AAC: ARM/NEON asm for VMUL2/4 functions

Originally committed as revision 21219 to svn://svn.ffmpeg.org/ffmpeg/trunk

2 files changed, 142 insertions, 0 deletions

diff --git a/libavcodec/aac.c b/libavcodec/aac.c
index b47b58e291..393f2e6594 100644
--- a/libavcodec/aac.c
+++ b/libavcodec/aac.c
@@ -93,6 +93,10 @@
 #include <math.h>
 #include <string.h>
 
+#if ARCH_ARM
+#   include "arm/aac.h"
+#endif
+
 union float754 {
     float f;
     uint32_t i;
@@ -862,6 +866,7 @@ static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
     }
 }
 
+#ifndef VMUL2
 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
                            const float *scale)
 {
@@ -870,7 +875,9 @@ static inline float *VMUL2(float *dst, const float *v, unsigned idx,
     *dst++ = v[idx>>4 & 15] * s;
     return dst;
 }
+#endif
 
+#ifndef VMUL4
 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
                            const float *scale)
 {
@@ -881,7 +888,9 @@ static inline float *VMUL4(float *dst, const float *v, unsigned idx,
     *dst++ = v[idx>>6 & 3] * s;
     return dst;
 }
+#endif
 
+#ifndef VMUL2S
 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
                             unsigned sign, const float *scale)
 {
@@ -896,7 +905,9 @@ static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
 
     return dst;
 }
+#endif
 
+#ifndef VMUL4S
 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
                             unsigned sign, const float *scale)
 {
@@ -921,6 +932,7 @@ static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
 
     return dst;
 }
+#endif
 
 /**
  * Decode spectral data; reference: table 4.50.
diff --git a/libavcodec/arm/aac.h b/libavcodec/arm/aac.h
new file mode 100644
index 0000000000..2ccb8cfb93
--- /dev/null
+++ b/libavcodec/arm/aac.h
@@ -0,0 +1,130 @@
+/*
+ * Copyright (c) 2010 Mans Rullgard <mans@mansr.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
+ */
+
+#if HAVE_NEON && HAVE_INLINE_ASM
+
+#define VMUL2 VMUL2
+static inline float *VMUL2(float *dst, const float *v, unsigned idx,
+                           const float *scale)
+{
+    unsigned v0, v1;
+    __asm__ volatile ("ubfx     %0,  %4,  #0, #4      \n\t"
+                      "ubfx     %1,  %4,  #4, #4      \n\t"
+                      "ldr      %0,  [%3, %0, lsl #2] \n\t"
+                      "ldr      %1,  [%3, %1, lsl #2] \n\t"
+                      "vld1.32  {d1[]},   [%5,:32]    \n\t"
+                      "vmov     d0,  %0,  %1          \n\t"
+                      "vmul.f32 d0,  d0,  d1          \n\t"
+                      "vst1.32  {d0},     [%2,:64]!   \n\t"
+                      : "=&r"(v0), "=&r"(v1), "+r"(dst)
+                      : "r"(v), "r"(idx), "r"(scale)
+                      : "d0", "d1");
+    return dst;
+}
+
+#define VMUL4 VMUL4
+static inline float *VMUL4(float *dst, const float *v, unsigned idx,
+                           const float *scale)
+{
+    unsigned v0, v1, v2, v3;
+    __asm__ volatile ("ubfx     %0,  %6,  #0, #2      \n\t"
+                      "ubfx     %1,  %6,  #2, #2      \n\t"
+                      "ldr      %0,  [%5, %0, lsl #2] \n\t"
+                      "ubfx     %2,  %6,  #4, #2      \n\t"
+                      "ldr      %1,  [%5, %1, lsl #2] \n\t"
+                      "ubfx     %3,  %6,  #6, #2      \n\t"
+                      "ldr      %2,  [%5, %2, lsl #2] \n\t"
+                      "vmov     d0,  %0,  %1          \n\t"
+                      "ldr      %3,  [%5, %3, lsl #2] \n\t"
+                      "vld1.32  {d2[],d3[]},[%7,:32]  \n\t"
+                      "vmov     d1,  %2,  %3          \n\t"
+                      "vmul.f32 q0,  q0,  q1          \n\t"
+                      "vst1.32  {q0},     [%4,:128]!  \n\t"
+                      : "=&r"(v0), "=&r"(v1), "=&r"(v2), "=&r"(v3), "+r"(dst)
+                      : "r"(v), "r"(idx), "r"(scale)
+                      : "d0", "d1", "d2", "d3");
+    return dst;
+}
+
+#define VMUL2S VMUL2S
+static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
+                            unsigned sign, const float *scale)
+{
+    unsigned v0, v1, v2, v3;
+    __asm__ volatile ("ubfx     %0,  %6,  #0, #4      \n\t"
+                      "ubfx     %1,  %6,  #4, #4      \n\t"
+                      "ldr      %0,  [%5, %0, lsl #2] \n\t"
+                      "lsl      %2,  %8,  #30         \n\t"
+                      "ldr      %1,  [%5, %1, lsl #2] \n\t"
+                      "lsl      %3,  %8,  #31         \n\t"
+                      "vmov     d0,  %0,  %1          \n\t"
+                      "bic      %2,  %2,  #1<<30      \n\t"
+                      "vld1.32  {d1[]},   [%7,:32]    \n\t"
+                      "vmov     d2,  %2,  %3          \n\t"
+                      "veor     d0,  d0,  d2          \n\t"
+                      "vmul.f32 d0,  d0,  d1          \n\t"
+                      "vst1.32  {d0},     [%4,:64]!   \n\t"
+                      : "=&r"(v0), "=&r"(v1), "=&r"(v2), "=&r"(v3), "+r"(dst)
+                      : "r"(v), "r"(idx), "r"(scale), "r"(sign)
+                      : "d0", "d1", "d2");
+    return dst;
+}
+
+#define VMUL4S VMUL4S
+static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
+                            unsigned sign, const float *scale)
+{
+    unsigned v0, v1, v2, v3, nz;
+    __asm__ volatile ("vld1.32  {d2[],d3[]},[%9,:32]  \n\t"
+                      "ubfx     %0,  %8,  #0, #2      \n\t"
+                      "ubfx     %1,  %8,  #2, #2      \n\t"
+                      "ldr      %0,  [%7, %0, lsl #2] \n\t"
+                      "ubfx     %2,  %8,  #4, #2      \n\t"
+                      "ldr      %1,  [%7, %1, lsl #2] \n\t"
+                      "ubfx     %3,  %8,  #6, #2      \n\t"
+                      "ldr      %2,  [%7, %2, lsl #2] \n\t"
+                      "vmov     d0,  %0,  %1          \n\t"
+                      "ldr      %3,  [%7, %3, lsl #2] \n\t"
+                      "lsr      %6,  %8,  #12         \n\t"
+                      "rbit     %6,  %6               \n\t"
+                      "vmov     d1,  %2,  %3          \n\t"
+                      "lsls     %6,  %6,  #1          \n\t"
+                      "and      %0,  %5,  #1<<31      \n\t"
+                      "lslcs    %5,  %5,  #1          \n\t"
+                      "lsls     %6,  %6,  #1          \n\t"
+                      "and      %1,  %5,  #1<<31      \n\t"
+                      "lslcs    %5,  %5,  #1          \n\t"
+                      "lsls     %6,  %6,  #1          \n\t"
+                      "and      %2,  %5,  #1<<31      \n\t"
+                      "lslcs    %5,  %5,  #1          \n\t"
+                      "vmov     d4,  %0,  %1          \n\t"
+                      "and      %3,  %5,  #1<<31      \n\t"
+                      "vmov     d5,  %2,  %3          \n\t"
+                      "veor     q0,  q0,  q2          \n\t"
+                      "vmul.f32 q0,  q0,  q1          \n\t"
+                      "vst1.32  {q0},     [%4,:128]!  \n\t"
+                      : "=&r"(v0), "=&r"(v1), "=&r"(v2), "=&r"(v3), "+r"(dst),
+                        "+r"(sign), "=r"(nz)
+                      : "r"(v), "r"(idx), "r"(scale)
+                      : "d0", "d1", "d2", "d3", "d4", "d5");
+    return dst;
+}
+
+#endif /* HAVE_NEON && HAVE_INLINE_ASM */