ref: b79f25b54635294508fca42e3766b805fb4fcf4e
dir: /vp8/common/x86/subpixel_mmx.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"
%define BLOCK_HEIGHT_WIDTH 4
%define vp8_filter_weight 128
%define VP8_FILTER_SHIFT 7
SECTION .text
;void vp8_filter_block1d_h6_mmx
;(
; unsigned char *src_ptr,
; unsigned short *output_ptr,
; unsigned int src_pixels_per_line,
; unsigned int pixel_step,
; unsigned int output_height,
; unsigned int output_width,
; short * vp8_filter
;)
globalsym(vp8_filter_block1d_h6_mmx)
sym(vp8_filter_block1d_h6_mmx):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 7
GET_GOT rbx
push rsi
push rdi
; end prolog
mov rdx, arg(6) ;vp8_filter
movq mm1, [rdx + 16] ; do both the negative taps first!!!
movq mm2, [rdx + 32] ;
movq mm6, [rdx + 48] ;
movq mm7, [rdx + 64] ;
mov rdi, arg(1) ;output_ptr
mov rsi, arg(0) ;src_ptr
movsxd rcx, dword ptr arg(4) ;output_height
movsxd rax, dword ptr arg(5) ;output_width ; destination pitch?
pxor mm0, mm0 ; mm0 = 00000000
.nextrow:
movq mm3, [rsi-2] ; mm3 = p-2..p5
movq mm4, mm3 ; mm4 = p-2..p5
psrlq mm3, 8 ; mm3 = p-1..p5
punpcklbw mm3, mm0 ; mm3 = p-1..p2
pmullw mm3, mm1 ; mm3 *= kernel 1 modifiers.
movq mm5, mm4 ; mm5 = p-2..p5
punpckhbw mm4, mm0 ; mm5 = p2..p5
pmullw mm4, mm7 ; mm5 *= kernel 4 modifiers
paddsw mm3, mm4 ; mm3 += mm5
movq mm4, mm5 ; mm4 = p-2..p5;
psrlq mm5, 16 ; mm5 = p0..p5;
punpcklbw mm5, mm0 ; mm5 = p0..p3
pmullw mm5, mm2 ; mm5 *= kernel 2 modifiers
paddsw mm3, mm5 ; mm3 += mm5
movq mm5, mm4 ; mm5 = p-2..p5
psrlq mm4, 24 ; mm4 = p1..p5
punpcklbw mm4, mm0 ; mm4 = p1..p4
pmullw mm4, mm6 ; mm5 *= kernel 3 modifiers
paddsw mm3, mm4 ; mm3 += mm5
; do outer positive taps
movd mm4, [rsi+3]
punpcklbw mm4, mm0 ; mm5 = p3..p6
pmullw mm4, [rdx+80] ; mm5 *= kernel 0 modifiers
paddsw mm3, mm4 ; mm3 += mm5
punpcklbw mm5, mm0 ; mm5 = p-2..p1
pmullw mm5, [rdx] ; mm5 *= kernel 5 modifiers
paddsw mm3, mm5 ; mm3 += mm5
paddsw mm3, [GLOBAL(rd)] ; mm3 += round value
psraw mm3, VP8_FILTER_SHIFT ; mm3 /= 128
packuswb mm3, mm0 ; pack and unpack to saturate
punpcklbw mm3, mm0 ;
movq [rdi], mm3 ; store the results in the destination
%if ABI_IS_32BIT
add rsi, dword ptr arg(2) ;src_pixels_per_line ; next line
add rdi, rax;
%else
movsxd r8, dword ptr arg(2) ;src_pixels_per_line
add rdi, rax;
add rsi, r8 ; next line
%endif
dec rcx ; decrement count
jnz .nextrow ; next row
; begin epilog
pop rdi
pop rsi
RESTORE_GOT
UNSHADOW_ARGS
pop rbp
ret
;void vp8_filter_block1dc_v6_mmx
;(
; short *src_ptr,
; unsigned char *output_ptr,
; int output_pitch,
; unsigned int pixels_per_line,
; unsigned int pixel_step,
; unsigned int output_height,
; unsigned int output_width,
; short * vp8_filter
;)
globalsym(vp8_filter_block1dc_v6_mmx)
sym(vp8_filter_block1dc_v6_mmx):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 8
GET_GOT rbx
push rsi
push rdi
; end prolog
movq mm5, [GLOBAL(rd)]
push rbx
mov rbx, arg(7) ;vp8_filter
movq mm1, [rbx + 16] ; do both the negative taps first!!!
