[FFmpeg-devel] [PATCH 4/5] avcodec/on2avc: Avoid indirection when calling float dsp function

[Andreas Rheinhardt]

Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt at gmail.com>
---
 libavcodec/on2avc.c | 35 ++++++++++++++++++-----------------
 1 file changed, 18 insertions(+), 17 deletions(-)

diff --git a/libavcodec/on2avc.c b/libavcodec/on2avc.c
index 625e733ca3..3b566e1e4b 100644
--- a/libavcodec/on2avc.c
+++ b/libavcodec/on2avc.c
@@ -46,7 +46,8 @@ enum WindowTypes {
 
 typedef struct On2AVCContext {
     AVCodecContext *avctx;
-    AVFloatDSPContext *fdsp;
+    void (*vector_fmul_window)(float *dst, const float *src0,
+                               const float *src1, const float *win, int len);
     FFTContext mdct, mdct_half, mdct_small;
     FFTContext fft128, fft256, fft512, fft1024;
     void (*wtf)(struct On2AVCContext *ctx, float *out, float *in, int size);
@@ -720,7 +721,7 @@ static int on2avc_reconstruct_channel_ext(On2AVCContext *c, AVFrame *dst, int of
         }
 
         memcpy(out, saved, 448 * sizeof(float));
-        c->fdsp->vector_fmul_window(wout, saved + 448, buf, c->short_win, 64);
+        c->vector_fmul_window(wout, saved + 448, buf, c->short_win, 64);
         memcpy(wout + 128,  buf + 64,         448 * sizeof(float));
         memcpy(saved,       buf + 512,        448 * sizeof(float));
         memcpy(saved + 448, buf + 7*128 + 64,  64 * sizeof(float));
@@ -756,20 +757,20 @@ static int on2avc_reconstruct_channel(On2AVCContext *c, int channel,
          c->prev_window_type == WINDOW_TYPE_LONG_STOP) &&
         (c->window_type == WINDOW_TYPE_LONG ||
          c->window_type == WINDOW_TYPE_LONG_START)) {
-        c->fdsp->vector_fmul_window(out, saved, buf, c->long_win, 512);
+        c->vector_fmul_window(out, saved, buf, c->long_win, 512);
     } else {
         float *wout = out + 448;
         memcpy(out, saved, 448 * sizeof(float));
 
         if (c->window_type == WINDOW_TYPE_8SHORT) {
-            c->fdsp->vector_fmul_window(wout + 0*128, saved + 448,      buf + 0*128, c->short_win, 64);
-            c->fdsp->vector_fmul_window(wout + 1*128, buf + 0*128 + 64, buf + 1*128, c->short_win, 64);
-            c->fdsp->vector_fmul_window(wout + 2*128, buf + 1*128 + 64, buf + 2*128, c->short_win, 64);
-            c->fdsp->vector_fmul_window(wout + 3*128, buf + 2*128 + 64, buf + 3*128, c->short_win, 64);
-            c->fdsp->vector_fmul_window(temp,         buf + 3*128 + 64, buf + 4*128, c->short_win, 64);
+            c->vector_fmul_window(wout + 0*128, saved + 448,      buf + 0*128, c->short_win, 64);
+            c->vector_fmul_window(wout + 1*128, buf + 0*128 + 64, buf + 1*128, c->short_win, 64);
+            c->vector_fmul_window(wout + 2*128, buf + 1*128 + 64, buf + 2*128, c->short_win, 64);
+            c->vector_fmul_window(wout + 3*128, buf + 2*128 + 64, buf + 3*128, c->short_win, 64);
+            c->vector_fmul_window(temp,         buf + 3*128 + 64, buf + 4*128, c->short_win, 64);
             memcpy(wout + 4*128, temp, 64 * sizeof(float));
         } else {
-            c->fdsp->vector_fmul_window(wout, saved + 448, buf, c->short_win, 64);
+            c->vector_fmul_window(wout, saved + 448, buf, c->short_win, 64);
             memcpy(wout + 128, buf + 64, 448 * sizeof(float));
         }
     }
@@ -778,9 +779,9 @@ static int on2avc_reconstruct_channel(On2AVCContext *c, int channel,
     switch (c->window_type) {
     case WINDOW_TYPE_8SHORT:
         memcpy(saved,       temp + 64,         64 * sizeof(float));
-        c->fdsp->vector_fmul_window(saved + 64,  buf + 4*128 + 64, buf + 5*128, c->short_win, 64);
-        c->fdsp->vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, c->short_win, 64);
-        c->fdsp->vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, c->short_win, 64);
+        c->vector_fmul_window(saved + 64,  buf + 4*128 + 64, buf + 5*128, c->short_win, 64);
+        c->vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, c->short_win, 64);
+        c->vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, c->short_win, 64);
         memcpy(saved + 448, buf + 7*128 + 64,  64 * sizeof(float));
         break;
     case WINDOW_TYPE_LONG_START:
@@ -906,6 +907,7 @@ static av_cold void on2avc_free_vlcs(On2AVCContext *c)
 static av_cold int on2avc_decode_init(AVCodecContext *avctx)
 {
     On2AVCContext *c = avctx->priv_data;
+    AVFloatDSPContext *fdsp;
     int i;
 
     if (avctx->channels > 2U) {
@@ -952,9 +954,11 @@ static av_cold int on2avc_decode_init(AVCodecContext *avctx)
     ff_fft_init(&c->fft256,  7, 0);
     ff_fft_init(&c->fft512,  8, 1);
     ff_fft_init(&c->fft1024, 9, 1);
-    c->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
-    if (!c->fdsp)
+    fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
+    if (!fdsp)
         return AVERROR(ENOMEM);
+    c->vector_fmul_window = fdsp->vector_fmul_window;
+    av_free(fdsp);
 
     if (init_vlc(&c->scale_diff, 9, ON2AVC_SCALE_DIFFS,
                  ff_on2avc_scale_diff_bits,  1, 1,
@@ -975,7 +979,6 @@ static av_cold int on2avc_decode_init(AVCodecContext *avctx)
 vlc_fail:
     av_log(avctx, AV_LOG_ERROR, "Cannot init VLC\n");
     on2avc_free_vlcs(c);
-    av_freep(&c->fdsp);
     return AVERROR(ENOMEM);
 }
 
@@ -991,8 +994,6 @@ static av_cold int on2avc_decode_close(AVCodecContext *avctx)
     ff_fft_end(&c->fft512);
     ff_fft_end(&c->fft1024);
 
-    av_freep(&c->fdsp);
-
     on2avc_free_vlcs(c);
 
     return 0;
-- 
2.25.1