---
libavcodec/fmtconvert.c | 8 ++++++++
libavcodec/fmtconvert.h | 11 +++++++++++
libavcodec/x86/fmtconvert.asm | 29 +++++++++++++++++++++++++++++
libavcodec/x86/fmtconvert_mmx.c | 5 +++++
4 files changed, 53 insertions(+)
diff --git a/libavcodec/fmtconvert.c b/libavcodec/fmtconvert.c
index 58fece7..4b683d2 100644
--- a/libavcodec/fmtconvert.c
+++ b/libavcodec/fmtconvert.c
@@ -29,6 +29,13 @@ static void int32_to_float_fmul_scalar_c(float *dst, const
int *src, float mul,
dst[i] = src[i] * mul;
}
+static void int32_to_int16_clipped_c(int16_t *dst, const int32_t *src, int len)
+{
+ int i;
+ for (i = 0; i < len; i++)
+ dst[i] = av_clip_int16(src[i]);
+}
+
static av_always_inline int float_to_int16_one(const float *src){
return av_clip_int16(lrintf(*src));
}
@@ -81,6 +88,7 @@ av_cold void ff_fmt_convert_init(FmtConvertContext *c,
AVCodecContext *avctx)
c->float_to_int16 = float_to_int16_c;
c->float_to_int16_interleave = float_to_int16_interleave_c;
c->float_interleave = ff_float_interleave_c;
+ c->int32_to_int16_clipped = int32_to_int16_clipped_c;
if (ARCH_ARM) ff_fmt_convert_init_arm(c, avctx);
if (HAVE_ALTIVEC) ff_fmt_convert_init_altivec(c, avctx);
diff --git a/libavcodec/fmtconvert.h b/libavcodec/fmtconvert.h
index b879ee6..824f70b 100644
--- a/libavcodec/fmtconvert.h
+++ b/libavcodec/fmtconvert.h
@@ -38,6 +38,17 @@ typedef struct FmtConvertContext {
void (*int32_to_float_fmul_scalar)(float *dst, const int *src, float mul,
int len);
/**
+ * Convert an array of int32_t to clipped int16_t.
+ * @param dst destination array of int16_t.
+ * constraints: 16-byte aligned
+ * @param src source array of int32_t.
+ * constraints: 16-byte aligned
+ * @param len number of elements to convert.
+ * constraints: multiple of 8
+ */
+ void (*int32_to_int16_clipped)(int16_t *dst, const int32_t *src, int len);
+
+ /**
* Convert an array of float to an array of int16_t.
*
* Convert floats from in the range [-32768.0,32767.0] to ints
diff --git a/libavcodec/x86/fmtconvert.asm b/libavcodec/x86/fmtconvert.asm
index 63befc9..05ce7c7 100644
--- a/libavcodec/x86/fmtconvert.asm
+++ b/libavcodec/x86/fmtconvert.asm
@@ -365,3 +365,32 @@ INIT_XMM
%define PUNPCKLDQ unpcklps
%define PUNPCKHDQ unpckhps
FLOAT_INTERLEAVE2 sse, 5
+
+;-------------------------------------------------------------------------------
+; void ff_int32_to_int16_clipped(int16_t *dst, const int32_t *src, int len);
+;-------------------------------------------------------------------------------
+
+%macro INT32_TO_INT16_CLIPPED 0
+cglobal int32_to_int16_clipped, 3, 3, 1, dst, src, len
+ add lenq, lenq
+ lea srcq, [srcq + 2 * lenq]
+ add dstq, lenq
+ neg lenq
+
+.loop:
+ mova m0, [srcq + 2 * lenq]
+ packssdw m0, [srcq + 2 * lenq + mmsize]
+ mova [dstq + lenq], m0
+ add lenq, mmsize
+ jl .loop
+%if mmsize == 8
+ RET
+%else
+ REP_RET
+%endif
+%endmacro
+
+INIT_MMX mmx
+INT32_TO_INT16_CLIPPED
+INIT_XMM sse2
+INT32_TO_INT16_CLIPPED
diff --git a/libavcodec/x86/fmtconvert_mmx.c b/libavcodec/x86/fmtconvert_mmx.c
index 42cb0bc..87cd504 100644
--- a/libavcodec/x86/fmtconvert_mmx.c
+++ b/libavcodec/x86/fmtconvert_mmx.c
@@ -43,6 +43,9 @@ void ff_float_to_int16_interleave6_sse(int16_t *dst, const
float **src, int len)
void ff_float_to_int16_interleave6_3dnow(int16_t *dst, const float **src, int
len);
void ff_float_to_int16_interleave6_3dn2(int16_t *dst, const float **src, int
len);
+void ff_int32_to_int16_clipped_mmx(int16_t *dst, const int32_t *src, int len);
+void ff_int32_to_int16_clipped_sse2(int16_t *dst, const int32_t *src, int len);
+
#define ff_float_to_int16_interleave6_sse2 ff_float_to_int16_interleave6_sse
#define FLOAT_TO_INT16_INTERLEAVE(cpu) \
@@ -115,6 +118,7 @@ void ff_fmt_convert_init_x86(FmtConvertContext *c,
AVCodecContext *avctx)
if (mm_flags & AV_CPU_FLAG_MMX) {
c->float_interleave = float_interleave_mmx;
+ c->int32_to_int16_clipped = ff_int32_to_int16_clipped_mmx;
if (HAVE_AMD3DNOW && mm_flags & AV_CPU_FLAG_3DNOW) {
if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
@@ -137,6 +141,7 @@ void ff_fmt_convert_init_x86(FmtConvertContext *c,
AVCodecContext *avctx)
c->int32_to_float_fmul_scalar = ff_int32_to_float_fmul_scalar_sse2;
c->float_to_int16 = ff_float_to_int16_sse2;
c->float_to_int16_interleave = float_to_int16_interleave_sse2;
+ c->int32_to_int16_clipped = ff_int32_to_int16_clipped_sse2;
}
}
#endif
--
1.7.10
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel
