Skip to content

GitLab

Commit e84136ba authored by Sebastian Dröge's avatar Sebastian Dröge
Browse files

videomixer: Port most blending related functions to orc 

Only remaining MMX implementation is the ARGB/BGRA/AYUV blending
for which we first need the orc compositing opcodes.
parent e87802ef
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 629 additions and 456 deletions
gst/videomixer/Makefile.am
View file @ e84136ba
plugin_LTLIBRARIES = libgstvideomixer.la
ORC_SOURCE=blendorc
include $(top_srcdir)/common/orc.mak
libgstvideomixer_la_SOURCES = videomixer.c blend.c
nodist_libgstvideomixer_la_SOURCES = $(ORC_NODIST_SOURCES)
libgstvideomixer_la_CFLAGS = $(GST_PLUGINS_BASE_CFLAGS) \
$(GST_BASE_CFLAGS) $(GST_CONTROLLER_CFLAGS) $(GST_CFLAGS) $(ORC_CFLAGS)
libgstvideomixer_la_LIBADD = $(GST_PLUGINS_BASE_LIBS) -lgstvideo-@GST_MAJORMINOR@ \
......
gst/videomixer/blend.c
View file @ e84136ba
This diff is collapsed.
gst/videomixer/blend_mmx.h
View file @ e84136ba
......@@ -122,229 +122,3 @@ NAME_BLEND (guint8 * dest, const guint8 * src, gint src_height, gint src_width,
}
#endif
#ifdef GENERIC
static inline void
_memcpy_u8_mmx (guint8 * dest, const guint8 * src, guint count)
{
/* *INDENT-OFF* */
__asm__ __volatile__ (
"1: \n\t"
"test $7, %0 \n\t"
"je 3f \n\t"
"2: \n\t"
"movb (%2), %%al \n\t"
"movb %%al, (%1) \n\t"
"inc %2 \n\t"
"inc %1 \n\t"
"dec %0 \n\t"
"test $7, %0 \n\t"
"jne 2b \n\t"
"3: \n\t"
"sar $3, %0 \n\t"
"cmp $0, %0 \n\t"
"je 5f \n\t"
"4: \n\t"
"movq (%2), %%mm0 \n\t"
"movq %%mm0, (%1) \n\t"
"add $8, %2 \n\t"
"add $8, %1 \n\t"
"dec %0 \n\t"
"jne 4b \n\t"
"5: \n\t"
"emms \n\t"
: "=r" (count), "=q" (dest), "=q" (src)
: "0" (count), "1" (dest), "2" (src)
: "memory", "al",
"st", "st(1)", "st(2)", "st(3)", "st(4)", "st(5)", "st(6)", "st(7)"
#ifdef __MMX__
, "mm0"
#endif
);
/* *INDENT-ON* */
}
static inline void
_memset_u8_mmx (guint8 * dest, guint val, guint count)
{
guint8 val8 = val;
guint64 val64;
val64 = (val << 24) | (val << 16) | (val << 8) | (val);
val64 = (val64 << 32) | val64;
/* *INDENT-OFF* */
__asm__ __volatile__ (
"1: \n\t"
"test $7, %0 \n\t"
"je 3f \n\t"
"2: \n\t"
"movb %4, (%1) \n\t"
"inc %1 \n\t"
"dec %0 \n\t"
"test $7, %0 \n\t"
"jne 2b \n\t"
"3: \n\t"
"sar $3, %0 \n\t"
"cmp $0, %0 \n\t"
"je 5f \n\t"
"movq %5, %%mm0 \n\t"
"4: \n\t"
"movq %%mm0, (%1) \n\t"
"add $8, %1 \n\t"
"dec %0 \n\t"
"jne 4b \n\t"
"5: \n\t"
"emms \n\t"
: "=r" (count), "=q" (dest)
: "0" (count), "1" (dest), "q" (val8), "m" (val64)
: "memory",
"st", "st(1)", "st(2)", "st(3)", "st(4)", "st(5)", "st(6)", "st(7)"
#ifdef __MMX__
, "mm0"
#endif
);
/* *INDENT-ON* */
}
static inline void
_memset_u32_mmx (guint32 * dest, guint32 val, guint count)
{
guint64 val64 = val;
val64 |= (val64 << 32);
/* *INDENT-OFF* */
__asm__ __volatile__ (
"1: \n\t"
"test $1, %0 \n\t"
