highperfocused/ffmpeg
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

lavc/vc1dsp: fix potential overflow in R-V V inv_trans_4

Browse Source 
Judging by the coefficients, the last round of add/sub can overflow
to 17 bits with a very small probability just as with the 8-point
transform. This is not observed under FATE, but better safe than sorry.
This commit is contained in:
Rémi Denis-Courmont
2024-06-27 22:00:05 +03:00
parent 349c49fd1b
commit b818dff8d8
1 changed files with 8 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
libavcodec/riscv/vc1dsp_rvv.S
View File

@@ -202,13 +202,14 @@ func ff_vc1_inv_trans_4_rvv, zve32x
vmul.vx v20, v1, t2 vmul.vx v20, v1, t2
vadd.vv v26, v14, v16 # t3 vadd.vv v26, v14, v16 # t3
vsub.vv v27, v18, v20 # t4 vsub.vv v27, v18, v20 # t4
vadd.vv v0, v24, v26 vwadd.vv v8, v24, v26
vsub.vv v1, v25, v27 vwsub.vv v10, v25, v27
vadd.vv v2, v25, v27 vwadd.vv v12, v25, v27
vsub.vv v3, v24, v26 vwsub.vv v14, v24, v26
.irp n,0,1,2,3 vnclip.wx v0, v8, t1
vssra.vx v\n, v\n, t1 # + 4 >> 3 or + 64 >> 7 vnclip.wx v1, v10, t1
.endr vnclip.wx v2, v12, t1
vnclip.wx v3, v14, t1
jr t0 jr t0
endfunc endfunc