[FFmpeg-devel] [PATCH] libavcodec/riscv:add RVV optimized for idct_32x32_8:
daichengrong at iscas.ac.cn
daichengrong at iscas.ac.cn
Tue Apr 15 10:34:24 EEST 2025
From: daichengrong <daichengrong at iscas.ac.cn>
riscv/hevcdsp_idct_rvv: Optimize idct_32x32_8
On Banana PI F3:
hevc_idct_32x32_8_c: 119579.3 ( 1.00x)
hevc_idct_32x32_8_rvv_i64: 51254.4 ( 2.33x)
Signed-off-by: daichengrong <daichengrong at iscas.ac.cn>
---
libavcodec/riscv/Makefile | 1 +
libavcodec/riscv/hevcdsp_idct_rvv.S | 1042 +++++++++++++++++++++++++++
libavcodec/riscv/hevcdsp_init.c | 52 +-
3 files changed, 1075 insertions(+), 20 deletions(-)
create mode 100644 libavcodec/riscv/hevcdsp_idct_rvv.S
diff --git a/libavcodec/riscv/Makefile b/libavcodec/riscv/Makefile
index a80d2fa2e7..dfc33afbee 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..f8dd2e5bf4
--- /dev/null
+++ b/libavcodec/riscv/hevcdsp_idct_rvv.S
@@ -0,0 +1,1042 @@
+/*
+ * Copyright (c) 2025 Institue 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"
+
+const trans, align=4
+ .2byte 64, 83, 64, 36
+ .2byte 89, 75, 50, 18
+ .2byte 90, 87, 80, 70
+ .2byte 57, 43, 25, 9
+ .2byte 90, 90, 88, 85
+ .2byte 82, 78, 73, 67
+ .2byte 61, 54, 46, 38
+ .2byte 31, 22, 13, 4
+endconst
+
+.macro sum_sub out, in, c, op, p
+ vsetivli t0, 4, e16, mf2, tu, ma
+ .ifc \op, +
+ .ifc \p, 2
+ vslidedown.vi v8, \in, 4
+ vwmacc.vx \out, \c, v8
+ .else
+ vwmacc.vx \out, \c, \in
+ .endif
+ .else
+ .ifc \p, 2
+ neg \c, \c
+ vslidedown.vi v8, \in, 4
+ vwmacc.vx \out, \c, v8
+ neg \c, \c
+ .else
+ neg \c, \c
+ vwmacc.vx \out, \c, \in
+ neg \c, \c
+ .endif
+ .endif
+.endm
+
+.macro add_member32 in, t0, index0, t1, index1, t2, index2, t3, index3, op0, op1, op2, op3, p
+ vsetivli t0, 1, e16, m1, tu, ma
+ vslidedown.vi v12, \t0, \index0
+ vmv.x.s s2, v12
+ vslidedown.vi v12, \t1, \index1
+ vmv.x.s s3, v12
+ vslidedown.vi v12, \t2, \index2
+ vmv.x.s s4, v12
+ vslidedown.vi v12, \t3, \index3
+ vmv.x.s s5, v12
+
+ sum_sub v24, \in, s2, \op0, \p
+ sum_sub v25, \in, s3, \op1, \p
+ sum_sub v26, \in, s4, \op2, \p
+ sum_sub v27, \in, s5, \op3, \p
+.endm
+
+.macro butterfly e, o, tmp_p, tmp_m
+ vsetivli t0, 4, e32, m1, tu, ma
+ vadd.vv \tmp_p, \e, \o
+ vsub.vv \tmp_m, \e, \o
+.endm
+
+.macro butterfly16 in0, in1, in2, in3, in4, in5, in6, in7
+ vsetivli t0, 4, e32, m1, tu, ma
+ vadd.vv v20, \in0, \in1
+ vsub.vv \in0, \in0, \in1
+ vadd.vv \in1, \in2, \in3
+ vsub.vv \in2, \in2, \in3
+ vadd.vv \in3, \in4, \in5
+ vsub.vv \in4, \in4, \in5
+ vadd.vv \in5, \in6, \in7
+ vsub.vv \in6, \in6, \in7
+.endm
+
+.macro multiply in
+ vsetivli t0, 1, e16, m1, tu, ma
+ vmv.x.s s2, \in
+ vslidedown.vi v12, \in, 1
+ vmv.x.s s3, v12
+ vslidedown.vi v12, \in, 2
+ vmv.x.s s4, v12
+ vslidedown.vi v12, \in, 3
+ vmv.x.s s5, v12
+
+ vsetivli t0, 4, e16, mf2, tu, ma
+ vwmul.vx v24, v4, s2
+ vwmul.vx v25, v4, s3
+ vwmul.vx v26, v4, s4
+ vwmul.vx v27, v4, s5
+.endm
+
+func tr_block1, zve64x
+ multiply v0
+ add_member32 v4, v0, 1, v1, 0, v1, 3, v2, 2, +, +, +, +, 2
+ add_member32 v5, v0, 2, v1, 3, v3, 0, v3, 2, +, +, +, -
+ add_member32 v5, v0, 3, v2, 2, v3, 2, v1, 3, +, +, -, -, 2
+ add_member32 v6, v1, 0, v3, 1, v2, 1, v0, 0, +, +, -, -
+ add_member32 v6, v1, 1, v3, 3, v1, 0, v1, 2, +, -, -, -, 2
+ add_member32 v7, v1, 2, v3, 0, v0, 0, v3, 1, +, -, -, -
+ add_member32 v7, v1, 3, v2, 1, v1, 1, v2, 3, +, -, -, +, 2
+ add_member32 v16, v2, 0, v1, 2, v2, 2, v1, 0, +, -, -, +
+ add_member32 v16, v2, 1, v0, 3, v3, 3, v0, 2, +, -, -, +, 2
+ add_member32 v17, v2, 2, v0, 1, v2, 3, v2, 1, +, -, +, +
+ add_member32 v17, v2, 3, v0, 2, v1, 2, v3, 3, +, -, +, -, 2
+ add_member32 v18, v3, 0, v1, 1, v0, 1, v2, 0, +, -, +, -
+ add_member32 v18, v3, 1, v2, 0, v0, 3, v0, 1, +, -, +, -, 2
+ add_member32 v19, v3, 2, v2, 3, v2, 0, v1, 1, +, -, +, -
+ add_member32 v19, v3, 3, v3, 2, v3, 1, v3, 0, +, -, +, -, 2
+ ret
+endfunc
+
+func tr_block2, zve64x
+ multiply v1
+ add_member32 v4, v3, 1, v3, 3, v3, 0, v2, 1, +, -, -, -, 2
+ add_member32 v5, v2, 1, v1, 0, v0, 0, v1, 1, -, -, -, -
+ add_member32 v5, v0, 0, v1, 2, v3, 1, v2, 3, -, -, -, +, 2
+ add_member32 v6, v2, 0, v3, 2, v1, 1, v0, 3, -, +, +, +
+ add_member32 v6, v3, 2, v0, 3, v1, 3, v3, 1, +, +, +, -, 2
+ add_member32 v7, v1, 1, v1, 3, v2, 3, v0, 0, +, +, -, -
+ add_member32 v7, v0, 3, v3, 1, v0, 1, v3, 3, +, -, -, +, 2
+ add_member32 v16, v3, 0, v0, 2, v3, 2, v0, 1, +, -, -, +
+ add_member32 v16, v2, 2, v2, 0, v1, 0, v3, 2, -, -, +, +, 2
+ add_member32 v17, v0, 1, v3, 0, v2, 0, v0, 2, -, +, +, -
+ add_member32 v17, v1, 3, v0, 1, v2, 2, v3, 0, -, +, -, -, 2
+ add_member32 v18, v3, 3, v2, 1, v0, 2, v1, 0, +, +, -, +
+ add_member32 v18, v1, 2, v2, 3, v3, 3, v2, 2, +, -, -, +, 2
+ add_member32 v19, v0, 2, v0, 1, v0, 3, v1, 2, +, -, +, -
+ add_member32 v19, v2, 3, v2, 2, v2, 1, v2, 0, +, -, +, -, 2
+ ret
+endfunc
+
+func tr_block3, zve64x
+ multiply v2
+ add_member32 v4, v1, 2, v0, 3, v0, 0, v0, 2, -, -, -, -, 2
+ add_member32 v5, v2, 2, v3, 3, v2, 3, v1, 2, -, -, +, +
+ add_member32 v5, v1, 0, v0, 2, v2, 1, v3, 3, +, +, +, -, 2
+ add_member32 v6, v3, 0, v2, 2, v0, 1, v1, 3, +, -, -, -
+ add_member32 v6, v0, 2, v2, 0, v3, 0, v0, 0, -, -, +, +, 2
+ add_member32 v7, v3, 2, v1, 0, v2, 0, v2, 2, -, +, +, -
+ add_member32 v7, v0, 0, v3, 2, v0, 2, v3, 0, +, +, -, -, 2
+ add_member32 v16, v3, 3, v0, 1, v3, 1, v0, 3, -, -, +, +
+ add_member32 v16, v0, 1, v2, 3, v1, 3, v1, 1, -, +, +, -, 2
+ add_member32 v17, v3, 1, v1, 3, v0, 3, v3, 2, +, +, -, +
+ add_member32 v17, v0, 3, v1, 1, v3, 2, v2, 0, +, -, +, +, 2
+ add_member32 v18, v2, 3, v3, 1, v1, 2, v0, 1, -, -, +, -
+ add_member32 v18, v1, 1, v0, 0, v1, 0, v2, 1, -, +, -, +, 2
+ add_member32 v19, v2, 1, v3, 0, v3, 3, v3, 1, +, -, +, +
+ add_member32 v19, v1, 3, v1, 2, v1, 1, v1, 0, +, -, +, -, 2
+ ret
+endfunc
+
+func tr_block4, zve64x
+ multiply v3
+ add_member32 v4, v1, 1, v2, 0, v2, 3, v3, 2, -, -, -, -, 2
+ add_member32 v5, v0, 0, v0, 3, v2, 0, v3, 1, +, +, +, +
+ add_member32 v5, v2, 0, v0, 0, v1, 1, v3, 0, -, -, -, -, 2
+ add_member32 v6, v3, 3, v1, 2, v0, 2, v2, 3, +, +, +, +
+ add_member32 v6, v2, 1, v2, 3, v0, 0, v2, 2, +, -, -, -, 2
+ add_member32 v7, v0, 2, v3, 3, v0, 3, v2, 1, -, -, +, +
+ add_member32 v7, v1, 0, v2, 2, v1, 2, v2, 0, +, +, -, -, 2
+ add_member32 v16, v2, 3, v1, 1, v2, 1, v1, 3, -, -, +, +
+ add_member32 v16, v3, 1, v0, 1, v3, 0, v1, 2, -, +, -, -, 2
+ add_member32 v17, v1, 2, v1, 0, v3, 3, v1, 1, +, -, +, +
+ add_member32 v17, v0, 1, v2, 1, v3, 1, v1, 0, -, +, +, -, 2
+ add_member32 v18, v1, 3, v3, 2, v2, 2, v0, 3, +, -, -, +
+ add_member32 v18, v3, 2, v3, 0, v1, 3, v0, 2, -, -, +, -, 2
+ add_member32 v19, v2, 2, v1, 3, v1, 0, v0, 1, -, +, -, +
+ add_member32 v19, v0, 3, v0, 2, v0, 1, v0, 0, +, -, +, -, 2
+ ret
+endfunc
+
+.macro butterfly32 in0, in1, in2, in3, out
+ vsetivli t0, 4, e32, m1, tu, ma
+ vadd.vv \out, \in0, \in1
+ vsub.vv \in0, \in0, \in1
+ vadd.vv \in1, \in2, \in3
+ vsub.vv \in2, \in2, \in3
+.endm
+
+.macro load16 in0, in1, in2, in3
+ vsetivli t0, 1, e64, m1, tu, ma
+ vle64.v v8, (a3)
+
+ vsetivli t0, 2, e64, m1, tu, ma
+ vslide1up.vx \in0, v8, zero
