[FFmpeg-devel] [PATCH v4] lavc/h264chroma: RISC-V V add motion compensation for 8x8 chroma blocks

Arnie Chang arnie.chang at sifive.com
Wed May 24 08:28:08 EEST 2023


Optimize the put and avg filtering for 8x8 chroma blocks

Signed-off-by: Arnie Chang <arnie.chang at sifive.com>
---
v4:
Assembly portion:
1. Fix issues raised during the code review
2. Initialize vxrm to ensure the rounding mode is as expected
Non-asm:
1. Put the function decarlations to h264_chroma_init_riscv.c

checkasm: using random seed 3191366013
RVVi32:
 - h264chroma.chroma_mc [OK]
checkasm: all 2 tests passed
avg_h264_chroma_mc1_8_c: 1821.5
avg_h264_chroma_mc1_8_rvv_i32: 482.5
put_h264_chroma_mc1_8_c: 1436.5
put_h264_chroma_mc1_8_rvv_i32: 390.5

 libavcodec/h264chroma.c                   |   2 +
 libavcodec/h264chroma.h                   |   1 +
 libavcodec/riscv/Makefile                 |   2 +
 libavcodec/riscv/h264_chroma_init_riscv.c |  41 +++
 libavcodec/riscv/h264_mc_chroma.S         | 307 ++++++++++++++++++++++
 5 files changed, 353 insertions(+)
 create mode 100644 libavcodec/riscv/h264_chroma_init_riscv.c
 create mode 100644 libavcodec/riscv/h264_mc_chroma.S

diff --git a/libavcodec/h264chroma.c b/libavcodec/h264chroma.c
index 60b86b6fba..1eeab7bc40 100644
--- a/libavcodec/h264chroma.c
+++ b/libavcodec/h264chroma.c
@@ -58,5 +58,7 @@ av_cold void ff_h264chroma_init(H264ChromaContext *c, int bit_depth)
     ff_h264chroma_init_mips(c, bit_depth);
 #elif ARCH_LOONGARCH64
     ff_h264chroma_init_loongarch(c, bit_depth);
+#elif ARCH_RISCV
+    ff_h264chroma_init_riscv(c, bit_depth);
 #endif
 }
diff --git a/libavcodec/h264chroma.h b/libavcodec/h264chroma.h
index b8f9c8f4fc..9c81c18a76 100644
--- a/libavcodec/h264chroma.h
+++ b/libavcodec/h264chroma.h
@@ -37,5 +37,6 @@ void ff_h264chroma_init_ppc(H264ChromaContext *c, int bit_depth);
 void ff_h264chroma_init_x86(H264ChromaContext *c, int bit_depth);
 void ff_h264chroma_init_mips(H264ChromaContext *c, int bit_depth);
 void ff_h264chroma_init_loongarch(H264ChromaContext *c, int bit_depth);
+void ff_h264chroma_init_riscv(H264ChromaContext *c, int bit_depth);
 
