ryan/mozjpeg
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge branch 'master' into dev

Browse Source
This commit is contained in:
DRC
2019-10-24 02:13:23 -05:00
parent 2ae2bd66b4 55de97207d
commit 95f4d6ef8b
99 changed files with 168 additions and 331 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
ChangeLog.md
View File

@@ -46,6 +46,17 @@ longer supports 32-bit Java virtual machines. Oracle no longer provides a
systems) is long obsolete. systems) is long obsolete.
2.0.4
=====
### Significant changes relative to 2.0.3:
1. Fixed a regression in the Windows packaging system (introduced by
2.0 beta1[2]) whereby, if both the 64-bit libjpeg-turbo SDK for GCC and the
64-bit libjpeg-turbo SDK for Visual C++ were installed on the same system, only
one of them could be uninstalled.
2.0.3 2.0.3
===== =====

16
appveyor.yml
View File

@@ -8,11 +8,23 @@ install:
7z x c:\installers\nasm-2.10.01-win32.zip -oc:\ > c:\installers\nasm.install.log 7z x c:\installers\nasm-2.10.01-win32.zip -oc:\ > c:\installers\nasm.install.log
if not exist c:\installers\i686-6.4.0-release-posix-dwarf-rt_v5-rev0.7z curl -fSL -o c:\installers\i686-6.4.0-release-posix-dwarf-rt_v5-rev0.7z "https://sourceforge.net/projects/mingw-w64/files/Toolchains targetting Win32/Personal Builds/mingw-builds/6.4.0/threads-posix/dwarf/i686-6.4.0-release-posix-dwarf-rt_v5-rev0.7z"
md "c:\Program Files (x86)\mingw-w64\i686-6.4.0-posix-dwarf-rt_v5-rev0"
7z x c:\installers\i686-6.4.0-release-posix-dwarf-rt_v5-rev0.7z -o"c:\Program Files (x86)\mingw-w64\i686-6.4.0-posix-dwarf-rt_v5-rev0" > c:\installers\mingw32.install.log
if not exist c:\installers\x86_64-6.4.0-release-posix-seh-rt_v5-rev0.7z curl -fSL -o c:\installers\x86_64-6.4.0-release-posix-seh-rt_v5-rev0.7z "https://sourceforge.net/projects/mingw-w64/files/Toolchains targetting Win64/Personal Builds/mingw-builds/6.4.0/threads-posix/seh/x86_64-6.4.0-release-posix-seh-rt_v5-rev0.7z"
md "c:\Program Files\mingw-w64\x86_64-6.4.0-posix-seh-rt_v5-rev0"
7z x c:\installers\x86_64-6.4.0-release-posix-seh-rt_v5-rev0.7z -o"c:\Program Files\mingw-w64\x86_64-6.4.0-posix-seh-rt_v5-rev0" > c:\installers\mingw64.install.log
set INCLUDE=c:\Program Files (x86)\Microsoft Visual Studio 10.0\VC\include;c:\Program Files (x86)\Microsoft SDKs\Windows\v7.1A\include set INCLUDE=c:\Program Files (x86)\Microsoft Visual Studio 10.0\VC\include;c:\Program Files (x86)\Microsoft SDKs\Windows\v7.1A\include
set LIB=c:\Program Files (x86)\Microsoft Visual Studio 10.0\VC\lib\amd64;c:\Program Files (x86)\Microsoft SDKs\Windows\v7.1A\lib\x64 set LIB=c:\Program Files (x86)\Microsoft Visual Studio 10.0\VC\lib\amd64;c:\Program Files (x86)\Microsoft SDKs\Windows\v7.1A\lib\x64
set PATH=c:\nasm-2.10.01;c:\Program Files (x86)\NSIS;c:\msys64\mingw32\bin;c:\msys64\usr\bin;c:\Program Files (x86)\Microsoft Visual Studio 10.0\VC\bin\amd64;c:\Program Files (x86)\Microsoft Visual Studio 10.0\Common7\IDE;c:\Program Files (x86)\Microsoft SDKs\Windows\v7.1A\bin\x64;c:\Program Files (x86)\Microsoft SDKs\Windows\v7.1A\bin;%PATH% set PATH=c:\nasm-2.10.01;c:\Program Files (x86)\NSIS;c:\msys64\usr\bin;c:\Program Files (x86)\Microsoft Visual Studio 10.0\VC\bin\amd64;c:\Program Files (x86)\Microsoft Visual Studio 10.0\Common7\IDE;c:\Program Files (x86)\Microsoft SDKs\Windows\v7.1A\bin\x64;c:\Program Files (x86)\Microsoft SDKs\Windows\v7.1A\bin;%PATH%
set MSYSTEM=MINGW32 set MSYSTEM=MINGW32
@@ -24,6 +36,8 @@ install:
cache: cache:
- c:\installers\nasm-2.10.01-win32.zip -> appveyor.yml - c:\installers\nasm-2.10.01-win32.zip -> appveyor.yml
- c:\installers\i686-6.4.0-release-posix-dwarf-rt_v5-rev0.7z -> appveyor.yml
- c:\installers\x86_64-6.4.0-release-posix-seh-rt_v5-rev0.7z -> appveyor.yml
build_script: build_script:
- cmd: >- - cmd: >-

2
cmakescripts/BuildPackages.cmake
View File

@@ -83,7 +83,7 @@ endif()
if(BITS EQUAL 64) if(BITS EQUAL 64)
set(INST_PLATFORM "${INST_PLATFORM} 64-bit") set(INST_PLATFORM "${INST_PLATFORM} 64-bit")
set(INST_NAME ${INST_NAME}64) set(INST_NAME ${INST_NAME}64)
set(INST_REG_NAME ${INST_DIR}64) set(INST_REG_NAME ${INST_REG_NAME}64)
set(INST_DEFS ${INST_DEFS} -DWIN64) set(INST_DEFS ${INST_DEFS} -DWIN64)
endif() endif()

4
jchuff.c
View File

@@ -44,8 +44,8 @@
*/ */
/* NOTE: Both GCC and Clang define __GNUC__ */ /* NOTE: Both GCC and Clang define __GNUC__ */
#if defined __GNUC__ && (defined __arm__ || defined __aarch64__) #if defined(__GNUC__) && (defined(__arm__) || defined(__aarch64__))
#if !defined __thumb__ || defined __thumb2__ #if !defined(__thumb__) || defined(__thumb2__)
#define USE_CLZ_INTRINSIC #define USE_CLZ_INTRINSIC
#endif #endif
#endif #endif

4
jcphuff.c
View File

@@ -52,8 +52,8 @@
*/ */
/* NOTE: Both GCC and Clang define __GNUC__ */ /* NOTE: Both GCC and Clang define __GNUC__ */
#if defined __GNUC__ && (defined __arm__ || defined __aarch64__) #if defined(__GNUC__) && (defined(__arm__) || defined(__aarch64__))
#if !defined __thumb__ || defined __thumb2__ #if !defined(__thumb__) || defined(__thumb2__)
#define USE_CLZ_INTRINSIC #define USE_CLZ_INTRINSIC
#endif #endif
#endif #endif

6
simd/i386/jccolext-avx2.asm
View File

@@ -13,8 +13,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jcolsamp.inc" %include "jcolsamp.inc"
@@ -110,12 +108,12 @@ EXTN(jsimd_rgb_ycc_convert_avx2):
test cl, SIZEOF_BYTE test cl, SIZEOF_BYTE
jz short .column_ld2 jz short .column_ld2
sub ecx, byte SIZEOF_BYTE sub ecx, byte SIZEOF_BYTE
movzx eax, BYTE [esi+ecx] movzx eax, byte [esi+ecx]
.column_ld2: .column_ld2:
test cl, SIZEOF_WORD test cl, SIZEOF_WORD
jz short .column_ld4 jz short .column_ld4
sub ecx, byte SIZEOF_WORD sub ecx, byte SIZEOF_WORD
movzx edx, WORD [esi+ecx] movzx edx, word [esi+ecx]
shl eax, WORD_BIT shl eax, WORD_BIT
or eax, edx or eax, edx
.column_ld4: .column_ld4:

10
simd/i386/jccolext-mmx.asm
View File

@@ -13,8 +13,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jcolsamp.inc" %include "jcolsamp.inc"
@@ -111,13 +109,13 @@ EXTN(jsimd_rgb_ycc_convert_mmx):
jz short .column_ld2 jz short .column_ld2
sub ecx, byte SIZEOF_BYTE sub ecx, byte SIZEOF_BYTE
xor eax, eax xor eax, eax
mov al, BYTE [esi+ecx] mov al, byte [esi+ecx]
.column_ld2: .column_ld2:
test cl, SIZEOF_WORD test cl, SIZEOF_WORD
jz short .column_ld4 jz short .column_ld4
sub ecx, byte SIZEOF_WORD sub ecx, byte SIZEOF_WORD
xor edx, edx xor edx, edx
mov dx, WORD [esi+ecx] mov dx, word [esi+ecx]
shl eax, WORD_BIT shl eax, WORD_BIT
or eax, edx or eax, edx
.column_ld4: .column_ld4:
@@ -127,7 +125,7 @@ EXTN(jsimd_rgb_ycc_convert_mmx):
test cl, SIZEOF_DWORD test cl, SIZEOF_DWORD
jz short .column_ld8 jz short .column_ld8
sub ecx, byte SIZEOF_DWORD sub ecx, byte SIZEOF_DWORD
movd mmG, DWORD [esi+ecx] movd mmG, dword [esi+ecx]
psllq mmA, DWORD_BIT psllq mmA, DWORD_BIT
por mmA, mmG por mmA, mmG
.column_ld8: .column_ld8:
@@ -197,7 +195,7 @@ EXTN(jsimd_rgb_ycc_convert_mmx):
test cl, SIZEOF_MMWORD/8 test cl, SIZEOF_MMWORD/8
jz short .column_ld2 jz short .column_ld2
sub ecx, byte SIZEOF_MMWORD/8 sub ecx, byte SIZEOF_MMWORD/8
movd mmA, DWORD [esi+ecx*RGB_PIXELSIZE] movd mmA, dword [esi+ecx*RGB_PIXELSIZE]
.column_ld2: .column_ld2:
test cl, SIZEOF_MMWORD/4 test cl, SIZEOF_MMWORD/4
jz short .column_ld4 jz short .column_ld4

6
simd/i386/jccolext-sse2.asm
View File

@@ -12,8 +12,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jcolsamp.inc" %include "jcolsamp.inc"
@@ -109,12 +107,12 @@ EXTN(jsimd_rgb_ycc_convert_sse2):
test cl, SIZEOF_BYTE test cl, SIZEOF_BYTE
jz short .column_ld2 jz short .column_ld2
sub ecx, byte SIZEOF_BYTE sub ecx, byte SIZEOF_BYTE
movzx eax, BYTE [esi+ecx] movzx eax, byte [esi+ecx]
.column_ld2: .column_ld2:
test cl, SIZEOF_WORD test cl, SIZEOF_WORD
jz short .column_ld4 jz short .column_ld4
sub ecx, byte SIZEOF_WORD sub ecx, byte SIZEOF_WORD
movzx edx, WORD [esi+ecx] movzx edx, word [esi+ecx]
shl eax, WORD_BIT shl eax, WORD_BIT
or eax, edx or eax, edx
.column_ld4: .column_ld4:

