mirror of
https://gitlab.freedesktop.org/gstreamer/gstreamer.git
synced 2024-07-05 06:05:53 +00:00
cog: Add fast paths for colorspace conversion
This commit is contained in:
David Schleef
parent
62d7f1f991
commit
135908dfe1
|
|
@ -756,6 +756,72 @@ convert_Y444_AYUV (CogFrame * dest, CogFrame * src)
|
|||
src->components[2].stride, dest->width, dest->height);
|
||||
}
|
||||
|
||||
static void
|
||||
convert_AYUV_ARGB (CogFrame * dest, CogFrame * src)
|
||||
{
|
||||
cogorc_convert_AYUV_ARGB (dest->components[0].data,
|
||||
dest->components[0].stride, src->components[0].data,
|
||||
src->components[0].stride, dest->width, dest->height);
|
||||
}
|
||||
|
||||
static void
|
||||
convert_AYUV_BGRA (CogFrame * dest, CogFrame * src)
|
||||
{
|
||||
cogorc_convert_AYUV_BGRA (dest->components[0].data,
|
||||
dest->components[0].stride, src->components[0].data,
|
||||
src->components[0].stride, dest->width, dest->height);
|
||||
}
|
||||
|
||||
static void
|
||||
convert_AYUV_ABGR (CogFrame * dest, CogFrame * src)
|
||||
{
|
||||
cogorc_convert_AYUV_ABGR (dest->components[0].data,
|
||||
dest->components[0].stride, src->components[0].data,
|
||||
src->components[0].stride, dest->width, dest->height);
|
||||
}
|
||||
|
||||
static void
|
||||
convert_AYUV_RGBA (CogFrame * dest, CogFrame * src)
|
||||
{
|
||||
cogorc_convert_AYUV_RGBA (dest->components[0].data,
|
||||
dest->components[0].stride, src->components[0].data,
|
||||
src->components[0].stride, dest->width, dest->height);
|
||||
}
|
||||
|
||||
static void
|
||||
convert_I420_BGRA (CogFrame * dest, CogFrame * src)
|
||||
{
|
||||
int i;
|
||||
int quality = 0;
|
||||
|
||||
if (quality > 3) {
|
||||
for (i = 0; i < dest->height; i++) {
|
||||
if (i & 1) {
|
||||
cogorc_convert_I420_BGRA_avg (COG_FRAME_DATA_GET_LINE (dest->components
|
||||
+ 0, i), COG_FRAME_DATA_GET_LINE (src->components + 0, i),
|
||||
COG_FRAME_DATA_GET_LINE (src->components + 1, i >> 1),
|
||||
COG_FRAME_DATA_GET_LINE (src->components + 1, (i >> 1) + 1),
|
||||
COG_FRAME_DATA_GET_LINE (src->components + 2, i >> 1),
|
||||
COG_FRAME_DATA_GET_LINE (src->components + 2, (i >> 1) + 1),
|
||||
dest->width);
|
||||
} else {
|
||||
cogorc_convert_I420_BGRA (COG_FRAME_DATA_GET_LINE (dest->components + 0,
|
||||
i), COG_FRAME_DATA_GET_LINE (src->components + 0, i),
|
||||
COG_FRAME_DATA_GET_LINE (src->components + 1, i >> 1),
|
||||
COG_FRAME_DATA_GET_LINE (src->components + 2, i >> 1), dest->width);
|
||||
}
|
