[FFmpeg-cvslog] hevc: Add hevc_get_pixel_4/8/12/16/24/32/48/64
Alexandra Hájková
git at videolan.org
Mon Dec 11 02:04:09 EET 2017
ffmpeg | branch: master | Alexandra Hájková <alexandra.khirnova at gmail.com> | Sun Dec 3 09:15:23 2017 +0100| [7993ec19af394fdc58ec64165bc0b12619543a5d] | committer: Martin Storsjö
hevc: Add hevc_get_pixel_4/8/12/16/24/32/48/64
Checkasm timings:
block size bitdepth C NEON
4 8 bit: 146.7 48.7
10 bit: 146.7 52.7
8 8 bit: 430.3 84.4
10 bit: 430.4 119.5
12 8 bit: 812.8 141.0
10 bit: 812.8 195.0
16 8 bit: 1499.1 268.0
10 bit: 1498.9 368.4
24 8 bit: 4394.2 574.8
10 bit: 3696.3 804.8
32 8 bit: 5108.6 568.9
10 bit: 4249.6 918.8
48 8 bit: 16819.6 2304.9
10 bit: 13882.0 3178.5
64 8 bit: 13490.8 1799.5
10 bit: 11018.5 2519.4
Signed-off-by: Martin Storsjö <martin at martin.st>
> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=7993ec19af394fdc58ec64165bc0b12619543a5d
---
libavcodec/arm/Makefile | 3 +-
libavcodec/arm/hevc_mc.S | 395 ++++++++++++++++++++++++++++++++++++++
libavcodec/arm/hevcdsp_init_arm.c | 67 +++++++
3 files changed, 464 insertions(+), 1 deletion(-)
diff --git a/libavcodec/arm/Makefile b/libavcodec/arm/Makefile
index b48745ad40..49e17ce0df 100644
--- a/libavcodec/arm/Makefile
+++ b/libavcodec/arm/Makefile
@@ -135,7 +135,8 @@ NEON-OBJS-$(CONFIG_AAC_DECODER) += arm/aacpsdsp_neon.o \
NEON-OBJS-$(CONFIG_APE_DECODER) += arm/apedsp_neon.o
NEON-OBJS-$(CONFIG_DCA_DECODER) += arm/dcadsp_neon.o \
arm/synth_filter_neon.o
-NEON-OBJS-$(CONFIG_HEVC_DECODER) += arm/hevc_idct.o
+NEON-OBJS-$(CONFIG_HEVC_DECODER) += arm/hevc_idct.o \
+ arm/hevc_mc.o
NEON-OBJS-$(CONFIG_RV30_DECODER) += arm/rv34dsp_neon.o
NEON-OBJS-$(CONFIG_RV40_DECODER) += arm/rv34dsp_neon.o \
arm/rv40dsp_neon.o
diff --git a/libavcodec/arm/hevc_mc.S b/libavcodec/arm/hevc_mc.S
new file mode 100644
index 0000000000..f499d294b9
--- /dev/null
+++ b/libavcodec/arm/hevc_mc.S
@@ -0,0 +1,395 @@
+/*
+ * ARM NEON optimised MC functions for HEVC decoding
+ *
+ * Copyright (c) 2017 Alexandra Hájková
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "libavutil/arm/asm.S"
+
+.macro get_pixels4 bitdepth
+function ff_hevc_get_pixels_4_\bitdepth\()_neon, export=1
+ at r0 dst, r1 dststride, r2 src, r3 srcstride
+ ldr r12, [sp] @height
+ cmp r12, #0
+ it eq
+ bxeq lr
+
+1:
+ .if \bitdepth == 8
+ vld1.32 {d0[0]}, [r2], r3
+ vld1.32 {d1[0]}, [r2], r3
+ vld1.32 {d2[0]}, [r2], r3
+ vld1.32 {d3[0]}, [r2], r3
+ vshll.u8 q8, d0, #6
+ vshll.u8 q9, d1, #6
+ vshll.u8 q10, d2, #6
+ vshll.u8 q11, d3, #6
+ .else
+ vld1.16 {d0}, [r2], r3
+ vld1.16 {d1}, [r2], r3
+ vld1.16 {d2}, [r2], r3
+ vld1.16 {d3}, [r2], r3
+ vshl.i16 d16, d0, #4
+ vshl.i16 d18, d1, #4
+ vshl.i16 d20, d2, #4
+ vshl.i16 d22, d3, #4
