Git Inbox Mirror of the ffmpeg-devel mailing list - see https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
 help / color / mirror / Atom feed
* [FFmpeg-devel] [PATCH] WIP: libavcodec/riscv: add RVV optimized idct_32x32_8 and idct_16x16_8 for HEVC (PR #20426)
@ 2025-09-04  3:07 CheryDan via ffmpeg-devel
  0 siblings, 0 replies; only message in thread
From: CheryDan via ffmpeg-devel @ 2025-09-04  3:07 UTC (permalink / raw)
  To: ffmpeg-devel; +Cc: CheryDan

PR #20426 opened by CheryDan
URL: https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20426
Patch URL: https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20426.patch

optimize 32x32 and 16x16 in IDCT of HEVC with RVV;
scale execution by VL.

On Banana PI F3(256-bit vectors):
    hevc_idct_16x16_8_c:                                 14159.4 ( 1.00x)
    hevc_idct_16x16_8_rvv_i32:                             644.3 (21.98x)

    hevc_idct_32x32_8_c:                                119064.2 ( 1.00x)
    hevc_idct_32x32_8_rvv_i32:                            3978.5 (29.93x)

Signed-off-by: daichengrong <daichengrong@iscas.ac.cn>


>From 47e6be21bc5b0b14a7e833d5f598385d29b4fb1e Mon Sep 17 00:00:00 2001
From: daichengrong <daichengrong@iscas.ac.cn>
Date: Thu, 4 Sep 2025 10:47:52 +0800
Subject: [PATCH] libavcodec/riscv: add RVV optimized idct_32x32_8 and
 idct_16x16_8 for HEVC

     riscv/hevcdsp_idct_rvv: Optimize idct_32x32_8 and idct_16x16_8

On Banana PI F3(256-bit vectors):
    hevc_idct_16x16_8_c:                                 14159.4 ( 1.00x)
    hevc_idct_16x16_8_rvv_i32:                             644.3 (21.98x)

    hevc_idct_32x32_8_c:                                119064.2 ( 1.00x)
    hevc_idct_32x32_8_rvv_i32:                            3978.5 (29.93x)

Signed-off-by: daichengrong <daichengrong@iscas.ac.cn>
---
 libavcodec/riscv/Makefile           |   1 +
 libavcodec/riscv/hevcdsp_idct_rvv.S | 668 ++++++++++++++++++++++++++++
 libavcodec/riscv/hevcdsp_init.c     |  41 +-
 3 files changed, 696 insertions(+), 14 deletions(-)
 create mode 100644 libavcodec/riscv/hevcdsp_idct_rvv.S