||||
}
|
||||
} else {
|
||||
for (i = 0; i < dest->height; i++) {
|
||||
cogorc_convert_I420_BGRA (COG_FRAME_DATA_GET_LINE (dest->components + 0,
|
||||
i), COG_FRAME_DATA_GET_LINE (src->components + 0, i),
|
||||
COG_FRAME_DATA_GET_LINE (src->components + 1, i >> 1),
|
||||
COG_FRAME_DATA_GET_LINE (src->components + 2, i >> 1), dest->width);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
|
@ -801,6 +867,17 @@ static CogColorspaceTransform transforms[] = {
|
|||
{GST_VIDEO_FORMAT_Y444, GST_VIDEO_FORMAT_UYVY, convert_Y444_UYVY},
|
||||
{GST_VIDEO_FORMAT_Y444, GST_VIDEO_FORMAT_AYUV, convert_Y444_AYUV},
|
||||
{GST_VIDEO_FORMAT_Y444, GST_VIDEO_FORMAT_Y42B, convert_Y444_Y42B},
|
||||
|
||||
{GST_VIDEO_FORMAT_AYUV, GST_VIDEO_FORMAT_ARGB, convert_AYUV_ARGB},
|
||||
{GST_VIDEO_FORMAT_AYUV, GST_VIDEO_FORMAT_BGRA, convert_AYUV_BGRA},
|
||||
{GST_VIDEO_FORMAT_AYUV, GST_VIDEO_FORMAT_xRGB, convert_AYUV_ARGB}, /* alias */
|
||||
{GST_VIDEO_FORMAT_AYUV, GST_VIDEO_FORMAT_BGRx, convert_AYUV_BGRA}, /* alias */
|
||||
{GST_VIDEO_FORMAT_AYUV, GST_VIDEO_FORMAT_ABGR, convert_AYUV_ABGR},
|
||||
{GST_VIDEO_FORMAT_AYUV, GST_VIDEO_FORMAT_RGBA, convert_AYUV_RGBA},
|
||||
{GST_VIDEO_FORMAT_AYUV, GST_VIDEO_FORMAT_xBGR, convert_AYUV_ABGR}, /* alias */
|
||||
{GST_VIDEO_FORMAT_AYUV, GST_VIDEO_FORMAT_RGBx, convert_AYUV_RGBA}, /* alias */
|
||||
|
||||
{GST_VIDEO_FORMAT_I420, GST_VIDEO_FORMAT_BGRA, convert_I420_BGRA},
|
||||
};
|
||||
|
||||
static GstFlowReturn
|
||||
|
|
@ -860,7 +937,7 @@ gst_cogcolorspace_transform (GstBaseTransform * base_transform,
|
|||
}
|
||||
}
|
||||
|
||||
GST_ERROR ("no match");
|
||||
GST_DEBUG ("no fastpath match %d %d", in_format, out_format);
|
||||
}
|
||||
|
||||
switch (out_format) {
|
||||
|
|
|
|||
|
|
@ -1026,30 +1026,401 @@ mergewl ayuv, ay, uv
|
|||
|
||||
|
||||
|
||||
.function cogorc_convert_AYUV_ARGB
|
||||
.flags 2d
|
||||
.dest 4 argb
|
||||
.source 4 ayuv
|
||||
.temp 2 t1
|
||||
.temp 2 t2
|
||||
.temp 1 a
|
||||
.temp 1 y
|
||||
.temp 1 u
|
||||
.temp 1 v
|
||||
.temp 2 wy
|
||||
.temp 2 wu
|
||||
.temp 2 wv
|
||||
.temp 2 wr
|
||||
.temp 2 wg
|
||||
.temp 2 wb
|
||||
.temp 1 r
|
||||
.temp 1 g
|
||||
.temp 1 b
|
||||
.temp 4 x
|
||||
.const 1 c8 8
|
||||
|
||||
x4 subb x, ayuv, 128
|
||||
splitlw t1, t2, x
|
||||
splitwb y, a, t2
|
||||
splitwb v, u, t1
|
||||
convsbw wy, y
|
||||
convsbw wu, u
|
||||
convsbw wv, v
|
||||
|
||||
mullw t1, wy, 42
|
||||
shrsw t1, t1, c8
|
||||
addssw wy, wy, t1
|
||||
|
||||
addssw wr, wy, wv
|
||||
mullw t1, wv, 103
|
||||
shrsw t1, t1, c8
|
||||
subssw wr, wr, t1
|
||||
addssw wr, wr, wv
|
||||
|
||||
addssw wb, wy, wu
|
||||
addssw wb, wb, wu
|
||||
mullw t1, wu, 4
|
||||
shrsw t1, t1, c8
|
||||
addssw wb, wb, t1
|
||||
|
||||
mullw t1, wu, 100
|
||||
shrsw t1, t1, c8
|
||||
subssw wg, wy, t1
|
||||
mullw t1, wv, 104
|
||||
shrsw t1, t1, c8
|
||||
subssw wg, wg, t1
|
||||
subssw wg, wg, t1
|
||||
|
||||
convssswb r, wr
|
||||
convssswb g, wg
|
||||
convssswb b, wb
|
||||
|
||||
mergebw t1, a, r
|
||||
mergebw t2, g, b
|
||||
mergewl x, t1, t2
|
||||
x4 addb argb, x, 128
|
||||
|
||||
|
||||
|
||||
.function cogorc_convert_AYUV_BGRA
|
||||
.flags 2d
|
||||
.dest 4 argb
|
||||
.source 4 ayuv
|
||||
.temp 2 t1
|
||||
.temp 2 t2
|
||||
.temp 1 a
|
||||
.temp 1 y
|
||||
.temp 1 u
|
||||
.temp 1 v
|
||||
.temp 2 wy
|
||||
.temp 2 wu
|
||||
.temp 2 wv
|
||||
.temp 2 wr
|
||||
.temp 2 wg
|
||||
.temp 2 wb
|
||||
.temp 1 r
|
||||
.temp 1 g
|
||||
.temp 1 b
|
||||
.temp 4 x
|
||||
.const 1 c8 8
|
||||
|
||||
x4 subb x, ayuv, 128
|
||||
splitlw t1, t2, x
|
||||
splitwb y, a, t2
|
||||
splitwb v, u, t1
|
||||
convsbw wy, y
|
||||
convsbw wu, u
|
||||
convsbw wv, v
|
||||
|
||||
mullw t1, wy, 42
|
||||
shrsw t1, t1, c8
|
||||
addssw wy, wy, t1
|
||||
|
||||
addssw wr, wy, wv
|
||||
mullw t1, wv, 103
|
||||
shrsw t1, t1, c8
|
||||
subssw wr, wr, t1
|
||||
addssw wr, wr, wv
|
||||
|
||||
addssw wb, wy, wu
|
||||
addssw wb, wb, wu
|
||||
mullw t1, wu, 4
|
||||
shrsw t1, t1, c8
|
||||
addssw wb, wb, t1
|
||||
|
||||
mullw t1, wu, 100
|
||||
shrsw t1, t1, c8
|
||||
subssw wg, wy, t1
|
||||
mullw t1, wv, 104
|
||||
shrsw t1, t1, c8
|
||||
subssw wg, wg, t1
|
||||
subssw wg, wg, t1
|
||||
|
||||
convssswb r, wr
|
||||
convssswb g, wg
|
||||
convssswb b, wb
|
||||
|
||||
mergebw t1, b, g
|
||||
mergebw t2, r, a
|
||||
mergewl x, t1, t2
|
||||
x4 addb argb, x, 128
|
||||
|
||||
|
||||
|
||||
|
||||
.function cogorc_convert_AYUV_ABGR
|
||||
.flags 2d
|
||||
.dest 4 argb
|
||||
.source 4 ayuv
|
||||
.temp 2 t1
|
||||
.temp 2 t2
|
||||
.temp 1 a
|
||||
.temp 1 y
|
||||
.temp 1 u
|
||||
.temp 1 v
|
||||
.temp 2 wy
|
||||
.temp 2 wu
|
||||
.temp 2 wv
|
||||
.temp 2 wr
|
||||
.temp 2 wg
|
||||
.temp 2 wb
|
||||
.temp 1 r
|
||||
.temp 1 g
|
||||
.temp 1 b
|
||||
.temp 4 x
|
||||
.const 1 c8 8
|
||||
|
||||
x4 subb x, ayuv, 128
|
||||
splitlw t1, t2, x
|
||||
splitwb y, a, t2
|
||||
splitwb v, u, t1
|
||||