+ .endif
+
+ vst1.16 {d16}, [r0, :64], r1
+ vst1.16 {d18}, [r0, :64], r1
+ vst1.16 {d20}, [r0, :64], r1
+ vst1.16 {d22}, [r0, :64], r1
+ subs r12, #4
+ bgt 1b
+
+ bx lr
+endfunc
+.endm
+
+.macro get_pixels8 bitdepth
+function ff_hevc_get_pixels_8_\bitdepth\()_neon, export=1
+ at r0 dst, r1 dststride, r2 src, r3 srcstride
+ ldr r12, [sp] @height
+ cmp r12, #0
+ it eq
+ bxeq lr
+
+1:
+ .if \bitdepth == 8
+ vld1.8 {d0}, [r2], r3
+ vld1.8 {d1}, [r2], r3
+ vld1.8 {d2}, [r2], r3
+ vld1.8 {d3}, [r2], r3
+ vshll.u8 q8, d0, #6
+ vshll.u8 q9, d1, #6
+ vshll.u8 q10, d2, #6
+ vshll.u8 q11, d3, #6
+ .else
+ vld1.16 {d16-d17}, [r2], r3
+ vld1.16 {d18-d19}, [r2], r3
+ vld1.16 {d20-d21}, [r2], r3
+ vld1.16 {d22-d23}, [r2], r3
+ vshl.i16 q8, q8, #4
+ vshl.i16 q9, q9, #4
+ vshl.i16 q10, q10, #4
+ vshl.i16 q11, q11, #4
+ .endif
+
+ vst1.16 {d16-d17}, [r0, :64], r1
+ vst1.16 {d18-d19}, [r0, :64], r1
+ vst1.16 {d20-d21}, [r0, :64], r1
+ vst1.16 {d22-d23}, [r0, :64], r1
+ subs r12, #4
+ bgt 1b
+
+ bx lr
+endfunc
+.endm
+
+.macro get_pixels12 bitdepth
+function ff_hevc_get_pixels_12_\bitdepth\()_neon, export=1
+ at r0 - dst, r1 - dststride, r2 - src, r3 - srcstride
+ ldr r12, [sp] @height
+ cmp r12, #0
+ it eq
+ bxeq lr
+ push {r4-r5, lr}
+ add r4, r0, #16
+
+1:
+ .if \bitdepth == 8
+ add r5, r2, #8
+ vld1.8 {d0}, [r2], r3
+ vld1.32 {d4[0]}, [r5], r3
+ vld1.8 {d1}, [r2], r3
+ vld1.32 {d5[0]}, [r5], r3
+ vld1.8 {d2}, [r2], r3
+ vld1.32 {d6[0]}, [r5], r3
+ vld1.8 {d3}, [r2], r3
+ vld1.32 {d7[0]}, [r5], r3
+ vshll.u8 q8, d0, #6
+ vshll.u8 q12, d4, #6
+ vshll.u8 q9, d1, #6
+ vshll.u8 q13, d5, #6
+ vshll.u8 q10, d2, #6
+ vshll.u8 q14, d6, #6
+ vshll.u8 q11, d3, #6
+ vshll.u8 q15, d7, #6
+ .else
+ add r5, r2, #16
+ vld1.16 {d16-d17}, [r2], r3
+ vld1.16 {d24}, [r5], r3
+ vld1.16 {d18-d19}, [r2], r3
+ vld1.16 {d26}, [r5], r3
+ vld1.16 {d20-d21}, [r2], r3
+ vld1.16 {d28}, [r5], r3
+ vld1.16 {d22-d23}, [r2], r3
+ vld1.16 {d30}, [r5], r3
+ vshl.i16 q8, q8, #4
+ vshl.i16 d24, d24, #4
+ vshl.i16 q9, q9, #4
+ vshl.i16 d26, d26, #4
+ vshl.i16 q10, q10, #4
+ vshl.i16 d28, d28, #4
+ vshl.i16 q11, q11, #4
+ vshl.i16 d30, d30, #4
+ .endif
+
+ vst1.16 {d16-d17}, [r0, :64], r1
+ vst1.16 {d24}, [r4, :64], r1
+ vst1.16 {d18-d19}, [r0, :64], r1
+ vst1.16 {d26}, [r4, :64], r1
+ vst1.16 {d20-d21}, [r0, :64], r1
+ vst1.16 {d28}, [r4, :64], r1
+ vst1.16 {d22-d23}, [r0, :64], r1
+ vst1.16 {d30}, [r4, :64], r1
+ subs r12, #4
+ bgt 1b
+
+ pop {r4-r5, pc}
+endfunc
+.endm
+
+ at 8 bitdepth case
+.macro process_8 load
+ vld1.8 {d0-d1}, [\load], r3
+ vld1.8 {d2-d3}, [\load], r3
+ vld1.8 {d4-d5}, [\load], r3
+ vld1.8 {d6-d7}, [\load], r3
+ vshll.u8 q8, d0, #6
+ vshll.u8 q9, d1, #6
+ vshll.u8 q10, d2, #6
+ vshll.u8 q11, d3, #6
