audioconvert: Use ORC for the float<->int32 conversion

This should speed up standard Vorbis encoding and decoding pipelines a bit.

Thanks to David Schleef for the assistance to get the ORC code right
and explaining everything.
This commit is contained in:
Sebastian Dröge 2010-08-24 11:11:49 +02:00
parent faebf174fb
commit 5e0706c74f
2 changed files with 92 additions and 4 deletions

View file

@ -41,6 +41,8 @@
#define orc_audio_convert_unpack_u32_be orc_audio_convert_unpack_u32_swap
#define orc_audio_convert_unpack_s32_le orc_audio_convert_unpack_s32
#define orc_audio_convert_unpack_s32_be orc_audio_convert_unpack_s32_swap
#define orc_audio_convert_unpack_float_le orc_audio_convert_unpack_float_s32
#define orc_audio_convert_unpack_float_be orc_audio_convert_unpack_float_s32_swap
#define orc_audio_convert_pack_u16_le orc_audio_convert_pack_u16
#define orc_audio_convert_pack_u16_be orc_audio_convert_pack_u16_swap
#define orc_audio_convert_pack_s16_le orc_audio_convert_pack_s16
@ -49,6 +51,8 @@
#define orc_audio_convert_pack_u32_be orc_audio_convert_pack_u32_swap
#define orc_audio_convert_pack_s32_le orc_audio_convert_pack_s32
#define orc_audio_convert_pack_s32_be orc_audio_convert_pack_s32_swap
#define orc_audio_convert_pack_float_le orc_audio_convert_pack_s32_float
#define orc_audio_convert_pack_float_be orc_audio_convert_pack_s32_float_swap
#else
#define orc_audio_convert_unpack_u16_be orc_audio_convert_unpack_u16
#define orc_audio_convert_unpack_u16_le orc_audio_convert_unpack_u16_swap
@ -58,6 +62,8 @@
#define orc_audio_convert_unpack_u32_le orc_audio_convert_unpack_u32_swap
#define orc_audio_convert_unpack_s32_be orc_audio_convert_unpack_s32
#define orc_audio_convert_unpack_s32_le orc_audio_convert_unpack_s32_swap
#define orc_audio_convert_unpack_float_be orc_audio_convert_unpack_float_s32
#define orc_audio_convert_unpack_float_le orc_audio_convert_unpack_float_s32_swap
#define orc_audio_convert_pack_u16_be orc_audio_convert_pack_u16
#define orc_audio_convert_pack_u16_le orc_audio_convert_pack_u16_swap
#define orc_audio_convert_pack_s16_be orc_audio_convert_pack_s16
@ -66,6 +72,8 @@
#define orc_audio_convert_pack_u32_le orc_audio_convert_pack_u32_swap
#define orc_audio_convert_pack_s32_be orc_audio_convert_pack_s32
#define orc_audio_convert_pack_s32_le orc_audio_convert_pack_s32_swap
#define orc_audio_convert_pack_float_be orc_audio_convert_pack_s32_float
#define orc_audio_convert_pack_float_le orc_audio_convert_pack_s32_float_swap
#endif
/* sign bit in the intermediate format */
@ -114,6 +122,14 @@ MAKE_UNPACK_FUNC_NAME (name) (type * src, gint32 * dst, gint s, gint count) \
} \
}
/* unpack from float to signed integer 32 with orc */
#define MAKE_UNPACK_FUNC_ORC_FI(name, type, READ_FUNC) \
static void \
MAKE_UNPACK_FUNC_NAME (name) (type * src, gint32 * dst, gint s, gint count) \
{ \
MAKE_ORC_UNPACK_FUNC_NAME (name) ((guint32 *) dst, src, count); \
}
/* unpack from float to float 64 (double) */
#define MAKE_UNPACK_FUNC_FF(name, type, FUNC) \
static void \
@ -162,8 +178,8 @@ MAKE_UNPACK_FUNC_ORC_II (u32_le, 4, SIGNED, READ32_FROM_LE);
MAKE_UNPACK_FUNC_ORC_II (s32_le, 4, 0, READ32_FROM_LE);
MAKE_UNPACK_FUNC_ORC_II (u32_be, 4, SIGNED, READ32_FROM_BE);
MAKE_UNPACK_FUNC_ORC_II (s32_be, 4, 0, READ32_FROM_BE);
MAKE_UNPACK_FUNC_FI (float_le, gfloat, GFLOAT_FROM_LE);
MAKE_UNPACK_FUNC_FI (float_be, gfloat, GFLOAT_FROM_BE);
MAKE_UNPACK_FUNC_ORC_FI (float_le, gfloat, GFLOAT_FROM_LE);
MAKE_UNPACK_FUNC_ORC_FI (float_be, gfloat, GFLOAT_FROM_BE);
MAKE_UNPACK_FUNC_FI (double_le, gdouble, GDOUBLE_FROM_LE);
MAKE_UNPACK_FUNC_FI (double_be, gdouble, GDOUBLE_FROM_BE);
MAKE_UNPACK_FUNC_FF (float_hq_le, gfloat, GFLOAT_FROM_LE);
@ -231,6 +247,15 @@ MAKE_PACK_FUNC_NAME (name) (gint32 *src, guint8 * dst, \
} \
}
/* pack from signed integer 32 to float using orc */
#define MAKE_PACK_FUNC_ORC_IF(name, type, FUNC) \
static void \
MAKE_PACK_FUNC_NAME (name) (gint32 * src, type * dst, gint scale, \
gint count) \
{ \
MAKE_PACK_FUNC_NAME_ORC (name) (dst, src, count); \
}
/* pack from signed integer 32 to float */
#define MAKE_PACK_FUNC_IF(name, type, FUNC) \
static void \
@ -309,8 +334,8 @@ MAKE_PACK_FUNC_ORC_II (u32_le, 4, SIGNED, WRITE32_TO_LE);
MAKE_PACK_FUNC_ORC_II (s32_le, 4, 0, WRITE32_TO_LE);
MAKE_PACK_FUNC_ORC_II (u32_be, 4, SIGNED, WRITE32_TO_BE);
MAKE_PACK_FUNC_ORC_II (s32_be, 4, 0, WRITE32_TO_BE);
MAKE_PACK_FUNC_IF (float_le, gfloat, GFLOAT_TO_LE);
MAKE_PACK_FUNC_IF (float_be, gfloat, GFLOAT_TO_BE);
MAKE_PACK_FUNC_ORC_IF (float_le, gfloat, GFLOAT_TO_LE);
MAKE_PACK_FUNC_ORC_IF (float_be, gfloat, GFLOAT_TO_BE);
MAKE_PACK_FUNC_IF (double_le, gdouble, GDOUBLE_TO_LE);
MAKE_PACK_FUNC_IF (double_be, gdouble, GDOUBLE_TO_BE);
MAKE_PACK_FUNC_FF (float_hq_le, gfloat, GFLOAT_TO_LE);

