Add cpuid.cpp

CMakeLists.txt

@@ -105,6 +105,7 @@ set(BIN_SRC src/app/CmdOptions.cpp
 
 set(LIB_SRC src/api.cpp
             src/api_c.cpp
+            src/cpuid.cpp
             src/BitSieve240.cpp
             src/FactorTable.cpp
             src/RiemannR.cpp

cmake/multiarch_avx512_vpopcnt.cmake

@@ -20,6 +20,9 @@ check_cxx_source_compiles("
       Error: AVX512 BMI2 multiarch not needed!
     #endif
 
+    #define ENABLE_MULTIARCH_AVX512_BMI2
+
+    #include <cpuid.hpp>
     #include <cpu_supports_avx512_bmi2.hpp>
     #include <immintrin.h>
     #include <stdint.h>

include/check_enable_cpuid_popcnt.hpp

@@ -0,0 +1,40 @@
+///
+/// @file  check_enable_cpuid_popcnt.hpp
+/// @brief Check if runtime POPCNT detection should be enabled on
+///        x86 and x64 CPUs.
+///
+/// Copyright (C) 2024 Kim Walisch, <[email protected]>
+///
+/// This file is distributed under the BSD License. See the COPYING
+/// file in the top level directory.
+///
+
+#ifndef CHECK_ENABLE_CPUID_POPCNT_HPP
+#define CHECK_ENABLE_CPUID_POPCNT_HPP
+
+// Enable CPUID on x86 and x86-64 CPUs
+#if defined(__x86_64__) || \
+    defined(__i386__) || \
+    defined(_M_X64) || \
+    defined(_M_IX86)
+
+// Both GCC and Clang (even Clang on Windows) define the __POPCNT__
+// macro if the user compiles with -mpopcnt. The __POPCNT__
+// macro is even defined if the user compiles with other flags
+// such as -mavx or -march=native.
+#if defined(__POPCNT__)
+  #define HAS_POPCNT
+// The MSVC compiler does not support a POPCNT macro, but if the user
+// compiles with e.g. /arch:AVX or /arch:AVX512 then MSVC defines
+// the __AVX__ macro and POPCNT is also supported.
+#elif defined(_MSC_VER) && defined(__AVX__)
+  #define HAS_POPCNT
+#endif
+
+#if !defined(HAS_POPCNT)
+  #define ENABLE_CPUID_POPCNT
+#endif
+
+#endif // x86 CPUs
+
+#endif

include/cpu_supports_avx512_bmi2.hpp

@@ -11,87 +11,16 @@
 #ifndef CPU_SUPPORTS_AVX512_BMI2_HPP
 #define CPU_SUPPORTS_AVX512_BMI2_HPP
 
-#include <cpuid.hpp>
+namespace primecount {
 
-#if defined(_MSC_VER)
-  #include <immintrin.h>
-#endif
-
-// CPUID bits documentation:
-// https://en.wikipedia.org/wiki/CPUID
-
-// %ebx bit flags
-#define bit_BMI2    (1 << 8)
-#define bit_AVX512F (1 << 16)
+bool run_cpuid_avx512_bmi2();
 
-// %ecx bit flags
-#define bit_AVX512_VPOPCNTDQ (1 << 14)
-
-// xgetbv bit flags
-#define XSTATE_SSE (1 << 1)
-#define XSTATE_YMM (1 << 2)
-#define XSTATE_ZMM (7 << 5)
+} // namespace
 