+ vshll.u8 q12, d4, #6
+ vshll.u8 q13, d5, #6
+ vshll.u8 q14, d6, #6
+ vshll.u8 q15, d7, #6
+.endm
+
+ at 10 bitdepth case
+.macro process_10 load
+ vld1.16 {d16-d19}, [\load], r3
+ vld1.16 {d20-d23}, [\load], r3
+ vld1.16 {d24-d27}, [\load], r3
+ vld1.16 {d28-d31}, [\load], r3
+ vshl.i16 q8, q8, #4
+ vshl.i16 q9, q9, #4
+ vshl.i16 q10, q10, #4
+ vshl.i16 q11, q11, #4
+ vshl.i16 q12, q12, #4
+ vshl.i16 q13, q13, #4
+ vshl.i16 q14, q14, #4
+ vshl.i16 q15, q15, #4
+.endm
+
+.macro store_4x16 store
+ vst1.16 {d16-d19}, [\store, :128], r1
+ vst1.16 {d20-d23}, [\store, :128], r1
+ vst1.16 {d24-d27}, [\store, :128], r1
+ vst1.16 {d28-d31}, [\store, :128], r1
+.endm
+
+.macro get_pixels16 bitdepth
+function ff_hevc_get_pixels_16_\bitdepth\()_neon, export=1
+ at r0 dst, r1 dststride, r2 src, r3 srcstride
+ ldr r12, [sp] @height
+ cmp r12, #0
+ it eq
+ bxeq lr
+
+1:
+ .if \bitdepth == 8
+ process_8 r2
+ .else
+ process_10 r2
+ .endif
+
+ store_4x16 r0
+ subs r12, #4
+ bgt 1b
+
+ bx lr
+endfunc
+.endm
+
+.macro get_pixels24 bitdepth
+function ff_hevc_get_pixels_24_\bitdepth\()_neon, export=1
+ at r0 dst, r1 dststride, r2 src, r3 srcstride
+ ldr r12, [sp] @height
+ cmp r12, #0
+ it eq
+ bxeq lr
+
+ push {r0-r4, lr}
+ push {r12}
+ bl X(ff_hevc_get_pixels_8_\bitdepth\()_neon)
+ pop {r12}
+ pop {r0-r4, lr}
+
+ .if \bitdepth == 8
+ add r2, #8
+ .else
+ add r2, #16
+ .endif
+ add r0, #16
+ b X(ff_hevc_get_pixels_16_\bitdepth\()_neon)
+endfunc
+.endm
+
+.macro get_pixels32 bitdepth
+function ff_hevc_get_pixels_32_\bitdepth\()_neon, export=1
+ at r0 dst, r1 dststride, r2 src, r3 srcstride
+ ldr r12, [sp] @height
+ cmp r12, #0
+ it eq
+ bxeq lr
+ push {r4-r5, lr}
+ .if \bitdepth == 8
+ add r4, r2, #16
+ .else
+ add r4, r2, #32
+ .endif
+ add r5, r0, #32
+
+1:
+ .if \bitdepth == 8
+ process_8 r2
+ .else
+ process_10 r2
+ .endif
+ store_4x16 r0
+
+ .if \bitdepth == 8
+ process_8 r4
+ .else
+ process_10 r4
+ .endif
+ store_4x16 r5
+
+ subs r12, #4
+ bgt 1b
+
+ pop {r4-r5, pc}
+endfunc
+.endm
+
+.macro get_pixels48 bitdepth
+function ff_hevc_get_pixels_48_\bitdepth\()_neon, export=1
+ at r0 dst, r1 dststride, r2 src, r3 srcstride
+ ldr r12, [sp] @height
+ cmp r12, #0
+ it eq
+ bxeq lr
+
+ push {r0-r4, lr}
+ push {r12}
+ bl X(ff_hevc_get_pixels_16_\bitdepth\()_neon)
+ pop {r12}
+ pop {r0-r4, lr}
+
+ .if \bitdepth == 8
+ add r2, #16
+ .else
+ add r2, #32
+ .endif
+ add r0, #32
+ b X(ff_hevc_get_pixels_32_\bitdepth\()_neon)
+endfunc
+.endm
+
+.macro get_pixels64 bitdepth
+function ff_hevc_get_pixels_64_\bitdepth\()_neon, export=1
+ at r0 dst, r1 dststride, r2 src, r3 srcstride
+ ldr r12, [sp] @height
+ cmp r12, #0
+ it eq
+ bxeq lr
+ push {r4-r9, lr}
+ .if \bitdepth == 8
+ add r4, r2, #16
+ add r6, r4, #16
+ add r8, r6, #16
+ .else
+ add r4, r2, #32
+ add r6, r4, #32
+ add r8, r6, #32
+ .endif
+ add r5, r0, #32
+ add r7, r5, #32
+ add r9, r7, #32
+
+1:
+ .if \bitdepth == 8
+ process_8 r2
+ .else
+ process_10 r2
+ .endif
+ store_4x16 r0
+
+ .if \bitdepth == 8
+ process_8 r4
+ .else
+ process_10 r4
+ .endif
+ store_4x16 r5
+
+ .if \bitdepth == 8
+ process_8 r6
+ .else
+ process_10 r6
+ .endif
+ store_4x16 r7
+ .if \bitdepth == 8
+ process_8 r8
+ .else
+ process_10 r8
+ .endif
+ store_4x16 r9
+
+ subs r12, #4
+ bgt 1b
+
+ pop {r4-r9, pc}
+
+endfunc
+.endm
+
+get_pixels4 8
+get_pixels4 10
+
+get_pixels8 8
+get_pixels8 10
+
+get_pixels12 8
+get_pixels12 10
+
+get_pixels16 8
+get_pixels16 10
+
+get_pixels24 8
+get_pixels24 10
+
+get_pixels32 8
+get_pixels32 10
+
+get_pixels48 8
+get_pixels48 10
+
+get_pixels64 8
+get_pixels64 10
diff --git a/libavcodec/arm/hevcdsp_init_arm.c b/libavcodec/arm/hevcdsp_init_arm.c
index a8b03eb356..60c211de25 100644
--- a/libavcodec/arm/hevcdsp_init_arm.c
+++ b/libavcodec/arm/hevcdsp_init_arm.c
@@ -61,6 +61,55 @@ void ff_hevc_idct_8x8_10_neon(int16_t *coeffs, int col_limit);
void ff_hevc_idct_16x16_10_neon(int16_t *coeffs, int col_limit);
void ff_hevc_idct_32x32_10_neon(int16_t *coeffs, int col_limit);
+void ff_hevc_get_pixels_4_8_neon(int16_t *dst, ptrdiff_t dststride, uint8_t *src,
+ ptrdiff_t srcstride, int height, int mx, int my,
+ int16_t *mcbuffer);
+void ff_hevc_get_pixels_4_10_neon(int16_t *dst, ptrdiff_t dststride, uint8_t *src,
+ ptrdiff_t srcstride, int height, int mx, int my,
+ int16_t *mcbuffer);
+void ff_hevc_get_pixels_8_8_neon(int16_t *dst, ptrdiff_t dststride, uint8_t *src,
+ ptrdiff_t srcstride, int height, int mx, int my,
+ int16_t *mcbuffer);
+void ff_hevc_get_pixels_8_10_neon(int16_t *dst, ptrdiff_t dststride, uint8_t *src,
+ ptrdiff_t srcstride, int height, int mx, int my,
+ int16_t *mcbuffer);
+void ff_hevc_get_pixels_12_8_neon(int16_t *dst, ptrdiff_t dststride, uint8_t *src,
+ ptrdiff_t srcstride, int height, int mx, int my,
+ int16_t *mcbuffer);
+void ff_hevc_get_pixels_12_10_neon(int16_t *dst, ptrdiff_t dststride, uint8_t *src,
+ ptrdiff_t srcstride, int height, int mx, int my,
+ int16_t *mcbuffer);
+void ff_hevc_get_pixels_16_8_neon(int16_t *dst, ptrdiff_t dststride, uint8_t *src,
+ ptrdiff_t srcstride, int height, int mx, int my,
+ int16_t *mcbuffer);
+void ff_hevc_get_pixels_16_10_neon(int16_t *dst, ptrdiff_t dststride, uint8_t *src,
+ ptrdiff_t srcstride, int height, int mx, int my,
+ int16_t *mcbuffer);
+void ff_hevc_get_pixels_24_8_neon(int16_t *dst, ptrdiff_t dststride, uint8_t *src,
+ ptrdiff_t srcstride, int height, int mx, int my,
+ int16_t *mcbuffer);