2
simd/i386/jccolor-avx2.asm
View File

@@ -13,8 +13,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"

2
simd/i386/jccolor-mmx.asm
View File

@@ -13,8 +13,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"

2
simd/i386/jccolor-sse2.asm
View File

@@ -12,8 +12,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"

2
simd/i386/jcgray-avx2.asm
View File

@@ -13,8 +13,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"

2
simd/i386/jcgray-mmx.asm
View File

@@ -13,8 +13,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"

2
simd/i386/jcgray-sse2.asm
View File

@@ -12,8 +12,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"

6
simd/i386/jcgryext-avx2.asm
View File

@@ -13,8 +13,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jcolsamp.inc" %include "jcolsamp.inc"
@@ -102,12 +100,12 @@ EXTN(jsimd_rgb_gray_convert_avx2):
test cl, SIZEOF_BYTE test cl, SIZEOF_BYTE
jz short .column_ld2 jz short .column_ld2
sub ecx, byte SIZEOF_BYTE sub ecx, byte SIZEOF_BYTE
movzx eax, BYTE [esi+ecx] movzx eax, byte [esi+ecx]
.column_ld2: .column_ld2:
test cl, SIZEOF_WORD test cl, SIZEOF_WORD
jz short .column_ld4 jz short .column_ld4
sub ecx, byte SIZEOF_WORD sub ecx, byte SIZEOF_WORD
movzx edx, WORD [esi+ecx] movzx edx, word [esi+ecx]
shl eax, WORD_BIT shl eax, WORD_BIT
or eax, edx or eax, edx
.column_ld4: .column_ld4:

10
simd/i386/jcgryext-mmx.asm
View File

@@ -13,8 +13,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jcolsamp.inc" %include "jcolsamp.inc"
@@ -103,13 +101,13 @@ EXTN(jsimd_rgb_gray_convert_mmx):
jz short .column_ld2 jz short .column_ld2
sub ecx, byte SIZEOF_BYTE sub ecx, byte SIZEOF_BYTE
xor eax, eax xor eax, eax
mov al, BYTE [esi+ecx] mov al, byte [esi+ecx]
.column_ld2: .column_ld2:
test cl, SIZEOF_WORD test cl, SIZEOF_WORD
jz short .column_ld4 jz short .column_ld4
sub ecx, byte SIZEOF_WORD sub ecx, byte SIZEOF_WORD
xor edx, edx xor edx, edx
mov dx, WORD [esi+ecx] mov dx, word [esi+ecx]
shl eax, WORD_BIT shl eax, WORD_BIT
or eax, edx or eax, edx
.column_ld4: .column_ld4:
@@ -119,7 +117,7 @@ EXTN(jsimd_rgb_gray_convert_mmx):
test cl, SIZEOF_DWORD test cl, SIZEOF_DWORD
jz short .column_ld8 jz short .column_ld8
sub ecx, byte SIZEOF_DWORD sub ecx, byte SIZEOF_DWORD
movd mmG, DWORD [esi+ecx] movd mmG, dword [esi+ecx]
psllq mmA, DWORD_BIT psllq mmA, DWORD_BIT
por mmA, mmG por mmA, mmG
.column_ld8: .column_ld8:
@@ -189,7 +187,7 @@ EXTN(jsimd_rgb_gray_convert_mmx):
test cl, SIZEOF_MMWORD/8 test cl, SIZEOF_MMWORD/8
jz short .column_ld2 jz short .column_ld2
sub ecx, byte SIZEOF_MMWORD/8 sub ecx, byte SIZEOF_MMWORD/8
movd mmA, DWORD [esi+ecx*RGB_PIXELSIZE] movd mmA, dword [esi+ecx*RGB_PIXELSIZE]
.column_ld2: .column_ld2:
test cl, SIZEOF_MMWORD/4 test cl, SIZEOF_MMWORD/4
jz short .column_ld4 jz short .column_ld4

6
simd/i386/jcgryext-sse2.asm
View File

@@ -12,8 +12,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jcolsamp.inc" %include "jcolsamp.inc"
@@ -101,12 +99,12 @@ EXTN(jsimd_rgb_gray_convert_sse2):
test cl, SIZEOF_BYTE test cl, SIZEOF_BYTE
jz short .column_ld2 jz short .column_ld2
sub ecx, byte SIZEOF_BYTE sub ecx, byte SIZEOF_BYTE
movzx eax, BYTE [esi+ecx] movzx eax, byte [esi+ecx]
.column_ld2: .column_ld2:
test cl, SIZEOF_WORD test cl, SIZEOF_WORD
jz short .column_ld4 jz short .column_ld4
sub ecx, byte SIZEOF_WORD sub ecx, byte SIZEOF_WORD
movzx edx, WORD [esi+ecx] movzx edx, word [esi+ecx]
shl eax, WORD_BIT shl eax, WORD_BIT
or eax, edx or eax, edx
.column_ld4: .column_ld4:

46
simd/i386/jchuff-sse2.asm
View File

@@ -17,8 +17,6 @@
; This file contains an SSE2 implementation for Huffman coding of one block. ; This file contains an SSE2 implementation for Huffman coding of one block.
; The following code is based directly on jchuff.c; see jchuff.c for more ; The following code is based directly on jchuff.c; see jchuff.c for more
; details. ; details.
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
@@ -197,8 +195,8 @@ EXTN(jsimd_huff_encode_one_block_sse2):
push ebp push ebp
mov esi, POINTER [eax+8] ; (working_state *state) mov esi, POINTER [eax+8] ; (working_state *state)
mov put_buffer, DWORD [esi+8] ; put_buffer = state->cur.put_buffer; mov put_buffer, dword [esi+8] ; put_buffer = state->cur.put_buffer;
mov put_bits, DWORD [esi+12] ; put_bits = state->cur.put_bits; mov put_bits, dword [esi+12] ; put_bits = state->cur.put_bits;
push esi ; esi is now scratch push esi ; esi is now scratch
get_GOT edx ; get GOT address get_GOT edx ; get GOT address
@@ -214,7 +212,7 @@ EXTN(jsimd_huff_encode_one_block_sse2):
; Encode the DC coefficient difference per section F.1.2.1 ; Encode the DC coefficient difference per section F.1.2.1
mov esi, POINTER [esp+block] ; block mov esi, POINTER [esp+block] ; block
movsx ecx, word [esi] ; temp = temp2 = block[0] - last_dc_val; movsx ecx, word [esi] ; temp = temp2 = block[0] - last_dc_val;
sub ecx, DWORD [eax+20] sub ecx, dword [eax+20]
mov esi, ecx mov esi, ecx
; This is a well-known technique for obtaining the absolute value ; This is a well-known technique for obtaining the absolute value
@@ -229,12 +227,12 @@ EXTN(jsimd_huff_encode_one_block_sse2):
; For a negative input, want temp2 = bitwise complement of abs(input) ; For a negative input, want temp2 = bitwise complement of abs(input)
; This code assumes we are on a two's complement machine ; This code assumes we are on a two's complement machine
add esi, edx ; temp2 += temp3; add esi, edx ; temp2 += temp3;
mov DWORD [esp+temp], esi ; backup temp2 in temp mov dword [esp+temp], esi ; backup temp2 in temp
; Find the number of bits needed for the magnitude of the coefficient ; Find the number of bits needed for the magnitude of the coefficient
movpic ebp, POINTER [esp+gotptr] ; load GOT address (ebp) movpic ebp, POINTER [esp+gotptr] ; load GOT address (ebp)
movzx edx, byte [GOTOFF(ebp, jpeg_nbits_table + ecx)] ; nbits = JPEG_NBITS(temp); movzx edx, byte [GOTOFF(ebp, jpeg_nbits_table + ecx)] ; nbits = JPEG_NBITS(temp);
mov DWORD [esp+temp2], edx ; backup nbits in temp2 mov dword [esp+temp2], edx ; backup nbits in temp2
; Emit the Huffman-coded symbol for the number of bits ; Emit the Huffman-coded symbol for the number of bits
mov ebp, POINTER [eax+24] ; After this point, arguments are not accessible anymore mov ebp, POINTER [eax+24] ; After this point, arguments are not accessible anymore
@@ -242,13 +240,13 @@ EXTN(jsimd_huff_encode_one_block_sse2):
movzx ecx, byte [ebp + edx + 1024] ; size = dctbl->ehufsi[nbits]; movzx ecx, byte [ebp + edx + 1024] ; size = dctbl->ehufsi[nbits];
EMIT_BITS eax ; EMIT_BITS(code, size) EMIT_BITS eax ; EMIT_BITS(code, size)
mov ecx, DWORD [esp+temp2] ; restore nbits mov ecx, dword [esp+temp2] ; restore nbits
; Mask off any extra bits in code ; Mask off any extra bits in code
mov eax, 1 mov eax, 1
shl eax, cl shl eax, cl
dec eax dec eax
and eax, DWORD [esp+temp] ; temp2 &= (((JLONG)1)<<nbits) - 1; and eax, dword [esp+temp] ; temp2 &= (((JLONG)1)<<nbits) - 1;
; Emit that number of bits of the value, if positive, ; Emit that number of bits of the value, if positive,
; or the complement of its magnitude, if negative. ; or the complement of its magnitude, if negative.
@@ -291,22 +289,22 @@ EXTN(jsimd_huff_encode_one_block_sse2):
jz near .ELOOP jz near .ELOOP
lea esi, [esi+ecx*2] ; k += r; lea esi, [esi+ecx*2] ; k += r;
shr edx, cl ; index >>= r; shr edx, cl ; index >>= r;
mov DWORD [esp+temp3], edx mov dword [esp+temp3], edx
.BRLOOP: .BRLOOP:
cmp ecx, 16 ; while (r > 15) { cmp ecx, 16 ; while (r > 15) {
jl near .ERLOOP jl near .ERLOOP
sub ecx, 16 ; r -= 16; sub ecx, 16 ; r -= 16;
mov DWORD [esp+temp], ecx mov dword [esp+temp], ecx
mov eax, INT [ebp + 240 * 4] ; code_0xf0 = actbl->ehufco[0xf0]; mov eax, INT [ebp + 240 * 4] ; code_0xf0 = actbl->ehufco[0xf0];
movzx ecx, byte [ebp + 1024 + 240] ; size_0xf0 = actbl->ehufsi[0xf0]; movzx ecx, byte [ebp + 1024 + 240] ; size_0xf0 = actbl->ehufsi[0xf0];
EMIT_BITS eax ; EMIT_BITS(code_0xf0, size_0xf0) EMIT_BITS eax ; EMIT_BITS(code_0xf0, size_0xf0)
mov ecx, DWORD [esp+temp] mov ecx, dword [esp+temp]
jmp .BRLOOP jmp .BRLOOP
.ERLOOP: .ERLOOP:
movsx eax, word [esi] ; temp = t1[k]; movsx eax, word [esi] ; temp = t1[k];
movpic edx, POINTER [esp+gotptr] ; load GOT address (edx) movpic edx, POINTER [esp+gotptr] ; load GOT address (edx)
movzx eax, byte [GOTOFF(edx, jpeg_nbits_table + eax)] ; nbits = JPEG_NBITS(temp); movzx eax, byte [GOTOFF(edx, jpeg_nbits_table + eax)] ; nbits = JPEG_NBITS(temp);
mov DWORD [esp+temp2], eax mov dword [esp+temp2], eax
; Emit Huffman symbol for run length / number of bits ; Emit Huffman symbol for run length / number of bits
shl ecx, 4 ; temp3 = (r << 4) + nbits; shl ecx, 4 ; temp3 = (r << 4) + nbits;
add ecx, eax add ecx, eax
@@ -316,13 +314,13 @@ EXTN(jsimd_huff_encode_one_block_sse2):
movsx edx, word [esi+DCTSIZE2*2] ; temp2 = t2[k]; movsx edx, word [esi+DCTSIZE2*2] ; temp2 = t2[k];
; Mask off any extra bits in code ; Mask off any extra bits in code
mov ecx, DWORD [esp+temp2] mov ecx, dword [esp+temp2]
mov eax, 1 mov eax, 1
shl eax, cl shl eax, cl
dec eax dec eax
and eax, edx ; temp2 &= (((JLONG)1)<<nbits) - 1; and eax, edx ; temp2 &= (((JLONG)1)<<nbits) - 1;
EMIT_BITS eax ; PUT_BITS(temp2, nbits) EMIT_BITS eax ; PUT_BITS(temp2, nbits)
mov edx, DWORD [esp+temp3] mov edx, dword [esp+temp3]
add esi, 2 ; ++k; add esi, 2 ; ++k;
shr edx, 1 ; index >>= 1; shr edx, 1 ; index >>= 1;
@@ -352,29 +350,29 @@ EXTN(jsimd_huff_encode_one_block_sse2):
shr edx, cl ; index >>= r; shr edx, cl ; index >>= r;
add ecx, eax add ecx, eax
lea esi, [esi+ecx*2] ; k += r; lea esi, [esi+ecx*2] ; k += r;
mov DWORD [esp+temp3], edx mov dword [esp+temp3], edx
jmp .BRLOOP2 jmp .BRLOOP2
.BLOOP2: .BLOOP2:
bsf ecx, edx ; r = __builtin_ctzl(index); bsf ecx, edx ; r = __builtin_ctzl(index);
jz near .ELOOP2 jz near .ELOOP2
lea esi, [esi+ecx*2] ; k += r; lea esi, [esi+ecx*2] ; k += r;
shr edx, cl ; index >>= r; shr edx, cl ; index >>= r;
mov DWORD [esp+temp3], edx mov dword [esp+temp3], edx
.BRLOOP2: .BRLOOP2:
cmp ecx, 16 ; while (r > 15) { cmp ecx, 16 ; while (r > 15) {
jl near .ERLOOP2 jl near .ERLOOP2
sub ecx, 16 ; r -= 16; sub ecx, 16 ; r -= 16;
mov DWORD [esp+temp], ecx mov dword [esp+temp], ecx
mov eax, INT [ebp + 240 * 4] ; code_0xf0 = actbl->ehufco[0xf0]; mov eax, INT [ebp + 240 * 4] ; code_0xf0 = actbl->ehufco[0xf0];
movzx ecx, byte [ebp + 1024 + 240] ; size_0xf0 = actbl->ehufsi[0xf0]; movzx ecx, byte [ebp + 1024 + 240] ; size_0xf0 = actbl->ehufsi[0xf0];
EMIT_BITS eax ; EMIT_BITS(code_0xf0, size_0xf0) EMIT_BITS eax ; EMIT_BITS(code_0xf0, size_0xf0)
mov ecx, DWORD [esp+temp] mov ecx, dword [esp+temp]
jmp .BRLOOP2 jmp .BRLOOP2
.ERLOOP2: .ERLOOP2:
movsx eax, word [esi] ; temp = t1[k]; movsx eax, word [esi] ; temp = t1[k];
bsr eax, eax ; nbits = 32 - __builtin_clz(temp); bsr eax, eax ; nbits = 32 - __builtin_clz(temp);
inc eax inc eax
mov DWORD [esp+temp2], eax mov dword [esp+temp2], eax
; Emit Huffman symbol for run length / number of bits ; Emit Huffman symbol for run length / number of bits
shl ecx, 4 ; temp3 = (r << 4) + nbits; shl ecx, 4 ; temp3 = (r << 4) + nbits;
add ecx, eax add ecx, eax
@@ -384,13 +382,13 @@ EXTN(jsimd_huff_encode_one_block_sse2):
movsx edx, word [esi+DCTSIZE2*2] ; temp2 = t2[k]; movsx edx, word [esi+DCTSIZE2*2] ; temp2 = t2[k];
; Mask off any extra bits in code ; Mask off any extra bits in code
mov ecx, DWORD [esp+temp2] mov ecx, dword [esp+temp2]
mov eax, 1 mov eax, 1
shl eax, cl shl eax, cl
dec eax dec eax
and eax, edx ; temp2 &= (((JLONG)1)<<nbits) - 1; and eax, edx ; temp2 &= (((JLONG)1)<<nbits) - 1;
EMIT_BITS eax ; PUT_BITS(temp2, nbits) EMIT_BITS eax ; PUT_BITS(temp2, nbits)
mov edx, DWORD [esp+temp3] mov edx, dword [esp+temp3]
add esi, 2 ; ++k; add esi, 2 ; ++k;
shr edx, 1 ; index >>= 1; shr edx, 1 ; index >>= 1;
@@ -407,8 +405,8 @@ EXTN(jsimd_huff_encode_one_block_sse2):
mov eax, [esp+buffer] mov eax, [esp+buffer]
pop esi pop esi
; Save put_buffer & put_bits ; Save put_buffer & put_bits
mov DWORD [esi+8], put_buffer ; state->cur.put_buffer = put_buffer; mov dword [esi+8], put_buffer ; state->cur.put_buffer = put_buffer;
mov DWORD [esi+12], put_bits ; state->cur.put_bits = put_bits; mov dword [esi+12], put_bits ; state->cur.put_bits = put_bits;
pop ebp pop ebp
pop edi pop edi

2
simd/i386/jcphuff-sse2.asm
View File

@@ -15,8 +15,6 @@
; ;
; This file contains an SSE2 implementation of data preparation for progressive ; This file contains an SSE2 implementation of data preparation for progressive
; Huffman encoding. See jcphuff.c for more details. ; Huffman encoding. See jcphuff.c for more details.
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"

2
simd/i386/jcsample-avx2.asm
View File

@@ -14,8 +14,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"

2
simd/i386/jcsample-mmx.asm
View File

@@ -13,8 +13,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"

2
simd/i386/jcsample-sse2.asm
View File

@@ -13,8 +13,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"

6
simd/i386/jdcolext-avx2.asm
View File

@@ -14,8 +14,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jcolsamp.inc" %include "jcolsamp.inc"
@@ -348,7 +346,7 @@ EXTN(jsimd_ycc_rgb_convert_avx2):
vmovd eax, xmmA vmovd eax, xmmA
cmp ecx, byte SIZEOF_WORD cmp ecx, byte SIZEOF_WORD
jb short .column_st1 jb short .column_st1
mov WORD [edi], ax mov word [edi], ax
add edi, byte SIZEOF_WORD add edi, byte SIZEOF_WORD
sub ecx, byte SIZEOF_WORD sub ecx, byte SIZEOF_WORD
shr eax, 16 shr eax, 16
@@ -357,7 +355,7 @@ EXTN(jsimd_ycc_rgb_convert_avx2):
; space. ; space.
test ecx, ecx test ecx, ecx
jz short .nextrow jz short .nextrow
mov BYTE [edi], al mov byte [edi], al
%else ; RGB_PIXELSIZE == 4 ; ----------- %else ; RGB_PIXELSIZE == 4 ; -----------

10
simd/i386/jdcolext-mmx.asm
View File

@@ -13,8 +13,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jcolsamp.inc" %include "jcolsamp.inc"
@@ -280,7 +278,7 @@ EXTN(jsimd_ycc_rgb_convert_mmx):
movd eax, mmA movd eax, mmA
cmp ecx, byte SIZEOF_DWORD cmp ecx, byte SIZEOF_DWORD
jb short .column_st2 jb short .column_st2
mov DWORD [edi+0*SIZEOF_DWORD], eax mov dword [edi+0*SIZEOF_DWORD], eax
psrlq mmA, DWORD_BIT psrlq mmA, DWORD_BIT
movd eax, mmA movd eax, mmA
sub ecx, byte SIZEOF_DWORD sub ecx, byte SIZEOF_DWORD
@@ -288,14 +286,14 @@ EXTN(jsimd_ycc_rgb_convert_mmx):
.column_st2: .column_st2:
cmp ecx, byte SIZEOF_WORD cmp ecx, byte SIZEOF_WORD
jb short .column_st1 jb short .column_st1
mov WORD [edi+0*SIZEOF_WORD], ax mov word [edi+0*SIZEOF_WORD], ax
shr eax, WORD_BIT shr eax, WORD_BIT
sub ecx, byte SIZEOF_WORD sub ecx, byte SIZEOF_WORD
add edi, byte SIZEOF_WORD add edi, byte SIZEOF_WORD
.column_st1: .column_st1:
cmp ecx, byte SIZEOF_BYTE cmp ecx, byte SIZEOF_BYTE
jb short .nextrow jb short .nextrow
mov BYTE [edi+0*SIZEOF_BYTE], al mov byte [edi+0*SIZEOF_BYTE], al
%else ; RGB_PIXELSIZE == 4 ; ----------- %else ; RGB_PIXELSIZE == 4 ; -----------
@@ -367,7 +365,7 @@ EXTN(jsimd_ycc_rgb_convert_mmx):
.column_st4: .column_st4:
cmp ecx, byte SIZEOF_MMWORD/8 cmp ecx, byte SIZEOF_MMWORD/8
jb short .nextrow jb short .nextrow
movd DWORD [edi+0*SIZEOF_DWORD], mmA movd dword [edi+0*SIZEOF_DWORD], mmA
%endif ; RGB_PIXELSIZE ; --------------- %endif ; RGB_PIXELSIZE ; ---------------

6
simd/i386/jdcolext-sse2.asm
View File

@@ -13,8 +13,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jcolsamp.inc" %include "jcolsamp.inc"
@@ -320,7 +318,7 @@ EXTN(jsimd_ycc_rgb_convert_sse2):
movd eax, xmmA movd eax, xmmA
cmp ecx, byte SIZEOF_WORD cmp ecx, byte SIZEOF_WORD
jb short .column_st1 jb short .column_st1
mov WORD [edi], ax mov word [edi], ax
add edi, byte SIZEOF_WORD add edi, byte SIZEOF_WORD
sub ecx, byte SIZEOF_WORD sub ecx, byte SIZEOF_WORD
shr eax, 16 shr eax, 16
@@ -329,7 +327,7 @@ EXTN(jsimd_ycc_rgb_convert_sse2):
; space. ; space.
test ecx, ecx test ecx, ecx
jz short .nextrow jz short .nextrow
mov BYTE [edi], al mov byte [edi], al
%else ; RGB_PIXELSIZE == 4 ; ----------- %else ; RGB_PIXELSIZE == 4 ; -----------

2
simd/i386/jdcolor-avx2.asm
View File

@@ -14,8 +14,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"

2
simd/i386/jdcolor-mmx.asm
View File

@@ -13,8 +13,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"

2
simd/i386/jdcolor-sse2.asm
View File

@@ -13,8 +13,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"

2
simd/i386/jdmerge-avx2.asm
View File

@@ -14,8 +14,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"

2
simd/i386/jdmerge-mmx.asm
View File

@@ -13,8 +13,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"

2
simd/i386/jdmerge-sse2.asm
View File

@@ -13,8 +13,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"

6
simd/i386/jdmrgext-avx2.asm
View File

@@ -14,8 +14,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jcolsamp.inc" %include "jcolsamp.inc"
@@ -354,7 +352,7 @@ EXTN(jsimd_h2v1_merged_upsample_avx2):
vmovd eax, xmmA vmovd eax, xmmA
cmp ecx, byte SIZEOF_WORD cmp ecx, byte SIZEOF_WORD
jb short .column_st1 jb short .column_st1
mov WORD [edi], ax mov word [edi], ax
add edi, byte SIZEOF_WORD add edi, byte SIZEOF_WORD
sub ecx, byte SIZEOF_WORD sub ecx, byte SIZEOF_WORD
shr eax, 16 shr eax, 16
@@ -363,7 +361,7 @@ EXTN(jsimd_h2v1_merged_upsample_avx2):
; space. ; space.
test ecx, ecx test ecx, ecx
jz short .endcolumn jz short .endcolumn
mov BYTE [edi], al mov byte [edi], al
%else ; RGB_PIXELSIZE == 4 ; ----------- %else ; RGB_PIXELSIZE == 4 ; -----------

10
simd/i386/jdmrgext-mmx.asm
View File

@@ -13,8 +13,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jcolsamp.inc" %include "jcolsamp.inc"
@@ -283,7 +281,7 @@ EXTN(jsimd_h2v1_merged_upsample_mmx):
movd eax, mmA movd eax, mmA
cmp ecx, byte SIZEOF_DWORD cmp ecx, byte SIZEOF_DWORD
jb short .column_st2 jb short .column_st2
mov DWORD [edi+0*SIZEOF_DWORD], eax mov dword [edi+0*SIZEOF_DWORD], eax
psrlq mmA, DWORD_BIT psrlq mmA, DWORD_BIT
movd eax, mmA movd eax, mmA
sub ecx, byte SIZEOF_DWORD sub ecx, byte SIZEOF_DWORD
@@ -291,14 +289,14 @@ EXTN(jsimd_h2v1_merged_upsample_mmx):
.column_st2: .column_st2:
cmp ecx, byte SIZEOF_WORD cmp ecx, byte SIZEOF_WORD
jb short .column_st1 jb short .column_st1
mov WORD [edi+0*SIZEOF_WORD], ax mov word [edi+0*SIZEOF_WORD], ax
shr eax, WORD_BIT shr eax, WORD_BIT
sub ecx, byte SIZEOF_WORD sub ecx, byte SIZEOF_WORD
add edi, byte SIZEOF_WORD add edi, byte SIZEOF_WORD
.column_st1: .column_st1:
cmp ecx, byte SIZEOF_BYTE cmp ecx, byte SIZEOF_BYTE
jb short .endcolumn jb short .endcolumn
mov BYTE [edi+0*SIZEOF_BYTE], al mov byte [edi+0*SIZEOF_BYTE], al
%else ; RGB_PIXELSIZE == 4 ; ----------- %else ; RGB_PIXELSIZE == 4 ; -----------
@@ -373,7 +371,7 @@ EXTN(jsimd_h2v1_merged_upsample_mmx):
.column_st4: .column_st4:
cmp ecx, byte SIZEOF_MMWORD/8 cmp ecx, byte SIZEOF_MMWORD/8
jb short .endcolumn jb short .endcolumn
movd DWORD [edi+0*SIZEOF_DWORD], mmA movd dword [edi+0*SIZEOF_DWORD], mmA
%endif ; RGB_PIXELSIZE ; --------------- %endif ; RGB_PIXELSIZE ; ---------------

6
simd/i386/jdmrgext-sse2.asm
View File

@@ -13,8 +13,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jcolsamp.inc" %include "jcolsamp.inc"
@@ -325,7 +323,7 @@ EXTN(jsimd_h2v1_merged_upsample_sse2):
movd eax, xmmA movd eax, xmmA
cmp ecx, byte SIZEOF_WORD cmp ecx, byte SIZEOF_WORD
jb short .column_st1 jb short .column_st1
mov WORD [edi], ax mov word [edi], ax
add edi, byte SIZEOF_WORD add edi, byte SIZEOF_WORD
sub ecx, byte SIZEOF_WORD sub ecx, byte SIZEOF_WORD
shr eax, 16 shr eax, 16
@@ -334,7 +332,7 @@ EXTN(jsimd_h2v1_merged_upsample_sse2):
; space. ; space.
test ecx, ecx test ecx, ecx
jz short .endcolumn jz short .endcolumn
mov BYTE [edi], al mov byte [edi], al
%else ; RGB_PIXELSIZE == 4 ; ----------- %else ; RGB_PIXELSIZE == 4 ; -----------

2
simd/i386/jdsample-avx2.asm
View File

@@ -14,8 +14,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"

2
simd/i386/jdsample-mmx.asm
View File

@@ -13,8 +13,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"

2
simd/i386/jdsample-sse2.asm
View File

@@ -13,8 +13,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"

2
simd/i386/jfdctflt-3dn.asm
View File

@@ -17,8 +17,6 @@
; This file contains a floating-point implementation of the forward DCT ; This file contains a floating-point implementation of the forward DCT
; (Discrete Cosine Transform). The following code is based directly on ; (Discrete Cosine Transform). The following code is based directly on
; the IJG's original jfdctflt.c; see the jfdctflt.c for more details. ; the IJG's original jfdctflt.c; see the jfdctflt.c for more details.
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"

2
simd/i386/jfdctflt-sse.asm
View File

@@ -17,8 +17,6 @@
; This file contains a floating-point implementation of the forward DCT ; This file contains a floating-point implementation of the forward DCT
; (Discrete Cosine Transform). The following code is based directly on ; (Discrete Cosine Transform). The following code is based directly on
; the IJG's original jfdctflt.c; see the jfdctflt.c for more details. ; the IJG's original jfdctflt.c; see the jfdctflt.c for more details.
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"

2
simd/i386/jfdctfst-mmx.asm
View File

@@ -18,8 +18,6 @@
; the forward DCT (Discrete Cosine Transform). The following code is ; the forward DCT (Discrete Cosine Transform). The following code is
; based directly on the IJG's original jfdctfst.c; see the jfdctfst.c ; based directly on the IJG's original jfdctfst.c; see the jfdctfst.c
; for more details. ; for more details.
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"

2
simd/i386/jfdctfst-sse2.asm
View File

@@ -18,8 +18,6 @@
; the forward DCT (Discrete Cosine Transform). The following code is ; the forward DCT (Discrete Cosine Transform). The following code is
; based directly on the IJG's original jfdctfst.c; see the jfdctfst.c ; based directly on the IJG's original jfdctfst.c; see the jfdctfst.c
; for more details. ; for more details.
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"

2
simd/i386/jfdctint-avx2.asm
View File

@@ -18,8 +18,6 @@
; forward DCT (Discrete Cosine Transform). The following code is based ; forward DCT (Discrete Cosine Transform). The following code is based
; directly on the IJG's original jfdctint.c; see the jfdctint.c for ; directly on the IJG's original jfdctint.c; see the jfdctint.c for
; more details. ; more details.
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"

2
simd/i386/jfdctint-mmx.asm
View File

@@ -18,8 +18,6 @@
; forward DCT (Discrete Cosine Transform). The following code is based ; forward DCT (Discrete Cosine Transform). The following code is based
; directly on the IJG's original jfdctint.c; see the jfdctint.c for ; directly on the IJG's original jfdctint.c; see the jfdctint.c for
; more details. ; more details.
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"

2
simd/i386/jfdctint-sse2.asm
View File

@@ -18,8 +18,6 @@
; forward DCT (Discrete Cosine Transform). The following code is based ; forward DCT (Discrete Cosine Transform). The following code is based
; directly on the IJG's original jfdctint.c; see the jfdctint.c for ; directly on the IJG's original jfdctint.c; see the jfdctint.c for
; more details. ; more details.
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"