 #endif /* AVCODEC_H264CHROMA_H */
diff --git a/libavcodec/riscv/Makefile b/libavcodec/riscv/Makefile
index 965942f4df..ee17a521fd 100644
--- a/libavcodec/riscv/Makefile
+++ b/libavcodec/riscv/Makefile
@@ -10,6 +10,8 @@ OBJS-$(CONFIG_BSWAPDSP) += riscv/bswapdsp_init.o \
 RVV-OBJS-$(CONFIG_BSWAPDSP) += riscv/bswapdsp_rvv.o
 OBJS-$(CONFIG_FMTCONVERT) += riscv/fmtconvert_init.o
 RVV-OBJS-$(CONFIG_FMTCONVERT) += riscv/fmtconvert_rvv.o
+OBJS-$(CONFIG_H264CHROMA) += riscv/h264_chroma_init_riscv.o
+RVV-OBJS-$(CONFIG_H264CHROMA) += riscv/h264_mc_chroma.o
 OBJS-$(CONFIG_IDCTDSP) += riscv/idctdsp_init.o
 RVV-OBJS-$(CONFIG_IDCTDSP) += riscv/idctdsp_rvv.o
 OBJS-$(CONFIG_OPUS_DECODER) += riscv/opusdsp_init.o
diff --git a/libavcodec/riscv/h264_chroma_init_riscv.c b/libavcodec/riscv/h264_chroma_init_riscv.c
new file mode 100644
index 0000000000..7c905edfcd
--- /dev/null
+++ b/libavcodec/riscv/h264_chroma_init_riscv.c
@@ -0,0 +1,41 @@
+/*
+ * Copyright (c) 2023 SiFive, Inc. All rights reserved.
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg 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.
+ *
+ * FFmpeg 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 FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include <stdint.h>
+
+#include "libavutil/attributes.h"
+#include "libavutil/riscv/cpu.h"
+#include "libavcodec/h264chroma.h"
+#include "config.h"
+
+void h264_put_chroma_mc8_rvv(uint8_t *p_dst, const uint8_t *p_src, ptrdiff_t stride, int h, int x, int y);
+void h264_avg_chroma_mc8_rvv(uint8_t *p_dst, const uint8_t *p_src, ptrdiff_t stride, int h, int x, int y);
+
+av_cold void ff_h264chroma_init_riscv(H264ChromaContext *c, int bit_depth)
+{
+#if HAVE_RVV
+    int flags = av_get_cpu_flags();
+
+    if (bit_depth == 8 && (flags & AV_CPU_FLAG_RVV_I32) && ff_get_rv_vlenb() >= 16) {
+        c->put_h264_chroma_pixels_tab[0] = h264_put_chroma_mc8_rvv;
+        c->avg_h264_chroma_pixels_tab[0] = h264_avg_chroma_mc8_rvv;
+    }
+#endif
+}
diff --git a/libavcodec/riscv/h264_mc_chroma.S b/libavcodec/riscv/h264_mc_chroma.S
new file mode 100644
index 0000000000..9fcd2e34b3
--- /dev/null
+++ b/libavcodec/riscv/h264_mc_chroma.S
@@ -0,0 +1,307 @@
+/*
+ * Copyright (c) 2023 SiFive, Inc. All rights reserved.
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg 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.
+ *
+ * FFmpeg 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 FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+#include "libavutil/riscv/asm.S"
+
+.macro  h264_chroma_mc8 type
+func h264_\type\()_chroma_mc8_rvv, zve32x
+        csrw            vxrm, zero
+        slli            t2, a5, 3
+        mulw            t1, a5, a4
+        sh3add          a5, a4, t2
+        slli            a4, a4, 3
+        sub             a5, t1, a5
+        sub             a7, a4, t1
+        addi            a6, a5, 64
+        sub             t0, t2, t1
+        vsetivli        t3, 8, e8, m1, ta, mu
+        beqz            t1, 2f
+        blez            a3, 8f
+        li              t4, 0
+        li              t2, 0
+        li              t5, 1
+        addi            a5, t3, 1
+        slli            t3, a2, 2
+1:                                # if (xy != 0)
+        add             a4, a1, t4
+        vsetvli         zero, a5, e8, m1, ta, ma
+        addi            t2, t2, 4
+        vle8.v          v10, (a4)
+        add             a4, a4, a2
+        vslide1down.vx  v11, v10, t5
+        vsetivli        zero, 8, e8, m1, ta, ma
+        vwmulu.vx       v8, v10, a6
+        vwmaccu.vx      v8, a7, v11
+        vsetvli         zero, a5, e8, m1, ta, ma
+        vle8.v          v12, (a4)
+        vsetivli        zero, 8, e8, m1, ta, ma
+        add             a4, a4, a2
+        vwmaccu.vx      v8, t0, v12
+        vsetvli         zero, a5, e8, m1, ta, ma
+        vslide1down.vx  v13, v12, t5
+        vsetivli        zero, 8, e8, m1, ta, ma
+        vwmulu.vx       v10, v12, a6
+        vwmaccu.vx      v8, t1, v13
+        vwmaccu.vx      v10, a7, v13
+        vsetvli         zero, a5, e8, m1, ta, ma
+        vle8.v          v14, (a4)
