generic-library/vpx
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity
c5f890af2c
vpx/vp8/encoder/x86/variance_impl_mmx.asm

852 lines
33 KiB
NASM
Raw Normal View History

Initial WebM release
2010-05-18 17:58:33 +02:00
;
Use WebM in copyright notice for consistency Changes 'The VP8 project' to 'The WebM project', for consistency with other webmproject.org repositories. Fixes issue #97. Change-Id: I37c13ed5fbdb9d334ceef71c6350e9febed9bbba
2010-09-09 14:16:39 +02:00
; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
Initial WebM release
2010-05-18 17:58:33 +02:00
;
cosmetics: trim trailing whitespace When the license headers were updated, they accidentally contained trailing whitespace, so unfortunately we have to touch all the files again. Change-Id: I236c05fade06589e417179c0444cb39b09e4200d
2010-06-18 18:39:21 +02:00
; Use of this source code is governed by a BSD-style license
LICENSE: update with latest text Change-Id: Ieebea089095d9073b3a94932791099f614ce120c
2010-06-04 22:19:40 +02:00
; that can be found in the LICENSE file in the root of the source
; tree. An additional intellectual property rights grant can be found
cosmetics: trim trailing whitespace When the license headers were updated, they accidentally contained trailing whitespace, so unfortunately we have to touch all the files again. Change-Id: I236c05fade06589e417179c0444cb39b09e4200d
2010-06-18 18:39:21 +02:00
; in the file PATENTS. All contributing project authors may
LICENSE: update with latest text Change-Id: Ieebea089095d9073b3a94932791099f614ce120c
2010-06-04 22:19:40 +02:00
; be found in the AUTHORS file in the root of the source tree.
Initial WebM release
2010-05-18 17:58:33 +02:00
;
%include "vpx_ports/x86_abi_support.asm"
;unsigned int vp8_get_mb_ss_mmx( short *src_ptr )
global sym(vp8_get_mb_ss_mmx)
sym(vp8_get_mb_ss_mmx):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 7
GET_GOT rbx
push rsi
push rdi
sub rsp, 8
; end prolog
mov rax, arg(0) ;src_ptr
mov rcx, 16
pxor mm4, mm4
Use local labels for jumps/loops in x86 assembly. Prepend . to local labels in assembly code. This allows non unique labels within a file. Also makes profiling information more informative by keeping the function name with the loop name. Change-Id: I7a983cb3a5ba2413d5dafd0a37936b268fb9e37f
2011-08-23 00:29:41 +02:00
.NEXTROW:
Initial WebM release
2010-05-18 17:58:33 +02:00
movq mm0, [rax]
movq mm1, [rax+8]
movq mm2, [rax+16]
movq mm3, [rax+24]
pmaddwd mm0, mm0
pmaddwd mm1, mm1
pmaddwd mm2, mm2
pmaddwd mm3, mm3
paddd mm4, mm0
paddd mm4, mm1
paddd mm4, mm2
paddd mm4, mm3
add rax, 32
dec rcx
Use local labels for jumps/loops in x86 assembly. Prepend . to local labels in assembly code. This allows non unique labels within a file. Also makes profiling information more informative by keeping the function name with the loop name. Change-Id: I7a983cb3a5ba2413d5dafd0a37936b268fb9e37f
2011-08-23 00:29:41 +02:00
ja .NEXTROW
Initial WebM release
2010-05-18 17:58:33 +02:00
movq QWORD PTR [rsp], mm4
;return sum[0]+sum[1];
movsxd rax, dword ptr [rsp]
movsxd rcx, dword ptr [rsp+4]
add rax, rcx
; begin epilog
add rsp, 8
