| 1 | /* intel_init.c - SSE2 optimized filter functions
|
|---|
| 2 | *
|
|---|
| 3 | * Copyright (c) 2018 Cosmin Truta
|
|---|
| 4 | * Copyright (c) 2016-2017 Glenn Randers-Pehrson
|
|---|
| 5 | * Written by Mike Klein and Matt Sarett, Google, Inc.
|
|---|
| 6 | * Derived from arm/arm_init.c
|
|---|
| 7 | *
|
|---|
| 8 | * This code is released under the libpng license.
|
|---|
| 9 | * For conditions of distribution and use, see the disclaimer
|
|---|
| 10 | * and license in png.h
|
|---|
| 11 | */
|
|---|
| 12 |
|
|---|
| 13 | #include "../pngpriv.h"
|
|---|
| 14 |
|
|---|
| 15 | #ifdef PNG_READ_SUPPORTED
|
|---|
| 16 | #if PNG_INTEL_SSE_IMPLEMENTATION > 0
|
|---|
| 17 |
|
|---|
| 18 | void
|
|---|
| 19 | png_init_filter_functions_sse2(png_structp pp, unsigned int bpp)
|
|---|
| 20 | {
|
|---|
| 21 | /* The techniques used to implement each of these filters in SSE operate on
|
|---|
| 22 | * one pixel at a time.
|
|---|
| 23 | * So they generally speed up 3bpp images about 3x, 4bpp images about 4x.
|
|---|
| 24 | * They can scale up to 6 and 8 bpp images and down to 2 bpp images,
|
|---|
| 25 | * but they'd not likely have any benefit for 1bpp images.
|
|---|
| 26 | * Most of these can be implemented using only MMX and 64-bit registers,
|
|---|
| 27 | * but they end up a bit slower than using the equally-ubiquitous SSE2.
|
|---|
| 28 | */
|
|---|
| 29 | png_debug(1, "in png_init_filter_functions_sse2");
|
|---|
| 30 | if (bpp == 3)
|
|---|
| 31 | {
|
|---|
| 32 | pp->read_filter[PNG_FILTER_VALUE_SUB-1] = png_read_filter_row_sub3_sse2;
|
|---|
| 33 | pp->read_filter[PNG_FILTER_VALUE_AVG-1] = png_read_filter_row_avg3_sse2;
|
|---|
| 34 | pp->read_filter[PNG_FILTER_VALUE_PAETH-1] =
|
|---|
| 35 | png_read_filter_row_paeth3_sse2;
|
|---|
| 36 | }
|
|---|
| 37 | else if (bpp == 4)
|
|---|
| 38 | {
|
|---|
| 39 | pp->read_filter[PNG_FILTER_VALUE_SUB-1] = png_read_filter_row_sub4_sse2;
|
|---|
| 40 | pp->read_filter[PNG_FILTER_VALUE_AVG-1] = png_read_filter_row_avg4_sse2;
|
|---|
| 41 | pp->read_filter[PNG_FILTER_VALUE_PAETH-1] =
|
|---|
| 42 | png_read_filter_row_paeth4_sse2;
|
|---|
| 43 | }
|
|---|
| 44 |
|
|---|
| 45 | /* No need optimize PNG_FILTER_VALUE_UP. The compiler should
|
|---|
| 46 | * autovectorize.
|
|---|
| 47 | */
|
|---|
| 48 | }
|
|---|
| 49 |
|
|---|
| 50 | #endif /* PNG_INTEL_SSE_IMPLEMENTATION > 0 */
|
|---|
| 51 | #endif /* PNG_READ_SUPPORTED */
|
|---|
