The function is only instantiated once, so there is no point
in keeping it in a template file.
---
libavcodec/x86/dsputil_mmx.c | 2 +-
libavcodec/x86/dsputil_mmx.h | 2 ++
libavcodec/x86/fpel_mmx.c | 21 +++++++++++++++++++++
libavcodec/x86/hpeldsp_init.c | 1 +
libavcodec/x86/rnd_template.c | 22 ----------------------
5 files changed, 25 insertions(+), 23 deletions(-)
diff --git a/libavcodec/x86/dsputil_mmx.c b/libavcodec/x86/dsputil_mmx.c
index 0ef0564..5cc2452 100644
--- a/libavcodec/x86/dsputil_mmx.c
+++ b/libavcodec/x86/dsputil_mmx.c
@@ -1034,7 +1034,7 @@ void ff_put_cavs_qpel8_mc00_mmx(uint8_t *dst, uint8_t
*src, ptrdiff_t stride)
void ff_avg_cavs_qpel8_mc00_mmx(uint8_t *dst, uint8_t *src, ptrdiff_t stride)
{
- avg_pixels8_mmx(dst, src, stride, 8);
+ ff_avg_pixels8_mmx(dst, src, stride, 8);
}
void ff_put_cavs_qpel16_mc00_mmx(uint8_t *dst, uint8_t *src, ptrdiff_t stride)
diff --git a/libavcodec/x86/dsputil_mmx.h b/libavcodec/x86/dsputil_mmx.h
index df0c85a..eb23377 100644
--- a/libavcodec/x86/dsputil_mmx.h
+++ b/libavcodec/x86/dsputil_mmx.h
@@ -156,6 +156,8 @@ void ff_put_pixels_clamped_mmx(const int16_t *block,
uint8_t *pixels, int line_s
void ff_put_signed_pixels_clamped_mmx(const int16_t *block, uint8_t *pixels,
int line_size);
+void ff_avg_pixels8_mmx(uint8_t *block, const uint8_t *pixels,
+ ptrdiff_t line_size, int h);
void ff_put_pixels8_mmx(uint8_t *block, const uint8_t *pixels,
ptrdiff_t line_size, int h);
void ff_put_pixels16_mmx(uint8_t *block, const uint8_t *pixels,
diff --git a/libavcodec/x86/fpel_mmx.c b/libavcodec/x86/fpel_mmx.c
index 3470cf2..a2845fa 100644
--- a/libavcodec/x86/fpel_mmx.c
+++ b/libavcodec/x86/fpel_mmx.c
@@ -30,6 +30,27 @@
#if HAVE_INLINE_ASM
+// in case more speed is needed - unroling would certainly help
+void ff_avg_pixels8_mmx(uint8_t *block, const uint8_t *pixels,
+ ptrdiff_t line_size, int h)
+{
+ MOVQ_BFE(mm6);
+ JUMPALIGN();
+ do {
+ __asm__ volatile(
+ "movq %0, %%mm0 \n\t"
+ "movq %1, %%mm1 \n\t"
+ PAVGB_MMX(%%mm0, %%mm1, %%mm2, %%mm6)
+ "movq %%mm2, %0 \n\t"
+ :"+m"(*block)
+ :"m"(*pixels)
+ :"memory");
+ pixels += line_size;
+ block += line_size;
+ }
+ while (--h);
+}
+
void ff_put_pixels8_mmx(uint8_t *block, const uint8_t *pixels,
ptrdiff_t line_size, int h)
{
diff --git a/libavcodec/x86/hpeldsp_init.c b/libavcodec/x86/hpeldsp_init.c
index 36e93c4..fba0755 100644
--- a/libavcodec/x86/hpeldsp_init.c
+++ b/libavcodec/x86/hpeldsp_init.c
@@ -74,6 +74,7 @@ void ff_avg_pixels8_xy2_mmxext(uint8_t *block, const uint8_t
*pixels,
void ff_avg_pixels8_xy2_3dnow(uint8_t *block, const uint8_t *pixels,
ptrdiff_t line_size, int h);
+#define avg_pixels8_mmx ff_avg_pixels8_mmx
#define put_pixels8_mmx ff_put_pixels8_mmx
#define put_pixels16_mmx ff_put_pixels16_mmx
#define put_no_rnd_pixels8_mmx ff_put_pixels8_mmx
diff --git a/libavcodec/x86/rnd_template.c b/libavcodec/x86/rnd_template.c
index 2ff77d7..08e8593 100644
--- a/libavcodec/x86/rnd_template.c
+++ b/libavcodec/x86/rnd_template.c
@@ -92,28 +92,6 @@ static void DEF(put, pixels8_xy2)(uint8_t *block, const
uint8_t *pixels, ptrdiff
}
// avg_pixels
-#ifndef NO_RND
-// in case more speed is needed - unroling would certainly help
-static void DEF(avg, pixels8)(uint8_t *block, const uint8_t *pixels, ptrdiff_t
line_size, int h)
-{
- MOVQ_BFE(mm6);
- JUMPALIGN();
- do {
- __asm__ volatile(
- "movq %0, %%mm0 \n\t"
- "movq %1, %%mm1 \n\t"
- OP_AVG(%%mm0, %%mm1, %%mm2, %%mm6)
- "movq %%mm2, %0 \n\t"
- :"+m"(*block)
- :"m"(*pixels)
- :"memory");
- pixels += line_size;
- block += line_size;
- }
- while (--h);
-}
-#endif /* NO_RND */
-
static void DEF(avg, pixels16)(uint8_t *block, const uint8_t *pixels,
ptrdiff_t line_size, int h)
{
MOVQ_BFE(mm6);
--
1.7.9.5
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel
