Skip to content

L
libyuv
Commit 9c02ba53 authored by fbarchard@google.com's avatar fbarchard@google.com
Browse files
Options

ARGBUnattenuateRow_C use same table as ARGBUnattenuateRow_SSE2. … 

ARGBUnattenuateRow_C use same table as ARGBUnattenuateRow_SSE2.  omit-frame-pointer for ScaleRowDown38_3_Int_NEON
BUG=35
TESTED=disable asm and run unittest on x86
Review URL: https://webrtc-codereview.appspot.com/585004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@268 16f28f9a-4ce2-e073-06de-1de4eb20be90
parent 600587d1
Hide whitespace changes
Inline Side-by-side
Showing with 21 additions and 102 deletions
README.chromium
View file @ 9c02ba53
Name: libyuv
URL: http://code.google.com/p/libyuv/
Version: 267
Version: 268
License: BSD
License File: LICENSE
......
include/libyuv/version.h
View file @ 9c02ba53
......@@ -11,7 +11,7 @@
#ifndef INCLUDE_LIBYUV_VERSION_H_
#define INCLUDE_LIBYUV_VERSION_H_
#define LIBYUV_VERSION 267
#define LIBYUV_VERSION 268
#endif // INCLUDE_LIBYUV_VERSION_H_
source/row.h
View file @ 9c02ba53
......@@ -110,6 +110,12 @@ typedef uint16 __attribute__((vector_size(16))) uvec16;
typedef uint32 __attribute__((vector_size(16))) uvec32;
#endif
#if defined(__APPLE__) || defined(__x86_64__)
#define OMITFP
#else
#define OMITFP __attribute__((optimize("omit-frame-pointer")))
#endif
void I420ToARGBRow_NEON(const uint8* y_buf,
const uint8* u_buf,
const uint8* v_buf,
......@@ -388,6 +394,8 @@ void ARGBAttenuateRow_C(const uint8* src_argb, uint8* dst_argb, int width);
void ARGBAttenuateRow_SSE2(const uint8* src_argb, uint8* dst_argb, int width);
void ARGBAttenuateRow_SSSE3(const uint8* src_argb, uint8* dst_argb, int width);
// Inverse table for unattenuate, shared by C and SSE2.
extern uint32 fixed_invtbl8[256];
void ARGBUnattenuateRow_C(const uint8* src_argb, uint8* dst_argb, int width);
void ARGBUnattenuateRow_SSE2(const uint8* src_argb, uint8* dst_argb, int width);
......
source/row_common.cc
View file @ 9c02ba53
......@@ -730,9 +730,9 @@ void ARGBAttenuateRow_C(const uint8* src_argb, uint8* dst_argb, int width) {
// r = (r * 255 + (a / 2)) / a;
// Reciprocal method is off by 1 on some values. ie 125
// 8.16 fixed point inverse table
#define T(a) 0x1000000 / a
static uint32 fixed_invtbl[256] = {
0, T(0x01), T(0x02), T(0x03), T(0x04), T(0x05), T(0x06), T(0x07),
#define T(a) 0x10000 / a
uint32 fixed_invtbl8[256] = {
0x0100, T(0x01), T(0x02), T(0x03), T(0x04), T(0x05), T(0x06), T(0x07),
T(0x08), T(0x09), T(0x0a), T(0x0b), T(0x0c), T(0x0d), T(0x0e), T(0x0f),
T(0x10), T(0x11), T(0x12), T(0x13), T(0x14), T(0x15), T(0x16), T(0x17),
T(0x18), T(0x19), T(0x1a), T(0x1b), T(0x1c), T(0x1d), T(0x1e), T(0x1f),
......@@ -763,7 +763,7 @@ static uint32 fixed_invtbl[256] = {
T(0xe0), T(0xe1), T(0xe2), T(0xe3), T(0xe4), T(0xe5), T(0xe6), T(0xe7),
T(0xe8), T(0xe9), T(0xea), T(0xeb), T(0xec), T(0xed), T(0xee), T(0xef),
T(0xf0), T(0xf1), T(0xf2), T(0xf3), T(0xf4), T(0xf5), T(0xf6), T(0xf7),
T(0xf8), T(0xf9), T(0xfa), T(0xfb), T(0xfc), T(0xfd), T(0xfe), T(0xff) };
T(0xf8), T(0xf9), T(0xfa), T(0xfb), T(0xfc), T(0xfd), T(0xfe), 0x0100 };
#undef T
void ARGBUnattenuateRow_C(const uint8* src_argb, uint8* dst_argb, int width) {
......@@ -773,10 +773,10 @@ void ARGBUnattenuateRow_C(const uint8* src_argb, uint8* dst_argb, int width) {
uint32 r = src_argb[2];
const uint32 a = src_argb[3];
if (a) {
const uint32 ia = fixed_invtbl[a]; // 8.16 fixed point
b = (b * ia + 0x8000) >> 16;
g = (g * ia + 0x8000) >> 16;
r = (r * ia + 0x8000) >> 16;
const uint32 ia = fixed_invtbl8[a]; // 8.16 fixed point
b = (b * ia) >> 8;
g = (g * ia) >> 8;
r = (r * ia) >> 8;
// Clamping should not be necessary but is free in assembly.
