summaryrefslogtreecommitdiff
path: root/libavcodec/x86
diff options
context:
space:
mode:
Diffstat (limited to 'libavcodec/x86')
-rw-r--r--libavcodec/x86/dsputil_init.c37
-rw-r--r--libavcodec/x86/dsputil_x86.h3
-rw-r--r--libavcodec/x86/dsputilenc_mmx.c15
3 files changed, 24 insertions, 31 deletions
diff --git a/libavcodec/x86/dsputil_init.c b/libavcodec/x86/dsputil_init.c
index 00f89b2525..288d1af0ad 100644
--- a/libavcodec/x86/dsputil_init.c
+++ b/libavcodec/x86/dsputil_init.c
@@ -518,11 +518,9 @@ do { \
} while (0)
static av_cold void dsputil_init_mmx(DSPContext *c, AVCodecContext *avctx,
- int cpu_flags)
+ int cpu_flags, unsigned high_bit_depth)
{
#if HAVE_MMX_INLINE
- const int high_bit_depth = avctx->bits_per_raw_sample > 8;
-
c->put_pixels_clamped = ff_put_pixels_clamped_mmx;
c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_mmx;
c->add_pixels_clamped = ff_add_pixels_clamped_mmx;
@@ -559,11 +557,9 @@ static av_cold void dsputil_init_mmx(DSPContext *c, AVCodecContext *avctx,
}
static av_cold void dsputil_init_mmxext(DSPContext *c, AVCodecContext *avctx,
- int cpu_flags)
+ int cpu_flags, unsigned high_bit_depth)
{
#if HAVE_MMXEXT_INLINE
- const int high_bit_depth = avctx->bits_per_raw_sample > 8;
-
if (!high_bit_depth && avctx->idct_algo == FF_IDCT_XVIDMMX) {
c->idct_put = ff_idct_xvid_mmxext_put;
c->idct_add = ff_idct_xvid_mmxext_add;
@@ -590,11 +586,9 @@ static av_cold void dsputil_init_mmxext(DSPContext *c, AVCodecContext *avctx,
}
static av_cold void dsputil_init_sse(DSPContext *c, AVCodecContext *avctx,
- int cpu_flags)
+ int cpu_flags, unsigned high_bit_depth)
{
#if HAVE_SSE_INLINE
- const int high_bit_depth = avctx->bits_per_raw_sample > 8;
-
c->vector_clipf = ff_vector_clipf_sse;
#if FF_API_XVMC
@@ -613,11 +607,9 @@ FF_ENABLE_DEPRECATION_WARNINGS
}
static av_cold void dsputil_init_sse2(DSPContext *c, AVCodecContext *avctx,
- int cpu_flags)
+ int cpu_flags, unsigned high_bit_depth)
{
#if HAVE_SSE2_INLINE
- const int high_bit_depth = avctx->bits_per_raw_sample > 8;
-
if (!high_bit_depth && avctx->idct_algo == FF_IDCT_XVIDMMX) {
c->idct_put = ff_idct_xvid_sse2_put;
c->idct_add = ff_idct_xvid_sse2_add;
@@ -639,7 +631,7 @@ static av_cold void dsputil_init_sse2(DSPContext *c, AVCodecContext *avctx,
}
static av_cold void dsputil_init_ssse3(DSPContext *c, AVCodecContext *avctx,
- int cpu_flags)
+ int cpu_flags, unsigned high_bit_depth)
{
#if HAVE_SSSE3_EXTERNAL
c->add_hfyu_left_prediction = ff_add_hfyu_left_prediction_ssse3;
@@ -653,14 +645,15 @@ static av_cold void dsputil_init_ssse3(DSPContext *c, AVCodecContext *avctx,
}
static av_cold void dsputil_init_sse4(DSPContext *c, AVCodecContext *avctx,
- int cpu_flags)
+ int cpu_flags, unsigned high_bit_depth)
{
#if HAVE_SSE4_EXTERNAL
c->vector_clip_int32 = ff_vector_clip_int32_sse4;
#endif /* HAVE_SSE4_EXTERNAL */
}
-av_cold void ff_dsputil_init_x86(DSPContext *c, AVCodecContext *avctx)
+av_cold void ff_dsputil_init_x86(DSPContext *c, AVCodecContext *avctx,
+ unsigned high_bit_depth)
{
int cpu_flags = av_get_cpu_flags();
@@ -670,23 +663,23 @@ av_cold void ff_dsputil_init_x86(DSPContext *c, AVCodecContext *avctx)
#endif
if (X86_MMX(cpu_flags))
- dsputil_init_mmx(c, avctx, cpu_flags);
