Add m-esr52 at 52.6.0

13 files changed, 3033 insertions, 0 deletions

diff --git a/media/libav/libavutil/x86/asm.h b/media/libav/libavutil/x86/asm.h
new file mode 100644
index 000000000..e30f5dbaf
--- /dev/null
+++ b/media/libav/libavutil/x86/asm.h
@@ -0,0 +1,112 @@
+/*
+ * copyright (c) 2006 Michael Niedermayer <michaelni@gmx.at>
+ *
+ * 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
+ */
+
+#ifndef AVUTIL_X86_ASM_H
+#define AVUTIL_X86_ASM_H
+
+#include <stdint.h>
+#include "config.h"
+
+typedef struct xmm_reg { uint64_t a, b; } xmm_reg;
+
+#if ARCH_X86_64
+#    define OPSIZE "q"
+#    define REG_a "rax"
+#    define REG_b "rbx"
+#    define REG_c "rcx"
+#    define REG_d "rdx"
+#    define REG_D "rdi"
+#    define REG_S "rsi"
+#    define PTR_SIZE "8"
+typedef int64_t x86_reg;
+
+#    define REG_SP "rsp"
+#    define REG_BP "rbp"
+#    define REGBP   rbp
+#    define REGa    rax
+#    define REGb    rbx
+#    define REGc    rcx
+#    define REGd    rdx
+#    define REGSP   rsp
+
+#elif ARCH_X86_32
+
+#    define OPSIZE "l"
+#    define REG_a "eax"
+#    define REG_b "ebx"
+#    define REG_c "ecx"
+#    define REG_d "edx"
+#    define REG_D "edi"
+#    define REG_S "esi"
+#    define PTR_SIZE "4"
+typedef int32_t x86_reg;
+
+#    define REG_SP "esp"
+#    define REG_BP "ebp"
+#    define REGBP   ebp
+#    define REGa    eax
+#    define REGb    ebx
+#    define REGc    ecx
+#    define REGd    edx
+#    define REGSP   esp
+#else
+typedef int x86_reg;
+#endif
+
+#define HAVE_7REGS (ARCH_X86_64 || (HAVE_EBX_AVAILABLE && HAVE_EBP_AVAILABLE))
+#define HAVE_6REGS (ARCH_X86_64 || (HAVE_EBX_AVAILABLE || HAVE_EBP_AVAILABLE))
+
+#if ARCH_X86_64 && defined(PIC)
+#    define BROKEN_RELOCATIONS 1
+#endif
+
+/*
+ * If gcc is not set to support sse (-msse) it will not accept xmm registers
+ * in the clobber list for inline asm. XMM_CLOBBERS takes a list of xmm
+ * registers to be marked as clobbered and evaluates to nothing if they are
+ * not supported, or to the list itself if they are supported. Since a clobber
+ * list may not be empty, XMM_CLOBBERS_ONLY should be used if the xmm
+ * registers are the only in the clobber list.
+ * For example a list with "eax" and "xmm0" as clobbers should become:
+ * : XMM_CLOBBERS("xmm0",) "eax"
+ * and a list with only "xmm0" should become:
+ * XMM_CLOBBERS_ONLY("xmm0")
+ */
+#if HAVE_XMM_CLOBBERS
+#    define XMM_CLOBBERS(...)        __VA_ARGS__
+#    define XMM_CLOBBERS_ONLY(...) : __VA_ARGS__
+#else
+#    define XMM_CLOBBERS(...)
+#    define XMM_CLOBBERS_ONLY(...)
+#endif
+
+/* Use to export labels from asm. */
+#define LABEL_MANGLE(a) EXTERN_PREFIX #a
+
+// Use rip-relative addressing if compiling PIC code on x86-64.
+#if ARCH_X86_64 && defined(PIC)
+#    define LOCAL_MANGLE(a) #a "(%%rip)"
+#else
+#    define LOCAL_MANGLE(a) #a
+#endif
+
+#define MANGLE(a) EXTERN_PREFIX LOCAL_MANGLE(a)
+
+#endif /* AVUTIL_X86_ASM_H */
diff --git a/media/libav/libavutil/x86/bswap.h b/media/libav/libavutil/x86/bswap.h
new file mode 100644
index 000000000..c73be9af8
--- /dev/null
+++ b/media/libav/libavutil/x86/bswap.h
@@ -0,0 +1,61 @@
+/*
+ * 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
+ */
+
+/**
+ * @file
+ * byte swapping routines
+ */
+
+#ifndef AVUTIL_X86_BSWAP_H
+#define AVUTIL_X86_BSWAP_H
+
+#include <stdint.h>
+#include "config.h"
+#include "libavutil/attributes.h"
+
+#if HAVE_INLINE_ASM
+
+#if !AV_GCC_VERSION_AT_LEAST(4,1)
+#define av_bswap16 av_bswap16
+static av_always_inline av_const unsigned av_bswap16(unsigned x)
+{
+    __asm__("rorw $8, %w0" : "+r"(x));
+    return x;
+}
+#endif /* !AV_GCC_VERSION_AT_LEAST(4,1) */
+
+#if !AV_GCC_VERSION_AT_LEAST(4,5)
+#define av_bswap32 av_bswap32
+static av_always_inline av_const uint32_t av_bswap32(uint32_t x)
+{
+    __asm__("bswap   %0" : "+r" (x));
+    return x;
+}
+
+#if ARCH_X86_64
+#define av_bswap64 av_bswap64
+static inline uint64_t av_const av_bswap64(uint64_t x)
+{
+    __asm__("bswap  %0": "=r" (x) : "0" (x));
+    return x;
+}
+#endif
+#endif /* !AV_GCC_VERSION_AT_LEAST(4,5) */
+
+#endif /* HAVE_INLINE_ASM */
+#endif /* AVUTIL_X86_BSWAP_H */
diff --git a/media/libav/libavutil/x86/cpu.c b/media/libav/libavutil/x86/cpu.c
new file mode 100644
index 000000000..8be6d9474
--- /dev/null
+++ b/media/libav/libavutil/x86/cpu.c
@@ -0,0 +1,218 @@
+/*
+ * CPU detection code, extracted from mmx.h
+ * (c)1997-99 by H. Dietz and R. Fisher
+ * Converted to C and improved by Fabrice Bellard.
+ *
+ * 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 <stdlib.h>
+#include <string.h>
+
+#include "libavutil/x86/asm.h"
+#include "libavutil/x86/cpu.h"
+#include "libavutil/cpu.h"
+#include "libavutil/cpu_internal.h"
+
+#if HAVE_YASM
+
+#define cpuid(index, eax, ebx, ecx, edx)        \
+    ff_cpu_cpuid(index, &eax, &ebx, &ecx, &edx)
+
+#define xgetbv(index, eax, edx)                 \
+    ff_cpu_xgetbv(index, &eax, &edx)
+
+#elif HAVE_INLINE_ASM
+
+/* ebx saving is necessary for PIC. gcc seems unable to see it alone */
+#define cpuid(index, eax, ebx, ecx, edx)                        \
+    __asm__ volatile (                                          \
+        "mov    %%"REG_b", %%"REG_S" \n\t"                      \
+        "cpuid                       \n\t"                      \
+        "xchg   %%"REG_b", %%"REG_S                             \
+        : "=a" (eax), "=S" (ebx), "=c" (ecx), "=d" (edx)        \
+        : "0" (index))
+
+#define xgetbv(index, eax, edx)                                 \
+    __asm__ (".byte 0x0f, 0x01, 0xd0" : "=a"(eax), "=d"(edx) : "c" (index))
+
+#define get_eflags(x)                           \
+    __asm__ volatile ("pushfl     \n"           \
+                      "pop    %0  \n"           \
+                      : "=r"(x))
+
+#define set_eflags(x)                           \
+    __asm__ volatile ("push    %0 \n"           \
+                      "popfl      \n"           \
+                      :: "r"(x))
+
+#endif /* HAVE_INLINE_ASM */
+
+#if ARCH_X86_64
+
+#define cpuid_test() 1
+
+#elif HAVE_YASM
+
+#define cpuid_test ff_cpu_cpuid_test
+
+#elif HAVE_INLINE_ASM
+
+static int cpuid_test(void)
+{
+    x86_reg a, c;
+
+    /* Check if CPUID is supported by attempting to toggle the ID bit in
+     * the EFLAGS register. */
+    get_eflags(a);
+    set_eflags(a ^ 0x200000);
+    get_eflags(c);
+
+    return a != c;
+}
+#endif
+
+/* Function to test if multimedia instructions are supported...  */
+int ff_get_cpu_flags_x86(void)
+{
+    int rval = 0;
+
+#ifdef cpuid
+
+    int eax, ebx, ecx, edx;
+    int max_std_level, max_ext_level, std_caps = 0, ext_caps = 0;
+    int family = 0, model = 0;
+    union { int i[3]; char c[12]; } vendor;
+
+    if (!cpuid_test())
+        return 0; /* CPUID not supported */
+
+    cpuid(0, max_std_level, vendor.i[0], vendor.i[2], vendor.i[1]);
+
+    if (max_std_level >= 1) {
+        cpuid(1, eax, ebx, ecx, std_caps);
+        family = ((eax >> 8) & 0xf) + ((eax >> 20) & 0xff);
+        model  = ((eax >> 4) & 0xf) + ((eax >> 12) & 0xf0);
+        if (std_caps & (1 << 15))
+            rval |= AV_CPU_FLAG_CMOV;
+        if (std_caps & (1 << 23))
+            rval |= AV_CPU_FLAG_MMX;
+        if (std_caps & (1 << 25))
+            rval |= AV_CPU_FLAG_MMXEXT;
+#if HAVE_SSE
+        if (std_caps & (1 << 25))
+            rval |= AV_CPU_FLAG_SSE;
+        if (std_caps & (1 << 26))
+            rval |= AV_CPU_FLAG_SSE2;
+        if (ecx & 1)
+            rval |= AV_CPU_FLAG_SSE3;
+        if (ecx & 0x00000200 )
+            rval |= AV_CPU_FLAG_SSSE3;
+        if (ecx & 0x00080000 )
+            rval |= AV_CPU_FLAG_SSE4;
+        if (ecx & 0x00100000 )
+            rval |= AV_CPU_FLAG_SSE42;
+#if HAVE_AVX
+        /* Check OXSAVE and AVX bits */
+        if ((ecx & 0x18000000) == 0x18000000) {
+            /* Check for OS support */
+            xgetbv(0, eax, edx);
+            if ((eax & 0x6) == 0x6) {
+                rval |= AV_CPU_FLAG_AVX;
+                if (ecx & 0x00001000)
+                    rval |= AV_CPU_FLAG_FMA3;
+            }
+        }
+#endif /* HAVE_AVX */
+#endif /* HAVE_SSE */
+    }
+    if (max_std_level >= 7) {
+        cpuid(7, eax, ebx, ecx, edx);
+#if HAVE_AVX2
+        if (ebx & 0x00000020)
+            rval |= AV_CPU_FLAG_AVX2;
+#endif /* HAVE_AVX2 */
+        /* BMI1/2 don't need OS support */
+        if (ebx & 0x00000008) {
+            rval |= AV_CPU_FLAG_BMI1;
+            if (ebx & 0x00000100)
+                rval |= AV_CPU_FLAG_BMI2;
+        }
+    }
+
+    cpuid(0x80000000, max_ext_level, ebx, ecx, edx);
+
+    if (max_ext_level >= 0x80000001) {
+        cpuid(0x80000001, eax, ebx, ecx, ext_caps);
+        if (ext_caps & (1U << 31))
+            rval |= AV_CPU_FLAG_3DNOW;
+        if (ext_caps & (1 << 30))
+            rval |= AV_CPU_FLAG_3DNOWEXT;
+        if (ext_caps & (1 << 23))
+            rval |= AV_CPU_FLAG_MMX;
+        if (ext_caps & (1 << 22))
+            rval |= AV_CPU_FLAG_MMXEXT;
+
+        /* Allow for selectively disabling SSE2 functions on AMD processors
+           with SSE2 support but not SSE4a. This includes Athlon64, some
+           Opteron, and some Sempron processors. MMX, SSE, or 3DNow! are faster
+           than SSE2 often enough to utilize this special-case flag.
+           AV_CPU_FLAG_SSE2 and AV_CPU_FLAG_SSE2SLOW are both set in this case
+           so that SSE2 is used unless explicitly disabled by checking
+           AV_CPU_FLAG_SSE2SLOW. */
+        if (!strncmp(vendor.c, "AuthenticAMD", 12) &&
+            rval & AV_CPU_FLAG_SSE2 && !(ecx & 0x00000040)) {
+            rval |= AV_CPU_FLAG_SSE2SLOW;
+        }
+
+        /* XOP and FMA4 use the AVX instruction coding scheme, so they can't be
+         * used unless the OS has AVX support. */
+        if (rval & AV_CPU_FLAG_AVX) {
+            if (ecx & 0x00000800)
+                rval |= AV_CPU_FLAG_XOP;
+            if (ecx & 0x00010000)
+                rval |= AV_CPU_FLAG_FMA4;
+        }
+    }
+
+    if (!strncmp(vendor.c, "GenuineIntel", 12)) {
+        if (family == 6 && (model == 9 || model == 13 || model == 14)) {
+            /* 6/9 (pentium-m "banias"), 6/13 (pentium-m "dothan"), and
+             * 6/14 (core1 "yonah") theoretically support sse2, but it's
+             * usually slower than mmx, so let's just pretend they don't.
