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

Lay the groundwork for 32-bit AVX2 SIMD support

Browse Source
This commit is contained in:
DRC
2016-07-05 16:19:26 -05:00
parent b5426c52c0
commit 7ee3ce9ada
45 changed files with 196 additions and 164 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
simd/jidctflt-sse.asm
View File

@@ -2,6 +2,7 @@
; jidctflt.asm - floating-point IDCT (SSE & MMX)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
; Copyright (C) 2016, D. R. Commander.
;
; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -35,7 +36,7 @@
; --------------------------------------------------------------------------
SECTION SEG_CONST
alignz 16
alignz 32
global EXTN(jconst_idct_float_sse)
EXTN(jconst_idct_float_sse):
@@ -47,7 +48,7 @@ PD_M2_613 times 4 dd -2.613125929752753055713286
PD_0_125 times 4 dd 0.125 ; 1/8
PB_CENTERJSAMP times 8 db CENTERJSAMPLE
alignz 16
alignz 32
; --------------------------------------------------------------------------
SECTION SEG_TEXT
@@ -71,7 +72,7 @@ PB_CENTERJSAMP times 8 db CENTERJSAMPLE
%define workspace wk(0)-DCTSIZE2*SIZEOF_FAST_FLOAT
; FAST_FLOAT workspace[DCTSIZE2]
align 16
align 32
global EXTN(jsimd_idct_float_sse)
EXTN(jsimd_idct_float_sse):
@@ -568,4 +569,4 @@ EXTN(jsimd_idct_float_sse):
; For some reason, the OS X linker does not honor the request to align the
; segment unless we do this.
align 16
align 32