Merge commit '52c9b0a6c0d02cff6caebcf6989e565e05b55200'

[ffmpeg] / libavcodec / aarch64 / vp8dsp_neon.S

diff --git a/libavcodec/aarch64/vp8dsp_neon.S b/libavcodec/aarch64/vp8dsp_neon.S
index fbe064e9aaaa70469ccf51ea69c086d304a84892..6a727e717c39a5cd5db384eae41b90d1901b8e3f 100644 (file)
--- a/libavcodec/aarch64/vp8dsp_neon.S
+++ b/libavcodec/aarch64/vp8dsp_neon.S
@@ -4,6 +4,7 @@
  * Copyright (c) 2010 Rob Clark <rob@ti.com>
  * Copyright (c) 2011 Mans Rullgard <mans@mansr.com>
  * Copyright (c) 2018 Magnus Röös <mla2.roos@gmail.com>
+ * Copyright (c) 2019 Martin Storsjo <martin@martin.st>
  *
  * This file is part of FFmpeg.
  *
@@ -25,10 +26,66 @@
 #include "libavutil/aarch64/asm.S"
 #include "neon.S"
 
+function ff_vp8_luma_dc_wht_neon, export=1
+        ld1             {v0.4h - v3.4h}, [x1]
+        movi            v30.8h, #0
+
+        add             v4.4h,  v0.4h,  v3.4h
+        add             v6.4h,  v1.4h,  v2.4h
+        st1             {v30.8h}, [x1], #16
+        sub             v7.4h,  v1.4h,  v2.4h
+        sub             v5.4h,  v0.4h,  v3.4h
+        st1             {v30.8h}, [x1]
+        add             v0.4h,  v4.4h,  v6.4h
+        add             v1.4h,  v5.4h,  v7.4h
+        sub             v2.4h,  v4.4h,  v6.4h
+        sub             v3.4h,  v5.4h,  v7.4h
+
+        movi            v16.4h, #3
+
+        transpose_4x4H  v0, v1, v2, v3, v4, v5, v6, v7
+
+        add             v0.4h,  v0.4h,  v16.4h
+
+        add             v4.4h,  v0.4h,  v3.4h
+        add             v6.4h,  v1.4h,  v2.4h
+        sub             v7.4h,  v1.4h,  v2.4h
+        sub             v5.4h,  v0.4h,  v3.4h
+        add             v0.4h,  v4.4h,  v6.4h
+        add             v1.4h,  v5.4h,  v7.4h
+        sub             v2.4h,  v4.4h,  v6.4h
+        sub             v3.4h,  v5.4h,  v7.4h
+
+        sshr            v0.4h,  v0.4h,  #3
+        sshr            v1.4h,  v1.4h,  #3
+        sshr            v2.4h,  v2.4h,  #3
+        sshr            v3.4h,  v3.4h,  #3
+
+        mov             x3,  #32
+        st1             {v0.h}[0],  [x0], x3
+        st1             {v1.h}[0],  [x0], x3
+        st1             {v2.h}[0],  [x0], x3
+        st1             {v3.h}[0],  [x0], x3
+        st1             {v0.h}[1],  [x0], x3
+        st1             {v1.h}[1],  [x0], x3
+        st1             {v2.h}[1],  [x0], x3
+        st1             {v3.h}[1],  [x0], x3
+        st1             {v0.h}[2],  [x0], x3
+        st1             {v1.h}[2],  [x0], x3
+        st1             {v2.h}[2],  [x0], x3
+        st1             {v3.h}[2],  [x0], x3
+        st1             {v0.h}[3],  [x0], x3
+        st1             {v1.h}[3],  [x0], x3
+        st1             {v2.h}[3],  [x0], x3
+        st1             {v3.h}[3],  [x0], x3
+
+        ret
+endfunc
+
 function ff_vp8_idct_add_neon, export=1
         ld1             {v0.8b - v3.8b},  [x1]
         mov             w4,  #20091
-        movk            w4,  #35468/2, lsl 16
+        movk            w4,  #35468/2, lsl #16
         dup             v4.2s, w4
 
         smull           v26.4s, v1.4h,  v4.h[0]
@@ -102,6 +159,58 @@ function ff_vp8_idct_add_neon, export=1
         ret
 endfunc
 
