diff mbox series

[FFmpeg-devel] lavc/vc1dsp: fix R-V V avg_mspel_pixels

Message ID 20240601091128.13694-1-remi@remlab.net
State Accepted
Commit 6c6bec04f3fc57674cb0dfca27ce415990a3f3e1
Headers show
Series [FFmpeg-devel] lavc/vc1dsp: fix R-V V avg_mspel_pixels | expand

Checks

Context Check Description
yinshiyou/make_loongarch64 success Make finished
yinshiyou/make_fate_loongarch64 success Make fate finished
andriy/make_x86 success Make finished
andriy/make_fate_x86 success Make fate finished

Commit Message

Rémi Denis-Courmont June 1, 2024, 9:11 a.m. UTC 
The 8x8 pixel arrays are not necessarily aligned to 64 bits, so the
current code leads to Bus error on real hardware. This reproducible
with FATE's vc1_ilaced_twomv test case.

The new "pessimist" code can trivially be shared for 16x16 pixel
arrays so we also do that. FWIW, this also nominally reduces the
hardware requirement from Zve64x to Zve32x.

T-Head C908:
vc1dsp.avg_vc1_mspel_pixels_tab[0][0]_c:      14.7
vc1dsp.avg_vc1_mspel_pixels_tab[0][0]_rvv_i32: 3.5
vc1dsp.avg_vc1_mspel_pixels_tab[1][0]_c:       3.7
vc1dsp.avg_vc1_mspel_pixels_tab[1][0]_rvv_i32: 1.5

SpacemiT X60:
vc1dsp.avg_vc1_mspel_pixels_tab[0][0]_c:      13.0
vc1dsp.avg_vc1_mspel_pixels_tab[0][0]_rvv_i32: 3.0
vc1dsp.avg_vc1_mspel_pixels_tab[1][0]_c:       3.2
vc1dsp.avg_vc1_mspel_pixels_tab[1][0]_rvv_i32: 1.2
---
 libavcodec/riscv/vc1dsp_init.c |  2 +-
 libavcodec/riscv/vc1dsp_rvv.S  | 40 +++++++++++++++-------------------
 2 files changed, 18 insertions(+), 24 deletions(-)

Comments

Rémi Denis-Courmont June 1, 2024, 2:56 p.m. UTC | #1 
Le lauantaina 1. kesäkuuta 2024, 12.11.28 EEST Rémi Denis-Courmont a écrit :
> The 8x8 pixel arrays are not necessarily aligned to 64 bits, so the
> current code leads to Bus error on real hardware. This reproducible
> with FATE's vc1_ilaced_twomv test case.

http://fate.ffmpeg.org/report.cgi?time=20240601134536&slot=rv64gcvb-linux-gnu-gcc
diff mbox series

Patch

diff --git a/libavcodec/riscv/vc1dsp_init.c b/libavcodec/riscv/vc1dsp_init.c
index f105a3a3c6..e4838fb347 100644
--- a/libavcodec/riscv/vc1dsp_init.c
+++ b/libavcodec/riscv/vc1dsp_init.c
@@ -56,10 +56,10 @@  av_cold void ff_vc1dsp_init_riscv(VC1DSPContext *dsp)
             dsp->vc1_inv_trans_4x8_dc = ff_vc1_inv_trans_4x8_dc_rvv;
             dsp->vc1_inv_trans_4x4_dc = ff_vc1_inv_trans_4x4_dc_rvv;
             dsp->avg_vc1_mspel_pixels_tab[0][0] = ff_avg_pixels16x16_rvv;
+            dsp->avg_vc1_mspel_pixels_tab[1][0] = ff_avg_pixels8x8_rvv;
             if (flags & AV_CPU_FLAG_RVV_I64) {
                 dsp->vc1_inv_trans_8x8_dc = ff_vc1_inv_trans_8x8_dc_rvv;
                 dsp->vc1_inv_trans_8x4_dc = ff_vc1_inv_trans_8x4_dc_rvv;
-                dsp->avg_vc1_mspel_pixels_tab[1][0] = ff_avg_pixels8x8_rvv;
             }
         }
         dsp->startcode_find_candidate = ff_startcode_find_candidate_rvv;
diff --git a/libavcodec/riscv/vc1dsp_rvv.S b/libavcodec/riscv/vc1dsp_rvv.S
index 1166f35cad..8b3a830a4a 100644
--- a/libavcodec/riscv/vc1dsp_rvv.S
+++ b/libavcodec/riscv/vc1dsp_rvv.S
@@ -132,31 +132,25 @@  endfunc
 .endm
 
 func ff_avg_pixels16x16_rvv, zve32x
-        csrwi         vxrm, 0
-        vsetivli      zero, 16, e8, m1, ta, ma
-        mspel_op_all  l a1 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
-        mspel_op_all  l a0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
-        vsetvli       t0, zero, e8, m8, ta, ma
-        sub           a0, a0, a2
-        vaaddu.vv     v0, v0, v16
-        neg           a2, a2
-        vaaddu.vv     v8, v8, v24
-        vsetivli      zero, 16, e8, m1, ta, ma
-        mspel_op_all  s a0 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
-
-        ret
+        li       t0, 16
+        vsetivli zero, 16, e8, m1, ta, ma
+        j        1f
 endfunc
 
-func ff_avg_pixels8x8_rvv, zve64x
-        csrwi         vxrm, 0
-        li            t0, 64
-        vsetivli      zero, 8, e8, mf2, ta, ma
-        vlse64.v      v16, (a1), a2
-        vlse64.v      v8, (a0), a2
-        vsetvli       zero, t0, e8, m4, ta, ma
-        vaaddu.vv     v16, v16, v8
-        vsetivli      zero, 8, e8, mf2, ta, ma
-        vsse64.v      v16, (a0), a2
+func ff_avg_pixels8x8_rvv, zve32x
+        li        t0, 8
+        vsetivli  zero, 8, e8, mf2, ta, ma
+1:
+        csrwi     vxrm, 0
+2:
+        vle8.v    v16, (a1)
+        addi      t0, t0, -1
+        vle8.v    v8, (a0)
+        add       a1, a1, a2
+        vaaddu.vv v16, v16, v8
+        vse8.v    v16, (a0)
+        add       a0, a0, a2
+        bnez      t0, 2b
 
         ret
 endfunc