ARM: NEON optimised vector_clip_int32()

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

libavcodec/arm/dsputil_init_neon.c

@@ -160,6 +160,8 @@ void ff_vector_fmul_add_neon(float *dst, const float *src0, const float *src1,
 
 void ff_vector_clipf_neon(float *dst, const float *src, float min, float max,
                           int len);
+void ff_vector_clip_int32_neon(int32_t *dst, const int32_t *src, int32_t min,
+                               int32_t max, unsigned int len);
 
 void ff_vorbis_inverse_coupling_neon(float *mag, float *ang, int blocksize);
 
@@ -316,6 +318,7 @@ void ff_dsputil_init_neon(DSPContext *c, AVCodecContext *avctx)
     c->vector_fmul_reverse        = ff_vector_fmul_reverse_neon;
     c->vector_fmul_add            = ff_vector_fmul_add_neon;
     c->vector_clipf               = ff_vector_clipf_neon;
+    c->vector_clip_int32          = ff_vector_clip_int32_neon;
 
     c->vector_fmul_sv_scalar[0] = ff_vector_fmul_sv_scalar_2_neon;
     c->vector_fmul_sv_scalar[1] = ff_vector_fmul_sv_scalar_4_neon;

libavcodec/arm/dsputil_neon.S

@@ -815,3 +815,19 @@ function ff_apply_window_int16_neon, export=1
 
         pop             {r4,pc}
 endfunc
+
+function ff_vector_clip_int32_neon, export=1
+        vdup.32         q0,  r2
+        vdup.32         q1,  r3
+        ldr             r2,  [sp]
+1:
+        vld1.32         {q2-q3},  [r1,:128]!
+        vmin.s32        q2,  q2,  q1
+        vmin.s32        q3,  q3,  q1
+        vmax.s32        q2,  q2,  q0
+        vmax.s32        q3,  q3,  q0
+        vst1.32         {q2-q3},  [r0,:128]!
+        subs            r2,  r2,  #8
+        bgt             1b
+        bx              lr
+endfunc