shithub: libvpx

Info  •  Files  •  Log  •  Branches

vp9_denoiser_neon,horizontal_add_s8x16: use vaddlv w/aarch64

this reduces the number of instructions to compute the sum

Change-Id: Icae4d4fb3e343d5b6e5a095c60ac6d171b3e7d54

--- a/vp9/encoder/arm/neon/vp9_denoiser_neon.c

+++ b/vp9/encoder/arm/neon/vp9_denoiser_neon.c

@@ -21,6 +21,9 @@

 // Compute the sum of all pixel differences of this MB.

 static INLINE int horizontal_add_s8x16(const int8x16_t v_sum_diff_total) {

+#if defined(__aarch64__)

+  return vaddlvq_s8(v_sum_diff_total);

+#else

   const int16x8_t fe_dc_ba_98_76_54_32_10 = vpaddlq_s8(v_sum_diff_total);

   const int32x4_t fedc_ba98_7654_3210 = vpaddlq_s16(fe_dc_ba_98_76_54_32_10);

   const int64x2_t fedcba98_76543210 = vpaddlq_s32(fedc_ba98_7654_3210);

@@ -28,6 +31,7 @@

                                 vget_low_s64(fedcba98_76543210));

   const int sum_diff = vget_lane_s32(vreinterpret_s32_s64(x), 0);

   return sum_diff;

+#endif

 }

 // Denoise a 16x1 vector.