+             * AV_CPU_FLAG_SSE2 is disabled and AV_CPU_FLAG_SSE2SLOW is
+             * enabled so that SSE2 is not used unless explicitly enabled
+             * by checking AV_CPU_FLAG_SSE2SLOW. The same situation
+             * applies for AV_CPU_FLAG_SSE3 and AV_CPU_FLAG_SSE3SLOW. */
+            if (rval & AV_CPU_FLAG_SSE2)
+                rval ^= AV_CPU_FLAG_SSE2SLOW | AV_CPU_FLAG_SSE2;
+            if (rval & AV_CPU_FLAG_SSE3)
+                rval ^= AV_CPU_FLAG_SSE3SLOW | AV_CPU_FLAG_SSE3;
+        }
+        /* The Atom processor has SSSE3 support, which is useful in many cases,
+         * but sometimes the SSSE3 version is slower than the SSE2 equivalent
+         * on the Atom, but is generally faster on other processors supporting
+         * SSSE3. This flag allows for selectively disabling certain SSSE3
+         * functions on the Atom. */
+        if (family == 6 && model == 28)
+            rval |= AV_CPU_FLAG_ATOM;
+    }
+
+#endif /* cpuid */
+
+    return rval;
+}
diff --git a/media/libav/libavutil/x86/cpu.h b/media/libav/libavutil/x86/cpu.h
new file mode 100644
index 000000000..50da30e38
--- /dev/null
+++ b/media/libav/libavutil/x86/cpu.h
@@ -0,0 +1,81 @@
+/*
+ * 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
+ */
+
+#ifndef AVUTIL_X86_CPU_H
+#define AVUTIL_X86_CPU_H
+
+#include "config.h"
+#include "libavutil/cpu.h"
+#include "libavutil/cpu_internal.h"
+
+#define AV_CPU_FLAG_AMD3DNOW    AV_CPU_FLAG_3DNOW
+#define AV_CPU_FLAG_AMD3DNOWEXT AV_CPU_FLAG_3DNOWEXT
+
+#define X86_AMD3DNOW(flags)         CPUEXT(flags, AMD3DNOW)
+#define X86_AMD3DNOWEXT(flags)      CPUEXT(flags, AMD3DNOWEXT)
+#define X86_MMX(flags)              CPUEXT(flags, MMX)
+#define X86_MMXEXT(flags)           CPUEXT(flags, MMXEXT)
+#define X86_SSE(flags)              CPUEXT(flags, SSE)
+#define X86_SSE2(flags)             CPUEXT(flags, SSE2)
+#define X86_SSE3(flags)             CPUEXT(flags, SSE3)
+#define X86_SSSE3(flags)            CPUEXT(flags, SSSE3)
+#define X86_SSE4(flags)             CPUEXT(flags, SSE4)
+#define X86_SSE42(flags)            CPUEXT(flags, SSE42)
+#define X86_AVX(flags)              CPUEXT(flags, AVX)
+#define X86_XOP(flags)              CPUEXT(flags, XOP)
+#define X86_FMA3(flags)             CPUEXT(flags, FMA3)
+#define X86_FMA4(flags)             CPUEXT(flags, FMA4)
+#define X86_AVX2(flags)             CPUEXT(flags, AVX2)
+
+#define EXTERNAL_AMD3DNOW(flags)    CPUEXT_SUFFIX(flags, _EXTERNAL, AMD3DNOW)
+#define EXTERNAL_AMD3DNOWEXT(flags) CPUEXT_SUFFIX(flags, _EXTERNAL, AMD3DNOWEXT)
+#define EXTERNAL_MMX(flags)         CPUEXT_SUFFIX(flags, _EXTERNAL, MMX)
+#define EXTERNAL_MMXEXT(flags)      CPUEXT_SUFFIX(flags, _EXTERNAL, MMXEXT)
+#define EXTERNAL_SSE(flags)         CPUEXT_SUFFIX(flags, _EXTERNAL, SSE)
+#define EXTERNAL_SSE2(flags)        CPUEXT_SUFFIX(flags, _EXTERNAL, SSE2)
+#define EXTERNAL_SSE3(flags)        CPUEXT_SUFFIX(flags, _EXTERNAL, SSE3)
+#define EXTERNAL_SSSE3(flags)       CPUEXT_SUFFIX(flags, _EXTERNAL, SSSE3)
+#define EXTERNAL_SSE4(flags)        CPUEXT_SUFFIX(flags, _EXTERNAL, SSE4)
+#define EXTERNAL_SSE42(flags)       CPUEXT_SUFFIX(flags, _EXTERNAL, SSE42)
+#define EXTERNAL_AVX(flags)         CPUEXT_SUFFIX(flags, _EXTERNAL, AVX)
+#define EXTERNAL_XOP(flags)         CPUEXT_SUFFIX(flags, _EXTERNAL, XOP)
+#define EXTERNAL_FMA3(flags)        CPUEXT_SUFFIX(flags, _EXTERNAL, FMA3)
+#define EXTERNAL_FMA4(flags)        CPUEXT_SUFFIX(flags, _EXTERNAL, FMA4)
+#define EXTERNAL_AVX2(flags)        CPUEXT_SUFFIX(flags, _EXTERNAL, AVX2)
+
+#define INLINE_AMD3DNOW(flags)      CPUEXT_SUFFIX(flags, _INLINE, AMD3DNOW)
+#define INLINE_AMD3DNOWEXT(flags)   CPUEXT_SUFFIX(flags, _INLINE, AMD3DNOWEXT)
+#define INLINE_MMX(flags)           CPUEXT_SUFFIX(flags, _INLINE, MMX)
+#define INLINE_MMXEXT(flags)        CPUEXT_SUFFIX(flags, _INLINE, MMXEXT)
+#define INLINE_SSE(flags)           CPUEXT_SUFFIX(flags, _INLINE, SSE)
+#define INLINE_SSE2(flags)          CPUEXT_SUFFIX(flags, _INLINE, SSE2)
+#define INLINE_SSE3(flags)          CPUEXT_SUFFIX(flags, _INLINE, SSE3)
+#define INLINE_SSSE3(flags)         CPUEXT_SUFFIX(flags, _INLINE, SSSE3)
+#define INLINE_SSE4(flags)          CPUEXT_SUFFIX(flags, _INLINE, SSE4)
+#define INLINE_SSE42(flags)         CPUEXT_SUFFIX(flags, _INLINE, SSE42)
+#define INLINE_AVX(flags)           CPUEXT_SUFFIX(flags, _INLINE, AVX)
+#define INLINE_XOP(flags)           CPUEXT_SUFFIX(flags, _INLINE, XOP)
+#define INLINE_FMA3(flags)          CPUEXT_SUFFIX(flags, _INLINE, FMA3)
+#define INLINE_FMA4(flags)          CPUEXT_SUFFIX(flags, _INLINE, FMA4)
+#define INLINE_AVX2(flags)          CPUEXT_SUFFIX(flags, _INLINE, AVX2)
+
+void ff_cpu_cpuid(int index, int *eax, int *ebx, int *ecx, int *edx);
+void ff_cpu_xgetbv(int op, int *eax, int *edx);
+int  ff_cpu_cpuid_test(void);
+
+#endif /* AVUTIL_X86_CPU_H */
diff --git a/media/libav/libavutil/x86/cpuid.asm b/media/libav/libavutil/x86/cpuid.asm
new file mode 100644
index 000000000..1cb8e94ea
--- /dev/null
+++ b/media/libav/libavutil/x86/cpuid.asm
@@ -0,0 +1,91 @@
+;*****************************************************************************
+;* Copyright (C) 2005-2010 x264 project
+;*
+;* Authors: Loren Merritt <lorenm@u.washington.edu>
+;*          Fiona Glaser <fiona@x264.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 "x86util.asm"
+
+SECTION .text
+
+;-----------------------------------------------------------------------------
+; void ff_cpu_cpuid(int index, int *eax, int *ebx, int *ecx, int *edx)
+;-----------------------------------------------------------------------------
+cglobal cpu_cpuid, 5,7
+    push rbx
+    push  r4
+    push  r3
+    push  r2
+    push  r1
+    mov  eax, r0d
+    xor  ecx, ecx
+    cpuid
+    pop   r4
+    mov [r4], eax
+    pop   r4
+    mov [r4], ebx
+    pop   r4
+    mov [r4], ecx
+    pop   r4
+    mov [r4], edx
+    pop  rbx
+    RET
+
+;-----------------------------------------------------------------------------
+; void ff_cpu_xgetbv(int op, int *eax, int *edx)
+;-----------------------------------------------------------------------------
+cglobal cpu_xgetbv, 3,7
+    push  r2
+    push  r1
+    mov  ecx, r0d
+    xgetbv
+    pop   r4
+    mov [r4], eax
+    pop   r4
+    mov [r4], edx
+    RET
+
+%if ARCH_X86_64 == 0
+;-----------------------------------------------------------------------------
+; int ff_cpu_cpuid_test(void)
+; return 0 if unsupported
+;-----------------------------------------------------------------------------
+cglobal cpu_cpuid_test
+    pushfd
+    push    ebx
+    push    ebp
+    push    esi
+    push    edi
+    pushfd
+    pop     eax
+    mov     ebx, eax
+    xor     eax, 0x200000
+    push    eax
+    popfd
+    pushfd
+    pop     eax
+    xor     eax, ebx
+    pop     edi
+    pop     esi
+    pop     ebp
+    pop     ebx
+    popfd
+    ret
+%endif
diff --git a/media/libav/libavutil/x86/cpuid.s b/media/libav/libavutil/x86/cpuid.s
new file mode 100644
index 000000000..1cb8e94ea
--- /dev/null
+++ b/media/libav/libavutil/x86/cpuid.s
@@ -0,0 +1,91 @@
+;*****************************************************************************
+;* Copyright (C) 2005-2010 x264 project
+;*
+;* Authors: Loren Merritt <lorenm@u.washington.edu>
+;*          Fiona Glaser <fiona@x264.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 "x86util.asm"
+
+SECTION .text
+
+;-----------------------------------------------------------------------------
+; void ff_cpu_cpuid(int index, int *eax, int *ebx, int *ecx, int *edx)
+;-----------------------------------------------------------------------------
+cglobal cpu_cpuid, 5,7
+    push rbx
+    push  r4
+    push  r3
+    push  r2
+    push  r1
+    mov  eax, r0d
+    xor  ecx, ecx
+    cpuid
+    pop   r4
+    mov [r4], eax
+    pop   r4
+    mov [r4], ebx
+    pop   r4
+    mov [r4], ecx
+    pop   r4
+    mov [r4], edx
+    pop  rbx
+    RET
+
+;-----------------------------------------------------------------------------
+; void ff_cpu_xgetbv(int op, int *eax, int *edx)
+;-----------------------------------------------------------------------------
+cglobal cpu_xgetbv, 3,7
+    push  r2
+    push  r1
+    mov  ecx, r0d
+    xgetbv
+    pop   r4
+    mov [r4], eax
+    pop   r4
+    mov [r4], edx
+    RET
+
+%if ARCH_X86_64 == 0
+;-----------------------------------------------------------------------------
+; int ff_cpu_cpuid_test(void)
+; return 0 if unsupported
+;-----------------------------------------------------------------------------
+cglobal cpu_cpuid_test
+    pushfd
+    push    ebx
+    push    ebp
+    push    esi
+    push    edi
+    pushfd
+    pop     eax
+    mov     ebx, eax
+    xor     eax, 0x200000
+    push    eax
+    popfd
+    pushfd
+    pop     eax
+    xor     eax, ebx
+    pop     edi
+    pop     esi
+    pop     ebp
+    pop     ebx
+    popfd
+    ret
+%endif
diff --git a/media/libav/libavutil/x86/emms.asm b/media/libav/libavutil/x86/emms.asm
new file mode 100644
index 000000000..a6851acc9
--- /dev/null
+++ b/media/libav/libavutil/x86/emms.asm
@@ -0,0 +1,30 @@
+;*****************************************************************************
+;* Copyright (C) 2013 Martin Storsjo
+;*
+;* 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 "x86util.asm"
+
+SECTION .text
+
+;-----------------------------------------------------------------------------
+; void avpriv_emms_yasm(void)
+;-----------------------------------------------------------------------------
+cvisible emms_yasm, 0, 0
+    emms
+    RET
diff --git a/media/libav/libavutil/x86/emms.h b/media/libav/libavutil/x86/emms.h
new file mode 100644
index 000000000..2ed9e5d09
--- /dev/null
+++ b/media/libav/libavutil/x86/emms.h
@@ -0,0 +1,45 @@
+/*
+ * 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
+ */
+
+#ifndef AVUTIL_X86_EMMS_H
+#define AVUTIL_X86_EMMS_H
+
+#include "config.h"
+#include "libavutil/attributes.h"
+
+void avpriv_emms_yasm(void);
+
+#if HAVE_MMX_INLINE
+#   define emms_c emms_c
+/**
+ * Empty mmx state.
+ * this must be called between any dsp function and float/double code.