"je 3f \n\t"
"2: \n\t"
"movl %4, (%1) \n\t"
"add $4, %1 \n\t"
"dec %0 \n\t"
"test $1, %0 \n\t"
"jne 2b \n\t"
"3: \n\t"
"sar $1, %0 \n\t"
"cmp $0, %0 \n\t"
"je 5f \n\t"
"movq %5, %%mm0 \n\t"
"4: \n\t"
"movq %%mm0, (%1) \n\t"
"add $8, %1 \n\t"
"dec %0 \n\t"
"jne 4b \n\t"
"5: \n\t"
"emms \n\t"
: "=r" (count), "=r" (dest)
: "0" (count), "1" (dest), "r" (val), "m" (val64)
: "memory",
"st", "st(1)", "st(2)", "st(3)", "st(4)", "st(5)", "st(6)", "st(7)"
#ifdef __MMX__
, "mm0"
#endif
);
/* *INDENT-ON* */
}
static inline void
_blend_u8_mmx (guint8 * dest, const guint8 * src,
gint src_stride, gint dest_stride, gint src_width, gint src_height,
gint dest_width, gint s_alpha)
{
gint i;
gint src_add = src_stride - src_width;
gint dest_add = dest_stride - src_width;
for (i = 0; i < src_height; i++) {
/* Do first 3 "odd" pixels */
while ((src_width & 0x03)) {
*dest = BLEND (*dest, *src, s_alpha);
dest++;
src++;
src_width--;
}
/* (P1 * (256 - A) + (P2 * A)) / 256
* => (P1 * 256 - P1 * A + P2 * A) / 256
* => (P1 * 256 + A * (P2 - P1) / 256
* => P1 + (A * (P2 - P1)) / 256
*/
/* *INDENT-OFF* */
__asm__ __volatile__ (
" mov %4 , %%eax \n\t" /* eax = s_alpha */
" movd %%eax , %%mm6 \n\t" /* mm6 = s_alpha */
" punpcklwd %%mm6 , %%mm6 \n\t" /* mm6 = 00 00 00 00 00 aa 00 aa, alpha scale factor */
" punpckldq %%mm6 , %%mm6 \n\t" /* mm6 = 00 aa 00 aa 00 aa 00 aa */
" pxor %%mm7 , %%mm7 \n\t" /* mm7 = 00 00 00 00 00 00 00 00 */
" movl %5 , %%ecx \n\t" /* ecx = src_width */
"1: \n\t"
" test $7 , %%ecx \n\t"
" je 2f \n\t"
/* do first 4 "odd" bytes */
" movd (%2) , %%mm2 \n\t" /* mm2 = src, 00 00 00 00 sv su sy sa */
" movd (%3) , %%mm1 \n\t" /* mm1 = dest, 00 00 00 00 dv du dy da */
" punpcklbw %%mm7 , %%mm2 \n\t"
" punpcklbw %%mm7 , %%mm1 \n\t"
" psubw %%mm1 , %%mm2 \n\t" /* mm2 = mm2 - mm1 */
" pmullw %%mm6 , %%mm2 \n\t" /* mm2 = a * mm2 */
" psllw $8 , %%mm1 \n\t" /* scale up */
" paddw %%mm1 , %%mm2 \n\t" /* mm2 = mm2 + mm1 */
" psrlw $8 , %%mm2 \n\t" /* scale down */
" packuswb %%mm2 , %%mm2 \n\t"
" movd %%mm2 , (%3) \n\t" /* dest = mm1 */
" add $4 , %1 \n\t"
" add $4 , %0 \n\t"
"2: \n\t"
" sar $3 , %%ecx \n\t" /* prepare for 8 bytes per loop */
" cmp $0 , %%ecx \n\t"
" je 4f \n\t"
"3: \n\t"
/* do even pixels */
" movq (%2) , %%mm2 \n\t" /* mm2 = src, sv1 su1 sy1 sa1 sv0 su0 sy0 sa0 */
" movq (%3) , %%mm1 \n\t" /* mm1 = dest, dv1 du1 dy1 da1 dv0 du0 dy0 da0 */
" movq %%mm2 , %%mm4 \n\t"
" movq %%mm1 , %%mm3 \n\t"
" punpcklbw %%mm7 , %%mm2 \n\t"
" punpckhbw %%mm7 , %%mm4 \n\t"
" punpcklbw %%mm7 , %%mm1 \n\t"
" punpckhbw %%mm7 , %%mm3 \n\t"
" psubw %%mm1 , %%mm2 \n\t" /* mm2 = mm2 - mm1 */
" psubw %%mm3 , %%mm4 \n\t" /* mm4 = mm4 - mm3 */
" pmullw %%mm6 , %%mm2 \n\t" /* mm2 = a * mm2 */
" pmullw %%mm6 , %%mm4 \n\t" /* mm2 = a * mm2 */
" psllw $8 , %%mm1 \n\t" /* scale up */
