Skip to content

L
libyuv
Commit c8a34d2e authored by ashok.bhat@gmail.com's avatar ashok.bhat@gmail.com
Browse files
Options

Row AArch64 Neon implementation - Part 9 

BUG=319
TESTED=libyuv_unittest
R=fbarchard@google.com

Change-Id: Id3af83a6efbd70b4a808a8442c3badbef749c0cc
Signed-off-by: 's avatarAshok Bhat <ashok.bhat@arm.com>

Review URL: https://webrtc-codereview.appspot.com/23769004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1092 16f28f9a-4ce2-e073-06de-1de4eb20be90
parent 093fd484
Hide whitespace changes
Inline Side-by-side
Showing with 33 additions and 43 deletions
include/libyuv/row.h
View file @ c8a34d2e
...@@ -305,9 +305,9 @@ extern "C" { ...@@ -305,9 +305,9 @@ extern "C" {
#define HAS_ARGBSHUFFLEROW_NEON #define HAS_ARGBSHUFFLEROW_NEON
#define HAS_I422TOYUY2ROW_NEON #define HAS_I422TOYUY2ROW_NEON
#define HAS_I422TOUYVYROW_NEON #define HAS_I422TOUYVYROW_NEON
// #define HAS_ARGBTORGB565ROW_NEON #define HAS_ARGBTORGB565ROW_NEON
// #define HAS_ARGBTOARGB1555ROW_NEON #define HAS_ARGBTOARGB1555ROW_NEON
// #define HAS_ARGBTOARGB4444ROW_NEON #define HAS_ARGBTOARGB4444ROW_NEON
#define HAS_ARGBTOYROW_NEON #define HAS_ARGBTOYROW_NEON
#define HAS_ARGBTOYJROW_NEON #define HAS_ARGBTOYJROW_NEON
#define HAS_ARGBTOUV444ROW_NEON #define HAS_ARGBTOUV444ROW_NEON
......
source/row_neon64.cc
View file @ c8a34d2e
...@@ -414,16 +414,11 @@ void I422ToRAWRow_NEON(const uint8* src_y, ...@@ -414,16 +414,11 @@ void I422ToRAWRow_NEON(const uint8* src_y,
#endif // HAS_I422TORAWROW_NEON #endif // HAS_I422TORAWROW_NEON
#define ARGBTORGB565 \ #define ARGBTORGB565 \
"vshr.u8 d20, d20, #3 \n" /* B */ \ "shll v0.8h, v22.8b, #8 \n" /* R */ \
"vshr.u8 d21, d21, #2 \n" /* G */ \ "shll v20.8h, v20.8b, #8 \n" /* B */ \
"vshr.u8 d22, d22, #3 \n" /* R */ \ "shll v21.8h, v21.8b, #8 \n" /* G */ \
"vmovl.u8 q8, d20 \n" /* B */ \ "sri v0.8h, v21.8h, #5 \n" /* RG */ \
"vmovl.u8 q9, d21 \n" /* G */ \ "sri v0.8h, v20.8h, #11 \n" /* RGB */
"vmovl.u8 q10, d22 \n" /* R */ \
"vshl.u16 q9, q9, #5 \n" /* G */ \
"vshl.u16 q10, q10, #11 \n" /* R */ \
"vorr q0, q8, q9 \n" /* BG */ \
"vorr q0, q0, q10 \n" /* BGR */
#ifdef HAS_I422TORGB565ROW_NEON #ifdef HAS_I422TORGB565ROW_NEON
void I422ToRGB565Row_NEON(const uint8* src_y, void I422ToRGB565Row_NEON(const uint8* src_y,
...@@ -462,19 +457,13 @@ void I422ToRGB565Row_NEON(const uint8* src_y, ...@@ -462,19 +457,13 @@ void I422ToRGB565Row_NEON(const uint8* src_y,
#endif // HAS_I422TORGB565ROW_NEON #endif // HAS_I422TORGB565ROW_NEON
#define ARGBTOARGB1555 \ #define ARGBTOARGB1555 \
"vshr.u8 q10, q10, #3 \n" /* B */ \ "shll v0.8h, v23.8b, #8 \n" /* A */ \