+ * for example sin(); dsp->idct_put(); emms_c(); cos()
+ */
+static av_always_inline void emms_c(void)
+{
+    __asm__ volatile ("emms" ::: "memory");
+}
+#elif HAVE_MMX && HAVE_MM_EMPTY
+#   include <mmintrin.h>
+#   define emms_c _mm_empty
+#elif HAVE_MMX_EXTERNAL
+#   define emms_c avpriv_emms_yasm
+#endif /* HAVE_MMX_INLINE */
+
+#endif /* AVUTIL_X86_EMMS_H */
diff --git a/media/libav/libavutil/x86/intreadwrite.h b/media/libav/libavutil/x86/intreadwrite.h
new file mode 100644
index 000000000..635096e56
--- /dev/null
+++ b/media/libav/libavutil/x86/intreadwrite.h
@@ -0,0 +1,97 @@
+/*
+ * Copyright (c) 2010 Alexander Strange <astrange@ithinksw.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
+ */
+
+#ifndef AVUTIL_X86_INTREADWRITE_H
+#define AVUTIL_X86_INTREADWRITE_H
+
+#include <stdint.h>
+#include "config.h"
+#include "libavutil/attributes.h"
+
+#if HAVE_MMX
+
+#if !HAVE_FAST_64BIT && defined(__MMX__)
+
+#define AV_COPY64 AV_COPY64
+static av_always_inline void AV_COPY64(void *d, const void *s)
+{
+    __asm__("movq   %1, %%mm0  \n\t"
+            "movq   %%mm0, %0  \n\t"
+            : "=m"(*(uint64_t*)d)
+            : "m" (*(const uint64_t*)s)
+            : "mm0");
+}
+
+#define AV_SWAP64 AV_SWAP64
+static av_always_inline void AV_SWAP64(void *a, void *b)
+{
+    __asm__("movq   %1, %%mm0  \n\t"
+            "movq   %0, %%mm1  \n\t"
+            "movq   %%mm0, %0  \n\t"
+            "movq   %%mm1, %1  \n\t"
+            : "+m"(*(uint64_t*)a), "+m"(*(uint64_t*)b)
+            ::"mm0", "mm1");
+}
+
+#define AV_ZERO64 AV_ZERO64
+static av_always_inline void AV_ZERO64(void *d)
+{
+    __asm__("pxor %%mm0, %%mm0  \n\t"
+            "movq %%mm0, %0     \n\t"
+            : "=m"(*(uint64_t*)d)
+            :: "mm0");
+}
+
+#endif /* !HAVE_FAST_64BIT && defined(__MMX__) */
+
+#ifdef __SSE__
+
+#define AV_COPY128 AV_COPY128
+static av_always_inline void AV_COPY128(void *d, const void *s)
+{
+    struct v {uint64_t v[2];};
+
+    __asm__("movaps   %1, %%xmm0  \n\t"
+            "movaps   %%xmm0, %0  \n\t"
+            : "=m"(*(struct v*)d)
+            : "m" (*(const struct v*)s)
+            : "xmm0");
+}
+
+#endif /* __SSE__ */
+
+#ifdef __SSE2__
+
+#define AV_ZERO128 AV_ZERO128
+static av_always_inline void AV_ZERO128(void *d)
+{
+    struct v {uint64_t v[2];};
+
+    __asm__("pxor %%xmm0, %%xmm0  \n\t"
+            "movdqa   %%xmm0, %0  \n\t"
+            : "=m"(*(struct v*)d)
+            :: "xmm0");
+}
+
+#endif /* __SSE2__ */
+
+#endif /* HAVE_MMX */
+
+#endif /* AVUTIL_X86_INTREADWRITE_H */
diff --git a/media/libav/libavutil/x86/moz.build b/media/libav/libavutil/x86/moz.build
new file mode 100644
index 000000000..978905d01
--- /dev/null
+++ b/media/libav/libavutil/x86/moz.build
@@ -0,0 +1,11 @@
+# -*- Mode: python; indent-tabs-mode: nil; tab-width: 40 -*-
+# vim: set filetype=python:
+# This Source Code Form is subject to the terms of the Mozilla Public
+# License, v. 2.0. If a copy of the MPL was not distributed with this
+# file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+SOURCES += [
+    'cpu.c',
+]
+
+include('/media/libav/libavcommon.mozbuild')
diff --git a/media/libav/libavutil/x86/timer.h b/media/libav/libavutil/x86/timer.h
new file mode 100644
index 000000000..cdd67ddbb
--- /dev/null
+++ b/media/libav/libavutil/x86/timer.h
@@ -0,0 +1,44 @@
+/*
+ * copyright (c) 2006 Michael Niedermayer <michaelni@gmx.at>
+ *
+ * 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
+ */
+
+#ifndef AVUTIL_X86_TIMER_H
+#define AVUTIL_X86_TIMER_H
+
+#include <stdint.h>
+
+#if HAVE_INLINE_ASM
+
+#define AV_READ_TIME read_time
+
+static inline uint64_t read_time(void)
+{
+    uint32_t a, d;
+    __asm__ volatile("rdtsc" : "=a" (a), "=d" (d));
+    return ((uint64_t)d << 32) + a;
+}
+
+#elif HAVE_RDTSC
+
+#include <intrin.h>
+#define AV_READ_TIME __rdtsc
+
+#endif /* HAVE_INLINE_ASM */
+
+#endif /* AVUTIL_X86_TIMER_H */
diff --git a/media/libav/libavutil/x86/x86inc.asm b/media/libav/libavutil/x86/x86inc.asm
new file mode 100644
index 000000000..2d4d1d3ca
--- /dev/null
+++ b/media/libav/libavutil/x86/x86inc.asm
@@ -0,0 +1,1472 @@
+;*****************************************************************************
+;* x86inc.asm: x264asm abstraction layer
+;*****************************************************************************
+;* Copyright (C) 2005-2013 x264 project
+;*
+;* Authors: Loren Merritt <lorenm@u.washington.edu>
+;*          Anton Mitrofanov <BugMaster@narod.ru>
+;*          Fiona Glaser <fiona@x264.com>
+;*          Henrik Gramner <henrik@gramner.com>
+;*
+;* Permission to use, copy, modify, and/or distribute this software for any
+;* purpose with or without fee is hereby granted, provided that the above
+;* copyright notice and this permission notice appear in all copies.
+;*
+;* THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+;* WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+;* MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+;* ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+;* WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+;* ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+;* OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+;*****************************************************************************
+
+; This is a header file for the x264ASM assembly language, which uses
+; NASM/YASM syntax combined with a large number of macros to provide easy
+; abstraction between different calling conventions (x86_32, win64, linux64).
+; It also has various other useful features to simplify writing the kind of
+; DSP functions that are most often used in x264.
+
+; Unlike the rest of x264, this file is available under an ISC license, as it
+; has significant usefulness outside of x264 and we want it to be available
+; to the largest audience possible.  Of course, if you modify it for your own
+; purposes to add a new feature, we strongly encourage contributing a patch
+; as this feature might be useful for others as well.  Send patches or ideas
+; to x264-devel@videolan.org .
+
+%ifndef private_prefix
+    %define private_prefix x264
+%endif
+
+%ifndef public_prefix
+    %define public_prefix private_prefix
+%endif
+
+%define WIN64  0
+%define UNIX64 0
+%if ARCH_X86_64
+    %ifidn __OUTPUT_FORMAT__,win32
+        %define WIN64  1
+    %elifidn __OUTPUT_FORMAT__,win64
+        %define WIN64  1
+    %elifidn __OUTPUT_FORMAT__,x64
+        %define WIN64  1
+    %else
+        %define UNIX64 1
+    %endif
+%endif
+
+%ifdef PREFIX
+    %define mangle(x) _ %+ x
+%else
+    %define mangle(x) x
+%endif
+
+; aout does not support align=
+; NOTE: This section is out of sync with x264, in order to
+; keep supporting OS/2.
+%macro SECTION_RODATA 0-1 16
+    %ifidn __OUTPUT_FORMAT__,aout
+        section .text
+    %else
+        SECTION .rodata align=%1
+    %endif
+%endmacro
+
+%macro SECTION_TEXT 0-1 16
+    %ifidn __OUTPUT_FORMAT__,aout
+        SECTION .text
+    %else
+        SECTION .text align=%1
+    %endif
+%endmacro
+
+%if WIN64
+    %define PIC
+%elif ARCH_X86_64 == 0
+; x86_32 doesn't require PIC.
+; Some distros prefer shared objects to be PIC, but nothing breaks if
+; the code contains a few textrels, so we'll skip that complexity.
+    %undef PIC
+%endif
+%ifdef PIC
+    default rel
+%endif
+
+%macro CPUNOP 1
+    %if HAVE_CPUNOP
+        CPU %1
+    %endif
+%endmacro
+
+; Always use long nops (reduces 0x90 spam in disassembly on x86_32)
+CPUNOP amdnop
+
+; Macros to eliminate most code duplication between x86_32 and x86_64:
+; Currently this works only for leaf functions which load all their arguments
+; into registers at the start, and make no other use of the stack. Luckily that
+; covers most of x264's asm.
+
+; PROLOGUE:
+; %1 = number of arguments. loads them from stack if needed.
+; %2 = number of registers used. pushes callee-saved regs if needed.
+; %3 = number of xmm registers used. pushes callee-saved xmm regs if needed.
+; %4 = (optional) stack size to be allocated. If not aligned (x86-32 ICC 10.x,
+;      MSVC or YMM), the stack will be manually aligned (to 16 or 32 bytes),
+;      and an extra register will be allocated to hold the original stack
+;      pointer (to not invalidate r0m etc.). To prevent the use of an extra
+;      register as stack pointer, request a negative stack size.
+; %4+/%5+ = list of names to define to registers
+; PROLOGUE can also be invoked by adding the same options to cglobal
+
+; e.g.
+; cglobal foo, 2,3,0, dst, src, tmp
+; declares a function (foo), taking two args (dst and src) and one local variable (tmp)
+
+; TODO Some functions can use some args directly from the stack. If they're the
+; last args then you can just not declare them, but if they're in the middle
+; we need more flexible macro.
+
+; RET:
+; Pops anything that was pushed by PROLOGUE, and returns.
+
+; REP_RET:
+; Use this instead of RET if it's a branch target.
+
+; registers:
+; rN and rNq are the native-size register holding function argument N
+; rNd, rNw, rNb are dword, word, and byte size
+; rNh is the high 8 bits of the word size
+; rNm is the original location of arg N (a register or on the stack), dword
+; rNmp is native size
+
+%macro DECLARE_REG 2-3
+    %define r%1q %2
+    %define r%1d %2d
+    %define r%1w %2w
+    %define r%1b %2b
+    %define r%1h %2h
+    %define %2q %2
+    %if %0 == 2
+        %define r%1m  %2d
+        %define r%1mp %2
+    %elif ARCH_X86_64 ; memory
+        %define r%1m [rstk + stack_offset + %3]
+        %define r%1mp qword r %+ %1 %+ m
+    %else
+        %define r%1m [rstk + stack_offset + %3]
+        %define r%1mp dword r %+ %1 %+ m
+    %endif
+    %define r%1  %2
+%endmacro
+
+%macro DECLARE_REG_SIZE 3
+    %define r%1q r%1
+    %define e%1q r%1
+    %define r%1d e%1
+    %define e%1d e%1
+    %define r%1w %1
+    %define e%1w %1
+    %define r%1h %3
+    %define e%1h %3
+    %define r%1b %2
+    %define e%1b %2
+%if ARCH_X86_64 == 0
+    %define r%1  e%1
+%endif
+%endmacro
+
+DECLARE_REG_SIZE ax, al, ah
+DECLARE_REG_SIZE bx, bl, bh
+DECLARE_REG_SIZE cx, cl, ch
+DECLARE_REG_SIZE dx, dl, dh
+DECLARE_REG_SIZE si, sil, null
+DECLARE_REG_SIZE di, dil, null
+DECLARE_REG_SIZE bp, bpl, null
+
+; t# defines for when per-arch register allocation is more complex than just function arguments
+
+%macro DECLARE_REG_TMP 1-*
+    %assign %%i 0
+    %rep %0
+        CAT_XDEFINE t, %%i, r%1
+        %assign %%i %%i+1
+        %rotate 1
+    %endrep
+%endmacro
+
+%macro DECLARE_REG_TMP_SIZE 0-*
+    %rep %0
+        %define t%1q t%1 %+ q
+        %define t%1d t%1 %+ d
+        %define t%1w t%1 %+ w
+        %define t%1h t%1 %+ h
+        %define t%1b t%1 %+ b
+        %rotate 1
+    %endrep
+%endmacro
+
+DECLARE_REG_TMP_SIZE 0,1,2,3,4,5,6,7,8,9,10,11,12,13,14
+
+%if ARCH_X86_64
+    %define gprsize 8
+%else
+    %define gprsize 4
+%endif
+
+%macro PUSH 1
+    push %1
+    %ifidn rstk, rsp
+        %assign stack_offset stack_offset+gprsize
+    %endif
+%endmacro
+
+%macro POP 1
+    pop %1
+    %ifidn rstk, rsp
+        %assign stack_offset stack_offset-gprsize
+    %endif
+%endmacro
+
+%macro PUSH_IF_USED 1-*
+    %rep %0
+        %if %1 < regs_used
+            PUSH r%1
+        %endif
+        %rotate 1
+    %endrep
+%endmacro
+
+%macro POP_IF_USED 1-*
+    %rep %0
+        %if %1 < regs_used
+            pop r%1
+        %endif
+        %rotate 1
+    %endrep
+%endmacro
+
+%macro LOAD_IF_USED 1-*
+    %rep %0
+        %if %1 < num_args
+            mov r%1, r %+ %1 %+ mp
+        %endif
+        %rotate 1
+    %endrep
+%endmacro
+
+%macro SUB 2
+    sub %1, %2
+    %ifidn %1, rstk
+        %assign stack_offset stack_offset+(%2)
+    %endif
+%endmacro
+
+%macro ADD 2
+    add %1, %2
+    %ifidn %1, rstk
+        %assign stack_offset stack_offset-(%2)
+    %endif
+%endmacro
+
+%macro movifnidn 2
+    %ifnidn %1, %2
+        mov %1, %2
+    %endif
+%endmacro
+
+%macro movsxdifnidn 2
+    %ifnidn %1, %2
+        movsxd %1, %2
+    %endif
+%endmacro
+
+%macro ASSERT 1
+    %if (%1) == 0
+        %error assert failed
+    %endif
+%endmacro
+
+%macro DEFINE_ARGS 0-*
+    %ifdef n_arg_names
+        %assign %%i 0
+        %rep n_arg_names
+            CAT_UNDEF arg_name %+ %%i, q
+            CAT_UNDEF arg_name %+ %%i, d
+            CAT_UNDEF arg_name %+ %%i, w
+            CAT_UNDEF arg_name %+ %%i, h
+            CAT_UNDEF arg_name %+ %%i, b
+            CAT_UNDEF arg_name %+ %%i, m
+            CAT_UNDEF arg_name %+ %%i, mp
+            CAT_UNDEF arg_name, %%i
+            %assign %%i %%i+1
+        %endrep
+    %endif
+
+    %xdefine %%stack_offset stack_offset
+    %undef stack_offset ; so that the current value of stack_offset doesn't get baked in by xdefine
+    %assign %%i 0
+    %rep %0
+        %xdefine %1q r %+ %%i %+ q
+        %xdefine %1d r %+ %%i %+ d
+        %xdefine %1w r %+ %%i %+ w
+        %xdefine %1h r %+ %%i %+ h
+        %xdefine %1b r %+ %%i %+ b
+        %xdefine %1m r %+ %%i %+ m
+        %xdefine %1mp r %+ %%i %+ mp
+        CAT_XDEFINE arg_name, %%i, %1
+        %assign %%i %%i+1
+        %rotate 1
+    %endrep
+    %xdefine stack_offset %%stack_offset
+    %assign n_arg_names %0
+%endmacro
+
+%macro ALLOC_STACK 1-2 0 ; stack_size, n_xmm_regs (for win64 only)
+    %ifnum %1
+        %if %1 != 0
+            %assign %%stack_alignment ((mmsize + 15) & ~15)
+            %assign stack_size %1
+            %if stack_size < 0
+                %assign stack_size -stack_size
+            %endif
+            %assign stack_size_padded stack_size
+            %if WIN64
+                %assign stack_size_padded stack_size_padded + 32 ; reserve 32 bytes for shadow space
+                %if mmsize != 8
+                    %assign xmm_regs_used %2
+                    %if xmm_regs_used > 8
+                        %assign stack_size_padded stack_size_padded + (xmm_regs_used-8)*16
+                    %endif
+                %endif
+            %endif
+            %if mmsize <= 16 && HAVE_ALIGNED_STACK
+                %assign stack_size_padded stack_size_padded + %%stack_alignment - gprsize - (stack_offset & (%%stack_alignment - 1))
+                SUB rsp, stack_size_padded
+            %else
+                %assign %%reg_num (regs_used - 1)
+                %xdefine rstk r %+ %%reg_num
+                ; align stack, and save original stack location directly above
+                ; it, i.e. in [rsp+stack_size_padded], so we can restore the
+                ; stack in a single instruction (i.e. mov rsp, rstk or mov
+                ; rsp, [rsp+stack_size_padded])
+                mov  rstk, rsp
+                %if %1 < 0 ; need to store rsp on stack
+                    sub  rsp, gprsize+stack_size_padded
+                    and  rsp, ~(%%stack_alignment-1)
+                    %xdefine rstkm [rsp+stack_size_padded]
+                    mov rstkm, rstk
+                %else ; can keep rsp in rstk during whole function
+                    sub  rsp, stack_size_padded
+                    and  rsp, ~(%%stack_alignment-1)
+                    %xdefine rstkm rstk
+                %endif
+            %endif
+            WIN64_PUSH_XMM
+        %endif
+    %endif
+%endmacro
+
+%macro SETUP_STACK_POINTER 1
+    %ifnum %1
+        %if %1 != 0 && (HAVE_ALIGNED_STACK == 0 || mmsize == 32)
+            %if %1 > 0
+                %assign regs_used (regs_used + 1)
+            %elif ARCH_X86_64 && regs_used == num_args && num_args <= 4 + UNIX64 * 2
+                %warning "Stack pointer will overwrite register argument"
+            %endif
+        %endif
+    %endif
+%endmacro
+
+%macro DEFINE_ARGS_INTERNAL 3+
+    %ifnum %2
+        DEFINE_ARGS %3
+    %elif %1 == 4
+        DEFINE_ARGS %2
+    %elif %1 > 4
+        DEFINE_ARGS %2, %3
+    %endif
+%endmacro
+
+%if WIN64 ; Windows x64 ;=================================================
+
+DECLARE_REG 0,  rcx
+DECLARE_REG 1,  rdx
+DECLARE_REG 2,  R8
+DECLARE_REG 3,  R9
+DECLARE_REG 4,  R10, 40
+DECLARE_REG 5,  R11, 48
+DECLARE_REG 6,  rax, 56
+DECLARE_REG 7,  rdi, 64
+DECLARE_REG 8,  rsi, 72
+DECLARE_REG 9,  rbx, 80
+DECLARE_REG 10, rbp, 88
+DECLARE_REG 11, R12, 96
+DECLARE_REG 12, R13, 104
+DECLARE_REG 13, R14, 112
+DECLARE_REG 14, R15, 120
+
+%macro PROLOGUE 2-5+ 0 ; #args, #regs, #xmm_regs, [stack_size,] arg_names...
+    %assign num_args %1
+    %assign regs_used %2
+    ASSERT regs_used >= num_args
+    SETUP_STACK_POINTER %4
+    ASSERT regs_used <= 15
+    PUSH_IF_USED 7, 8, 9, 10, 11, 12, 13, 14
+    ALLOC_STACK %4, %3
+    %if mmsize != 8 && stack_size == 0
+        WIN64_SPILL_XMM %3
+    %endif
+    LOAD_IF_USED 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14
+    DEFINE_ARGS_INTERNAL %0, %4, %5
+%endmacro
+
+%macro WIN64_PUSH_XMM 0
+    ; Use the shadow space to store XMM6 and XMM7, the rest needs stack space allocated.
+    %if xmm_regs_used > 6
+        movaps [rstk + stack_offset +  8], xmm6
+    %endif
+    %if xmm_regs_used > 7
+        movaps [rstk + stack_offset + 24], xmm7
+    %endif
+    %if xmm_regs_used > 8
+        %assign %%i 8
+        %rep xmm_regs_used-8
+            movaps [rsp + (%%i-8)*16 + stack_size + 32], xmm %+ %%i
+            %assign %%i %%i+1
+        %endrep
+    %endif
+%endmacro
+
+%macro WIN64_SPILL_XMM 1
+    %assign xmm_regs_used %1
+    ASSERT xmm_regs_used <= 16
+    %if xmm_regs_used > 8
+        %assign stack_size_padded (xmm_regs_used-8)*16 + (~stack_offset&8) + 32
+        SUB rsp, stack_size_padded
+    %endif
+    WIN64_PUSH_XMM
+%endmacro
+
+%macro WIN64_RESTORE_XMM_INTERNAL 1
+    %assign %%pad_size 0
+    %if xmm_regs_used > 8
+        %assign %%i xmm_regs_used
+        %rep xmm_regs_used-8
+            %assign %%i %%i-1
+            movaps xmm %+ %%i, [%1 + (%%i-8)*16 + stack_size + 32]
+        %endrep
+    %endif
+    %if stack_size_padded > 0
+        %if stack_size > 0 && (mmsize == 32 || HAVE_ALIGNED_STACK == 0)
+            mov rsp, rstkm
+        %else
+            add %1, stack_size_padded
+            %assign %%pad_size stack_size_padded
+        %endif
+    %endif
+    %if xmm_regs_used > 7
+        movaps xmm7, [%1 + stack_offset - %%pad_size + 24]
+    %endif
+    %if xmm_regs_used > 6
+        movaps xmm6, [%1 + stack_offset - %%pad_size +  8]
+    %endif
+%endmacro
+
+%macro WIN64_RESTORE_XMM 1
+    WIN64_RESTORE_XMM_INTERNAL %1
+    %assign stack_offset (stack_offset-stack_size_padded)
+    %assign xmm_regs_used 0
+%endmacro
+
+%define has_epilogue regs_used > 7 || xmm_regs_used > 6 || mmsize == 32 || stack_size > 0
+
+%macro RET 0
+    WIN64_RESTORE_XMM_INTERNAL rsp
+    POP_IF_USED 14, 13, 12, 11, 10, 9, 8, 7
+%if mmsize == 32
+    vzeroupper
+%endif
+    AUTO_REP_RET
+%endmacro
+
+%elif ARCH_X86_64 ; *nix x64 ;=============================================
+
+DECLARE_REG 0,  rdi
+DECLARE_REG 1,  rsi
+DECLARE_REG 2,  rdx
+DECLARE_REG 3,  rcx
+DECLARE_REG 4,  R8
+DECLARE_REG 5,  R9
+DECLARE_REG 6,  rax, 8
+DECLARE_REG 7,  R10, 16
+DECLARE_REG 8,  R11, 24
+DECLARE_REG 9,  rbx, 32
+DECLARE_REG 10, rbp, 40
+DECLARE_REG 11, R12, 48
+DECLARE_REG 12, R13, 56
+DECLARE_REG 13, R14, 64
+DECLARE_REG 14, R15, 72
+
+%macro PROLOGUE 2-5+ ; #args, #regs, #xmm_regs, [stack_size,] arg_names...
+    %assign num_args %1
+    %assign regs_used %2
+    ASSERT regs_used >= num_args
+    SETUP_STACK_POINTER %4
+    ASSERT regs_used <= 15
+    PUSH_IF_USED 9, 10, 11, 12, 13, 14
+    ALLOC_STACK %4
+    LOAD_IF_USED 6, 7, 8, 9, 10, 11, 12, 13, 14
+    DEFINE_ARGS_INTERNAL %0, %4, %5
+%endmacro
+
+%define has_epilogue regs_used > 9 || mmsize == 32 || stack_size > 0
+
+%macro RET 0
+%if stack_size_padded > 0
+%if mmsize == 32 || HAVE_ALIGNED_STACK == 0
+    mov rsp, rstkm
+%else
+    add rsp, stack_size_padded
+%endif
+%endif
+    POP_IF_USED 14, 13, 12, 11, 10, 9
+%if mmsize == 32
+    vzeroupper
+%endif
+    AUTO_REP_RET
+%endmacro
+
+%else ; X86_32 ;==============================================================
+
+DECLARE_REG 0, eax, 4
+DECLARE_REG 1, ecx, 8
+DECLARE_REG 2, edx, 12
+DECLARE_REG 3, ebx, 16
+DECLARE_REG 4, esi, 20
+DECLARE_REG 5, edi, 24
+DECLARE_REG 6, ebp, 28
+%define rsp esp
+
+%macro DECLARE_ARG 1-*
+    %rep %0
+        %define r%1m [rstk + stack_offset + 4*%1 + 4]
+        %define r%1mp dword r%1m
+        %rotate 1
+    %endrep
+%endmacro
+
+DECLARE_ARG 7, 8, 9, 10, 11, 12, 13, 14
+
+%macro PROLOGUE 2-5+ ; #args, #regs, #xmm_regs, [stack_size,] arg_names...
+    %assign num_args %1
+    %assign regs_used %2
+    ASSERT regs_used >= num_args
+    %if num_args > 7
+        %assign num_args 7
+    %endif
+    %if regs_used > 7
+        %assign regs_used 7
+    %endif
+    SETUP_STACK_POINTER %4
+    ASSERT regs_used <= 7
+    PUSH_IF_USED 3, 4, 5, 6
+    ALLOC_STACK %4
+    LOAD_IF_USED 0, 1, 2, 3, 4, 5, 6
+    DEFINE_ARGS_INTERNAL %0, %4, %5
+%endmacro
+
+%define has_epilogue regs_used > 3 || mmsize == 32 || stack_size > 0
+
+%macro RET 0
+%if stack_size_padded > 0
+%if mmsize == 32 || HAVE_ALIGNED_STACK == 0
+    mov rsp, rstkm
+%else
+    add rsp, stack_size_padded
+%endif
+%endif
+    POP_IF_USED 6, 5, 4, 3
+%if mmsize == 32
+    vzeroupper
+%endif
+    AUTO_REP_RET
+%endmacro
+
+%endif ;======================================================================
+
+%if WIN64 == 0
+%macro WIN64_SPILL_XMM 1
+%endmacro
+%macro WIN64_RESTORE_XMM 1
+%endmacro
+%macro WIN64_PUSH_XMM 0
+%endmacro
+%endif
+
+; On AMD cpus <=K10, an ordinary ret is slow if it immediately follows either
+; a branch or a branch target. So switch to a 2-byte form of ret in that case.
+; We can automatically detect "follows a branch", but not a branch target.
+; (SSSE3 is a sufficient condition to know that your cpu doesn't have this problem.)
+%macro REP_RET 0
+    %if has_epilogue
+        RET
+    %else
+        rep ret
+    %endif
+%endmacro
+
+%define last_branch_adr $$
+%macro AUTO_REP_RET 0
+    %ifndef cpuflags
+        times ((last_branch_adr-$)>>31)+1 rep ; times 1 iff $ != last_branch_adr.
+    %elif notcpuflag(ssse3)
+        times ((last_branch_adr-$)>>31)+1 rep
+    %endif
+    ret
+%endmacro
+
+%macro BRANCH_INSTR 0-*
+    %rep %0
+        %macro %1 1-2 %1
+            %2 %1
+            %%branch_instr:
+            %xdefine last_branch_adr %%branch_instr
+        %endmacro
+        %rotate 1
+    %endrep
+%endmacro
+
+BRANCH_INSTR jz, je, jnz, jne, jl, jle, jnl, jnle, jg, jge, jng, jnge, ja, jae, jna, jnae, jb, jbe, jnb, jnbe, jc, jnc, js, jns, jo, jno, jp, jnp
+
+%macro TAIL_CALL 2 ; callee, is_nonadjacent
+    %if has_epilogue
+        call %1
+        RET
+    %elif %2
+        jmp %1
+    %endif
+%endmacro
+
+;=============================================================================
+; arch-independent part
+;=============================================================================
+
+%assign function_align 16
+
+; Begin a function.
+; Applies any symbol mangling needed for C linkage, and sets up a define such that
+; subsequent uses of the function name automatically refer to the mangled version.
+; Appends cpuflags to the function name if cpuflags has been specified.
+; The "" empty default parameter is a workaround for nasm, which fails if SUFFIX
+; is empty and we call cglobal_internal with just %1 %+ SUFFIX (without %2).
+%macro cglobal 1-2+ "" ; name, [PROLOGUE args]
+    cglobal_internal 1, %1 %+ SUFFIX, %2
+%endmacro
+%macro cvisible 1-2+ "" ; name, [PROLOGUE args]
+    cglobal_internal 0, %1 %+ SUFFIX, %2
+%endmacro
+%macro cglobal_internal 2-3+
+    %if %1
+        %xdefine %%FUNCTION_PREFIX private_prefix
+        %xdefine %%VISIBILITY hidden
+    %else
+        %xdefine %%FUNCTION_PREFIX public_prefix
+        %xdefine %%VISIBILITY
+    %endif
+    %ifndef cglobaled_%2
+        %xdefine %2 mangle(%%FUNCTION_PREFIX %+ _ %+ %2)
+        %xdefine %2.skip_prologue %2 %+ .skip_prologue
+        CAT_XDEFINE cglobaled_, %2, 1
+    %endif
+    %xdefine current_function %2
+    %ifidn __OUTPUT_FORMAT__,elf
+        global %2:function %%VISIBILITY
+    %else
+        global %2
+    %endif
+    align function_align
+    %2:
+    RESET_MM_PERMUTATION        ; needed for x86-64, also makes disassembly somewhat nicer
+    %xdefine rstk rsp           ; copy of the original stack pointer, used when greater alignment than the known stack alignment is required
+    %assign stack_offset 0      ; stack pointer offset relative to the return address
+    %assign stack_size 0        ; amount of stack space that can be freely used inside a function
+    %assign stack_size_padded 0 ; total amount of allocated stack space, including space for callee-saved xmm registers on WIN64 and alignment padding
+    %assign xmm_regs_used 0     ; number of XMM registers requested, used for dealing with callee-saved registers on WIN64
+    %ifnidn %3, ""
+        PROLOGUE %3
+    %endif
+%endmacro
+
+%macro cextern 1
+    %xdefine %1 mangle(private_prefix %+ _ %+ %1)
+    CAT_XDEFINE cglobaled_, %1, 1
+    extern %1
+%endmacro
+
+; like cextern, but without the prefix
+%macro cextern_naked 1
+    %xdefine %1 mangle(%1)
+    CAT_XDEFINE cglobaled_, %1, 1
+    extern %1
+%endmacro
+
+%macro const 1-2+
+    %xdefine %1 mangle(private_prefix %+ _ %+ %1)
+    %ifidn __OUTPUT_FORMAT__,elf
+        global %1:data hidden
+    %else
+        global %1
+    %endif
+    %1: %2
+%endmacro
+
+; This is needed for ELF, otherwise the GNU linker assumes the stack is
+; executable by default.
+%ifidn __OUTPUT_FORMAT__,elf
+SECTION .note.GNU-stack noalloc noexec nowrite progbits
+%endif
+
+; cpuflags
+
+%assign cpuflags_mmx      (1<<0)
+%assign cpuflags_mmx2     (1<<1) | cpuflags_mmx
+%assign cpuflags_3dnow    (1<<2) | cpuflags_mmx
+%assign cpuflags_3dnowext (1<<3) | cpuflags_3dnow
+%assign cpuflags_sse      (1<<4) | cpuflags_mmx2
+%assign cpuflags_sse2     (1<<5) | cpuflags_sse
+%assign cpuflags_sse2slow (1<<6) | cpuflags_sse2
+%assign cpuflags_sse3     (1<<7) | cpuflags_sse2
+%assign cpuflags_ssse3    (1<<8) | cpuflags_sse3
+%assign cpuflags_sse4     (1<<9) | cpuflags_ssse3
+%assign cpuflags_sse42    (1<<10)| cpuflags_sse4
+%assign cpuflags_avx      (1<<11)| cpuflags_sse42
+%assign cpuflags_xop      (1<<12)| cpuflags_avx
+%assign cpuflags_fma4     (1<<13)| cpuflags_avx
+%assign cpuflags_avx2     (1<<14)| cpuflags_avx
+%assign cpuflags_fma3     (1<<15)| cpuflags_avx
+
+%assign cpuflags_cache32  (1<<16)
+%assign cpuflags_cache64  (1<<17)
+%assign cpuflags_slowctz  (1<<18)
+%assign cpuflags_lzcnt    (1<<19)
+%assign cpuflags_aligned  (1<<20) ; not a cpu feature, but a function variant
+%assign cpuflags_atom     (1<<21)
+%assign cpuflags_bmi1     (1<<22)|cpuflags_lzcnt
+%assign cpuflags_bmi2     (1<<23)|cpuflags_bmi1
+
+%define    cpuflag(x) ((cpuflags & (cpuflags_ %+ x)) == (cpuflags_ %+ x))
+%define notcpuflag(x) ((cpuflags & (cpuflags_ %+ x)) != (cpuflags_ %+ x))
+
+; Takes up to 2 cpuflags from the above list.
+; All subsequent functions (up to the next INIT_CPUFLAGS) is built for the specified cpu.
+; You shouldn't need to invoke this macro directly, it's a subroutine for INIT_MMX &co.
+%macro INIT_CPUFLAGS 0-2
+    CPUNOP amdnop
+    %if %0 >= 1
+        %xdefine cpuname %1
+        %assign cpuflags cpuflags_%1
+        %if %0 >= 2
+            %xdefine cpuname %1_%2
+            %assign cpuflags cpuflags | cpuflags_%2
+        %endif
+        %xdefine SUFFIX _ %+ cpuname
+        %if cpuflag(avx)
+            %assign avx_enabled 1
+        %endif
+        %if (mmsize == 16 && notcpuflag(sse2)) || (mmsize == 32 && notcpuflag(avx2))
+            %define mova movaps
+            %define movu movups
+            %define movnta movntps
+        %endif
+        %if cpuflag(aligned)
+            %define movu mova
+        %elifidn %1, sse3
+            %define movu lddqu
+        %endif
+        %if notcpuflag(sse2)
+            CPUNOP basicnop
+        %endif
+    %else
+        %xdefine SUFFIX
+        %undef cpuname
+        %undef cpuflags
+    %endif
+%endmacro
+
+; Merge mmx and sse*
+; m# is a simd regsiter of the currently selected size
+; xm# is the corresponding xmmreg (if selcted xmm or ymm size), or mmreg (if selected mmx)
+; ym# is the corresponding ymmreg (if selcted xmm or ymm size), or mmreg (if selected mmx)
+; (All 3 remain in sync through SWAP.)
+
+%macro CAT_XDEFINE 3
+    %xdefine %1%2 %3
+%endmacro
+
+%macro CAT_UNDEF 2
+    %undef %1%2
+%endmacro
+
+%macro INIT_MMX 0-1+
+    %assign avx_enabled 0
+    %define RESET_MM_PERMUTATION INIT_MMX %1
+    %define mmsize 8
+    %define num_mmregs 8
+    %define mova movq
+    %define movu movq
+    %define movh movd
+    %define movnta movntq
+    %assign %%i 0
+    %rep 8
+    CAT_XDEFINE m, %%i, mm %+ %%i
+    CAT_XDEFINE nmm, %%i, %%i
+    %assign %%i %%i+1
+    %endrep
+    %rep 8
+    CAT_UNDEF m, %%i
+    CAT_UNDEF nmm, %%i
+    %assign %%i %%i+1
+    %endrep
+    INIT_CPUFLAGS %1
+%endmacro
+
+%macro INIT_XMM 0-1+
+    %assign avx_enabled 0
+    %define RESET_MM_PERMUTATION INIT_XMM %1
+    %define mmsize 16
+    %define num_mmregs 8
+    %if ARCH_X86_64
+    %define num_mmregs 16
+    %endif
+    %define mova movdqa
+    %define movu movdqu
+    %define movh movq
+    %define movnta movntdq
+    %assign %%i 0
+    %rep num_mmregs
+    CAT_XDEFINE m, %%i, xmm %+ %%i
+    CAT_XDEFINE nxmm, %%i, %%i
+    %assign %%i %%i+1
+    %endrep
+    INIT_CPUFLAGS %1
+%endmacro
+
+%macro INIT_YMM 0-1+
+    %assign avx_enabled 1
+    %define RESET_MM_PERMUTATION INIT_YMM %1
+    %define mmsize 32
+    %define num_mmregs 8
+    %if ARCH_X86_64
+    %define num_mmregs 16
+    %endif
+    %define mova movdqa
+    %define movu movdqu
+    %undef movh
+    %define movnta movntdq
+    %assign %%i 0
+    %rep num_mmregs
+    CAT_XDEFINE m, %%i, ymm %+ %%i
+    CAT_XDEFINE nymm, %%i, %%i
+    %assign %%i %%i+1
+    %endrep
+    INIT_CPUFLAGS %1
+%endmacro
+
+INIT_XMM
+
+%macro DECLARE_MMCAST 1
+    %define  mmmm%1   mm%1
+    %define  mmxmm%1  mm%1
+    %define  mmymm%1  mm%1
+    %define xmmmm%1   mm%1
+    %define xmmxmm%1 xmm%1
+    %define xmmymm%1 xmm%1
+    %define ymmmm%1   mm%1
+    %define ymmxmm%1 ymm%1
+    %define ymmymm%1 ymm%1
+    %define xm%1 xmm %+ m%1
+    %define ym%1 ymm %+ m%1
+%endmacro
+
+%assign i 0
+%rep 16
+    DECLARE_MMCAST i
+%assign i i+1
+%endrep
+
+; I often want to use macros that permute their arguments. e.g. there's no
+; efficient way to implement butterfly or transpose or dct without swapping some
+; arguments.
+;
+; I would like to not have to manually keep track of the permutations:
+; If I insert a permutation in the middle of a function, it should automatically
+; change everything that follows. For more complex macros I may also have multiple
+; implementations, e.g. the SSE2 and SSSE3 versions may have different permutations.
+;
+; Hence these macros. Insert a PERMUTE or some SWAPs at the end of a macro that
+; permutes its arguments. It's equivalent to exchanging the contents of the
+; registers, except that this way you exchange the register names instead, so it
+; doesn't cost any cycles.
+
+%macro PERMUTE 2-* ; takes a list of pairs to swap
+%rep %0/2
+    %xdefine %%tmp%2 m%2
+    %rotate 2
+%endrep
+%rep %0/2
+    %xdefine m%1 %%tmp%2
+    CAT_XDEFINE n, m%1, %1
+    %rotate 2
+%endrep
+%endmacro
+
+%macro SWAP 2+ ; swaps a single chain (sometimes more concise than pairs)
+%ifnum %1 ; SWAP 0, 1, ...
+    SWAP_INTERNAL_NUM %1, %2
+%else ; SWAP m0, m1, ...
+    SWAP_INTERNAL_NAME %1, %2
+%endif
+%endmacro
+
+%macro SWAP_INTERNAL_NUM 2-*
+    %rep %0-1
+        %xdefine %%tmp m%1
+        %xdefine m%1 m%2
+        %xdefine m%2 %%tmp
+        CAT_XDEFINE n, m%1, %1
+        CAT_XDEFINE n, m%2, %2
+    %rotate 1
+    %endrep
+%endmacro
+
+%macro SWAP_INTERNAL_NAME 2-*
+    %xdefine %%args n %+ %1
+    %rep %0-1
+        %xdefine %%args %%args, n %+ %2
+    %rotate 1
+    %endrep
+    SWAP_INTERNAL_NUM %%args
+%endmacro
+
+; If SAVE_MM_PERMUTATION is placed at the end of a function, then any later
+; calls to that function will automatically load the permutation, so values can
+; be returned in mmregs.
