[FFmpeg-devel] [PATCH] address speed regression in h264 deblocking filter after PAFF

[Andreas Öman]

Hi

Michael Niedermayer wrote:
> and what else does change if you use av_always_inline?
> anything unrelated gcc stops inlining?

I don't spot any significant changes except the actual
deblocking functions.

diff of sorted output from nm with and without av_always_inline:

@@ -78,9 +78,6 @@
  000002bf t filter_mb_edgev
  000002e7 t h264_luma_dc_dequant_idct_c
  000002ed t flush_dpb
-00000379 t filter_mb_fast_simple
-00000384 t filter_mb_fast_interlace
-0000038c t filter_mb_fast_mode2
  00000398 r cabac_context_init_I
  000003a0 t filter_mb_edgeh
  0000048b t svq3_luma_dc_dequant_idct_c
@@ -94,25 +91,27 @@
  000008b8 t svq3_decode_frame
  00000943 t filter_mb_fast_mbaff
  00000ac8 r cabac_context_init_PB
-00000aee t decode_mb_skip
+00000b24 t decode_mb_skip
  00000baf t hl_motion
  00000d98 t decode_scaling_matrices
  00000dd9 t decode_init
  00000f55 t decode_ref_pic_list_reordering
-00000fde t hl_decode_mb_simple
+00000fdd t hl_decode_mb_simple
  00001232 t decode_cabac_residual
  00001845 t decode_frame
  00001989 t decode_nal_units
  00001a10 t hl_decode_mb_complex
  00001b3a t fill_default_ref_list
  00001b6f t decode_seq_parameter_set
-00001d13 t svq3_mc_dir
+00001d11 t svq3_mc_dir
  00001d9e t mc_part
  00001f3c t filter_mb
  00001f3f t fill_caches
  000020b7 t pred_direct_motion
-000023a8 t svq3_decode_mb
-00002777 t filter_mb_fast_internal
+00002418 t svq3_decode_mb
+00002bc9 t filter_mb_fast_mode2
  00002c27 t decode_slice_header
+00002c27 t filter_mb_fast_simple
+00002d7a t filter_mb_fast_interlace
  000067e6 t decode_mb_cavlc
  000071cc t decode_mb_cabac