[FFmpeg-devel] [PATCH 06/10] swscale/arm/yuv2rgb: only process one line at a time for the yuv420p and nv{12, 21} formats
Matthieu Bouron
matthieu.bouron at gmail.com
Fri Mar 25 23:46:01 CET 2016
From: Matthieu Bouron <matthieu.bouron at stupeflix.com>
---
libswscale/arm/yuv2rgb_neon.S | 89 ++++++++++++-------------------------------
1 file changed, 24 insertions(+), 65 deletions(-)
diff --git a/libswscale/arm/yuv2rgb_neon.S b/libswscale/arm/yuv2rgb_neon.S
index ef7b0a6..8abb986 100644
--- a/libswscale/arm/yuv2rgb_neon.S
+++ b/libswscale/arm/yuv2rgb_neon.S
@@ -105,16 +105,6 @@
compute_16px r2, d14, d15, \ofmt
.endm
-.macro process_2l_16px ofmt
- compute_premult d28, d29, d30, d31
-
- vld1.8 {q7}, [r4]! @ first line of luma
- compute_16px r2, d14, d15, \ofmt
-
- vld1.8 {q7}, [r12]! @ second line of luma
- compute_16px r11, d14, d15, \ofmt
-.endm
-
.macro load_args_nvx
push {r4-r12, lr}
vpush {q4-q7}
@@ -127,13 +117,9 @@
ldr r10,[sp, #128] @ r10 = y_coeff
vdup.16 d0, r10 @ d0 = y_coeff
vld1.16 {d1}, [r8] @ d1 = *table
- add r11, r2, r3 @ r11 = dst + linesize (dst2)
- add r12, r4, r5 @ r12 = srcY + linesizeY (srcY2)
- lsl r3, r3, #1
- lsl r5, r5, #1
- sub r3, r3, r0, lsl #2 @ r3 = linesize * 2 - width * 4 (padding)
- sub r5, r5, r0 @ r5 = linesizeY * 2 - width (paddingY)
- sub r7, r7, r0 @ r7 = linesizeC - width (paddingC)
+ sub r3, r3, r0, lsl #2 @ r3 = linesize - width * 4 (padding)
+ sub r5, r5, r0 @ r5 = linesizeY - width (paddingY)
+ sub r7, r7, r0 @ r7 = linesizeC - width (paddingC)
.endm
.macro load_args_yuv420p
@@ -142,26 +128,6 @@
ldr r4, [sp, #104] @ r4 = srcY
ldr r5, [sp, #108] @ r5 = linesizeY
ldr r6, [sp, #112] @ r6 = srcU
- ldr r8, [sp, #128] @ r8 = table
- ldr r9, [sp, #132] @ r9 = y_offset
- ldr r10,[sp, #136] @ r10 = y_coeff
- vdup.16 d0, r10 @ d0 = y_coeff
- vld1.16 {d1}, [r8] @ d1 = *table
- add r11, r2, r3 @ r11 = dst + linesize (dst2)
- add r12, r4, r5 @ r12 = srcY + linesizeY (srcY2)
- lsl r3, r3, #1
- lsl r5, r5, #1
- sub r3, r3, r0, lsl #2 @ r3 = linesize * 2 - width * 4 (padding)
- sub r5, r5, r0 @ r5 = linesizeY * 2 - width (paddingY)
- ldr r10,[sp, #120] @ r10 = srcV
-.endm
-
-.macro load_args_yuv422p
- push {r4-r12, lr}
- vpush {q4-q7}
- ldr r4, [sp, #104] @ r4 = srcY
- ldr r5, [sp, #108] @ r5 = linesizeY
- ldr r6, [sp, #112] @ r6 = srcU
ldr r7, [sp, #116] @ r7 = linesizeU
ldr r12,[sp, #124] @ r12 = linesizeV
ldr r8, [sp, #128] @ r8 = table
@@ -176,6 +142,10 @@
ldr r10,[sp, #120] @ r10 = srcV
.endm
+.macro load_args_yuv422p
