Merge remote-tracking branch 'qatar/master'

* qatar/master:
  float_dsp: ppc: add a separate header for Altivec function prototypes
  ARM: fix float_dsp breakage from d5a7229
  Add a float DSP framework to libavutil
  PPC: Move types_altivec.h and util_altivec.h from libavcodec to libavutil
  ARM: Move asm.S from libavcodec to libavutil
  vc1dsp: mark put/avg_vc1_mspel_mc() always_inline

Merged-by: Michael Niedermayer <michaelni@gmx.at>

20 files changed, 962 insertions, 0 deletions

diff --git a/libavutil/Makefile b/libavutil/Makefile
index a4498313a9..43c6177733 100644
--- a/libavutil/Makefile
+++ b/libavutil/Makefile
@@ -63,6 +63,7 @@ OBJS = adler32.o                                                        \
        eval.o                                                           \
        fifo.o                                                           \
        file.o                                                           \
+       float_dsp.o                                                      \
        imgutils.o                                                       \
        intfloat_readwrite.o                                             \
        inverse.o                                                        \
diff --git a/libavutil/arm/Makefile b/libavutil/arm/Makefile
index e600383b9c..ac7eca6751 100644
--- a/libavutil/arm/Makefile
+++ b/libavutil/arm/Makefile
@@ -1 +1,8 @@
 OBJS += arm/cpu.o                                                       \
+        arm/float_dsp_init_arm.o                                        \
+
+ARMVFP-OBJS += arm/float_dsp_init_vfp.o                                 \
+               arm/float_dsp_vfp.o                                      \
+
+NEON-OBJS += arm/float_dsp_init_neon.o                                  \
+             arm/float_dsp_neon.o                                       \
diff --git a/libavutil/arm/asm.S b/libavutil/arm/asm.S
new file mode 100644
index 0000000000..28d52cfd53
--- /dev/null
+++ b/libavutil/arm/asm.S
@@ -0,0 +1,231 @@
+/*
+ * Copyright (c) 2008 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
+ */
+
+#include "config.h"
+
+#ifdef __ELF__
+#   define ELF
+#else
+#   define ELF @
+#endif
+
+#if CONFIG_THUMB
+#   define A @
+#   define T
+#else
+#   define A
+#   define T @
+#endif
+
+#if   HAVE_NEON
+        .arch           armv7-a
+#elif HAVE_ARMV6T2
+        .arch           armv6t2
+#elif HAVE_ARMV6
+        .arch           armv6
+#elif HAVE_ARMV5TE
+        .arch           armv5te
+#endif
+
+#if   HAVE_NEON
+        .fpu            neon
+#elif HAVE_ARMVFP
+        .fpu            vfp
+#endif
+
+        .syntax unified
+T       .thumb
+
+.macro  require8 val=1
+ELF     .eabi_attribute 24, \val
+.endm
+
+.macro  preserve8 val=1
+ELF     .eabi_attribute 25, \val
+.endm
+
+.macro  function name, export=0
+    .macro endfunc
+ELF     .size   \name, . - \name
+        .endfunc
+        .purgem endfunc
+    .endm
+        .text
+        .align          2
+    .if \export
+        .global EXTERN_ASM\name
+EXTERN_ASM\name:
+    .endif
+ELF     .type   \name, %function
+        .func   \name
+\name:
+.endm
+
+.macro  const   name, align=2
+    .macro endconst
+ELF     .size   \name, . - \name
+        .purgem endconst
+    .endm
+        .section        .rodata
+        .align          \align
+\name:
+.endm
+
+#if !HAVE_ARMV6T2
+.macro  movw    rd, val
+        mov     \rd, \val &  255
+        orr     \rd, \val & ~255
+.endm
+#endif
+
+.macro  mov32   rd, val
+#if HAVE_ARMV6T2
+        movw            \rd, #(\val) & 0xffff
+    .if (\val) >> 16
+        movt            \rd, #(\val) >> 16
+    .endif
+#else
+        ldr             \rd, =\val
+#endif
+.endm
+
+.macro  movrel rd, val
+#if HAVE_ARMV6T2 && !CONFIG_PIC && !defined(__APPLE__)
+        movw            \rd, #:lower16:\val
+        movt            \rd, #:upper16:\val
+#else
+        ldr             \rd, =\val
+#endif
+.endm
+
+.macro  ldr_pre         rt,  rn,  rm:vararg
+A       ldr             \rt, [\rn, \rm]!
