Chiebot-Mirror
/
FFmpeg
8
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Merge commit 'a36947c167d7278b891453083b57dc56b7a7f5c5'

Browse Source 
* commit 'a36947c167d7278b891453083b57dc56b7a7f5c5':
  dsputilenc_mmx: K&R formatting cosmetics

Conflicts:
	libavcodec/x86/dsputilenc_mmx.c

Merged-by: Michael Niedermayer <michaelni@gmx.at>
pull/293/head
Michael Niedermayer 11 years ago
parent d926c4b240 a36947c167
commit b7a5f5dc66
1 changed files with 770 additions and 727 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 117
      libavcodec/x86/dsputilenc_mmx.c

117
libavcodec/x86/dsputilenc_mmx.c
Unescape View File

@ -34,14 +34,17 @@
void ff_get_pixels_mmx(int16_t *block, const uint8_t *pixels, int line_size);
void ff_get_pixels_sse2(int16_t *block, const uint8_t *pixels, int line_size);
void ff_diff_pixels_mmx(int16_t *block, const uint8_t *s1, const uint8_t *s2, int stride);
void ff_diff_pixels_mmx(int16_t *block, const uint8_t *s1, const uint8_t *s2,
int stride);
int ff_pix_sum16_mmx(uint8_t *pix, int line_size);
int ff_pix_norm1_mmx(uint8_t *pix, int line_size);
#if HAVE_INLINE_ASM
static int sse8_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
static int sse8_mmx(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h)
{
int tmp;
__asm__ volatile (
"movl %4, %%ecx \n"
"shr $1, %%ecx \n"
@ -98,11 +101,15 @@ static int sse8_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int
: "+r" (pix1), "+r" (pix2), "=r" (tmp)
: "r" ((x86_reg) line_size), "m" (h)
: "%ecx");
return tmp;
}
static int sse16_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
static int sse16_mmx(void *v, uint8_t *pix1, uint8_t *pix2,
int line_size, int h)
{
int tmp;
__asm__ volatile (
"movl %4, %%ecx\n"
"pxor %%mm0, %%mm0\n" /* mm0 = 0 */
@ -158,11 +165,14 @@ static int sse16_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int
: "+r" (pix1), "+r" (pix2), "=r" (tmp)
: "r" ((x86_reg) line_size), "m" (h)
: "%ecx");
return tmp;
}
static int hf_noise8_mmx(uint8_t * pix1, int line_size, int h) {
static int hf_noise8_mmx(uint8_t *pix1, int line_size, int h)
{
int tmp;
__asm__ volatile (
"movl %3, %%ecx\n"
"pxor %%mm7, %%mm7\n"
@ -283,12 +293,15 @@ static int hf_noise8_mmx(uint8_t * pix1, int line_size, int h) {
: "+r" (pix1), "=r" (tmp)
: "r" ((x86_reg) line_size), "g" (h - 2)
: "%ecx");
return tmp;
}
static int hf_noise16_mmx(uint8_t * pix1, int line_size, int h) {
static int hf_noise16_mmx(uint8_t *pix1, int line_size, int h)
{
int tmp;
uint8_t *pix = pix1;
__asm__ volatile (
"movl %3, %%ecx\n"
"pxor %%mm7, %%mm7\n"
@ -397,31 +410,46 @@ static int hf_noise16_mmx(uint8_t * pix1, int line_size, int h) {
: "+r" (pix1), "=r" (tmp)
: "r" ((x86_reg) line_size), "g" (h - 2)
: "%ecx");
return tmp + hf_noise8_mmx(pix + 8, line_size, h);
}
static int nsse16_mmx(void *p, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
static int nsse16_mmx(void *p, uint8_t *pix1, uint8_t *pix2,
int line_size, int h)
{
MpegEncContext *c = p;
int score1, score2;
if(c) score1 = c->dsp.sse[0](c, pix1, pix2, line_size, h);
else score1 = sse16_mmx(c, pix1, pix2, line_size, h);