if (b > 255) {
b = 255;
......
source/row_posix.cc
View file @ 9c02ba53
......@@ -1231,12 +1231,6 @@ void ABGRToUVRow_Unaligned_SSSE3(const uint8* src_abgr0, int src_stride_abgr,
#define YG 74 /* static_cast<int8>(1.164 * 64 + 0.5) */
#if defined(__APPLE__) || defined(__x86_64__)
#define OMITFP
#else
#define OMITFP __attribute__((optimize("omit-frame-pointer")))
#endif
struct {
vec8 kUVToB;
vec8 kUVToG;
......@@ -2479,48 +2473,6 @@ void ARGBAttenuateRow_SSSE3(const uint8* src_argb, uint8* dst_argb, int width) {
#endif // HAS_ARGBATTENUATE_SSSE3
#ifdef HAS_ARGBUNATTENUATE_SSE2
// Divide source RGB by alpha and store to destination.
// b = (b * 255 + (a / 2)) / a;
// g = (g * 255 + (a / 2)) / a;
// r = (r * 255 + (a / 2)) / a;
// Reciprocal method is off by 1 on some values. ie 125
// 8.16 fixed point inverse table
#define T(a) 0x10000 / a
CONST uint32 fixed_invtbl8[256] = {
0x100, 0xffff, T(0x02), T(0x03), T(0x04), T(0x05), T(0x06), T(0x07),
T(0x08), T(0x09), T(0x0a), T(0x0b), T(0x0c), T(0x0d), T(0x0e), T(0x0f),
T(0x10), T(0x11), T(0x12), T(0x13), T(0x14), T(0x15), T(0x16), T(0x17),
T(0x18), T(0x19), T(0x1a), T(0x1b), T(0x1c), T(0x1d), T(0x1e), T(0x1f),
T(0x20), T(0x21), T(0x22), T(0x23), T(0x24), T(0x25), T(0x26), T(0x27),
T(0x28), T(0x29), T(0x2a), T(0x2b), T(0x2c), T(0x2d), T(0x2e), T(0x2f),
T(0x30), T(0x31), T(0x32), T(0x33), T(0x34), T(0x35), T(0x36), T(0x37),
T(0x38), T(0x39), T(0x3a), T(0x3b), T(0x3c), T(0x3d), T(0x3e), T(0x3f),
T(0x40), T(0x41), T(0x42), T(0x43), T(0x44), T(0x45), T(0x46), T(0x47),
T(0x48), T(0x49), T(0x4a), T(0x4b), T(0x4c), T(0x4d), T(0x4e), T(0x4f),
T(0x50), T(0x51), T(0x52), T(0x53), T(0x54), T(0x55), T(0x56), T(0x57),
T(0x58), T(0x59), T(0x5a), T(0x5b), T(0x5c), T(0x5d), T(0x5e), T(0x5f),
T(0x60), T(0x61), T(0x62), T(0x63), T(0x64), T(0x65), T(0x66), T(0x67),
