEATURE_EM64T)) ? INTEL_X86_64 : INTEL_PENTIUM_IV);
+    default: // Unknown family 6 CPU.
       break;
-    }
     break;
     }
   default:
-    break; /*"generic"*/
+    break; // Unknown.
   }
 }
 
 static void getAMDProcessorTypeAndSubtype(unsigned Family, unsigned Model,
                                           unsigned Features, unsigned *Type,
                                           unsigned *Subtype) {
   // FIXME: this poorly matches the generated SubtargetFeatureKV table.  There
   // appears to be no way to generate the wide variety of AMD-specific targets
   // from the information returned from CPUID.
   switch (Family) {
-  case 4:
-    *Type = AMD_i486;
-    break;
-  case 5:
-    *Type = AMDPENTIUM;
-    switch (Model) {
-    case 6:
-    case 7:
-      *Subtype = AMDPENTIUM_K6;
-      break; // "k6"
-    case 8:
-      *Subtype = AMDPENTIUM_K62;
-      break; // "k6-2"
-    case 9:
-    case 13:
-      *Subtype = AMDPENTIUM_K63;
-      break; // "k6-3"
-    case 10:
-      *Subtype = AMDPENTIUM_GEODE;
-      break; // "geode"
-    }
-    break;
-  case 6:
-    *Type = AMDATHLON;
-    switch (Model) {
-    case 4:
-      *Subtype = AMDATHLON_TBIRD;
-      break; // "athlon-tbird"
-    case 6:
-    case 7:
-    case 8:
-      *Subtype = AMDATHLON_MP;
-      break; // "athlon-mp"
-    case 10:
-      *Subtype = AMDATHLON_XP;
-      break; // "athlon-xp"
-    }
-    break;
-  case 15:
-    *Type = AMDATHLON;
-    if (Features & (1 << FEATURE_SSE3)) {
-      *Subtype = AMDATHLON_K8SSE3;
-      break; // "k8-sse3"
-    }
-    switch (Model) {
-    case 1:
-      *Subtype = AMDATHLON_OPTERON;
-      break; // "opteron"
-    case 5:
-      *Subtype = AMDATHLON_FX;
-      break; // "athlon-fx"; also opteron
-    default:
-      *Subtype = AMDATHLON_64;
-      break; // "athlon64"
-    }
-    break;
   case 16:
     *Type = AMDFAM10H; // "amdfam10"
     switch (Model) {
     case 2:
       *Subtype = AMDFAM10H_BARCELONA;
       break;
     case 4:
       *Subtype = AMDFAM10H_SHANGHAI;
       break;
     case 8:
       *Subtype = AMDFAM10H_ISTANBUL;
       break;
     }
     break;
   case 20:
-    *Type = AMDFAM14H;
-    *Subtype = AMD_BTVER1;
+    *Type = AMD_BTVER1;
     break; // "btver1";
   case 21:
     *Type = AMDFAM15H;
     if (Model >= 0x60 && Model <= 0x7f) {
       *Subtype = AMDFAM15H_BDVER4;
-      break; // "bdver4"; 50h-6Fh: Excavator
+      break; // "bdver4"; 60h-7Fh: Excavator
     }
     if (Model >= 0x30 && Model <= 0x3f) {
       *Subtype = AMDFAM15H_BDVER3;
       break; // "bdver3"; 30h-3Fh: Steamroller
     }
     if (Model >= 0x10 && Model <= 0x1f) {
       *Subtype = AMDFAM15H_BDVER2;
       break; // "bdver2"; 10h-1Fh: Piledriver
case 23:
     *Type = AMDFAM17H;
     *Subtype = AMDFAM17H_ZNVER1;
     break;
   default:
     break; // "generic"
   }
 }
 