+function ff_vp8_idct_dc_add4uv_neon, export=1
+        movi            v0.4h,  #0
+        mov             x3,     #32
+        ld1r            {v16.4h},  [x1]
+        st1             {v0.h}[0], [x1], x3
+        ld1r            {v17.4h},  [x1]
+        st1             {v0.h}[0], [x1], x3
+        ld1r            {v18.4h},  [x1]
+        st1             {v0.h}[0], [x1], x3
+        ld1r            {v19.4h},  [x1]
+        st1             {v0.h}[0], [x1], x3
+        ins             v16.d[1],  v17.d[0]
+        ins             v18.d[1],  v19.d[0]
+        mov             x3,  x0
+        srshr           v16.8h,    v16.8h,  #3            // dc >>= 3
+        ld1             {v0.8b},   [x0], x2
+        srshr           v18.8h,    v18.8h,  #3
+        ld1             {v1.8b},   [x0], x2
+        uaddw           v20.8h,    v16.8h, v0.8b
+        ld1             {v2.8b},   [x0], x2
+        uaddw           v0.8h,     v16.8h, v1.8b
+        ld1             {v3.8b},   [x0], x2
+        uaddw           v22.8h,    v16.8h, v2.8b
+        ld1             {v4.8b},   [x0], x2
+        uaddw           v2.8h,     v16.8h, v3.8b
+        ld1             {v5.8b},   [x0], x2
+        uaddw           v24.8h,    v18.8h, v4.8b
+        ld1             {v6.8b},   [x0], x2
+        uaddw           v4.8h,     v18.8h, v5.8b
+        ld1             {v7.8b},   [x0], x2
+        uaddw           v26.8h,    v18.8h, v6.8b
+        sqxtun          v20.8b,    v20.8h
+        uaddw           v6.8h,     v18.8h, v7.8b
+        sqxtun          v21.8b,    v0.8h
+        sqxtun          v22.8b,    v22.8h
+        st1             {v20.8b},  [x3], x2
+        sqxtun          v23.8b,    v2.8h
+        st1             {v21.8b},  [x3], x2
+        sqxtun          v24.8b,    v24.8h
+        st1             {v22.8b},  [x3], x2
+        sqxtun          v25.8b,    v4.8h
+        st1             {v23.8b},  [x3], x2
+        sqxtun          v26.8b,    v26.8h
+        st1             {v24.8b},  [x3], x2
+        sqxtun          v27.8b,    v6.8h
+        st1             {v25.8b},  [x3], x2
+        st1             {v26.8b},  [x3], x2
+        st1             {v27.8b},  [x3], x2
+
+        ret
+endfunc
+
 function ff_vp8_idct_dc_add4y_neon, export=1
         movi            v0.16b,  #0
         mov             x3,  #32
@@ -581,7 +690,7 @@ function ff_put_vp8_pixels16_neon, export=1
         st1             {v1.16b},     [x0], x1
         st1             {v2.16b},     [x0], x1
         st1             {v3.16b},     [x0], x1
-        bgt             1b
+        b.gt            1b
         ret
 endfunc
 
@@ -596,7 +705,7 @@ function ff_put_vp8_pixels8_neon, export=1
         st1             {v0.d}[1], [x0], x1
         st1             {v1.8b},   [x0], x1
         st1             {v1.d}[1], [x0], x1
-        bgt             1b
+        b.gt            1b
         ret
 endfunc
 
