[host] memory performance improvements

This commit is contained in:
Geoffrey McRae 2017-11-17 11:54:42 +11:00
parent 534e9425f1
commit 9de71ed2da
2 changed files with 72 additions and 2 deletions

View File

@ -260,7 +260,7 @@ bool NvFBC::GrabFrame(struct FrameInfo & frame)
uint8_t *src = (uint8_t *)m_frameBuffer + dataOffset; uint8_t *src = (uint8_t *)m_frameBuffer + dataOffset;
uint8_t *dst = (uint8_t *)frame.buffer; uint8_t *dst = (uint8_t *)frame.buffer;
for(unsigned int y = 0; y < frame.height; ++y, dst += dataWidth, src += m_grabInfo.dwBufferWidth * 3) for(unsigned int y = 0; y < frame.height; ++y, dst += dataWidth, src += m_grabInfo.dwBufferWidth * 3)
memcpy_s(dst, dataWidth, src, dataWidth); Util::Memcpy64(dst, src, dataWidth);
return true; return true;
} }

View File

@ -60,7 +60,7 @@ public:
return defaultPath; return defaultPath;
} }
static void BGRAtoRGB(uint8_t * orig, size_t imagesize, uint8_t * dest) static inline void BGRAtoRGB(uint8_t * orig, size_t imagesize, uint8_t * dest)
{ {
assert((uintptr_t)orig % 16 == 0); assert((uintptr_t)orig % 16 == 0);
assert((uintptr_t)dest % 16 == 0); assert((uintptr_t)dest % 16 == 0);
@ -86,6 +86,9 @@ public:
uint8_t *end = orig + imagesize * 4; uint8_t *end = orig + imagesize * 4;
for (; orig != end; orig += 64, dest += 48) for (; orig != end; orig += 64, dest += 48)
{ {
_mm_prefetch((char *)(orig + 128), _MM_HINT_NTA);
_mm_prefetch((char *)(orig + 192), _MM_HINT_NTA);
__m128i v0 = _mm_shuffle_epi8(_mm_load_si128((__m128i *)&orig[0 ]), mask_right); __m128i v0 = _mm_shuffle_epi8(_mm_load_si128((__m128i *)&orig[0 ]), mask_right);
__m128i v1 = _mm_shuffle_epi8(_mm_load_si128((__m128i *)&orig[16]), mask_left ); __m128i v1 = _mm_shuffle_epi8(_mm_load_si128((__m128i *)&orig[16]), mask_left );
__m128i v2 = _mm_shuffle_epi8(_mm_load_si128((__m128i *)&orig[32]), mask_left ); __m128i v2 = _mm_shuffle_epi8(_mm_load_si128((__m128i *)&orig[32]), mask_left );
@ -100,4 +103,71 @@ public:
_mm_stream_si128((__m128i *)&dest[32], v2); _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);
}
}
}; };