ARM: NEON optimised vector_fmac_scalar()

Signed-off-by: Mans Rullgard <mans@mansr.com>

libavcodec/arm/dsputil_init_neon.c

@@ -143,6 +143,8 @@ void ff_vector_fmul_window_neon(float *dst, const float *src0,
                                 const float *src1, const float *win, int len);
 void ff_vector_fmul_scalar_neon(float *dst, const float *src, float mul,
                                 int len);
+void ff_vector_fmac_scalar_neon(float *dst, const float *src, float mul,
+                                int len);
 void ff_butterflies_float_neon(float *v1, float *v2, int len);
 float ff_scalarproduct_float_neon(const float *v1, const float *v2, int len);
 void ff_vector_fmul_reverse_neon(float *dst, const float *src0,
@@ -305,6 +307,7 @@ void ff_dsputil_init_neon(DSPContext *c, AVCodecContext *avctx)
     c->vector_fmul                = ff_vector_fmul_neon;
     c->vector_fmul_window         = ff_vector_fmul_window_neon;
     c->vector_fmul_scalar         = ff_vector_fmul_scalar_neon;
+    c->vector_fmac_scalar         = ff_vector_fmac_scalar_neon;
     c->butterflies_float          = ff_butterflies_float_neon;
     c->scalarproduct_float        = ff_scalarproduct_float_neon;
     c->vector_fmul_reverse        = ff_vector_fmul_reverse_neon;

libavcodec/arm/dsputil_neon.S

@@ -587,6 +587,54 @@ NOVFP   vdup.32         q8,  r2
         .unreq          len
 endfunc
 
+function ff_vector_fmac_scalar_neon, export=1
+VFP     len .req r2
+VFP     acc .req r3
+NOVFP   len .req r3
+NOVFP   acc .req r2
+VFP     vdup.32         q15, d0[0]
+NOVFP   vdup.32         q15, r2
+        bics            r12, len, #15
+        mov             acc, r0
+        beq             3f
+        vld1.32         {q0},     [r1,:128]!
+        vld1.32         {q8},     [acc,:128]!
+        vld1.32         {q1},     [r1,:128]!
+        vld1.32         {q9},     [acc,:128]!
+1:      vmla.f32        q8,  q0,  q15
+        vld1.32         {q2},     [r1,:128]!
+        vld1.32         {q10},    [acc,:128]!
+        vmla.f32        q9,  q1,  q15
+        vld1.32         {q3},     [r1,:128]!
+        vld1.32         {q11},    [acc,:128]!
+        vmla.f32        q10, q2,  q15
+        vst1.32         {q8},     [r0,:128]!
+        vmla.f32        q11, q3,  q15
+        vst1.32         {q9},     [r0,:128]!
+        subs            r12, r12, #16
+        beq             2f
+        vld1.32         {q0},     [r1,:128]!
+        vld1.32         {q8},     [acc,:128]!
+        vst1.32         {q10},    [r0,:128]!
+        vld1.32         {q1},     [r1,:128]!
+        vld1.32         {q9},     [acc,:128]!
+        vst1.32         {q11},    [r0,:128]!
+        b               1b
+2:      vst1.32         {q10},    [r0,:128]!
+        vst1.32         {q11},    [r0,:128]!
+        ands            len, len, #15
+        it              eq
+        bxeq            lr
+3:      vld1.32         {q0},     [r1,:128]!
+        vld1.32         {q8},     [acc,:128]!
+        vmla.f32        q8,  q0,  q15
+        vst1.32         {q8},     [r0,:128]!
+        subs            len, len, #4
+        bgt             3b
+        bx              lr
+        .unreq          len
+endfunc
+
 function ff_butterflies_float_neon, export=1
 1:      vld1.32         {q0},[r0,:128]
         vld1.32         {q1},[r1,:128]