 namespace {
 
-// Get Value of Extended Control Register
-inline int get_xcr0()
-{
-  int xcr0;
-
-#if defined(_MSC_VER)
-  xcr0 = (int) _xgetbv(0);
-#else
-  __asm__ ("xgetbv" : "=a" (xcr0) : "c" (0) : "%edx" );
-#endif
-
-  return xcr0;
-}
-
-inline bool run_cpuid_avx512_bmi2()
-{
-  int eax = 1;
-  int ebx = 0;
-  int ecx = 0;
-  int edx = 0;
-
-  run_cpuid(&eax, &ebx, &ecx, &edx);
-
-  int osxsave_mask = (1 << 27);
-
-  // Ensure OS supports extended processor state management
-  if ((ecx & osxsave_mask) != osxsave_mask)
-    return false;
-
-  int ymm_mask = XSTATE_SSE | XSTATE_YMM;
-  int zmm_mask = XSTATE_SSE | XSTATE_YMM | XSTATE_ZMM;
-  int xcr0 = get_xcr0();
-
-  // Check AVX OS support
-  if ((xcr0 & ymm_mask) != ymm_mask)
-    return false;
-
-  // Check AVX512 OS support
-  if ((xcr0 & zmm_mask) != zmm_mask)
-    return false;
-
-  eax = 7;
-  ebx = 0;
-  ecx = 0;
-  edx = 0;
-
-  run_cpuid(&eax, &ebx, &ecx, &edx);
-
-  if ((ebx & bit_BMI2) != bit_BMI2)
-    return false;
-
-  // AVX512F, AVX512VPOPCNTDQ
-  return ((ebx & bit_AVX512F) == bit_AVX512F &&
-          (ecx & bit_AVX512_VPOPCNTDQ) == bit_AVX512_VPOPCNTDQ);
-}
-
 /// Initialized at startup
-bool cpu_supports_avx512_bmi2 = run_cpuid_avx512_bmi2();
+bool cpu_supports_avx512_bmi2 = primecount::run_cpuid_avx512_bmi2();
 
 } // namespace
 

include/cpu_supports_popcnt.hpp

@@ -11,52 +11,23 @@
 #ifndef CPU_SUPPORTS_POPCNT_HPP
 #define CPU_SUPPORTS_POPCNT_HPP
 
-// Enable CPUID on x86 and x86-64 CPUs
-#if defined(__x86_64__) || \
-    defined(__i386__) || \
-    defined(_M_X64) || \
-    defined(_M_IX86)
-
-// Both GCC and Clang (even Clang on Windows) define the __POPCNT__
-// macro if the user compiles with -mpopcnt. The __POPCNT__
-// macro is even defined if the user compiles with other flags
-// such as -mavx or -march=native.
-#if defined(__POPCNT__)
-  #define HAS_POPCNT
-// The MSVC compiler does not support a POPCNT macro, but if the user
-// compiles with e.g. /arch:AVX or /arch:AVX512 then MSVC defines
-// the __AVX__ macro and POPCNT is also supported.
-#elif defined(_MSC_VER) && defined(__AVX__)
-  #define HAS_POPCNT
-#endif
-
-#if !defined(HAS_POPCNT)
+#include <check_enable_cpuid_popcnt.hpp>
 
-#include <cpuid.hpp>
-#define ENABLE_CPUID_POPCNT
+#if defined(ENABLE_CPUID_POPCNT)
 
-namespace {
+namespace primecount {
 
-inline bool run_cpuid_supports_popcnt()
-{
-  int eax = 1;
-  int ebx = 0;
-  int ecx = 0;
-  int edx = 0;
+bool run_cpuid_supports_popcnt();
 
-  run_cpuid(&eax, &ebx, &ecx, &edx);
+} // namespace
 
-  // https://en.wikipedia.org/wiki/CPUID
-  int bit_POPCNT = 1 << 23;
-  return (ecx & bit_POPCNT) == bit_POPCNT;
-}
+namespace {
 
 /// Initialized at startup
-bool cpu_supports_popcnt = run_cpuid_supports_popcnt();
+bool cpu_supports_popcnt = primecount::run_cpuid_supports_popcnt();
 
 } // namespace
 
-#endif // !defined(HAS_POPCNT)
-#endif // CPUID
+#endif // ENABLE_CPUID_POPCNT
 
 #endif

include/cpuid.hpp

@@ -11,6 +11,8 @@
 #ifndef CPUID_HPP
 #define CPUID_HPP
 
+#include <check_enable_cpuid_popcnt.hpp>
+
 #if defined(_MSC_VER)
   #include <intrin.h>
 #endif
@@ -23,7 +25,6 @@ inline void run_cpuid(int* eax, int* ebx, int* ecx, int* edx)
 
   int abcd[4];
   __cpuidex(abcd, *eax, *ecx);
-
   *eax = abcd[0];
   *ebx = abcd[1];
   *ecx = abcd[2];
@@ -37,26 +38,125 @@ inline void run_cpuid(int* eax, int* ebx, int* ecx, int* edx)
     "movl %%ebx, %%edi;"
     "cpuid;"
     "xchgl %%ebx, %%edi;"
-    : "=D" (*ebx),
-      "+a" (*eax),
-      "+c" (*ecx),
-      "=d" (*edx)
+    : "=D" (*ebx), "+a" (*eax), "+c" (*ecx), "=d" (*edx)
   );
 