diff --git a/libavcodec/riscv/Makefile b/libavcodec/riscv/Makefile
index 736f873fe8..7b1a3f079b 100644
--- a/libavcodec/riscv/Makefile
+++ b/libavcodec/riscv/Makefile
@@ -36,6 +36,7 @@ RVV-OBJS-$(CONFIG_H264DSP) += riscv/h264addpx_rvv.o riscv/h264dsp_rvv.o \
 OBJS-$(CONFIG_H264QPEL) += riscv/h264qpel_init.o
 RVV-OBJS-$(CONFIG_H264QPEL) += riscv/h264qpel_rvv.o
 OBJS-$(CONFIG_HEVC_DECODER) += riscv/hevcdsp_init.o
+OBJS-$(CONFIG_HEVC_DECODER) += riscv/hevcdsp_idct_rvv.o
 RVV-OBJS-$(CONFIG_HEVC_DECODER)  += riscv/h26x/h2656_inter_rvv.o
 OBJS-$(CONFIG_HUFFYUV_DECODER) += riscv/huffyuvdsp_init.o
 RVV-OBJS-$(CONFIG_HUFFYUV_DECODER) += riscv/huffyuvdsp_rvv.o
diff --git a/libavcodec/riscv/hevcdsp_idct_rvv.S b/libavcodec/riscv/hevcdsp_idct_rvv.S
new file mode 100644
index 0000000000..3e3981dcab
--- /dev/null
+++ b/libavcodec/riscv/hevcdsp_idct_rvv.S
@@ -0,0 +1,668 @@
+/*
+ * Copyright (c) 2025 Institute of Software, Chinese Academy of Sciences (ISCAS).
+ *
+ * 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  lx rd, addr
+#if (__riscv_xlen == 32)
+        lw      \rd, \addr
+#elif (__riscv_xlen == 64)
+        ld      \rd, \addr
+#else
+        lq      \rd, \addr
+#endif
+.endm
+
+.macro  sx rd, addr
+#if (__riscv_xlen == 32)
+        sw      \rd, \addr
+#elif (__riscv_xlen == 64)
+        sd      \rd, \addr
+#else
+        sq      \rd, \addr
+#endif
+.endm
+
+
+.macro load_trans_4x4
+        li a2, 64
+        li a3, 83
+
+        li a5, 36
+        li a6, -64
+        li a7, -83
+.endm
+
+.macro load_trans_8x4
+        li s6, 89
+        li s7, 75
+        li s8, 50
+        li s9, 18
+
+        li s2, -89
+        li s4, -50
+        li s5, -18
+.endm
+
+.macro load_trans_16x4
+        li x12, 90
+        li x13, 87
+        li x14, 80
+        li x15, 70
+
+        li x16, 57
+        li x17, 43
+        li x18, 25
+        li x19, 9
+
+        li x20, -90
+        li x21, -87
+        li x22, -80
+        li x23, -70
+
+        li x24, -57
+        li x25, -43
+        li x26, -25
+        li x27, -9
+.endm
+
+.macro load_trans_32x4
+        li x12, 90
+        li x13, 90
+        li x14, 88
+        li x15, 85
+
+        li x16, 82
+        li x17, 78
+        li x18, 73
+        li x19, 67
+
+        li x20, 61
+        li x21, 54
+        li x22, 46
+        li x23, 38
+
+        li x24, 31
+        li x25, 22
+        li x26, 13
+        li x27, 4
+.endm
+
+.macro add_member32 in, t0, t1, t2, t3, op0, op1, op2, op3
+        .ifc \op0, -
+                neg    t0, \t0
+        .endif
+        .ifc \op1, -
+                neg    t1, \t1
+        .endif
+        .ifc \op2, -
+                neg    t2, \t2
+        .endif
+        .ifc \op3, -
+                neg    t3, \t3
+        .endif
+
+        .ifc \op0, -
+                vwmacc.vx    v24, t0, \in
+        .else
+                vwmacc.vx    v24, \t0, \in
+        .endif
+        .ifc \op1, -
+                vwmacc.vx    v25, t1, \in
+        .else
+                vwmacc.vx    v25, \t1, \in
+        .endif
+        .ifc \op2, -
+                vwmacc.vx    v26, t2, \in
+        .else
+                vwmacc.vx    v26, \t2, \in
+        .endif
+        .ifc \op3, -
+                vwmacc.vx    v27, t3, \in
+        .else
+                vwmacc.vx    v27, \t3, \in
+        .endif
+.endm
+
+.macro tr_block1
+        vwmul.vx        v24, v4, x12
+        vwmul.vx        v25, v4, x13