pop rdi
pop rsi
RESTORE_GOT
UNSHADOW_ARGS
pop rbp
ret
;unsigned int vp8_get8x8var_mmx
;(
; unsigned char *src_ptr,
; int source_stride,
; unsigned char *ref_ptr,
; int recon_stride,
; unsigned int *SSE,
; int *Sum
;)
global sym(vp8_get8x8var_mmx)
sym(vp8_get8x8var_mmx):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 6
push rsi
push rdi
push rbx
sub rsp, 16
; end prolog
pxor mm5, mm5 ; Blank mmx6
pxor mm6, mm6 ; Blank mmx7
pxor mm7, mm7 ; Blank mmx7
mov rax, arg(0) ;[src_ptr] ; Load base addresses
mov rbx, arg(2) ;[ref_ptr]
movsxd rcx, dword ptr arg(1) ;[source_stride]
movsxd rdx, dword ptr arg(3) ;[recon_stride]
; Row 1
movq mm0, [rax] ; Copy eight bytes to mm0
movq mm1, [rbx] ; Copy eight bytes to mm1
movq mm2, mm0 ; Take copies
movq mm3, mm1 ; Take copies
punpcklbw mm0, mm6 ; unpack to higher prrcision
punpcklbw mm1, mm6
punpckhbw mm2, mm6 ; unpack to higher prrcision
punpckhbw mm3, mm6
psubsw mm0, mm1 ; A-B (low order) to MM0
psubsw mm2, mm3 ; A-B (high order) to MM2
paddw mm5, mm0 ; accumulate differences in mm5
paddw mm5, mm2 ; accumulate differences in mm5
pmaddwd mm0, mm0 ; square and accumulate
pmaddwd mm2, mm2 ; square and accumulate
add rbx,rdx ; Inc pointer into ref data
add rax,rcx ; Inc pointer into the new data
movq mm1, [rbx] ; Copy eight bytes to mm1
paddd mm7, mm0 ; accumulate in mm7
paddd mm7, mm2 ; accumulate in mm7
; Row 2
movq mm0, [rax] ; Copy eight bytes to mm0
movq mm2, mm0 ; Take copies
movq mm3, mm1 ; Take copies
punpcklbw mm0, mm6 ; unpack to higher prrcision
punpcklbw mm1, mm6
punpckhbw mm2, mm6 ; unpack to higher prrcision
punpckhbw mm3, mm6
psubsw mm0, mm1 ; A-B (low order) to MM0
psubsw mm2, mm3 ; A-B (high order) to MM2
paddw mm5, mm0 ; accumulate differences in mm5
paddw mm5, mm2 ; accumulate differences in mm5
pmaddwd mm0, mm0 ; square and accumulate
pmaddwd mm2, mm2 ; square and accumulate
add rbx,rdx ; Inc pointer into ref data
add rax,rcx ; Inc pointer into the new data
movq mm1, [rbx] ; Copy eight bytes to mm1
paddd mm7, mm0 ; accumulate in mm7
paddd mm7, mm2 ; accumulate in mm7
; Row 3
movq mm0, [rax] ; Copy eight bytes to mm0
movq mm2, mm0 ; Take copies
movq mm3, mm1 ; Take copies
punpcklbw mm0, mm6 ; unpack to higher prrcision
punpcklbw mm1, mm6
punpckhbw mm2, mm6 ; unpack to higher prrcision
punpckhbw mm3, mm6
psubsw mm0, mm1 ; A-B (low order) to MM0
psubsw mm2, mm3 ; A-B (high order) to MM2
paddw mm5, mm0 ; accumulate differences in mm5
paddw mm5, mm2 ; accumulate differences in mm5
pmaddwd mm0, mm0 ; square and accumulate
pmaddwd mm2, mm2 ; square and accumulate
add rbx,rdx ; Inc pointer into ref data
add rax,rcx ; Inc pointer into the new data
movq mm1, [rbx] ; Copy eight bytes to mm1
paddd mm7, mm0 ; accumulate in mm7
paddd mm7, mm2 ; accumulate in mm7
; Row 4
movq mm0, [rax] ; Copy eight bytes to mm0
movq mm2, mm0 ; Take copies
movq mm3, mm1 ; Take copies
punpcklbw mm0, mm6 ; unpack to higher prrcision
punpcklbw mm1, mm6
punpckhbw mm2, mm6 ; unpack to higher prrcision
punpckhbw mm3, mm6
psubsw mm0, mm1 ; A-B (low order) to MM0
psubsw mm2, mm3 ; A-B (high order) to MM2
paddw mm5, mm0 ; accumulate differences in mm5
paddw mm5, mm2 ; accumulate differences in mm5
pmaddwd mm0, mm0 ; square and accumulate
pmaddwd mm2, mm2 ; square and accumulate
add rbx,rdx ; Inc pointer into ref data
add rax,rcx ; Inc pointer into the new data
movq mm1, [rbx] ; Copy eight bytes to mm1
paddd mm7, mm0 ; accumulate in mm7
paddd mm7, mm2 ; accumulate in mm7
; Row 5
movq mm0, [rax] ; Copy eight bytes to mm0
movq mm2, mm0 ; Take copies
movq mm3, mm1 ; Take copies
punpcklbw mm0, mm6 ; unpack to higher prrcision
punpcklbw mm1, mm6
punpckhbw mm2, mm6 ; unpack to higher prrcision
punpckhbw mm3, mm6
psubsw mm0, mm1 ; A-B (low order) to MM0
psubsw mm2, mm3 ; A-B (high order) to MM2
paddw mm5, mm0 ; accumulate differences in mm5
paddw mm5, mm2 ; accumulate differences in mm5
pmaddwd mm0, mm0 ; square and accumulate
pmaddwd mm2, mm2 ; square and accumulate
add rbx,rdx ; Inc pointer into ref data
add rax,rcx ; Inc pointer into the new data
movq mm1, [rbx] ; Copy eight bytes to mm1
; movq mm4, [rbx + rdx]
paddd mm7, mm0 ; accumulate in mm7
paddd mm7, mm2 ; accumulate in mm7
; Row 6
movq mm0, [rax] ; Copy eight bytes to mm0
movq mm2, mm0 ; Take copies
movq mm3, mm1 ; Take copies
punpcklbw mm0, mm6 ; unpack to higher prrcision
punpcklbw mm1, mm6
punpckhbw mm2, mm6 ; unpack to higher prrcision
punpckhbw mm3, mm6
psubsw mm0, mm1 ; A-B (low order) to MM0
psubsw mm2, mm3 ; A-B (high order) to MM2
paddw mm5, mm0 ; accumulate differences in mm5
paddw mm5, mm2 ; accumulate differences in mm5
pmaddwd mm0, mm0 ; square and accumulate
pmaddwd mm2, mm2 ; square and accumulate
add rbx,rdx ; Inc pointer into ref data
add rax,rcx ; Inc pointer into the new data
movq mm1, [rbx] ; Copy eight bytes to mm1
paddd mm7, mm0 ; accumulate in mm7
paddd mm7, mm2 ; accumulate in mm7
; Row 7
movq mm0, [rax] ; Copy eight bytes to mm0
movq mm2, mm0 ; Take copies
movq mm3, mm1 ; Take copies
punpcklbw mm0, mm6 ; unpack to higher prrcision
punpcklbw mm1, mm6
punpckhbw mm2, mm6 ; unpack to higher prrcision
punpckhbw mm3, mm6
psubsw mm0, mm1 ; A-B (low order) to MM0
psubsw mm2, mm3 ; A-B (high order) to MM2
paddw mm5, mm0 ; accumulate differences in mm5
paddw mm5, mm2 ; accumulate differences in mm5
pmaddwd mm0, mm0 ; square and accumulate
pmaddwd mm2, mm2 ; square and accumulate
add rbx,rdx ; Inc pointer into ref data
add rax,rcx ; Inc pointer into the new data
movq mm1, [rbx] ; Copy eight bytes to mm1
paddd mm7, mm0 ; accumulate in mm7
paddd mm7, mm2 ; accumulate in mm7
; Row 8
movq mm0, [rax] ; Copy eight bytes to mm0
movq mm2, mm0 ; Take copies
movq mm3, mm1 ; Take copies
punpcklbw mm0, mm6 ; unpack to higher prrcision
punpcklbw mm1, mm6
punpckhbw mm2, mm6 ; unpack to higher prrcision
punpckhbw mm3, mm6
psubsw mm0, mm1 ; A-B (low order) to MM0
psubsw mm2, mm3 ; A-B (high order) to MM2
paddw mm5, mm0 ; accumulate differences in mm5
paddw mm5, mm2 ; accumulate differences in mm5
pmaddwd mm0, mm0 ; square and accumulate
pmaddwd mm2, mm2 ; square and accumulate
add rbx,rdx ; Inc pointer into ref data
add rax,rcx ; Inc pointer into the new data
paddd mm7, mm0 ; accumulate in mm7
paddd mm7, mm2 ; accumulate in mm7
; Now accumulate the final results.