"vshr.u8 d22, d22, #3 \n" /* R */ \ "shll v22.8h, v22.8b, #8 \n" /* R */ \
"vshr.u8 d23, d23, #7 \n" /* A */ \ "shll v20.8h, v20.8b, #8 \n" /* B */ \
"vmovl.u8 q8, d20 \n" /* B */ \ "shll v21.8h, v21.8b, #8 \n" /* G */ \
"vmovl.u8 q9, d21 \n" /* G */ \ "sri v0.8h, v22.8h, #1 \n" /* AR */ \
"vmovl.u8 q10, d22 \n" /* R */ \ "sri v0.8h, v21.8h, #6 \n" /* ARG */ \
"vmovl.u8 q11, d23 \n" /* A */ \ "sri v0.8h, v20.8h, #11 \n" /* ARGB */
"vshl.u16 q9, q9, #5 \n" /* G */ \
"vshl.u16 q10, q10, #10 \n" /* R */ \
"vshl.u16 q11, q11, #15 \n" /* A */ \
"vorr q0, q8, q9 \n" /* BG */ \
"vorr q1, q10, q11 \n" /* RA */ \
"vorr q0, q0, q1 \n" /* BGRA */
#ifdef HAS_I422TOARGB1555ROW_NEON #ifdef HAS_I422TOARGB1555ROW_NEON
void I422ToARGB1555Row_NEON(const uint8* src_y, void I422ToARGB1555Row_NEON(const uint8* src_y,
...@@ -514,13 +503,14 @@ void I422ToARGB1555Row_NEON(const uint8* src_y, ...@@ -514,13 +503,14 @@ void I422ToARGB1555Row_NEON(const uint8* src_y,
#endif // HAS_I422TOARGB1555ROW_NEON #endif // HAS_I422TOARGB1555ROW_NEON
#define ARGBTOARGB4444 \ #define ARGBTOARGB4444 \
"vshr.u8 d20, d20, #4 \n" /* B */ \ /* Input v20.8b<=B, v21.8b<=G, v22.8b<=R, v23.8b<=A, v4.8b<=0x0f */ \
"vbic.32 d21, d21, d4 \n" /* G */ \ "ushr v20.8b, v20.8b, #4 \n" /* B */ \
"vshr.u8 d22, d22, #4 \n" /* R */ \ "bic v21.8b, v21.8b, v4.8b \n" /* G */ \
"vbic.32 d23, d23, d4 \n" /* A */ \ "ushr v22.8b, v22.8b, #4 \n" /* R */ \
"vorr d0, d20, d21 \n" /* BG */ \ "bic v23.8b, v23.8b, v4.8b \n" /* A */ \
"vorr d1, d22, d23 \n" /* RA */ \ "orr v0.8b, v20.8b, v21.8b \n" /* BG */ \
"vzip.u8 d0, d1 \n" /* BGRA */ "orr v1.8b, v22.8b, v23.8b \n" /* RA */ \
"zip1 v0.16b, v0.16b, v1.16b \n" /* BGRA */
#ifdef HAS_I422TOARGB4444ROW_NEON #ifdef HAS_I422TOARGB4444ROW_NEON
void I422ToARGB4444Row_NEON(const uint8* src_y, void I422ToARGB4444Row_NEON(const uint8* src_y,
...@@ -1516,17 +1506,17 @@ void ARGBToRGB565Row_NEON(const uint8* src_argb, uint8* dst_rgb565, int pix) { ...@@ -1516,17 +1506,17 @@ void ARGBToRGB565Row_NEON(const uint8* src_argb, uint8* dst_rgb565, int pix) {
".p2align 2 \n" ".p2align 2 \n"
"1: \n" "1: \n"
MEMACCESS(0) MEMACCESS(0)
"vld4.8 {d20, d21, d22, d23}, [%0]! \n" // load 8 pixels of ARGB. "ld4 {v20.8b-v23.8b}, [%0], #32 \n" // load 8 pixels of ARGB.
"subs %2, %2, #8 \n" // 8 processed per loop. "subs %2, %2, #8 \n" // 8 processed per loop.
ARGBTORGB565 ARGBTORGB565
MEMACCESS(1) MEMACCESS(1)
"vst1.8 {q0}, [%1]! \n" // store 8 pixels RGB565. "st1 {v0.16b}, [%1], #16 \n" // store 8 pixels RGB565.
"b.gt 1b \n" "b.gt 1b \n"
: "+r"(src_argb), // %0 : "+r"(src_argb), // %0
"+r"(dst_rgb565), // %1 "+r"(dst_rgb565), // %1