34
simd/i386/jidctflt-3dn.asm
View File

@@ -17,8 +17,6 @@
; This file contains a floating-point implementation of the inverse DCT ; This file contains a floating-point implementation of the inverse DCT
; (Discrete Cosine Transform). The following code is based directly on ; (Discrete Cosine Transform). The following code is based directly on
; the IJG's original jidctflt.c; see the jidctflt.c for more details. ; the IJG's original jidctflt.c; see the jidctflt.c for more details.
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"
@@ -92,23 +90,23 @@ EXTN(jsimd_idct_float_3dnow):
alignx 16, 7 alignx 16, 7
.columnloop: .columnloop:
%ifndef NO_ZERO_COLUMN_TEST_FLOAT_3DNOW %ifndef NO_ZERO_COLUMN_TEST_FLOAT_3DNOW
mov eax, DWORD [DWBLOCK(1,0,esi,SIZEOF_JCOEF)] mov eax, dword [DWBLOCK(1,0,esi,SIZEOF_JCOEF)]
or eax, DWORD [DWBLOCK(2,0,esi,SIZEOF_JCOEF)] or eax, dword [DWBLOCK(2,0,esi,SIZEOF_JCOEF)]
jnz short .columnDCT jnz short .columnDCT
pushpic ebx ; save GOT address pushpic ebx ; save GOT address
mov ebx, DWORD [DWBLOCK(3,0,esi,SIZEOF_JCOEF)] mov ebx, dword [DWBLOCK(3,0,esi,SIZEOF_JCOEF)]
mov eax, DWORD [DWBLOCK(4,0,esi,SIZEOF_JCOEF)] mov eax, dword [DWBLOCK(4,0,esi,SIZEOF_JCOEF)]
or ebx, DWORD [DWBLOCK(5,0,esi,SIZEOF_JCOEF)] or ebx, dword [DWBLOCK(5,0,esi,SIZEOF_JCOEF)]
or eax, DWORD [DWBLOCK(6,0,esi,SIZEOF_JCOEF)] or eax, dword [DWBLOCK(6,0,esi,SIZEOF_JCOEF)]
or ebx, DWORD [DWBLOCK(7,0,esi,SIZEOF_JCOEF)] or ebx, dword [DWBLOCK(7,0,esi,SIZEOF_JCOEF)]
or eax, ebx or eax, ebx
poppic ebx ; restore GOT address poppic ebx ; restore GOT address
jnz short .columnDCT jnz short .columnDCT
; -- AC terms all zero ; -- AC terms all zero
movd mm0, DWORD [DWBLOCK(0,0,esi,SIZEOF_JCOEF)] movd mm0, dword [DWBLOCK(0,0,esi,SIZEOF_JCOEF)]
punpcklwd mm0, mm0 punpcklwd mm0, mm0
psrad mm0, (DWORD_BIT-WORD_BIT) psrad mm0, (DWORD_BIT-WORD_BIT)
@@ -135,10 +133,10 @@ EXTN(jsimd_idct_float_3dnow):
; -- Even part ; -- Even part
movd mm0, DWORD [DWBLOCK(0,0,esi,SIZEOF_JCOEF)] movd mm0, dword [DWBLOCK(0,0,esi,SIZEOF_JCOEF)]
movd mm1, DWORD [DWBLOCK(2,0,esi,SIZEOF_JCOEF)] movd mm1, dword [DWBLOCK(2,0,esi,SIZEOF_JCOEF)]
movd mm2, DWORD [DWBLOCK(4,0,esi,SIZEOF_JCOEF)] movd mm2, dword [DWBLOCK(4,0,esi,SIZEOF_JCOEF)]
movd mm3, DWORD [DWBLOCK(6,0,esi,SIZEOF_JCOEF)] movd mm3, dword [DWBLOCK(6,0,esi,SIZEOF_JCOEF)]
punpcklwd mm0, mm0 punpcklwd mm0, mm0
punpcklwd mm1, mm1 punpcklwd mm1, mm1
@@ -182,10 +180,10 @@ EXTN(jsimd_idct_float_3dnow):
; -- Odd part ; -- Odd part
movd mm2, DWORD [DWBLOCK(1,0,esi,SIZEOF_JCOEF)] movd mm2, dword [DWBLOCK(1,0,esi,SIZEOF_JCOEF)]
movd mm3, DWORD [DWBLOCK(3,0,esi,SIZEOF_JCOEF)] movd mm3, dword [DWBLOCK(3,0,esi,SIZEOF_JCOEF)]
movd mm5, DWORD [DWBLOCK(5,0,esi,SIZEOF_JCOEF)] movd mm5, dword [DWBLOCK(5,0,esi,SIZEOF_JCOEF)]
movd mm1, DWORD [DWBLOCK(7,0,esi,SIZEOF_JCOEF)] movd mm1, dword [DWBLOCK(7,0,esi,SIZEOF_JCOEF)]
punpcklwd mm2, mm2 punpcklwd mm2, mm2
punpcklwd mm3, mm3 punpcklwd mm3, mm3

6
simd/i386/jidctflt-sse.asm
View File

@@ -17,8 +17,6 @@
; This file contains a floating-point implementation of the inverse DCT ; This file contains a floating-point implementation of the inverse DCT
; (Discrete Cosine Transform). The following code is based directly on ; (Discrete Cosine Transform). The following code is based directly on
; the IJG's original jidctflt.c; see the jidctflt.c for more details. ; the IJG's original jidctflt.c; see the jidctflt.c for more details.
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"
@@ -102,8 +100,8 @@ EXTN(jsimd_idct_float_sse):
alignx 16, 7 alignx 16, 7
.columnloop: .columnloop:
%ifndef NO_ZERO_COLUMN_TEST_FLOAT_SSE %ifndef NO_ZERO_COLUMN_TEST_FLOAT_SSE
mov eax, DWORD [DWBLOCK(1,0,esi,SIZEOF_JCOEF)] mov eax, dword [DWBLOCK(1,0,esi,SIZEOF_JCOEF)]
or eax, DWORD [DWBLOCK(2,0,esi,SIZEOF_JCOEF)] or eax, dword [DWBLOCK(2,0,esi,SIZEOF_JCOEF)]
jnz near .columnDCT jnz near .columnDCT
movq mm0, MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)] movq mm0, MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)]

6
simd/i386/jidctflt-sse2.asm
View File

@@ -17,8 +17,6 @@
; This file contains a floating-point implementation of the inverse DCT ; This file contains a floating-point implementation of the inverse DCT
; (Discrete Cosine Transform). The following code is based directly on ; (Discrete Cosine Transform). The following code is based directly on
; the IJG's original jidctflt.c; see the jidctflt.c for more details. ; the IJG's original jidctflt.c; see the jidctflt.c for more details.
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"
@@ -102,8 +100,8 @@ EXTN(jsimd_idct_float_sse2):
alignx 16, 7 alignx 16, 7
.columnloop: .columnloop:
%ifndef NO_ZERO_COLUMN_TEST_FLOAT_SSE %ifndef NO_ZERO_COLUMN_TEST_FLOAT_SSE
mov eax, DWORD [DWBLOCK(1,0,esi,SIZEOF_JCOEF)] mov eax, dword [DWBLOCK(1,0,esi,SIZEOF_JCOEF)]
or eax, DWORD [DWBLOCK(2,0,esi,SIZEOF_JCOEF)] or eax, dword [DWBLOCK(2,0,esi,SIZEOF_JCOEF)]
jnz near .columnDCT jnz near .columnDCT
movq xmm1, XMM_MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)] movq xmm1, XMM_MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)]

6
simd/i386/jidctfst-mmx.asm
View File

@@ -18,8 +18,6 @@
; the inverse DCT (Discrete Cosine Transform). The following code is ; the inverse DCT (Discrete Cosine Transform). The following code is
; based directly on the IJG's original jidctfst.c; see the jidctfst.c ; based directly on the IJG's original jidctfst.c; see the jidctfst.c
; for more details. ; for more details.
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"
@@ -123,8 +121,8 @@ EXTN(jsimd_idct_ifast_mmx):
alignx 16, 7 alignx 16, 7
.columnloop: .columnloop:
%ifndef NO_ZERO_COLUMN_TEST_IFAST_MMX %ifndef NO_ZERO_COLUMN_TEST_IFAST_MMX
mov eax, DWORD [DWBLOCK(1,0,esi,SIZEOF_JCOEF)] mov eax, dword [DWBLOCK(1,0,esi,SIZEOF_JCOEF)]
or eax, DWORD [DWBLOCK(2,0,esi,SIZEOF_JCOEF)] or eax, dword [DWBLOCK(2,0,esi,SIZEOF_JCOEF)]
jnz short .columnDCT jnz short .columnDCT
movq mm0, MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)] movq mm0, MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)]

6
simd/i386/jidctfst-sse2.asm
View File

@@ -18,8 +18,6 @@
; the inverse DCT (Discrete Cosine Transform). The following code is ; the inverse DCT (Discrete Cosine Transform). The following code is
; based directly on the IJG's original jidctfst.c; see the jidctfst.c ; based directly on the IJG's original jidctfst.c; see the jidctfst.c
; for more details. ; for more details.
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"
@@ -118,8 +116,8 @@ EXTN(jsimd_idct_ifast_sse2):
mov esi, JCOEFPTR [coef_block(eax)] ; inptr mov esi, JCOEFPTR [coef_block(eax)] ; inptr
%ifndef NO_ZERO_COLUMN_TEST_IFAST_SSE2 %ifndef NO_ZERO_COLUMN_TEST_IFAST_SSE2
mov eax, DWORD [DWBLOCK(1,0,esi,SIZEOF_JCOEF)] mov eax, dword [DWBLOCK(1,0,esi,SIZEOF_JCOEF)]
or eax, DWORD [DWBLOCK(2,0,esi,SIZEOF_JCOEF)] or eax, dword [DWBLOCK(2,0,esi,SIZEOF_JCOEF)]
jnz near .columnDCT jnz near .columnDCT
movdqa xmm0, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)] movdqa xmm0, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)]

6
simd/i386/jidctint-avx2.asm
View File

@@ -18,8 +18,6 @@
; inverse DCT (Discrete Cosine Transform). The following code is based ; inverse DCT (Discrete Cosine Transform). The following code is based
; directly on the IJG's original jidctint.c; see the jidctint.c for ; directly on the IJG's original jidctint.c; see the jidctint.c for
; more details. ; more details.
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"
@@ -320,8 +318,8 @@ EXTN(jsimd_idct_islow_avx2):
mov esi, JCOEFPTR [coef_block(eax)] ; inptr mov esi, JCOEFPTR [coef_block(eax)] ; inptr
%ifndef NO_ZERO_COLUMN_TEST_ISLOW_AVX2 %ifndef NO_ZERO_COLUMN_TEST_ISLOW_AVX2
mov eax, DWORD [DWBLOCK(1,0,esi,SIZEOF_JCOEF)] mov eax, dword [DWBLOCK(1,0,esi,SIZEOF_JCOEF)]
or eax, DWORD [DWBLOCK(2,0,esi,SIZEOF_JCOEF)] or eax, dword [DWBLOCK(2,0,esi,SIZEOF_JCOEF)]
jnz near .columnDCT jnz near .columnDCT
movdqa xmm0, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)] movdqa xmm0, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)]

6
simd/i386/jidctint-mmx.asm
View File

@@ -18,8 +18,6 @@
; inverse DCT (Discrete Cosine Transform). The following code is based ; inverse DCT (Discrete Cosine Transform). The following code is based
; directly on the IJG's original jidctint.c; see the jidctint.c for ; directly on the IJG's original jidctint.c; see the jidctint.c for
; more details. ; more details.
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"
@@ -136,8 +134,8 @@ EXTN(jsimd_idct_islow_mmx):
alignx 16, 7 alignx 16, 7
.columnloop: .columnloop:
%ifndef NO_ZERO_COLUMN_TEST_ISLOW_MMX %ifndef NO_ZERO_COLUMN_TEST_ISLOW_MMX
mov eax, DWORD [DWBLOCK(1,0,esi,SIZEOF_JCOEF)] mov eax, dword [DWBLOCK(1,0,esi,SIZEOF_JCOEF)]
or eax, DWORD [DWBLOCK(2,0,esi,SIZEOF_JCOEF)] or eax, dword [DWBLOCK(2,0,esi,SIZEOF_JCOEF)]
jnz short .columnDCT jnz short .columnDCT
movq mm0, MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)] movq mm0, MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)]