+ vsetivli t0, 1, e64, m1, tu, ma
+ vle64.v \in0, (a1)
+ add a1, a1, a2
+ add a3, a3, a2
+
+ vsetivli t0, 1, e64, m1, tu, ma
+ vle64.v v8, (a3)
+ vsetivli t0, 2, e64, m1, tu, ma
+ vslide1up.vx \in1, v8, zero
+ vsetivli t0, 1, e64, m1, tu, ma
+ vle64.v \in1, (a1)
+ add a1, a1, a2
+ add a3, a3, a2
+
+ vsetivli t0, 1, e64, m1, tu, ma
+ vle64.v v8, (a3)
+ vsetivli t0, 2, e64, m1, tu, ma
+ vslide1up.vx \in2, v8, zero
+ vsetivli t0, 1, e64, m1, tu, ma
+ vle64.v \in2, (a1)
+ add a1, a1, a2
+ add a3, a3, a2
+
+ vsetivli t0, 1, e64, m1, tu, ma
+ vle64.v v8, (a3)
+ vsetivli t0, 2, e64, m1, tu, ma
+ vslide1up.vx \in3, v8, zero
+ vsetivli t0, 1, e64, m1, tu, ma
+ vle64.v \in3, (a1)
+ add a1, a1, a2
+ add a3, a3, a2
+.endm
+
+.macro store16 in0, in1, in2, in3, rx
+ vsetivli t0, 1, e64, m1, tu, ma
+ vse64.v \in0, (a1)
+ vsetivli t0, 2, e64, m1, tu, ma
+ vslide1down.vx v8, \in0, zero
+ vsetivli t0, 1, e64, m1, tu, ma
+ vse64.v v8, (a3)
+ add a1, a1, a2
+ add a3, a3, \rx
+
+ vsetivli t0, 1, e64, m1, tu, ma
+ vse64.v \in1, (a1)
+ vsetivli t0, 2, e64, m1, tu, ma
+ vslide1down.vx v8, \in1, zero
+ vsetivli t0, 1, e64, m1, tu, ma
+ vse64.v v8, (a3)
+ add a1, a1, a2
+ add a3, a3, \rx
+
+ vsetivli t0, 1, e64, m1, tu, ma
+ vse64.v \in2, (a1)
+ vsetivli t0, 2, e64, m1, tu, ma
+ vslide1down.vx v8, \in2, zero
+ vsetivli t0, 1, e64, m1, tu, ma
+ vse64.v v8, (a3)
+ add a1, a1, a2
+ add a3, a3, \rx
+
+ vsetivli t0, 1, e64, m1, tu, ma
+ vse64.v \in3, (a1)
+ vsetivli t0, 2, e64, m1, tu, ma
+ vslide1down.vx v8, \in3, zero
+ vsetivli t0, 1, e64, m1, tu, ma
+ vse64.v v8, (a3)
+ add a1, a1, a2
+ add a3, a3, \rx
+
+.endm
+
+.macro load32
+ addi a1, a5, 64
+ addi a3, a1, 128
+ li a2, 256
+
+ vsetivli t0, 1, e64, m1, tu, ma
+ vle64.v v8, (a3)
+ vsetivli t0, 2, e64, m1, tu, ma
+ vslide1up.vx v4, v8, zero
+ vsetivli t0, 1, e64, m1, tu, ma
+ vle64.v v4, (a1)
+ add a1, a1, a2
+ add a3, a3, a2
+
+ vsetivli t0, 1, e64, m1, tu, ma
+ vle64.v v8, (a3)
+ vsetivli t0, 2, e64, m1, tu, ma
+ vslide1up.vx v5, v8, zero
+ vsetivli t0, 1, e64, m1, tu, ma
+ vle64.v v5, (a1)
+ add a1, a1, a2
+ add a3, a3, a2
+
+ vsetivli t0, 1, e64, m1, tu, ma
+ vle64.v v8, (a3)
+ vsetivli t0, 2, e64, m1, tu, ma
+ vslide1up.vx v6, v8, zero
+ vsetivli t0, 1, e64, m1, tu, ma
+ vle64.v v6, (a1)
+ add a1, a1, a2
+ add a3, a3, a2
+
+ vsetivli t0, 1, e64, m1, tu, ma
+ vle64.v v8, (a3)
+ vsetivli t0, 2, e64, m1, tu, ma
+ vslide1up.vx v7, v8, zero
+ vsetivli t0, 1, e64, m1, tu, ma
+ vle64.v v7, (a1)
+ add a1, a1, a2
+ add a3, a3, a2
+
+ vsetivli t0, 1, e64, m1, tu, ma
+ vle64.v v8, (a3)
+ vsetivli t0, 2, e64, m1, tu, ma
+ vslide1up.vx v16, v8, zero
+ vsetivli t0, 1, e64, m1, tu, ma
+ vle64.v v16, (a1)
+ add a1, a1, a2
+ add a3, a3, a2
+
+ vsetivli t0, 1, e64, m1, tu, ma
+ vle64.v v8, (a3)
+ vsetivli t0, 2, e64, m1, tu, ma
+ vslide1up.vx v17, v8, zero
+ vsetivli t0, 1, e64, m1, tu, ma
+ vle64.v v17, (a1)
+ add a1, a1, a2
+ add a3, a3, a2
+
+ vsetivli t0, 1, e64, m1, tu, ma
+ vle64.v v8, (a3)
+ vsetivli t0, 2, e64, m1, tu, ma
+ vslide1up.vx v18, v8, zero
+ vsetivli t0, 1, e64, m1, tu, ma
+ vle64.v v18, (a1)
+ add a1, a1, a2
+ add a3, a3, a2
+
+ vsetivli t0, 1, e64, m1, tu, ma
+ vle64.v v8, (a3)
+ vsetivli t0, 2, e64, m1, tu, ma
+ vslide1up.vx v19, v8, zero
+ vsetivli t0, 1, e64, m1, tu, ma
+ vle64.v v19, (a1)
+ add a1, a1, a2
+ add a3, a3, a2
+
+.endm
+
+
+
+.macro add_member in, tt0, tt1, tt2, tt3, tt4, tt5, tt6, tt7, op0, op1, op2, op3, op4, op5, op6, op7, p
+ sum_sub v21, \in, \tt0, \op0, \p
+ sum_sub v22, \in, \tt1, \op1, \p