View file

@ -113,7 +113,51 @@ xorl d1, t1, c1
swapl t1, s1
shll d1, t1, p1
.function orc_audio_convert_unpack_float_s32
.source 4 s1 gfloat
.dest 4 d1 guint32
.temp 4 t1
.temp 4 t2
.temp 4 t3
.temp 4 t4
loadl t1, s1
# multiply with 2147483647.0
mulf t1, t1, 0x4F000000
# add 0.5 for rounding
addf t1, t1, 0x3F000000
convfl t2, t1
# if overflow, t3 = ~0
cmpeql t3, t2, -2147483648
# if negative, t4 = ~0
shrsl t4, t1, 31
# if overflow and !negative, t4 = ~0
andnl t4, t4, t3
# 0x80000000 + ~0 = 0x7fffffff
addl d1, t2, t4
.function orc_audio_convert_unpack_float_s32_swap
.source 4 s1 gfloat
.dest 4 d1 guint32
.temp 4 t1
.temp 4 t2
.temp 4 t3
.temp 4 t4
swapl t1, s1
# multiply with 2147483647.0
mulf t1, t1, 0x4F000000
# add 0.5 for rounding
addf t1, t1, 0x3F000000
convfl t2, t1
# if overflow, t3 = ~0
cmpeql t3, t2, -2147483648
# if negative, t4 = ~0
shrsl t4, t1, 31
# if overflow and !negative, t4 = ~0
andnl t4, t4, t3
# 0x80000000 + ~0 = 0x7fffffff
addl d1, t2, t4
.function orc_audio_convert_pack_u8
.dest 1 d1 guint8
@ -231,4 +275,23 @@ swapl d1, t1
shrsl t1, s1, p1
swapl d1, t1
.function orc_audio_convert_pack_s32_float
.dest 4 d1 gfloat
.source 4 s1 gint32
.temp 4 t1
convlf t1, s1
# divide by 2147483647.0
divf t1, t1, 0x4F000000
storel d1, t1
.function orc_audio_convert_pack_s32_float_swap
.dest 4 d1 gfloat
.source 4 s1 gint32
.temp 4 t1
convlf t1, s1
# divide by 2147483647.0
divf t1, t1, 0x4F000000
swapl d1, t1