+ dsputil_init_mmx(c, avctx, cpu_flags, high_bit_depth);
if (X86_MMXEXT(cpu_flags))
- dsputil_init_mmxext(c, avctx, cpu_flags);
+ dsputil_init_mmxext(c, avctx, cpu_flags, high_bit_depth);
if (X86_SSE(cpu_flags))
- dsputil_init_sse(c, avctx, cpu_flags);
+ dsputil_init_sse(c, avctx, cpu_flags, high_bit_depth);
if (X86_SSE2(cpu_flags))
- dsputil_init_sse2(c, avctx, cpu_flags);
+ dsputil_init_sse2(c, avctx, cpu_flags, high_bit_depth);
if (EXTERNAL_SSSE3(cpu_flags))
- dsputil_init_ssse3(c, avctx, cpu_flags);
+ dsputil_init_ssse3(c, avctx, cpu_flags, high_bit_depth);
if (EXTERNAL_SSE4(cpu_flags))
- dsputil_init_sse4(c, avctx, cpu_flags);
+ dsputil_init_sse4(c, avctx, cpu_flags, high_bit_depth);
if (CONFIG_ENCODERS)
- ff_dsputilenc_init_mmx(c, avctx);
+ ff_dsputilenc_init_mmx(c, avctx, high_bit_depth);
}
diff --git a/libavcodec/x86/dsputil_x86.h b/libavcodec/x86/dsputil_x86.h
index 8f1fc17474..8f8ea052b3 100644
--- a/libavcodec/x86/dsputil_x86.h
+++ b/libavcodec/x86/dsputil_x86.h
@@ -104,7 +104,8 @@
"psubb "#regb", "#regr" \n\t" \
"psubb "#regd", "#regp" \n\t"
-void ff_dsputilenc_init_mmx(DSPContext *c, AVCodecContext *avctx);
+void ff_dsputilenc_init_mmx(DSPContext *c, AVCodecContext *avctx,
+ unsigned high_bit_depth);
void ff_dsputil_init_pix_mmx(DSPContext *c, AVCodecContext *avctx);
void ff_add_pixels_clamped_mmx(const int16_t *block, uint8_t *pixels,
diff --git a/libavcodec/x86/dsputilenc_mmx.c b/libavcodec/x86/dsputilenc_mmx.c
index 2c320371b0..99f094e76e 100644
--- a/libavcodec/x86/dsputilenc_mmx.c
+++ b/libavcodec/x86/dsputilenc_mmx.c
@@ -986,16 +986,15 @@ hadamard_func(mmxext)
hadamard_func(sse2)
hadamard_func(ssse3)
-av_cold void ff_dsputilenc_init_mmx(DSPContext *c, AVCodecContext *avctx)
+av_cold void ff_dsputilenc_init_mmx(DSPContext *c, AVCodecContext *avctx,
+ unsigned high_bit_depth)
{
int cpu_flags = av_get_cpu_flags();
const int dct_algo = avctx->dct_algo;
#if HAVE_YASM
- int bit_depth = avctx->bits_per_raw_sample;
-
if (EXTERNAL_MMX(cpu_flags)) {
- if (bit_depth <= 8)
+ if (!high_bit_depth)
c->get_pixels = ff_get_pixels_mmx;
c->diff_pixels = ff_diff_pixels_mmx;
c->pix_sum = ff_pix_sum16_mmx;
@@ -1003,13 +1002,13 @@ av_cold void ff_dsputilenc_init_mmx(DSPContext *c, AVCodecContext *avctx)
}
if (EXTERNAL_SSE2(cpu_flags))
- if (bit_depth <= 8)
+ if (!high_bit_depth)
c->get_pixels = ff_get_pixels_sse2;
#endif /* HAVE_YASM */
#if HAVE_INLINE_ASM
if (INLINE_MMX(cpu_flags)) {
- if (avctx->bits_per_raw_sample <= 8 &&
+ if (!high_bit_depth &&
(dct_algo == FF_DCT_AUTO || dct_algo == FF_DCT_MMX))
c->fdct = ff_fdct_mmx;
@@ -1039,7 +1038,7 @@ av_cold void ff_dsputilenc_init_mmx(DSPContext *c, AVCodecContext *avctx)
}
if (INLINE_MMXEXT(cpu_flags)) {
- if (avctx->bits_per_raw_sample <= 8 &&
+ if (!high_bit_depth &&
(dct_algo == FF_DCT_AUTO || dct_algo == FF_DCT_MMX))
c->fdct = ff_fdct_mmxext;
@@ -1054,7 +1053,7 @@ av_cold void ff_dsputilenc_init_mmx(DSPContext *c, AVCodecContext *avctx)
}
if (INLINE_SSE2(cpu_flags)) {
- if (avctx->bits_per_raw_sample <= 8 &&
+ if (!high_bit_depth &&
(dct_algo == FF_DCT_AUTO || dct_algo == FF_DCT_MMX))
c->fdct = ff_fdct_sse2;
generated by cgit v1.2.3 (git 2.39.1) at 2024-06-28 21:12:41 +0000