+void ff_hevc_get_pixels_24_10_neon(int16_t *dst, ptrdiff_t dststride, uint8_t *src,
+ ptrdiff_t srcstride, int height, int mx, int my,
+ int16_t *mcbuffer);
+void ff_hevc_get_pixels_32_8_neon(int16_t *dst, ptrdiff_t dststride, uint8_t *src,
+ ptrdiff_t srcstride, int height, int mx, int my,
+ int16_t *mcbuffer);
+void ff_hevc_get_pixels_32_10_neon(int16_t *dst, ptrdiff_t dststride, uint8_t *src,
+ ptrdiff_t srcstride, int height, int mx, int my,
+ int16_t *mcbuffer);
+void ff_hevc_get_pixels_48_8_neon(int16_t *dst, ptrdiff_t dststride, uint8_t *src,
+ ptrdiff_t srcstride, int height, int mx, int my,
+ int16_t *mcbuffer);
+void ff_hevc_get_pixels_48_10_neon(int16_t *dst, ptrdiff_t dststride, uint8_t *src,
+ ptrdiff_t srcstride, int height, int mx, int my,
+ int16_t *mcbuffer);
+void ff_hevc_get_pixels_64_8_neon(int16_t *dst, ptrdiff_t dststride, uint8_t *src,
+ ptrdiff_t srcstride, int height, int mx, int my,
+ int16_t *mcbuffer);
+void ff_hevc_get_pixels_64_10_neon(int16_t *dst, ptrdiff_t dststride, uint8_t *src,
+ ptrdiff_t srcstride, int height, int mx, int my,
+ int16_t *mcbuffer);
+
av_cold void ff_hevc_dsp_init_arm(HEVCDSPContext *c, int bit_depth)
{
int cpu_flags = av_get_cpu_flags();
@@ -81,6 +130,15 @@ av_cold void ff_hevc_dsp_init_arm(HEVCDSPContext *c, int bit_depth)
c->idct[1] = ff_hevc_idct_8x8_8_neon;
c->idct[2] = ff_hevc_idct_16x16_8_neon;
c->idct[3] = ff_hevc_idct_32x32_8_neon;
+
+ c->put_hevc_qpel[0][0][0] = ff_hevc_get_pixels_4_8_neon;
+ c->put_hevc_qpel[0][0][1] = ff_hevc_get_pixels_8_8_neon;
+ c->put_hevc_qpel[0][0][2] = ff_hevc_get_pixels_12_8_neon;
+ c->put_hevc_qpel[0][0][3] = ff_hevc_get_pixels_16_8_neon;
+ c->put_hevc_qpel[0][0][4] = ff_hevc_get_pixels_24_8_neon;
+ c->put_hevc_qpel[0][0][5] = ff_hevc_get_pixels_32_8_neon;
+ c->put_hevc_qpel[0][0][6] = ff_hevc_get_pixels_48_8_neon;
+ c->put_hevc_qpel[0][0][7] = ff_hevc_get_pixels_64_8_neon;
}
if (bit_depth == 10) {
c->add_residual[0] = ff_hevc_add_residual_4x4_10_neon;
@@ -97,6 +155,15 @@ av_cold void ff_hevc_dsp_init_arm(HEVCDSPContext *c, int bit_depth)
c->idct[1] = ff_hevc_idct_8x8_10_neon;
c->idct[2] = ff_hevc_idct_16x16_10_neon;
c->idct[3] = ff_hevc_idct_32x32_10_neon;
+
+ c->put_hevc_qpel[0][0][0] = ff_hevc_get_pixels_4_10_neon;
+ c->put_hevc_qpel[0][0][1] = ff_hevc_get_pixels_8_10_neon;
+ c->put_hevc_qpel[0][0][2] = ff_hevc_get_pixels_12_10_neon;
+ c->put_hevc_qpel[0][0][3] = ff_hevc_get_pixels_16_10_neon;
+ c->put_hevc_qpel[0][0][4] = ff_hevc_get_pixels_24_10_neon;
+ c->put_hevc_qpel[0][0][5] = ff_hevc_get_pixels_32_10_neon;
+ c->put_hevc_qpel[0][0][6] = ff_hevc_get_pixels_48_10_neon;
+ c->put_hevc_qpel[0][0][7] = ff_hevc_get_pixels_64_10_neon;
}
}
}
More information about the ffmpeg-cvslog
mailing list