movq QWORD PTR [rsp+8], mm5 ; copy back accumulated results into normal memory
movq QWORD PTR [rsp], mm7 ; copy back accumulated results into normal memory
movsx rdx, WORD PTR [rsp+8]
movsx rcx, WORD PTR [rsp+10]
movsx rbx, WORD PTR [rsp+12]
movsx rax, WORD PTR [rsp+14]
add rdx, rcx
add rbx, rax
add rdx, rbx ;XSum
movsxd rax, DWORD PTR [rsp]
movsxd rcx, DWORD PTR [rsp+4]
add rax, rcx ;XXSum
mov rsi, arg(4) ;SSE
mov rdi, arg(5) ;Sum
mov dword ptr [rsi], eax
mov dword ptr [rdi], edx
xor rax, rax ; return 0
; begin epilog
add rsp, 16
pop rbx
pop rdi
pop rsi
UNSHADOW_ARGS
pop rbp
ret
;unsigned int
;vp8_get4x4var_mmx
;(
; unsigned char *src_ptr,
; int source_stride,
; unsigned char *ref_ptr,
; int recon_stride,
; unsigned int *SSE,
; int *Sum
;)
global sym(vp8_get4x4var_mmx)
sym(vp8_get4x4var_mmx):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 6
push rsi
push rdi
push rbx
sub rsp, 16
; end prolog
pxor mm5, mm5 ; Blank mmx6
pxor mm6, mm6 ; Blank mmx7
pxor mm7, mm7 ; Blank mmx7
mov rax, arg(0) ;[src_ptr] ; Load base addresses
mov rbx, arg(2) ;[ref_ptr]
movsxd rcx, dword ptr arg(1) ;[source_stride]
movsxd rdx, dword ptr arg(3) ;[recon_stride]
; Row 1
movq mm0, [rax] ; Copy eight bytes to mm0
movq mm1, [rbx] ; Copy eight bytes to mm1
punpcklbw mm0, mm6 ; unpack to higher prrcision
punpcklbw mm1, mm6
psubsw mm0, mm1 ; A-B (low order) to MM0
paddw mm5, mm0 ; accumulate differences in mm5
pmaddwd mm0, mm0 ; square and accumulate
add rbx,rdx ; Inc pointer into ref data
add rax,rcx ; Inc pointer into the new data
movq mm1, [rbx] ; Copy eight bytes to mm1
paddd mm7, mm0 ; accumulate in mm7
; Row 2
movq mm0, [rax] ; Copy eight bytes to mm0
punpcklbw mm0, mm6 ; unpack to higher prrcision
punpcklbw mm1, mm6
psubsw mm0, mm1 ; A-B (low order) to MM0
paddw mm5, mm0 ; accumulate differences in mm5
pmaddwd mm0, mm0 ; square and accumulate
add rbx,rdx ; Inc pointer into ref data
add rax,rcx ; Inc pointer into the new data
movq mm1, [rbx] ; Copy eight bytes to mm1
paddd mm7, mm0 ; accumulate in mm7
; Row 3
movq mm0, [rax] ; Copy eight bytes to mm0
punpcklbw mm0, mm6 ; unpack to higher prrcision
punpcklbw mm1, mm6
psubsw mm0, mm1 ; A-B (low order) to MM0
paddw mm5, mm0 ; accumulate differences in mm5
pmaddwd mm0, mm0 ; square and accumulate
add rbx,rdx ; Inc pointer into ref data
add rax,rcx ; Inc pointer into the new data
movq mm1, [rbx] ; Copy eight bytes to mm1
paddd mm7, mm0 ; accumulate in mm7
; Row 4
movq mm0, [rax] ; Copy eight bytes to mm0
punpcklbw mm0, mm6 ; unpack to higher prrcision
punpcklbw mm1, mm6
psubsw mm0, mm1 ; A-B (low order) to MM0
paddw mm5, mm0 ; accumulate differences in mm5
pmaddwd mm0, mm0 ; square and accumulate
paddd mm7, mm0 ; accumulate in mm7
; Now accumulate the final results.
movq QWORD PTR [rsp+8], mm5 ; copy back accumulated results into normal memory
movq QWORD PTR [rsp], mm7 ; copy back accumulated results into normal memory
movsx rdx, WORD PTR [rsp+8]
movsx rcx, WORD PTR [rsp+10]
movsx rbx, WORD PTR [rsp+12]
movsx rax, WORD PTR [rsp+14]
add rdx, rcx
add rbx, rax
add rdx, rbx ;XSum
movsxd rax, DWORD PTR [rsp]
movsxd rcx, DWORD PTR [rsp+4]
add rax, rcx ;XXSum
mov rsi, arg(4) ;SSE
mov rdi, arg(5) ;Sum
mov dword ptr [rsi], eax
mov dword ptr [rdi], edx
xor rax, rax ; return 0
; begin epilog
add rsp, 16
pop rbx
pop rdi
pop rsi
UNSHADOW_ARGS
pop rbp
ret
;unsigned int
;vp8_get4x4sse_cs_mmx
;(
; unsigned char *src_ptr,
; int source_stride,
; unsigned char *ref_ptr,
; int recon_stride
;)
global sym(vp8_get4x4sse_cs_mmx)
sym(vp8_get4x4sse_cs_mmx):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 4
push rsi
push rdi
push rbx
; end prolog
pxor mm6, mm6 ; Blank mmx7
pxor mm7, mm7 ; Blank mmx7
mov rax, arg(0) ;[src_ptr] ; Load base addresses
mov rbx, arg(2) ;[ref_ptr]
movsxd rcx, dword ptr arg(1) ;[source_stride]
movsxd rdx, dword ptr arg(3) ;[recon_stride]
; Row 1
movd mm0, [rax] ; Copy eight bytes to mm0
movd mm1, [rbx] ; Copy eight bytes to mm1
punpcklbw mm0, mm6 ; unpack to higher prrcision
punpcklbw mm1, mm6
psubsw mm0, mm1 ; A-B (low order) to MM0
pmaddwd mm0, mm0 ; square and accumulate
add rbx,rdx ; Inc pointer into ref data
add rax,rcx ; Inc pointer into the new data
movd mm1, [rbx] ; Copy eight bytes to mm1
paddd mm7, mm0 ; accumulate in mm7
; Row 2
movd mm0, [rax] ; Copy eight bytes to mm0
punpcklbw mm0, mm6 ; unpack to higher prrcision
punpcklbw mm1, mm6
psubsw mm0, mm1 ; A-B (low order) to MM0
pmaddwd mm0, mm0 ; square and accumulate
add rbx,rdx ; Inc pointer into ref data
add rax,rcx ; Inc pointer into the new data
movd mm1, [rbx] ; Copy eight bytes to mm1
paddd mm7, mm0 ; accumulate in mm7
; Row 3
movd mm0, [rax] ; Copy eight bytes to mm0
punpcklbw mm1, mm6
punpcklbw mm0, mm6 ; unpack to higher prrcision
psubsw mm0, mm1 ; A-B (low order) to MM0
pmaddwd mm0, mm0 ; square and accumulate
add rbx,rdx ; Inc pointer into ref data
add rax,rcx ; Inc pointer into the new data
movd mm1, [rbx] ; Copy eight bytes to mm1
paddd mm7, mm0 ; accumulate in mm7
; Row 4
movd mm0, [rax] ; Copy eight bytes to mm0
punpcklbw mm0, mm6 ; unpack to higher prrcision
punpcklbw mm1, mm6
psubsw mm0, mm1 ; A-B (low order) to MM0
pmaddwd mm0, mm0 ; square and accumulate
paddd mm7, mm0 ; accumulate in mm7
movq mm0, mm7 ;
psrlq mm7, 32
paddd mm0, mm7
nasm: match instruction length (movd/movq) to parameters nasm requires the instruction length (movd/movq) to match to its parameters. I find it more clear to really use 64bit instructions when we use 64bit registers in the assembly. Provide nasm compatibility. No binary change by this patch with yasm on {x86_64,i686}-fedora13-linux-gnu. Few longer opcodes with nasm on {x86_64,i686}-fedora13-linux-gnu have been checked as safe. Change-Id: Id9b1a5cdfb1bc05697e523c317a296df43d42a91
2010-10-04 23:19:33 +02:00
movq rax, mm0
Initial WebM release
2010-05-18 17:58:33 +02:00
; begin epilog
pop rbx
pop rdi
pop rsi
UNSHADOW_ARGS
pop rbp
ret
%define mmx_filter_shift 7
;void vp8_filter_block2d_bil4x4_var_mmx
;(
; unsigned char *ref_ptr,
; int ref_pixels_per_line,
; unsigned char *src_ptr,
; int src_pixels_per_line,
; unsigned short *HFilter,
; unsigned short *VFilter,
; int *sum,
; unsigned int *sumsquared
;)
global sym(vp8_filter_block2d_bil4x4_var_mmx)
sym(vp8_filter_block2d_bil4x4_var_mmx):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 8
GET_GOT rbx
push rsi
push rdi
sub rsp, 16
; end prolog
pxor mm6, mm6 ;
pxor mm7, mm7 ;
mov rax, arg(4) ;HFilter ;
mov rdx, arg(5) ;VFilter ;
mov rsi, arg(0) ;ref_ptr ;
mov rdi, arg(2) ;src_ptr ;
mov rcx, 4 ;
pxor mm0, mm0 ;
movd mm1, [rsi] ;
movd mm3, [rsi+1] ;
punpcklbw mm1, mm0 ;
pmullw mm1, [rax] ;
punpcklbw mm3, mm0 ;
pmullw mm3, [rax+8] ;
paddw mm1, mm3 ;
nasm: address labels 'rel label' vice 'wrt rip' nasm does not support `label wrt rip', it requires `rel label'. It is still fully compatible with yasm. Provide nasm compatibility. No binary change by this patch with yasm on {x86_64,i686}-fedora13-linux-gnu. Few longer opcodes with nasm on {x86_64,i686}-fedora13-linux-gnu have been checked as safe. Change-Id: I488773a4e930a56e43b0cc72d867ee5291215f50
2010-10-04 23:18:58 +02:00
paddw mm1, [GLOBAL(mmx_bi_rd)] ;
Initial WebM release
2010-05-18 17:58:33 +02:00
psraw mm1, mmx_filter_shift ;
movq mm5, mm1
%if ABI_IS_32BIT
add rsi, dword ptr arg(1) ;ref_pixels_per_line ;
%else
movsxd r8, dword ptr arg(1) ;ref_pixels_per_line ;
add rsi, r8
%endif
Use local labels for jumps/loops in x86 assembly. Prepend . to local labels in assembly code. This allows non unique labels within a file. Also makes profiling information more informative by keeping the function name with the loop name. Change-Id: I7a983cb3a5ba2413d5dafd0a37936b268fb9e37f
2011-08-23 00:29:41 +02:00
.filter_block2d_bil4x4_var_mmx_loop:
Initial WebM release
2010-05-18 17:58:33 +02:00
movd mm1, [rsi] ;
movd mm3, [rsi+1] ;
punpcklbw mm1, mm0 ;
pmullw mm1, [rax] ;
punpcklbw mm3, mm0 ;
pmullw mm3, [rax+8] ;
paddw mm1, mm3 ;
nasm: address labels 'rel label' vice 'wrt rip' nasm does not support `label wrt rip', it requires `rel label'. It is still fully compatible with yasm. Provide nasm compatibility. No binary change by this patch with yasm on {x86_64,i686}-fedora13-linux-gnu. Few longer opcodes with nasm on {x86_64,i686}-fedora13-linux-gnu have been checked as safe. Change-Id: I488773a4e930a56e43b0cc72d867ee5291215f50
2010-10-04 23:18:58 +02:00
paddw mm1, [GLOBAL(mmx_bi_rd)] ;
Initial WebM release
2010-05-18 17:58:33 +02:00
psraw mm1, mmx_filter_shift ;
movq mm3, mm5 ;
movq mm5, mm1 ;
pmullw mm3, [rdx] ;
pmullw mm1, [rdx+8] ;
paddw mm1, mm3 ;
nasm: address labels 'rel label' vice 'wrt rip' nasm does not support `label wrt rip', it requires `rel label'. It is still fully compatible with yasm. Provide nasm compatibility. No binary change by this patch with yasm on {x86_64,i686}-fedora13-linux-gnu. Few longer opcodes with nasm on {x86_64,i686}-fedora13-linux-gnu have been checked as safe. Change-Id: I488773a4e930a56e43b0cc72d867ee5291215f50
2010-10-04 23:18:58 +02:00
paddw mm1, [GLOBAL(mmx_bi_rd)] ;
Initial WebM release
2010-05-18 17:58:33 +02:00
psraw mm1, mmx_filter_shift ;
movd mm3, [rdi] ;
punpcklbw mm3, mm0 ;
psubw mm1, mm3 ;
paddw mm6, mm1 ;
pmaddwd mm1, mm1 ;
paddd mm7, mm1 ;
%if ABI_IS_32BIT
add rsi, dword ptr arg(1) ;ref_pixels_per_line ;
add rdi, dword ptr arg(3) ;src_pixels_per_line ;
%else
movsxd r8, dword ptr arg(1) ;ref_pixels_per_line
movsxd r9, dword ptr arg(3) ;src_pixels_per_line
add rsi, r8
add rdi, r9
%endif
sub rcx, 1 ;
Use local labels for jumps/loops in x86 assembly. Prepend . to local labels in assembly code. This allows non unique labels within a file. Also makes profiling information more informative by keeping the function name with the loop name. Change-Id: I7a983cb3a5ba2413d5dafd0a37936b268fb9e37f
2011-08-23 00:29:41 +02:00
jnz .filter_block2d_bil4x4_var_mmx_loop ;
Initial WebM release
2010-05-18 17:58:33 +02:00
pxor mm3, mm3 ;
pxor mm2, mm2 ;
punpcklwd mm2, mm6 ;
punpckhwd mm3, mm6 ;
paddd mm2, mm3 ;
movq mm6, mm2 ;
psrlq mm6, 32 ;
paddd mm2, mm6 ;
psrad mm2, 16 ;
movq mm4, mm7 ;
psrlq mm4, 32 ;
paddd mm4, mm7 ;
mov rdi, arg(6) ;sum
mov rsi, arg(7) ;sumsquared
movd dword ptr [rdi], mm2 ;
movd dword ptr [rsi], mm4 ;
; begin epilog
add rsp, 16
pop rdi
pop rsi
RESTORE_GOT
UNSHADOW_ARGS
pop rbp
ret
;void vp8_filter_block2d_bil_var_mmx
;(
; unsigned char *ref_ptr,
; int ref_pixels_per_line,
; unsigned char *src_ptr,
; int src_pixels_per_line,
; unsigned int Height,
; unsigned short *HFilter,
; unsigned short *VFilter,
; int *sum,
; unsigned int *sumsquared
;)
global sym(vp8_filter_block2d_bil_var_mmx)
sym(vp8_filter_block2d_bil_var_mmx):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 9
GET_GOT rbx
push rsi
push rdi
sub rsp, 16
; end prolog
pxor mm6, mm6 ;
pxor mm7, mm7 ;
mov rax, arg(5) ;HFilter ;
mov rdx, arg(6) ;VFilter ;
mov rsi, arg(0) ;ref_ptr ;
mov rdi, arg(2) ;src_ptr ;
movsxd rcx, dword ptr arg(4) ;Height ;
pxor mm0, mm0 ;
movq mm1, [rsi] ;
movq mm3, [rsi+1] ;
movq mm2, mm1 ;
movq mm4, mm3 ;
punpcklbw mm1, mm0 ;
punpckhbw mm2, mm0 ;
pmullw mm1, [rax] ;
pmullw mm2, [rax] ;
punpcklbw mm3, mm0 ;
punpckhbw mm4, mm0 ;
pmullw mm3, [rax+8] ;
pmullw mm4, [rax+8] ;
paddw mm1, mm3 ;
paddw mm2, mm4 ;
nasm: address labels 'rel label' vice 'wrt rip' nasm does not support `label wrt rip', it requires `rel label'. It is still fully compatible with yasm. Provide nasm compatibility. No binary change by this patch with yasm on {x86_64,i686}-fedora13-linux-gnu. Few longer opcodes with nasm on {x86_64,i686}-fedora13-linux-gnu have been checked as safe. Change-Id: I488773a4e930a56e43b0cc72d867ee5291215f50
2010-10-04 23:18:58 +02:00
paddw mm1, [GLOBAL(mmx_bi_rd)] ;
Initial WebM release
2010-05-18 17:58:33 +02:00
psraw mm1, mmx_filter_shift ;
nasm: address labels 'rel label' vice 'wrt rip' nasm does not support `label wrt rip', it requires `rel label'. It is still fully compatible with yasm. Provide nasm compatibility. No binary change by this patch with yasm on {x86_64,i686}-fedora13-linux-gnu. Few longer opcodes with nasm on {x86_64,i686}-fedora13-linux-gnu have been checked as safe. Change-Id: I488773a4e930a56e43b0cc72d867ee5291215f50
2010-10-04 23:18:58 +02:00
paddw mm2, [GLOBAL(mmx_bi_rd)] ;
Initial WebM release
2010-05-18 17:58:33 +02:00
psraw mm2, mmx_filter_shift ;
movq mm5, mm1
packuswb mm5, mm2 ;
%if ABI_IS_32BIT
add rsi, dword ptr arg(1) ;ref_pixels_per_line
%else
movsxd r8, dword ptr arg(1) ;ref_pixels_per_line
add rsi, r8
%endif
Use local labels for jumps/loops in x86 assembly. Prepend . to local labels in assembly code. This allows non unique labels within a file. Also makes profiling information more informative by keeping the function name with the loop name. Change-Id: I7a983cb3a5ba2413d5dafd0a37936b268fb9e37f
2011-08-23 00:29:41 +02:00
.filter_block2d_bil_var_mmx_loop:
Initial WebM release
2010-05-18 17:58:33 +02:00
movq mm1, [rsi] ;
movq mm3, [rsi+1] ;
movq mm2, mm1 ;
movq mm4, mm3 ;
punpcklbw mm1, mm0 ;
punpckhbw mm2, mm0 ;
pmullw mm1, [rax] ;
pmullw mm2, [rax] ;
punpcklbw mm3, mm0 ;
punpckhbw mm4, mm0 ;
pmullw mm3, [rax+8] ;
pmullw mm4, [rax+8] ;
paddw mm1, mm3 ;
paddw mm2, mm4 ;
