[host] altered to use common memcpySSE

This commit is contained in:
Geoffrey McRae 2017-11-17 14:10:33 +11:00
parent 05d83d2311
commit ad9b78a7cc
2 changed files with 2 additions and 68 deletions

View file

@ -22,6 +22,7 @@ using namespace Capture;
#include <string>
#include "common\debug.h"
#include "common\memcpySSE.h"
#include "Util.h"
#ifdef _WIN64
@ -260,7 +261,7 @@ bool NvFBC::GrabFrame(struct FrameInfo & frame)
uint8_t *src = (uint8_t *)m_frameBuffer + dataOffset;
uint8_t *dst = (uint8_t *)frame.buffer;
for(unsigned int y = 0; y < frame.height; ++y, dst += dataWidth, src += m_grabInfo.dwBufferWidth * 3)
Util::Memcpy64(dst, src, dataWidth);
memcpySSE(dst, src, dataWidth);
return true;
}

View file

@ -103,71 +103,4 @@ public:
_mm_stream_si128((__m128i *)&dest[32], v2);
}
}
static inline void Memcpy64(void * dst, void * src, size_t length)
{
// check if we can't perform an aligned copy
if (((uintptr_t)src & 0xF) != ((uintptr_t)dst & 0xF))
{
static bool unalignedDstWarn = false;
if (!unalignedDstWarn)
{
DEBUG_WARN("Memcpy64 unable to perform aligned copy, performance will suffer");
unalignedDstWarn = true;
}
memcpy(dst, src, length);
return;
}
// check if the source needs slight alignment
{
uint8_t * _src = (uint8_t *)src;
unsigned int count = (16 - ((uintptr_t)src & 0xF)) & 0xF;
static bool unalignedSrcWarn = false;
if (count > 0)
{
if (!unalignedSrcWarn)
{
DEBUG_WARN("Memcpy64 unaligned source, performance will suffer");
unalignedSrcWarn = true;
}
uint8_t * _dst = (uint8_t *)dst;
for (unsigned int i = count; i > 0; --i)
*_dst++ = *_src++;
src = _src;
dst = _dst;
length -= count;
}
}
// perform the SMID copy
__m128i * _src = (__m128i *)src;
__m128i * _dst = (__m128i *)dst;
__m128i * _end = (__m128i *)src + (length / 16);
for (; _src != _end; _src += 8, _dst += 8)
{
_mm_prefetch((char *)(_src + 16), _MM_HINT_NTA);
_mm_prefetch((char *)(_src + 24), _MM_HINT_NTA);
__m128i v0 = _mm_load_si128(_src + 0);
__m128i v1 = _mm_load_si128(_src + 1);
__m128i v2 = _mm_load_si128(_src + 2);
__m128i v3 = _mm_load_si128(_src + 3);
__m128i v4 = _mm_load_si128(_src + 4);
__m128i v5 = _mm_load_si128(_src + 5);
__m128i v6 = _mm_load_si128(_src + 6);
__m128i v7 = _mm_load_si128(_src + 7);
_mm_stream_si128(_dst + 0, v0);
_mm_stream_si128(_dst + 1, v1);
_mm_stream_si128(_dst + 2, v2);
_mm_stream_si128(_dst + 3, v3);
_mm_stream_si128(_dst + 4, v4);
_mm_stream_si128(_dst + 5, v5);
_mm_stream_si128(_dst + 6, v6);
_mm_stream_si128(_dst + 7, v7);
}
}
};