libvpx: Pull from upstream

source/libvpx/vp9/common/x86/vp9_subpixel_8t_ssse3.asm

 ;
 ;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
 ;
 ;  Use of this source code is governed by a BSD-style license
 ;  that can be found in the LICENSE file in the root of the source
 ;  tree. An additional intellectual property rights grant can be found
 ;  in the file PATENTS.  All contributing project authors may
 ;  be found in the AUTHORS file in the root of the source tree.
 ;
 
 
 %include "vpx_ports/x86_abi_support.asm"
 
-;/************************************************************************************
-; Notes: filter_block1d_h6 applies a 6 tap filter horizontally to the input pixels. The
-; input pixel array has output_height rows. This routine assumes that output_height is an
-; even number. This function handles 8 pixels in horizontal direction, calculating ONE
-; rows each iteration to take advantage of the 128 bits operations.
-;
-; This is an implementation of some of the SSE optimizations first seen in ffvp8
-;
-;*************************************************************************************/
-
-
 %macro VERTx4 1
     mov         rdx, arg(5)                 ;filter ptr
     mov         rsi, arg(0)                 ;src_ptr
     mov         rdi, arg(2)                 ;output_ptr
     mov         rcx, 0x0400040
 
     movdqa      xmm4, [rdx]                 ;load filters
     movd        xmm5, rcx
     packsswb    xmm4, xmm4
     pshuflw     xmm0, xmm4, 0b              ;k0_k1
@@ skipping 39 matching lines @@
 
     movd        xmm6, [rsi + rbx]           ;G
     movd        xmm7, [rax + rbx]           ;H
 
     pmaddubsw   xmm0, k0k1
     pmaddubsw   xmm2, k2k3
     punpcklbw   xmm6, xmm7                  ;G H
     pmaddubsw   xmm4, k4k5
     pmaddubsw   xmm6, k6k7
 
+    movdqa      xmm1, xmm2
     paddsw      xmm0, xmm6
+    pmaxsw      xmm2, xmm4
+    pminsw      xmm4, xmm1
+    paddsw      xmm0, xmm4
     paddsw      xmm0, xmm2
-    paddsw      xmm0, xmm4
+
     paddsw      xmm0, krd
-
     psraw       xmm0, 7
     packuswb    xmm0, xmm0
 
     add         rsi,  rdx
     add         rax,  rdx
 %if %1
     movd        xmm1, [rdi]
     pavgb       xmm0, xmm1
 %endif
     movd        [rdi], xmm0
@@ skipping 60 matching lines @@
     movq        xmm6, [rsi + rbx]           ;G
     movq        xmm7, [rax + rbx]           ;H
 
     pmaddubsw   xmm0, k0k1
     pmaddubsw   xmm2, k2k3
     punpcklbw   xmm6, xmm7                  ;G H
     pmaddubsw   xmm4, k4k5
     pmaddubsw   xmm6, k6k7
 
     paddsw      xmm0, xmm6
+    movdqa      xmm1, xmm2
+    pmaxsw      xmm2, xmm4
+    pminsw      xmm4, xmm1
+    paddsw      xmm0, xmm4
     paddsw      xmm0, xmm2
-    paddsw      xmm0, xmm4
+
     paddsw      xmm0, krd
-
     psraw       xmm0, 7
     packuswb    xmm0, xmm0
 
     add         rsi,  rdx
     add         rax,  rdx
 %if %1
     movq        xmm1, [rdi]
     pavgb       xmm0, xmm1
 %endif
     movq        [rdi], xmm0
@@ skipping 61 matching lines @@
     movq        xmm6, [rsi + rbx]           ;G
     movq        xmm7, [rax + rbx]           ;H
 
     pmaddubsw   xmm0, k0k1
     pmaddubsw   xmm2, k2k3
     punpcklbw   xmm6, xmm7                  ;G H
     pmaddubsw   xmm4, k4k5
     pmaddubsw   xmm6, k6k7
 