nasm: address labels 'rel label' vice 'wrt rip' nasm does not support `label wrt rip', it requires `rel label'. It is still fully compatible with yasm. Provide nasm compatibility. No binary change by this patch with yasm on {x86_64,i686}-fedora13-linux-gnu. Few longer opcodes with nasm on {x86_64,i686}-fedora13-linux-gnu have been checked as safe. Change-Id: I488773a4e930a56e43b0cc72d867ee5291215f50
2010-10-04 23:18:58 +02:00
paddw mm1, [GLOBAL(mmx_bi_rd)] ;
Initial WebM release
2010-05-18 17:58:33 +02:00
psraw mm1, mmx_filter_shift ;
nasm: address labels 'rel label' vice 'wrt rip' nasm does not support `label wrt rip', it requires `rel label'. It is still fully compatible with yasm. Provide nasm compatibility. No binary change by this patch with yasm on {x86_64,i686}-fedora13-linux-gnu. Few longer opcodes with nasm on {x86_64,i686}-fedora13-linux-gnu have been checked as safe. Change-Id: I488773a4e930a56e43b0cc72d867ee5291215f50
2010-10-04 23:18:58 +02:00
paddw mm2, [GLOBAL(mmx_bi_rd)] ;
Initial WebM release
2010-05-18 17:58:33 +02:00
psraw mm2, mmx_filter_shift ;
movq mm3, mm5 ;
movq mm4, mm5 ;
punpcklbw mm3, mm0 ;
punpckhbw mm4, mm0 ;
movq mm5, mm1 ;
packuswb mm5, mm2 ;
pmullw mm3, [rdx] ;
pmullw mm4, [rdx] ;
pmullw mm1, [rdx+8] ;
pmullw mm2, [rdx+8] ;
paddw mm1, mm3 ;
paddw mm2, mm4 ;
nasm: address labels 'rel label' vice 'wrt rip' nasm does not support `label wrt rip', it requires `rel label'. It is still fully compatible with yasm. Provide nasm compatibility. No binary change by this patch with yasm on {x86_64,i686}-fedora13-linux-gnu. Few longer opcodes with nasm on {x86_64,i686}-fedora13-linux-gnu have been checked as safe. Change-Id: I488773a4e930a56e43b0cc72d867ee5291215f50
2010-10-04 23:18:58 +02:00
paddw mm1, [GLOBAL(mmx_bi_rd)] ;
paddw mm2, [GLOBAL(mmx_bi_rd)] ;
Initial WebM release
2010-05-18 17:58:33 +02:00
psraw mm1, mmx_filter_shift ;
psraw mm2, mmx_filter_shift ;
movq mm3, [rdi] ;
movq mm4, mm3 ;
punpcklbw mm3, mm0 ;
punpckhbw mm4, mm0 ;
psubw mm1, mm3 ;
psubw mm2, mm4 ;
paddw mm6, mm1 ;
pmaddwd mm1, mm1 ;
paddw mm6, mm2 ;
pmaddwd mm2, mm2 ;
paddd mm7, mm1 ;
paddd mm7, mm2 ;
%if ABI_IS_32BIT
add rsi, dword ptr arg(1) ;ref_pixels_per_line ;
add rdi, dword ptr arg(3) ;src_pixels_per_line ;
%else
movsxd r8, dword ptr arg(1) ;ref_pixels_per_line ;
movsxd r9, dword ptr arg(3) ;src_pixels_per_line ;
add rsi, r8
add rdi, r9
%endif
sub rcx, 1 ;
Use local labels for jumps/loops in x86 assembly. Prepend . to local labels in assembly code. This allows non unique labels within a file. Also makes profiling information more informative by keeping the function name with the loop name. Change-Id: I7a983cb3a5ba2413d5dafd0a37936b268fb9e37f
2011-08-23 00:29:41 +02:00
jnz .filter_block2d_bil_var_mmx_loop ;
Initial WebM release
2010-05-18 17:58:33 +02:00
pxor mm3, mm3 ;
pxor mm2, mm2 ;
punpcklwd mm2, mm6 ;
punpckhwd mm3, mm6 ;
paddd mm2, mm3 ;
movq mm6, mm2 ;
psrlq mm6, 32 ;
paddd mm2, mm6 ;
psrad mm2, 16 ;
movq mm4, mm7 ;
psrlq mm4, 32 ;
paddd mm4, mm7 ;
mov rdi, arg(7) ;sum
mov rsi, arg(8) ;sumsquared
movd dword ptr [rdi], mm2 ;
movd dword ptr [rsi], mm4 ;
; begin epilog
add rsp, 16
pop rdi
pop rsi
RESTORE_GOT
UNSHADOW_ARGS
pop rbp
ret
SECTION_RODATA
;short mmx_bi_rd[4] = { 64, 64, 64, 64};
align 16
mmx_bi_rd:
times 4 dw 64
Reference in New Issue Copy Permalink