convsbw wy, y
|
||||
convsbw wu, u
|
||||
convsbw wv, v
|
||||
|
||||
mullw t1, wy, 42
|
||||
shrsw t1, t1, c8
|
||||
addssw wy, wy, t1
|
||||
|
||||
addssw wr, wy, wv
|
||||
mullw t1, wv, 103
|
||||
shrsw t1, t1, c8
|
||||
subssw wr, wr, t1
|
||||
addssw wr, wr, wv
|
||||
|
||||
addssw wb, wy, wu
|
||||
addssw wb, wb, wu
|
||||
mullw t1, wu, 4
|
||||
shrsw t1, t1, c8
|
||||
addssw wb, wb, t1
|
||||
|
||||
mullw t1, wu, 100
|
||||
shrsw t1, t1, c8
|
||||
subssw wg, wy, t1
|
||||
mullw t1, wv, 104
|
||||
shrsw t1, t1, c8
|
||||
subssw wg, wg, t1
|
||||
subssw wg, wg, t1
|
||||
|
||||
convssswb r, wr
|
||||
convssswb g, wg
|
||||
convssswb b, wb
|
||||
|
||||
mergebw t1, a, b
|
||||
mergebw t2, g, r
|
||||
mergewl x, t1, t2
|
||||
x4 addb argb, x, 128
|
||||
|
||||
|
||||
|
||||
.function cogorc_convert_AYUV_RGBA
|
||||
.flags 2d
|
||||
.dest 4 argb
|
||||
.source 4 ayuv
|
||||
.temp 2 t1
|
||||
.temp 2 t2
|
||||
.temp 1 a
|
||||
.temp 1 y
|
||||
.temp 1 u
|
||||
.temp 1 v
|
||||
.temp 2 wy
|
||||
.temp 2 wu
|
||||
.temp 2 wv
|
||||
.temp 2 wr
|
||||
.temp 2 wg
|
||||
.temp 2 wb
|
||||
.temp 1 r
|
||||
.temp 1 g
|
||||
.temp 1 b
|
||||
.temp 4 x
|
||||
.const 1 c8 8
|
||||
|
||||
x4 subb x, ayuv, 128
|
||||
splitlw t1, t2, x
|
||||
splitwb y, a, t2
|
||||
splitwb v, u, t1
|
||||
convsbw wy, y
|
||||
convsbw wu, u
|
||||
convsbw wv, v
|
||||
|
||||
mullw t1, wy, 42
|
||||
shrsw t1, t1, c8
|
||||
addssw wy, wy, t1
|
||||
|
||||
addssw wr, wy, wv
|
||||
mullw t1, wv, 103
|
||||
shrsw t1, t1, c8
|
||||
subssw wr, wr, t1
|
||||
addssw wr, wr, wv
|
||||
|
||||
addssw wb, wy, wu
|
||||
addssw wb, wb, wu
|
||||
mullw t1, wu, 4
|
||||
shrsw t1, t1, c8
|
||||
addssw wb, wb, t1
|
||||
|
||||
mullw t1, wu, 100
|
||||
shrsw t1, t1, c8
|
||||
subssw wg, wy, t1
|
||||
mullw t1, wv, 104
|
||||
shrsw t1, t1, c8
|
||||
subssw wg, wg, t1
|
||||
subssw wg, wg, t1
|
||||
|
||||
convssswb r, wr
|
||||
convssswb g, wg
|
||||
convssswb b, wb
|
||||
|
||||
mergebw t1, r, g
|
||||
mergebw t2, b, a
|
||||
mergewl x, t1, t2
|
||||
x4 addb argb, x, 128
|
||||
|
||||
|
||||
|
||||
.function cogorc_convert_I420_BGRA
|
||||
.dest 4 argb
|
||||
.source 1 y
|
||||
.source 1 u
|
||||
.source 1 v
|
||||
.temp 2 t1
|
||||
.temp 2 t2
|
||||
.temp 1 t3
|
||||
.temp 2 wy
|
||||
.temp 2 wu
|
||||
.temp 2 wv
|
||||
.temp 2 wr
|
||||
.temp 2 wg
|
||||
.temp 2 wb
|
||||
.temp 1 r
|
||||
.temp 1 g
|
||||
.temp 1 b
|
||||
.temp 4 x
|
||||
.const 1 c8 8
|
||||
.const 1 c128 128
|
||||
|
||||
subb t3, y, c128
|
||||
convsbw wy, t3
|
||||
loadupib t3, u
|
||||
subb t3, t3, c128
|
||||
convsbw wu, t3
|
||||
loadupib t3, v
|
||||
subb t3, t3, c128
|
||||
convsbw wv, t3
|
||||
|
||||
mullw t1, wy, 42
|
||||
shrsw t1, t1, c8
|
||||
addssw wy, wy, t1
|
||||
|
||||
addssw wr, wy, wv
|
||||
mullw t1, wv, 103
|
||||
shrsw t1, t1, c8
|
||||
subssw wr, wr, t1
|
||||
addssw wr, wr, wv
|
||||
|
||||
addssw wb, wy, wu
|
||||
addssw wb, wb, wu
|
||||
mullw t1, wu, 4
|
||||
shrsw t1, t1, c8
|
||||
addssw wb, wb, t1
|
||||
|
||||
mullw t1, wu, 100
|
||||
shrsw t1, t1, c8
|
||||
subssw wg, wy, t1
|
||||
mullw t1, wv, 104
|
||||
shrsw t1, t1, c8
|
||||
subssw wg, wg, t1
|
||||
subssw wg, wg, t1
|
||||
|
||||
convssswb r, wr
|
||||
convssswb g, wg
|
||||
convssswb b, wb
|
||||
|
||||
mergebw t1, b, g
|
||||
mergebw t2, r, 255
|
||||
mergewl x, t1, t2
|
||||
x4 addb argb, x, c128
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
.function cogorc_convert_I420_BGRA_avg
|
||||
.dest 4 argb
|
||||
.source 1 y
|
||||
.source 1 u1
|
||||
.source 1 u2
|
||||
.source 1 v1
|
||||
.source 1 v2
|
||||
.temp 2 t1
|
||||
.temp 2 t2
|
||||
.temp 1 t3
|
||||
.temp 1 t4
|
||||
.temp 2 wy
|
||||
.temp 2 wu
|
||||
.temp 2 wv
|
||||
.temp 2 wr
|
||||
.temp 2 wg
|
||||
.temp 2 wb
|
||||
.temp 1 r
|
||||
.temp 1 g
|
||||
.temp 1 b
|
||||
.temp 4 x
|
||||
.const 1 c8 8
|
||||
.const 1 c128 128
|
||||
|
||||
subb t3, y, c128
|
||||
convsbw wy, t3
|
||||
loadupib t3, u1
|
||||
loadupib t4, u2
|
||||
avgub t3, t3, t4
|
||||
subb t3, t3, c128
|
||||
convsbw wu, t3
|
||||
loadupib t3, v1
|
||||
loadupib t4, v2
|
||||
avgub t3, t3, t4
|
||||
subb t3, t3, c128
|
||||
convsbw wv, t3
|
||||
|
||||
mullw t1, wy, 42
|
||||
shrsw t1, t1, c8
|
||||
addssw wy, wy, t1
|
||||
|
||||
addssw wr, wy, wv
|
||||
mullw t1, wv, 103
|
||||
shrsw t1, t1, c8
|
||||
subssw wr, wr, t1
|
||||
addssw wr, wr, wv
|
||||
|
||||
addssw wb, wy, wu
|
||||
addssw wb, wb, wu
|
||||
mullw t1, wu, 4
|
||||
shrsw t1, t1, c8
|
||||
addssw wb, wb, t1
|
||||
|
||||
mullw t1, wu, 100
|
||||
shrsw t1, t1, c8
|
||||
subssw wg, wy, t1
|
||||
mullw t1, wv, 104
|
||||
shrsw t1, t1, c8
|
||||
subssw wg, wg, t1
|
||||
subssw wg, wg, t1
|
||||
|
||||
convssswb r, wr
|
||||
convssswb g, wg
|
||||
convssswb b, wb
|
||||
|
||||
mergebw t1, b, g
|
||||
mergebw t2, r, 255
|
||||
mergewl x, t1, t2
|
||||
x4 addb argb, x, c128
|
||||
|
||||
|
||||
|
||||
|
|
|
|||
Loading…
Reference in a new issue