6
simd/i386/jidctint-sse2.asm
View File

@@ -18,8 +18,6 @@
; inverse DCT (Discrete Cosine Transform). The following code is based ; inverse DCT (Discrete Cosine Transform). The following code is based
; directly on the IJG's original jidctint.c; see the jidctint.c for ; directly on the IJG's original jidctint.c; see the jidctint.c for
; more details. ; more details.
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"
@@ -131,8 +129,8 @@ EXTN(jsimd_idct_islow_sse2):
mov esi, JCOEFPTR [coef_block(eax)] ; inptr mov esi, JCOEFPTR [coef_block(eax)] ; inptr
%ifndef NO_ZERO_COLUMN_TEST_ISLOW_SSE2 %ifndef NO_ZERO_COLUMN_TEST_ISLOW_SSE2
mov eax, DWORD [DWBLOCK(1,0,esi,SIZEOF_JCOEF)] mov eax, dword [DWBLOCK(1,0,esi,SIZEOF_JCOEF)]
or eax, DWORD [DWBLOCK(2,0,esi,SIZEOF_JCOEF)] or eax, dword [DWBLOCK(2,0,esi,SIZEOF_JCOEF)]
jnz near .columnDCT jnz near .columnDCT
movdqa xmm0, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)] movdqa xmm0, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)]

18
simd/i386/jidctred-mmx.asm
View File

@@ -18,8 +18,6 @@
; output: either 4x4 or 2x2 pixels from an 8x8 DCT block. ; output: either 4x4 or 2x2 pixels from an 8x8 DCT block.
; The following code is based directly on the IJG's original jidctred.c; ; The following code is based directly on the IJG's original jidctred.c;
; see the jidctred.c for more details. ; see the jidctred.c for more details.
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"
@@ -144,8 +142,8 @@ EXTN(jsimd_idct_4x4_mmx):
alignx 16, 7 alignx 16, 7
.columnloop: .columnloop:
%ifndef NO_ZERO_COLUMN_TEST_4X4_MMX %ifndef NO_ZERO_COLUMN_TEST_4X4_MMX
mov eax, DWORD [DWBLOCK(1,0,esi,SIZEOF_JCOEF)] mov eax, dword [DWBLOCK(1,0,esi,SIZEOF_JCOEF)]
or eax, DWORD [DWBLOCK(2,0,esi,SIZEOF_JCOEF)] or eax, dword [DWBLOCK(2,0,esi,SIZEOF_JCOEF)]
jnz short .columnDCT jnz short .columnDCT
movq mm0, MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)] movq mm0, MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)]
@@ -464,16 +462,16 @@ EXTN(jsimd_idct_4x4_mmx):
mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW] mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW]
mov esi, JSAMPROW [edi+2*SIZEOF_JSAMPROW] mov esi, JSAMPROW [edi+2*SIZEOF_JSAMPROW]
movd DWORD [edx+eax*SIZEOF_JSAMPLE], mm1 movd dword [edx+eax*SIZEOF_JSAMPLE], mm1
movd DWORD [esi+eax*SIZEOF_JSAMPLE], mm0 movd dword [esi+eax*SIZEOF_JSAMPLE], mm0
psrlq mm1, 4*BYTE_BIT psrlq mm1, 4*BYTE_BIT
psrlq mm0, 4*BYTE_BIT psrlq mm0, 4*BYTE_BIT
mov edx, JSAMPROW [edi+1*SIZEOF_JSAMPROW] mov edx, JSAMPROW [edi+1*SIZEOF_JSAMPROW]
mov esi, JSAMPROW [edi+3*SIZEOF_JSAMPROW] mov esi, JSAMPROW [edi+3*SIZEOF_JSAMPROW]
movd DWORD [edx+eax*SIZEOF_JSAMPLE], mm1 movd dword [edx+eax*SIZEOF_JSAMPLE], mm1
movd DWORD [esi+eax*SIZEOF_JSAMPLE], mm0 movd dword [esi+eax*SIZEOF_JSAMPLE], mm0
emms ; empty MMX state emms ; empty MMX state
@@ -688,8 +686,8 @@ EXTN(jsimd_idct_2x2_mmx):
mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW] mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW]
mov esi, JSAMPROW [edi+1*SIZEOF_JSAMPROW] mov esi, JSAMPROW [edi+1*SIZEOF_JSAMPROW]
mov WORD [edx+eax*SIZEOF_JSAMPLE], bx mov word [edx+eax*SIZEOF_JSAMPLE], bx
mov WORD [esi+eax*SIZEOF_JSAMPLE], cx mov word [esi+eax*SIZEOF_JSAMPLE], cx
emms ; empty MMX state emms ; empty MMX state

10
simd/i386/jidctred-sse2.asm
View File

@@ -18,8 +18,6 @@
; output: either 4x4 or 2x2 pixels from an 8x8 DCT block. ; output: either 4x4 or 2x2 pixels from an 8x8 DCT block.
; The following code is based directly on the IJG's original jidctred.c; ; The following code is based directly on the IJG's original jidctred.c;
; see the jidctred.c for more details. ; see the jidctred.c for more details.
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"
@@ -139,8 +137,8 @@ EXTN(jsimd_idct_4x4_sse2):
mov esi, JCOEFPTR [coef_block(eax)] ; inptr mov esi, JCOEFPTR [coef_block(eax)] ; inptr
%ifndef NO_ZERO_COLUMN_TEST_4X4_SSE2 %ifndef NO_ZERO_COLUMN_TEST_4X4_SSE2
mov eax, DWORD [DWBLOCK(1,0,esi,SIZEOF_JCOEF)] mov eax, dword [DWBLOCK(1,0,esi,SIZEOF_JCOEF)]
or eax, DWORD [DWBLOCK(2,0,esi,SIZEOF_JCOEF)] or eax, dword [DWBLOCK(2,0,esi,SIZEOF_JCOEF)]
jnz short .columnDCT jnz short .columnDCT
movdqa xmm0, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)] movdqa xmm0, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)]
@@ -578,8 +576,8 @@ EXTN(jsimd_idct_2x2_sse2):
mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW] mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW]
mov esi, JSAMPROW [edi+1*SIZEOF_JSAMPROW] mov esi, JSAMPROW [edi+1*SIZEOF_JSAMPROW]
mov WORD [edx+eax*SIZEOF_JSAMPLE], bx mov word [edx+eax*SIZEOF_JSAMPLE], bx
mov WORD [esi+eax*SIZEOF_JSAMPLE], cx mov word [esi+eax*SIZEOF_JSAMPLE], cx
pop edi pop edi
pop esi pop esi

2
simd/i386/jquant-3dn.asm
View File

@@ -13,8 +13,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"

2
simd/i386/jquant-mmx.asm
View File

@@ -13,8 +13,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"

2
simd/i386/jquant-sse.asm
View File

@@ -13,8 +13,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"

2
simd/i386/jquantf-sse2.asm
View File

@@ -13,8 +13,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"

2
simd/i386/jquanti-avx2.asm
View File

@@ -14,8 +14,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"

2
simd/i386/jquanti-sse2.asm
View File

@@ -13,8 +13,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"

2
simd/i386/jsimdcpu.asm
View File

@@ -13,8 +13,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"

2
simd/nasm/jcolsamp.inc
View File

@@ -7,8 +7,6 @@
; Based on the x86 SIMD extension for IJG JPEG library ; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru. ; Copyright (C) 1999-2006, MIYASAKA Masaru.
; For conditions of distribution and use, see copyright notice in jsimdext.inc ; For conditions of distribution and use, see copyright notice in jsimdext.inc
;
; [TAB8]
; -------------------------------------------------------------------------- ; --------------------------------------------------------------------------

2
simd/nasm/jdct.inc
View File

@@ -7,8 +7,6 @@
; Based on the x86 SIMD extension for IJG JPEG library ; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru. ; Copyright (C) 1999-2006, MIYASAKA Masaru.
; For conditions of distribution and use, see copyright notice in jsimdext.inc ; For conditions of distribution and use, see copyright notice in jsimdext.inc
;
; [TAB8]
; Each IDCT routine is responsible for range-limiting its results and ; Each IDCT routine is responsible for range-limiting its results and
; converting them to unsigned form (0..MAXJSAMPLE). The raw outputs could ; converting them to unsigned form (0..MAXJSAMPLE). The raw outputs could

24
simd/nasm/jsimdext.inc
View File

@@ -25,8 +25,6 @@
; 2. Altered source versions must be plainly marked as such, and must not be ; 2. Altered source versions must be plainly marked as such, and must not be
; misrepresented as being the original software. ; misrepresented as being the original software.
; 3. This notice may not be removed or altered from any source distribution. ; 3. This notice may not be removed or altered from any source distribution.
;
; [TAB8]
; ========================================================================== ; ==========================================================================
; System-dependent configurations ; System-dependent configurations
@@ -204,19 +202,19 @@ section .note.GNU-stack noalloc noexec nowrite progbits
%define XMM_DWORD %define XMM_DWORD
%define XMM_MMWORD %define XMM_MMWORD
%define SIZEOF_BYTE 1 ; sizeof(BYTE) %define SIZEOF_BYTE 1 ; sizeof(byte)
%define SIZEOF_WORD 2 ; sizeof(WORD) %define SIZEOF_WORD 2 ; sizeof(word)
%define SIZEOF_DWORD 4 ; sizeof(DWORD) %define SIZEOF_DWORD 4 ; sizeof(dword)
%define SIZEOF_QWORD 8 ; sizeof(QWORD) %define SIZEOF_QWORD 8 ; sizeof(qword)
%define SIZEOF_OWORD 16 ; sizeof(OWORD) %define SIZEOF_OWORD 16 ; sizeof(oword)
%define SIZEOF_YWORD 32 ; sizeof(YWORD) %define SIZEOF_YWORD 32 ; sizeof(yword)
%define BYTE_BIT 8 ; CHAR_BIT in C %define BYTE_BIT 8 ; CHAR_BIT in C
%define WORD_BIT 16 ; sizeof(WORD)*BYTE_BIT %define WORD_BIT 16 ; sizeof(word)*BYTE_BIT
%define DWORD_BIT 32 ; sizeof(DWORD)*BYTE_BIT %define DWORD_BIT 32 ; sizeof(dword)*BYTE_BIT
%define QWORD_BIT 64 ; sizeof(QWORD)*BYTE_BIT %define QWORD_BIT 64 ; sizeof(qword)*BYTE_BIT
%define OWORD_BIT 128 ; sizeof(OWORD)*BYTE_BIT %define OWORD_BIT 128 ; sizeof(oword)*BYTE_BIT
%define YWORD_BIT 256 ; sizeof(YWORD)*BYTE_BIT %define YWORD_BIT 256 ; sizeof(yword)*BYTE_BIT
; -------------------------------------------------------------------------- ; --------------------------------------------------------------------------
; External Symbol Name ; External Symbol Name