+ sum_sub v23, \in, \tt2, \op2, \p
+ sum_sub v24, \in, \tt3, \op3, \p
+ sum_sub v25, \in, \tt4, \op4, \p
+ sum_sub v26, \in, \tt5, \op5, \p
+ sum_sub v27, \in, \tt6, \op6, \p
+ sum_sub v28, \in, \tt7, \op7, \p
+.endm
+
+.macro scale_store shift
+ vsetivli t0, 8, e16, m1, tu, ma
+ vle16.v v28, (a4)
+ addi a4, a4, 2*8
+ vle16.v v29, (a4)
+ addi a4, a4, 2*8
+ vle16.v v30, (a4)
+ addi a4, a4, 2*8
+ vle16.v v31, (a4)
+ addi a4, a4, 2*8
+
+ butterfly32 v28, v24, v29, v25, v2
+ butterfly32 v30, v26, v31, v27, v3
+
+ scale v20, v21, v22, v23, v2, v28, v24, v29, v3, v30, v26, v31, \shift
+
+ transpose16_4x4_2 v20, v21, v22, v23, v24, v25, v26, v27, v28, v29
+
+ store16 v20, v21, v22, v23, t1
+
+ vsetivli t0, 4, e16, m1, tu, ma
+ vle16.v v2, (t2)
+ addi t2, t2, 8
+ vle16.v v3, (t2)
+ addi t2, t2, -8
+.endm
+
+.macro store_to_stack off1, off2, in0, in2, in4, in6, in7, in5, in3, in1
+ li a7, \off1
+ add a1, sp, a7
+ li a7, \off2
+ add a3, sp, a7
+ li a2, -16
+ li a4, 16
+
+ vsetivli t0, 4, e32, m1, tu, ma
+ vse32.v \in0, (a1)
+ add a1, a1, a4
+ vse32.v \in1, (a3)
+ add a3, a3, a2
+ vse32.v \in2, (a1)
+ add a1, a1, a4
+ vse32.v \in3, (a3)
+ add a3, a3, a2
+ vse32.v \in4, (a1)
+ add a1, a1, a4
+ vse32.v \in5, (a3)
+ add a3, a3, a2
+ vse32.v \in6, (a1)
+ vse32.v \in7, (a3)
+.endm
+
+.macro transpose16_4x4_2 r0, r1, r2, r3, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5
+ // lower halves
+ vsetivli t0, 1, e16, m1, tu, ma
+ vmv.v.v \tmp0\(), \r0\()
+ vsetivli t0, 2, e16, m1, tu, ma
+ vslideup.vi \tmp0\(), \r1\(), 1
+ vslidedown.vi v8, \r0\(), 2
+ vsetivli t0, 3, e16, m1, tu, ma
+ vslideup.vi \tmp0\(), v8, 2
+ vslidedown.vi v8, \r1\(), 2
+ vsetivli t0, 4, e16, m1, tu, ma
+ vslideup.vi \tmp0\(), v8, 3
+ vsetivli t0, 1, e16, m1, tu, ma
+ vslidedown.vi v8, \r0\(), 1
+ vmv.v.v \tmp1\(), v8
+ vslidedown.vi v8, \r1\(), 1
+ vsetivli t0, 2, e16, m1, tu, ma
+ vslideup.vi \tmp1\(), v8, 1
+ vslidedown.vi v8, \r0\(), 3
+ vsetivli t0, 3, e16, m1, tu, ma
+ vslideup.vi \tmp1\(), v8, 2
+ vslidedown.vi v8, \r1\(), 3
+ vsetivli t0, 4, e16, m1, tu, ma
+ vslideup.vi \tmp1\(), v8, 3
+ vsetivli t0, 1, e16, m1, tu, ma
+ vmv.v.v \tmp2\(), \r2\()
+ vsetivli t0, 2, e16, m1, tu, ma
+ vslideup.vi \tmp2\(), \r3\(), 1
+ vslidedown.vi v8, \r2\(), 2
+ vsetivli t0, 3, e16, m1, tu, ma
+ vslideup.vi \tmp2\(), v8, 2
+ vslidedown.vi v8, \r3\(), 2
+ vsetivli t0, 4, e16, m1, tu, ma
+ vslideup.vi \tmp2\(), v8, 3
+ vsetivli t0, 1, e16, m1, tu, ma
+ vslidedown.vi v8, \r2\(), 1
+ vmv.v.v \tmp3\(), v8
+ vslidedown.vi v8, \r3\(), 1
+ vsetivli t0, 2, e16, m1, tu, ma
+ vslideup.vi \tmp3\(), v8, 1
+ vslidedown.vi v8, \r2\(), 3
+ vsetivli t0, 3, e16, m1, tu, ma
+ vslideup.vi \tmp3\(), v8, 2
+ vslidedown.vi v8, \r3\(), 3
+ vsetivli t0, 4, e16, m1, tu, ma
+ vslideup.vi \tmp3\(), v8, 3
+ vsetivli t0, 1, e32, m1, tu, ma
+ vmv.v.v \tmp4\(), \tmp0\()
+ vsetivli t0, 2, e32, m1, tu, ma
+ vslideup.vi \tmp4\(), \tmp2\(), 1
+ vsetivli t0, 1, e32, m1, tu, ma
+ vslidedown.vi v8, \tmp0\(), 1
+ vmv.v.v \tmp5\(), v8
+ vslidedown.vi v8, \tmp2\(), 1
+ vsetivli t0, 2, e32, m1, tu, ma
+ vslideup.vi \tmp5\(), v8, 1
+ vsetivli t0, 1, e32, m1, tu, ma
+ vmv.v.v \tmp0\(), \tmp1\()
+ vsetivli t0, 2, e32, m1, tu, ma
+ vslideup.vi \tmp0\(), \tmp3\(), 1
+ vsetivli t0, 1, e32, m1, tu, ma
+ vslidedown.vi v8, \tmp1\(), 1
+ vmv.v.v \tmp2\(), v8
+ vslidedown.vi v8, \tmp3\(), 1
+ vsetivli t0, 2, e32, m1, tu, ma
+ vslideup.vi \tmp2\(), v8, 1
+
+ vsetivli t0, 1, e64, m1, tu, ma
+ vmv.v.v \r0\(), \tmp4\()
+ vmv.v.v \r2\(), \tmp5\()
+ vmv.v.v \r1\(), \tmp0\()