-static unsigned getAvailableFeatures(unsigned ECX, unsigned EDX,
-                                     unsigned MaxLeaf) {
+static void getAvailableFeatures(unsigned ECX, unsigned EDX, unsigned MaxLeaf,
+                                 unsigned *FeaturesOut) {
   unsigned Features = 0;
   unsigned EAX, EBX;
-  Features |= (((EDX >> 23) & 1) << FEATURE_MMX);
-  Features |= (((EDX >> 25) & 1) << FEATURE_SSE);
-  Features |= (((EDX >> 26) & 1) << FEATURE_SSE2);
-  Features |= (((ECX >> 0) & 1) << FEATURE_SSE3);
-  Features |= (((ECX >> 9) & 1) << FEATURE_SSSE3);
-  Features |= (((ECX >> 19) & 1) << FEATURE_SSE4_1);
-  Features |= (((ECX >> 20) & 1) << FEATURE_SSE4_2);
-  Features |= (((ECX >> 22) & 1) << FEATURE_MOVBE);
+
+  if ((EDX >> 15) & 1)
+    Features |= 1 << FEATURE_CMOV;
+  if ((EDX >> 23) & 1)
+    Features |= 1 << FEATURE_MMX;
+  if ((EDX >> 25) & 1)
+    Features |= 1 << FEATURE_SSE;
+  if ((EDX >> 26) & 1)
+    Features |= 1 << FEATURE_SSE2;
+
+  if ((ECX >> 0) & 1)
+    Features |= 1 << FEATURE_SSE3;
+  if ((ECX >> 1) & 1)
+    Features |= 1 << FEATURE_PCLMUL;
+  if ((ECX >> 9) & 1)
+    Features |= 1 << FEATURE_SSSE3;
+  if ((ECX >> 12) & 1)
+    Features |= 1 << FEATURE_FMA;
+  if ((ECX >> 19) & 1)
+    Features |= 1 << FEATURE_SSE4_1;
+  if ((ECX >> 20) & 1)
+    Features |= 1 << FEATURE_SSE4_2;
+  if ((ECX >> 23) & 1)
+    Features |= 1 << FEATURE_POPCNT;
+  if ((ECX >> 25) & 1)
+    Features |= 1 << FEATURE_AES;
 
   // If CPUID indicates support for XSAVE, XRESTORE and AVX, and XGETBV
   // indicates that the AVX registers will be saved and restored on context
   // switch, then we have full AVX support.
   const unsigned AVXBits = (1 << 27) | (1 << 28);
   bool HasAVX = ((ECX & AVXBits) == AVXBits) && !getX86XCR0(&EAX, &EDX) &&
                 ((EAX & 0x6) == 0x6);
   bool HasAVX512Save = HasAVX && ((EAX & 0xe0) == 0xe0);
+
+  if (HasAVX)
+    Features |= 1 << FEATURE_AVX;
+
   bool HasLeaf7 =
       MaxLeaf >= 0x7 && !getX86CpuIDAndInfoEx(0x7, 0x0, &EAX, &EBX, &ECX, &EDX);
-  bool HasADX = HasLeaf7 && ((EBX >> 19) & 1);
-  bool HasAVX2 = HasAVX && HasLeaf7 && (EBX & 0x20);
-  bool HasAVX512 = HasLeaf7 && HasAVX512Save && ((EBX >> 16) & 1);
-  Features |= (HasAVX << FEATURE_AVX);
-  Features |= (HasAVX2 << FEATURE_AVX2);
-  Features |= (HasAVX512 << FEATURE_AVX512);
-  Features |= (HasAVX512Save << FEATURE_AVX512SAVE);
-  Features |= (HasADX << FEATURE_ADX);
+
+  if (HasLeaf7 && ((EBX >> 3) & 1))
+    Features |= 1 << FEATURE_BMI;
+  if (HasLeaf7 && ((EBX >> 5) & 1) && HasAVX)
+    Features |= 1 << FEATURE_AVX2;
+  if (HasLeaf7 && ((EBX >> 9) & 1))
+    Features |= 1 << FEATURE_BMI2;
+  if (HasLeaf7 && ((EBX >> 16) & 1) && HasAVX512Save)
+    Features |= 1 << FEATURE_AVX512F;
+  if (HasLeaf7 && ((EBX >> 17) & 1) && HasAVX512Save)
+    Features |= 1 << FEATURE_AVX512DQ;
+  if (HasLeaf7 && ((EBX >> 21) & 1) && HasAVX512Save)
+    Features |= 1 << FEATURE_AVX512IFMA;
+  if (HasLeaf7 && ((EBX >> 26) & 1) && HasAVX512Save)
+    Features |= 1 << FEATURE_AVX512PF;
+  if (HasLeaf7 && ((EBX >> 27) & 1) && HasAVX512Save)
+    Features |= 1 << FEATURE_AVX512ER;
+  if (HasLeaf7 && ((EBX >> 28) & 1) && HasAVX512Save)
+    Features |= 1 << FEATURE_AVX512CD;
+  if (HasLeaf7 && ((EBX >> 30) & 1) && HasAVX512Save)
+    Features |= 1 << FEATURE_AVX512BW;
+  if (HasLeaf7 && ((EBX >> 31) & 1) && HasAVX512Save)
+    Features |= 1 << FEATURE_AVX512VL;
+
+  if (HasLeaf7 && ((ECX >> 1) & 1) && HasAVX512Save)
+    Features |= 1 << FEATURE_AVX512VBMI;
+  if (HasLeaf7 && ((ECX >> 14) & 1) && HasAVX512Save)
+    Features |= 1 << FEATURE_AVX512VPOPCNTDQ;
+
+  if (HasLeaf7 && ((EDX >> 2) & 1) && HasAVX512Save)
+    Features |= 1 << FEATURE_AVX5124VNNIW;
+  if (HasLeaf7 && ((EDX >> 3) & 1) && HasAVX512Save)
+    Features |= 1 << FEATURE_AVX5124FMAPS;
 