6
simd/x86_64/jccolext-avx2.asm
View File

@@ -14,8 +14,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jcolsamp.inc" %include "jcolsamp.inc"
@@ -97,12 +95,12 @@ EXTN(jsimd_rgb_ycc_convert_avx2):
test cl, SIZEOF_BYTE test cl, SIZEOF_BYTE
jz short .column_ld2 jz short .column_ld2
sub rcx, byte SIZEOF_BYTE sub rcx, byte SIZEOF_BYTE
movzx rax, BYTE [rsi+rcx] movzx rax, byte [rsi+rcx]
.column_ld2: .column_ld2:
test cl, SIZEOF_WORD test cl, SIZEOF_WORD
jz short .column_ld4 jz short .column_ld4
sub rcx, byte SIZEOF_WORD sub rcx, byte SIZEOF_WORD
movzx rdx, WORD [rsi+rcx] movzx rdx, word [rsi+rcx]
shl rax, WORD_BIT shl rax, WORD_BIT
or rax, rdx or rax, rdx
.column_ld4: .column_ld4:

6
simd/x86_64/jccolext-sse2.asm
View File

@@ -13,8 +13,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jcolsamp.inc" %include "jcolsamp.inc"
@@ -96,12 +94,12 @@ EXTN(jsimd_rgb_ycc_convert_sse2):
test cl, SIZEOF_BYTE test cl, SIZEOF_BYTE
jz short .column_ld2 jz short .column_ld2
sub rcx, byte SIZEOF_BYTE sub rcx, byte SIZEOF_BYTE
movzx rax, BYTE [rsi+rcx] movzx rax, byte [rsi+rcx]
.column_ld2: .column_ld2:
test cl, SIZEOF_WORD test cl, SIZEOF_WORD
jz short .column_ld4 jz short .column_ld4
sub rcx, byte SIZEOF_WORD sub rcx, byte SIZEOF_WORD
movzx rdx, WORD [rsi+rcx] movzx rdx, word [rsi+rcx]
shl rax, WORD_BIT shl rax, WORD_BIT
or rax, rdx or rax, rdx
.column_ld4: .column_ld4:

2
simd/x86_64/jccolor-avx2.asm
View File

@@ -13,8 +13,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"

2
simd/x86_64/jccolor-sse2.asm
View File

@@ -12,8 +12,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"

2
simd/x86_64/jcgray-avx2.asm
View File

@@ -13,8 +13,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"

2
simd/x86_64/jcgray-sse2.asm
View File

@@ -12,8 +12,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"

6
simd/x86_64/jcgryext-avx2.asm
View File

@@ -14,8 +14,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jcolsamp.inc" %include "jcolsamp.inc"
@@ -89,12 +87,12 @@ EXTN(jsimd_rgb_gray_convert_avx2):
test cl, SIZEOF_BYTE test cl, SIZEOF_BYTE
jz short .column_ld2 jz short .column_ld2
sub rcx, byte SIZEOF_BYTE sub rcx, byte SIZEOF_BYTE
movzx rax, BYTE [rsi+rcx] movzx rax, byte [rsi+rcx]
.column_ld2: .column_ld2:
test cl, SIZEOF_WORD test cl, SIZEOF_WORD
jz short .column_ld4 jz short .column_ld4
sub rcx, byte SIZEOF_WORD sub rcx, byte SIZEOF_WORD
movzx rdx, WORD [rsi+rcx] movzx rdx, word [rsi+rcx]
shl rax, WORD_BIT shl rax, WORD_BIT
or rax, rdx or rax, rdx
.column_ld4: .column_ld4:

6
simd/x86_64/jcgryext-sse2.asm
View File

@@ -13,8 +13,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jcolsamp.inc" %include "jcolsamp.inc"
@@ -88,12 +86,12 @@ EXTN(jsimd_rgb_gray_convert_sse2):
test cl, SIZEOF_BYTE test cl, SIZEOF_BYTE
jz short .column_ld2 jz short .column_ld2
sub rcx, byte SIZEOF_BYTE sub rcx, byte SIZEOF_BYTE
movzx rax, BYTE [rsi+rcx] movzx rax, byte [rsi+rcx]
.column_ld2: .column_ld2:
test cl, SIZEOF_WORD test cl, SIZEOF_WORD
jz short .column_ld4 jz short .column_ld4
sub rcx, byte SIZEOF_WORD sub rcx, byte SIZEOF_WORD
movzx rdx, WORD [rsi+rcx] movzx rdx, word [rsi+rcx]
shl rax, WORD_BIT shl rax, WORD_BIT
or rax, rdx or rax, rdx
.column_ld4: .column_ld4:

6
simd/x86_64/jchuff-sse2.asm
View File

@@ -18,8 +18,6 @@
; This file contains an SSE2 implementation for Huffman coding of one block. ; This file contains an SSE2 implementation for Huffman coding of one block.
; The following code is based directly on jchuff.c; see jchuff.c for more ; The following code is based directly on jchuff.c; see jchuff.c for more
; details. ; details.
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
@@ -201,7 +199,7 @@ EXTN(jsimd_huff_encode_one_block_sse2):
mov buffer, r11 ; r11 is now sratch mov buffer, r11 ; r11 is now sratch
mov put_buffer, MMWORD [r10+SIZEOF_POINTER*2] ; put_buffer = state->cur.put_buffer; mov put_buffer, MMWORD [r10+SIZEOF_POINTER*2] ; put_buffer = state->cur.put_buffer;
mov put_bits, DWORD [r10+SIZEOF_POINTER*2+8] ; put_bits = state->cur.put_bits; mov put_bits, dword [r10+SIZEOF_POINTER*2+8] ; put_bits = state->cur.put_bits;
push r10 ; r10 is now scratch push r10 ; r10 is now scratch
; Encode the DC coefficient difference per section F.1.2.1 ; Encode the DC coefficient difference per section F.1.2.1
@@ -334,7 +332,7 @@ EXTN(jsimd_huff_encode_one_block_sse2):
pop r10 pop r10
; Save put_buffer & put_bits ; Save put_buffer & put_bits
mov MMWORD [r10+SIZEOF_POINTER*2], put_buffer ; state->cur.put_buffer = put_buffer; mov MMWORD [r10+SIZEOF_POINTER*2], put_buffer ; state->cur.put_buffer = put_buffer;
mov DWORD [r10+SIZEOF_POINTER*2+8], put_bits ; state->cur.put_bits = put_bits; mov dword [r10+SIZEOF_POINTER*2+8], put_bits ; state->cur.put_bits = put_bits;
pop rbx pop rbx
uncollect_args 6 uncollect_args 6

2
simd/x86_64/jcphuff-sse2.asm
View File

@@ -16,8 +16,6 @@
; ;
; This file contains an SSE2 implementation of data preparation for progressive ; This file contains an SSE2 implementation of data preparation for progressive
; Huffman encoding. See jcphuff.c for more details. ; Huffman encoding. See jcphuff.c for more details.
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"

2
simd/x86_64/jcsample-avx2.asm
View File

@@ -15,8 +15,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"

2
simd/x86_64/jcsample-sse2.asm
View File

@@ -14,8 +14,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"

6
simd/x86_64/jdcolext-avx2.asm
View File

@@ -15,8 +15,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jcolsamp.inc" %include "jcolsamp.inc"
@@ -335,7 +333,7 @@ EXTN(jsimd_ycc_rgb_convert_avx2):
vmovd eax, xmmA vmovd eax, xmmA
cmp rcx, byte SIZEOF_WORD cmp rcx, byte SIZEOF_WORD
jb short .column_st1 jb short .column_st1
mov WORD [rdi], ax mov word [rdi], ax
add rdi, byte SIZEOF_WORD add rdi, byte SIZEOF_WORD
sub rcx, byte SIZEOF_WORD sub rcx, byte SIZEOF_WORD
shr rax, 16 shr rax, 16
@@ -344,7 +342,7 @@ EXTN(jsimd_ycc_rgb_convert_avx2):
; space. ; space.
test rcx, rcx test rcx, rcx
jz short .nextrow jz short .nextrow
mov BYTE [rdi], al mov byte [rdi], al
%else ; RGB_PIXELSIZE == 4 ; ----------- %else ; RGB_PIXELSIZE == 4 ; -----------

6
simd/x86_64/jdcolext-sse2.asm
View File

@@ -14,8 +14,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jcolsamp.inc" %include "jcolsamp.inc"
@@ -307,7 +305,7 @@ EXTN(jsimd_ycc_rgb_convert_sse2):
movd eax, xmmA movd eax, xmmA
cmp rcx, byte SIZEOF_WORD cmp rcx, byte SIZEOF_WORD
jb short .column_st1 jb short .column_st1
mov WORD [rdi], ax mov word [rdi], ax
add rdi, byte SIZEOF_WORD add rdi, byte SIZEOF_WORD
sub rcx, byte SIZEOF_WORD sub rcx, byte SIZEOF_WORD
shr rax, 16 shr rax, 16
@@ -316,7 +314,7 @@ EXTN(jsimd_ycc_rgb_convert_sse2):
; space. ; space.
test rcx, rcx test rcx, rcx
jz short .nextrow jz short .nextrow
mov BYTE [rdi], al mov byte [rdi], al
%else ; RGB_PIXELSIZE == 4 ; ----------- %else ; RGB_PIXELSIZE == 4 ; -----------

2
simd/x86_64/jdcolor-avx2.asm
View File

@@ -14,8 +14,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"

2
simd/x86_64/jdcolor-sse2.asm
View File

@@ -13,8 +13,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"

2
simd/x86_64/jdmerge-avx2.asm
View File

@@ -14,8 +14,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"

2
simd/x86_64/jdmerge-sse2.asm
View File

@@ -13,8 +13,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"

6
simd/x86_64/jdmrgext-avx2.asm
View File

@@ -15,8 +15,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jcolsamp.inc" %include "jcolsamp.inc"
@@ -340,7 +338,7 @@ EXTN(jsimd_h2v1_merged_upsample_avx2):
vmovd eax, xmmA vmovd eax, xmmA
cmp rcx, byte SIZEOF_WORD cmp rcx, byte SIZEOF_WORD
jb short .column_st1 jb short .column_st1
mov WORD [rdi], ax mov word [rdi], ax
add rdi, byte SIZEOF_WORD add rdi, byte SIZEOF_WORD
sub rcx, byte SIZEOF_WORD sub rcx, byte SIZEOF_WORD
shr rax, 16 shr rax, 16
@@ -349,7 +347,7 @@ EXTN(jsimd_h2v1_merged_upsample_avx2):
; space. ; space.
test rcx, rcx test rcx, rcx
jz short .endcolumn jz short .endcolumn
mov BYTE [rdi], al mov byte [rdi], al
%else ; RGB_PIXELSIZE == 4 ; ----------- %else ; RGB_PIXELSIZE == 4 ; -----------

