On Fri, Mar 25, 2016 at 11:46 PM, Matthieu Bouron <matthieu.bou...@gmail.com
> wrote:
> From: Matthieu Bouron <matthieu.bou...@stupeflix.com>
>
> ---
> libswscale/arm/yuv2rgb_neon.S | 154
> +++++++++++++++++++-----------------------
> 1 file changed, 69 insertions(+), 85 deletions(-)
>
Patch updated (resolve a conflict with the updated version of patch 06/10).
From d06a5437f9042e0b350556e9642d52866284e7a8 Mon Sep 17 00:00:00 2001
From: Matthieu Bouron <matthieu.bou...@stupeflix.com>
Date: Wed, 23 Mar 2016 14:10:45 +0000
Subject: [PATCH 08/10] swscale/arm/yuv2rgb: re-organize the code like its
aarch64 counter part
---
libswscale/arm/yuv2rgb_neon.S | 154 +++++++++++++++++++-----------------------
1 file changed, 69 insertions(+), 85 deletions(-)
diff --git a/libswscale/arm/yuv2rgb_neon.S b/libswscale/arm/yuv2rgb_neon.S
index 6279637..6a15778 100644
--- a/libswscale/arm/yuv2rgb_neon.S
+++ b/libswscale/arm/yuv2rgb_neon.S
@@ -21,90 +21,6 @@
#include "libavutil/arm/asm.S"
-
-.macro compute_premult half_u1, half_u2, half_v1, half_v2
- vmov d2, \half_u1 @ copy left q14 to left q1
- vmov d3, \half_u1 @ copy left q14 to right q1
- vmov d4, \half_u2 @ copy right q14 to left q2
- vmov d5, \half_u2 @ copy right q14 to right q2
-
- vmov d6, \half_v1 @ copy left q15 to left q3
- vmov d7, \half_v1 @ copy left q15 to right q3
- vmov d8, \half_v2 @ copy right q15 to left q4
- vmov d9, \half_v2 @ copy right q15 to right q4
-
- vzip.16 d2, d3 @ U1U1U2U2U3U3U4U4
- vzip.16 d4, d5 @ U5U5U6U6U7U7U8U8
-
- vzip.16 d6, d7 @ V1V1V2V2V3V3V4V4
- vzip.16 d8, d9 @ V5V5V6V6V7V7V8V8
-
- vmul.s16 q8, q3, d1[0] @ V * v2r (left, red)
- vmul.s16 q9, q4, d1[0] @ V * v2r (right, red)
- vmul.s16 q10, q1, d1[1] @ U * u2g
- vmul.s16 q11, q2, d1[1] @ U * u2g
- vmla.s16 q10, q3, d1[2] @ U * u2g + V * v2g (left, green)
- vmla.s16 q11, q4, d1[2] @ U * u2g + V * v2g (right, green)
- vmul.s16 q12, q1, d1[3] @ U * u2b (left, blue)
- vmul.s16 q13, q2, d1[3] @ U * u2b (right, blue)
-.endm
-
-.macro compute_color dst_comp1 dst_comp2 pre1 pre2
- vadd.s16 q1, q14, \pre1
- vadd.s16 q2, q15, \pre2
- vqrshrun.s16 \dst_comp1, q1, #6
- vqrshrun.s16 \dst_comp2, q2, #6
-.endm
-
-.macro compute_rgba r1 r2 g1 g2 b1 b2 a1 a2
- compute_color \r1, \r2, q8, q9
- compute_color \g1, \g2, q10, q11
- compute_color \b1, \b2, q12, q13
- vmov.u8 \a1, #255
- vmov.u8 \a2, #255
-.endm
-
-.macro compute_16px dst y0 y1 ofmt
- vmovl.u8 q14, \y0 @ 8px of y
- vmovl.u8 q15, \y1 @ 8px of y
-
- vdup.16 q5, r9 @ q5 = y_offset
- vmov d14, d0 @ q7 = y_coeff
- vmov d15, d0 @ q7 = y_coeff
-
- vsub.s16 q14, q5
- vsub.s16 q15, q5
-
- vmul.s16 q14, q7 @ q14 = (srcY - y_offset) * y_coeff (left)
- vmul.s16 q15, q7 @ q15 = (srcY - y_offset) * y_coeff (right)
-
-
-.ifc \ofmt,argb
- compute_rgba d7, d11, d8, d12, d9, d13, d6, d10
-.endif
-
-.ifc \ofmt,rgba
- compute_rgba d6, d10, d7, d11, d8, d12, d9, d13
-.endif
-
-.ifc \ofmt,abgr
- compute_rgba d9, d13, d8, d12, d7, d11, d6, d10
-.endif
-
-.ifc \ofmt,bgra
- compute_rgba d8, d12, d7, d11, d6, d10, d9, d13
-.endif
- vst4.8 {q3, q4}, [\dst,:128]!
- vst4.8 {q5, q6}, [\dst,:128]!
-
-.endm
-
-.macro process_1l_16px ofmt
- compute_premult d28, d29, d30, d31
- vld1.8 {q7}, [r4]!
- compute_16px r2, d14, d15, \ofmt
-.endm
-
.macro load_args_nv12
push {r4-r12, lr}
vpush {q4-q7}
@@ -200,6 +116,21 @@
add r10,r10,r12 @ srcV += paddingV
.endm
+.macro compute_color dst_comp1 dst_comp2 pre1 pre2
+ vadd.s16 q1, q14, \pre1
+ vadd.s16 q2, q15, \pre2
+ vqrshrun.s16 \dst_comp1, q1, #6
+ vqrshrun.s16 \dst_comp2, q2, #6
+.endm
+
+.macro compute_rgba r1 r2 g1 g2 b1 b2 a1 a2
+ compute_color \r1, \r2, q8, q9
+ compute_color \g1, \g2, q10, q11
+ compute_color \b1, \b2, q12, q13
+ vmov.u8 \a1, #255
+ vmov.u8 \a2, #255
+.endm
+
.macro declare_func ifmt ofmt
function ff_\ifmt\()_to_\ofmt\()_neon, export=1
load_args_\ifmt
@@ -210,7 +141,60 @@ function ff_\ifmt\()_to_\ofmt\()_neon, export=1
pld [r4, #64*3]
vmov.i8 d10, #128
load_chroma_\ifmt
- process_1l_16px \ofmt
+ vmov d2, d28 @ copy left q14 to left q1
+ vmov d3, d28 @ copy left q14 to right q1
+ vmov d4, d29 @ copy right q14 to left q2
+ vmov d5, d29 @ copy right q14 to right q2
+
+ vmov d6, d30 @ copy left q15 to left q3
+ vmov d7, d30 @ copy left q15 to right q3
+ vmov d8, d31 @ copy right q15 to left q4
+ vmov d9, d31 @ copy right q15 to right q4
+
+ vzip.16 d2, d3 @ U1U1U2U2U3U3U4U4
+ vzip.16 d4, d5 @ U5U5U6U6U7U7U8U8
+
+ vzip.16 d6, d7 @ V1V1V2V2V3V3V4V4
+ vzip.16 d8, d9 @ V5V5V6V6V7V7V8V8
+
+ vmul.s16 q8, q3, d1[0] @ V * v2r (left, red)
+ vmul.s16 q9, q4, d1[0] @ V * v2r (right, red)
+ vmul.s16 q10, q1, d1[1] @ U * u2g
+ vmul.s16 q11, q2, d1[1] @ U * u2g
+ vmla.s16 q10, q3, d1[2] @ U * u2g + V * v2g (left, green)
+ vmla.s16 q11, q4, d1[2] @ U * u2g + V * v2g (right, green)
+ vmul.s16 q12, q1, d1[3] @ U * u2b (left, blue)
+ vmul.s16 q13, q2, d1[3] @ U * u2b (right, blue)
+
+ vld1.8 {q7}, [r4]!
+ vmovl.u8 q14, d14 @ 8px of y
+ vmovl.u8 q15, d15 @ 8px of y
+ vdup.16 q5, r9 @ q5 = y_offset
+ vmov d14, d0 @ q7 = y_coeff
+ vmov d15, d0 @ q7 = y_coeff
+ vsub.s16 q14, q5
+ vsub.s16 q15, q5
+ vmul.s16 q14, q7 @ q14 = (srcY - y_offset) * y_coeff (left)
+ vmul.s16 q15, q7 @ q15 = (srcY - y_offset) * y_coeff (right)
+
+.ifc \ofmt,argb
+ compute_rgba d7, d11, d8, d12, d9, d13, d6, d10
+.endif
+
+.ifc \ofmt,rgba
+ compute_rgba d6, d10, d7, d11, d8, d12, d9, d13
+.endif
+
+.ifc \ofmt,abgr
+ compute_rgba d9, d13, d8, d12, d7, d11, d6, d10
+.endif
+
+.ifc \ofmt,bgra
+ compute_rgba d8, d12, d7, d11, d6, d10, d9, d13
+.endif
+
+ vst4.8 {q3, q4}, [r2,:128]!
+ vst4.8 {q5, q6}, [r2,:128]!
subs r8, r8, #16 @ width -= 16
bgt 2b
add r2, r2, r3 @ dst += padding
--
2.7.4
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