+        vwmul.vx        v26, v4, x14
+        vwmul.vx        v27, v4, x15
+
+        add_member32    v12, x13, x16, x19, x22,  +,  +,  +,  +
+        add_member32    v5,  x14, x19, x24, x26,  +,  +,  +,  -
+        add_member32    v13, x15, x22, x26, x19,  +,  +,  -,  -
+
+        add_member32    v6,  x16, x25, x21, x12,  +,  +,  -,  -
+        add_member32    v14, x17, x27, x16, x18,  +,  -,  -,  -
+        add_member32    v7,  x18, x24, x12, x25,  +,  -,  -,  -
+        add_member32    v15, x19, x21, x17, x23,  +,  -,  -,  +
+
+        add_member32    v16, x20, x18, x22, x16,  +,  -,  -,  +
+        add_member32    v20, x21, x15, x27, x14,  +,  -,  -,  +
+        add_member32    v17, x22, x13, x23, x21,  +,  -,  +,  +
+        add_member32    v21, x23, x14, x18, x27,  +,  -,  +,  -
+
+        add_member32    v18, x24, x17, x13, x20,  +,  -,  +,  -
+        add_member32    v22, x25, x20, x15, x13,  +,  -,  +,  -
+        add_member32    v19, x26, x23, x20, x17,  +,  -,  +,  -
+        add_member32    v23, x27, x26, x25, x24,  +,  -,  +,  -
+.endm
+
+.macro tr_block2
+        vwmul.vx        v24, v4, x16
+        vwmul.vx        v25, v4, x17
+        vwmul.vx        v26, v4, x18
+        vwmul.vx        v27, v4, x19
+
+        add_member32    v12, x25, x27, x24, x21,  +,  -,  -,  -
+        add_member32    v5,  x21, x16, x12, x17,  -,  -,  -,  -
+        add_member32    v13, x12, x18, x25, x23,  -,  -,  -,  +
+
+        add_member32    v6,  x20, x26, x17, x15,  -,  +,  +,  +
+        add_member32    v14, x26, x15, x19, x25,  +,  +,  +,  -
+        add_member32    v7,  x17, x19, x23, x12,  +,  +,  -,  -
+        add_member32    v15, x15, x25, x13, x27,  +,  -,  -,  +
+
+        add_member32    v16, x24, x14, x26, x13,  +,  -,  -,  +
+        add_member32    v20, x22, x20, x16, x26,  -,  -,  +,  +
+        add_member32    v17, x13, x24, x20, x14,  -,  +,  +,  -
+        add_member32    v21, x19, x13, x22, x24,  -,  +,  -,  -
+
+        add_member32    v18, x27, x21, x14, x16,  +,  +,  -,  +
+        add_member32    v22, x18, x23, x27, x22,  +,  -,  -,  +
+        add_member32    v19, x14, x13, x15, x18,  +,  -,  +,  -
+        add_member32    v23, x23, x22, x21, x20,  +,  -,  +,  -
+.endm
+
+.macro tr_block3
+        vwmul.vx        v24, v4, x20
+        vwmul.vx        v25, v4, x21
+        vwmul.vx        v26, v4, x22
+        vwmul.vx        v27, v4, x23
+
+        add_member32    v12, x18, x15, x12, x14,  -,  -,  -,  -
+        add_member32    v5,  x22, x27, x23, x18,  -,  -,  +,  +
+        add_member32    v13, x16, x14, x21, x27,  +,  +,  +,  -
+
+        add_member32    v6,  x24, x22, x13, x19,  +,  -,  -,  -
+        add_member32    v14, x14, x20, x24, x12,  -,  -,  +,  +
+        add_member32    v7,  x26, x16, x20, x22,  -,  +,  +,  -
+        add_member32    v15, x12, x26, x14, x24,  +,  +,  -,  -
+
+        add_member32    v16, x27, x13, x25, x15,  -,  -,  +,  +
+        add_member32    v20, x13, x23, x19, x17,  -,  +,  +,  -
+        add_member32    v17, x25, x19, x15, x26,  +,  +,  -,  +
+        add_member32    v21, x15, x17, x26, x20,  +,  -,  +,  +
+
+        add_member32    v18, x23, x25, x18, x13,  -,  -,  +,  -
+        add_member32    v22, x17, x12, x16, x21,  -,  +,  -,  +
+        add_member32    v19, x21, x24, x27, x25,  +,  -,  +,  +
+        add_member32    v23, x19, x18, x17, x16,  +,  -,  +,  -
+.endm
+
