add ARM64 inline-asm to 128bit REDC function

Author: hurchalla

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).

modular_arithmetic/CMakeLists.txt

@@ -75,7 +75,7 @@ include(FetchContent)
 FetchContent_Declare(
     hurchalla_util
     GIT_REPOSITORY https://github.com/hurchalla/util.git
-    GIT_TAG        ba38c2c5e1164e0e7ab480e4fc959c3a43183c0e
+    GIT_TAG        6901743704ac1caf4e99090ce52e52a40147ba82
 )
 FetchContent_MakeAvailable(hurchalla_util)
 

montgomery_arithmetic/CMakeLists.txt

@@ -79,7 +79,7 @@ include(FetchContent)
 FetchContent_Declare(
     hurchalla_util
     GIT_REPOSITORY https://github.com/hurchalla/util.git
-    GIT_TAG        ba38c2c5e1164e0e7ab480e4fc959c3a43183c0e
+    GIT_TAG        6901743704ac1caf4e99090ce52e52a40147ba82
 )
 FetchContent_MakeAvailable(hurchalla_util)
 

montgomery_arithmetic/include/hurchalla/montgomery_arithmetic/detail/experimental/montgomery_two_pow/testbench_montgomery_two_pow.cpp

@@ -941,7 +941,7 @@ using namespace hurchalla;
    constexpr int NUM_TEST_REPETITIONS = 10;
 
 
-#if 0
+#if 1
    std::cout << "\nbegin benchmarks - array two_pow\n";
 
    // warm up call
@@ -955,17 +955,17 @@ using namespace hurchalla;
      for (size_t j=0; j<timingA[i].size(); ++j) {
 
       timingA[i][j].push_back(
-         bench_array_two_pow<0, 31, 3, MontType, false>(static_cast<U>(maxU - range), range, dummy, mmbr[i], seed, ebr[i]));
+         bench_array_two_pow<0, 30, 3, MontType, false>(static_cast<U>(maxU - range), range, dummy, mmbr[i], seed, ebr[i]));
       timingA[i][j].push_back(
-         bench_array_two_pow<0, 31, 4, MontType, false>(static_cast<U>(maxU - range), range, dummy, mmbr[i], seed, ebr[i]));
+         bench_array_two_pow<0, 30, 4, MontType, false>(static_cast<U>(maxU - range), range, dummy, mmbr[i], seed, ebr[i]));
       timingA[i][j].push_back(
-         bench_array_two_pow<0, 31, 5, MontType, false>(static_cast<U>(maxU - range), range, dummy, mmbr[i], seed, ebr[i]));
+         bench_array_two_pow<0, 30, 5, MontType, false>(static_cast<U>(maxU - range), range, dummy, mmbr[i], seed, ebr[i]));
       timingA[i][j].push_back(
-         bench_array_two_pow<0, 31, 6, MontType, false>(static_cast<U>(maxU - range), range, dummy, mmbr[i], seed, ebr[i]));
+         bench_array_two_pow<0, 30, 6, MontType, false>(static_cast<U>(maxU - range), range, dummy, mmbr[i], seed, ebr[i]));
       timingA[i][j].push_back(
-         bench_array_two_pow<0, 31, 7, MontType, false>(static_cast<U>(maxU - range), range, dummy, mmbr[i], seed, ebr[i]));
+         bench_array_two_pow<0, 30, 7, MontType, false>(static_cast<U>(maxU - range), range, dummy, mmbr[i], seed, ebr[i]));
       timingA[i][j].push_back(
-         bench_array_two_pow<0, 31, 8, MontType, false>(static_cast<U>(maxU - range), range, dummy, mmbr[i], seed, ebr[i]));
+         bench_array_two_pow<0, 30, 8, MontType, false>(static_cast<U>(maxU - range), range, dummy, mmbr[i], seed, ebr[i]));
 
 
 #if 0
@@ -1779,7 +1779,7 @@ std::cout << "Timings By Test Type:\n";
 
 
 
-#if 1
+#if 0
    std::cout << "\nbegin benchmarks - scalar two_pow\n";
 
    //  warm up to get cpu boost (or throttle) going
@@ -1796,7 +1796,7 @@ std::cout << "Timings By Test Type:\n";
        // format is bench_range<TABLE_BITS, USE_SLIDING_WINDOW_OPTIMIZATION, CODE_SECTION,
        //                       MontType, USE_SQUARING_VALUE_OPTIMIZATION>
       timings[i][j].push_back(
-         bench_range<0, false, 22, MontType, false>(static_cast<U>(maxU - range), range, dummy, mmbr[i], seed, ebr[i]));
+         bench_range<0, false, 34, MontType, true>(static_cast<U>(maxU - range), range, dummy, mmbr[i], seed, ebr[i]));
 
 #if 0
 // This is a copy/paste of the "best of best" code sections from further below (nothing is new here).