+        vsetivli        zero, 8, e8, m1, ta, ma
+        add             a4, a4, a2
+        vwmaccu.vx      v10, t0, v14
+        vsetvli         zero, a5, e8, m1, ta, ma
+        vslide1down.vx  v15, v14, t5
+        vsetivli        zero, 8, e8, m1, ta, ma
+        vwmulu.vx       v12, v14, a6
+        vwmaccu.vx      v10, t1, v15
+        vwmaccu.vx      v12, a7, v15
+        vsetvli         zero, a5, e8, m1, ta, ma
+        vle8.v          v14, (a4)
+        vsetivli        zero, 8, e8, m1, ta, ma
+        add             a4, a4, a2
+        vwmaccu.vx      v12, t0, v14
+        vsetvli         zero, a5, e8, m1, ta, ma
+        vslide1down.vx  v15, v14, t5
+        vsetivli        zero, 8, e8, m1, ta, ma
+        vwmulu.vx       v16, v14, a6
+        vwmaccu.vx      v12, t1, v15
+        vwmaccu.vx      v16, a7, v15
+        vsetvli         zero, a5, e8, m1, ta, ma
+        vle8.v          v14, (a4)
+        vsetivli        zero, 8, e8, m1, ta, ma
+        add             a4, a0, t4
+        add             t4, t4, t3
+        vwmaccu.vx      v16, t0, v14
+        vsetvli         zero, a5, e8, m1, ta, ma
+        vslide1down.vx  v14, v14, t5
+        vsetivli        zero, 8, e8, m1, ta, ma
+        vnclipu.wi      v15, v8, 6
+        vwmaccu.vx      v16, t1, v14
+  .ifc \type,avg
+        vle8.v          v9, (a4)
+        vaaddu.vv       v15, v15, v9
+  .endif
+        vse8.v          v15, (a4)
+        add             a4, a4, a2
+        vnclipu.wi      v8, v10, 6
+  .ifc \type,avg
+        vle8.v          v9, (a4)
+        vaaddu.vv       v8, v8, v9
+  .endif
+        vse8.v          v8, (a4)
+        add             a4, a4, a2
+        vnclipu.wi      v8, v12, 6
+  .ifc \type,avg
+        vle8.v          v9, (a4)
+        vaaddu.vv       v8, v8, v9
+  .endif
+        vse8.v          v8, (a4)
+        add             a4, a4, a2
+        vnclipu.wi      v8, v16, 6
+  .ifc \type,avg
+        vle8.v          v9, (a4)
+        vaaddu.vv       v8, v8, v9
+  .endif
+        vse8.v          v8, (a4)
+        blt             t2, a3, 1b
+        j               8f
+2:
+        bnez            a4, 4f
+        beqz            t2, 4f
+        blez            a3, 8f
+        li              a4, 0
+        li              t1, 0
+        slli            a7, a2, 2
+3:                                # if ((x8 - xy) == 0 && (y8 -xy) != 0)
+        add             a5, a1, a4
+        vsetvli         zero, zero, e8, m1, ta, ma
+        addi            t1, t1, 4
+        vle8.v          v8, (a5)
+        add             a5, a5, a2
+        add             t2, a5, a2
+        vwmulu.vx       v10, v8, a6
+        vle8.v          v8, (a5)
+        vwmulu.vx       v12, v8, a6
+        vle8.v          v9, (t2)
+        add             t2, t2, a2
+        add             a5, t2, a2
+        vwmaccu.vx      v10, t0, v8
+        vle8.v          v8, (t2)
+        vle8.v          v14, (a5)
+        add             a5, a0, a4
+        add             a4, a4, a7
+        vwmaccu.vx      v12, t0, v9
+        vnclipu.wi      v15, v10, 6
+        vwmulu.vx       v10, v9, a6
+  .ifc \type,avg
+        vle8.v          v16, (a5)
+        vaaddu.vv       v15, v15, v16
+  .endif
+        vse8.v          v15, (a5)
+        add             a5, a5, a2
+        vnclipu.wi      v9, v12, 6
+        vwmaccu.vx      v10, t0, v8
+        vwmulu.vx       v12, v8, a6
+  .ifc \type,avg
+        vle8.v          v16, (a5)
+        vaaddu.vv       v9, v9, v16
+  .endif
+        vse8.v          v9, (a5)
+        add             a5, a5, a2
+        vnclipu.wi      v8, v10, 6
+        vwmaccu.vx      v12, t0, v14
+  .ifc \type,avg
+        vle8.v          v16, (a5)
+        vaaddu.vv       v8, v8, v16
+  .endif
+        vse8.v          v8, (a5)
+        add             a5, a5, a2
+        vnclipu.wi      v8, v12, 6
+  .ifc \type,avg
+        vle8.v          v16, (a5)
+        vaaddu.vv       v8, v8, v16
+  .endif
+        vse8.v          v8, (a5)
+        blt             t1, a3, 3b
+        j               8f
+4:
+        beqz            a4, 6f
+        bnez            t2, 6f