+.macro tr_block4
+        vwmul.vx        v24, v4, x24
+        vwmul.vx        v25, v4, x25
+        vwmul.vx        v26, v4, x26
+        vwmul.vx        v27, v4, x27
+
+        add_member32    v12, x17, x20, x23, x26,  -,  -,  -,  -
+        add_member32    v5,  x12, x15, x20, x25,  +,  +,  +,  +
+        add_member32    v13, x20, x12, x17, x24,  -,  -,  -,  -
+
+        add_member32    v6,  x27, x18, x14, x23,  +,  +,  +,  +
+        add_member32    v14, x21, x23, x12, x22,  +,  -,  -,  -
+        add_member32    v7,  x14, x27, x15, x21,  -,  -,  +,  +
+        add_member32    v15, x16, x22, x18, x20,  +,  +,  -,  -
+
+        add_member32    v16, x23, x17, x21, x19,  -,  -,  +,  +
+        add_member32    v20, x25, x13, x24, x18,  -,  +,  -,  -
+        add_member32    v17, x18, x16, x27, x17,  +,  -,  +,  +
+        add_member32    v21, x13, x21, x25, x16,  -,  +,  +,  -
+
+        add_member32    v18, x19, x26, x22, x15,  +,  -,  -,  +
+        add_member32    v22, x26, x24, x19, x14,  -,  -,  +,  -
+        add_member32    v19, x22, x19, x16, x13,  -,  +,  -,  +
+        add_member32    v23, x15, x14, x13, x12,  +,  -,  +,  -
+.endm
+
+.macro butterfly e, o, tmp_p, tmp_m
+        vadd.vv             \tmp_p, \e, \o
+        vsub.vv             \tmp_m, \e, \o
+.endm
+
+.macro butterfly16 out0, out1, out2, out3, out4, out5, out6, out7, in0, in1, in2, in3, in4, in5, in6, in7
+        vadd.vv              \out0, \in0, \in1
+        vsub.vv              \out1, \in0, \in1
+        vadd.vv              \out2, \in2, \in3
+        vsub.vv              \out3, \in2, \in3
+
+        vadd.vv              \out4, \in4, \in5
+        vsub.vv              \out5, \in4, \in5
+        vadd.vv              \out6, \in6, \in7
+        vsub.vv              \out7, \in6, \in7
+.endm
+
+.macro butterfly32 in0, in1, in2, in3, out0, out1, out2, out3
+        vadd.vv             \out0, \in0, \in1
+        vsub.vv             \out1, \in0, \in1
+
+        vadd.vv             \out2, \in2, \in3
+        vsub.vv             \out3, \in2, \in3
+.endm
+
+.macro load16xN_rvv
+        addi t0,  a0, 64
+        addi a2, t0, 256 * 1
+        addi a3, t0, 256 * 2
+        addi a4, t0, 256 * 3
+        vle16.v v4, (t0)
+        vle16.v v5, (a2)
+        vle16.v v6, (a3)
+        vle16.v v7, (a4)
+
+        addi a5, t0, 256 * 4
+        addi a6, t0, 256 * 5
+        addi a7, t0, 256 * 6
+        addi s9, t0, 256 * 7
+        vle16.v v16, (a5)
+        vle16.v v17, (a6)
+        vle16.v v18, (a7)
+        vle16.v v19, (s9)
+
+        addi t1,  t0, 128
+        addi s2, t1, 256 * 1
+        addi s3, t1, 256 * 2
+        addi s4, t1, 256 * 3
+        vle16.v v12, (t1)
+        vle16.v v13, (s2)
+        vle16.v v14, (s3)
+        vle16.v v15, (s4)
+
+        addi s5, t1, 256 * 4
+        addi s6, t1, 256 * 5
+        addi s7, t1, 256 * 6
+        addi s8, t1, 256 * 7
+        vle16.v v20, (s5)
+        vle16.v v21, (s6)
+        vle16.v v22, (s7)
+        vle16.v v23, (s8)
+.endm
+
+.macro load16_rvv in0, in1, in2, in3, off1, off2, step, in4, in5, in6, in7
+        addi t0,  a0, \off1
+        addi a2, t0, \step * 1
+        addi a3, t0, \step * 2
+        addi a4, t0, \step * 3
+        vle16.v \in0, (t0)
+        vle16.v \in1, (a2)
+        vle16.v \in2, (a3)
+        vle16.v \in3, (a4)
+
+        addi t1,  a0, \off2
+        addi s2, t1, \step * 1
+        addi s3, t1, \step * 2
+        addi s4, t1, \step * 3
+        vle16.v \in4, (t1)
+        vle16.v \in5, (s2)
+        vle16.v \in6, (s3)
+        vle16.v \in7, (s4)
+.endm
+
+.macro reload16 reload_offset
+        add t0, t4, \reload_offset
+        add t1, t0, t5
+        add t2, t1, t5
+        add t3, t2, t5
+
+        vle32.v        v28, (t0)
+        vle32.v        v29, (t1)
+        vle32.v        v30, (t2)
+        vle32.v        v31, (t3)
+.endm
+
+.macro scale out0, out1, out2, out3, out4, out5, out6, out7, in0, in1, in2, in3, in4, in5, in6, in7, shift
+        vnclip.wi       \out0\(), \in0\(), \shift
+        vnclip.wi       \out1\(), \in2\(), \shift
+        vnclip.wi       \out2\(), \in4\(), \shift
+        vnclip.wi       \out3\(), \in6\(), \shift
+
+        vnclip.wi       \out4\(), \in1\(), \shift
+        vnclip.wi       \out5\(), \in3\(), \shift
+        vnclip.wi       \out6\(), \in5\(), \shift
+        vnclip.wi       \out7\(), \in7\(), \shift
+.endm
+
+.macro scale_store_rvv shift, off, reload_offset
+        vsetvli         zero, zero, e32, m1, ta, ma
+        reload16        \reload_offset
+
+        butterfly32     v28, v24, v29, v25, v0, v1, v2, v3
+        butterfly32     v30, v26, v31, v27, v8, v9, v10, v11
+        
+        vsetvli         zero, zero, e16, mf2, ta, ma
+        scale           v24, v25, v26, v27,  v31, v30, v29, v28,  v0, v1, v2, v3,  v8, v9, v10, v11, \shift
+
+        li t2, 32*2
+        add t0, a1, \off
+        vssseg4e16.v v24, (t0), t2
+        
+        addi t0, a1, (64-4*2)-\off
+        vssseg4e16.v v28, (t0), t2
+.endm
+
+.macro add_member in, tt0, tt1, tt2, tt3, tt4, tt5, tt6, tt7
+        vwmacc.vx    v8, \tt0, \in
+        vwmacc.vx    v9, \tt1, \in
+        vwmacc.vx    v10, \tt2, \in
+        vwmacc.vx    v11, \tt3, \in
+
+        vwmacc.vx    v12, \tt4, \in
+        vwmacc.vx    v13, \tt5, \in
+        vwmacc.vx    v14, \tt6, \in
+        vwmacc.vx    v15, \tt7, \in
+.endm
+
+.macro tr_16xN_rvv name, shift, step
+func func_tr_16xN_\name\()_rvv, zve32x
+        load16_rvv     v0, v1, v2, v3, 0, \step * 64, \step * (2 * 64), v4, v5, v6, v7,
+        load16_rvv      v24, v25, v26, v27, \step * 32, \step * 3 * 32, \step * (2 * 64), v28, v29, v30, v31
+
+        load_trans_16x4
+
+        vwmul.vx            v8, v24, x12
+        vwmul.vx            v9, v24, x13
+        vwmul.vx            v10, v24, x14
+        vwmul.vx            v11, v24, x15
+
+        vwmul.vx            v12, v24, x16
+        vwmul.vx            v13, v24, x17
+        vwmul.vx            v14, v24, x18
+        vwmul.vx            v15, v24, x19
+
+        add_member       v28, x13, x16, x19, x25, x22, x20, x23, x26
+        add_member       v25, x14, x19, x23, x21, x26, x16, x12, x17
+        add_member       v29, x15, x25, x21, x19, x12, x18, x22, x24
+
+        add_member       v26, x16, x22, x26, x12, x27, x21, x17, x15
+        add_member       v30, x17, x20, x16, x18, x21, x15, x19, x22
+        add_member       v27, x18, x23, x12, x22, x17, x19, x24, x13
+        add_member       v31, x19, x26, x17, x24, x15, x22, x13, x20
+
+        vsetvli         zero, zero, e32, m1, ta, ma
+        vsext.vf2       v19, v0
+        vsll.vi         v28, v19, 6
+        vmv.v.v         v29, v28
+
+        vsetvli    zero, zero, e16, mf2, ta, ma
+        load_trans_4x4
+        vwmul.vx    v30, v1, a3
+        vwmul.vx    v31, v1, a5
+        vwmacc.vx    v28, a2, v2
+
+        vwmacc.vx    v29, a6, v2
+        vwmacc.vx    v30, a5, v3
+        vwmacc.vx    v31, a7, v3
+
+        load_trans_8x4
+        vwmul.vx    v20, v4, s6
+        vwmul.vx    v21, v4, s7
+        vwmul.vx    v22, v4, s8
+        vwmul.vx    v23, v4, s9
+
+        vwmacc.vx    v20, s7, v5
+        vwmacc.vx    v21, s5, v5
+        vwmacc.vx    v22, s2, v5
+        vwmacc.vx    v23, s4, v5
+
+        vwmacc.vx    v20, s8, v6
+        vwmacc.vx    v21, s2, v6
+        vwmacc.vx    v22, s9, v6
+        vwmacc.vx    v23, s7, v6
+
+        vwmacc.vx    v20, s9, v7
+        vwmacc.vx    v21, s4, v7
+        vwmacc.vx    v22, s7, v7
+        vwmacc.vx    v23, s2, v7
+
+        vsetvli    zero, zero, e32, m1, ta, ma
+        vadd.vv             v24, v28, v30
+        vadd.vv             v25, v29, v31
+        vsub.vv             v26, v29, v31
+        vsub.vv             v27, v28, v30
+
+        butterfly       v24, v20, v0, v7
+        butterfly       v25, v21, v1, v6
+        butterfly       v26, v22, v2, v5
+        butterfly       v27, v23, v3, v4
+
+        butterfly16       v16, v31, v17, v30,  v18, v29, v19, v28,     v0, v8, v1, v9,  v2, v10, v3, v11
+        butterfly16       v20, v27, v21, v26,  v22, v25, v23, v24      v4, v12, v5, v13,  v6, v14, v7, v15
+
+.if \shift > 0
+        vsetvli        zero, zero, e16, mf2, ta, ma
+        scale           v0, v1, v2, v3,  v15, v14,v13, v12,      v16, v31, v17, v30,  v18, v29, v19, v28, \shift
+
+        scale            v4, v5, v6, v7, v11, v10, v9, v8,      v20, v27, v21, v26,  v22, v25, v23, v24 , \shift
+
+        li t0, 16*2
+        addi t1, a1, 8
+        addi t2, a1, 16
+        addi t3, a1, 24
+        vssseg4e16.v v0, (a1), t0
+        vssseg4e16.v v4, (t1), t0
+
+        vssseg4e16.v v8, (t2), t0
+        vssseg4e16.v v12, (t3), t0
+.else
+        add             t1, t4, t5
+        add             t2, t1, t5
+        add             t3, t2, t5
+        vse32.v         v16, (t4)
+        vse32.v         v17, (t1)
+        vse32.v         v18, (t2)
+        vse32.v         v19, (t3)
+
+        add             t0, t3, t5
+        add             t1, t0, t5
+        add             t2, t1, t5
+        add             t3, t2, t5
+        vse32.v         v20, (t0)
+        vse32.v         v21, (t1)
+        vse32.v         v22, (t2)
+        vse32.v         v23, (t3)
+
+        add             t0, t3, t5
+        add             t1, t0, t5
+        add             t2, t1, t5
+        add             t3, t2, t5
+        vse32.v         v24, (t0)
+        vse32.v         v25, (t1)
+        vse32.v         v26, (t2)
+        vse32.v         v27, (t3)
+
+        add             t0, t3, t5
+        add             t1, t0, t5
+        add             t2, t1, t5
+        add             t3, t2, t5
+        vse32.v         v28, (t0)
+        vse32.v         v29, (t1)
+        vse32.v         v30, (t2)
+        vse32.v         v31, (t3)
+.endif
+        ret
+endfunc
+.endm
+
+tr_16xN_rvv noscale, 0, 4
+
+tr_16xN_rvv firstpass, 7, 1
+tr_16xN_rvv secondpass_8, 20 - 8, 1
+
+.macro tr_32xN_rvv name, shift
+func func_tr_32xN_\name\()_rvv, zve32x
+        vsetvli    zero, zero, e16, mf2, ta, ma
+        load16xN_rvv
+
+        load_trans_32x4
+
+        tr_block1
+        li t3, 0
+        scale_store_rvv         \shift, 0, t3
+
+        tr_block2
+        slli    t3, t5, 2
+        scale_store_rvv         \shift, 8, t3
+
+        tr_block3
+        scale_store_rvv         \shift, 16, s1
+
+        tr_block4
+        li      t0, 12
+        mul     t3, t5, t0
+        scale_store_rvv          \shift, 24, t3
+
+        ret
+endfunc
+.endm
+
+tr_32xN_rvv firstpass, 7
+tr_32xN_rvv secondpass_8, 20 - 8
+
+.macro idct_32x32 bitdepth
+func ff_hevc_idct_32x32_\bitdepth\()_rvv, zve32x
+
+        addi    sp, sp, -(__riscv_xlen / 8)*13
+        sx  ra, (__riscv_xlen / 8)*(12)(sp)
+.irp i, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
+        sx    s\i, (__riscv_xlen / 8)*(11-\i)(sp)
+.endr
+        mv t6, a0
+        csrwi   vxrm, 1
+
+        li      t0, 2048
+        sub     t4, sp, t0
+        sub     sp, t4, t0
+
+        li      t0, 32
+        vsetvli        t0, t0, e16, mf2, ta, ma
+        slli    s0, t0, 1
+        slli    t5, t0, 2
+        slli    s1, t0, 5
+
+        mv      a1, sp
+1:
+        jal     func_tr_16xN_noscale_rvv
+        jal     func_tr_32xN_firstpass_rvv
+
+        add     a0, a0, s0
+
+        slli    t0, s1, 1
+        add     a1, a1, t0
+
+        addi     a3, t6, (32 * 2)
+        bgt      a3, a0, 1b
+
+        mv      a0, sp
+        mv      a1, t6
+1:
+        jal     func_tr_16xN_noscale_rvv
+        jal     func_tr_32xN_secondpass_\bitdepth\()_rvv
+
+        add     a0, a0, s0
+
+        slli    t0, s1, 1
+        add     a1, a1, t0
+
+        addi     a3, sp, (32 * 2)
+        bgt      a3, a0, 1b
+
+        li              t0, 4096
+        add             sp,  sp,  t0
+
+        lx  ra, (__riscv_xlen / 8)*(12)(sp)
+.irp i, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
+        lx    s\i, (__riscv_xlen / 8)*(11-\i)(sp)
+.endr
+        addi    sp, sp, (__riscv_xlen / 8)*13
+        ret
+endfunc
+.endm
+
+idct_32x32 8
+
+.macro idct_16x16 bitdepth
+func ff_hevc_idct_16x16_\bitdepth\()_rvv, zve32x
+
+        addi    sp, sp, -(__riscv_xlen / 8)*13
+        sx  ra, (__riscv_xlen / 8)*(12)(sp)
+.irp i, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
+        sx    s\i, (__riscv_xlen / 8)*(11-\i)(sp)
+.endr
+        mv t6, a0
+
+        csrwi   vxrm, 1
+
+        li      t0, 512
+        sub     sp, sp, t0
+
+        # li t0, 16
+        vsetivli        t0, 16, e16, mf2, ta, ma
+
+        slli    s0, t0, 1
+        slli    t5, t0, 2
+        slli    s1, t0, 5
+
+        mv      a1, sp
+1:
+        jal     func_tr_16xN_firstpass_rvv
+
+        add     a0, a0, s0
+        add     a1, a1, s1
+
+        addi     a3, t6, (16 * 2)
+        bgt      a3, a0, 1b
+
+        mv      a0, sp
+        mv      a1, t6
+1:
+        jal     func_tr_16xN_secondpass_\bitdepth\()_rvv
+
+        add     a0, a0, s0
+
+        add     a1, a1, s1
+
+        addi     a3, sp, (16 * 2)
+        bgt      a3, a0, 1b
+
+        add             sp,  sp, 512
+
+        lx  ra, (__riscv_xlen / 8)*(12)(sp)
+.irp i, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
+        lx    s\i, (__riscv_xlen / 8)*(11-\i)(sp)
+.endr
+        addi    sp, sp, (__riscv_xlen / 8)*13
+        ret
+endfunc
+.endm
+
+idct_16x16 8
diff --git a/libavcodec/riscv/hevcdsp_init.c b/libavcodec/riscv/hevcdsp_init.c
index 70bc8ebea7..c73f784f39 100644
--- a/libavcodec/riscv/hevcdsp_init.c
+++ b/libavcodec/riscv/hevcdsp_init.c
@@ -27,6 +27,9 @@
 #include "libavcodec/hevc/dsp.h"
 #include "libavcodec/riscv/h26x/h2656dsp.h"
 
+void ff_hevc_idct_16x16_8_rvv(int16_t *coeffs, int col_limit);
+void ff_hevc_idct_32x32_8_rvv(int16_t *coeffs, int col_limit);
+
 #define RVV_FNASSIGN(member, v, h, fn, ext) \
         member[1][v][h] = ff_h2656_put_pixels_##8_##ext;  \
         member[3][v][h] = ff_h2656_put_pixels_##8_##ext;  \
@@ -40,27 +43,37 @@ void ff_hevc_dsp_init_riscv(HEVCDSPContext *c, const int bit_depth)
     const int flags = av_get_cpu_flags();
     int vlenb;
 
-    if (!(flags & AV_CPU_FLAG_RVV_I32) || !(flags & AV_CPU_FLAG_RVB))
-        return;
-
     vlenb = ff_get_rv_vlenb();
-    if (vlenb >= 32) {
+
+    if (flags & AV_CPU_FLAG_RVV_I32)
         switch (bit_depth) {
             case 8:
-                RVV_FNASSIGN(c->put_hevc_qpel, 0, 0, pel_pixels, rvv_256);
-                RVV_FNASSIGN(c->put_hevc_epel, 0, 0, pel_pixels, rvv_256);
+                c->idct[3]                     = ff_hevc_idct_32x32_8_rvv;
+                c->idct[2]                     = ff_hevc_idct_16x16_8_rvv;
                 break;
             default:
                 break;
         }
-    } else if (vlenb >= 16) {
-        switch (bit_depth) {
-            case 8:
-                RVV_FNASSIGN(c->put_hevc_qpel, 0, 0, pel_pixels, rvv_128);
-                RVV_FNASSIGN(c->put_hevc_epel, 0, 0, pel_pixels, rvv_128);
-                break;
-            default:
-                break;
+
+    if ((flags & AV_CPU_FLAG_RVV_I32) && (flags & AV_CPU_FLAG_RVB)){
+        if (vlenb >= 32) {
+            switch (bit_depth) {
+                case 8:
+                    RVV_FNASSIGN(c->put_hevc_qpel, 0, 0, pel_pixels, rvv_256);
+                    RVV_FNASSIGN(c->put_hevc_epel, 0, 0, pel_pixels, rvv_256);
+                    break;
+                default:
+                    break;
+            }
+        } else if (vlenb >= 16) {
+            switch (bit_depth) {
+                case 8:
+                    RVV_FNASSIGN(c->put_hevc_qpel, 0, 0, pel_pixels, rvv_128);
+                    RVV_FNASSIGN(c->put_hevc_epel, 0, 0, pel_pixels, rvv_128);
+                    break;
+                default:
+                    break;
+            }
         }
     }
 #endif
-- 
2.49.1

_______________________________________________
ffmpeg-devel mailing list -- ffmpeg-devel@ffmpeg.org
To unsubscribe send an email to ffmpeg-devel-leave@ffmpeg.org

^ permalink raw reply	[flat|nested] only message in thread

only message in thread, other threads:[~2025-09-04  3:07 UTC | newest]

Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2025-09-04  3:07 [FFmpeg-devel] [PATCH] WIP: libavcodec/riscv: add RVV optimized idct_32x32_8 and idct_16x16_8 for HEVC (PR #20426) CheryDan via ffmpeg-devel

Git Inbox Mirror of the ffmpeg-devel mailing list - see https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

This inbox may be cloned and mirrored by anyone:

	git clone --mirror https://master.gitmailbox.com/ffmpegdev/0 ffmpegdev/git/0.git

	# If you have public-inbox 1.1+ installed, you may
	# initialize and index your mirror using the following commands:
	public-inbox-init -V2 ffmpegdev ffmpegdev/ https://master.gitmailbox.com/ffmpegdev \
		ffmpegdev@gitmailbox.com
	public-inbox-index ffmpegdev

Example config snippet for mirrors.


AGPL code for this site: git clone https://public-inbox.org/public-inbox.git