From: "Ronald S. Bultje" <[email protected]>
---
libavcodec/dct-test.c | 2 +-
libavcodec/x86/dsputilenc_mmx.c | 80 +++++++++++++++++++++++----------------
libavcodec/x86/fdct_mmx.c | 4 ++
libavcodec/x86/motion_est_mmx.c | 6 +++
libavcodec/x86/mpegvideo_mmx.c | 6 +++
5 files changed, 64 insertions(+), 34 deletions(-)
diff --git a/libavcodec/dct-test.c b/libavcodec/dct-test.c
index 5046544..9e19e0c 100644
--- a/libavcodec/dct-test.c
+++ b/libavcodec/dct-test.c
@@ -85,7 +85,7 @@ static const struct algo fdct_tab[] = {
{ "IJG-AAN-INT", ff_fdct_ifast, SCALE_PERM },
{ "IJG-LLM-INT", ff_jpeg_fdct_islow_8, NO_PERM },
-#if HAVE_MMX
+#if HAVE_MMX && HAVE_INLINE_ASM
{ "MMX", ff_fdct_mmx, NO_PERM, AV_CPU_FLAG_MMX
},
{ "MMX2", ff_fdct_mmx2, NO_PERM, AV_CPU_FLAG_MMX2
},
{ "SSE2", ff_fdct_sse2, NO_PERM, AV_CPU_FLAG_SSE2
},
diff --git a/libavcodec/x86/dsputilenc_mmx.c b/libavcodec/x86/dsputilenc_mmx.c
index 47fa5ca..3cac979 100644
--- a/libavcodec/x86/dsputilenc_mmx.c
+++ b/libavcodec/x86/dsputilenc_mmx.c
@@ -30,6 +30,8 @@
#include "dsputil_mmx.h"
+#if HAVE_INLINE_ASM
+
static void get_pixels_mmx(DCTELEM *block, const uint8_t *pixels, int
line_size)
{
__asm__ volatile(
@@ -323,8 +325,6 @@ static int sse16_mmx(void *v, uint8_t * pix1, uint8_t *
pix2, int line_size, int
return tmp;
}
-int ff_sse16_sse2(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int
h);
-
static int hf_noise8_mmx(uint8_t * pix1, int line_size, int h) {
int tmp;
__asm__ volatile (
@@ -925,17 +925,6 @@ static void sub_hfyu_median_prediction_mmx2(uint8_t *dst,
const uint8_t *src1, c
"paddusw "#t", "#a" \n\t"\
"movd "#a", "#dst" \n\t"\
-#define hadamard_func(cpu) \
-int ff_hadamard8_diff_##cpu (void *s, uint8_t *src1, uint8_t *src2, \
- int stride, int h); \
-int ff_hadamard8_diff16_##cpu(void *s, uint8_t *src1, uint8_t *src2, \
- int stride, int h);
-
-hadamard_func(mmx)
-hadamard_func(mmx2)
-hadamard_func(sse2)
-hadamard_func(ssse3)
-
#define DCT_SAD4(m,mm,o)\
"mov"#m" "#o"+ 0(%1), "#mm"2 \n\t"\
"mov"#m" "#o"+16(%1), "#mm"3 \n\t"\
@@ -1094,10 +1083,26 @@ static int ssd_int8_vs_int16_mmx(const int8_t *pix1,
const int16_t *pix2, int si
#undef PHADDD
#endif //HAVE_SSSE3
+#endif /* HAVE_INLINE_ASM */
+
+int ff_sse16_sse2(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int
h);
+
+#define hadamard_func(cpu) \
+int ff_hadamard8_diff_##cpu (void *s, uint8_t *src1, uint8_t *src2, \
+ int stride, int h); \
+int ff_hadamard8_diff16_##cpu(void *s, uint8_t *src1, uint8_t *src2, \
