yukijoou/kwin
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

kwin: Add an SSE2 vertex upload path

Browse source 
Use MOVNTDQ instructions to write vertex data into the buffer object.
This commit is contained in:
Fredrik Höglund 2013-03-19 16:59:43 +01:00
parent cb3f1b7776
commit d6fadfa91c
1 changed files with 70 additions and 15 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

55
libkwineffects/kwineffects.cpp
View file

@ -48,6 +48,23 @@ along with this program. If not, see <http://www.gnu.org/licenses/>.
#include <xcb/xfixes.h>
#endif
#if defined(__GNUC__)
# define KWIN_ALIGN(n) __attribute((aligned(n)))
# if defined(__SSE2__)
# define HAVE_SSE2
# endif
#elif defined(__INTEL_COMPILER)
# define KWIN_ALIGN(n) __declspec(align(n))
# define HAVE_SSE2
#else
# define KWIN_ALIGN(n)
#endif
#ifdef HAVE_SSE2
# include <emmintrin.h>
#endif
namespace KWin
{
@ -1055,6 +1072,43 @@ void WindowQuadList::makeInterleavedArrays(GLVertex2D *vertices, const QMatrix4x
GLVertex2D *vertex = vertices;
#ifdef HAVE_SSE2
if (!(intptr_t(vertex) & 0xf)) {
for (int i = 0; i < count(); i++) {
const WindowQuad &quad = at(i);
KWIN_ALIGN(16) GLVertex2D v[4];
for (int j = 0; j < 4; j++) {
const WindowVertex &wv = quad[j];
v[j].position = QVector2D(wv.x(), wv.y());
v[j].texcoord = QVector2D(wv.u(), wv.v()) * coeff + offset;
}
const __m128i *srcP = (const __m128i *) &v;
__m128i *dstP = (__m128i *) vertex;
__m128i src[4];
src[0] = _mm_load_si128(&srcP[0]); // Top-left
src[1] = _mm_load_si128(&srcP[1]); // Top-right
src[2] = _mm_load_si128(&srcP[2]); // Bottom-right
src[3] = _mm_load_si128(&srcP[3]); // Bottom-left
// First triangle
_mm_stream_si128(&dstP[0], src[1]); // Top-right
_mm_stream_si128(&dstP[1], src[0]); // Top-left
_mm_stream_si128(&dstP[2], src[3]); // Bottom-left
// Second triangle
_mm_stream_si128(&dstP[3], src[3]); // Bottom-left
_mm_stream_si128(&dstP[4], src[2]); // Bottom-right
_mm_stream_si128(&dstP[5], src[1]); // Top-right
vertex += 6;
}
} else
#endif
{
for (int i = 0; i < count(); i++) {
const WindowQuad &quad = at(i);
GLVertex2D v[4]; // Four unique vertices / quad
@ -1077,6 +1131,7 @@ void WindowQuadList::makeInterleavedArrays(GLVertex2D *vertices, const QMatrix4x
*(vertex++) = v[1]; // Top-right
}
}
}
void WindowQuadList::makeArrays(float **vertices, float **texcoords, const QSizeF &size, bool yInverted) const
{