+%macro SAVE_MM_PERMUTATION 0-1
+    %if %0
+        %xdefine %%f %1_m
+    %else
+        %xdefine %%f current_function %+ _m
+    %endif
+    %assign %%i 0
+    %rep num_mmregs
+        CAT_XDEFINE %%f, %%i, m %+ %%i
+    %assign %%i %%i+1
+    %endrep
+%endmacro
+
+%macro LOAD_MM_PERMUTATION 1 ; name to load from
+    %ifdef %1_m0
+        %assign %%i 0
+        %rep num_mmregs
+            CAT_XDEFINE m, %%i, %1_m %+ %%i
+            CAT_XDEFINE n, m %+ %%i, %%i
+        %assign %%i %%i+1
+        %endrep
+    %endif
+%endmacro
+
+; Append cpuflags to the callee's name iff the appended name is known and the plain name isn't
+%macro call 1
+    call_internal %1 %+ SUFFIX, %1
+%endmacro
+%macro call_internal 2
+    %xdefine %%i %2
+    %ifndef cglobaled_%2
+        %ifdef cglobaled_%1
+            %xdefine %%i %1
+        %endif
+    %endif
+    call %%i
+    LOAD_MM_PERMUTATION %%i
+%endmacro
+
+; Substitutions that reduce instruction size but are functionally equivalent
+%macro add 2
+    %ifnum %2
+        %if %2==128
+            sub %1, -128
+        %else
+            add %1, %2
+        %endif
+    %else
+        add %1, %2
+    %endif
+%endmacro
+
+%macro sub 2
+    %ifnum %2
+        %if %2==128
+            add %1, -128
+        %else
+            sub %1, %2
+        %endif
+    %else
+        sub %1, %2
+    %endif
+%endmacro
+
+;=============================================================================
+; AVX abstraction layer
+;=============================================================================
+
+%assign i 0
+%rep 16
+    %if i < 8
+        CAT_XDEFINE sizeofmm, i, 8
+    %endif
+    CAT_XDEFINE sizeofxmm, i, 16
+    CAT_XDEFINE sizeofymm, i, 32
+%assign i i+1
+%endrep
+%undef i
+
+%macro CHECK_AVX_INSTR_EMU 3-*
+    %xdefine %%opcode %1
+    %xdefine %%dst %2
+    %rep %0-2
+        %ifidn %%dst, %3
+            %error non-avx emulation of ``%%opcode'' is not supported
+        %endif
+        %rotate 1
+    %endrep
+%endmacro
+
+;%1 == instruction
+;%2 == 1 if float, 0 if int
+;%3 == 1 if non-destructive or 4-operand (xmm, xmm, xmm, imm), 0 otherwise
+;%4 == 1 if commutative (i.e. doesn't matter which src arg is which), 0 if not
+;%5+: operands
+%macro RUN_AVX_INSTR 5-8+
+    %ifnum sizeof%6
+        %assign __sizeofreg sizeof%6
+    %elifnum sizeof%5
+        %assign __sizeofreg sizeof%5
+    %else
+        %assign __sizeofreg mmsize
+    %endif
+    %assign __emulate_avx 0
+    %if avx_enabled && __sizeofreg >= 16
+        %xdefine __instr v%1
+    %else
+        %xdefine __instr %1
+        %if %0 >= 7+%3
+            %assign __emulate_avx 1
+        %endif
+    %endif
+
+    %if __emulate_avx
+        %xdefine __src1 %6
+        %xdefine __src2 %7
+        %ifnidn %5, %6
+            %if %0 >= 8
+                CHECK_AVX_INSTR_EMU {%1 %5, %6, %7, %8}, %5, %7, %8
+            %else
+                CHECK_AVX_INSTR_EMU {%1 %5, %6, %7}, %5, %7
+            %endif
+            %if %4 && %3 == 0
+                %ifnid %7
+                    ; 3-operand AVX instructions with a memory arg can only have it in src2,
+                    ; whereas SSE emulation prefers to have it in src1 (i.e. the mov).
+                    ; So, if the instruction is commutative with a memory arg, swap them.
+                    %xdefine __src1 %7
+                    %xdefine __src2 %6
+                %endif
+            %endif
+            %if __sizeofreg == 8
+                MOVQ %5, __src1
+            %elif %2
+                MOVAPS %5, __src1
+            %else
+                MOVDQA %5, __src1
+            %endif
+        %endif
+        %if %0 >= 8
+            %1 %5, __src2, %8
+        %else
+            %1 %5, __src2
+        %endif
+    %elif %0 >= 8
+        __instr %5, %6, %7, %8
+    %elif %0 == 7
+        __instr %5, %6, %7
+    %elif %0 == 6
+        __instr %5, %6
+    %else
+        __instr %5
+    %endif
+%endmacro
+
+;%1 == instruction
+;%2 == 1 if float, 0 if int
+;%3 == 1 if non-destructive or 4-operand (xmm, xmm, xmm, imm), 0 otherwise
+;%4 == 1 if commutative (i.e. doesn't matter which src arg is which), 0 if not
+%macro AVX_INSTR 1-4 0, 1, 0
+    %macro %1 1-9 fnord, fnord, fnord, fnord, %1, %2, %3, %4
+        %ifidn %2, fnord
+            RUN_AVX_INSTR %6, %7, %8, %9, %1
+        %elifidn %3, fnord
+            RUN_AVX_INSTR %6, %7, %8, %9, %1, %2
+        %elifidn %4, fnord
+            RUN_AVX_INSTR %6, %7, %8, %9, %1, %2, %3
+        %elifidn %5, fnord
+            RUN_AVX_INSTR %6, %7, %8, %9, %1, %2, %3, %4
+        %else
+            RUN_AVX_INSTR %6, %7, %8, %9, %1, %2, %3, %4, %5
+        %endif
+    %endmacro
+%endmacro
+
+; Instructions with both VEX and non-VEX encodings
+; Non-destructive instructions are written without parameters
+AVX_INSTR addpd, 1, 0, 1
+AVX_INSTR addps, 1, 0, 1
+AVX_INSTR addsd, 1, 0, 1
+AVX_INSTR addss, 1, 0, 1
+AVX_INSTR addsubpd, 1, 0, 0
+AVX_INSTR addsubps, 1, 0, 0
+AVX_INSTR aesdec, 0, 0, 0
+AVX_INSTR aesdeclast, 0, 0, 0
+AVX_INSTR aesenc, 0, 0, 0
+AVX_INSTR aesenclast, 0, 0, 0
+AVX_INSTR aesimc
+AVX_INSTR aeskeygenassist
+AVX_INSTR andnpd, 1, 0, 0
+AVX_INSTR andnps, 1, 0, 0
+AVX_INSTR andpd, 1, 0, 1
+AVX_INSTR andps, 1, 0, 1
+AVX_INSTR blendpd, 1, 0, 0
+AVX_INSTR blendps, 1, 0, 0
+AVX_INSTR blendvpd, 1, 0, 0
+AVX_INSTR blendvps, 1, 0, 0
+AVX_INSTR cmppd, 1, 1, 0
+AVX_INSTR cmpps, 1, 1, 0
+AVX_INSTR cmpsd, 1, 1, 0
+AVX_INSTR cmpss, 1, 1, 0
+AVX_INSTR comisd
+AVX_INSTR comiss
+AVX_INSTR cvtdq2pd
+AVX_INSTR cvtdq2ps
+AVX_INSTR cvtpd2dq
+AVX_INSTR cvtpd2ps
+AVX_INSTR cvtps2dq
+AVX_INSTR cvtps2pd
+AVX_INSTR cvtsd2si
+AVX_INSTR cvtsd2ss
+AVX_INSTR cvtsi2sd
+AVX_INSTR cvtsi2ss
+AVX_INSTR cvtss2sd
+AVX_INSTR cvtss2si
+AVX_INSTR cvttpd2dq
+AVX_INSTR cvttps2dq
+AVX_INSTR cvttsd2si
+AVX_INSTR cvttss2si
+AVX_INSTR divpd, 1, 0, 0
+AVX_INSTR divps, 1, 0, 0
+AVX_INSTR divsd, 1, 0, 0
+AVX_INSTR divss, 1, 0, 0
+AVX_INSTR dppd, 1, 1, 0
+AVX_INSTR dpps, 1, 1, 0
+AVX_INSTR extractps
+AVX_INSTR haddpd, 1, 0, 0
+AVX_INSTR haddps, 1, 0, 0
+AVX_INSTR hsubpd, 1, 0, 0
+AVX_INSTR hsubps, 1, 0, 0
+AVX_INSTR insertps, 1, 1, 0
+AVX_INSTR lddqu
+AVX_INSTR ldmxcsr
+AVX_INSTR maskmovdqu
+AVX_INSTR maxpd, 1, 0, 1
+AVX_INSTR maxps, 1, 0, 1
+AVX_INSTR maxsd, 1, 0, 1
+AVX_INSTR maxss, 1, 0, 1
+AVX_INSTR minpd, 1, 0, 1
+AVX_INSTR minps, 1, 0, 1
+AVX_INSTR minsd, 1, 0, 1
+AVX_INSTR minss, 1, 0, 1
+AVX_INSTR movapd
+AVX_INSTR movaps
+AVX_INSTR movd
+AVX_INSTR movddup
+AVX_INSTR movdqa
+AVX_INSTR movdqu
+AVX_INSTR movhlps, 1, 0, 0
+AVX_INSTR movhpd, 1, 0, 0
+AVX_INSTR movhps, 1, 0, 0
+AVX_INSTR movlhps, 1, 0, 0
+AVX_INSTR movlpd, 1, 0, 0
+AVX_INSTR movlps, 1, 0, 0
+AVX_INSTR movmskpd
+AVX_INSTR movmskps
+AVX_INSTR movntdq
+AVX_INSTR movntdqa
+AVX_INSTR movntpd
+AVX_INSTR movntps
+AVX_INSTR movq
+AVX_INSTR movsd, 1, 0, 0
+AVX_INSTR movshdup
+AVX_INSTR movsldup
+AVX_INSTR movss, 1, 0, 0
+AVX_INSTR movupd
+AVX_INSTR movups
+AVX_INSTR mpsadbw, 0, 1, 0
+AVX_INSTR mulpd, 1, 0, 1
+AVX_INSTR mulps, 1, 0, 1
+AVX_INSTR mulsd, 1, 0, 1
+AVX_INSTR mulss, 1, 0, 1
+AVX_INSTR orpd, 1, 0, 1
+AVX_INSTR orps, 1, 0, 1
+AVX_INSTR pabsb
+AVX_INSTR pabsd
+AVX_INSTR pabsw
+AVX_INSTR packsswb, 0, 0, 0
+AVX_INSTR packssdw, 0, 0, 0
+AVX_INSTR packuswb, 0, 0, 0
+AVX_INSTR packusdw, 0, 0, 0
+AVX_INSTR paddb, 0, 0, 1
+AVX_INSTR paddw, 0, 0, 1
+AVX_INSTR paddd, 0, 0, 1
+AVX_INSTR paddq, 0, 0, 1
+AVX_INSTR paddsb, 0, 0, 1
+AVX_INSTR paddsw, 0, 0, 1
+AVX_INSTR paddusb, 0, 0, 1
+AVX_INSTR paddusw, 0, 0, 1
+AVX_INSTR palignr, 0, 1, 0
+AVX_INSTR pand, 0, 0, 1
+AVX_INSTR pandn, 0, 0, 0
+AVX_INSTR pavgb, 0, 0, 1
+AVX_INSTR pavgw, 0, 0, 1
+AVX_INSTR pblendvb, 0, 0, 0
+AVX_INSTR pblendw, 0, 1, 0
+AVX_INSTR pclmulqdq, 0, 1, 0
+AVX_INSTR pcmpestri
+AVX_INSTR pcmpestrm
+AVX_INSTR pcmpistri
+AVX_INSTR pcmpistrm
+AVX_INSTR pcmpeqb, 0, 0, 1
+AVX_INSTR pcmpeqw, 0, 0, 1
+AVX_INSTR pcmpeqd, 0, 0, 1
+AVX_INSTR pcmpeqq, 0, 0, 1
+AVX_INSTR pcmpgtb, 0, 0, 0
+AVX_INSTR pcmpgtw, 0, 0, 0
+AVX_INSTR pcmpgtd, 0, 0, 0
+AVX_INSTR pcmpgtq, 0, 0, 0
+AVX_INSTR pextrb
+AVX_INSTR pextrd
+AVX_INSTR pextrq
+AVX_INSTR pextrw
+AVX_INSTR phaddw, 0, 0, 0
+AVX_INSTR phaddd, 0, 0, 0
+AVX_INSTR phaddsw, 0, 0, 0
+AVX_INSTR phminposuw
+AVX_INSTR phsubw, 0, 0, 0
+AVX_INSTR phsubd, 0, 0, 0
+AVX_INSTR phsubsw, 0, 0, 0
+AVX_INSTR pinsrb, 0, 1, 0
+AVX_INSTR pinsrd, 0, 1, 0
+AVX_INSTR pinsrq, 0, 1, 0
+AVX_INSTR pinsrw, 0, 1, 0
+AVX_INSTR pmaddwd, 0, 0, 1
+AVX_INSTR pmaddubsw, 0, 0, 0
+AVX_INSTR pmaxsb, 0, 0, 1
+AVX_INSTR pmaxsw, 0, 0, 1
+AVX_INSTR pmaxsd, 0, 0, 1
+AVX_INSTR pmaxub, 0, 0, 1
+AVX_INSTR pmaxuw, 0, 0, 1
+AVX_INSTR pmaxud, 0, 0, 1
