hevc: Add NEON IDCT DC functions for bitdepth 8

Signed-off-by: Alexandra Hájková <alexandra@khirnov.net>
Signed-off-by: Martin Storsjö <martin@martin.st>

libavcodec/arm/hevc_idct.S

 /*
  * ARM NEON optimised IDCT functions for HEVC decoding
+ *
+ * Copyright (c) 2014 Seppo Tomperi <seppo.tomperi@vtt.fi>
  * Copyright (c) 2017 Alexandra Hájková
  *
  * This file is part of Libav.
@@ -28,6 +30,82 @@ const trans, align=4
         .short 57, 43, 25, 9
 endconst
 
+function ff_hevc_idct_4x4_dc_8_neon, export=1
+        ldrsh           r1, [r0]
+        ldr             r2, =0x20
+        add             r1, #1
+        asr             r1, #1
+        add             r1, r2
+        asr             r1, #6
+        vdup.16         q0, r1
+        vdup.16         q1, r1
+        vst1.16         {q0, q1}, [r0, :128]
+        bx              lr
+endfunc
+
+function ff_hevc_idct_8x8_dc_8_neon, export=1
+        ldrsh           r1, [r0]
+        ldr             r2, =0x20
+        add             r1, #1
+        asr             r1, #1
+        add             r1, r2
+        asr             r1, #6
+        vdup.16         q8, r1
+        vdup.16         q9, r1
+        vmov.16         q10, q8
+        vmov.16         q11, q8
+        vmov.16         q12, q8
+        vmov.16         q13, q8
+        vmov.16         q14, q8
+        vmov.16         q15, q8
+        vstm            r0, {q8-q15}
+        bx              lr
+endfunc
+
+function ff_hevc_idct_16x16_dc_8_neon, export=1
+        ldrsh           r1, [r0]
+        ldr             r2, =0x20
+        add             r1, #1
+        asr             r1, #1
+        add             r1, r2
+        asr             r1, #6
+        vdup.16         q8, r1
+        vdup.16         q9, r1
+        vmov.16         q10, q8
+        vmov.16         q11, q8
+        vmov.16         q12, q8
+        vmov.16         q13, q8
+        vmov.16         q14, q8
+        vmov.16         q15, q8
+        vstm            r0!, {q8-q15}
+        vstm            r0!, {q8-q15}
+        vstm            r0!, {q8-q15}
+        vstm            r0, {q8-q15}
+        bx              lr
+endfunc
+
+function ff_hevc_idct_32x32_dc_8_neon, export=1
+        ldrsh           r1, [r0]
+        ldr             r2, =0x20
+        add             r1, #1
+        asr             r1, #1
+        add             r1, r2
+        asr             r1, #6
+        mov             r3, #16
+        vdup.16         q8, r1
+        vdup.16         q9, r1
+        vmov.16         q10, q8
+        vmov.16         q11, q8
+        vmov.16         q12, q8
+        vmov.16         q13, q8
+        vmov.16         q14, q8
+        vmov.16         q15, q8
+1:      subs            r3, #1
+        vstm            r0!, {q8-q15}
+        bne             1b
+        bx              lr
+endfunc
+
 .macro sum_sub out, in, c, op
   .ifc \op, +
         vmlal.s16       \out, \in, \c

libavcodec/arm/hevcdsp_init_arm.c

@@ -25,6 +25,11 @@
 
 #include "libavcodec/hevcdsp.h"
 
+void ff_hevc_idct_4x4_dc_8_neon(int16_t *coeffs);
+void ff_hevc_idct_8x8_dc_8_neon(int16_t *coeffs);
+void ff_hevc_idct_16x16_dc_8_neon(int16_t *coeffs);
+void ff_hevc_idct_32x32_dc_8_neon(int16_t *coeffs);
+
 void ff_hevc_idct_4x4_8_neon(int16_t *coeffs, int col_limit);
 void ff_hevc_idct_8x8_8_neon(int16_t *coeffs, int col_limit);
 void ff_hevc_idct_16x16_8_neon(int16_t *coeffs, int col_limit);
@@ -38,6 +43,11 @@ av_cold void ff_hevc_dsp_init_arm(HEVCDSPContext *c, int bit_depth)
 
     if (have_neon(cpu_flags)) {
         if (bit_depth == 8) {
+            c->idct_dc[0] = ff_hevc_idct_4x4_dc_8_neon;
+            c->idct_dc[1] = ff_hevc_idct_8x8_dc_8_neon;
+            c->idct_dc[2] = ff_hevc_idct_16x16_dc_8_neon;
+            c->idct_dc[3] = ff_hevc_idct_32x32_dc_8_neon;
+
             c->idct[0] = ff_hevc_idct_4x4_8_neon;
             c->idct[1] = ff_hevc_idct_8x8_8_neon;
             c->idct[2] = ff_hevc_idct_16x16_8_neon;