dct32: Add AVX implementation of 32-point DCT

libavcodec/mpegaudiodec.c

@@ -69,9 +69,9 @@ typedef struct MPADecodeContext {
     uint32_t free_format_next_header;
     GetBitContext gb;
     GetBitContext in_gb;
-    DECLARE_ALIGNED(16, MPA_INT, synth_buf)[MPA_MAX_CHANNELS][512 * 2];
+    DECLARE_ALIGNED(32, MPA_INT, synth_buf)[MPA_MAX_CHANNELS][512 * 2];
     int synth_buf_offset[MPA_MAX_CHANNELS];
-    DECLARE_ALIGNED(16, INTFLOAT, sb_samples)[MPA_MAX_CHANNELS][36][SBLIMIT];
+    DECLARE_ALIGNED(32, INTFLOAT, sb_samples)[MPA_MAX_CHANNELS][36][SBLIMIT];
     INTFLOAT mdct_buf[MPA_MAX_CHANNELS][SBLIMIT * 18]; /* previous samples, for layer 3 MDCT */
     GranuleDef granules[2][2]; /* Used in Layer 3 */
 #ifdef DEBUG

libavcodec/x86/fft.c

@@ -57,7 +57,9 @@ av_cold void ff_fft_init_mmx(FFTContext *s)
 av_cold void ff_dct_init_mmx(DCTContext *s)
 {
     int has_vectors = av_get_cpu_flags();
-    if (has_vectors & AV_CPU_FLAG_SSE && HAVE_SSE)
+    if (has_vectors & AV_CPU_FLAG_AVX && HAVE_AVX)
+        s->dct32 = ff_dct32_float_avx;
+    else if (has_vectors & AV_CPU_FLAG_SSE && HAVE_SSE)
         s->dct32 = ff_dct32_float_sse;
 }
 #endif

libavcodec/x86/fft.h

@@ -35,5 +35,6 @@ void ff_imdct_calc_sse(FFTContext *s, FFTSample *output, const FFTSample *input)
 void ff_imdct_half_sse(FFTContext *s, FFTSample *output, const FFTSample *input);
 void ff_imdct_half_avx(FFTContext *s, FFTSample *output, const FFTSample *input);
 void ff_dct32_float_sse(FFTSample *out, const FFTSample *in);
+void ff_dct32_float_avx(FFTSample *out, const FFTSample *in);
 
 #endif /* AVCODEC_X86_FFT_H */