Merge pull request #11351 from JosJuice/bitset-cpp20

Common: Use C++20 <bit> header in BitSet.h

Source/Core/Common/BitSet.h

@@ -2,89 +2,15 @@
 
 #pragma once
 
+#include <bit>
 #include <cstddef>
 #include <initializer_list>
 #include <type_traits>
+
 #include "Common/CommonTypes.h"
 
-#ifdef _WIN32
-
-#include <intrin.h>
-
 namespace Common
 {
-template <typename T>
-constexpr int CountSetBits(T v)
-{
-  // 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;
-}
-inline int LeastSignificantSetBit(u8 val)
-{
-  unsigned long index;
-  _BitScanForward(&index, val);
-  return (int)index;
-}
-inline int LeastSignificantSetBit(u16 val)
-{
-  unsigned long index;
-  _BitScanForward(&index, val);
-  return (int)index;
-}
-inline int LeastSignificantSetBit(u32 val)
-{
-  unsigned long index;
-  _BitScanForward(&index, val);
-  return (int)index;
-}
-inline int LeastSignificantSetBit(u64 val)
-{
-  unsigned long index;
-  _BitScanForward64(&index, val);
-  return (int)index;
-}
-#else
-namespace Common
-{
-constexpr int CountSetBits(u8 val)
-{
-  return __builtin_popcount(val);
-}
-constexpr int CountSetBits(u16 val)
-{
-  return __builtin_popcount(val);
-}
-constexpr int CountSetBits(u32 val)
-{
-  return __builtin_popcount(val);
-}
-constexpr int CountSetBits(u64 val)
-{
-  return __builtin_popcountll(val);
-}
-inline int LeastSignificantSetBit(u8 val)
-{
-  return __builtin_ctz(val);
-}
-inline int LeastSignificantSetBit(u16 val)
-{
-  return __builtin_ctz(val);
-}
-inline int LeastSignificantSetBit(u32 val)
-{
-  return __builtin_ctz(val);
-}
-inline int LeastSignificantSetBit(u64 val)
-{
-  return __builtin_ctzll(val);
-}
-#endif
-
 // Similar to std::bitset, this is a class which encapsulates a bitset, i.e.
 // using the set bits of an integer to represent a set of integers.  Like that
 // class, it acts like an array of bools:
@@ -146,7 +72,7 @@ public:
       }
       else
       {
-        int bit = LeastSignificantSetBit(m_val);
+        int bit = std::countr_zero(m_val);
         m_val &= ~(1 << bit);
         m_bit = bit;
       }
@@ -203,7 +129,7 @@ public:
   // 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 unsigned int Count() const { return std::popcount(m_val); }
   constexpr Iterator begin() const { return ++Iterator(m_val, 0); }
   constexpr Iterator end() const { return Iterator(m_val, -1); }
   IntTy m_val;

Source/Core/VideoBackends/D3D/D3DState.cpp

@@ -5,9 +5,9 @@
 
 #include <algorithm>
 #include <array>
+#include <bit>
 
 #include "Common/Assert.h"
-#include "Common/BitSet.h"
 #include "Common/CommonTypes.h"
 #include "Common/Logging/Log.h"
 #include "Common/MsgHandler.h"
@@ -163,8 +163,8 @@ void StateManager::Apply()
 
 void StateManager::ApplyTextures()
 {
-  const int textureMaskShift = Common::LeastSignificantSetBit((u32)DirtyFlag_Texture0);
+  const int textureMaskShift = std::countr_zero((u32)DirtyFlag_Texture0);
-  const int samplerMaskShift = Common::LeastSignificantSetBit((u32)DirtyFlag_Sampler0);
+  const int samplerMaskShift = std::countr_zero((u32)DirtyFlag_Sampler0);
 
   u32 dirtyTextures =
       (m_dirtyFlags &
@@ -178,7 +178,7 @@ void StateManager::ApplyTextures()
       samplerMaskShift;
   while (dirtyTextures)
   {
-    const int index = Common::LeastSignificantSetBit(dirtyTextures);
+    const int index = std::countr_zero(dirtyTextures);
     if (m_current.textures[index] != m_pending.textures[index])
     {
       D3D::context->PSSetShaderResources(index, 1, &m_pending.textures[index]);
@@ -190,7 +190,7 @@ void StateManager::ApplyTextures()
 
   while (dirtySamplers)
   {
-    const int index = Common::LeastSignificantSetBit(dirtySamplers);
+    const int index = std::countr_zero(dirtySamplers);
     if (m_current.samplers[index] != m_pending.samplers[index])
     {
       D3D::context->PSSetSamplers(index, 1, &m_pending.samplers[index]);
@@ -221,7 +221,7 @@ void StateManager::SetTextureByMask(u32 textureSlotMask, ID3D11ShaderResourceVie
 {
   while (textureSlotMask)
   {
-    const int index = Common::LeastSignificantSetBit(textureSlotMask);
+    const int index = std::countr_zero(textureSlotMask);
     SetTexture(index, srv);
     textureSlotMask &= ~(1 << index);
   }

Source/Core/VideoCommon/VertexLoaderBase.cpp

@@ -4,6 +4,7 @@
 #include "VideoCommon/VertexLoaderBase.h"
 
 #include <array>
+#include <bit>
 #include <cstring>
 #include <memory>
 #include <string>
@@ -12,7 +13,6 @@
 #include <fmt/format.h>
 
 #include "Common/Assert.h"
-#include "Common/BitSet.h"
 #include "Common/CommonTypes.h"
 #include "Common/Logging/Log.h"
 #include "Common/MsgHandler.h"
@@ -99,7 +99,7 @@ u32 VertexLoaderBase::GetVertexSize(const TVtxDesc& vtx_desc, const VAT& vtx_att
   u32 size = 0;
 
   // Each enabled TexMatIdx adds one byte, as does PosMatIdx
-  size += Common::CountSetBits(vtx_desc.low.Hex & 0x1FF);
+  size += std::popcount(vtx_desc.low.Hex & 0x1FF);
 
   const u32 pos_size = VertexLoader_Position::GetSize(vtx_desc.low.Position, vtx_attr.g0.PosFormat,
                                                       vtx_attr.g0.PosElements);