movq mm2, [rbx + 32] ;
movq mm6, [rbx + 48] ;
movq mm7, [rbx + 64] ;
movsxd rdx, dword ptr arg(3) ;pixels_per_line
mov rdi, arg(1) ;output_ptr
mov rsi, arg(0) ;src_ptr
sub rsi, rdx
sub rsi, rdx
movsxd rcx, DWORD PTR arg(5) ;output_height
movsxd rax, DWORD PTR arg(2) ;output_pitch ; destination pitch?
pxor mm0, mm0 ; mm0 = 00000000
.nextrow_cv:
movq mm3, [rsi+rdx] ; mm3 = p0..p8 = row -1
pmullw mm3, mm1 ; mm3 *= kernel 1 modifiers.
movq mm4, [rsi + 4*rdx] ; mm4 = p0..p3 = row 2
pmullw mm4, mm7 ; mm4 *= kernel 4 modifiers.
paddsw mm3, mm4 ; mm3 += mm4
movq mm4, [rsi + 2*rdx] ; mm4 = p0..p3 = row 0
pmullw mm4, mm2 ; mm4 *= kernel 2 modifiers.
paddsw mm3, mm4 ; mm3 += mm4
movq mm4, [rsi] ; mm4 = p0..p3 = row -2
pmullw mm4, [rbx] ; mm4 *= kernel 0 modifiers.
paddsw mm3, mm4 ; mm3 += mm4
add rsi, rdx ; move source forward 1 line to avoid 3 * pitch
movq mm4, [rsi + 2*rdx] ; mm4 = p0..p3 = row 1
pmullw mm4, mm6 ; mm4 *= kernel 3 modifiers.
paddsw mm3, mm4 ; mm3 += mm4
movq mm4, [rsi + 4*rdx] ; mm4 = p0..p3 = row 3
pmullw mm4, [rbx +80] ; mm4 *= kernel 3 modifiers.
paddsw mm3, mm4 ; mm3 += mm4
paddsw mm3, mm5 ; mm3 += round value
psraw mm3, VP8_FILTER_SHIFT ; mm3 /= 128
packuswb mm3, mm0 ; pack and saturate
movd [rdi],mm3 ; store the results in the destination
; the subsequent iterations repeat 3 out of 4 of these reads. Since the
; recon block should be in cache this shouldn't cost much. Its obviously
; avoidable!!!.
lea rdi, [rdi+rax] ;
dec rcx ; decrement count
jnz .nextrow_cv ; next row
pop rbx
; begin epilog
pop rdi
pop rsi
RESTORE_GOT
UNSHADOW_ARGS
pop rbp
ret
SECTION_RODATA
align 16
rd:
times 4 dw 0x40
align 16
global HIDDEN_DATA(sym(vp8_six_tap_x86))
sym(vp8_six_tap_x86):
times 8 dw 0
times 8 dw 0
times 8 dw 128
times 8 dw 0
times 8 dw 0
times 8 dw 0
times 8 dw 0
times 8 dw -6
times 8 dw 123
times 8 dw 12
times 8 dw -1
times 8 dw 0
times 8 dw 2
times 8 dw -11
times 8 dw 108
times 8 dw 36
times 8 dw -8
times 8 dw 1
times 8 dw 0
times 8 dw -9
times 8 dw 93
times 8 dw 50
times 8 dw -6
times 8 dw 0
times 8 dw 3
times 8 dw -16
times 8 dw 77
times 8 dw 77
times 8 dw -16
times 8 dw 3
times 8 dw 0
times 8 dw -6
times 8 dw 50
times 8 dw 93
times 8 dw -9
times 8 dw 0
times 8 dw 1
times 8 dw -8
times 8 dw 36
times 8 dw 108
times 8 dw -11
times 8 dw 2
times 8 dw 0
times 8 dw -1
times 8 dw 12
times 8 dw 123
times 8 dw -6
times 8 dw 0