Since the SBR noise application optimisations are currently restricted
to hardware with 128-bit vectors, and use a quadruple multipler, they
can load up to 16 32-bit elements. But the "loads" are of 2 segments,
or 16 pairs of single precision float.
Thus we need to expand the dupiclated section of the noise table from
2x8 to 2x16 to avoid overflows.
---
libavcodec/sbrdsp_template.c | 6 ++++++
1 file changed, 6 insertions(+)
diff --git a/libavcodec/sbrdsp_template.c b/libavcodec/sbrdsp_template.c
index 80d535cbfa..0f731ba50d 100644
--- a/libavcodec/sbrdsp_template.c
+++ b/libavcodec/sbrdsp_template.c
@@ -371,4 +371,10 @@ const attribute_visibility_hidden DECLARE_ALIGNED(16,
INTFLOAT, AAC_RENAME(ff_sb
{Q31( 0.14130051758487f), Q31(-0.95090983575689f)}, {Q31(-0.47005496701697f),
Q31(-0.37340549728647f)},
{Q31( 0.80705063769351f), Q31( 0.29653668284408f)}, {Q31(-0.38981478896926f),
Q31( 0.89572605717087f)},
{Q31(-0.01053049862020f), Q31(-0.66959058036166f)}, {Q31(-0.91266367957293f),
Q31(-0.11522938140034f)},
+#if ARCH_RISCV
+{Q31( 0.54840422910309f), Q31( 0.75221367176302f)}, {Q31( 0.40009252867955f),
Q31(-0.98929400334421f)},
+{Q31(-0.99867974711855f), Q31(-0.88147068645358f)}, {Q31(-0.95531076805040f),
Q31( 0.90908757154593f)},
+{Q31(-0.45725933317144f), Q31(-0.56716323646760f)}, {Q31(-0.72929675029275f),
Q31(-0.98008272727324f)},
+{Q31( 0.75622801399036f), Q31( 0.20950329995549f)}, {Q31( 0.07069442601050f),
Q31(-0.78247898470706f)},
+#endif
};
--
2.45.1
_______________________________________________
ffmpeg-devel mailing list
[email protected]
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
[email protected] with subject "unsubscribe".