+        blez            a3, 8f
+        li              a4, 0
+        li              t2, 0
+        addi            t0, t3, 1
+        slli            t1, a2, 2
+5:                               # if ((x8 - xy) != 0 && (y8 -xy) == 0)
+        add             a5, a1, a4
+        vsetvli         zero, t0, e8, m1, ta, ma
+        addi            t2, t2, 4
+        vle8.v          v8, (a5)
+        add             a5, a5, a2
+        vslide1down.vx  v9, v8, t5
+        vsetivli        zero, 8, e8, m1, ta, ma
+        vwmulu.vx       v10, v8, a6
+        vwmaccu.vx      v10, a7, v9
+        vsetvli         zero, t0, e8, m1, ta, ma
+        vle8.v          v8, (a5)
+        add             a5, a5, a2
+        vslide1down.vx  v9, v8, t5
+        vsetivli        zero, 8, e8, m1, ta, ma
+        vwmulu.vx       v12, v8, a6
+        vwmaccu.vx      v12, a7, v9
+        vsetvli         zero, t0, e8, m1, ta, ma
+        vle8.v          v8, (a5)
+        add             a5, a5, a2
+        vslide1down.vx  v9, v8, t5
+        vsetivli        zero, 8, e8, m1, ta, ma
+        vwmulu.vx       v14, v8, a6
+        vwmaccu.vx      v14, a7, v9
+        vsetvli         zero, t0, e8, m1, ta, ma
+        vle8.v          v8, (a5)
+        add             a5, a0, a4
+        add             a4, a4, t1
+        vslide1down.vx  v9, v8, t5
+        vsetivli        zero, 8, e8, m1, ta, ma
+        vnclipu.wi      v16, v10, 6
+  .ifc \type,avg
+        vle8.v          v18, (a5)
+        vaaddu.vv       v16, v16, v18
+  .endif
+        vse8.v          v16, (a5)
+        add             a5, a5, a2
+        vnclipu.wi      v10, v12, 6
+        vwmulu.vx       v12, v8, a6
+  .ifc \type,avg
+        vle8.v          v18, (a5)
+        vaaddu.vv       v10, v10, v18
+  .endif
+        vse8.v          v10, (a5)
+        add             a5, a5, a2
+        vnclipu.wi      v8, v14, 6
+        vwmaccu.vx      v12, a7, v9
+  .ifc \type,avg
+        vle8.v          v18, (a5)
+        vaaddu.vv       v8, v8, v18
+  .endif
+        vse8.v          v8, (a5)
+        add             a5, a5, a2
+        vnclipu.wi      v8, v12, 6
+  .ifc \type,avg
+        vle8.v          v18, (a5)
+        vaaddu.vv       v8, v8, v18
+  .endif
+        vse8.v          v8, (a5)
+        blt             t2, a3, 5b
+        j               8f
+6:
+        blez            a3, 8f
+        li              a4, 0
+        li              t2, 0
+        slli            a7, a2, 2
+7:                               # the final else, none of the above conditions are met
+        add             t0, a1, a4
+        vsetvli         zero, zero, e8, m1, ta, ma
+        add             a5, a0, a4
+        add             a4, a4, a7
+        addi            t2, t2, 4
+        vle8.v          v8, (t0)
+        add             t0, t0, a2
+        add             t1, t0, a2
+        vwmulu.vx       v10, v8, a6
+        vle8.v          v8, (t0)
+        add             t0, t1, a2
+        vle8.v          v9, (t1)
+        vle8.v          v12, (t0)
+        vnclipu.wi      v13, v10, 6
+        vwmulu.vx       v10, v8, a6
+  .ifc \type,avg
+        vle8.v          v18, (a5)
+        vaaddu.vv       v13, v13, v18
+  .endif
+        vse8.v          v13, (a5)
+        add             a5, a5, a2
+        vnclipu.wi      v8, v10, 6
+        vwmulu.vx       v10, v9, a6
+  .ifc \type,avg
+        vle8.v          v18, (a5)
+        vaaddu.vv       v8, v8, v18
+  .endif
+        vse8.v          v8, (a5)
+        add             a5, a5, a2
+        vnclipu.wi      v8, v10, 6
+        vwmulu.vx       v10, v12, a6
+  .ifc \type,avg
+        vle8.v          v18, (a5)
+        vaaddu.vv       v8, v8, v18
+  .endif
+        vse8.v          v8, (a5)
+        add             a5, a5, a2
+        vnclipu.wi      v8, v10, 6
+  .ifc \type,avg
+        vle8.v          v18, (a5)
+        vaaddu.vv       v8, v8, v18
+  .endif
+        vse8.v          v8, (a5)
+        blt             t2, a3, 7b
+8:
+        ret
+endfunc
+.endm
+
+h264_chroma_mc8 put
+h264_chroma_mc8 avg
-- 
2.17.1



More information about the ffmpeg-devel mailing list