+T       add             \rn, \rn, \rm
+T       ldr             \rt, [\rn]
+.endm
+
+.macro  ldr_dpre        rt,  rn,  rm:vararg
+A       ldr             \rt, [\rn, -\rm]!
+T       sub             \rn, \rn, \rm
+T       ldr             \rt, [\rn]
+.endm
+
+.macro  ldr_nreg        rt,  rn,  rm:vararg
+A       ldr             \rt, [\rn, -\rm]
+T       sub             \rt, \rn, \rm
+T       ldr             \rt, [\rt]
+.endm
+
+.macro  ldr_post        rt,  rn,  rm:vararg
+A       ldr             \rt, [\rn], \rm
+T       ldr             \rt, [\rn]
+T       add             \rn, \rn, \rm
+.endm
+
+.macro  ldrd_reg        rt,  rt2, rn,  rm
+A       ldrd            \rt, \rt2, [\rn, \rm]
+T       add             \rt, \rn, \rm
+T       ldrd            \rt, \rt2, [\rt]
+.endm
+
+.macro  ldrd_post       rt,  rt2, rn,  rm
+A       ldrd            \rt, \rt2, [\rn], \rm
+T       ldrd            \rt, \rt2, [\rn]
+T       add             \rn, \rn, \rm
+.endm
+
+.macro  ldrh_pre        rt,  rn,  rm
+A       ldrh            \rt, [\rn, \rm]!
+T       add             \rn, \rn, \rm
+T       ldrh            \rt, [\rn]
+.endm
+
+.macro  ldrh_dpre       rt,  rn,  rm
+A       ldrh            \rt, [\rn, -\rm]!
+T       sub             \rn, \rn, \rm
+T       ldrh            \rt, [\rn]
+.endm
+
+.macro  ldrh_post       rt,  rn,  rm
+A       ldrh            \rt, [\rn], \rm
+T       ldrh            \rt, [\rn]
+T       add             \rn, \rn, \rm
+.endm
+
+.macro  ldrb_post       rt,  rn,  rm
+A       ldrb            \rt, [\rn], \rm
+T       ldrb            \rt, [\rn]
+T       add             \rn, \rn, \rm
+.endm
+
+.macro  str_post       rt,  rn,  rm:vararg
+A       str             \rt, [\rn], \rm
+T       str             \rt, [\rn]
+T       add             \rn, \rn, \rm
+.endm
+
+.macro  strb_post       rt,  rn,  rm:vararg
+A       strb            \rt, [\rn], \rm
+T       strb            \rt, [\rn]
+T       add             \rn, \rn, \rm
+.endm
+
+.macro  strd_post       rt,  rt2, rn,  rm
+A       strd            \rt, \rt2, [\rn], \rm
+T       strd            \rt, \rt2, [\rn]
+T       add             \rn, \rn, \rm
+.endm
+
+.macro  strh_pre        rt,  rn,  rm
+A       strh            \rt, [\rn, \rm]!
+T       add             \rn, \rn, \rm
+T       strh            \rt, [\rn]
+.endm
+
+.macro  strh_dpre       rt,  rn,  rm
+A       strh            \rt, [\rn, -\rm]!
