ac3enc: NEON optimised sum_square_butterfly_float

libavcodec/arm/ac3dsp_init_arm.c

@@ -33,6 +33,10 @@ void ff_ac3_sum_square_butterfly_int32_neon(int64_t sum[4],
                                             const int32_t *coef0,
                                             const int32_t *coef1,
                                             int len);
+void ff_ac3_sum_square_butterfly_float_neon(float sum[4],
+                                            const float *coef0,
+                                            const float *coef1,
+                                            int len);
 
 void ff_ac3_bit_alloc_calc_bap_armv6(int16_t *mask, int16_t *psd,
                                      int start, int end,
@@ -57,5 +61,6 @@ av_cold void ff_ac3dsp_init_arm(AC3DSPContext *c, int bit_exact)
         c->float_to_fixed24      = ff_float_to_fixed24_neon;
         c->extract_exponents     = ff_ac3_extract_exponents_neon;
         c->sum_square_butterfly_int32 = ff_ac3_sum_square_butterfly_int32_neon;
+        c->sum_square_butterfly_float = ff_ac3_sum_square_butterfly_float_neon;
     }
 }

libavcodec/arm/ac3dsp_neon.S

@@ -132,3 +132,23 @@ function ff_ac3_sum_square_butterfly_int32_neon, export=1
         vst1.64         {q0-q1},  [r0]
         bx              lr
 endfunc
+
+function ff_ac3_sum_square_butterfly_float_neon, export=1
+        vmov.f32        q0,  #0.0
+        vmov.f32        q1,  #0.0
+1:
+        vld1.32         {d16},    [r1]!
+        vld1.32         {d17},    [r2]!
+        vadd.f32        d18, d16, d17
+        vsub.f32        d19, d16, d17
+        vmla.f32        d0,  d16, d16
+        vmla.f32        d1,  d17, d17
+        vmla.f32        d2,  d18, d18
+        vmla.f32        d3,  d19, d19
+        subs            r3,  r3,  #2
+        bgt             1b
+        vpadd.f32       d0,  d0,  d1
+        vpadd.f32       d1,  d2,  d3
+        vst1.32         {q0},     [r0]
+        bx              lr
+endfunc