yyc12345/stb
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

stb_image: Add more of the SSE2 skeleton.

Browse Source
This commit is contained in:
Fabian Giesen 2014-12-18 07:03:21 -08:00
parent c6a3235995
commit fb2c841bb8
1 changed files with 30 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

35
stb_image.h
View File

@ -191,6 +191,19 @@
// //
// The three functions you must define are "read" (reads some bytes of data), // The three functions you must define are "read" (reads some bytes of data),
// "skip" (skips some bytes of data), "eof" (reports if the stream is at the end). // "skip" (skips some bytes of data), "eof" (reports if the stream is at the end).
//
// ===========================================================================
//
// SIMD support
//
// The JPEG decoder will automatically use SIMD kernels where supported,
// replacing the STBI_SIMD-do-it-yourself interface from previous versions.
// The code will automatically detect if the required SIMD instructions are
// available, and fall back to the generic C version where they're not.
//
// The supplied kernels are designed to produce results that are bit-identical
// to the C versions. Nevertheless, if you want to disable this functionality,
// define STBI_NO_SIMD.
#ifndef STBI_NO_STDIO #ifndef STBI_NO_STDIO
@ -404,6 +417,21 @@ typedef unsigned char validate_uint32[sizeof(stbi__uint32)==4 ? 1 : -1];
#define stbi_lrot(x,y) (((x) << (y)) | ((x) >> (32 - (y)))) #define stbi_lrot(x,y) (((x) << (y)) | ((x) >> (32 - (y))))
#endif #endif
#if !defined(STBI_NO_SIMD) && (defined(__x86_64__) || defined(_M_X64) || defined(__i386) || defined(_M_IX86))
#define STBI_SSE2
#include <emmintrin.h>
#ifdef _MSC_VER
#define STBI_SIMD_ALIGN(type, name) __declspec(align(16)) type name
#else // assume GCC-style if not VC++
#define STBI_SIMD_ALIGN(type, name) type name __attribute__((aligned(16)))
#endif
#endif
#ifndef STBI_SIMD_ALIGN
#define STBI_SIMD_ALIGN(type, name) type name
#endif
/////////////////////////////////////////////// ///////////////////////////////////////////////
// //
// stbi__context struct and start_xxx functions // stbi__context struct and start_xxx functions
@ -1445,10 +1473,7 @@ static int stbi__parse_entropy_coded_data(stbi__jpeg *z)
stbi__jpeg_reset(z); stbi__jpeg_reset(z);
if (z->scan_n == 1) { if (z->scan_n == 1) {
int i,j; int i,j;
#ifdef STBI_SIMD STBI_SIMD_ALIGN(short, data[64]);
__declspec(align(16))
#endif
short data[64];
int n = z->order[0]; int n = z->order[0];
// non-interleaved data, we just need to process one block at a time, // non-interleaved data, we just need to process one block at a time,
// in trivial scanline order // in trivial scanline order
@ -1473,7 +1498,7 @@ static int stbi__parse_entropy_coded_data(stbi__jpeg *z)
} }
} else { // interleaved! } else { // interleaved!
int i,j,k,x,y; int i,j,k,x,y;
short data[64]; STBI_SIMD_ALIGN(short, data[64]);
for (j=0; j < z->img_mcu_y; ++j) { for (j=0; j < z->img_mcu_y; ++j) {
for (i=0; i < z->img_mcu_x; ++i) { for (i=0; i < z->img_mcu_x; ++i) {
// scan an interleaved mcu... process scan_n components in order // scan an interleaved mcu... process scan_n components in order