6
simd/x86_64/jdmrgext-sse2.asm
View File

@@ -14,8 +14,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jcolsamp.inc" %include "jcolsamp.inc"
@@ -311,7 +309,7 @@ EXTN(jsimd_h2v1_merged_upsample_sse2):
movd eax, xmmA movd eax, xmmA
cmp rcx, byte SIZEOF_WORD cmp rcx, byte SIZEOF_WORD
jb short .column_st1 jb short .column_st1
mov WORD [rdi], ax mov word [rdi], ax
add rdi, byte SIZEOF_WORD add rdi, byte SIZEOF_WORD
sub rcx, byte SIZEOF_WORD sub rcx, byte SIZEOF_WORD
shr rax, 16 shr rax, 16
@@ -320,7 +318,7 @@ EXTN(jsimd_h2v1_merged_upsample_sse2):
; space. ; space.
test rcx, rcx test rcx, rcx
jz short .endcolumn jz short .endcolumn
mov BYTE [rdi], al mov byte [rdi], al
%else ; RGB_PIXELSIZE == 4 ; ----------- %else ; RGB_PIXELSIZE == 4 ; -----------

2
simd/x86_64/jdsample-avx2.asm
View File

@@ -15,8 +15,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"

2
simd/x86_64/jdsample-sse2.asm
View File

@@ -14,8 +14,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"

2
simd/x86_64/jfdctflt-sse.asm
View File

@@ -17,8 +17,6 @@
; This file contains a floating-point implementation of the forward DCT ; This file contains a floating-point implementation of the forward DCT
; (Discrete Cosine Transform). The following code is based directly on ; (Discrete Cosine Transform). The following code is based directly on
; the IJG's original jfdctflt.c; see the jfdctflt.c for more details. ; the IJG's original jfdctflt.c; see the jfdctflt.c for more details.
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"

2
simd/x86_64/jfdctfst-sse2.asm
View File

@@ -18,8 +18,6 @@
; the forward DCT (Discrete Cosine Transform). The following code is ; the forward DCT (Discrete Cosine Transform). The following code is
; based directly on the IJG's original jfdctfst.c; see the jfdctfst.c ; based directly on the IJG's original jfdctfst.c; see the jfdctfst.c
; for more details. ; for more details.
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"

2
simd/x86_64/jfdctint-avx2.asm
View File

@@ -18,8 +18,6 @@
; forward DCT (Discrete Cosine Transform). The following code is based ; forward DCT (Discrete Cosine Transform). The following code is based
; directly on the IJG's original jfdctint.c; see the jfdctint.c for ; directly on the IJG's original jfdctint.c; see the jfdctint.c for
; more details. ; more details.
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"

2
simd/x86_64/jfdctint-sse2.asm
View File

@@ -18,8 +18,6 @@
; forward DCT (Discrete Cosine Transform). The following code is based ; forward DCT (Discrete Cosine Transform). The following code is based
; directly on the IJG's original jfdctint.c; see the jfdctint.c for ; directly on the IJG's original jfdctint.c; see the jfdctint.c for
; more details. ; more details.
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"

6
simd/x86_64/jidctflt-sse2.asm
View File

@@ -18,8 +18,6 @@
; This file contains a floating-point implementation of the inverse DCT ; This file contains a floating-point implementation of the inverse DCT
; (Discrete Cosine Transform). The following code is based directly on ; (Discrete Cosine Transform). The following code is based directly on
; the IJG's original jidctflt.c; see the jidctflt.c for more details. ; the IJG's original jidctflt.c; see the jidctflt.c for more details.
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"
@@ -96,8 +94,8 @@ EXTN(jsimd_idct_float_sse2):
mov rcx, DCTSIZE/4 ; ctr mov rcx, DCTSIZE/4 ; ctr
.columnloop: .columnloop:
%ifndef NO_ZERO_COLUMN_TEST_FLOAT_SSE %ifndef NO_ZERO_COLUMN_TEST_FLOAT_SSE
mov eax, DWORD [DWBLOCK(1,0,rsi,SIZEOF_JCOEF)] mov eax, dword [DWBLOCK(1,0,rsi,SIZEOF_JCOEF)]
or eax, DWORD [DWBLOCK(2,0,rsi,SIZEOF_JCOEF)] or eax, dword [DWBLOCK(2,0,rsi,SIZEOF_JCOEF)]
jnz near .columnDCT jnz near .columnDCT
movq xmm1, XMM_MMWORD [MMBLOCK(1,0,rsi,SIZEOF_JCOEF)] movq xmm1, XMM_MMWORD [MMBLOCK(1,0,rsi,SIZEOF_JCOEF)]

6
simd/x86_64/jidctfst-sse2.asm
View File

@@ -19,8 +19,6 @@
; the inverse DCT (Discrete Cosine Transform). The following code is ; the inverse DCT (Discrete Cosine Transform). The following code is
; based directly on the IJG's original jidctfst.c; see the jidctfst.c ; based directly on the IJG's original jidctfst.c; see the jidctfst.c
; for more details. ; for more details.
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"
@@ -112,8 +110,8 @@ EXTN(jsimd_idct_ifast_sse2):
mov rsi, r11 ; inptr mov rsi, r11 ; inptr
%ifndef NO_ZERO_COLUMN_TEST_IFAST_SSE2 %ifndef NO_ZERO_COLUMN_TEST_IFAST_SSE2
mov eax, DWORD [DWBLOCK(1,0,rsi,SIZEOF_JCOEF)] mov eax, dword [DWBLOCK(1,0,rsi,SIZEOF_JCOEF)]
or eax, DWORD [DWBLOCK(2,0,rsi,SIZEOF_JCOEF)] or eax, dword [DWBLOCK(2,0,rsi,SIZEOF_JCOEF)]
jnz near .columnDCT jnz near .columnDCT
movdqa xmm0, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)] movdqa xmm0, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)]

6
simd/x86_64/jidctint-avx2.asm
View File

@@ -19,8 +19,6 @@
; inverse DCT (Discrete Cosine Transform). The following code is based ; inverse DCT (Discrete Cosine Transform). The following code is based
; directly on the IJG's original jidctint.c; see the jidctint.c for ; directly on the IJG's original jidctint.c; see the jidctint.c for
; more details. ; more details.
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"
@@ -293,8 +291,8 @@ EXTN(jsimd_idct_islow_avx2):
; ---- Pass 1: process columns. ; ---- Pass 1: process columns.
%ifndef NO_ZERO_COLUMN_TEST_ISLOW_AVX2 %ifndef NO_ZERO_COLUMN_TEST_ISLOW_AVX2
mov eax, DWORD [DWBLOCK(1,0,r11,SIZEOF_JCOEF)] mov eax, dword [DWBLOCK(1,0,r11,SIZEOF_JCOEF)]
or eax, DWORD [DWBLOCK(2,0,r11,SIZEOF_JCOEF)] or eax, dword [DWBLOCK(2,0,r11,SIZEOF_JCOEF)]
jnz near .columnDCT jnz near .columnDCT
movdqa xmm0, XMMWORD [XMMBLOCK(1,0,r11,SIZEOF_JCOEF)] movdqa xmm0, XMMWORD [XMMBLOCK(1,0,r11,SIZEOF_JCOEF)]

6
simd/x86_64/jidctint-sse2.asm
View File

@@ -19,8 +19,6 @@
; inverse DCT (Discrete Cosine Transform). The following code is based ; inverse DCT (Discrete Cosine Transform). The following code is based
; directly on the IJG's original jidctint.c; see the jidctint.c for ; directly on the IJG's original jidctint.c; see the jidctint.c for
; more details. ; more details.
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"
@@ -125,8 +123,8 @@ EXTN(jsimd_idct_islow_sse2):
mov rsi, r11 ; inptr mov rsi, r11 ; inptr
%ifndef NO_ZERO_COLUMN_TEST_ISLOW_SSE2 %ifndef NO_ZERO_COLUMN_TEST_ISLOW_SSE2
mov eax, DWORD [DWBLOCK(1,0,rsi,SIZEOF_JCOEF)] mov eax, dword [DWBLOCK(1,0,rsi,SIZEOF_JCOEF)]
or eax, DWORD [DWBLOCK(2,0,rsi,SIZEOF_JCOEF)] or eax, dword [DWBLOCK(2,0,rsi,SIZEOF_JCOEF)]
jnz near .columnDCT jnz near .columnDCT
movdqa xmm0, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)] movdqa xmm0, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)]

10
simd/x86_64/jidctred-sse2.asm
View File

@@ -19,8 +19,6 @@
; output: either 4x4 or 2x2 pixels from an 8x8 DCT block. ; output: either 4x4 or 2x2 pixels from an 8x8 DCT block.
; The following code is based directly on the IJG's original jidctred.c; ; The following code is based directly on the IJG's original jidctred.c;
; see the jidctred.c for more details. ; see the jidctred.c for more details.
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"
@@ -133,8 +131,8 @@ EXTN(jsimd_idct_4x4_sse2):
mov rsi, r11 ; inptr mov rsi, r11 ; inptr
%ifndef NO_ZERO_COLUMN_TEST_4X4_SSE2 %ifndef NO_ZERO_COLUMN_TEST_4X4_SSE2
mov eax, DWORD [DWBLOCK(1,0,rsi,SIZEOF_JCOEF)] mov eax, dword [DWBLOCK(1,0,rsi,SIZEOF_JCOEF)]
or eax, DWORD [DWBLOCK(2,0,rsi,SIZEOF_JCOEF)] or eax, dword [DWBLOCK(2,0,rsi,SIZEOF_JCOEF)]
jnz short .columnDCT jnz short .columnDCT
movdqa xmm0, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)] movdqa xmm0, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)]
@@ -563,8 +561,8 @@ EXTN(jsimd_idct_2x2_sse2):
mov rdxp, JSAMPROW [rdi+0*SIZEOF_JSAMPROW] mov rdxp, JSAMPROW [rdi+0*SIZEOF_JSAMPROW]
mov rsip, JSAMPROW [rdi+1*SIZEOF_JSAMPROW] mov rsip, JSAMPROW [rdi+1*SIZEOF_JSAMPROW]
mov WORD [rdx+rax*SIZEOF_JSAMPLE], bx mov word [rdx+rax*SIZEOF_JSAMPLE], bx
mov WORD [rsi+rax*SIZEOF_JSAMPLE], cx mov word [rsi+rax*SIZEOF_JSAMPLE], cx
pop rbx pop rbx
uncollect_args 4 uncollect_args 4

2
simd/x86_64/jquantf-sse2.asm
View File

@@ -14,8 +14,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"

2
simd/x86_64/jquanti-avx2.asm
View File

@@ -15,8 +15,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"

2
simd/x86_64/jquanti-sse2.asm
View File

@@ -14,8 +14,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"
%include "jdct.inc" %include "jdct.inc"

2
simd/x86_64/jsimdcpu.asm
View File

@@ -14,8 +14,6 @@
; assembler (including Borland's Turbo Assembler). ; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or ; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208 ; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc" %include "jsimdext.inc"