+T       sub             \rn, \rn, \rm
+T       strh            \rt, [\rn]
+.endm
+
+.macro  strh_post       rt,  rn,  rm
+A       strh            \rt, [\rn], \rm
+T       strh            \rt, [\rn]
+T       add             \rn, \rn, \rm
+.endm
+
+.macro  strh_dpost       rt,  rn,  rm
+A       strh            \rt, [\rn], -\rm
+T       strh            \rt, [\rn]
+T       sub             \rn, \rn, \rm
+.endm
+
+#if HAVE_VFP_ARGS
+        .eabi_attribute 28, 1
+#   define VFP
+#   define NOVFP @
+#else
+#   define VFP   @
+#   define NOVFP
+#endif
+
+#define GLUE(a, b) a ## b
+#define JOIN(a, b) GLUE(a, b)
+#define X(s) JOIN(EXTERN_ASM, s)
diff --git a/libavutil/arm/float_dsp_arm.h b/libavutil/arm/float_dsp_arm.h
new file mode 100644
index 0000000000..f3fafe326d
--- /dev/null
+++ b/libavutil/arm/float_dsp_arm.h
@@ -0,0 +1,29 @@
+/*
+ * Copyright (c) 2009 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
+ */
+
+#ifndef AVUTIL_ARM_FLOAT_DSP_ARM_H
+#define AVUTIL_ARM_FLOAT_DSP_ARM_H
+
+#include "libavutil/float_dsp.h"
+
+void ff_float_dsp_init_vfp (AVFloatDSPContext *fdsp);
+void ff_float_dsp_init_neon(AVFloatDSPContext *fdsp);
+
+#endif /* AVUTIL_ARM_FLOAT_DSP_ARM_H */
diff --git a/libavutil/arm/float_dsp_init_arm.c b/libavutil/arm/float_dsp_init_arm.c
new file mode 100644
index 0000000000..ab636dbcd2
--- /dev/null
+++ b/libavutil/arm/float_dsp_init_arm.c
@@ -0,0 +1,33 @@
+/*
+ * ARM optimized DSP utils
+ *
+ * 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/cpu.h"
+#include "libavutil/float_dsp.h"
+#include "float_dsp_arm.h"
+
+void ff_float_dsp_init_arm(AVFloatDSPContext *fdsp)
+{
+    int cpu_flags = av_get_cpu_flags();
+
+    if (have_vfp(cpu_flags))
+        ff_float_dsp_init_vfp(fdsp);
+    if (have_neon(cpu_flags))
+        ff_float_dsp_init_neon(fdsp);
+}
diff --git a/libavutil/arm/float_dsp_init_neon.c b/libavutil/arm/float_dsp_init_neon.c
new file mode 100644
index 0000000000..fa6d0d7d15
--- /dev/null
+++ b/libavutil/arm/float_dsp_init_neon.c
@@ -0,0 +1,32 @@
+/*
+ * ARM NEON optimised Float DSP functions
+ * Copyright (c) 2008 Mans Rullgard <mans@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 <stdint.h>
+
+#include "libavutil/float_dsp.h"
+#include "float_dsp_arm.h"
+
+void ff_vector_fmul_neon(float *dst, const float *src0, const float *src1, int len);
+
+void ff_float_dsp_init_neon(AVFloatDSPContext *fdsp)
+{
+    fdsp->vector_fmul = ff_vector_fmul_neon;
+}
diff --git a/libavutil/arm/float_dsp_init_vfp.c b/libavutil/arm/float_dsp_init_vfp.c
new file mode 100644
index 0000000000..dfde3fdb64
--- /dev/null
+++ b/libavutil/arm/float_dsp_init_vfp.c
@@ -0,0 +1,34 @@
+/*
+ * Copyright (c) 2008 Siarhei Siamashka <ssvb@users.sourceforge.net>
+ *
+ * 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/cpu.h"
+#include "libavutil/float_dsp.h"
+#include "float_dsp_arm.h"
+
+void ff_vector_fmul_vfp(float *dst, const float *src0, const float *src1,
+                        int len);
+
+void ff_float_dsp_init_vfp(AVFloatDSPContext *fdsp)
+{
+    int cpu_flags = av_get_cpu_flags();
+
+    if (!have_vfpv3(cpu_flags))
+        fdsp->vector_fmul = ff_vector_fmul_vfp;
+}
diff --git a/libavutil/arm/float_dsp_neon.S b/libavutil/arm/float_dsp_neon.S
new file mode 100644
index 0000000000..d66fa09424
--- /dev/null
+++ b/libavutil/arm/float_dsp_neon.S
@@ -0,0 +1,64 @@
+/*
+ * ARM NEON optimised Float DSP functions
+ * Copyright (c) 2008 Mans Rullgard <mans@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 "config.h"
+#include "asm.S"
+
+        preserve8
+
+function ff_vector_fmul_neon, export=1
+        subs            r3,  r3,  #8
+        vld1.32         {d0-d3},  [r1,:128]!