" psllw $8 , %%mm3 \n\t" /* scale up */
" paddw %%mm1 , %%mm2 \n\t" /* mm2 = mm2 + mm1 */
" paddw %%mm3 , %%mm4 \n\t" /* mm4 = mm4 + mm3 */
" psrlw $8 , %%mm2 \n\t" /* scale down */
" psrlw $8 , %%mm4 \n\t" /* scale down */
" packuswb %%mm4 , %%mm2 \n\t"
" movq %%mm2 , (%3) \n\t"
" add $8 , %0 \n\t"
" add $8 , %1 \n\t"
" dec %%ecx \n\t"
" jne 3b \n\t"
"4: \n\t"
:"=r" (src), "=r" (dest)
:"0" (src), "1" (dest), "m" (s_alpha), "m" (src_width)
:"%eax", "%ecx", "memory",
"st", "st(1)", "st(2)", "st(3)", "st(4)", "st(5)", "st(6)", "st(7)"
#ifdef __MMX__
, "mm1", "mm2", "mm3", "mm4", "mm6", "mm7"
#endif
);
/* *INDENT-ON* */
src += src_add;
dest += dest_add;
}
__asm__ __volatile__ ("emms");
}
#endif
gst/videomixer/blendorc-dist.c 0 → 100644
View file @ e84136ba
/* autogenerated from blendorc.orc */
#ifdef HAVE_CONFIG_H
#include "config.h"
#endif
#ifndef DISABLE_ORC
#include <orc/orc.h>
#endif
#include <glib.h>
#ifndef _ORC_INTEGER_TYPEDEFS_
#define _ORC_INTEGER_TYPEDEFS_
#if defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L
#include <stdint.h>
typedef int8_t orc_int8;
typedef int16_t orc_int16;
typedef int32_t orc_int32;
typedef int64_t orc_int64;
typedef uint8_t orc_uint8;
typedef uint16_t orc_uint16;
typedef uint32_t orc_uint32;
typedef uint64_t orc_uint64;
#elif defined(_MSC_VER)
typedef signed __int8 orc_int8;
typedef signed __int16 orc_int16;
typedef signed __int32 orc_int32;
typedef signed __int64 orc_int64;
typedef unsigned __int8 orc_uint8;
typedef unsigned __int16 orc_uint16;
typedef unsigned __int32 orc_uint32;
typedef unsigned __int64 orc_uint64;
#else
#include <limits.h>
typedef signed char orc_int8;
typedef short orc_int16;
typedef int orc_int32;
typedef unsigned char orc_uint8;
typedef unsigned short orc_uint16;
typedef unsigned int orc_uint32;
#if INT_MAX == LONG_MAX
typedef long long orc_int64;
typedef unsigned long long orc_uint64;
#else
typedef long orc_int64;
typedef unsigned long orc_uint64;
#endif
#endif
typedef union
{
orc_int32 i;
float f;
} orc_union32;
typedef union
{
orc_int64 i;
double f;
} orc_union64;
#endif
void orc_splat_u32 (guint32 * d1, int p1, int n);
void orc_memcpy_u32 (guint32 * d1, const guint32 * s1, int n);
void orc_blend_u8 (guint8 * d1, int d1_stride, const guint8 * s1, int s1_stride,
int p1, int n, int m);
/* begin Orc C target preamble */
#define ORC_CLAMP(x,a,b) ((x)<(a) ? (a) : ((x)>(b) ? (b) : (x)))
#define ORC_ABS(a) ((a)<0 ? -(a) : (a))
#define ORC_MIN(a,b) ((a)<(b) ? (a) : (b))
#define ORC_MAX(a,b) ((a)>(b) ? (a) : (b))
#define ORC_SB_MAX 127
#define ORC_SB_MIN (-1-ORC_SB_MAX)
#define ORC_UB_MAX 255
#define ORC_UB_MIN 0
#define ORC_SW_MAX 32767
#define ORC_SW_MIN (-1-ORC_SW_MAX)
#define ORC_UW_MAX 65535
#define ORC_UW_MIN 0
#define ORC_SL_MAX 2147483647
#define ORC_SL_MIN (-1-ORC_SL_MAX)