@@ -743,7 +852,7 @@ endfunc
 
 
 // note: worst case sum of all 6-tap filter values * 255 is 0x7f80 so 16 bit
-// arithmatic can be used to apply filters
+// arithmetic can be used to apply filters
 const   subpel_filters, align=4
         .short     0,   6, 123,  12,   1,   0,   0,   0
         .short     2,  11, 108,  36,   8,   1,   0,   0
@@ -778,7 +887,7 @@ function ff_put_vp8_epel16_v6_neon, export=1
         st1             {v1.1d - v2.1d}, [x0], x1
         st1             {v3.1d - v4.1d}, [x0], x1
         subs            x4, x4, #2
-        bne             1b
+        b.ne            1b
 
         ret
 endfunc
@@ -797,7 +906,7 @@ function ff_put_vp8_epel16_h6_neon, export=1
         st1             {v1.16b}, [x0], x1
 
         subs            w4, w4, #1
-        bne             1b
+        b.ne            1b
         ret
 endfunc
 
@@ -821,7 +930,7 @@ function ff_put_vp8_epel16_h6v6_neon, export=1
         vp8_epel16_h6   v1, v1, v2
         st1             {v1.16b}, [x7], #16
         subs            x16, x16, #1
-        bne             1b
+        b.ne            1b
 
 
         // second pass (vertical):
@@ -842,7 +951,7 @@ function ff_put_vp8_epel16_h6v6_neon, export=1
 
         st1             {v2.16b}, [x0], x1
         subs            x4, x4, #1
-        bne             2b
+        b.ne            2b
 
         add             sp,  sp,  #336+16
         ret
@@ -869,7 +978,7 @@ function ff_put_vp8_epel8_h6v6_neon, export=1
 
         st1             {v1.8b}, [x7], #8
         subs            x16, x16, #1
-        bne             1b
+        b.ne            1b
 
         // second pass (vertical):
         sxtw            x6,  w6
@@ -888,7 +997,7 @@ function ff_put_vp8_epel8_h6v6_neon, export=1
         st1             {v1.8b}, [x0], x1
         st1             {v2.8b}, [x0], x1
         subs            x4, x4, #2
-        bne             2b
+        b.ne            2b
 
         add             sp,  sp,  #168+16
         ret
@@ -915,7 +1024,7 @@ function ff_put_vp8_epel8_h4v6_neon, export=1
 
         st1             {v1.8b}, [x7], #8
         subs            x16, x16, #1
-        bne             1b
+        b.ne            1b
 
         // second pass (vertical):
         sxtw            x6,  w6
@@ -934,7 +1043,7 @@ function ff_put_vp8_epel8_h4v6_neon, export=1
         st1             {v1.8b}, [x0], x1
         st1             {v2.8b}, [x0], x1
         subs            x4, x4, #2
-        bne             2b
+        b.ne            2b
 
         add             sp,  sp,  #168+16
         ret
@@ -962,7 +1071,7 @@ function ff_put_vp8_epel8_h4v4_neon, export=1
 
         st1             {v1.8b}, [x7], #8
         subs            x16, x16, #1
-        bne             1b
+        b.ne            1b
 
         // second pass (vertical):
         sxtw            x6,  w6
@@ -979,7 +1088,7 @@ function ff_put_vp8_epel8_h4v4_neon, export=1
         st1             {v1.d}[0], [x0], x1
         st1             {v1.d}[1], [x0], x1
         subs            x4, x4, #2
-        bne             2b
+        b.ne            2b
 
         add             sp,  sp,  #168+16
         ret
@@ -1007,7 +1116,7 @@ function ff_put_vp8_epel8_h6v4_neon, export=1
 
         st1             {v1.8b}, [x7], #8
         subs            x16, x16, #1
-        bne             1b
+        b.ne            1b
 
         // second pass (vertical):
         sxtw            x6,  w6
@@ -1024,7 +1133,7 @@ function ff_put_vp8_epel8_h6v4_neon, export=1
         st1             {v1.d}[0], [x0], x1
         st1             {v1.d}[1], [x0], x1
         subs            x4, x4, #2
-        bne             2b
+        b.ne            2b
 
         add             sp,  sp,  #168+16
         ret