score2= hf_noise16_mmx(pix1, line_size, h) - hf_noise16_mmx(pix2, line_size, h);
if(c) return score1 + FFABS(score2)*c->avctx->nsse_weight;
else return score1 + FFABS(score2)*8;
if (c)
score1 = c->dsp.sse[0](c, pix1, pix2, line_size, h);
else
score1 = sse16_mmx(c, pix1, pix2, line_size, h);
score2 = hf_noise16_mmx(pix1, line_size, h) -
hf_noise16_mmx(pix2, line_size, h);
if (c)
return score1 + FFABS(score2) * c->avctx->nsse_weight;
else
return score1 + FFABS(score2) * 8;
}
static int nsse8_mmx(void *p, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
static int nsse8_mmx(void *p, uint8_t *pix1, uint8_t *pix2,
int line_size, int h)
{
MpegEncContext *c = p;
int score1 = sse8_mmx(c, pix1, pix2, line_size, h);
int score2= hf_noise8_mmx(pix1, line_size, h) - hf_noise8_mmx(pix2, line_size, h);
int score2 = hf_noise8_mmx(pix1, line_size, h) -
hf_noise8_mmx(pix2, line_size, h);
if(c) return score1 + FFABS(score2)*c->avctx->nsse_weight;
else return score1 + FFABS(score2)*8;
if (c)
return score1 + FFABS(score2) * c->avctx->nsse_weight;
else
return score1 + FFABS(score2) * 8;
}
static int vsad_intra16_mmx(void *v, uint8_t * pix, uint8_t * dummy, int line_size, int h) {
static int vsad_intra16_mmx(void *v, uint8_t *pix, uint8_t *dummy,
int line_size, int h)
{
int tmp;
av_assert2((((int) pix) & 7) == 0);
@ -478,6 +506,7 @@ static int vsad_intra16_mmx(void *v, uint8_t * pix, uint8_t * dummy, int line_si
: "+r" (pix), "=r" (tmp)
: "r" ((x86_reg) line_size), "m" (h)
: "%ecx");
return tmp & 0xFFFF;
}
#undef SUM
@ -520,11 +549,14 @@ static int vsad_intra16_mmxext(void *v, uint8_t *pix, uint8_t *dummy,
: "+r" (pix), "=r" (tmp)
: "r" ((x86_reg) line_size), "m" (h)
: "%ecx");
return tmp;
}
#undef SUM
static int vsad16_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
static int vsad16_mmx(void *v, uint8_t *pix1, uint8_t *pix2,
int line_size, int h)
{
int tmp;
av_assert2((((int) pix1) & 7) == 0);
@ -598,6 +630,7 @@ static int vsad16_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, in
: "+r" (pix1), "+r" (pix2), "=r" (tmp)
: "r" ((x86_reg) line_size), "m" (h)
: "%ecx");
return tmp & 0x7FFF;
}
#undef SUM
@ -657,12 +690,15 @@ static int vsad16_mmxext(void *v, uint8_t *pix1, uint8_t *pix2,
: "+r" (pix1), "+r" (pix2), "=r" (tmp)
: "r" ((x86_reg) line_size), "m" (h)
: "%ecx");
return tmp;
}
#undef SUM
static void diff_bytes_mmx(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int w){
static void diff_bytes_mmx(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int w)
{
x86_reg i = 0;
if (w >= 16)
__asm__ volatile (
"1: \n\t"
@ -678,8 +714,8 @@ static void diff_bytes_mmx(uint8_t *dst, const uint8_t *src1, const uint8_t *src
"cmp %4, %0 \n\t"
" jb 1b \n\t"
: "+r" (i)
: "r"(src1), "r"(src2), "r"(dst), "r"((x86_reg)w-15)
);
: "r" (src1), "r" (src2), "r" (dst), "r" ((x86_reg) w - 15));
for (; i < w; i++)
dst[i + 0] = src1[i + 0] - src2[i + 0];
}
@ -713,8 +749,7 @@ static void sub_hfyu_median_prediction_mmxext(uint8_t *dst, const uint8_t *src1,
"cmp %4, %0 \n\t"
" jb 1b \n\t"
: "+r" (i)
: "r"(src1), "r"(src2), "r"(dst), "r"((x86_reg)w)
);
: "r" (src1), "r" (src2), "r" (dst), "r" ((x86_reg) w));