+AVX_INSTR pminsb, 0, 0, 1
+AVX_INSTR pminsw, 0, 0, 1
+AVX_INSTR pminsd, 0, 0, 1
+AVX_INSTR pminub, 0, 0, 1
+AVX_INSTR pminuw, 0, 0, 1
+AVX_INSTR pminud, 0, 0, 1
+AVX_INSTR pmovmskb
+AVX_INSTR pmovsxbw
+AVX_INSTR pmovsxbd
+AVX_INSTR pmovsxbq
+AVX_INSTR pmovsxwd
+AVX_INSTR pmovsxwq
+AVX_INSTR pmovsxdq
+AVX_INSTR pmovzxbw
+AVX_INSTR pmovzxbd
+AVX_INSTR pmovzxbq
+AVX_INSTR pmovzxwd
+AVX_INSTR pmovzxwq
+AVX_INSTR pmovzxdq
+AVX_INSTR pmuldq, 0, 0, 1
+AVX_INSTR pmulhrsw, 0, 0, 1
+AVX_INSTR pmulhuw, 0, 0, 1
+AVX_INSTR pmulhw, 0, 0, 1
+AVX_INSTR pmullw, 0, 0, 1
+AVX_INSTR pmulld, 0, 0, 1
+AVX_INSTR pmuludq, 0, 0, 1
+AVX_INSTR por, 0, 0, 1
+AVX_INSTR psadbw, 0, 0, 1
+AVX_INSTR pshufb, 0, 0, 0
+AVX_INSTR pshufd
+AVX_INSTR pshufhw
+AVX_INSTR pshuflw
+AVX_INSTR psignb, 0, 0, 0
+AVX_INSTR psignw, 0, 0, 0
+AVX_INSTR psignd, 0, 0, 0
+AVX_INSTR psllw, 0, 0, 0
+AVX_INSTR pslld, 0, 0, 0
+AVX_INSTR psllq, 0, 0, 0
+AVX_INSTR pslldq, 0, 0, 0
+AVX_INSTR psraw, 0, 0, 0
+AVX_INSTR psrad, 0, 0, 0
+AVX_INSTR psrlw, 0, 0, 0
+AVX_INSTR psrld, 0, 0, 0
+AVX_INSTR psrlq, 0, 0, 0
+AVX_INSTR psrldq, 0, 0, 0
+AVX_INSTR psubb, 0, 0, 0
+AVX_INSTR psubw, 0, 0, 0
+AVX_INSTR psubd, 0, 0, 0
+AVX_INSTR psubq, 0, 0, 0
+AVX_INSTR psubsb, 0, 0, 0
+AVX_INSTR psubsw, 0, 0, 0
+AVX_INSTR psubusb, 0, 0, 0
+AVX_INSTR psubusw, 0, 0, 0
+AVX_INSTR ptest
+AVX_INSTR punpckhbw, 0, 0, 0
+AVX_INSTR punpckhwd, 0, 0, 0
+AVX_INSTR punpckhdq, 0, 0, 0
+AVX_INSTR punpckhqdq, 0, 0, 0
+AVX_INSTR punpcklbw, 0, 0, 0
+AVX_INSTR punpcklwd, 0, 0, 0
+AVX_INSTR punpckldq, 0, 0, 0
+AVX_INSTR punpcklqdq, 0, 0, 0
+AVX_INSTR pxor, 0, 0, 1
+AVX_INSTR rcpps, 1, 0, 0
+AVX_INSTR rcpss, 1, 0, 0
+AVX_INSTR roundpd
+AVX_INSTR roundps
+AVX_INSTR roundsd
+AVX_INSTR roundss
+AVX_INSTR rsqrtps, 1, 0, 0
+AVX_INSTR rsqrtss, 1, 0, 0
+AVX_INSTR shufpd, 1, 1, 0
+AVX_INSTR shufps, 1, 1, 0
+AVX_INSTR sqrtpd, 1, 0, 0
+AVX_INSTR sqrtps, 1, 0, 0
+AVX_INSTR sqrtsd, 1, 0, 0
+AVX_INSTR sqrtss, 1, 0, 0
+AVX_INSTR stmxcsr
+AVX_INSTR subpd, 1, 0, 0
+AVX_INSTR subps, 1, 0, 0
+AVX_INSTR subsd, 1, 0, 0
+AVX_INSTR subss, 1, 0, 0
+AVX_INSTR ucomisd
+AVX_INSTR ucomiss
+AVX_INSTR unpckhpd, 1, 0, 0
+AVX_INSTR unpckhps, 1, 0, 0
+AVX_INSTR unpcklpd, 1, 0, 0
+AVX_INSTR unpcklps, 1, 0, 0
+AVX_INSTR xorpd, 1, 0, 1
+AVX_INSTR xorps, 1, 0, 1
+
+; 3DNow instructions, for sharing code between AVX, SSE and 3DN
+AVX_INSTR pfadd, 1, 0, 1
+AVX_INSTR pfsub, 1, 0, 0
+AVX_INSTR pfmul, 1, 0, 1
+
+; base-4 constants for shuffles
+%assign i 0
+%rep 256
+    %assign j ((i>>6)&3)*1000 + ((i>>4)&3)*100 + ((i>>2)&3)*10 + (i&3)
+    %if j < 10
+        CAT_XDEFINE q000, j, i
+    %elif j < 100
+        CAT_XDEFINE q00, j, i
+    %elif j < 1000
+        CAT_XDEFINE q0, j, i
+    %else
+        CAT_XDEFINE q, j, i
+    %endif
+%assign i i+1
+%endrep
+%undef i
+%undef j
+
+%macro FMA_INSTR 3
+    %macro %1 4-7 %1, %2, %3
+        %if cpuflag(xop)
+            v%5 %1, %2, %3, %4
+        %else
+            %6 %1, %2, %3
+            %7 %1, %4
+        %endif
+    %endmacro
+%endmacro
+
+FMA_INSTR  pmacsdd,  pmulld, paddd
+FMA_INSTR  pmacsww,  pmullw, paddw
+FMA_INSTR pmadcswd, pmaddwd, paddd
+
+; tzcnt is equivalent to "rep bsf" and is backwards-compatible with bsf.
+; This lets us use tzcnt without bumping the yasm version requirement yet.
+%define tzcnt rep bsf
+
+; convert FMA4 to FMA3 if possible
+%macro FMA4_INSTR 4
+    %macro %1 4-8 %1, %2, %3, %4
+        %if cpuflag(fma4)
+            v%5 %1, %2, %3, %4
+        %elifidn %1, %2
+            v%6 %1, %4, %3 ; %1 = %1 * %3 + %4
+        %elifidn %1, %3
+            v%7 %1, %2, %4 ; %1 = %2 * %1 + %4
+        %elifidn %1, %4
+            v%8 %1, %2, %3 ; %1 = %2 * %3 + %1
+        %else
+            %error fma3 emulation of ``%5 %1, %2, %3, %4'' is not supported
+        %endif
+    %endmacro
+%endmacro
+
+FMA4_INSTR fmaddpd, fmadd132pd, fmadd213pd, fmadd231pd
+FMA4_INSTR fmaddps, fmadd132ps, fmadd213ps, fmadd231ps
+FMA4_INSTR fmaddsd, fmadd132sd, fmadd213sd, fmadd231sd
+FMA4_INSTR fmaddss, fmadd132ss, fmadd213ss, fmadd231ss
+
+FMA4_INSTR fmaddsubpd, fmaddsub132pd, fmaddsub213pd, fmaddsub231pd
+FMA4_INSTR fmaddsubps, fmaddsub132ps, fmaddsub213ps, fmaddsub231ps
+FMA4_INSTR fmsubaddpd, fmsubadd132pd, fmsubadd213pd, fmsubadd231pd
+FMA4_INSTR fmsubaddps, fmsubadd132ps, fmsubadd213ps, fmsubadd231ps
+
+FMA4_INSTR fmsubpd, fmsub132pd, fmsub213pd, fmsub231pd
+FMA4_INSTR fmsubps, fmsub132ps, fmsub213ps, fmsub231ps
+FMA4_INSTR fmsubsd, fmsub132sd, fmsub213sd, fmsub231sd
+FMA4_INSTR fmsubss, fmsub132ss, fmsub213ss, fmsub231ss
+
+FMA4_INSTR fnmaddpd, fnmadd132pd, fnmadd213pd, fnmadd231pd
+FMA4_INSTR fnmaddps, fnmadd132ps, fnmadd213ps, fnmadd231ps
+FMA4_INSTR fnmaddsd, fnmadd132sd, fnmadd213sd, fnmadd231sd
+FMA4_INSTR fnmaddss, fnmadd132ss, fnmadd213ss, fnmadd231ss
+
+FMA4_INSTR fnmsubpd, fnmsub132pd, fnmsub213pd, fnmsub231pd
+FMA4_INSTR fnmsubps, fnmsub132ps, fnmsub213ps, fnmsub231ps
+FMA4_INSTR fnmsubsd, fnmsub132sd, fnmsub213sd, fnmsub231sd
+FMA4_INSTR fnmsubss, fnmsub132ss, fnmsub213ss, fnmsub231ss
+
+; workaround: vpbroadcastq is broken in x86_32 due to a yasm bug
+%if ARCH_X86_64 == 0
+%macro vpbroadcastq 2
+%if sizeof%1 == 16
+    movddup %1, %2
+%else
+    vbroadcastsd %1, %2
+%endif
+%endmacro
+%endif
diff --git a/media/libav/libavutil/x86/x86util.asm b/media/libav/libavutil/x86/x86util.asm
new file mode 100644
index 000000000..11779cf07
--- /dev/null
+++ b/media/libav/libavutil/x86/x86util.asm
@@ -0,0 +1,680 @@
+;*****************************************************************************
+;* x86util.asm
+;*****************************************************************************
+;* Copyright (C) 2008-2010 x264 project
+;*
+;* Authors: Loren Merritt <lorenm@u.washington.edu>
+;*          Holger Lubitz <holger@lubitz.org>
+;*
+;* 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 private_prefix ff
+%define public_prefix  avpriv
+%define cpuflags_mmxext cpuflags_mmx2
+
+%include "libavutil/x86/x86inc.asm"
+
+%macro SBUTTERFLY 4
+%if avx_enabled == 0
+    mova      m%4, m%2
+    punpckl%1 m%2, m%3
+    punpckh%1 m%4, m%3
+%else
+    punpckh%1 m%4, m%2, m%3
+    punpckl%1 m%2, m%3
+%endif
+    SWAP %3, %4
+%endmacro
+
+%macro SBUTTERFLY2 4
+    punpckl%1 m%4, m%2, m%3
+    punpckh%1 m%2, m%2, m%3
+    SWAP %2, %4, %3
+%endmacro
+
+%macro SBUTTERFLYPS 3
+    unpcklps m%3, m%1, m%2
+    unpckhps m%1, m%1, m%2
+    SWAP %1, %3, %2
+%endmacro
+
+%macro TRANSPOSE4x4B 5
+    SBUTTERFLY bw, %1, %2, %5
+    SBUTTERFLY bw, %3, %4, %5
+    SBUTTERFLY wd, %1, %3, %5
+    SBUTTERFLY wd, %2, %4, %5
+    SWAP %2, %3
+%endmacro
+
+%macro TRANSPOSE4x4W 5
+    SBUTTERFLY wd, %1, %2, %5
+    SBUTTERFLY wd, %3, %4, %5
+    SBUTTERFLY dq, %1, %3, %5
+    SBUTTERFLY dq, %2, %4, %5
+    SWAP %2, %3
+%endmacro
+
+%macro TRANSPOSE2x4x4W 5
+    SBUTTERFLY wd,  %1, %2, %5
+    SBUTTERFLY wd,  %3, %4, %5
+    SBUTTERFLY dq,  %1, %3, %5
+    SBUTTERFLY dq,  %2, %4, %5
+    SBUTTERFLY qdq, %1, %2, %5
+    SBUTTERFLY qdq, %3, %4, %5
+%endmacro
+
+%macro TRANSPOSE4x4D 5
+    SBUTTERFLY dq,  %1, %2, %5
+    SBUTTERFLY dq,  %3, %4, %5
+    SBUTTERFLY qdq, %1, %3, %5
+    SBUTTERFLY qdq, %2, %4, %5
+    SWAP %2, %3
+%endmacro
+
+; identical behavior to TRANSPOSE4x4D, but using SSE1 float ops
+%macro TRANSPOSE4x4PS 5
+    SBUTTERFLYPS %1, %2, %5
+    SBUTTERFLYPS %3, %4, %5
+    movlhps m%5, m%1, m%3
+    movhlps m%3, m%1
+    SWAP %5, %1
+    movlhps m%5, m%2, m%4
+    movhlps m%4, m%2
+    SWAP %5, %2, %3
+%endmacro
+
+%macro TRANSPOSE8x8W 9-11
+%if ARCH_X86_64
+    SBUTTERFLY wd,  %1, %2, %9
+    SBUTTERFLY wd,  %3, %4, %9
+    SBUTTERFLY wd,  %5, %6, %9
+    SBUTTERFLY wd,  %7, %8, %9
+    SBUTTERFLY dq,  %1, %3, %9
+    SBUTTERFLY dq,  %2, %4, %9
+    SBUTTERFLY dq,  %5, %7, %9
+    SBUTTERFLY dq,  %6, %8, %9
+    SBUTTERFLY qdq, %1, %5, %9
+    SBUTTERFLY qdq, %2, %6, %9
+    SBUTTERFLY qdq, %3, %7, %9
+    SBUTTERFLY qdq, %4, %8, %9
+    SWAP %2, %5
+    SWAP %4, %7
+%else
+; in:  m0..m7, unless %11 in which case m6 is in %9
+; out: m0..m7, unless %11 in which case m4 is in %10
+; spills into %9 and %10
+%if %0<11
+    movdqa %9, m%7
+%endif
+    SBUTTERFLY wd,  %1, %2, %7
+    movdqa %10, m%2
+    movdqa m%7, %9
+    SBUTTERFLY wd,  %3, %4, %2
+    SBUTTERFLY wd,  %5, %6, %2
+    SBUTTERFLY wd,  %7, %8, %2
+    SBUTTERFLY dq,  %1, %3, %2
+    movdqa %9, m%3
+    movdqa m%2, %10
+    SBUTTERFLY dq,  %2, %4, %3
+    SBUTTERFLY dq,  %5, %7, %3
+    SBUTTERFLY dq,  %6, %8, %3
+    SBUTTERFLY qdq, %1, %5, %3
+    SBUTTERFLY qdq, %2, %6, %3
+    movdqa %10, m%2
+    movdqa m%3, %9
+    SBUTTERFLY qdq, %3, %7, %2
+    SBUTTERFLY qdq, %4, %8, %2
