Reduce code stink (#1818)

CRC32.cpp:
Make table initialization compile time

DSi_NAND.cpp:
Fix file close / unmount / disk close on error
~L427: Remove redundant calls, as they are immediately rendered useless by `rem` being overwritten

NDS.cpp / FreeBIOS.h:
Remove unneeded size values in header
Remove unneeded memset's as they are initialized anyway

sha1.c / sha1.h:
Fix useless warning

Wifi.cpp:
Remove unneeded includes

DSi.cpp:
Reduce ugly casts
Deduplicate code

qt_sdl/main.cpp:
silence clang switch statement warning

qt_sdl/main.h:
fix override warnings

dolphin/BitSet.h:
use msvc extensions only when appropriate, fix broken bit set count under _WIN32

1 files changed, 59 insertions, 51 deletions

diff --git a/src/dolphin/BitSet.h b/src/dolphin/BitSet.h
index d32b020..09cc1ce 100644
--- a/src/dolphin/BitSet.h
+++ b/src/dolphin/BitSet.h
@@ -3,85 +3,97 @@
 #pragma once
 
 #include <cstddef>
+#include <bitset>
 #include <initializer_list>
 #include <type_traits>
 #include "../types.h"
 
-#ifdef _WIN32
-
-#include <intrin.h>
-
 namespace Common
 {
-template <typename T>
-constexpr int CountSetBits(T v)
+#if defined(__GNUC__) || defined(__clang__)
+__attribute((always_inline)) static constexpr int CountSetBits(u8 val)
 {
-  // from https://graphics.stanford.edu/~seander/bithacks.html
-  // GCC has this built in, but MSVC's intrinsic will only emit the actual
-  // POPCNT instruction, which we're not depending on
-  v = v - ((v >> 1) & (T) ~(T)0 / 3);
-  v = (v & (T) ~(T)0 / 15 * 3) + ((v >> 2) & (T) ~(T)0 / 15 * 3);
-  v = (v + (v >> 4)) & (T) ~(T)0 / 255 * 15;
-  return (T)(v * ((T) ~(T)0 / 255)) >> (sizeof(T) - 1) * 8;
+  return __builtin_popcount(val);
 }
-inline int LeastSignificantSetBit(u8 val)
+__attribute((always_inline)) static constexpr int CountSetBits(u16 val)
 {
-  unsigned long index;
-  _BitScanForward(&index, val);
-  return (int)index;
+  return __builtin_popcount(val);
 }
-inline int LeastSignificantSetBit(u16 val)
+__attribute((always_inline)) static constexpr int CountSetBits(u32 val)
 {
-  unsigned long index;
-  _BitScanForward(&index, val);
-  return (int)index;
+  return __builtin_popcount(val);
 }
-inline int LeastSignificantSetBit(u32 val)
+__attribute((always_inline)) static constexpr int CountSetBits(u64 val)
 {
-  unsigned long index;
-  _BitScanForward(&index, val);
-  return (int)index;
+  return __builtin_popcountll(val);
 }
-inline int LeastSignificantSetBit(u64 val)
+__attribute((always_inline)) static constexpr int LeastSignificantSetBit(u8 val)
 {
-  unsigned long index;
-  _BitScanForward64(&index, val);
-  return (int)index;
+  return __builtin_ctz(val);
 }
-#else
-namespace Common
+__attribute((always_inline)) static constexpr int LeastSignificantSetBit(u16 val)
 {
-constexpr int CountSetBits(u8 val)
+  return __builtin_ctz(val);
+}
+__attribute((always_inline)) static constexpr int LeastSignificantSetBit(u32 val)
 {
-  return __builtin_popcount(val);
+  return __builtin_ctz(val);
 }
-constexpr int CountSetBits(u16 val)
+__attribute((always_inline)) static constexpr int LeastSignificantSetBit(u64 val)
 {
-  return __builtin_popcount(val);
+  return __builtin_ctzll(val);
 }
-constexpr int CountSetBits(u32 val)
+#elif defined(_MSC_VER)
+#include <intrin.h>
+// MSVC __popcnt doesn't switch between hardware availability like gcc does, can't use it, let C++ implementation handle it
+__forceinline static int CountSetBits(u8 val)
 {
-  return __builtin_popcount(val);
+  return std::bitset<8>(val).count();
 }
-constexpr int CountSetBits(u64 val)
+__forceinline static int CountSetBits(u16 val)
 {
-  return __builtin_popcountll(val);
+  return std::bitset<16>(val).count();
 }
-inline int LeastSignificantSetBit(u8 val)
+__forceinline static int CountSetBits(u32 val)
 {
-  return __builtin_ctz(val);
+  return std::bitset<32>(val).count();
 }
-inline int LeastSignificantSetBit(u16 val)
+__forceinline static int CountSetBits(u64 val)
 {
-  return __builtin_ctz(val);
+  return std::bitset<64>(val).count();
 }
-inline int LeastSignificantSetBit(u32 val)
+__forceinline static int LeastSignificantSetBit(u8 val)
 {
-  return __builtin_ctz(val);
+  unsigned long count;
+  _BitScanForward(&count, val);
+  return count;
 }
-inline int LeastSignificantSetBit(u64 val)
+__forceinline static int LeastSignificantSetBit(u16 val)
 {
-  return __builtin_ctzll(val);
+  unsigned long count;
+  _BitScanForward(&count, val);
+  return count;
+}
+__forceinline static int LeastSignificantSetBit(u32 val)
+{
+  unsigned long count;
+  _BitScanForward(&count, val);
+  return count;
+}
+__forceinline static int LeastSignificantSetBit(u64 val)
+{
+#if defined(_WIN64)
+  unsigned long count;
+  _BitScanForward64(&count, val);
+  return count;
+#else
+  unsigned long tmp;
+  _BitScanForward(&tmp, (u32)(val & 0x00000000FFFFFFFFull));
+  if (tmp)
+    return tmp;
+  _BitScanForward(&tmp, (u32)((val & 0xFFFFFFFF00000000ull) >> 32));
+  return tmp ? tmp + 32 : 0;
+#endif
 }
 #endif
 
@@ -201,10 +213,6 @@ public:
   BitSet& operator^=(BitSet other) { return *this = *this ^ other; }
   BitSet& operator<<=(IntTy shift) { return *this = *this << shift; }
   BitSet& operator>>=(IntTy shift) { return *this = *this >> shift; }
-  // Warning: Even though on modern CPUs this is a single fast instruction,
-  // Dolphin's official builds do not currently assume POPCNT support on x86,
-  // so slower explicit bit twiddling is generated.  Still should generally
-  // be faster than a loop.
   constexpr unsigned int Count() const { return CountSetBits(m_val); }
   constexpr Iterator begin() const { return ++Iterator(m_val, 0); }
   constexpr Iterator end() const { return Iterator(m_val, -1); }