+        vld1.32         {d4-d7},  [r2,:128]!
+        vmul.f32        q8,  q0,  q2
+        vmul.f32        q9,  q1,  q3
+        beq             3f
+        bics            ip,  r3,  #15
+        beq             2f
+1:      subs            ip,  ip,  #16
+        vld1.32         {d0-d1},  [r1,:128]!
+        vld1.32         {d4-d5},  [r2,:128]!
+        vmul.f32        q10, q0,  q2
+        vld1.32         {d2-d3},  [r1,:128]!
+        vld1.32         {d6-d7},  [r2,:128]!
+        vmul.f32        q11, q1,  q3
+        vst1.32         {d16-d19},[r0,:128]!
+        vld1.32         {d0-d1},  [r1,:128]!
+        vld1.32         {d4-d5},  [r2,:128]!
+        vmul.f32        q8,  q0,  q2
+        vld1.32         {d2-d3},  [r1,:128]!
+        vld1.32         {d6-d7},  [r2,:128]!
+        vmul.f32        q9,  q1,  q3
+        vst1.32         {d20-d23},[r0,:128]!
+        bne             1b
+        ands            r3,  r3,  #15
+        beq             3f
+2:      vld1.32         {d0-d1},  [r1,:128]!
+        vld1.32         {d4-d5},  [r2,:128]!
+        vst1.32         {d16-d17},[r0,:128]!
+        vmul.f32        q8,  q0,  q2
+        vld1.32         {d2-d3},  [r1,:128]!
+        vld1.32         {d6-d7},  [r2,:128]!
+        vst1.32         {d18-d19},[r0,:128]!
+        vmul.f32        q9,  q1,  q3
+3:      vst1.32         {d16-d19},[r0,:128]!
+        bx              lr
+endfunc
diff --git a/libavutil/arm/float_dsp_vfp.S b/libavutil/arm/float_dsp_vfp.S
new file mode 100644
index 0000000000..db63e5a675
--- /dev/null
+++ b/libavutil/arm/float_dsp_vfp.S
@@ -0,0 +1,68 @@
+/*
+ * Copyright (c) 2008 Siarhei Siamashka <ssvb@users.sourceforge.net>
+ *
+ * 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 "config.h"
+#include "asm.S"
+
+/**
+ * Assume that len is a positive number and is multiple of 8
+ */
+@ void ff_vector_fmul_vfp(float *dst, const float *src0, const float *src1, int len)
+function ff_vector_fmul_vfp, export=1
+        vpush           {d8-d15}
+        fmrx            r12, fpscr
+        orr             r12, r12, #(3 << 16) /* set vector size to 4 */
+        fmxr            fpscr, r12
+
+        vldmia          r1!, {s0-s3}
+        vldmia          r2!, {s8-s11}
+        vldmia          r1!, {s4-s7}
+        vldmia          r2!, {s12-s15}
+        vmul.f32        s8,  s0,  s8
+1:
+        subs            r3,  r3,  #16
+        vmul.f32        s12, s4,  s12
+        itttt           ge
+        vldmiage        r1!, {s16-s19}
+        vldmiage        r2!, {s24-s27}
+        vldmiage        r1!, {s20-s23}
+        vldmiage        r2!, {s28-s31}
+        it              ge
+        vmulge.f32      s24, s16, s24
+        vstmia          r0!, {s8-s11}
+        vstmia          r0!, {s12-s15}
+        it              ge
+        vmulge.f32      s28, s20, s28
+        itttt           gt
+        vldmiagt        r1!, {s0-s3}
+        vldmiagt        r2!, {s8-s11}
+        vldmiagt        r1!, {s4-s7}
+        vldmiagt        r2!, {s12-s15}
+        ittt            ge
