[libc][stdfix] Implement fxdivi functions

Signed-off-by: Shreeyash Pandey <[email protected]>

Author: bojle

libc/config/linux/riscv/entrypoints.txt

@@ -986,6 +986,7 @@ if(LIBC_COMPILER_HAS_FIXED_POINT)
     libc.src.stdfix.idivulr
     libc.src.stdfix.idivuk
     libc.src.stdfix.idivulk
+    libc.src.stdfix.rdivi
   )
 endif()
 

libc/config/linux/x86_64/entrypoints.txt

@@ -1019,6 +1019,7 @@ if(LIBC_COMPILER_HAS_FIXED_POINT)
     libc.src.stdfix.idivulr
     libc.src.stdfix.idivuk
     libc.src.stdfix.idivulk
+    libc.src.stdfix.rdivi
   )
 endif()
 

libc/include/stdfix.yaml

@@ -544,3 +544,11 @@ functions:
     arguments:
       - type: unsigned long accum
     guard: LIBC_COMPILER_HAS_FIXED_POINT
+  - name: rdivi
+    standards:
+      - stdc_ext
+    return_type: fract
+    arguments:
+      - type: int
+      - type: int
+    guard: LIBC_COMPILER_HAS_FIXED_POINT

libc/src/__support/fixed_point/fx_bits.h

@@ -224,6 +224,47 @@ idiv(T x, T y) {
   return static_cast<XType>(result);
 }
 
+inline long accum nrstep(long accum d, long accum x0) {
+  auto v = x0 * (2 - (d * x0));
+  return v;
+}
+
+/* Divide the two integers and return a fixed_point value
+ *
+ * For reference, see:
+ * https://en.wikipedia.org/wiki/Division_algorithm#Newton%E2%80%93Raphson_division
+ * https://stackoverflow.com/a/9231996
+ */
+template <typename XType> LIBC_INLINE constexpr XType divi(int n, int d) {
+  // If the value of the second operand of the / operator is zero, the
+  // behavior is undefined. Ref: ISO/IEC TR 18037:2008(E) p.g. 16
+  LIBC_CRASH_ON_VALUE(d, 0);
+
+  unsigned int nv = static_cast<unsigned int>(n);
+  unsigned int dv = static_cast<unsigned int>(d);
+  unsigned int clz = cpp::countl_zero<unsigned int>(dv) - 1;
+  unsigned long int scaled_val = dv << clz;
+  /* Scale denominator to be in the range of [0.5,1] */
+  FXBits<long accum> d_scaled{scaled_val};
+  unsigned long int scaled_val_n = nv << clz;
+  /* Scale the numerator as much as the denominator to maintain correctness of
+   * the original equation
+   */
+  FXBits<long accum> n_scaled{scaled_val_n};
+  long accum n_scaled_val = n_scaled.get_val();
+  long accum d_scaled_val = d_scaled.get_val();
+  /* x0 = (48/17) - (32/17) * d_n */
+  long accum a = 2.8235lk; /* 48/17 */
+  long accum b = 1.8823lk; /* 32/17 */
+  long accum initial_approx = a - (b * d_scaled_val);
+  long accum val = nrstep(d_scaled_val, initial_approx);
+  val = nrstep(d_scaled_val, val);
+  val = nrstep(d_scaled_val, val);
+  val = nrstep(d_scaled_val, val);
+  long accum res = n_scaled_val * val;
+  return static_cast<XType>(res);
+}
+
 } // namespace fixed_point
 } // namespace LIBC_NAMESPACE_DECL
 

libc/src/stdfix/CMakeLists.txt

@@ -89,6 +89,20 @@ foreach(suffix IN ITEMS r lr k lk ur ulr uk ulk)
   )
 endforeach()
 
+foreach(suffix IN ITEMS r)
+  add_entrypoint_object(
+    ${suffix}divi
+    HDRS
+      ${suffix}divi.h
+    SRCS
+      ${suffix}divi.cpp
+    COMPILE_OPTIONS
+      ${libc_opt_high_flag}
+    DEPENDS
+      libc.src.__support.fixed_point.fx_bits
+  )
+endforeach()
+
 add_entrypoint_object(
   uhksqrtus
   HDRS

libc/test/src/stdfix/CMakeLists.txt

@@ -120,6 +120,22 @@ foreach(suffix IN ITEMS r lr k lk ur ulr uk ulk)
   )
 endforeach()
 
+foreach(suffix IN ITEMS r)
+  add_libc_test(
+    ${suffix}divi_test
+    SUITE
+      libc-stdfix-tests
+    HDRS
+      DivITest.h
+    SRCS
+      ${suffix}divi_test.cpp
+    DEPENDS
+      libc.src.stdfix.${suffix}divi
+      libc.src.__support.fixed_point.fx_bits
+      libc.hdr.signal_macros
+  )
+endforeach()
+
 add_libc_test(
   uhksqrtus_test
   SUITE