Skip to content

L
libyuv
Commit 0cc0b4df authored by fbarchard@google.com's avatar fbarchard@google.com
Browse files
Options

Neon Color Matrix avoid overflow 

BUG=214
TEST=planar_tests
Review URL: https://webrtc-codereview.appspot.com/1276004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@644 16f28f9a-4ce2-e073-06de-1de4eb20be90
parent 0057aeb1
Hide whitespace changes
Inline Side-by-side
Showing with 34 additions and 26 deletions
README.chromium
View file @ 0cc0b4df
Name: libyuv
URL: http://code.google.com/p/libyuv/
Version: 643
Version: 644
License: BSD
License File: LICENSE
......
include/libyuv/row.h
View file @ 0cc0b4df
......@@ -260,8 +260,7 @@ extern "C" {
#define HAS_ARGBADDROW_NEON
#define HAS_ARGBATTENUATEROW_NEON
#define HAS_ARGBBLENDROW_NEON
// TODO(fbarchard): fix and enable
// #define HAS_ARGBCOLORMATRIXROW_NEON
#define HAS_ARGBCOLORMATRIXROW_NEON
#define HAS_ARGBGRAYROW_NEON
#define HAS_ARGBINTERPOLATEROW_NEON
#define HAS_ARGBMIRRORROW_NEON
......
include/libyuv/version.h
View file @ 0cc0b4df
......@@ -11,6 +11,6 @@
#ifndef INCLUDE_LIBYUV_VERSION_H_ // NOLINT
#define INCLUDE_LIBYUV_VERSION_H_
#define LIBYUV_VERSION 643
#define LIBYUV_VERSION 644
#endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT
source/cpu_id.cc
View file @ 0cc0b4df
......@@ -172,7 +172,7 @@ int InitCpuFlags(void) {
((cpu_info7[1] & 0x00000200) ? kCpuHasERMS : 0) |
kCpuHasX86;
#ifdef HAS_XGETBV
if ((cpu_info1[2] & 0x18000000) == 0x18000000 && // AVX and OSSave
if ((cpu_info1[2] & 0x18000000) == 0x18000000 && // AVX and OSSave
(XGetBV(kXCR_XFEATURE_ENABLED_MASK) & 0x06) == 0x06) { // Saves YMM.
cpu_info_ |= ((cpu_info7[1] & 0x00000020) ? kCpuHasAVX2 : 0) |
kCpuHasAVX;
......
source/row_neon.cc
View file @ 0cc0b4df
......@@ -2487,32 +2487,41 @@ void ARGBColorMatrixRow_NEON(uint8* dst_argb, const int8* matrix_argb,
"1: \n"
"vld4.8 {d16, d18, d20, d22}, [%0] \n" // load 8 ARGB pixels.
"subs %1, %1, #8 \n" // 8 processed per loop.
"vmovl.u8 q2, d16 \n" // b (0 .. 255) 16 bit
"vmovl.u8 q3, d18 \n"
"vmovl.u8 q8, d20 \n"
"vmovl.u8 q9, d22 \n"
"vmul.s16 q12, q2, d0[0] \n" // B to Matrix B
"vmla.s16 q12, q3, d0[1] \n" // G
"vmla.s16 q12, q8, d0[2] \n" // R
"vmla.s16 q12, q9, d0[3] \n" // A
"vmul.s16 q13, q2, d1[0] \n" // B to Matrix G
"vmla.s16 q13, q3, d1[1] \n" // G
"vmla.s16 q13, q8, d1[2] \n" // R
"vmla.s16 q13, q9, d1[3] \n" // A
"vmul.s16 q14, q2, d2[0] \n" // B to Matrix R
"vmla.s16 q14, q3, d2[1] \n" // G
"vmla.s16 q14, q8, d2[2] \n" // R
"vmla.s16 q14, q9, d2[3] \n" // A
"vmovl.u8 q8, d16 \n" // b (0 .. 255) 16 bit
"vmovl.u8 q9, d18 \n" // g
"vmovl.u8 q10, d20 \n" // r
"vmovl.u8 q15, d22 \n" // a
"vmul.s16 q12, q8, d0[0] \n" // B = B * Matrix B
"vmul.s16 q13, q8, d1[0] \n" // G = B * Matrix G
"vmul.s16 q14, q8, d2[0] \n" // R = B * Matrix R
"vmul.s16 q4, q9, d0[1] \n" // B += G * Matrix B
"vmul.s16 q5, q9, d1[1] \n" // G += G * Matrix G
"vmul.s16 q6, q9, d2[1] \n" // R += G * Matrix R
"vqadd.s16 q12, q12, q4 \n" // Accumulate B
"vqadd.s16 q13, q13, q5 \n" // Accumulate G
"vqadd.s16 q14, q14, q6 \n" // Accumulate R
"vmul.s16 q4, q10, d0[2] \n" // B += R * Matrix B
"vmul.s16 q5, q10, d1[2] \n" // G += R * Matrix G
"vmul.s16 q6, q10, d2[2] \n" // R += R * Matrix R
"vqadd.s16 q12, q12, q4 \n" // Accumulate B
"vqadd.s16 q13, q13, q5 \n" // Accumulate G
"vqadd.s16 q14, q14, q6 \n" // Accumulate R
"vmul.s16 q4, q15, d0[3] \n" // B += A * Matrix B
"vmul.s16 q5, q15, d1[3] \n" // G += A * Matrix G
"vmul.s16 q6, q15, d2[3] \n" // R += A * Matrix R
"vqadd.s16 q12, q12, q4 \n" // Accumulate B
"vqadd.s16 q13, q13, q5 \n" // Accumulate G
"vqadd.s16 q14, q14, q6 \n" // Accumulate R
"vqshrun.s16 d16, q12, #7 \n" // 16 bit to 8 bit B
"vqshrun.s16 d18, q13, #7 \n" // 16 bit to 8 bit G
"vqshrun.s16 d20, q14, #7 \n" // 16 bit to 8 bit R
"vqshrun.s16 d20, q14, #7 \n" // 16 bit to 8 bit G
"vst4.8 {d16, d18, d20, d22}, [%0]! \n" // store 8 ARGB pixels.
"bgt 1b \n"
: "+r"(dst_argb), // %0
"+r"(width) // %1
: "r"(matrix_argb) // %2
: "cc", "memory", "q0", "q1", "q2", "q3", "q8", "q9",
"q10", "q11", "q12", "q13", "q14"
: "cc", "memory", "q0", "q1", "q2", "q3", "q4", "q5", "q6", "q8", "q9",
"q10", "q11", "q12", "q13", "q14", "q15"
);
}
......
source/scale.cc
View file @ 0cc0b4df
......@@ -2512,7 +2512,7 @@ static void ScaleFilterCols_C(uint8* dst_ptr, const uint8* src_ptr,
}
}
static const int kMaxInputWidth = 2560;
static const int kMaxInputWidth = 2880;
#if defined(HAS_SCALEFILTERROWS_SSE2)
// Filter row to 3/4
......
source/scale_argb.cc
View file @ 0cc0b4df
......@@ -1019,7 +1019,7 @@ static void ScaleARGBDownEven(int src_width, int src_height,
// interpolation.
// Maximum width handled by 2 pass Bilinear.
static const int kMaxInputWidth = 2560;
static const int kMaxInputWidth = 2880;
static void ScaleARGBBilinear(int src_width, int src_height,
int dst_width, int dst_height,
int src_stride, int dst_stride,
......
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