arm: vp9itxfm: Only reload the idct coeffs for the iadst_idct combination
authorMartin Storsjö <martin@martin.st>
Mon, 9 Jan 2017 22:15:14 +0000 (00:15 +0200)
committerMichael Niedermayer <michael@niedermayer.cc>
Sat, 14 Jan 2017 20:13:27 +0000 (21:13 +0100)
This avoids reloading them if they haven't been clobbered, if the
first pass also was idct.

This is similar to what was done in the aarch64 version.

This is cherrypicked from libav commit
3c87039a404c5659ae9bf7454a04e186532eb40b.

Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
libavcodec/arm/vp9itxfm_neon.S

index 0097f5f..d5b8495 100644 (file)
@@ -814,7 +814,7 @@ A       and             r7,  sp,  #15
         mov             r3,  #\i
         bl              \txfm1\()16_1d_4x16_pass1_neon
 .endr
-.ifc \txfm2,idct
+.ifc \txfm1\()_\txfm2,iadst_idct
         movrel          r12, idct_coeffs
         vld1.16         {q0-q1}, [r12,:128]
 .endif