+        vmulge.f32      s8,  s0,  s8
+        vstmiage        r0!, {s24-s27}
+        vstmiage        r0!, {s28-s31}
+        bgt             1b
+
+        bic             r12, r12, #(7 << 16) /* set vector size back to 1 */
+        fmxr            fpscr, r12
+        vpop            {d8-d15}
+        bx              lr
+endfunc
diff --git a/libavutil/float_dsp.c b/libavutil/float_dsp.c
new file mode 100644
index 0000000000..039dd07d36
--- /dev/null
+++ b/libavutil/float_dsp.c
@@ -0,0 +1,42 @@
+/*
+ * 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 "config.h"
+
+#include "float_dsp.h"
+
+static void vector_fmul_c(float *dst, const float *src0, const float *src1,
+                          int len)
+{
+    int i;
+    for (i = 0; i < len; i++)
+        dst[i] = src0[i] * src1[i];
+}
+
+void avpriv_float_dsp_init(AVFloatDSPContext *fdsp, int bit_exact)
+{
+    fdsp->vector_fmul = vector_fmul_c;
+
+#if ARCH_ARM
+    ff_float_dsp_init_arm(fdsp);
+#elif ARCH_PPC
+    ff_float_dsp_init_ppc(fdsp, bit_exact);
+#elif ARCH_X86
+    ff_float_dsp_init_x86(fdsp);
+#endif
+}
diff --git a/libavutil/float_dsp.h b/libavutil/float_dsp.h
new file mode 100644
index 0000000000..02c4ab7bde
--- /dev/null
+++ b/libavutil/float_dsp.h
@@ -0,0 +1,53 @@
+/*
+ * 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
+ */
+
+#ifndef AVUTIL_FLOAT_DSP_H
+#define AVUTIL_FLOAT_DSP_H
+
+typedef struct AVFloatDSPContext {
+    /**
+     * Calculate the product of two vectors of floats and store the result in
+     * a vector of floats.
+     *
+     * @param dst  output vector
+     *             constraints: 32-byte aligned
+     * @param src0 first input vector
+     *             constraints: 32-byte aligned
+     * @param src1 second input vector
+     *             constraints: 32-byte aligned
+     * @param len  number of elements in the input
+     *             constraints: multiple of 16
+     */
+    void (*vector_fmul)(float *dst, const float *src0, const float *src1,
+                        int len);
+} AVFloatDSPContext;
+
+/**
+ * Initialize a float DSP context.
+ *
+ * @param fdsp    float DSP context
+ * @param strict  setting to non-zero avoids using functions which may not be IEEE-754 compliant
+ */
+void avpriv_float_dsp_init(AVFloatDSPContext *fdsp, int strict);
+
+
+void ff_float_dsp_init_arm(AVFloatDSPContext *fdsp);
+void ff_float_dsp_init_ppc(AVFloatDSPContext *fdsp, int strict);
+void ff_float_dsp_init_x86(AVFloatDSPContext *fdsp);
+
+#endif /* AVUTIL_FLOAT_DSP_H */
diff --git a/libavutil/ppc/Makefile b/libavutil/ppc/Makefile
index 5b18b08957..4fd8d6d57e 100644
--- a/libavutil/ppc/Makefile
+++ b/libavutil/ppc/Makefile
@@ -1 +1,4 @@
 OBJS += ppc/cpu.o                                                       \
+        ppc/float_dsp_init.o                                            \
+
+ALTIVEC-OBJS += ppc/float_dsp_altivec.o                                 \
diff --git a/libavutil/ppc/float_dsp_altivec.c b/libavutil/ppc/float_dsp_altivec.c
new file mode 100644
index 0000000000..6340e6c17b
--- /dev/null