+ vmv.v.v \r3\(), \tmp2\()
+
+ vsetivli t0, 1, e16, m1, tu, ma
+ vmv.v.v \tmp0\(), \r3\()
+ vsetivli t0, 2, e16, m1, tu, ma
+ vslideup.vi \tmp0\(), \r2\(), 1
+ vslidedown.vi v8, \r3\(), 2
+ vsetivli t0, 3, e16, m1, tu, ma
+ vslideup.vi \tmp0\(), v8, 2
+ vslidedown.vi v8, \r2\(), 2
+ vsetivli t0, 4, e16, m1, tu, ma
+ vslideup.vi \tmp0\(), v8, 3
+ vslidedown.vi v8, \r3\(), 4
+ vsetivli t0, 5, e16, m1, tu, ma
+ vslideup.vi \tmp0\(), v8, 4
+ vslidedown.vi v8, \r2\(), 4
+ vsetivli t0, 6, e16, m1, tu, ma
+ vslideup.vi \tmp0\(), v8, 5
+ vslidedown.vi v8, \r3\(), 6
+ vsetivli t0, 7, e16, m1, tu, ma
+ vslideup.vi \tmp0\(), v8, 6
+ vslidedown.vi v8, \r2\(), 6
+ vsetivli t0, 8, e16, m1, tu, ma
+ vslideup.vi \tmp0\(), v8, 7
+ vsetivli t0, 1, e16, m1, tu, ma
+ vslidedown.vi v8, \r3\(), 1
+ vmv.v.v \tmp1\(), v8
+ vslidedown.vi v8, \r2\(), 1
+ vsetivli t0, 2, e16, m1, tu, ma
+ vslideup.vi \tmp1\(), v8, 1
+ vslidedown.vi v8, \r3\(), 3
+ vsetivli t0, 3, e16, m1, tu, ma
+ vslideup.vi \tmp1\(), v8, 2
+ vslidedown.vi v8, \r2\(), 3
+ vsetivli t0, 4, e16, m1, tu, ma
+ vslideup.vi \tmp1\(), v8, 3
+ vslidedown.vi v8, \r3\(), 5
+ vsetivli t0, 5, e16, m1, tu, ma
+ vslideup.vi \tmp1\(), v8, 4
+ vslidedown.vi v8, \r2\(), 5
+ vsetivli t0, 6, e16, m1, tu, ma
+ vslideup.vi \tmp1\(), v8, 5
+ vslidedown.vi v8, \r3\(), 7
+ vsetivli t0, 7, e16, m1, tu, ma
+ vslideup.vi \tmp1\(), v8, 6
+ vslidedown.vi v8, \r2\(), 7
+ vsetivli t0, 8, e16, m1, tu, ma
+ vslideup.vi \tmp1\(), v8, 7
+ vsetivli t0, 1, e16, m1, tu, ma
+ vmv.v.v \tmp2\(), \r1\()
+ vsetivli t0, 2, e16, m1, tu, ma
+ vslideup.vi \tmp2\(), \r0\(), 1
+ vslidedown.vi v8, \r1\(), 2
+ vsetivli t0, 3, e16, m1, tu, ma
+ vslideup.vi \tmp2\(), v8, 2
+ vslidedown.vi v8, \r0\(), 2
+ vsetivli t0, 4, e16, m1, tu, ma
+ vslideup.vi \tmp2\(), v8, 3
+ vslidedown.vi v8, \r1\(), 4
+ vsetivli t0, 5, e16, m1, tu, ma
+ vslideup.vi \tmp2\(), v8, 4
+ vslidedown.vi v8, \r0\(), 4
+ vsetivli t0, 6, e16, m1, tu, ma
+ vslideup.vi \tmp2\(), v8, 5
+ vslidedown.vi v8, \r1\(), 6
+ vsetivli t0, 7, e16, m1, tu, ma
+ vslideup.vi \tmp2\(), v8, 6
+ vslidedown.vi v8, \r0\(), 6
+ vsetivli t0, 8, e16, m1, tu, ma
+ vslideup.vi \tmp2\(), v8, 7
+ vsetivli t0, 1, e16, m1, tu, ma
+ vslidedown.vi v8, \r1\(), 1
+ vmv.v.v \tmp3\(), v8
+ vslidedown.vi v8, \r0\(), 1
+ vsetivli t0, 2, e16, m1, tu, ma
+ vslideup.vi \tmp3\(), v8, 1
+ vslidedown.vi v8, \r1\(), 3
+ vsetivli t0, 3, e16, m1, tu, ma
+ vslideup.vi \tmp3\(), v8, 2
+ vslidedown.vi v8, \r0\(), 3
+ vsetivli t0, 4, e16, m1, tu, ma
+ vslideup.vi \tmp3\(), v8, 3
+ vslidedown.vi v8, \r1\(), 5
+ vsetivli t0, 5, e16, m1, tu, ma
+ vslideup.vi \tmp3\(), v8, 4
+ vslidedown.vi v8, \r0\(), 5
+ vsetivli t0, 6, e16, m1, tu, ma
+ vslideup.vi \tmp3\(), v8, 5
+ vslidedown.vi v8, \r1\(), 7
+ vsetivli t0, 7, e16, m1, tu, ma
+ vslideup.vi \tmp3\(), v8, 6
+ vslidedown.vi v8, \r0\(), 7
+ vsetivli t0, 8, e16, m1, tu, ma
+ vslideup.vi \tmp3\(), v8, 7
+ vsetivli t0, 1, e32, m1, tu, ma
+ vmv.v.v \tmp4\(), \tmp0\()
+ vsetivli t0, 2, e32, m1, tu, ma
+ vslideup.vi \tmp4\(), \tmp2\(), 1
+ vslidedown.vi v8, \tmp0\(), 2
+ vsetivli t0, 3, e32, m1, tu, ma
+ vslideup.vi \tmp4\(), v8, 2
+ vslidedown.vi v8, \tmp2\(), 2
+ vsetivli t0, 4, e32, m1, tu, ma
+ vslideup.vi \tmp4\(), v8, 3
+ vsetivli t0, 1, e32, m1, tu, ma
+ vslidedown.vi v8, \tmp0\(), 1
+ vmv.v.v \tmp5\(), v8
+ vslidedown.vi v8, \tmp2\(), 1
+ vsetivli t0, 2, e32, m1, tu, ma
+ vslideup.vi \tmp5\(), v8, 1
+ vslidedown.vi v8, \tmp0\(), 3
+ vsetivli t0, 3, e32, m1, tu, ma
+ vslideup.vi \tmp5\(), v8, 2
+ vslidedown.vi v8, \tmp2\(), 3
+ vsetivli t0, 4, e32, m1, tu, ma
+ vslideup.vi \tmp5\(), v8, 3
+ vsetivli t0, 1, e32, m1, tu, ma
+ vmv.v.v \tmp0\(), \tmp1\()
+ vsetivli t0, 2, e32, m1, tu, ma
+ vslideup.vi \tmp0\(), \tmp3\(), 1
+ vslidedown.vi v8, \tmp1\(), 2
+ vsetivli t0, 3, e32, m1, tu, ma
+ vslideup.vi \tmp0\(), v8, 2
+ vslidedown.vi v8, \tmp3\(), 2
+ vsetivli t0, 4, e32, m1, tu, ma
+ vslideup.vi \tmp0\(), v8, 3
+ vsetivli t0, 1, e32, m1, tu, ma
+ vslidedown.vi v8, \tmp1\(), 1
+ vmv.v.v \tmp2\(), v8
+ vslidedown.vi v8, \tmp3\(), 1
+ vsetivli t0, 2, e32, m1, tu, ma
+ vslideup.vi \tmp2\(), v8, 1
+ vslidedown.vi v8, \tmp1\(), 3
+ vsetivli t0, 3, e32, m1, tu, ma
+ vslideup.vi \tmp2\(), v8, 2
+ vslidedown.vi v8, \tmp3\(), 3
+ vsetivli t0, 4, e32, m1, tu, ma
+ vslideup.vi \tmp2\(), v8, 3
+
+ vsetivli t0, 1, e64, m1, tu, ma
+ vslidedown.vi v8, \tmp4\(), 1
+ vsetivli t0, 2, e64, m1, tu, ma
+ vslideup.vi \r3\(), v8, 1
+
+ vsetivli t0, 1, e64, m1, tu, ma
+ vslidedown.vi v8, \tmp5\(), 1
+ vsetivli t0, 2, e64, m1, tu, ma
+ vslideup.vi \r1\(), v8, 1
+
+ vsetivli t0, 1, e64, m1, tu, ma
+ vslidedown.vi v8, \tmp0\(), 1
+ vsetivli t0, 2, e64, m1, tu, ma
+ vslideup.vi \r2\(), v8, 1
+
+ vsetivli t0, 1, e64, m1, tu, ma
+ vslidedown.vi v8, \tmp2\(), 1
+ vsetivli t0, 2, e64, m1, tu, ma
+ vslideup.vi \r0\(), v8, 1
+.endm
+
+.macro tr16_8x4 in0, in1, in2, in3, offset
+ tr_4x4_8 \in0, \in1, \in2, \in3, v24, v25, v26, v27
+
+ vsetivli t0, 1, e16, m1, tu, ma
+ vmv.x.s s2, v0
+ vslidedown.vi v12, v0, 1
+ vmv.x.s s3, v12
+ vslidedown.vi v12, v0, 2
+ vmv.x.s s4, v12
+ vslidedown.vi v12, v0, 3
+ vmv.x.s s5, v12
+ vslidedown.vi v12, v0, 4
+ vmv.x.s s6, v12
+ vslidedown.vi v12, v0, 5
+ vmv.x.s s7, v12
+ vslidedown.vi v12, v0, 6
+ vmv.x.s s8, v12
+ vslidedown.vi v12, v0, 7
+ vmv.x.s s9, v12
+
+ vsetivli t0, 4, e16, mf2, tu, ma
+ vslidedown.vi v8, \in0, 4
+ vwmul.vx v28, v8, s6
+ vslidedown.vi v8, \in0, 4
+ vwmul.vx v29, v8, s7
+ vslidedown.vi v8, \in0, 4
+ vwmul.vx v30, v8, s8
+ vslidedown.vi v8, \in0, 4
+ vwmul.vx v31, v8, s9
+
+ sum_sub v28, \in1, s7, +, 2
+ sum_sub v29, \in1, s9, -, 2
+ sum_sub v30, \in1, s6, -, 2
+ sum_sub v31, \in1, s8, -, 2
+ sum_sub v28, \in2, s8, +, 2
+ sum_sub v29, \in2, s6, -, 2
+ sum_sub v30, \in2, s9, +, 2
+ sum_sub v31, \in2, s7, +, 2
+ sum_sub v28, \in3, s9, +, 2
+ sum_sub v29, \in3, s8, -, 2
+ sum_sub v30, \in3, s7, +, 2
+ sum_sub v31, \in3, s6, -, 2
+
+ butterfly v24, v28, v16, v23
+ butterfly v25, v29, v17, v22
+ butterfly v26, v30, v18, v21
+ butterfly v27, v31, v19, v20
+
+ li a7, \offset
+ add a4, sp, a7
+
+ vsetivli t0, 4, e32, m1, tu, ma
+ vse32.v v16, (a4)
+ add a4, a4, 16
+ vse32.v v17, (a4)
+ add a4, a4, 16
+ vse32.v v18, (a4)
+ add a4, a4, 16
+ vse32.v v19, (a4)
+ add a4, a4, 16
+
+ vse32.v v20, (a4)
+ add a4, a4, 16
+ vse32.v v21, (a4)
+ add a4, a4, 16
+ vse32.v v22, (a4)
+ add a4, a4, 16
+ vse32.v v23, (a4)
+ add a4, a4, 16
+
+ add a4, a4, -64
+.endm
+
+.macro scale out0, out1, out2, out3, in0, in1, in2, in3, in4, in5, in6, in7, shift
+ vsetivli t0, 4, e16, mf2, tu, ma
+ vnclip.wi \out0\(), \in0\(), \shift
+ vsetivli t0, 1, e64, m1, tu, ma
+ vmv.x.s a7, \out0\()
+ vsetivli t0, 4, e16, mf2, tu, ma
+ vnclip.wi v8, \in1\(), \shift
+ vsetivli t0, 2, e64, m1, tu, ma
+ vslide1up.vx \out0\(), v8, a7
+ vsetivli t0, 4, e16, mf2, tu, ma
+ vnclip.wi \out1\(), \in2\(), \shift
+ vsetivli t0, 1, e64, m1, tu, ma
+ vmv.x.s a7, \out1\()
+ vsetivli t0, 4, e16, mf2, tu, ma
+ vnclip.wi v8, \in3\(), \shift
+ vsetivli t0, 2, e64, m1, tu, ma
+ vslide1up.vx \out1\(), v8, a7
+ vsetivli t0, 4, e16, mf2, tu, ma
+ vnclip.wi \out2\(), \in4\(), \shift
+ vsetivli t0, 1, e64, m1, tu, ma
+ vmv.x.s a7, \out2\()
+ vsetivli t0, 4, e16, mf2, tu, ma
+ vnclip.wi v8, \in5\(), \shift
+ vsetivli t0, 2, e64, m1, tu, ma
+ vslide1up.vx \out2\(), v8, a7
+ vsetivli t0, 4, e16, mf2, tu, ma
+ vnclip.wi \out3\(), \in6\(), \shift
+ vsetivli t0, 1, e64, m1, tu, ma
+ vmv.x.s a7, \out3\()
+ vsetivli t0, 4, e16, mf2, tu, ma
+ vnclip.wi v8, \in7\(), \shift
+ vsetivli t0, 2, e64, m1, tu, ma
+ vslide1up.vx \out3\(), v8, a7
+
+.endm
+
+.macro tr_4x4_8 in0, in1, in2, in3, out0, out1, out2, out3, p1, p2
+ vsetivli t0, 4, e16, m1, tu, ma
+ vwcvt.x.x.v v8, \in0
+ vsetivli t0, 4, e32, m1, tu, ma
+ vsll.vi v28, v8, 6
+
+ vsetivli t0, 16, e8, m1, tu, ma
+ vmv.v.v v29, v28
+
+ vsetivli t0, 1, e16, m1, tu, ma
+ vmv.x.s s2, v0
+ vslidedown.vi v12, v0, 1
+ vmv.x.s s3, v12
+ vslidedown.vi v12, v0, 3
+ vmv.x.s s5, v12
+ vsetivli t0, 4, e16, mf2, tu, ma
+ vwmul.vx v30, \in1, s3
+ vwmul.vx v31, \in1, s5
+ vwmacc.vx v28, s2, \in2
+ neg s2, s2
+ vwmacc.vx v29, s2, \in2
+ neg s2, s2
+ vwmacc.vx v30, s5, \in3
+ neg s3, s3
+ vwmacc.vx v31, s3, \in3
+ neg s3, s3
+
+ vsetivli t0, 4, e32, m1, tu, ma
+ vadd.vv \out0, v28, v30
+ vadd.vv \out1, v29, v31
+ vsub.vv \out2, v29, v31
+ vsub.vv \out3, v28, v30
+.endm
+
+.macro tr_16x4 name, shift, offset, step
+func func_tr_16x4_\name, zve64x
+ mv a1, a5
+ addi a3, a5, \step * 64
+ li a2, \step * 128
+
+ load16 v16, v17, v18, v19
+
+ lla a1, trans
+
+ vsetivli t0, 8, e16, m1, tu, ma
+ vle16.v v0, (a1)
+
+ tr16_8x4 v16, v17, v18, v19, \offset
+
+ addi a1, a5, \step * 32
+ addi a3, a5, \step * 3 *32
+
+ li a2, \step * 128
+
+ load16 v20, v17, v18, v19
+
+ lla a1, trans
+ addi a1, a1, 16
+
+ vsetivli t0, 8, e16, m1, tu, ma
+ vle16.v v1, (a1)
+
+ vsetivli t0, 1, e16, m1, tu, ma
+ vmv.x.s s2, v1
+ vslidedown.vi v12, v1, 1
+ vmv.x.s s3, v12
+ vslidedown.vi v12, v1, 2
+ vmv.x.s s4, v12
+ vslidedown.vi v12, v1, 3
+ vmv.x.s s5, v12
+ vslidedown.vi v12, v1, 4
+ vmv.x.s s6, v12
+ vslidedown.vi v12, v1, 5
+ vmv.x.s s7, v12
+ vslidedown.vi v12, v1, 6
+ vmv.x.s s8, v12
+ vslidedown.vi v12, v1, 7
+ vmv.x.s s9, v12
+
+ vsetivli t0, 4, e16, mf2, tu, ma
+ vwmul.vx v21, v20, s2
+ vwmul.vx v22, v20, s3
+ vwmul.vx v23, v20, s4
+ vwmul.vx v24, v20, s5
+ vwmul.vx v25, v20, s6
+ vwmul.vx v26, v20, s7
+ vwmul.vx v27, v20, s8
+ vwmul.vx v28, v20, s9
+
+ vsetivli t0, 1, e16, m1, tu, ma
+ vmv.x.s s2, v1
+ vslidedown.vi v12, v1, 1
+ vmv.x.s s3, v12
+ vslidedown.vi v12, v1, 2
+ vmv.x.s s4, v12
+ vslidedown.vi v12, v1, 3
+ vmv.x.s s5, v12
+ vslidedown.vi v12, v1, 4
+ vmv.x.s s6, v12
+ vslidedown.vi v12, v1, 5
+ vmv.x.s s7, v12
+ vslidedown.vi v12, v1, 6
+ vmv.x.s s8, v12
+ vslidedown.vi v12, v1, 7
+ vmv.x.s s9, v12
+
+ add_member v20, s3, s6, s9, s7, s4, s2, s5, s8, +, +, +, -, -, -, -, -, 2
+ add_member v17, s4, s9, s5, s3, s8, s6, s2, s7, +, +, -, -, -, +, +, +
+ add_member v17, s5, s7, s3, s9, s2, s8, s4, s6, +, -, -, +, +, +, -, -, 2
+ add_member v18, s6, s4, s8, s2, s9, s3, s7, s5, +, -, -, +, -, -, +, +
+ add_member v18, s7, s2, s6, s8, s3, s5, s9, s4, +, -, +, +, -, +, +, -, 2
+ add_member v19, s8, s5, s2, s4, s7, s9, s6, s3, +, -, +, -, +, +, -, +
+ add_member v19, s9, s8, s7, s6, s5, s4, s3, s2, +, -, +, -, +, -, +, -, 2
+
+ li a7, \offset
+ add a4, sp, a7
+
+ vsetivli t0, 4, e32, m1, tu, ma
+ vle32.v v16, (a4)
+ addi a4, a4, 16
+ vle32.v v17, (a4)
+ addi a4, a4, 16
+ vle32.v v18, (a4)
+ addi a4, a4, 16
+ vle32.v v19, (a4)
+ addi a4, a4, 16
+
+ butterfly16 v16, v21, v17, v22, v18, v23, v19, v24
+ .if \shift > 0
+ scale v29, v30, v31, v24, v20, v16, v21, v17, v22, v18, v23, v19, \shift
+
+ transpose16_4x4_2 v29, v30, v31, v24, v2, v3, v4, v5, v6, v7
+
+ mv a1, a6
+ addi a3, a6, 24 +3*32
+ li a2, 32
+ li a4, -32
+
+ store16 v29, v30, v31, v24, a4
+ .else
+ store_to_stack \offset, (\offset + 240), v20, v21, v22, v23, v19, v18, v17, v16
+ .endif
+
+ li a7, \offset+64
+ add a4, sp, a7
+
+ vsetivli t0, 4, e32, m1, tu, ma
+ vle32.v v16, (a4)
+ addi a4, a4, 16
+ vle32.v v17, (a4)
+ addi a4, a4, 16
+ vle32.v v18, (a4)
+ addi a4, a4, 16
+ vle32.v v19, (a4)
+ addi a4, a4, 16
+
+ butterfly16 v16, v25, v17, v26, v18, v27, v19, v28
+ .if \shift > 0
+ scale v29, v30, v31, v20, v20, v16, v25, v17, v26, v18, v27, v19, \shift
+ transpose16_4x4_2 v29, v30, v31, v20, v2, v3, v4, v5, v6, v7
+
+ add a1, a6, 8
+ add a3, a6, (16 + 3 * 32)
+ li a2, 32
+ li a4, -32
+ store16 v29, v30, v31, v20, a4
+ .else
+ store_to_stack (\offset + 64), (\offset + 176), v20, v25, v26, v27, v19, v18, v17, v16
+ .endif
+
+ ret
+endfunc
+.endm
+
+tr_16x4 noscale, 0, 2048, 4
+
+.macro tr_32x4 name, shift
+func func_tr_32x4_\name, zve64x
+ mv t3, ra
+ jal func_tr_16x4_noscale
+
+ load32
+
+ lla t2, trans
+ addi t2, t2, 32
+
+ vsetivli t0, 4, e16, m1, tu, ma
+ vle16.v v0, (t2)
+ addi t2, t2, 2*4
+ vle16.v v1, (t2)
+ addi t2, t2, 2*4
+ vle16.v v2, (t2)
+ addi t2, t2, 2*4
+ vle16.v v3, (t2)
+ addi t2, t2, -2*4
+
+ li a7, 2048
+ add a4, sp, a7
+
+ li a2, 64
+ li t1, -64
+
+ jal tr_block1
+ mv a1, t4
+ addi a3, t4, (56 + 3 * 64)
+ scale_store \shift
+
+ jal tr_block2
+ addi a1, t4, 8
+ addi a3, t4, (48 + 3 * 64)
+ scale_store \shift
+
+ jal tr_block3
+ addi a1, t4, 16
+ addi a3, t4, (40 + 3 * 64)
+ scale_store \shift
+
+ jal tr_block4
+ addi a1, t4, 24
+ addi a3, t4, (32 + 3 * 64)
+ scale_store \shift
+
+ jr t3
+endfunc
+.endm
+
+tr_32x4 firstpass, 7
+tr_32x4 secondpass_8, 20 - 8
+
+.macro idct_32x32 bitdepth
+func ff_hevc_idct_32x32_\bitdepth\()_rvv, zve64x
+ mv t6, ra
+ addi sp,sp,-8*13
+ sd ra,8*12(sp)
+ sd s2,8*9(sp)
+ sd s3,8*8(sp)
+ sd s4,8*7(sp)
+ sd s5,8*6(sp)
+ sd s6,8*5(sp)
+ sd s7,8*4(sp)
+ sd s8,8*3(sp)
+ sd s9,8*2(sp)
+ sd s10,8*1(sp)
+ sd s11,8*0(sp)
+
+ csrwi vxrm, 1
+ li a7, 2432
+ sub sp, sp, a7
+
+.irp i, 0, 1, 2, 3, 4, 5, 6, 7
+ li a7, 8 * \i
+ add a5, a0, a7
+
+ li a7, 8 * \i * 32
+ add t4, sp, a7
+ jal func_tr_32x4_firstpass
+.endr
+
+.irp i, 0, 1, 2, 3, 4, 5, 6, 7
+ addi a5, sp, 8 * \i
+ addi t4, a0, 8 * \i * 32
+ jal func_tr_32x4_secondpass_\bitdepth
+.endr
+
+ li a7, 2432
+ add sp, sp, a7
+ ld ra,8*12(sp)
+ ld s2,8*9(sp)
+ ld s3,8*8(sp)
+ ld s4,8*7(sp)
+ ld s5,8*6(sp)
+ ld s6,8*5(sp)
+ ld s7,8*4(sp)
+ ld s8,8*3(sp)
+ ld s9,8*2(sp)
+ ld s10,8*1(sp)
+ ld s11,8*0(sp)
+ addi sp,sp,8*13
+
+ jr t6
+endfunc
+.endm
+
+idct_32x32 8
diff --git a/libavcodec/riscv/hevcdsp_init.c b/libavcodec/riscv/hevcdsp_init.c
index 1d8326a573..6dfb889eec 100644
--- a/libavcodec/riscv/hevcdsp_init.c
+++ b/libavcodec/riscv/hevcdsp_init.c
@@ -27,6 +27,8 @@
#include "libavcodec/hevc/dsp.h"
#include "libavcodec/riscv/h26x/h2656dsp.h"
+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 +42,37 @@ 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) {
- 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;
+
+ if (flags & AV_CPU_FLAG_RVV_I64)
+ if (vlenb >= 16)
+ switch (bit_depth) {
+ case 8:
+ c->idct[3] = ff_hevc_idct_32x32_8_rvv;
+ 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
--
2.25.1
More information about the ffmpeg-devel
mailing list