+ load_args_yuv420p
+.endm
+
.macro declare_func ifmt ofmt
function ff_\ifmt\()_to_\ofmt\()_neon, export=1
@@ -205,35 +175,30 @@ function ff_\ifmt\()_to_\ofmt\()_neon, export=1
vmov.i8 d10, #128
.ifc \ifmt,nv12
- pld [r12, #64*3]
-
vld2.8 {d2, d3}, [r6]! @ q1: interleaved chroma line
vsubl.u8 q14, d2, d10 @ q14 = U - 128
vsubl.u8 q15, d3, d10 @ q15 = V - 128
- process_2l_16px \ofmt
+ process_1l_16px \ofmt
.endif
.ifc \ifmt,nv21
- pld [r12, #64*3]
-
vld2.8 {d2, d3}, [r6]! @ q1: interleaved chroma line
vsubl.u8 q14, d3, d10 @ q14 = U - 128
vsubl.u8 q15, d2, d10 @ q15 = V - 128
- process_2l_16px \ofmt
+ process_1l_16px \ofmt
.endif
.ifc \ifmt,yuv420p
pld [r10, #64*3]
- pld [r12, #64*3]
vld1.8 d2, [r6]! @ d2: chroma red line
vld1.8 d3, [r10]! @ d3: chroma blue line
vsubl.u8 q14, d2, d10 @ q14 = U - 128
vsubl.u8 q15, d3, d10 @ q15 = V - 128
- process_2l_16px \ofmt
+ process_1l_16px \ofmt
.endif
.ifc \ifmt,yuv422p
@@ -254,35 +219,29 @@ function ff_\ifmt\()_to_\ofmt\()_neon, export=1
add r4, r4, r5 @ srcY += paddingY
.ifc \ifmt,nv12
- add r11, r11, r3 @ dst2 += padding
- add r12, r12, r5 @ srcY2 += paddingY
-
- add r6, r6, r7 @ srcC += paddingC
+ tst r1, #1
+ subeq r6, r6, r0 @ if (height % 2 == 0) paddingU -= width
+ addne r6, r7 @ else paddingU += linesizeU - width
- subs r1, r1, #2 @ height -= 2
+ subs r1, r1, #1 @ height -= 1
.endif
.ifc \ifmt,nv21
- add r11, r11, r3 @ dst2 += padding
- add r12, r12, r5 @ srcY2 += paddingY
+ tst r1, #1
+ subeq r6, r6, r0 @ if (height % 2 == 0) paddingU -= width
+ addne r6, r7 @ else paddingU += linesizeU - width
- add r6, r6, r7 @ srcC += paddingC
- subs r1, r1, #2 @ height -= 2
+ subs r1, r1, #1 @ height -= 1
.endif
.ifc \ifmt,yuv420p
- add r11, r11, r3 @ dst2 += padding
- add r12, r12, r5 @ srcY2 += paddingY
+ tst r1, #1
+ subeq r6, r6, r0, lsr #1 @ if (height % 2 == 0) paddingU -= (width / 2)
+ addne r6, r7 @ else paddingU += linesizeU - (width / 2)
+ subeq r10, r10, r0, lsr #1 @ if (height % 2 == 0) paddingU -= (width / 2)
+ addne r10, r12 @ else paddingV = linesizeV - (width / 2)
- ldr r7, [sp, #116] @ r7 = linesizeU
- sub r7, r7, r0, lsr #1 @ r7 = linesizeU - width / 2 (paddingU)
- add r6, r6, r7 @ srcU += paddingU
-
- ldr r7, [sp, #124] @ r7 = linesizeV
- sub r7, r7, r0, lsr #1 @ r7 = linesizeV - width / 2 (paddingV)
- add r10, r10, r7 @ srcV += paddingV
-
- subs r1, r1, #2 @ height -= 2
+ subs r1, r1, #1 @ height -= 1
.endif
.ifc \ifmt,yuv422p
--
2.7.4
More information about the ffmpeg-devel
mailing list