T(0x68), T(0x69), T(0x6a), T(0x6b), T(0x6c), T(0x6d), T(0x6e), T(0x6f),
T(0x70), T(0x71), T(0x72), T(0x73), T(0x74), T(0x75), T(0x76), T(0x77),
T(0x78), T(0x79), T(0x7a), T(0x7b), T(0x7c), T(0x7d), T(0x7e), T(0x7f),
T(0x80), T(0x81), T(0x82), T(0x83), T(0x84), T(0x85), T(0x86), T(0x87),
T(0x88), T(0x89), T(0x8a), T(0x8b), T(0x8c), T(0x8d), T(0x8e), T(0x8f),
T(0x90), T(0x91), T(0x92), T(0x93), T(0x94), T(0x95), T(0x96), T(0x97),
T(0x98), T(0x99), T(0x9a), T(0x9b), T(0x9c), T(0x9d), T(0x9e), T(0x9f),
T(0xa0), T(0xa1), T(0xa2), T(0xa3), T(0xa4), T(0xa5), T(0xa6), T(0xa7),
T(0xa8), T(0xa9), T(0xaa), T(0xab), T(0xac), T(0xad), T(0xae), T(0xaf),
T(0xb0), T(0xb1), T(0xb2), T(0xb3), T(0xb4), T(0xb5), T(0xb6), T(0xb7),
T(0xb8), T(0xb9), T(0xba), T(0xbb), T(0xbc), T(0xbd), T(0xbe), T(0xbf),
T(0xc0), T(0xc1), T(0xc2), T(0xc3), T(0xc4), T(0xc5), T(0xc6), T(0xc7),
T(0xc8), T(0xc9), T(0xca), T(0xcb), T(0xcc), T(0xcd), T(0xce), T(0xcf),
T(0xd0), T(0xd1), T(0xd2), T(0xd3), T(0xd4), T(0xd5), T(0xd6), T(0xd7),
T(0xd8), T(0xd9), T(0xda), T(0xdb), T(0xdc), T(0xdd), T(0xde), T(0xdf),
T(0xe0), T(0xe1), T(0xe2), T(0xe3), T(0xe4), T(0xe5), T(0xe6), T(0xe7),
T(0xe8), T(0xe9), T(0xea), T(0xeb), T(0xec), T(0xed), T(0xee), T(0xef),
T(0xf0), T(0xf1), T(0xf2), T(0xf3), T(0xf4), T(0xf5), T(0xf6), T(0xf7),
T(0xf8), T(0xf9), T(0xfa), T(0xfb), T(0xfc), T(0xfd), T(0xfe), 0x100 };
#undef T
// Unattenuate 4 pixels at a time.
// aligned to 16 bytes
void ARGBUnattenuateRow_SSE2(const uint8* src_argb, uint8* dst_argb,
......
source/row_win.cc
View file @ 9c02ba53
......@@ -2499,48 +2499,6 @@ void ARGBAttenuateRow_SSSE3(const uint8* src_argb, uint8* dst_argb, int width) {
#endif // HAS_ARGBATTENUATE_SSSE3
#ifdef HAS_ARGBUNATTENUATE_SSE2
// Divide source RGB by alpha and store to destination.