+ int stride, int h);
+
+hadamard_func(mmx)
+hadamard_func(mmx2)
+hadamard_func(sse2)
+hadamard_func(ssse3)
void ff_dsputilenc_init_mmx(DSPContext* c, AVCodecContext *avctx)
{
int mm_flags = av_get_cpu_flags();
+
+#if HAVE_INLINE_ASM
int bit_depth = avctx->bits_per_raw_sample;
if (mm_flags & AV_CPU_FLAG_MMX) {
@@ -1121,11 +1126,6 @@ void ff_dsputilenc_init_mmx(DSPContext* c,
AVCodecContext *avctx)
c->diff_bytes= diff_bytes_mmx;
c->sum_abs_dctelem= sum_abs_dctelem_mmx;
-#if HAVE_YASM
- c->hadamard8_diff[0]= ff_hadamard8_diff16_mmx;
- c->hadamard8_diff[1]= ff_hadamard8_diff_mmx;
-#endif
-
c->pix_norm1 = pix_norm1_mmx;
c->sse[0] = sse16_mmx;
c->sse[1] = sse8_mmx;
@@ -1146,10 +1146,6 @@ void ff_dsputilenc_init_mmx(DSPContext* c,
AVCodecContext *avctx)
if (mm_flags & AV_CPU_FLAG_MMX2) {
-#if HAVE_YASM
- c->hadamard8_diff[0]= ff_hadamard8_diff16_mmx2;
- c->hadamard8_diff[1]= ff_hadamard8_diff_mmx2;
-#endif
c->sum_abs_dctelem= sum_abs_dctelem_mmx2;
c->vsad[4]= vsad_intra16_mmx2;
@@ -1164,13 +1160,6 @@ void ff_dsputilenc_init_mmx(DSPContext* c,
AVCodecContext *avctx)
if (bit_depth <= 8)
c->get_pixels = get_pixels_sse2;
c->sum_abs_dctelem= sum_abs_dctelem_sse2;
-#if HAVE_YASM
- c->sse[0] = ff_sse16_sse2;
-#if HAVE_ALIGNED_STACK
- c->hadamard8_diff[0]= ff_hadamard8_diff16_sse2;
- c->hadamard8_diff[1]= ff_hadamard8_diff_sse2;
-#endif
-#endif
}
#if HAVE_SSSE3
@@ -1180,10 +1169,6 @@ void ff_dsputilenc_init_mmx(DSPContext* c,
AVCodecContext *avctx)
}
c->add_8x8basis= add_8x8basis_ssse3;
c->sum_abs_dctelem= sum_abs_dctelem_ssse3;
-#if HAVE_YASM && HAVE_ALIGNED_STACK
- c->hadamard8_diff[0]= ff_hadamard8_diff16_ssse3;
- c->hadamard8_diff[1]= ff_hadamard8_diff_ssse3;
-#endif
}
#endif
@@ -1194,6 +1179,35 @@ void ff_dsputilenc_init_mmx(DSPContext* c,
AVCodecContext *avctx)
c->add_8x8basis= add_8x8basis_3dnow;
}
}
+#endif /* HAVE_INLINE_ASM */
+
+#if HAVE_YASM
+ if (mm_flags & AV_CPU_FLAG_MMX) {
+ c->hadamard8_diff[0] = ff_hadamard8_diff16_mmx;
+ c->hadamard8_diff[1] = ff_hadamard8_diff_mmx;
+
+ if (mm_flags & AV_CPU_FLAG_MMX2) {
+ c->hadamard8_diff[0] = ff_hadamard8_diff16_mmx2;
+ c->hadamard8_diff[1] = ff_hadamard8_diff_mmx2;
+ }
+
+ if (mm_flags & AV_CPU_FLAG_SSE2){
+ c->sse[0] = ff_sse16_sse2;
+
+#if HAVE_ALIGNED_STACK
+ c->hadamard8_diff[0] = ff_hadamard8_diff16_sse2;
+ c->hadamard8_diff[1] = ff_hadamard8_diff_sse2;
+#endif
+ }
+
+#if HAVE_SSSE3 && HAVE_ALIGNED_STACK
+ if (mm_flags & AV_CPU_FLAG_SSSE3) {
