[FFmpeg-cvslog] ARM: NEON fixed-point FFT

Mans Rullgard git at videolan.org
Mon Apr 4 03:01:17 CEST 2011


ffmpeg | branch: master | Mans Rullgard <mans at mansr.com> | Sun Mar 27 10:26:58 2011 +0100| [dba9852935f31d81d16bf26ee21d164d5d5f5764] | committer: Mans Rullgard

ARM: NEON fixed-point FFT

Signed-off-by: Mans Rullgard <mans at mansr.com>

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=dba9852935f31d81d16bf26ee21d164d5d5f5764
---

 libavcodec/arm/Makefile             |    2 +
 libavcodec/arm/fft_fixed_init_arm.c |   32 +++++
 libavcodec/arm/fft_fixed_neon.S     |  261 +++++++++++++++++++++++++++++++++++
 libavcodec/fft.c                    |    1 +
 libavcodec/fft.h                    |    4 +
 5 files changed, 300 insertions(+), 0 deletions(-)

diff --git a/libavcodec/arm/Makefile b/libavcodec/arm/Makefile
index 3b77a55..2876e8f 100644
--- a/libavcodec/arm/Makefile
+++ b/libavcodec/arm/Makefile
@@ -16,6 +16,7 @@ OBJS-$(CONFIG_H264PRED)                += arm/h264pred_init_arm.o
 OBJS                                   += arm/dsputil_init_arm.o        \
                                           arm/dsputil_arm.o             \
                                           arm/fft_init_arm.o            \
+                                          arm/fft_fixed_init_arm.o      \
                                           arm/fmtconvert_init_arm.o     \
                                           arm/jrevdct_arm.o             \
                                           arm/mpegvideo_arm.o           \
@@ -41,6 +42,7 @@ OBJS-$(HAVE_IWMMXT)                    += arm/dsputil_iwmmxt.o          \
                                           arm/mpegvideo_iwmmxt.o        \
 
 NEON-OBJS-$(CONFIG_FFT)                += arm/fft_neon.o                \
+                                          arm/fft_fixed_neon.o          \
 
 NEON-OBJS-$(CONFIG_MDCT)               += arm/mdct_neon.o               \
 
diff --git a/libavcodec/arm/fft_fixed_init_arm.c b/libavcodec/arm/fft_fixed_init_arm.c
new file mode 100644
index 0000000..916bf29
--- /dev/null
+++ b/libavcodec/arm/fft_fixed_init_arm.c
@@ -0,0 +1,32 @@
+/*
+ * Copyright (c) 2009 Mans Rullgard <mans at mansr.com>
+ *
+ * This file is part of Libav.
+ *
+ * Libav 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.
+ *
+ * Libav 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 Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#define CONFIG_FFT_FLOAT 0
+#include "libavcodec/fft.h"
+
+void ff_fft_fixed_calc_neon(FFTContext *s, FFTComplex *z);
+
+av_cold void ff_fft_fixed_init_arm(FFTContext *s)
+{
+    if (HAVE_NEON) {
+        s->fft_permutation = FF_FFT_PERM_SWAP_LSBS;
+        s->fft_calc        = ff_fft_fixed_calc_neon;
+    }
+}
diff --git a/libavcodec/arm/fft_fixed_neon.S b/libavcodec/arm/fft_fixed_neon.S
new file mode 100644
index 0000000..14884d3
--- /dev/null
+++ b/libavcodec/arm/fft_fixed_neon.S
@@ -0,0 +1,261 @@
+/*
+ * Copyright (c) 2011 Mans Rullgard <mans at mansr.com>
+ *
+ * This file is part of Libav.
+ *
+ * Libav 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.
+ *
+ * Libav 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 Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "asm.S"
+
+.macro  bflies          d0,  d1,  r0,  r1
+        vrev64.32       \r0, \d1                @ t5, t6, t1, t2
+        vhsub.s16       \r1, \d1, \r0           @ t1-t5, t2-t6, t5-t1, t6-t2
+        vhadd.s16       \r0, \d1, \r0           @ t1+t5, t2+t6, t5+t1, t6+t2
+        vext.16         \r1, \r1, \r1, #1       @ t2-t6, t5-t1, t6-t2, t1-t5
+        vtrn.32         \r0, \r1                @ t1+t5, t2+t6, t2-t6, t5-t1
+                                                @ t5,    t6,    t4,    t3
+        vhsub.s16       \d1, \d0, \r0
+        vhadd.s16       \d0, \d0, \r0
+.endm
+
+.macro  transform01     q0,  q1,  d3,  c0,  c1,  r0,  w0,  w1
+        vrev32.16       \r0, \d3
+        vmull.s16       \w0, \d3, \c0
+        vmlal.s16       \w0, \r0, \c1
+        vshrn.s32       \d3, \w0, #15
+        bflies          \q0, \q1, \w0, \w1
+.endm
+
+.macro  transform2      d0,  d1,  d2,  d3,  q0,  q1,  c0,  c1,  c2,  c3, \
+                        r0,  r1,  w0,  w1
+        vrev32.16       \r0, \d1
+        vrev32.16       \r1, \d3
+        vmull.s16       \w0, \d1, \c0
+        vmlal.s16       \w0, \r0, \c1
+        vmull.s16       \w1, \d3, \c2
+        vmlal.s16       \w1, \r1, \c3
+        vshrn.s32       \d1, \w0, #15
+        vshrn.s32       \d3, \w1, #15
+        bflies          \q0, \q1, \w0, \w1
+.endm
+
+.macro  fft4            d0,  d1,  r0,  r1
+        vhsub.s16       \r0, \d0, \d1           @ t3, t4, t8, t7
+        vhsub.s16       \r1, \d1, \d0
+        vhadd.s16       \d0, \d0, \d1           @ t1, t2, t6, t5
+        vmov.i64        \d1, #0xffff<<32
+        vbit            \r0, \r1, \d1
+        vrev64.16       \r1, \r0                @ t7, t8, t4, t3
+        vtrn.32         \r0, \r1                @ t3, t4, t7, t8
+        vtrn.32         \d0, \r0                @ t1, t2, t3, t4, t6, t5, t8, t7
+        vhsub.s16       \d1, \d0, \r0           @ r2, i2, r3, i1
+        vhadd.s16       \d0, \d0, \r0           @ r0, i0, r1, i3
+.endm
+
+.macro  fft8            d0,  d1,  d2,  d3,  q0,  q1,  c0,  c1,  r0,  r1, w0, w1
+        fft4            \d0, \d1, \r0, \r1
+        vtrn.32         \d0, \d1                @ z0, z2, z1, z3
+        vhadd.s16       \r0, \d2, \d3           @ t1, t2, t3, t4
+        vhsub.s16       \d3, \d2, \d3           @ z5, z7
+        vmov            \d2, \r0
+        transform01     \q0, \q1, \d3, \c0, \c1, \r0, \w0, \w1
+.endm
+
+function fft4_neon
+        vld1.16         {d0-d1},  [r0,:128]
+        fft4            d0,  d1,  d2,  d3
+        vst1.16         {d0-d1},  [r0,:128]
+        bx              lr
+endfunc
+
+function fft8_neon
+        vld1.16         {d0-d3},  [r0,:128]
+        movrel          r1,  coefs
+        vld1.16         {d30},    [r1,:64]
+        vdup.16         d31, d30[0]
+        fft8            d0,  d1,  d2,  d3,  q0,  q1,  d31, d30, d20, d21, q8, q9
+        vtrn.32         d0,  d1
+        vtrn.32         d2,  d3
+        vst1.16         {d0-d3},  [r0,:128]
+        bx              lr
+endfunc
+
+function fft16_neon
+        vld1.16         {d0-d3},  [r0,:128]!
+        vld1.16         {d4-d7},  [r0,:128]
+        movrel          r1,  coefs
+        sub             r0,  r0,  #32
+        vld1.16         {d28-d31},[r1,:128]
+        vdup.16         d31, d28[0]
+        fft8            d0,  d1,  d2,  d3,  q0,  q1,  d31, d28, d20, d21, q8, q9
+        vswp            d5,  d6
+        fft4            q2,  q3,  q8,  q9
+        vswp            d5,  d6
+        vtrn.32         q0,  q1             @ z0, z4, z2, z6, z1, z5, z3, z7
+        vtrn.32         q2,  q3             @ z8, z12,z10,z14,z9, z13,z11,z15
+        vswp            d1,  d2
+        vdup.16         d31, d28[0]
+        transform01     q0,  q2,  d5,  d31, d28, d20, q8, q9
+        vdup.16         d26, d29[0]
+        vdup.16         d27, d30[0]
+        transform2      d2,  d6,  d3,  d7,  q1,  q3,  d26, d30, d27, d29, \
+                        d20, d21, q8,  q9
+        vtrn.32         q0,  q1
+        vtrn.32         q2,  q3
+        vst1.16         {d0-d3},  [r0,:128]!
+        vst1.16         {d4-d7},  [r0,:128]
+        bx              lr
+endfunc
+
+function fft_pass_neon
+        push            {r4,lr}
+        movrel          lr,  coefs + 24
+        vld1.16         {d30},    [lr,:64]
+        lsl             r12, r2,  #3
+        vmov            d31, d30
+        add             r3,  r1,  r2,  lsl #2
+        mov             lr,  #-8
+        sub             r3,  r3,  #2
+        mov             r4,  r0
+        vld1.16         {d27[]},  [r3,:16]
+        sub             r3,  r3,  #6
+        vld1.16         {q0},     [r4,:128], r12
+        vld1.16         {q1},     [r4,:128], r12
+        vld1.16         {q2},     [r4,:128], r12
+        vld1.16         {q3},     [r4,:128], r12
+        vld1.16         {d28},    [r1,:64]!
+        vld1.16         {d29},    [r3,:64], lr
+        vswp            d1,  d2
+        vswp            d5,  d6
+        vtrn.32         d0,  d1
+        vtrn.32         d4,  d5
+        vdup.16         d25, d28[1]
+        vmul.s16        d27, d27, d31
+        transform01     q0,  q2,  d5,  d25, d27, d20, q8,  q9
+        b               2f
+1:
+        mov             r4,  r0
+        vdup.16         d26, d29[0]
+        vld1.16         {q0},     [r4,:128], r12
+        vld1.16         {q1},     [r4,:128], r12
+        vld1.16         {q2},     [r4,:128], r12
+        vld1.16         {q3},     [r4,:128], r12
+        vld1.16         {d28},    [r1,:64]!
+        vld1.16         {d29},    [r3,:64], lr
+        vswp            d1,  d2
+        vswp            d5,  d6
+        vtrn.32         d0,  d1
+        vtrn.32         d4,  d5
+        vdup.16         d24, d28[0]
+        vdup.16         d25, d28[1]
+        vdup.16         d27, d29[3]
+        vmul.s16        q13, q13, q15
+        transform2      d0,  d4,  d1,  d5,  q0,  q2,  d24, d26, d25, d27, \
+                        d16, d17, q9,  q10
+2:
+        vtrn.32         d2,  d3
+        vtrn.32         d6,  d7
+        vdup.16         d24, d28[2]
+        vdup.16         d26, d29[2]
+        vdup.16         d25, d28[3]
+        vdup.16         d27, d29[1]
+        vmul.s16        q13, q13, q15
+        transform2      d2,  d6,  d3,  d7,  q1,  q3,  d24, d26, d25, d27, \
+                        d16, d17, q9,  q10
+        vtrn.32         d0,  d1
+        vtrn.32         d2,  d3
+        vtrn.32         d4,  d5
+        vtrn.32         d6,  d7
+        vswp            d1,  d2
+        vswp            d5,  d6
+        mov             r4,  r0
+        vst1.16         {q0},     [r4,:128], r12
+        vst1.16         {q1},     [r4,:128], r12
+        vst1.16         {q2},     [r4,:128], r12
+        vst1.16         {q3},     [r4,:128], r12
+        add             r0,  r0,  #16
+        subs            r2,  r2,  #2
+        bgt             1b
+        pop             {r4,pc}
+endfunc
+
+#define F_SQRT1_2   23170
+#define F_COS_16_1  30274
+#define F_COS_16_3  12540
+
+const   coefs, align=4
+        .short          F_SQRT1_2, -F_SQRT1_2, -F_SQRT1_2,  F_SQRT1_2
+        .short          F_COS_16_1,-F_COS_16_1,-F_COS_16_1, F_COS_16_1
+        .short          F_COS_16_3,-F_COS_16_3,-F_COS_16_3, F_COS_16_3
+        .short          1,         -1,         -1,          1
+endconst
+
+.macro  def_fft n, n2, n4
+function fft\n\()_neon
+        push            {r4, lr}
+        mov             r4,  r0
+        bl              fft\n2\()_neon
+        add             r0,  r4,  #\n4*2*4
+        bl              fft\n4\()_neon
+        add             r0,  r4,  #\n4*3*4
+        bl              fft\n4\()_neon
+        mov             r0,  r4
+        pop             {r4, lr}
+        movrel          r1,  X(ff_cos_\n\()_fixed)
+        mov             r2,  #\n4/2
+        b               fft_pass_neon
+endfunc
+.endm
+
+        def_fft    32,    16,     8
+        def_fft    64,    32,    16
+        def_fft   128,    64,    32
+        def_fft   256,   128,    64
+        def_fft   512,   256,   128
+        def_fft  1024,   512,   256
+        def_fft  2048,  1024,   512
+        def_fft  4096,  2048,  1024
+        def_fft  8192,  4096,  2048
+        def_fft 16384,  8192,  4096
+        def_fft 32768, 16384,  8192
+        def_fft 65536, 32768, 16384
+
+function ff_fft_fixed_calc_neon, export=1
+        ldr             r2,  [r0]
+        sub             r2,  r2,  #2
+        movrel          r3,  fft_fixed_tab_neon
+        ldr             r3,  [r3, r2, lsl #2]
+        mov             r0,  r1
+        bx              r3
+endfunc
+
+const   fft_fixed_tab_neon
+        .word fft4_neon
+        .word fft8_neon
+        .word fft16_neon
+        .word fft32_neon
+        .word fft64_neon
+        .word fft128_neon
+        .word fft256_neon
+        .word fft512_neon
+        .word fft1024_neon
+        .word fft2048_neon
+        .word fft4096_neon
+        .word fft8192_neon
+        .word fft16384_neon
+        .word fft32768_neon
+        .word fft65536_neon
+endconst
diff --git a/libavcodec/fft.c b/libavcodec/fft.c
index 19da483..077f471 100644
--- a/libavcodec/fft.c
+++ b/libavcodec/fft.c
@@ -126,6 +126,7 @@ av_cold int ff_fft_init(FFTContext *s, int nbits, int inverse)
     if (CONFIG_MDCT)  s->mdct_calcw = s->mdct_calc;
 #else
     if (CONFIG_MDCT)  s->mdct_calcw = ff_mdct_calcw_c;
+    if (ARCH_ARM)     ff_fft_fixed_init_arm(s);
 #endif
 
     for(j=4; j<=nbits; j++) {
diff --git a/libavcodec/fft.h b/libavcodec/fft.h
index 0f324cf..a4fee00 100644
--- a/libavcodec/fft.h
+++ b/libavcodec/fft.h
@@ -132,9 +132,13 @@ void ff_init_ff_cos_tabs(int index);
  */
 int ff_fft_init(FFTContext *s, int nbits, int inverse);
 
+#if CONFIG_FFT_FLOAT
 void ff_fft_init_altivec(FFTContext *s);
 void ff_fft_init_mmx(FFTContext *s);
 void ff_fft_init_arm(FFTContext *s);
+#else
+void ff_fft_fixed_init_arm(FFTContext *s);
+#endif
 
 void ff_fft_end(FFTContext *s);
 



More information about the ffmpeg-cvslog mailing list