   unsigned MaxExtLevel;
   getX86CpuIDAndInfo(0x80000000, &MaxExtLevel, &EBX, &ECX, &EDX);
 
   bool HasExtLeaf1 = MaxExtLevel >= 0x80000001 &&
                      !getX86CpuIDAndInfo(0x80000001, &EAX, &EBX, &ECX, &EDX);
-  if (HasExtLeaf1)
-    Features |= (((EDX >> 29) & 0x1) << FEATURE_EM64T);
+  if (HasExtLeaf1 && ((ECX >> 6) & 1))
+    Features |= 1 << FEATURE_SSE4_A;
+  if (HasExtLeaf1 && ((ECX >> 11) & 1))
+    Features |= 1 << FEATURE_XOP;
+  if (HasExtLeaf1 && ((ECX >> 16) & 1))
+    Features |= 1 << FEATURE_FMA4;
 
-  return Features;
+  *FeaturesOut = Features;
 }
 
 #if defined(HAVE_INIT_PRIORITY)
 #define CONSTRUCTOR_ATTRIBUTE __attribute__((__constructor__ 101))
 #elif __has_attribute(__constructor__)
 #define CONSTRUCTOR_ATTRIBUTE __attribute__((__constructor__))
 #else
 // FIXME: For MSVC, we should make a function pointer global in .CRT$X?? so that
if (getX86CpuIDAndInfo(0, &MaxLeaf, &Vendor, &ECX, &EDX) || MaxLeaf < 1) {
     __cpu_model.__cpu_vendor = VENDOR_OTHER;
     return -1;
   }
   getX86CpuIDAndInfo(1, &EAX, &EBX, &ECX, &EDX);
   detectX86FamilyModel(EAX, &Family, &Model);
   Brand_id = EBX & 0xff;
 
   /* Find available features. */
-  Features = getAvailableFeatures(ECX, EDX, MaxLeaf);
-  __cpu_model.__cpu_features[0] = Features;
+  getAvailableFeatures(ECX, EDX, MaxLeaf, &__cpu_model.__cpu_features[0]);
 
   if (Vendor == SIG_INTEL) {
     /* Get CPU type.  */
     getIntelProcessorTypeAndSubtype(Family, Model, Brand_id, Features,
                                     &(__cpu_model.__cpu_type),
                                     &(__cpu_model.__cpu_subtype));
     __cpu_model.__cpu_vendor = VENDOR_INTEL;
   } else if (Vendor == SIG_AMD) {