// b = (b * 255 + (a / 2)) / a;
// g = (g * 255 + (a / 2)) / a;
// r = (r * 255 + (a / 2)) / a;
// Reciprocal method is off by 1 on some values. ie 125
// 8.16 fixed point inverse table
#define T(a) 0x10000 / a
static uint32 fixed_invtbl8[256] = {
0x100, 0xffff, T(0x02), T(0x03), T(0x04), T(0x05), T(0x06), T(0x07),
T(0x08), T(0x09), T(0x0a), T(0x0b), T(0x0c), T(0x0d), T(0x0e), T(0x0f),
T(0x10), T(0x11), T(0x12), T(0x13), T(0x14), T(0x15), T(0x16), T(0x17),
T(0x18), T(0x19), T(0x1a), T(0x1b), T(0x1c), T(0x1d), T(0x1e), T(0x1f),
T(0x20), T(0x21), T(0x22), T(0x23), T(0x24), T(0x25), T(0x26), T(0x27),
T(0x28), T(0x29), T(0x2a), T(0x2b), T(0x2c), T(0x2d), T(0x2e), T(0x2f),
T(0x30), T(0x31), T(0x32), T(0x33), T(0x34), T(0x35), T(0x36), T(0x37),
T(0x38), T(0x39), T(0x3a), T(0x3b), T(0x3c), T(0x3d), T(0x3e), T(0x3f),
T(0x40), T(0x41), T(0x42), T(0x43), T(0x44), T(0x45), T(0x46), T(0x47),
T(0x48), T(0x49), T(0x4a), T(0x4b), T(0x4c), T(0x4d), T(0x4e), T(0x4f),
T(0x50), T(0x51), T(0x52), T(0x53), T(0x54), T(0x55), T(0x56), T(0x57),
T(0x58), T(0x59), T(0x5a), T(0x5b), T(0x5c), T(0x5d), T(0x5e), T(0x5f),
T(0x60), T(0x61), T(0x62), T(0x63), T(0x64), T(0x65), T(0x66), T(0x67),
T(0x68), T(0x69), T(0x6a), T(0x6b), T(0x6c), T(0x6d), T(0x6e), T(0x6f),
T(0x70), T(0x71), T(0x72), T(0x73), T(0x74), T(0x75), T(0x76), T(0x77),
T(0x78), T(0x79), T(0x7a), T(0x7b), T(0x7c), T(0x7d), T(0x7e), T(0x7f),
T(0x80), T(0x81), T(0x82), T(0x83), T(0x84), T(0x85), T(0x86), T(0x87),
T(0x88), T(0x89), T(0x8a), T(0x8b), T(0x8c), T(0x8d), T(0x8e), T(0x8f),
T(0x90), T(0x91), T(0x92), T(0x93), T(0x94), T(0x95), T(0x96), T(0x97),
T(0x98), T(0x99), T(0x9a), T(0x9b), T(0x9c), T(0x9d), T(0x9e), T(0x9f),
T(0xa0), T(0xa1), T(0xa2), T(0xa3), T(0xa4), T(0xa5), T(0xa6), T(0xa7),
T(0xa8), T(0xa9), T(0xaa), T(0xab), T(0xac), T(0xad), T(0xae), T(0xaf),
T(0xb0), T(0xb1), T(0xb2), T(0xb3), T(0xb4), T(0xb5), T(0xb6), T(0xb7),
T(0xb8), T(0xb9), T(0xba), T(0xbb), T(0xbc), T(0xbd), T(0xbe), T(0xbf),
T(0xc0), T(0xc1), T(0xc2), T(0xc3), T(0xc4), T(0xc5), T(0xc6), T(0xc7),
T(0xc8), T(0xc9), T(0xca), T(0xcb), T(0xcc), T(0xcd), T(0xce), T(0xcf),
T(0xd0), T(0xd1), T(0xd2), T(0xd3), T(0xd4), T(0xd5), T(0xd6), T(0xd7),
T(0xd8), T(0xd9), T(0xda), T(0xdb), T(0xdc), T(0xdd), T(0xde), T(0xdf),
T(0xe0), T(0xe1), T(0xe2), T(0xe3), T(0xe4), T(0xe5), T(0xe6), T(0xe7),
T(0xe8), T(0xe9), T(0xea), T(0xeb), T(0xec), T(0xed), T(0xee), T(0xef),
T(0xf0), T(0xf1), T(0xf2), T(0xf3), T(0xf4), T(0xf5), T(0xf6), T(0xf7),
T(0xf8), T(0xf9), T(0xfa), T(0xfb), T(0xfc), T(0xfd), T(0xfe), 0x100 };
#undef T
// Unattenuate 4 pixels at a time.
// aligned to 16 bytes
__declspec(naked) __declspec(align(16))
......
source/scale.cc
View file @ 9c02ba53
......@@ -298,8 +298,9 @@ static void ScaleRowDown38_NEON(const uint8* src_ptr, int,
}
// 32x3 -> 12x1
static void ScaleRowDown38_3_Int_NEON(const uint8* src_ptr, int src_stride,
uint8* dst_ptr, int dst_width) {
static void OMITFP ScaleRowDown38_3_Int_NEON(const uint8* src_ptr,
int src_stride,
uint8* dst_ptr, int dst_width) {
asm volatile (
"vld1.u16 {q13}, [%4] \n"
"vld1.u8 {q14}, [%5] \n"
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment