[FFmpeg-cvslog] swscale/arm: add ff_nv{12, 21}_to_{argb, rgba, abgr, bgra}_neon

Matthieu Bouron git at videolan.org
Mon Nov 23 14:49:56 CET 2015


ffmpeg | branch: master | Matthieu Bouron <matthieu.bouron at stupeflix.com> | Thu Nov  5 14:49:57 2015 +0100| [46feb66972bb9e872aebf15b9fe6ee9075e2615a] | committer: Clément Bœsch

swscale/arm: add ff_nv{12,21}_to_{argb,rgba,abgr,bgra}_neon

Signed-off-by: Matthieu Bouron <matthieu.bouron at stupeflix.com>
Signed-off-by: Clément Bœsch <clement at stupeflix.com>

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=46feb66972bb9e872aebf15b9fe6ee9075e2615a
---

 libswscale/arm/Makefile           |    3 +-
 libswscale/arm/swscale_unscaled.c |   66 +++++++++++++++
 libswscale/arm/yuv2rgb_neon.S     |  162 +++++++++++++++++++++++++++++++++++++
 libswscale/swscale_unscaled.c     |    4 +-
 4 files changed, 232 insertions(+), 3 deletions(-)

diff --git a/libswscale/arm/Makefile b/libswscale/arm/Makefile
index 8b5a97b..97b3561 100644
--- a/libswscale/arm/Makefile
+++ b/libswscale/arm/Makefile
@@ -1,4 +1,5 @@
-# OBJS        += arm/swscale_unscaled.o
+OBJS        += arm/swscale_unscaled.o
 
 # NEON-OBJS   += arm/rgb2yuv_neon_32.o
 # NEON-OBJS   += arm/rgb2yuv_neon_16.o
+NEON-OBJS   += arm/yuv2rgb_neon.o
diff --git a/libswscale/arm/swscale_unscaled.c b/libswscale/arm/swscale_unscaled.c
index 04be762..5d0c8e9 100644
--- a/libswscale/arm/swscale_unscaled.c
+++ b/libswscale/arm/swscale_unscaled.c
@@ -23,6 +23,7 @@
 #include "libswscale/swscale_internal.h"
 #include "libavutil/arm/cpu.h"
 
+#if 0
 extern void rgbx_to_nv12_neon_32(const uint8_t *src, uint8_t *y, uint8_t *chroma,
                 int width, int height,
                 int y_stride, int c_stride, int src_stride,
@@ -60,8 +61,69 @@ static int rgbx_to_nv12_neon_16_wrapper(SwsContext *context, const uint8_t *src[
 
     return 0;
 }
+#endif
+
+#define DECLARE_FF_NVX_TO_RGBX_FUNCS(ifmt, ofmt)                                            \
+int ff_##ifmt##_to_##ofmt##_neon(int w, int h,                                              \
+                                 uint8_t *dst, int linesize,                                \
+                                 const uint8_t *srcY, int linesizeY,                        \
+                                 const uint8_t *srcC, int linesizeC,                        \
+                                 const int16_t *table,                                      \
+                                 int y_offset,                                              \
+                                 int y_coeff);                                              \
+                                                                                            \
+static int ifmt##_to_##ofmt##_neon_wrapper(SwsContext *c, const uint8_t *src[],             \
+                                           int srcStride[], int srcSliceY, int srcSliceH,   \
+                                           uint8_t *dst[], int dstStride[]) {               \
+    const int16_t yuv2rgb_table[] = {                                                       \
+        c->yuv2rgb_v2r_coeff,                                                               \
+        c->yuv2rgb_u2g_coeff,                                                               \
+        c->yuv2rgb_v2g_coeff,                                                               \
+        c->yuv2rgb_u2b_coeff,                                                               \
+    };                                                                                      \
+                                                                                            \
+    ff_##ifmt##_to_##ofmt##_neon(c->srcW, srcSliceH,                                        \
+                                 dst[0] +  srcSliceY      * dstStride[0], dstStride[0],     \
+                                 src[0] +  srcSliceY      * srcStride[0], srcStride[0],     \
+                                 src[1] + (srcSliceY / 2) * srcStride[1], srcStride[1],     \
+                                 yuv2rgb_table,                                             \
+                                 c->yuv2rgb_y_offset >> 9,                                  \
+                                 c->yuv2rgb_y_coeff);                                       \
+                                                                                            \
+    return 0;                                                                               \
+}
+
+#define DECLARE_FF_NVX_TO_ALL_RGBX_FUNCS(nvx)                                               \
+DECLARE_FF_NVX_TO_RGBX_FUNCS(nvx, argb)                                                     \
+DECLARE_FF_NVX_TO_RGBX_FUNCS(nvx, rgba)                                                     \
+DECLARE_FF_NVX_TO_RGBX_FUNCS(nvx, abgr)                                                     \
+DECLARE_FF_NVX_TO_RGBX_FUNCS(nvx, bgra)                                                     \
+
+DECLARE_FF_NVX_TO_ALL_RGBX_FUNCS(nv12)
+DECLARE_FF_NVX_TO_ALL_RGBX_FUNCS(nv21)
+
+/* We need a 16 pixel width alignment. This constraint can easily be removed
+ * for input reading but for the output which is 4-bytes per pixel (RGBA) the
+ * assembly might be writing as much as 4*15=60 extra bytes at the end of the
+ * line, which won't fit the 32-bytes buffer alignment. */
+#define SET_FF_NVX_TO_RGBX_FUNC(ifmt, IFMT, ofmt, OFMT) do {                                \
+    if (c->srcFormat == AV_PIX_FMT_##IFMT                                                   \
+        && c->dstFormat == AV_PIX_FMT_##OFMT                                                \
+        && !(c->srcH & 1)                                                                   \
+        && !(c->srcW & 15)) {                                                               \
+        c->swscale = ifmt##_to_##ofmt##_neon_wrapper;                                       \
+    }                                                                                       \
+} while (0)
+
+#define SET_FF_NVX_TO_ALL_RGBX_FUNC(nvx, NVX) do {                                          \
+    SET_FF_NVX_TO_RGBX_FUNC(nvx, NVX, argb, ARGB);                                          \
+    SET_FF_NVX_TO_RGBX_FUNC(nvx, NVX, rgba, RGBA);                                          \
+    SET_FF_NVX_TO_RGBX_FUNC(nvx, NVX, abgr, ABGR);                                          \
+    SET_FF_NVX_TO_RGBX_FUNC(nvx, NVX, bgra, BGRA);                                          \
+} while (0)
 
 static void get_unscaled_swscale_neon(SwsContext *c) {
+#if 0
     int accurate_rnd = c->flags & SWS_ACCURATE_RND;
     if (c->srcFormat == AV_PIX_FMT_RGBA
             && c->dstFormat == AV_PIX_FMT_NV12
@@ -69,6 +131,10 @@ static void get_unscaled_swscale_neon(SwsContext *c) {
         c->swscale = accurate_rnd ? rgbx_to_nv12_neon_32_wrapper
                         : rgbx_to_nv12_neon_16_wrapper;
     }
+#endif
+
+    SET_FF_NVX_TO_ALL_RGBX_FUNC(nv12, NV12);
+    SET_FF_NVX_TO_ALL_RGBX_FUNC(nv21, NV21);
 }
 
 void ff_get_unscaled_swscale_arm(SwsContext *c)
diff --git a/libswscale/arm/yuv2rgb_neon.S b/libswscale/arm/yuv2rgb_neon.S
new file mode 100644
index 0000000..39b18cd
--- /dev/null
+++ b/libswscale/arm/yuv2rgb_neon.S
@@ -0,0 +1,162 @@
+/*
+ * Copyright (c) 2015 Matthieu Bouron <matthieu.bouron stupeflix.com>
+ * Copyright (c) 2015 Clément Bœsch <clement stupeflix.com>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "libavutil/arm/asm.S"
+
+.macro compute_premult half_u half_v
+    vmov                d2, \half_u                                    @ copy left q14 to left q1
+    vmov                d3, \half_u                                    @ copy left q14 to right q1
+    vmov                d4, \half_v                                    @ copy left q15 to left q2
+    vmov                d5, \half_v                                    @ copy left q15 to right q2
+
+    vzip.16             d2, d3                                         @ U1U1U2U2U3U3U4U4
+    vzip.16             d4, d5                                         @ V1V1V2V2V3V3V4V4
+
+    vmull.s16           q8,  d4, d1[0]                                 @  V * v2r             (left,  red)
+    vmull.s16           q9,  d5, d1[0]                                 @  V * v2r             (right, red)
+    vmull.s16           q10, d2, d1[1]                                 @  U * u2g
+    vmull.s16           q11, d3, d1[1]                                 @  U * u2g
+    vmlal.s16           q10, d4, d1[2]                                 @  U * u2g + V * v2g   (left,  green)
+    vmlal.s16           q11, d5, d1[2]                                 @  U * u2g + V * v2g   (right, green)
+    vmull.s16           q12, d2, d1[3]                                 @  U * u2b             (left,  blue)
+    vmull.s16           q13, d3, d1[3]                                 @  U * u2b             (right, blue)
+.endm
+
+.macro compute_color dst_comp pre1 pre2
+    vadd.s32            q3, q1, \pre1
+    vadd.s32            q4, q2, \pre2
+    vqrshrun.s32        d10, q3, #13
+    vqrshrun.s32        d11, q4, #13                                   @ q5 = ({q3,q4} + (1<<12)) >> 13
+    vqmovn.u16          \dst_comp, q5                                  @ saturate 16bit -> 8bit
+.endm
+
+.macro compute_rgba r g b a
+    compute_color       \r, q8,  q9
+    compute_color       \g, q10, q11
+    compute_color       \b, q12, q13
+    vmov.u8             \a, #255
+.endm
+
+.macro compute_half_line dst half_y ofmt
+    vmovl.u8            q7, \half_y                                    @ 8px of Y
+    vdup.16             q5, r9
+    vsub.s16            q7, q5
+    vmull.s16           q1, d14, d0                                    @ q1 = (srcY - y_offset) * y_coeff (left)
+    vmull.s16           q2, d15, d0                                    @ q2 = (srcY - y_offset) * y_coeff (right)
+
+.ifc \ofmt,argb
+    compute_rgba        d13, d14, d15, d12
+.endif
+
+.ifc \ofmt,rgba
+    compute_rgba        d12, d13, d14, d15
+.endif
+
+.ifc \ofmt,abgr
+    compute_rgba        d15, d14, d13, d12
+.endif
+
+.ifc \ofmt,bgra
+    compute_rgba        d14, d13, d12, d15
+.endif
+
+    vst4.8              {q6, q7}, [\dst,:128]!
+.endm
+
+.macro declare_func ifmt ofmt
+function ff_\ifmt\()_to_\ofmt\()_neon, export=1
+    push                {r4-r12, lr}
+    vpush               {q4-q7}
+    ldr                 r4, [sp, #104]                                 @ r4  = srcY
+    ldr                 r5, [sp, #108]                                 @ r5  = linesizeY
+    ldr                 r6, [sp, #112]                                 @ r6  = srcC
+    ldr                 r7, [sp, #116]                                 @ r7  = linesizeC
+    ldr                 r8, [sp, #120]                                 @ r8  = table
+    ldr                 r9, [sp, #124]                                 @ r9  = y_offset
+    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
+    lsl                 r8, r0, #2
+    sub                 r3, r3, r8                                     @ r3 = linesize  * 2 - width * 4 (padding)
+    sub                 r5, r5, r0                                     @ r5 = linesizeY * 2 - width     (paddingY)
+    sub                 r7, r7, r0                                     @ r7 = linesizeC     - width     (paddingC)
+1:
+    mov                 r8, r0                                         @ r8 = width
+2:
+    pld [r6, #64*3]
+    pld [r4, #64*3]
+    pld [r12, #64*3]
+
+    vld2.8              {d2, d3}, [r6]!                                @ q1: interleaved chroma line
+    vmov.i8             d10, #128
+.ifc \ifmt,nv12
+    vsubl.u8            q14, d2, d10                                   @ q14 = U - 128
+    vsubl.u8            q15, d3, d10                                   @ q15 = V - 128
+.else
+    vsubl.u8            q14, d3, d10                                   @ q14 = U - 128
+    vsubl.u8            q15, d2, d10                                   @ q15 = V - 128
+.endif
+
+    compute_premult     d28, d30
+
+    vld1.8              {q7}, [r4]!                                    @ first line of luma
+    vmov                d28, d15                                       @ save right of the first line of luma for later use
+    compute_half_line   r2, d14, \ofmt
+
+    vld1.8              {q7}, [r12]!                                   @ second line of luma
+    vmov                d30, d15                                       @ save right of the second line of luma for later use
+    compute_half_line   r11, d14, \ofmt
+
+    compute_premult     d29, d31
+    compute_half_line   r2,  d28, \ofmt
+    compute_half_line   r11, d30, \ofmt
+
+    subs                r8, r8, #16                                    @ width -= 16
+    bgt                 2b
+
+    add                 r2, r2, r3                                     @ dst   += padding
+    add                 r4, r4, r5                                     @ srcY  += paddingY
+    add                 r11, r11, r3                                   @ dst2  += padding
+    add                 r12, r12, r5                                   @ srcY2 += paddingY
+    add                 r6, r6, r7                                     @ srcC  += paddingC
+
+    subs                r1, r1, #2                                     @ height -= 2
+    bgt                 1b
+
+    vpop                {q4-q7}
+    pop                 {r4-r12, lr}
+    mov                 pc, lr
+endfunc
+.endm
+
+.macro declare_rgb_funcs ifmt
+    declare_func \ifmt, argb
+    declare_func \ifmt, rgba
+    declare_func \ifmt, abgr
+    declare_func \ifmt, bgra
+.endm
+
+declare_rgb_funcs nv12
+declare_rgb_funcs nv21
diff --git a/libswscale/swscale_unscaled.c b/libswscale/swscale_unscaled.c
index f387f52..89cb90d 100644
--- a/libswscale/swscale_unscaled.c
+++ b/libswscale/swscale_unscaled.c
@@ -1762,8 +1762,8 @@ void ff_get_unscaled_swscale(SwsContext *c)
 
     if (ARCH_PPC)
         ff_get_unscaled_swscale_ppc(c);
-//     if (ARCH_ARM)
-//         ff_get_unscaled_swscale_arm(c);
+     if (ARCH_ARM)
+         ff_get_unscaled_swscale_arm(c);
 }
 
 /* Convert the palette to the same packed 32-bit format as the palette */



More information about the ffmpeg-cvslog mailing list