swresample/x86/rematrix_init: Avoid allocation for native_simd_one

Reviewed-by: Michael Niedermayer <michael@niedermayer.cc>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>

libswresample/rematrix.c

@@ -562,7 +562,6 @@ av_cold int swri_rematrix_init(SwrContext *s){
 av_cold void swri_rematrix_free(SwrContext *s){
     av_freep(&s->native_matrix);
     av_freep(&s->native_simd_matrix);
-    av_freep(&s->native_simd_one);
 }
 
 int swri_rematrix(SwrContext *s, AudioData *out, AudioData *in, int len, int mustcopy){

libswresample/swresample.c

@@ -688,7 +688,7 @@ static int swr_convert_internal(struct SwrContext *s, AudioData *out, int out_co
 
                     if(len1)
                         for(ch=0; ch<preout->ch_count; ch++)
-                            s->mix_2_1_simd(conv_src->ch[ch], preout->ch[ch], s->dither.noise.ch[ch] + s->dither.noise.bps * s->dither.noise_pos, s->native_simd_one, 0, 0, len1);
+                            s->mix_2_1_simd(conv_src->ch[ch], preout->ch[ch], s->dither.noise.ch[ch] + s->dither.noise.bps * s->dither.noise_pos, &s->native_simd_one, 0, 0, len1);
                     if(out_count != len1)
                         for(ch=0; ch<preout->ch_count; ch++)
                             s->mix_2_1_f(conv_src->ch[ch] + off, preout->ch[ch] + off, s->dither.noise.ch[ch] + s->dither.noise.bps * s->dither.noise_pos + off, &s->native_one, 0, 0, out_count - len1);

libswresample/swresample_internal.h

@@ -179,7 +179,10 @@ struct SwrContext {
         double d;
     } native_one;
     uint8_t *native_matrix;
-    uint8_t *native_simd_one;
+    union {
+        int16_t i16[2];
+        float   f;
+    } native_simd_one;
     uint8_t *native_simd_matrix;
     uint8_t matrix_ch[SWR_CH_MAX][SWR_CH_MAX+1];    ///< Lists of input channels per output channel that have non zero rematrixing coefficients
     mix_1_1_func_type *mix_1_1_f;

libswresample/x86/rematrix_init.c

@@ -48,8 +48,7 @@ av_cold int swri_rematrix_init_x86(struct SwrContext *s){
             s->mix_2_1_simd = ff_mix_2_1_a_int16_sse2;
         }
         s->native_simd_matrix = av_calloc(num,  2 * sizeof(int16_t));
-        s->native_simd_one    = av_mallocz(2 * sizeof(int16_t));
-        if (!s->native_simd_matrix || !s->native_simd_one)
+        if (!s->native_simd_matrix)
             return AVERROR(ENOMEM);
 
         for(i=0; i<nb_out; i++){
@@ -63,8 +62,8 @@ av_cold int swri_rematrix_init_x86(struct SwrContext *s){
                     ((((int*)s->native_matrix)[i * nb_in + j]) + (1<<sh>>1)) >> sh;
             }
         }
-        ((int16_t*)s->native_simd_one)[1] = 14;
-        ((int16_t*)s->native_simd_one)[0] = 16384;
+        s->native_simd_one.i16[1] = 14;
+        s->native_simd_one.i16[0] = 16384;
     } else if(s->midbuf.fmt == AV_SAMPLE_FMT_FLTP){
         if(EXTERNAL_SSE(mm_flags)) {
             s->mix_1_1_simd = ff_mix_1_1_a_float_sse;
@@ -75,11 +74,10 @@ av_cold int swri_rematrix_init_x86(struct SwrContext *s){
             s->mix_2_1_simd = ff_mix_2_1_a_float_avx;
         }
         s->native_simd_matrix = av_calloc(num, sizeof(float));
-        s->native_simd_one = av_mallocz(sizeof(float));
-        if (!s->native_simd_matrix || !s->native_simd_one)
+        if (!s->native_simd_matrix)
             return AVERROR(ENOMEM);
         memcpy(s->native_simd_matrix, s->native_matrix, num * sizeof(float));
-        memcpy(s->native_simd_one, &s->native_one.f, sizeof(float));
+        s->native_simd_one.f = s->native_one.f;
     }
 #endif