arm: Implement x264_denoise_dct_neon

[x264] / common / arm / quant-a.S

diff --git a/common/arm/quant-a.S b/common/arm/quant-a.S
index 38045bc63efb9f2e17e1f5b32b69af4d72383de8..e63170e30e0e95d50ea36d22a3ba9d21a40a579e 100644 (file)
--- a/common/arm/quant-a.S
+++ b/common/arm/quant-a.S
@@ -1,9 +1,10 @@
 /****************************************************************************
  * quant.S: arm quantization and level-run
  *****************************************************************************
- * Copyright (C) 2009-2013 x264 project
+ * Copyright (C) 2009-2015 x264 project
  *
  * Authors: David Conrad <lessen42@gmail.com>
+ *          Janne Grunau <janne-x264@jannau.net>
  *
  * This program is free software; you can redistribute it and/or modify
  * it under the terms of the GNU General Public License as published by
@@ -25,8 +26,6 @@
 
 #include "asm.S"
 
-.fpu neon
-
 .section .rodata
 .align 4
 pmovmskb_byte:
@@ -80,7 +79,7 @@ function x264_quant_2x2_dc_neon
     vsub.s16    d3,  d3,  d0
     vst1.64     {d3}, [r0,:64]
     QUANT_END   d3
-.endfunc
+endfunc
 
 // quant_4x4_dc( int16_t dct[16], int mf, int bias )
 function x264_quant_4x4_dc_neon
@@ -92,7 +91,7 @@ function x264_quant_4x4_dc_neon
     QUANT_TWO   q0,  q0,  d4,  d5,  d4,  d5,  q0
     vorr        d0,  d0,  d1
     QUANT_END   d0
-.endfunc
+endfunc
 
 // quant_4x4( int16_t dct[16], uint16_t mf[16], uint16_t bias[16] )
 function x264_quant_4x4_neon
@@ -104,7 +103,7 @@ function x264_quant_4x4_neon
     QUANT_TWO   q0,  q1,  d4,  d5,  d6,  d7, q0
     vorr        d0,  d0,  d1
     QUANT_END   d0
-.endfunc
+endfunc
 
 // quant_4x4x4( int16_t dct[4][16], uint16_t mf[16], uint16_t bias[16] )
 function x264_quant_4x4x4_neon
@@ -145,7 +144,7 @@ function x264_quant_4x4x4_neon
     orrne       r0,  #8
     vpop        {d8-d15}
     bx          lr
-.endfunc
+endfunc
 
 // quant_8x8( int16_t dct[64], uint16_t mf[64], uint16_t bias[64] )
 function x264_quant_8x8_neon
@@ -165,7 +164,7 @@ function x264_quant_8x8_neon
 .endr
     vorr        d0,  d0,  d1
     QUANT_END   d0
-.endfunc
+endfunc
 
 .macro DEQUANT_START mf_size offset dc=no
     mov         r3,  #0x2b
@@ -257,7 +256,7 @@ dequant_\size\()_rshift_loop:
     bgt         dequant_\size\()_rshift_loop
 .endif
     bx          lr
-.endfunc
+endfunc
 .endm
 
 DEQUANT 4x4, 4
@@ -307,7 +306,7 @@ dequant_4x4_dc_rshift:
     vmovn.s32   d3,  q13
     vst1.16     {d0-d3},   [r0,:128]
     bx          lr
-.endfunc
+endfunc
 
 
 // int coeff_last( int16_t *l )
@@ -319,16 +318,28 @@ function x264_coeff_last4_arm
     lsrs        r2,  r2,  #16
     addne       r0,  r0,  #1
     bx          lr
-.endfunc
+endfunc
+
+function x264_coeff_last8_arm
+    ldrd        r2,  r3,  [r0, #8]
+    orrs        ip,  r2,  r3
+    movne       r0,  #4
+    ldrdeq      r2,  r3,  [r0]
+    moveq       r0,  #0
+    tst         r3,  r3
+    addne       r0,  #2
+    movne       r2,  r3
+    lsrs        r2,  r2,  #16
+    addne       r0,  r0,  #1
+    bx          lr
+endfunc
 
 .macro COEFF_LAST_1x size
 function x264_coeff_last\size\()_neon
 .if \size == 15
     sub         r0,  r0,  #2
-    vld1.64     {d0-d3}, [r0]
-.else
-    vld1.64     {d0-d3}, [r0,:128]
 .endif
+    vld1.64     {d0-d3}, [r0,:128]
     vtst.16     q0,  q0
     vtst.16     q1,  q1
     vshrn.u16   d0,  q0,  #8
@@ -344,7 +355,7 @@ function x264_coeff_last\size\()_neon
     subslt      r0,  r3,  r0,  lsr #2
     movlt       r0,  #0
     bx          lr
-.endfunc
+endfunc
 .endm
 
 COEFF_LAST_1x 15
@@ -393,4 +404,32 @@ function x264_coeff_last64_neon
     subslt      r0,  ip,  r0
     movlt       r0,  #0
     bx          lr
-.endfunc
+endfunc
+
+function x264_denoise_dct_neon
+1:  subs        r3,  r3,  #16
+    vld1.16     {q0,  q1},  [r0]
+    vld1.32     {q12, q13}, [r1]!
+    vld1.32     {q14, q15}, [r1]
+    sub         r1,  #32
+    vabs.s16    q8,  q0
+    vabs.s16    q9,  q1
+    vld1.16     {q2, q3}, [r2]!
+    vclt.s16    q10, q0,  #0
+    vclt.s16    q11, q1,  #0
+    vaddw.u16   q12, q12, d16
+    vaddw.u16   q13, q13, d17
+    vqsub.u16   q0,  q8,  q2
+    vqsub.u16   q1,  q9,  q3
+    vaddw.u16   q14, q14, d18
+    vaddw.u16   q15, q15, d19
+    vneg.s16    q8,  q0
+    vneg.s16    q9,  q1
+    vbsl        q10, q8,  q0
+    vbsl        q11, q9,  q1
+    vst1.32     {q12, q13}, [r1]!
+    vst1.32     {q14, q15}, [r1]!
+    vst1.16     {q10, q11}, [r0]!
+    bgt         1b
+    bx          lr
+endfunc