 #else
 
   __asm__ (
     "cpuid"
-    : "=a" (*eax),
-      "=b" (*ebx),
-      "=c" (*ecx),
-      "=d" (*edx)
+    : "=a" (*eax), "=b" (*ebx), "=c" (*ecx), "=d" (*edx)
     : "a" (*eax), "c" (*ecx)
   );
 
 #endif
 }
 
+#if defined(ENABLE_MULTIARCH_AVX512_BMI2)
+
+// Get Value of Extended Control Register
+inline int get_xcr0()
+{
+  int xcr0;
+
+#if defined(_MSC_VER)
+  xcr0 = (int) _xgetbv(0);
+#else
+  __asm__ ("xgetbv" : "=a" (xcr0) : "c" (0) : "%edx" );
+#endif
+
+  return xcr0;
+}
+
+#endif
+
+} // namespace
+
+#if defined(ENABLE_CPUID_POPCNT)
+
+namespace primecount {
+
+bool run_cpuid_supports_popcnt()
+{
+  int eax = 1;
+  int ebx = 0;
+  int ecx = 0;
+  int edx = 0;
+
+  run_cpuid(&eax, &ebx, &ecx, &edx);
+
+  // https://en.wikipedia.org/wiki/CPUID
+  int bit_POPCNT = 1 << 23;
+  return (ecx & bit_POPCNT) == bit_POPCNT;
+}
+
 } // namespace
 
 #endif
+
+#if defined(ENABLE_MULTIARCH_AVX512_BMI2)
+
+// CPUID bits documentation:
+// https://en.wikipedia.org/wiki/CPUID
+
+// %ebx bit flags
+#define bit_BMI2    (1 << 8)
+#define bit_AVX512F (1 << 16)
+
+// %ecx bit flags
+#define bit_AVX512_VPOPCNTDQ (1 << 14)
+
+// xgetbv bit flags
+#define XSTATE_SSE (1 << 1)
+#define XSTATE_YMM (1 << 2)
+#define XSTATE_ZMM (7 << 5)
+
+namespace primecount {
+
+bool run_cpuid_avx512_bmi2()
+{
+  int eax = 1;
+  int ebx = 0;
+  int ecx = 0;
+  int edx = 0;
+
+  run_cpuid(&eax, &ebx, &ecx, &edx);
+
+  int osxsave_mask = (1 << 27);
+
+  // Ensure OS supports extended processor state management
+  if ((ecx & osxsave_mask) != osxsave_mask)
+    return false;
+
+  int ymm_mask = XSTATE_SSE | XSTATE_YMM;
+  int zmm_mask = XSTATE_SSE | XSTATE_YMM | XSTATE_ZMM;
+  int xcr0 = get_xcr0();
+
+  // Check AVX OS support
+  if ((xcr0 & ymm_mask) != ymm_mask)
+    return false;
+
+  // Check AVX512 OS support
+  if ((xcr0 & zmm_mask) != zmm_mask)
+    return false;
+
+  eax = 7;
+  ebx = 0;
+  ecx = 0;
+  edx = 0;
+
+  run_cpuid(&eax, &ebx, &ecx, &edx);
+
+  if ((ebx & bit_BMI2) != bit_BMI2)
+    return false;
+
+  // AVX512F, AVX512VPOPCNTDQ
+  return ((ebx & bit_AVX512F) == bit_AVX512F &&
+          (ecx & bit_AVX512_VPOPCNTDQ) == bit_AVX512_VPOPCNTDQ);
+}
+
+} // namespace
+
+#endif
+
+#endif

src/cpuid.cpp

@@ -0,0 +1,19 @@
+///
+/// @file  cpuid.cpp
+/// @brief CPUID for x86 and x86-64 CPUs.
+///
+/// Copyright (C) 2024 Kim Walisch, <[email protected]>
+///
+/// This file is distributed under the BSD License. See the COPYING
+/// file in the top level directory.
+///
+
+// Enable CPUID on x86 and x86-64 CPUs
+#if defined(__x86_64__) || \
+    defined(__i386__) || \
+    defined(_M_X64) || \
+    defined(_M_IX86)
+
+#include <cpuid.hpp>
+
+#endif