     paddsw      xmm0, xmm6
+    movdqa      xmm1, xmm2
+    pmaxsw      xmm2, xmm4
+    pminsw      xmm4, xmm1
+    paddsw      xmm0, xmm4
     paddsw      xmm0, xmm2
-    paddsw      xmm0, xmm4
+
     paddsw      xmm0, krd
-
     psraw       xmm0, 7
     packuswb    xmm0, xmm0
 %if %1
     movq        xmm1, [rdi]
     pavgb       xmm0, xmm1
 %endif
     movq        [rdi], xmm0
 
     movq        xmm0, [rsi + 8]             ;A
     movq        xmm1, [rsi + rdx + 8]       ;B
@@ skipping 263 matching lines @@
     RESTORE_XMM
     UNSHADOW_ARGS
     pop         rbp
     ret
 
 ;~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 %macro HORIZx4_ROW 2
     movdqa      %2,   %1
     pshufb      %1,   [GLOBAL(shuf_t0t1)]
     pshufb      %2,   [GLOBAL(shuf_t2t3)]
-    pmaddubsw   %1,   xmm6
-    pmaddubsw   %2,   xmm7
+    pmaddubsw   %1,   k0k1k4k5
+    pmaddubsw   %2,   k2k3k6k7
 
-    paddsw      %1,   %2
-    movdqa      %2,   %1
+    movdqa      xmm4, %1
+    movdqa      xmm5, %2
+    psrldq      %1,   8
     psrldq      %2,   8
-    paddsw      %1,   %2
-    paddsw      %1,   xmm5
+    movdqa      xmm6, xmm5
+
+    paddsw      xmm4, %2
+    pmaxsw      xmm5, %1
+    pminsw      %1, xmm6
+    paddsw      %1, xmm4
+    paddsw      %1, xmm5
+
+    paddsw      %1,   krd
     psraw       %1,   7
     packuswb    %1,   %1
 %endm
 
 %macro HORIZx4 1
     mov         rdx, arg(5)                 ;filter ptr
     mov         rsi, arg(0)                 ;src_ptr
     mov         rdi, arg(2)                 ;output_ptr
     mov         rcx, 0x0400040
 
     movdqa      xmm4, [rdx]                 ;load filters
     movq        xmm5, rcx
     packsswb    xmm4, xmm4
     pshuflw     xmm6, xmm4, 0b              ;k0_k1
     pshufhw     xmm6, xmm6, 10101010b       ;k0_k1_k4_k5
     pshuflw     xmm7, xmm4, 01010101b       ;k2_k3
     pshufhw     xmm7, xmm7, 11111111b       ;k2_k3_k6_k7
     pshufd      xmm5, xmm5, 0               ;rounding
 
+    movdqa      k0k1k4k5, xmm6
+    movdqa      k2k3k6k7, xmm7
+    movdqa      krd, xmm5
+
     movsxd      rax, dword ptr arg(1)       ;src_pixels_per_line
     movsxd      rdx, dword ptr arg(3)       ;output_pitch
     movsxd      rcx, dword ptr arg(4)       ;output_height
     shr         rcx, 1
 .loop:
     ;Do two rows once
     movq        xmm0,   [rsi - 3]           ;load src
     movq        xmm1,   [rsi + 5]
     movq        xmm2,   [rsi + rax - 3]
     movq        xmm3,   [rsi + rax + 5]
@@ skipping 46 matching lines @@
     pshufb      %1,   [GLOBAL(shuf_t0t1)]
     pshufb      %2,   [GLOBAL(shuf_t2t3)]
     pshufb      %3,   [GLOBAL(shuf_t4t5)]
     pshufb      %4,   [GLOBAL(shuf_t6t7)]
 
     pmaddubsw   %1,   k0k1
     pmaddubsw   %2,   k2k3
     pmaddubsw   %3,   k4k5
     pmaddubsw   %4,   k6k7
 
+    paddsw      %1,   %4
+    movdqa      %4,   %2
+    pmaxsw      %2,   %3
+    pminsw      %3,   %4
+    paddsw      %1,   %3
     paddsw      %1,   %2
-    paddsw      %1,   %4
-    paddsw      %1,   %3
+
     paddsw      %1,   krd
     psraw       %1,   7
     packuswb    %1,   %1
 %endm
 