"+r"(pix) // %2 "+r"(pix) // %2
: :
: "cc", "memory", "q0", "q8", "q9", "q10", "q11" : "cc", "memory", "v0", "v20", "v21", "v22", "v23"
); );
} }
#endif // HAS_ARGBTORGB565ROW_NEON #endif // HAS_ARGBTORGB565ROW_NEON
...@@ -1538,17 +1528,17 @@ void ARGBToARGB1555Row_NEON(const uint8* src_argb, uint8* dst_argb1555, ...@@ -1538,17 +1528,17 @@ void ARGBToARGB1555Row_NEON(const uint8* src_argb, uint8* dst_argb1555,
".p2align 2 \n" ".p2align 2 \n"
"1: \n" "1: \n"
MEMACCESS(0) MEMACCESS(0)
"vld4.8 {d20, d21, d22, d23}, [%0]! \n" // load 8 pixels of ARGB. "ld4 {v20.8b-v23.8b}, [%0], #32 \n" // load 8 pixels of ARGB.
"subs %2, %2, #8 \n" // 8 processed per loop. "subs %2, %2, #8 \n" // 8 processed per loop.
ARGBTOARGB1555 ARGBTOARGB1555
MEMACCESS(1) MEMACCESS(1)
"vst1.8 {q0}, [%1]! \n" // store 8 pixels ARGB1555. "st1 {v0.16b}, [%1], #16 \n" // store 8 pixels ARGB1555.
"b.gt 1b \n" "b.gt 1b \n"
: "+r"(src_argb), // %0 : "+r"(src_argb), // %0
"+r"(dst_argb1555), // %1 "+r"(dst_argb1555), // %1
"+r"(pix) // %2 "+r"(pix) // %2
: :
: "cc", "memory", "q0", "q8", "q9", "q10", "q11" : "cc", "memory", "v0", "v20", "v21", "v22", "v23"
); );
} }
#endif // HAS_ARGBTOARGB1555ROW_NEON #endif // HAS_ARGBTOARGB1555ROW_NEON
...@@ -1557,21 +1547,21 @@ void ARGBToARGB1555Row_NEON(const uint8* src_argb, uint8* dst_argb1555, ...@@ -1557,21 +1547,21 @@ void ARGBToARGB1555Row_NEON(const uint8* src_argb, uint8* dst_argb1555,
void ARGBToARGB4444Row_NEON(const uint8* src_argb, uint8* dst_argb4444, void ARGBToARGB4444Row_NEON(const uint8* src_argb, uint8* dst_argb4444,
int pix) { int pix) {
asm volatile ( asm volatile (
"vmov.u8 d4, #0x0f \n" // bits to clear with vbic. "movi v4.16b, #0x0f \n" // bits to clear with vbic.
".p2align 2 \n" ".p2align 2 \n"
"1: \n" "1: \n"
MEMACCESS(0) MEMACCESS(0)
"vld4.8 {d20, d21, d22, d23}, [%0]! \n" // load 8 pixels of ARGB. "ld4 {v20.8b-v23.8b}, [%0], #32 \n" // load 8 pixels of ARGB.
"subs %2, %2, #8 \n" // 8 processed per loop. "subs %2, %2, #8 \n" // 8 processed per loop.
ARGBTOARGB4444 ARGBTOARGB4444
MEMACCESS(1) MEMACCESS(1)
"vst1.8 {q0}, [%1]! \n" // store 8 pixels ARGB4444. "st1 {v0.16b}, [%1], #16 \n" // store 8 pixels ARGB4444.
"b.gt 1b \n" "b.gt 1b \n"
: "+r"(src_argb), // %0 : "+r"(src_argb), // %0
"+r"(dst_argb4444), // %1 "+r"(dst_argb4444), // %1
"+r"(pix) // %2 "+r"(pix) // %2
: :
: "cc", "memory", "q0", "q8", "q9", "q10", "q11" : "cc", "memory", "v0", "v1", "v4", "v20", "v21", "v22", "v23"
); );
} }
#endif // HAS_ARGBTOARGB4444ROW_NEON #endif // HAS_ARGBTOARGB4444ROW_NEON
......
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