l = *left;
lt = *left_top;
@ -743,9 +778,10 @@ static void sub_hfyu_median_prediction_mmxext(uint8_t *dst, const uint8_t *src1,
MMABS(a,z) \
"paddusw " #a ", " #sum " \n\t"
/* FIXME: HSUM_* saturates at 64k, while an 8x8 hadamard or dct block can get up to
* about 100k on extreme inputs. But that's very unlikely to occur in natural video,
* and it's even more unlikely to not have any alternative mvs/modes with lower cost. */
/* FIXME: HSUM_* saturates at 64k, while an 8x8 hadamard or dct block can get
* up to about 100k on extreme inputs. But that's very unlikely to occur in
* natural video, and it's even more unlikely to not have any alternative
* mvs/modes with lower cost. */
#define HSUM_MMX(a, t, dst) \
"movq " #a ", " #t " \n\t" \
"psrlq $32, " #a " \n\t" \
@ -800,13 +836,13 @@ static void sub_hfyu_median_prediction_mmxext(uint8_t *dst, const uint8_t *src1,
HSUM(%%xmm0, %%xmm1, %0)
#define DCT_SAD_FUNC(cpu) \
static int sum_abs_dctelem_##cpu(int16_t *block){\
static int sum_abs_dctelem_ ## cpu(int16_t *block) \
{ \
int sum; \
__asm__ volatile ( \
DCT_SAD \
:"=r"(sum) \
:"r"(block)\
);\
:"r"(block)); \
return sum & 0xFFFF; \
}
@ -836,9 +872,12 @@ DCT_SAD_FUNC(ssse3)
#undef HSUM
#undef DCT_SAD
static int ssd_int8_vs_int16_mmx(const int8_t *pix1, const int16_t *pix2, int size){
static int ssd_int8_vs_int16_mmx(const int8_t *pix1, const int16_t *pix2,
int size)
{
int sum;
x86_reg i = size;
__asm__ volatile (
"pxor %%mm4, %%mm4 \n"
"1: \n"
@ -862,8 +901,8 @@ static int ssd_int8_vs_int16_mmx(const int8_t *pix1, const int16_t *pix2, int si
"paddd %%mm3, %%mm4 \n"
"movd %%mm4, %1 \n"
: "+r" (i), "=r" (sum)
:"r"(pix1), "r"(pix2)
);
: "r" (pix1), "r" (pix2));
return sum;
}
@ -871,10 +910,11 @@ static int ssd_int8_vs_int16_mmx(const int8_t *pix1, const int16_t *pix2, int si
"movq " #a ", " #t " \n\t" \
"psrlq $32, " #a " \n\t" \
"paddd " #t ", " #a " \n\t"
/*
pmulhw: dst[0-15]=(src[0-15]*dst[0-15])[16-31]
pmulhrw: dst[0-15]=(src[0-15]*dst[0-15] + 0x8000)[16-31]
pmulhrsw: dst[0-15]=(src[0-15]*dst[0-15] + 0x4000)[15-30]
* pmulhw: dst[0 - 15] = (src[0 - 15] * dst[0 - 15])[16 - 31]
* pmulhrw: dst[0 - 15] = (src[0 - 15] * dst[0 - 15] + 0x8000)[16 - 31]
* pmulhrsw: dst[0 - 15] = (src[0 - 15] * dst[0 - 15] + 0x4000)[15 - 30]
*/
#define PMULHRW(x, y, s, o) \
"pmulhw " #s ", " #x " \n\t" \
@ -913,9 +953,12 @@ static int ssd_int8_vs_int16_mmx(const int8_t *pix1, const int16_t *pix2, int si
#define DEF(x) x ## _ssse3
#define SET_RND(x)
#define SCALE_OFFSET -1
#define PHADDD(a, t) \
"pshufw $0x0E, " #a ", " #t " \n\t" \
"paddd "#t", "#a" \n\t" /* faster than phaddd on core2 */
/* faster than phaddd on core2 */ \
"paddd " #t ", " #a " \n\t"
#define PMULHRW(x, y, s, o) \
"pmulhrsw " #s ", " #x " \n\t" \
"pmulhrsw " #s ", " #y " \n\t"
@ -957,9 +1000,9 @@ av_cold void ff_dsputilenc_init_mmx(DSPContext *c, AVCodecContext *avctx)
c->get_pixels = ff_get_pixels_mmx;
c->diff_pixels = ff_diff_pixels_mmx;
c->pix_sum = ff_pix_sum16_mmx;
c->pix_norm1 = ff_pix_norm1_mmx;
}
if (EXTERNAL_SSE2(cpu_flags))
if (bit_depth <= 8)
c->get_pixels = ff_get_pixels_sse2;

Write Preview
Loading…
Cancel
Save