+ c->hadamard8_diff[0] = ff_hadamard8_diff16_ssse3;
+ c->hadamard8_diff[1] = ff_hadamard8_diff_ssse3;
+ }
+#endif
+ }
+#endif /* HAVE_YASM */
ff_dsputil_init_pix_mmx(c, avctx);
}
diff --git a/libavcodec/x86/fdct_mmx.c b/libavcodec/x86/fdct_mmx.c
index cc3036b..3614fd1 100644
--- a/libavcodec/x86/fdct_mmx.c
+++ b/libavcodec/x86/fdct_mmx.c
@@ -34,6 +34,8 @@
#include "libavutil/x86_cpu.h"
#include "libavcodec/dsputil.h"
+#if HAVE_INLINE_ASM
+
//////////////////////////////////////////////////////////////////////
//
// constants for the forward DCT
@@ -579,3 +581,5 @@ void ff_fdct_sse2(int16_t *block)
fdct_col_sse2(block, block1, 0);
fdct_row_sse2(block1, block);
}
+
+#endif /* HAVE_INLINE_ASM */
diff --git a/libavcodec/x86/motion_est_mmx.c b/libavcodec/x86/motion_est_mmx.c
index a522a5e..5aed655 100644
--- a/libavcodec/x86/motion_est_mmx.c
+++ b/libavcodec/x86/motion_est_mmx.c
@@ -26,6 +26,8 @@
#include "libavcodec/dsputil.h"
#include "dsputil_mmx.h"
+#if HAVE_INLINE_ASM
+
DECLARE_ASM_CONST(8, uint64_t, round_tab)[3]={
0x0000000000000000ULL,
0x0001000100010001ULL,
@@ -422,8 +424,11 @@ static int sad16_xy2_ ## suf(void *v, uint8_t *blk2,
uint8_t *blk1, int stride,
PIX_SAD(mmx)
PIX_SAD(mmx2)
+#endif /* HAVE_INLINE_ASM */
+
void ff_dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx)
{
+#if HAVE_INLINE_ASM
int mm_flags = av_get_cpu_flags();
if (mm_flags & AV_CPU_FLAG_MMX) {
@@ -458,4 +463,5 @@ void ff_dsputil_init_pix_mmx(DSPContext* c, AVCodecContext
*avctx)
if ((mm_flags & AV_CPU_FLAG_SSE2) && !(mm_flags & AV_CPU_FLAG_3DNOW) &&
avctx->codec_id != CODEC_ID_SNOW) {
c->sad[0]= sad16_sse2;
}
+#endif /* HAVE_INLINE_ASM */
}
diff --git a/libavcodec/x86/mpegvideo_mmx.c b/libavcodec/x86/mpegvideo_mmx.c
index dcce486..a242c19 100644
--- a/libavcodec/x86/mpegvideo_mmx.c
+++ b/libavcodec/x86/mpegvideo_mmx.c
@@ -29,6 +29,8 @@
#include "libavcodec/mpegvideo.h"
#include "dsputil_mmx.h"
+#if HAVE_INLINE_ASM
+
extern uint16_t ff_inv_zigzag_direct16[64];
@@ -626,8 +628,11 @@ static void denoise_dct_sse2(MpegEncContext *s, DCTELEM
*block){
#include "mpegvideo_mmx_template.c"
#endif
+#endif /* HAVE_INLINE_ASM */
+
void ff_MPV_common_init_mmx(MpegEncContext *s)
{
+#if HAVE_INLINE_ASM
int mm_flags = av_get_cpu_flags();
if (mm_flags & AV_CPU_FLAG_MMX) {
@@ -662,4 +667,5 @@ void ff_MPV_common_init_mmx(MpegEncContext *s)
}
}
}
+#endif /* HAVE_INLINE_ASM */
}
--
1.7.9.5
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel
