Update modular_arithmetic lib

Author: kimwalisch

lib/modular_arithmetic/build_tests.sh

@@ -18,9 +18,10 @@
 #    number of jobs will be used.
 # -r specifies to run all tests after the build.  Without -r, no tests will run.
 # -a specifies you want to compile the code using typically helpful (how much it
-#    helps depends on your compiler) inline asm optimizations, which makes for
-#    the fastest binaries but of course has the downsides of inline asm -
-#    primarily that inline asm is extremely difficult to properly test.
+#    helps depends on your compiler) inline asm optimizations, which usually
+#    makes the fastest binaries.  It is probably better to use -u instead for
+#    building these tests, since if we want to test some of the asm, we probably
+#    want to test all of it.
 # -u specifies that you want to compile the code using all available inline asm
 #    routines, so that the tests will cover all of them (this is not expected to
 #    result in the fastest binaries).

lib/modular_arithmetic/modular_arithmetic/CMakeLists.txt

@@ -75,7 +75,7 @@ include(FetchContent)
 FetchContent_Declare(
     hurchalla_util
     GIT_REPOSITORY https://github.com/hurchalla/util.git
-    GIT_TAG        38272aeb2b19a8bce2f96c11bf6e0ecbd9eed25b
+    GIT_TAG        9fac434b586717052c648339eb0f0f89d23e0298
 )
 FetchContent_MakeAvailable(hurchalla_util)
 

lib/modular_arithmetic/montgomery_arithmetic/CMakeLists.txt

@@ -79,7 +79,7 @@ include(FetchContent)
 FetchContent_Declare(
     hurchalla_util
     GIT_REPOSITORY https://github.com/hurchalla/util.git
-    GIT_TAG        38272aeb2b19a8bce2f96c11bf6e0ecbd9eed25b
+    GIT_TAG        9fac434b586717052c648339eb0f0f89d23e0298
 )
 FetchContent_MakeAvailable(hurchalla_util)
 

lib/modular_arithmetic/montgomery_arithmetic/include/hurchalla/montgomery_arithmetic/detail/MontyCommonBase.h

@@ -20,6 +20,7 @@
 #include "hurchalla/montgomery_arithmetic/detail/MontyTags.h"
 #include "hurchalla/util/traits/ut_numeric_limits.h"
 #include "hurchalla/util/unsigned_multiply_to_hilo_product.h"
+#include "hurchalla/util/unsigned_square_to_hilo_product.h"
 #include "hurchalla/util/compiler_macros.h"
 #include "hurchalla/util/branchless_shift_left.h"
 #include "hurchalla/util/branchless_shift_right.h"
@@ -540,8 +541,7 @@ class MontyCommonBase {
         HPBC_CLOCKWORK_INVARIANT2(r_squared_mod_n_ < n_);
         namespace hc = ::hurchalla;
         T u_lo;
-        T u_hi = hc::unsigned_multiply_to_hilo_product(u_lo,
-                                            r_squared_mod_n_, r_squared_mod_n_);
+        T u_hi = hc::unsigned_square_to_hilo_product(u_lo, r_squared_mod_n_);
         HPBC_CLOCKWORK_ASSERT2(u_hi < n_);  // verify that (u_hi*R + u_lo) < n*R
         T result = hc::REDC_standard(u_hi, u_lo, n_, inv_n_, PTAG());
         HPBC_CLOCKWORK_POSTCONDITION2(result < n_);

lib/modular_arithmetic/montgomery_arithmetic/include/hurchalla/montgomery_arithmetic/detail/MontyFullRange.h

@@ -19,6 +19,7 @@
 #include "hurchalla/modular_arithmetic/absolute_value_difference.h"
 #include "hurchalla/util/traits/ut_numeric_limits.h"
 #include "hurchalla/util/unsigned_multiply_to_hilo_product.h"
+#include "hurchalla/util/unsigned_square_to_hilo_product.h"
 #include "hurchalla/util/conditional_select.h"
 #include "hurchalla/util/cselect_on_bit.h"
 #include "hurchalla/util/compiler_macros.h"
@@ -277,7 +278,7 @@ class MontyFullRange final :
         namespace hc = ::hurchalla;
         T a = sv.getbits();
         T sqlo;
-        T sqhi = hc::unsigned_multiply_to_hilo_product(sqlo, a, a);
+        T sqhi = hc::unsigned_square_to_hilo_product(sqlo, a);
         T a_or_zero = sv.get_subtrahend();
         T u_hi = static_cast<T>(sqhi - a_or_zero - a_or_zero);
         T u_lo = sqlo;
@@ -305,7 +306,7 @@ class MontyFullRange final :
         namespace hc = ::hurchalla;
         T a = sv.getbits();
         T sqlo;
-        T sqhi = hc::unsigned_multiply_to_hilo_product(sqlo, a, a);
+        T sqhi = hc::unsigned_square_to_hilo_product(sqlo, a);
         T a_or_zero = sv.get_subtrahend();
         T u_hi = static_cast<T>(sqhi - a_or_zero - a_or_zero);
         T u_lo = sqlo;
@@ -349,15 +350,15 @@ class MontyFullRange final :
 
     // return the high word of the product, and write the low word of the
     // product to u_lo.
-    HURCHALLA_FORCE_INLINE T multiplyToHiLo(T& u_lo, V x, V y) const
+    HURCHALLA_FORCE_INLINE T multiplyToHiLo(T& HURCHALLA_RESTRICT u_lo, V x, V y) const
     {
         namespace hc = ::hurchalla;
         return hc::unsigned_multiply_to_hilo_product(u_lo, x.get(), y.get());
     }
-    HURCHALLA_FORCE_INLINE T squareToHiLo(T& u_lo, V x) const
+    HURCHALLA_FORCE_INLINE T squareToHiLo(T& HURCHALLA_RESTRICT u_lo, V x) const
     {
         namespace hc = ::hurchalla;
-        return hc::unsigned_multiply_to_hilo_product(u_lo, x.get(), x.get());
+        return hc::unsigned_square_to_hilo_product(u_lo, x.get());
     }
     HURCHALLA_FORCE_INLINE bool isValid(V x) const
     {

lib/modular_arithmetic/montgomery_arithmetic/include/hurchalla/montgomery_arithmetic/detail/MontyHalfRange.h

@@ -20,6 +20,7 @@
 #include "hurchalla/util/traits/ut_numeric_limits.h"
 #include "hurchalla/util/traits/extensible_make_signed.h"
 #include "hurchalla/util/signed_multiply_to_hilo_product.h"
+#include "hurchalla/util/signed_square_to_hilo_product.h"
 #include "hurchalla/util/conditional_select.h"
 #include "hurchalla/util/cselect_on_bit.h"
 #include "hurchalla/util/compiler_macros.h"
@@ -699,11 +700,11 @@ class MontyHalfRange final :
         HPBC_CLOCKWORK_POSTCONDITION2(0 <= u_hi && u_hi < n_);
         return u_hi;
     }
-    HURCHALLA_FORCE_INLINE T squareToHiLo(T& u_lo, V x) const
+    HURCHALLA_FORCE_INLINE T squareToHiLo(T& HURCHALLA_RESTRICT u_lo, V x) const
     {
         HPBC_CLOCKWORK_PRECONDITION2(isValid(x));
         namespace hc = ::hurchalla;
-        S tmp_hi = hc::signed_multiply_to_hilo_product(u_lo, x.get(), x.get());
+        S tmp_hi = hc::signed_square_to_hilo_product(u_lo, x.get());
         // The same logic as given in multiplyToHiLo shows that
         // -(n+1)/2 <= tmp_hi <= (n-1)/2.  But additionally, since the square
         // of an integer is always >= 0, we therefore know

lib/modular_arithmetic/montgomery_arithmetic/include/hurchalla/montgomery_arithmetic/detail/MontyQuarterRange.h

@@ -21,6 +21,7 @@
 #include "hurchalla/util/traits/ut_numeric_limits.h"
 #include "hurchalla/util/traits/extensible_make_signed.h"
 #include "hurchalla/util/unsigned_multiply_to_hilo_product.h"
+#include "hurchalla/util/unsigned_square_to_hilo_product.h"
 #include "hurchalla/util/cselect_on_bit.h"
 #include "hurchalla/util/compiler_macros.h"
 #include "hurchalla/modular_arithmetic/detail/clockwork_programming_by_contract.h"
@@ -490,15 +491,15 @@ class MontyQuarterRange final : public
 
     // return the high word of the product, and write the low word of the
     // product to u_lo.
-    HURCHALLA_FORCE_INLINE T multiplyToHiLo(T& u_lo, V x, V y) const
+    HURCHALLA_FORCE_INLINE T multiplyToHiLo(T& HURCHALLA_RESTRICT u_lo, V x, V y) const
     {
         namespace hc = ::hurchalla;
         return hc::unsigned_multiply_to_hilo_product(u_lo, x.get(), y.get());
     }
-    HURCHALLA_FORCE_INLINE T squareToHiLo(T& u_lo, V x) const
+    HURCHALLA_FORCE_INLINE T squareToHiLo(T& HURCHALLA_RESTRICT u_lo, V x) const
     {
         namespace hc = ::hurchalla;
-        return hc::unsigned_multiply_to_hilo_product(u_lo, x.get(), x.get());
+        return hc::unsigned_square_to_hilo_product(u_lo, x.get());
     }
     HURCHALLA_FORCE_INLINE bool isValid(V x) const
     {

lib/modular_arithmetic/montgomery_arithmetic/include/hurchalla/montgomery_arithmetic/detail/experimental/MontyFullRangeMasked.h

@@ -20,6 +20,7 @@
 #include "hurchalla/util/conditional_select.h"
 #include "hurchalla/util/cselect_on_bit.h"
 #include "hurchalla/util/unsigned_multiply_to_hilo_product.h"
+#include "hurchalla/util/unsigned_square_to_hilo_product.h"
 #include "hurchalla/util/compiler_macros.h"
 #include "hurchalla/modular_arithmetic/detail/clockwork_programming_by_contract.h"
 #include <type_traits>
@@ -423,12 +424,12 @@ class MontyFullRangeMasked final :
 
     // This function computes the full two-word product of x*x.  It returns the
     // high word of x*x, and writes the low word of x*x to u_lo.
-    HURCHALLA_FORCE_INLINE T squareToHiLo(T& u_lo, V x) const
+    HURCHALLA_FORCE_INLINE T squareToHiLo(T& HURCHALLA_RESTRICT u_lo, V x) const
     {
         HPBC_CLOCKWORK_PRECONDITION2(isValid(x));
         T a = x.getbits();
         T umlo;
-        T umhi = ::hurchalla::unsigned_multiply_to_hilo_product(umlo, a, a);
+        T umhi = ::hurchalla::unsigned_square_to_hilo_product(umlo, a);
         T masked_a = static_cast<T>(x.getmask() & a);
         T result_hi = static_cast<T>(umhi - masked_a - masked_a);
         u_lo = umlo;
@@ -511,9 +512,9 @@ class MontyFullRangeMasked final :
         // Recall that asqrHi and asqrLo are simply any values that satisfy
         // asqrHi*R + asqrLo == a*a, with 0 <= asqrHi < R  and  0 <= asqrLo < R.
         // When we compute
-        // T umlo; T umhi = unsigned_multiply_to_hilo_product(umlo, a, a);
+        // T umlo; T umhi = unsigned_square_to_hilo_product(umlo, a);
         // umhi and umlo satisfy these requirements, since all type T variables
-        // have bounds [0, R), and unsigned_multiply_to_hilo_product() computes
+        // have bounds [0, R), and unsigned_square_to_hilo_product() computes
         // the full two-word product of a*a.
         // Therefore, since
         // x*x == (asqrHi + s*(R-2*a))*R + asqrLo,  we can substitute and get
@@ -565,7 +566,7 @@ class MontyFullRangeMasked final :
         // We can express all of this in code via
         // T a = x.getbits();
         // T umlo;
-        // T umhi = ::hurchalla::unsigned_multiply_to_hilo_product(umlo, a, a);
+        // T umhi = ::hurchalla::unsigned_square_to_hilo_product(umlo, a);
         // T neg2a = static_cast<T>(-2) * a;
         // T result_hi = umhi + (x.getmask() & neg2a);
         // u_lo = umlo;

lib/modular_arithmetic/montgomery_arithmetic/include/hurchalla/montgomery_arithmetic/detail/experimental/montgomery_pow_2kary/testbench_montgomery_pow_2kary.cpp

@@ -8,6 +8,7 @@
 #include "hurchalla/montgomery_arithmetic/low_level_api/detail/platform_specific/impl_array_get_Rsquared_mod_n.h"
 #include "hurchalla/montgomery_arithmetic/MontgomeryForm.h"
 #include "hurchalla/montgomery_arithmetic/montgomery_form_aliases.h"
+#include "hurchalla/util/unsigned_square_to_hilo_product.h"
 #include "hurchalla/util/count_leading_zeros.h"
 #include "hurchalla/util/traits/ut_numeric_limits.h"
 #include "hurchalla/util/traits/safely_promote_unsigned.h"
@@ -796,7 +797,7 @@ bench_array_pow(U min, U range, U& totalU, unsigned int max_modulus_bits_reduce,
          for (size_t j=0; j < ARRAY_SIZE; ++j) {
             U modulus = tmpvec[i+j];
             U u_lo;
-            U u_hi = hurchalla::unsigned_multiply_to_hilo_product(u_lo, r_mod_n[j], r_mod_n[j]);
+            U u_hi = hurchalla::unsigned_square_to_hilo_product(u_lo, r_mod_n[j]);
             U remainder;
             //U quotient = div_2U_by_1U(u_hi, u_lo, modulus, remainder);
             div_2U_by_1U(u_hi, u_lo, modulus, remainder);

lib/modular_arithmetic/montgomery_arithmetic/include/hurchalla/montgomery_arithmetic/detail/experimental/montgomery_two_pow/experimental_montgomery_two_pow.h

@@ -3572,7 +3572,7 @@ if HURCHALLA_CPP17_CONSTEXPR (CODE_SECTION == 0) {
         if (n <= MASKBIG) {
             size_t loindex = static_cast<size_t>(n);
             HPBC_CLOCKWORK_ASSERT2(loindex < ut_numeric_limits<RU>::digits);
-#if defined(__GNUC__) && !defined(__clang__)
+#if defined(__GNUC__) && (__GNUC__ >= 14) && !defined(__clang__)
 #  pragma GCC diagnostic push
 #  pragma GCC diagnostic ignored "-Wnrvo"
 #endif
@@ -4183,6 +4183,7 @@ if HURCHALLA_CPP17_CONSTEXPR (CODE_SECTION == 0) {
         }
 
         while (shift >= P) {
+            static_assert(P > 0, "");
             shift -= P;
 
             if HURCHALLA_CPP17_CONSTEXPR (USE_SQUARING_VALUE_OPTIMIZATION) {