 %macro HORIZx8 1
     mov         rdx, arg(5)                 ;filter ptr
     mov         rsi, arg(0)                 ;src_ptr
     mov         rdi, arg(2)                 ;output_ptr
     mov         rcx, 0x0400040
@@ skipping 125 matching lines @@
 
     pmaddubsw   xmm0,   k0k1
     pmaddubsw   xmm1,   k2k3
     pmaddubsw   xmm2,   k4k5
     pmaddubsw   xmm3,   k6k7
     pmaddubsw   xmm4,   k0k1
     pmaddubsw   xmm5,   k2k3
     pmaddubsw   xmm6,   k4k5
     pmaddubsw   xmm7,   k6k7
 
+    paddsw      xmm0,   xmm3
+    movdqa      xmm3,   xmm1
+    pmaxsw      xmm1,   xmm2
+    pminsw      xmm2,   xmm3
+    paddsw      xmm0,   xmm2
     paddsw      xmm0,   xmm1
-    paddsw      xmm0,   xmm3
-    paddsw      xmm0,   xmm2
+
+    paddsw      xmm4,   xmm7
+    movdqa      xmm7,   xmm5
+    pmaxsw      xmm5,   xmm6
+    pminsw      xmm6,   xmm7
+    paddsw      xmm4,   xmm6
     paddsw      xmm4,   xmm5
-    paddsw      xmm4,   xmm7
-    paddsw      xmm4,   xmm6
 
     paddsw      xmm0,   krd
     paddsw      xmm4,   krd
     psraw       xmm0,   7
     psraw       xmm4,   7
     packuswb    xmm0,   xmm0
     packuswb    xmm4,   xmm4
     punpcklqdq  xmm0,   xmm4
 %if %1
     movdqa      xmm1,   [rdi]
@@ skipping 21 matching lines @@
 sym(vp9_filter_block1d4_h8_ssse3):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 6
     SAVE_XMM 7
     GET_GOT     rbx
     push        rsi
     push        rdi
     ; end prolog
 
+    ALIGN_STACK 16, rax
+    sub         rsp, 16 * 3
+    %define k0k1k4k5 [rsp + 16 * 0]
+    %define k2k3k6k7 [rsp + 16 * 1]
+    %define krd      [rsp + 16 * 2]
+
     HORIZx4 0
 
+    add rsp, 16 * 3
+    pop rsp
     ; begin epilog
     pop rdi
     pop rsi
     RESTORE_GOT
     RESTORE_XMM
     UNSHADOW_ARGS
     pop         rbp
     ret
 
 ;void vp9_filter_block1d8_h8_ssse3
@@ skipping 84 matching lines @@
 sym(vp9_filter_block1d4_h8_avg_ssse3):
     push        rbp
     mov         rbp, rsp
     SHADOW_ARGS_TO_STACK 6
     SAVE_XMM 7
     GET_GOT     rbx
     push        rsi
     push        rdi
     ; end prolog
 
+    ALIGN_STACK 16, rax
+    sub         rsp, 16 * 3
+    %define k0k1k4k5 [rsp + 16 * 0]
+    %define k2k3k6k7 [rsp + 16 * 1]
+    %define krd      [rsp + 16 * 2]
+
     HORIZx4 1
 
+    add rsp, 16 * 3
+    pop rsp
     ; begin epilog
     pop rdi
     pop rsi
     RESTORE_GOT
     RESTORE_XMM
     UNSHADOW_ARGS
     pop         rbp
     ret
 
 global sym(vp9_filter_block1d8_h8_avg_ssse3) PRIVATE
@@ skipping 67 matching lines @@
     db  0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8
 align 16
 shuf_t2t3:
     db  2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10
 align 16
 shuf_t4t5:
     db  4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12
 align 16
 shuf_t6t7:
     db  6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, 12, 13, 13, 14