[libc] Make BigInt bitwise shift consistent with regular integral s…

…emantics. (#87874)

This patch removes the test for cases where the shift operand is greater
or equal to the bit width of the number. This is done for two reasons,
first it makes `BigInt` consistent with regular integral bitwise shift
semantics, and second it makes the shift operation faster. The shift
operation is on the critical path for `exp` and `log` operations, see
#86137 (comment).

libc/src/__support/FPUtil/dyadic_float.h

@@ -122,15 +122,17 @@ template <size_t Bits> struct DyadicFloat {
 
     int exp_lo = exp_hi - static_cast<int>(PRECISION) - 1;
 
-    MantissaType m_hi(mantissa >> shift);
+    MantissaType m_hi =
+        shift >= MantissaType::BITS ? MantissaType(0) : mantissa >> shift;
 
     T d_hi = FPBits<T>::create_value(
                  sign, exp_hi,
                  (static_cast<output_bits_t>(m_hi) & FPBits<T>::SIG_MASK) |
                      IMPLICIT_MASK)
                  .get_val();
 
-    MantissaType round_mask = MantissaType(1) << (shift - 1);
+    MantissaType round_mask =
+        shift > MantissaType::BITS ? 0 : MantissaType(1) << (shift - 1);
     MantissaType sticky_mask = round_mask - MantissaType(1);
 
     bool round_bit = !(mantissa & round_mask).is_zero();

libc/src/__support/big_int.h

@@ -249,18 +249,14 @@ LIBC_INLINE constexpr bool is_negative(cpp::array<word, N> &array) {
 enum Direction { LEFT, RIGHT };
 
 // A bitwise shift on an array of elements.
-// TODO: Make the result UB when 'offset' is greater or equal to the number of
-// bits in 'array'. This will allow for better code performance.
+// 'offset' must be less than TOTAL_BITS (i.e., sizeof(word) * CHAR_BIT * N)
+// otherwise the behavior is undefined.
 template <Direction direction, bool is_signed, typename word, size_t N>
 LIBC_INLINE constexpr cpp::array<word, N> shift(cpp::array<word, N> array,
                                                 size_t offset) {
   static_assert(direction == LEFT || direction == RIGHT);
   constexpr size_t WORD_BITS = cpp::numeric_limits<word>::digits;
   constexpr size_t TOTAL_BITS = N * WORD_BITS;
-  if (LIBC_UNLIKELY(offset == 0))
-    return array;
-  if (LIBC_UNLIKELY(offset >= TOTAL_BITS))
-    return {};
 #ifdef LIBC_TYPES_HAS_INT128
   if constexpr (TOTAL_BITS == 128) {
     using type = cpp::conditional_t<is_signed, __int128_t, __uint128_t>;
@@ -272,6 +268,8 @@ LIBC_INLINE constexpr cpp::array<word, N> shift(cpp::array<word, N> array,
     return cpp::bit_cast<cpp::array<word, N>>(tmp);
   }
 #endif
+  if (LIBC_UNLIKELY(offset == 0))
+    return array;
   const bool is_neg = is_signed && is_negative(array);
   constexpr auto at = [](size_t index) -> int {
     // reverse iteration when direction == LEFT.

libc/test/src/__support/big_int_test.cpp

@@ -192,7 +192,7 @@ TYPED_TEST(LlvmLibcUIntClassTest, Masks, Types) {
 
 TYPED_TEST(LlvmLibcUIntClassTest, CountBits, Types) {
   if constexpr (!T::SIGNED) {
-    for (size_t i = 0; i <= T::BITS; ++i) {
+    for (size_t i = 0; i < T::BITS; ++i) {
       const auto l_one = T::all_ones() << i; // 0b111...000
       const auto r_one = T::all_ones() >> i; // 0b000...111
       const int zeros = i;
@@ -559,10 +559,6 @@ TEST(LlvmLibcUIntClassTest, ShiftLeftTests) {
   LL_UInt128 result5({0, 0x2468ace000000000});
   EXPECT_EQ((val2 << 100), result5);
 
-  LL_UInt128 result6({0, 0});
-  EXPECT_EQ((val2 << 128), result6);
-  EXPECT_EQ((val2 << 256), result6);
-
   LL_UInt192 val3({1, 0, 0});
   LL_UInt192 result7({0, 1, 0});
   EXPECT_EQ((val3 << 64), result7);
@@ -589,10 +585,6 @@ TEST(LlvmLibcUIntClassTest, ShiftRightTests) {
   LL_UInt128 result5({0x0000000001234567, 0});
   EXPECT_EQ((val2 >> 100), result5);
 
-  LL_UInt128 result6({0, 0});
-  EXPECT_EQ((val2 >> 128), result6);
-  EXPECT_EQ((val2 >> 256), result6);
-
   LL_UInt128 v1({0x1111222233334444, 0xaaaabbbbccccdddd});
   LL_UInt128 r1({0xaaaabbbbccccdddd, 0});
   EXPECT_EQ((v1 >> 64), r1);