+++ b/libavutil/ppc/float_dsp_altivec.c
@@ -0,0 +1,38 @@
+/*
+ * Copyright (c) 2006 Luca Barbato <lu_zero@gentoo.org>
+ *
+ * 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 "util_altivec.h"
+#include "float_dsp_altivec.h"
+
+void ff_vector_fmul_altivec(float *dst, const float *src0, const float *src1,
+                            int len)
+{
+    int i;
+    vector float d0, d1, s, zero = (vector float)vec_splat_u32(0);
+    for (i = 0; i < len - 7; i += 8) {
+        d0 = vec_ld( 0, src0 + i);
+        s  = vec_ld( 0, src1 + i);
+        d1 = vec_ld(16, src0 + i);
+        d0 = vec_madd(d0, s, zero);
+        d1 = vec_madd(d1, vec_ld(16, src1 + i), zero);
+        vec_st(d0,  0, dst + i);
+        vec_st(d1, 16, dst + i);
+    }
+}
diff --git a/libavutil/ppc/float_dsp_altivec.h b/libavutil/ppc/float_dsp_altivec.h
new file mode 100644
index 0000000000..20c89c2338
--- /dev/null
+++ b/libavutil/ppc/float_dsp_altivec.h
@@ -0,0 +1,27 @@
+/*
+ * Copyright (c) 2006 Luca Barbato <lu_zero@gentoo.org>
+ *
+ * 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
+ */
+
+#ifndef AVUTIL_PPC_FLOAT_DSP_ALTIVEC_H
+#define AVUTIL_PPC_FLOAT_DSP_ALTIVEC_H
+
+extern void ff_vector_fmul_altivec(float *dst, const float *src0,
+                                   const float *src1, int len);
+
+#endif /* AVUTIL_PPC_FLOAT_DSP_ALTIVEC_H */
diff --git a/libavutil/ppc/float_dsp_init.c b/libavutil/ppc/float_dsp_init.c
new file mode 100644
index 0000000000..d0ae788779
--- /dev/null
+++ b/libavutil/ppc/float_dsp_init.c
@@ -0,0 +1,36 @@
+/*
+ * Copyright (c) 2006 Luca Barbato <lu_zero@gentoo.org>
+ *
+ * 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 "config.h"
+#include "libavutil/cpu.h"
+#include "libavutil/float_dsp.h"
+#include "float_dsp_altivec.h"
+
+void ff_float_dsp_init_ppc(AVFloatDSPContext *fdsp, int bit_exact)
+{
+#if HAVE_ALTIVEC
+    int mm_flags = av_get_cpu_flags();
+
+    if (!(mm_flags & AV_CPU_FLAG_ALTIVEC))
+        return;
+
+    fdsp->vector_fmul = ff_vector_fmul_altivec;
+#endif
+}
diff --git a/libavutil/ppc/types_altivec.h b/libavutil/ppc/types_altivec.h
new file mode 100644
index 0000000000..69d8957509
--- /dev/null
+++ b/libavutil/ppc/types_altivec.h
@@ -0,0 +1,47 @@
+/*
+ * Copyright (c) 2006 Guillaume Poirier <gpoirier@mplayerhq.hu>
+ *
+ * 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
+ */
+
+#ifndef AVUTIL_PPC_TYPES_ALTIVEC_H
+#define AVUTIL_PPC_TYPES_ALTIVEC_H
+
+/***********************************************************************
+ * Vector types
+ **********************************************************************/
+#define vec_u8  vector unsigned char
+#define vec_s8  vector signed char
+#define vec_u16 vector unsigned short
+#define vec_s16 vector signed short
+#define vec_u32 vector unsigned int
+#define vec_s32 vector signed int
+#define vec_f   vector float
+
+/***********************************************************************
+ * Null vector
+ **********************************************************************/
+#define LOAD_ZERO const vec_u8 zerov = vec_splat_u8( 0 )
+
+#define zero_u8v  (vec_u8)  zerov
+#define zero_s8v  (vec_s8)  zerov
+#define zero_u16v (vec_u16) zerov
+#define zero_s16v (vec_s16) zerov
+#define zero_u32v (vec_u32) zerov
+#define zero_s32v (vec_s32) zerov
+
+#endif /* AVUTIL_PPC_TYPES_ALTIVEC_H */
diff --git a/libavutil/ppc/util_altivec.h b/libavutil/ppc/util_altivec.h
new file mode 100644
index 0000000000..7fe31504ed
--- /dev/null
+++ b/libavutil/ppc/util_altivec.h
@@ -0,0 +1,118 @@
+/*
+ * 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
+ */
+
+/**
+ * @file
+ * Contains misc utility macros and inline functions
+ */
+
+#ifndef AVUTIL_PPC_UTIL_ALTIVEC_H
+#define AVUTIL_PPC_UTIL_ALTIVEC_H
+
+#include <stdint.h>
+
+#include "config.h"
+
+#if HAVE_ALTIVEC_H
+#include <altivec.h>
+#endif
+
+#include "types_altivec.h"
+
+// used to build registers permutation vectors (vcprm)
+// the 's' are for words in the _s_econd vector
+#define WORD_0 0x00,0x01,0x02,0x03
+#define WORD_1 0x04,0x05,0x06,0x07
+#define WORD_2 0x08,0x09,0x0a,0x0b
+#define WORD_3 0x0c,0x0d,0x0e,0x0f
+#define WORD_s0 0x10,0x11,0x12,0x13
+#define WORD_s1 0x14,0x15,0x16,0x17
+#define WORD_s2 0x18,0x19,0x1a,0x1b
+#define WORD_s3 0x1c,0x1d,0x1e,0x1f
+
+#define vcprm(a,b,c,d) (const vector unsigned char){WORD_ ## a, WORD_ ## b, WORD_ ## c, WORD_ ## d}
+#define vcii(a,b,c,d) (const vector float){FLOAT_ ## a, FLOAT_ ## b, FLOAT_ ## c, FLOAT_ ## d}
+
+// vcprmle is used to keep the same index as in the SSE version.
+// it's the same as vcprm, with the index inversed
+// ('le' is Little Endian)
+#define vcprmle(a,b,c,d) vcprm(d,c,b,a)
+
+// used to build inverse/identity vectors (vcii)
+// n is _n_egative, p is _p_ositive
+#define FLOAT_n -1.
+#define FLOAT_p 1.
+
+
+// Transpose 8x8 matrix of 16-bit elements (in-place)
+#define TRANSPOSE8(a,b,c,d,e,f,g,h) \
+do { \
+    vector signed short A1, B1, C1, D1, E1, F1, G1, H1; \
+    vector signed short A2, B2, C2, D2, E2, F2, G2, H2; \
+ \
+    A1 = vec_mergeh (a, e); \
+    B1 = vec_mergel (a, e); \
+    C1 = vec_mergeh (b, f); \
+    D1 = vec_mergel (b, f); \
+    E1 = vec_mergeh (c, g); \
+    F1 = vec_mergel (c, g); \
+    G1 = vec_mergeh (d, h); \
+    H1 = vec_mergel (d, h); \
+ \
+    A2 = vec_mergeh (A1, E1); \
+    B2 = vec_mergel (A1, E1); \
+    C2 = vec_mergeh (B1, F1); \
+    D2 = vec_mergel (B1, F1); \
+    E2 = vec_mergeh (C1, G1); \
+    F2 = vec_mergel (C1, G1); \
+    G2 = vec_mergeh (D1, H1); \
+    H2 = vec_mergel (D1, H1); \
+ \
+    a = vec_mergeh (A2, E2); \
+    b = vec_mergel (A2, E2); \
+    c = vec_mergeh (B2, F2); \
+    d = vec_mergel (B2, F2); \
+    e = vec_mergeh (C2, G2); \
+    f = vec_mergel (C2, G2); \
+    g = vec_mergeh (D2, H2); \
+    h = vec_mergel (D2, H2); \
+} while (0)
+
+
+/** @brief loads unaligned vector @a *src with offset @a offset
+    and returns it */
+static inline vector unsigned char unaligned_load(int offset, uint8_t *src)
+{
+    register vector unsigned char first = vec_ld(offset, src);
+    register vector unsigned char second = vec_ld(offset+15, src);
+    register vector unsigned char mask = vec_lvsl(offset, src);
+    return vec_perm(first, second, mask);
+}
+
+/**
+ * loads vector known misalignment
+ * @param perm_vec the align permute vector to combine the two loads from lvsl
+ */
+static inline vec_u8 load_with_perm_vec(int offset, uint8_t *src, vec_u8 perm_vec)
+{
+    vec_u8 a = vec_ld(offset, src);
+    vec_u8 b = vec_ld(offset+15, src);
+    return vec_perm(a, b, perm_vec);
+}
+
+#endif /* AVUTIL_PPC_UTIL_ALTIVEC_H */
diff --git a/libavutil/x86/Makefile b/libavutil/x86/Makefile
index de8a341442..4546353669 100644
--- a/libavutil/x86/Makefile
+++ b/libavutil/x86/Makefile
@@ -1 +1,4 @@
 OBJS += x86/cpu.o                                                       \
+        x86/float_dsp_init.o                                            \
+
+YASM-OBJS += x86/float_dsp.o                                            \
diff --git a/libavutil/x86/float_dsp.asm b/libavutil/x86/float_dsp.asm
new file mode 100644
index 0000000000..53be7ab99a
--- /dev/null
+++ b/libavutil/x86/float_dsp.asm
@@ -0,0 +1,55 @@
+;*****************************************************************************
+;* x86-optimized Float DSP functions
+;*
+;* 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 "x86inc.asm"
+
+SECTION .text
+
+;-----------------------------------------------------------------------------
+; void vector_fmul(float *dst, const float *src0, const float *src1, int len)
+;-----------------------------------------------------------------------------
+%macro VECTOR_FMUL 0
+cglobal vector_fmul, 4,4,2, dst, src0, src1, len
+    lea       lenq, [lend*4 - 2*mmsize]
+ALIGN 16
+.loop
+    mova      m0,   [src0q + lenq]
+    mova      m1,   [src0q + lenq + mmsize]
+    mulps     m0, m0, [src1q + lenq]
+    mulps     m1, m1, [src1q + lenq + mmsize]
+    mova      [dstq + lenq], m0
+    mova      [dstq + lenq + mmsize], m1
+
+    sub       lenq, 2*mmsize
+    jge       .loop
+%if mmsize == 32
+    vzeroupper
+    RET
+%else
+    REP_RET
+%endif
+%endmacro
+
+INIT_XMM sse
+VECTOR_FMUL
+%if HAVE_AVX
+INIT_YMM avx
+VECTOR_FMUL
+%endif
diff --git a/libavutil/x86/float_dsp_init.c b/libavutil/x86/float_dsp_init.c
new file mode 100644
index 0000000000..8f6980cbc2
--- /dev/null
+++ b/libavutil/x86/float_dsp_init.c
@@ -0,0 +1,41 @@
+/*
+ * 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 "config.h"
+
+#include "libavutil/cpu.h"
+#include "libavutil/float_dsp.h"
+
+extern void ff_vector_fmul_sse(float *dst, const float *src0, const float *src1,
+                               int len);
+extern void ff_vector_fmul_avx(float *dst, const float *src0, const float *src1,
+                               int len);
+
+void ff_float_dsp_init_x86(AVFloatDSPContext *fdsp)
+{
+#if HAVE_YASM
+    int mm_flags = av_get_cpu_flags();
+
+    if (mm_flags & AV_CPU_FLAG_SSE && HAVE_SSE) {
+        fdsp->vector_fmul = ff_vector_fmul_sse;
+    }
+    if (mm_flags & AV_CPU_FLAG_AVX && HAVE_AVX) {
+        fdsp->vector_fmul = ff_vector_fmul_avx;
+    }
+#endif
+}