diff options
Diffstat (limited to 'third_party/aom/aom_dsp/x86/aom_subpixel_bilinear_ssse3.asm')
| -rw-r--r-- | third_party/aom/aom_dsp/x86/aom_subpixel_bilinear_ssse3.asm | 267 |
1 files changed, 0 insertions, 267 deletions
diff --git a/third_party/aom/aom_dsp/x86/aom_subpixel_bilinear_ssse3.asm b/third_party/aom/aom_dsp/x86/aom_subpixel_bilinear_ssse3.asm deleted file mode 100644 index 59edc49a9..000000000 --- a/third_party/aom/aom_dsp/x86/aom_subpixel_bilinear_ssse3.asm +++ /dev/null @@ -1,267 +0,0 @@ -; -; Copyright (c) 2016, Alliance for Open Media. All rights reserved -; -; This source code is subject to the terms of the BSD 2 Clause License and -; the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License -; was not distributed with this source code in the LICENSE file, you can -; obtain it at www.aomedia.org/license/software. If the Alliance for Open -; Media Patent License 1.0 was not distributed with this source code in the -; PATENTS file, you can obtain it at www.aomedia.org/license/patent. -; - -; - -%include "aom_ports/x86_abi_support.asm" - -%macro GET_PARAM_4 0 - mov rdx, arg(5) ;filter ptr - mov rsi, arg(0) ;src_ptr - mov rdi, arg(2) ;output_ptr - mov ecx, 0x01000100 - - movdqa xmm3, [rdx] ;load filters - psrldq xmm3, 6 - packsswb xmm3, xmm3 - pshuflw xmm3, xmm3, 0b ;k3_k4 - - movd xmm2, ecx ;rounding_shift - pshufd xmm2, xmm2, 0 - - movsxd rax, DWORD PTR arg(1) ;pixels_per_line - movsxd rdx, DWORD PTR arg(3) ;out_pitch - movsxd rcx, DWORD PTR arg(4) ;output_height -%endm - -%macro APPLY_FILTER_4 1 - punpcklbw xmm0, xmm1 - pmaddubsw xmm0, xmm3 - - pmulhrsw xmm0, xmm2 ;rounding(+64)+shift(>>7) - packuswb xmm0, xmm0 ;pack to byte - -%if %1 - movd xmm1, [rdi] - pavgb xmm0, xmm1 -%endif - movd [rdi], xmm0 - lea rsi, [rsi + rax] - lea rdi, [rdi + rdx] - dec rcx -%endm - -%macro GET_PARAM 0 - mov rdx, arg(5) ;filter ptr - mov rsi, arg(0) ;src_ptr - mov rdi, arg(2) ;output_ptr - mov ecx, 0x01000100 - - movdqa xmm7, [rdx] ;load filters - psrldq xmm7, 6 - packsswb xmm7, xmm7 - pshuflw xmm7, xmm7, 0b ;k3_k4 - punpcklwd xmm7, xmm7 - - movd xmm6, ecx ;rounding_shift - pshufd xmm6, xmm6, 0 - - movsxd rax, DWORD PTR arg(1) ;pixels_per_line - movsxd rdx, DWORD PTR arg(3) ;out_pitch - movsxd rcx, DWORD PTR arg(4) ;output_height -%endm - -%macro APPLY_FILTER_8 1 - punpcklbw xmm0, xmm1 - pmaddubsw xmm0, xmm7 - - pmulhrsw xmm0, xmm6 ;rounding(+64)+shift(>>7) - packuswb xmm0, xmm0 ;pack back to byte - -%if %1 - movq xmm1, [rdi] - pavgb xmm0, xmm1 -%endif - movq [rdi], xmm0 ;store the result - - lea rsi, [rsi + rax] - lea rdi, [rdi + rdx] - dec rcx -%endm - -%macro APPLY_FILTER_16 1 - punpcklbw xmm0, xmm1 - punpckhbw xmm2, xmm1 - pmaddubsw xmm0, xmm7 - pmaddubsw xmm2, xmm7 - - pmulhrsw xmm0, xmm6 ;rounding(+64)+shift(>>7) - pmulhrsw xmm2, xmm6 - packuswb xmm0, xmm2 ;pack back to byte - -%if %1 - movdqu xmm1, [rdi] - pavgb xmm0, xmm1 -%endif - movdqu [rdi], xmm0 ;store the result - - lea rsi, [rsi + rax] - lea rdi, [rdi + rdx] - dec rcx -%endm - -SECTION .text - -global sym(aom_filter_block1d4_v2_ssse3) PRIVATE -sym(aom_filter_block1d4_v2_ssse3): - push rbp - mov rbp, rsp - SHADOW_ARGS_TO_STACK 6 - push rsi - push rdi - ; end prolog - - GET_PARAM_4 -.loop: - movd xmm0, [rsi] ;load src - movd xmm1, [rsi + rax] - - APPLY_FILTER_4 0 - jnz .loop - - ; begin epilog - pop rdi - pop rsi - UNSHADOW_ARGS - pop rbp - ret - -global sym(aom_filter_block1d8_v2_ssse3) PRIVATE -sym(aom_filter_block1d8_v2_ssse3): - push rbp - mov rbp, rsp - SHADOW_ARGS_TO_STACK 6 - SAVE_XMM 7 - push rsi - push rdi - ; end prolog - - GET_PARAM -.loop: - movq xmm0, [rsi] ;0 - movq xmm1, [rsi + rax] ;1 - - APPLY_FILTER_8 0 - jnz .loop - - ; begin epilog - pop rdi - pop rsi - RESTORE_XMM - UNSHADOW_ARGS - pop rbp - ret - -global sym(aom_filter_block1d16_v2_ssse3) PRIVATE -sym(aom_filter_block1d16_v2_ssse3): - push rbp - mov rbp, rsp - SHADOW_ARGS_TO_STACK 6 - SAVE_XMM 7 - push rsi - push rdi - ; end prolog - - GET_PARAM -.loop: - movdqu xmm0, [rsi] ;0 - movdqu xmm1, [rsi + rax] ;1 - movdqa xmm2, xmm0 - - APPLY_FILTER_16 0 - jnz .loop - - ; begin epilog - pop rdi - pop rsi - RESTORE_XMM - UNSHADOW_ARGS - pop rbp - ret - -global sym(aom_filter_block1d4_h2_ssse3) PRIVATE -sym(aom_filter_block1d4_h2_ssse3): - push rbp - mov rbp, rsp - SHADOW_ARGS_TO_STACK 6 - push rsi - push rdi - ; end prolog - - GET_PARAM_4 -.loop: - movdqu xmm0, [rsi] ;load src - movdqa xmm1, xmm0 - psrldq xmm1, 1 - - APPLY_FILTER_4 0 - jnz .loop - - ; begin epilog - pop rdi - pop rsi - UNSHADOW_ARGS - pop rbp - ret - -global sym(aom_filter_block1d8_h2_ssse3) PRIVATE -sym(aom_filter_block1d8_h2_ssse3): - push rbp - mov rbp, rsp - SHADOW_ARGS_TO_STACK 6 - SAVE_XMM 7 - push rsi - push rdi - ; end prolog - - GET_PARAM -.loop: - movdqu xmm0, [rsi] ;load src - movdqa xmm1, xmm0 - psrldq xmm1, 1 - - APPLY_FILTER_8 0 - jnz .loop - - ; begin epilog - pop rdi - pop rsi - RESTORE_XMM - UNSHADOW_ARGS - pop rbp - ret - -global sym(aom_filter_block1d16_h2_ssse3) PRIVATE -sym(aom_filter_block1d16_h2_ssse3): - push rbp - mov rbp, rsp - SHADOW_ARGS_TO_STACK 6 - SAVE_XMM 7 - push rsi - push rdi - ; end prolog - - GET_PARAM -.loop: - movdqu xmm0, [rsi] ;load src - movdqu xmm1, [rsi + 1] - movdqa xmm2, xmm0 - - APPLY_FILTER_16 0 - jnz .loop - - ; begin epilog - pop rdi - pop rsi - RESTORE_XMM - UNSHADOW_ARGS - pop rbp - ret |