#define ORC_UL_MAX 4294967295U
#define ORC_UL_MIN 0
#define ORC_CLAMP_SB(x) ORC_CLAMP(x,ORC_SB_MIN,ORC_SB_MAX)
#define ORC_CLAMP_UB(x) ORC_CLAMP(x,ORC_UB_MIN,ORC_UB_MAX)
#define ORC_CLAMP_SW(x) ORC_CLAMP(x,ORC_SW_MIN,ORC_SW_MAX)
#define ORC_CLAMP_UW(x) ORC_CLAMP(x,ORC_UW_MIN,ORC_UW_MAX)
#define ORC_CLAMP_SL(x) ORC_CLAMP(x,ORC_SL_MIN,ORC_SL_MAX)
#define ORC_CLAMP_UL(x) ORC_CLAMP(x,ORC_UL_MIN,ORC_UL_MAX)
#define ORC_SWAP_W(x) ((((x)&0xff)<<8) | (((x)&0xff00)>>8))
#define ORC_SWAP_L(x) ((((x)&0xff)<<24) | (((x)&0xff00)<<8) | (((x)&0xff0000)>>8) | (((x)&0xff000000)>>24))
#define ORC_PTR_OFFSET(ptr,offset) ((void *)(((unsigned char *)(ptr)) + (offset)))
/* end Orc C target preamble */
/* orc_splat_u32 */
#ifdef DISABLE_ORC
void
orc_splat_u32 (guint32 * d1, int p1, int n)
{
int i;
orc_union32 var0;
orc_union32 *ptr0;
const orc_union32 var24 = { p1 };
ptr0 = (orc_union32 *) d1;
for (i = 0; i < n; i++) {
/* 0: copyl */
var0.i = var24.i;
*ptr0 = var0;
ptr0++;
}
}
#else
static void
_backup_orc_splat_u32 (OrcExecutor * ex)
{
int i;
int n = ex->n;
orc_union32 var0;
orc_union32 *ptr0;
const orc_union32 var24 = *(orc_union32 *) (ex->params + 24);
ptr0 = (orc_union32 *) ex->arrays[0];
for (i = 0; i < n; i++) {
/* 0: copyl */
var0.i = var24.i;
*ptr0 = var0;
ptr0++;
}
}
void
orc_splat_u32 (guint32 * d1, int p1, int n)
{
OrcExecutor _ex, *ex = &_ex;
static int p_inited = 0;
static OrcProgram *p = 0;
void (*func) (OrcExecutor *);
if (!p_inited) {
orc_once_mutex_lock ();
if (!p_inited) {
OrcCompileResult result;
p = orc_program_new ();
orc_program_set_name (p, "orc_splat_u32");
orc_program_set_backup_function (p, _backup_orc_splat_u32);
orc_program_add_destination (p, 4, "d1");
orc_program_add_parameter (p, 4, "p1");
orc_program_append (p, "copyl", ORC_VAR_D1, ORC_VAR_P1, ORC_VAR_D1);
result = orc_program_compile (p);
}
p_inited = TRUE;
orc_once_mutex_unlock ();
}
ex->program = p;
ex->n = n;
ex->arrays[ORC_VAR_D1] = d1;
ex->params[ORC_VAR_P1] = p1;
func = p->code_exec;
func (ex);
}
#endif
/* orc_memcpy_u32 */
#ifdef DISABLE_ORC
void
orc_memcpy_u32 (guint32 * d1, const guint32 * s1, int n)
{
int i;
orc_union32 var0;
orc_union32 *ptr0;
orc_union32 var4;
const orc_union32 *ptr4;
ptr0 = (orc_union32 *) d1;
ptr4 = (orc_union32 *) s1;
for (i = 0; i < n; i++) {
var4 = *ptr4;
ptr4++;
/* 0: copyl */
var0.i = var4.i;
*ptr0 = var0;
ptr0++;
}
}
#else
static void
_backup_orc_memcpy_u32 (OrcExecutor * ex)
{
int i;
int n = ex->n;
orc_union32 var0;
orc_union32 *ptr0;
orc_union32 var4;
const orc_union32 *ptr4;
ptr0 = (orc_union32 *) ex->arrays[0];
ptr4 = (orc_union32 *) ex->arrays[4];
for (i = 0; i < n; i++) {
var4 = *ptr4;
ptr4++;
/* 0: copyl */
var0.i = var4.i;
*ptr0 = var0;
ptr0++;
}
}
void
orc_memcpy_u32 (guint32 * d1, const guint32 * s1, int n)
{
OrcExecutor _ex, *ex = &_ex;
