libavcodec: arm: Add a NEON implementation of pixblockdsp

Cortex A7     A8     A9    A53   A72
get_pixels_c:                144.7  146.0  143.0  137.7   69.0
get_pixels_armv6:            112.0  106.7   90.2   95.0   72.5
get_pixels_neon:              69.0   29.7   68.7   40.2   19.0
get_pixels_unaligned_c:      144.7  146.2  143.0  137.7   69.0
get_pixels_unaligned_neon:    77.0   36.5   72.5   48.5   19.0
diff_pixels_c:               376.7  319.7  265.5  307.7  148.0
diff_pixels_armv6:           179.0  159.5  205.5  139.0  142.0
diff_pixels_neon:             69.0   40.2   77.5   53.2   26.0
diff_pixels_unaligned_c:     376.7  319.7  265.5  307.7  148.0
diff_pixels_unaligned_neon:   85.0   54.5   93.5   66.7   26.0

Signed-off-by: Martin Storsjö <martin@martin.st>
This commit is contained in:
Martin Storsjö
2020-05-13 13:57:59 +03:00
parent 5bdffced0a
commit b252178321
3 changed files with 88 additions and 0 deletions

View File

@@ -29,6 +29,15 @@ void ff_get_pixels_armv6(int16_t *block, const uint8_t *pixels,
void ff_diff_pixels_armv6(int16_t *block, const uint8_t *s1,
const uint8_t *s2, ptrdiff_t stride);
void ff_get_pixels_neon(int16_t *block, const uint8_t *pixels,
ptrdiff_t stride);
void ff_get_pixels_unaligned_neon(int16_t *block, const uint8_t *pixels,
ptrdiff_t stride);
void ff_diff_pixels_neon(int16_t *block, const uint8_t *s1,
const uint8_t *s2, ptrdiff_t stride);
void ff_diff_pixels_unaligned_neon(int16_t *block, const uint8_t *s1,
const uint8_t *s2, ptrdiff_t stride);
av_cold void ff_pixblockdsp_init_arm(PixblockDSPContext *c,
AVCodecContext *avctx,
unsigned high_bit_depth)
@@ -40,4 +49,13 @@ av_cold void ff_pixblockdsp_init_arm(PixblockDSPContext *c,
c->get_pixels = ff_get_pixels_armv6;
c->diff_pixels = ff_diff_pixels_armv6;
}
if (have_neon(cpu_flags)) {
if (!high_bit_depth) {
c->get_pixels_unaligned = ff_get_pixels_unaligned_neon;
c->get_pixels = ff_get_pixels_neon;
}
c->diff_pixels_unaligned = ff_diff_pixels_unaligned_neon;
c->diff_pixels = ff_diff_pixels_neon;
}
}