rooty/dxvk
1
0
Fork 0
mirror of https://github.com/doitsujin/dxvk.git synced 2025-02-23 10:54:14 +01:00
Code Issues Projects Releases Wiki Activity

[util] Add more optimal popcnt implementation

Browse Source
This commit is contained in:
Philip Rebohle 2025-02-18 14:02:13 +01:00
parent 4282829f38
commit 4c8ee300b5
2 changed files with 9 additions and 13 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
src/dxvk/dxvk_image.cpp
View File

@ -268,7 +268,7 @@ namespace dxvk {
uint16_t mipMask = ((1u << subresources.levelCount) - 1u) << subresources.baseMipLevel;
for (uint32_t i = subresources.baseArrayLayer; i < subresources.baseArrayLayer + subresources.layerCount; i++) {
m_uninitializedSubresourceCount -= bit::popcnt(m_uninitializedMipsPerLayer[i] & mipMask);
m_uninitializedSubresourceCount -= bit::popcnt(uint16_t(m_uninitializedMipsPerLayer[i] & mipMask));
m_uninitializedMipsPerLayer[i] &= ~mipMask;
}

20
src/util/util_bit.h
View File

@ -54,19 +54,15 @@ namespace dxvk::bit {
return (value >> fst) & ~(~T(0) << (lst - fst + 1));
}
inline uint32_t popcntStep(uint32_t n, uint32_t mask, uint32_t shift) {
return (n & mask) + ((n & ~mask) >> shift);
template<typename T>
T popcnt(T n) {
n -= ((n >> 1u) & T(0x5555555555555555ull));
n = (n & T(0x3333333333333333ull)) + ((n >> 2u) & T(0x3333333333333333ull));
n = (n + (n >> 4u)) & T(0x0f0f0f0f0f0f0f0full);
n *= T(0x0101010101010101ull);
return n >> (8u * (sizeof(T) - 1u));
}
inline uint32_t popcnt(uint32_t n) {
n = popcntStep(n, 0x55555555, 1);
n = popcntStep(n, 0x33333333, 2);
n = popcntStep(n, 0x0F0F0F0F, 4);
n = popcntStep(n, 0x00FF00FF, 8);
n = popcntStep(n, 0x0000FFFF, 16);
return n;
}
inline uint32_t tzcnt(uint32_t n) {
#if defined(_MSC_VER) && !defined(__clang__)
return _tzcnt_u32(n);