static int p_inited = 0;
static OrcProgram *p = 0;
void (*func) (OrcExecutor *);
if (!p_inited) {
orc_once_mutex_lock ();
if (!p_inited) {
OrcCompileResult result;
p = orc_program_new ();
orc_program_set_name (p, "orc_memcpy_u32");
orc_program_set_backup_function (p, _backup_orc_memcpy_u32);
orc_program_add_destination (p, 4, "d1");
orc_program_add_source (p, 4, "s1");
orc_program_append (p, "copyl", ORC_VAR_D1, ORC_VAR_S1, ORC_VAR_D1);
result = orc_program_compile (p);
}
p_inited = TRUE;
orc_once_mutex_unlock ();
}
ex->program = p;
ex->n = n;
ex->arrays[ORC_VAR_D1] = d1;
ex->arrays[ORC_VAR_S1] = (void *) s1;
func = p->code_exec;
func (ex);
}
#endif
/* orc_blend_u8 */
#ifdef DISABLE_ORC
void
orc_blend_u8 (guint8 * d1, int d1_stride, const guint8 * s1, int s1_stride,
int p1, int n, int m)
{
int i;
int j;
orc_int8 var0;
orc_int8 *ptr0;
orc_int8 var4;
const orc_int8 *ptr4;
const orc_int8 var16 = 8;
const orc_int16 var24 = p1;
orc_int16 var32;
orc_int16 var33;
orc_int16 var34;
orc_int16 var35;
orc_int16 var36;
orc_int16 var37;
orc_int16 var38;
for (j = 0; j < m; j++) {
ptr0 = ORC_PTR_OFFSET (d1, d1_stride * j);
ptr4 = ORC_PTR_OFFSET (s1, s1_stride * j);
for (i = 0; i < n; i++) {
var0 = *ptr0;
var4 = *ptr4;
ptr4++;
/* 0: convubw */
var32 = (orc_uint8) var0;
/* 1: convubw */
var33 = (orc_uint8) var4;
/* 2: subw */
var34 = var33 - var32;
/* 3: mullw */
var35 = (var34 * var24) & 0xffff;
/* 4: shlw */
var36 = var32 << var16;
/* 5: addw */
var37 = var36 + var35;
/* 6: shruw */
var38 = ((orc_uint16) var37) >> var16;
/* 7: convsuswb */
var0 = ORC_CLAMP_UB (var38);
*ptr0 = var0;
ptr0++;
}
}
}
#else
static void
_backup_orc_blend_u8 (OrcExecutor * ex)
{
int i;
int j;
int n = ex->n;
int m = ex->params[ORC_VAR_A1];
orc_int8 var0;
orc_int8 *ptr0;
orc_int8 var4;
const orc_int8 *ptr4;
const orc_int8 var16 = 8;
const orc_int16 var24 = ex->params[24];
orc_int16 var32;
orc_int16 var33;
orc_int16 var34;
orc_int16 var35;
orc_int16 var36;
orc_int16 var37;
orc_int16 var38;
for (j = 0; j < m; j++) {
ptr0 = ORC_PTR_OFFSET (ex->arrays[0], ex->params[0] * j);
ptr4 = ORC_PTR_OFFSET (ex->arrays[4], ex->params[4] * j);
for (i = 0; i < n; i++) {
var0 = *ptr0;
var4 = *ptr4;
ptr4++;
/* 0: convubw */
var32 = (orc_uint8) var0;
/* 1: convubw */
var33 = (orc_uint8) var4;
/* 2: subw */
var34 = var33 - var32;
/* 3: mullw */
var35 = (var34 * var24) & 0xffff;
/* 4: shlw */
var36 = var32 << var16;
/* 5: addw */
var37 = var36 + var35;
/* 6: shruw */
var38 = ((orc_uint16) var37) >> var16;
/* 7: convsuswb */
var0 = ORC_CLAMP_UB (var38);
*ptr0 = var0;
ptr0++;
}
}
}
void
orc_blend_u8 (guint8 * d1, int d1_stride, const guint8 * s1, int s1_stride,
int p1, int n, int m)
{
OrcExecutor _ex, *ex = &_ex;
static int p_inited = 0;
static OrcProgram *p = 0;
void (*func) (OrcExecutor *);