Preliminary MSVC int128_t support

ChangeLog

@@ -1,14 +1,15 @@
-Changes in primecount-7.13, 2024-04-08
+Changes in primecount-7.13, 2024-04-09
 
-* Delete ci-sage.yml, it has not been updated in 3 years and I
-  don't want to maintain it. I feel like this CI script should
-  be part of SageMath, not primecount.
+* Add preliminary MSVC 128-bit support.
 * Added AVX512 runtime dispatching for x86 CPUs.
 * CMakeLists.txt: New WITH_MULTIARCH option (default ON).
 * Sieve_count.cpp: New AVX512 popcount algorithm for x86 CPUs.
 * Sieve_count.cpp: New ARM SVE popcount algorithm.
 * int128.cmake: Improve int128_t support for Windows.
 * OpenMP.cmake: Improve LLVM/Clang OpenMP detection.
+* Delete ci-sage.yml, it has not been updated in 3 years and I
+  don't want to maintain it. I feel like this CI script should
+  be part of SageMath, not primecount.
 
 Changes in primecount-7.12, 2024-04-01
 

cmake/int128_t.cmake

@@ -21,15 +21,27 @@ set(CMAKE_REQUIRED_INCLUDES "${PROJECT_SOURCE_DIR}/include")
 check_cxx_source_compiles("
     #include <int128_t.hpp>
     #include <algorithm>
+
+    using namespace primecount;
+
     int main() {
-        using namespace primecount;
         int128_t x = int128_t(1) << 100;
         int128_t y = 1000;
-        x /= 123;
-        if (std::min(x, y) == y)
-            return 0;
-        else
+        unsigned divider = 123;
+        x = x / divider;
+        x /= y;
+        x = x + y;
+
+        if (std::min(x, y) != y)
             return 1;
+
+        long double z = ((long double) y) + 1.59867;
+        int128_t iz = (int128_t) z;
+
+        if (std::max(y, iz) != iz)
+            return 1;
+
+        return 0;
     }" int128)
 
 if(NOT int128)

include/int128_t.hpp

@@ -14,6 +14,9 @@
 #define INT128_T_HPP
 
 #include <stdint.h>
+#include <string>
+#include <limits>
+#include <type_traits>
 
 /// If INT128_MAX is defined we know that int128_t and
 /// uint128_t are available in <stdint.h>.
@@ -27,14 +30,15 @@ namespace primecount {
 
 using maxint_t = int128_t ;
 using maxuint_t = uint128_t;
+using std::to_string;
 
 } // namespace
 
 #elif defined(__SIZEOF_INT128__) && \
      !defined(DISABLE_INT128)
 
 #define HAVE_INT128_T
-#define HAVE_NON_STANDARD__INT128_T
+#define ENABLE_INT128_TO_STRING
 
 #include <ostream>
 
@@ -45,8 +49,39 @@ using uint128_t = __uint128_t;
 using maxint_t = __int128_t ;
 using maxuint_t = __uint128_t;
 
-/// std::cout support for int128_t.
-/// These functions are defined in util.cpp.
+/// These functions are defined in util.cpp
+std::string to_string(int128_t x);
+std::string to_string(uint128_t x);
+
+std::ostream& operator<<(std::ostream& stream, int128_t n);
+std::ostream& operator<<(std::ostream& stream, uint128_t n);
+
+} // namespace
+
+#elif __has_include(<__msvc_int128.hpp>) && \
+     !defined(DISABLE_INT128)
+
+#define HAVE_INT128_T
+#define ENABLE_INT128_TO_STRING
+
+// Unofficial/undocumented MSVC int128_t support:
+// https://github.com/microsoft/STL/blob/main/stl/inc/__msvc_int128.hpp
+// https://developercommunity.visualstudio.com/t/support-for-128-bit-integer-type/879048
+// https://stackoverflow.com/a/76440171
+#include <__msvc_int128.hpp>
+#include <ostream>
+
+namespace primecount {
+
+using int128_t = std::_Signed128;
+using uint128_t = std::_Unsigned128;
+using maxint_t = std::_Signed128;
+using maxuint_t = std::_Unsigned128;
+
+/// These functions are defined in util.cpp
+std::string to_string(int128_t x);
+std::string to_string(uint128_t x);
+
 std::ostream& operator<<(std::ostream& stream, int128_t n);
 std::ostream& operator<<(std::ostream& stream, uint128_t n);
 
@@ -58,14 +93,12 @@ namespace primecount {
 
 typedef int64_t maxint_t;
 typedef uint64_t maxuint_t;
+using std::to_string;
 
 } // namespace
 
 #endif
 
-#include <limits>
-#include <type_traits>
-
 // Portable C++ type traits that support int128_t and uint128_t.
 // This is required for GCC/Clang if the user compiles with -std=c++*
 // instead of -std=gnu++* and also for LLVM/Clang on Windows.

src/api.cpp

@@ -16,7 +16,6 @@
 #include <macros.hpp>
 #include <PiTable.hpp>
 #include <print.hpp>
-#include <to_string.hpp>
 
 #include <cmath>
 #include <string>
@@ -189,7 +188,7 @@ maxint_t get_max_x(double alpha_y)
 {
 #ifdef HAVE_INT128_T
   double max_x = std::pow((1ull << 62) * alpha_y, 3.0 / 2.0);
-  return (int128_t) max_x; 
+  return (int128_t) max_x;
 #else
   unused_param(alpha_y); 
   return port::numeric_limits<int64_t>::max();

src/app/main.cpp

@@ -31,7 +31,6 @@
 #include <PhiTiny.hpp>
 #include <print.hpp>
 #include <S.hpp>
-#include <to_string.hpp>
 
 #include <stdint.h>
 #include <exception>

src/deleglise-rivat/pi_deleglise_rivat.cpp

@@ -13,7 +13,7 @@
 ///        method, Revista do DETUA, vol. 4, no. 6, March 2006,
 ///        pp. 759-768.
 ///
-/// Copyright (C) 2022 Kim Walisch, <[email protected]>
+/// Copyright (C) 2024 Kim Walisch, <[email protected]>
 ///
 /// This file is distributed under the BSD License. See the COPYING
 /// file in the top level directory.
@@ -27,7 +27,6 @@
 #include <macros.hpp>
 #include <print.hpp>
 #include <S.hpp>
-#include <to_string.hpp>
 
 #include <stdint.h>
 #include <string>

src/gourdon/pi_gourdon.cpp

@@ -9,7 +9,7 @@
 ///        Xavier Gourdon formula:
 ///        pi(x) = A - B + C + D + Phi0 + Sigma
 ///
-/// Copyright (C) 2022 Kim Walisch, <[email protected]>
+/// Copyright (C) 2024 Kim Walisch, <[email protected]>
 ///
 /// This file is distributed under the BSD License. See the COPYING
 /// file in the top level directory.
@@ -22,7 +22,6 @@
 #include <macros.hpp>
 #include <PhiTiny.hpp>
 #include <print.hpp>
-#include <to_string.hpp>
 
 #include <stdint.h>
 #include <algorithm>

src/util.cpp

@@ -3,7 +3,7 @@
 ///        This file contains helper functions and global variables
 ///        that are initialized with default settings.
 ///
-/// Copyright (C) 2022 Kim Walisch, <[email protected]>
+/// Copyright (C) 2024 Kim Walisch, <[email protected]>
 ///
 /// This file is distributed under the BSD License. See the COPYING
 /// file in the top level directory.
@@ -15,7 +15,6 @@
 #include <int128_t.hpp>
 #include <imath.hpp>
 #include <macros.hpp>
-#include <to_string.hpp>
 
 #include <algorithm>
 #include <chrono>
@@ -60,7 +59,7 @@ namespace primecount {
 /// standard int128_t type is missing in <stdint.h>. We need to
 /// define a few functions that are not supported by the C++ STL.
 ///
-#if defined(HAVE_NON_STANDARD__INT128_T)
+#if defined(ENABLE_INT128_TO_STRING)
 
 std::string to_string(uint128_t n)
 {

test/fast_div.cpp

@@ -33,6 +33,7 @@ int main()
   std::uniform_int_distribution<int32_t> dist_i32(1, port::numeric_limits<int32_t>::max());
   std::uniform_int_distribution<uint64_t> dist_u64(0, port::numeric_limits<uint64_t>::max());
 
+  // Test unsigned/signed
   for (int i = 0; i < 10000; i++)
   {
     uint64_t x = dist_i32(gen);
@@ -52,18 +53,23 @@ int main()
 
 #ifdef HAVE_INT128_T
 
-  std::uniform_int_distribution<int128_t> dist_i128(0, port::numeric_limits<int128_t>::max());
+  std::uniform_int_distribution<uint64_t> dist_u62(0, uint64_t(1ull << 62));
 
+  // Test signed/signed
   for (int i = 0; i < 10000; i++)
   {
+    // Test x < 2^64
     int128_t x = dist_u64(gen);
      int32_t y = dist_i32(gen);
     int128_t res = fast_div(x, y);
 
     std::cout << "fast_div(" << x << ", " << y << ") = " << res;
     check(res == x / y);
 
-    x = dist_i128(gen);
+    // Test x > 2^64
+    int128_t low = dist_u64(gen);
+    int128_t high = int128_t(dist_u62(gen)) << 64;
+    x = high | low;
     y = dist_i32(gen);
     res = fast_div(x, y);