+    SWAP %2, %5
+    SWAP %4, %7
+%if %0<11
+    movdqa m%5, %10
+%endif
+%endif
+%endmacro
+
+; PABSW macro assumes %1 != %2, while ABS1/2 macros work in-place
+%macro PABSW 2
+%if cpuflag(ssse3)
+    pabsw      %1, %2
+%elif cpuflag(mmxext)
+    pxor    %1, %1
+    psubw   %1, %2
+    pmaxsw  %1, %2
+%else
+    pxor       %1, %1
+    pcmpgtw    %1, %2
+    pxor       %2, %1
+    psubw      %2, %1
+    SWAP       %1, %2
+%endif
+%endmacro
+
+%macro PSIGNW_MMX 2
+    pxor       %1, %2
+    psubw      %1, %2
+%endmacro
+
+%macro PSIGNW_SSSE3 2
+    psignw     %1, %2
+%endmacro
+
+%macro ABS1 2
+%if cpuflag(ssse3)
+    pabsw   %1, %1
+%elif cpuflag(mmxext) ; a, tmp
+    pxor    %2, %2
+    psubw   %2, %1
+    pmaxsw  %1, %2
+%else ; a, tmp
+    pxor       %2, %2
+    pcmpgtw    %2, %1
+    pxor       %1, %2
+    psubw      %1, %2
+%endif
+%endmacro
+
+%macro ABS2 4
+%if cpuflag(ssse3)
+    pabsw   %1, %1
+    pabsw   %2, %2
+%elif cpuflag(mmxext) ; a, b, tmp0, tmp1
+    pxor    %3, %3
+    pxor    %4, %4
+    psubw   %3, %1
+    psubw   %4, %2
+    pmaxsw  %1, %3
+    pmaxsw  %2, %4
+%else ; a, b, tmp0, tmp1
+    pxor       %3, %3
+    pxor       %4, %4
+    pcmpgtw    %3, %1
+    pcmpgtw    %4, %2
+    pxor       %1, %3
+    pxor       %2, %4
+    psubw      %1, %3
+    psubw      %2, %4
+%endif
+%endmacro
+
+%macro ABSB 2 ; source mmreg, temp mmreg (unused for ssse3)
+%if cpuflag(ssse3)
+    pabsb   %1, %1
+%else
+    pxor    %2, %2
+    psubb   %2, %1
+    pminub  %1, %2
+%endif
+%endmacro
+
+%macro ABSB2 4 ; src1, src2, tmp1, tmp2 (tmp1/2 unused for SSSE3)
+%if cpuflag(ssse3)
+    pabsb   %1, %1
+    pabsb   %2, %2
+%else
+    pxor    %3, %3
+    pxor    %4, %4
+    psubb   %3, %1
+    psubb   %4, %2
+    pminub  %1, %3
+    pminub  %2, %4
+%endif
+%endmacro
+
+%macro ABSD2_MMX 4
+    pxor    %3, %3
+    pxor    %4, %4
+    pcmpgtd %3, %1
+    pcmpgtd %4, %2
+    pxor    %1, %3
+    pxor    %2, %4
+    psubd   %1, %3
+    psubd   %2, %4
+%endmacro
+
+%macro ABS4 6
+    ABS2 %1, %2, %5, %6
+    ABS2 %3, %4, %5, %6
+%endmacro
+
+%macro SPLATB_LOAD 3
+%if cpuflag(ssse3)
+    movd      %1, [%2-3]
+    pshufb    %1, %3
+%else
+    movd      %1, [%2-3] ;to avoid crossing a cacheline
+    punpcklbw %1, %1
+    SPLATW    %1, %1, 3
+%endif
+%endmacro
+
+%macro SPLATB_REG 3
+%if cpuflag(ssse3)
+    movd      %1, %2d
+    pshufb    %1, %3
+%else
+    movd      %1, %2d
+    punpcklbw %1, %1
+    SPLATW    %1, %1, 0
+%endif
+%endmacro
+
+%macro PALIGNR 4-5
+%if cpuflag(ssse3)
+%if %0==5
+    palignr %1, %2, %3, %4
+%else
+    palignr %1, %2, %3
+%endif
+%elif cpuflag(mmx) ; [dst,] src1, src2, imm, tmp
+    %define %%dst %1
+%if %0==5
+%ifnidn %1, %2
+    mova    %%dst, %2
+%endif
+    %rotate 1
+%endif
+%ifnidn %4, %2
+    mova    %4, %2
+%endif
+%if mmsize==8
+    psllq   %%dst, (8-%3)*8
+    psrlq   %4, %3*8
+%else
+    pslldq  %%dst, 16-%3
+    psrldq  %4, %3
+%endif
+    por     %%dst, %4
+%endif
+%endmacro
+
+%macro PAVGB 2
+%if cpuflag(mmxext)
+    pavgb   %1, %2
+%elif cpuflag(3dnow)
+    pavgusb %1, %2
+%endif
+%endmacro
+
+%macro PSHUFLW 1+
+    %if mmsize == 8
+        pshufw %1
+    %else
+        pshuflw %1
+    %endif
+%endmacro
+
+%macro PSWAPD 2
+%if cpuflag(mmxext)
+    pshufw    %1, %2, q1032
+%elif cpuflag(3dnowext)
+    pswapd    %1, %2
+%elif cpuflag(3dnow)
+    movq      %1, %2
+    psrlq     %1, 32
+    punpckldq %1, %2
+%endif
+%endmacro
+
+%macro DEINTB 5 ; mask, reg1, mask, reg2, optional src to fill masks from
+%ifnum %5
+    pand   m%3, m%5, m%4 ; src .. y6 .. y4
+    pand   m%1, m%5, m%2 ; dst .. y6 .. y4
+%else
+    mova   m%1, %5
+    pand   m%3, m%1, m%4 ; src .. y6 .. y4
+    pand   m%1, m%1, m%2 ; dst .. y6 .. y4
+%endif
+    psrlw  m%2, 8        ; dst .. y7 .. y5
+    psrlw  m%4, 8        ; src .. y7 .. y5
+%endmacro
+
+%macro SUMSUB_BA 3-4
+%if %0==3
+    padd%1  m%2, m%3
+    padd%1  m%3, m%3
+    psub%1  m%3, m%2
+%else
+%if avx_enabled == 0
+    mova    m%4, m%2
+    padd%1  m%2, m%3
+    psub%1  m%3, m%4
+%else
+    padd%1  m%4, m%2, m%3
+    psub%1  m%3, m%2
+    SWAP    %2, %4
+%endif
+%endif
+%endmacro
+
+%macro SUMSUB_BADC 5-6
+%if %0==6
+    SUMSUB_BA %1, %2, %3, %6
+    SUMSUB_BA %1, %4, %5, %6
+%else
+    padd%1  m%2, m%3
+    padd%1  m%4, m%5
+    padd%1  m%3, m%3
+    padd%1  m%5, m%5
+    psub%1  m%3, m%2
+    psub%1  m%5, m%4
+%endif
+%endmacro
+
+%macro SUMSUB2_AB 4
+%ifnum %3
+    psub%1  m%4, m%2, m%3
+    psub%1  m%4, m%3
+    padd%1  m%2, m%2
+    padd%1  m%2, m%3
+%else
+    mova    m%4, m%2
+    padd%1  m%2, m%2
+    padd%1  m%2, %3
+    psub%1  m%4, %3
+    psub%1  m%4, %3
+%endif
+%endmacro
+
+%macro SUMSUB2_BA 4
+%if avx_enabled == 0
+    mova    m%4, m%2
+    padd%1  m%2, m%3
+    padd%1  m%2, m%3
+    psub%1  m%3, m%4
+    psub%1  m%3, m%4
+%else
+    padd%1  m%4, m%2, m%3
+    padd%1  m%4, m%3
+    psub%1  m%3, m%2
+    psub%1  m%3, m%2
+    SWAP     %2,  %4
+%endif
+%endmacro
+
+%macro SUMSUBD2_AB 5
+%ifnum %4
+    psra%1  m%5, m%2, 1  ; %3: %3>>1
+    psra%1  m%4, m%3, 1  ; %2: %2>>1
+    padd%1  m%4, m%2     ; %3: %3>>1+%2
+    psub%1  m%5, m%3     ; %2: %2>>1-%3
+    SWAP     %2, %5
+    SWAP     %3, %4
+%else
+    mova    %5, m%2
+    mova    %4, m%3
+    psra%1  m%3, 1  ; %3: %3>>1
+    psra%1  m%2, 1  ; %2: %2>>1
+    padd%1  m%3, %5 ; %3: %3>>1+%2
+    psub%1  m%2, %4 ; %2: %2>>1-%3
+%endif
+%endmacro
+
+%macro DCT4_1D 5
+%ifnum %5
+    SUMSUB_BADC w, %4, %1, %3, %2, %5
+    SUMSUB_BA   w, %3, %4, %5
+    SUMSUB2_AB  w, %1, %2, %5
+    SWAP %1, %3, %4, %5, %2
+%else
+    SUMSUB_BADC w, %4, %1, %3, %2
+    SUMSUB_BA   w, %3, %4
+    mova     [%5], m%2
+    SUMSUB2_AB  w, %1, [%5], %2
+    SWAP %1, %3, %4, %2
+%endif
+%endmacro
+
+%macro IDCT4_1D 6-7
+%ifnum %6
+    SUMSUBD2_AB %1, %3, %5, %7, %6
+    ; %3: %3>>1-%5 %5: %3+%5>>1
+    SUMSUB_BA   %1, %4, %2, %7
+    ; %4: %2+%4 %2: %2-%4
+    SUMSUB_BADC %1, %5, %4, %3, %2, %7
+    ; %5: %2+%4 + (%3+%5>>1)
+    ; %4: %2+%4 - (%3+%5>>1)
+    ; %3: %2-%4 + (%3>>1-%5)
+    ; %2: %2-%4 - (%3>>1-%5)
+%else
+%ifidn %1, w
+    SUMSUBD2_AB %1, %3, %5, [%6], [%6+16]
+%else
+    SUMSUBD2_AB %1, %3, %5, [%6], [%6+32]
+%endif
+    SUMSUB_BA   %1, %4, %2
+    SUMSUB_BADC %1, %5, %4, %3, %2
+%endif
+    SWAP %2, %5, %4
+    ; %2: %2+%4 + (%3+%5>>1) row0
+    ; %3: %2-%4 + (%3>>1-%5) row1
+    ; %4: %2-%4 - (%3>>1-%5) row2
+    ; %5: %2+%4 - (%3+%5>>1) row3
+%endmacro
+
+
+%macro LOAD_DIFF 5
+%ifidn %3, none
+    movh       %1, %4
+    movh       %2, %5
+    punpcklbw  %1, %2
+    punpcklbw  %2, %2
+    psubw      %1, %2
+%else
+    movh       %1, %4
+    punpcklbw  %1, %3
+    movh       %2, %5
+    punpcklbw  %2, %3
+    psubw      %1, %2
+%endif
+%endmacro
+
+%macro STORE_DCT 6
+    movq   [%5+%6+ 0], m%1
+    movq   [%5+%6+ 8], m%2
+    movq   [%5+%6+16], m%3
+    movq   [%5+%6+24], m%4
+    movhps [%5+%6+32], m%1
+    movhps [%5+%6+40], m%2
+    movhps [%5+%6+48], m%3
+    movhps [%5+%6+56], m%4
+%endmacro
+
+%macro LOAD_DIFF_8x4P 7-10 r0,r2,0 ; 4x dest, 2x temp, 2x pointer, increment?
+    LOAD_DIFF m%1, m%5, m%7, [%8],      [%9]
+    LOAD_DIFF m%2, m%6, m%7, [%8+r1],   [%9+r3]
+    LOAD_DIFF m%3, m%5, m%7, [%8+2*r1], [%9+2*r3]
+    LOAD_DIFF m%4, m%6, m%7, [%8+r4],   [%9+r5]
+%if %10
+    lea %8, [%8+4*r1]
+    lea %9, [%9+4*r3]
+%endif
+%endmacro
+
+%macro DIFFx2 6-7
+    movh       %3, %5
+    punpcklbw  %3, %4
+    psraw      %1, 6
+    paddsw     %1, %3
+    movh       %3, %6
+    punpcklbw  %3, %4
+    psraw      %2, 6
+    paddsw     %2, %3
+    packuswb   %2, %1
+%endmacro
+
+%macro STORE_DIFF 4
+    movh       %2, %4
+    punpcklbw  %2, %3
+    psraw      %1, 6
+    paddsw     %1, %2
+    packuswb   %1, %1
+    movh       %4, %1
+%endmacro
+
+%macro STORE_DIFFx2 8 ; add1, add2, reg1, reg2, zero, shift, source, stride
+    movh       %3, [%7]
+    movh       %4, [%7+%8]
+    psraw      %1, %6
+    psraw      %2, %6
+    punpcklbw  %3, %5
+    punpcklbw  %4, %5
+    paddw      %3, %1
+    paddw      %4, %2
+    packuswb   %3, %5
+    packuswb   %4, %5
+    movh     [%7], %3
+    movh  [%7+%8], %4
+%endmacro
+
+%macro PMINUB 3 ; dst, src, ignored
+%if cpuflag(mmxext)
+    pminub   %1, %2
+%else ; dst, src, tmp
+    mova     %3, %1
+    psubusb  %3, %2
+    psubb    %1, %3
+%endif
+%endmacro
+
+%macro SPLATW 2-3 0
+%if mmsize == 16
+    pshuflw    %1, %2, (%3)*0x55
+    punpcklqdq %1, %1
+%elif cpuflag(mmxext)
+    pshufw     %1, %2, (%3)*0x55
+%else
+    %ifnidn %1, %2
+        mova       %1, %2
+    %endif
+    %if %3 & 2
+        punpckhwd  %1, %1
+    %else
+        punpcklwd  %1, %1
+    %endif
+    %if %3 & 1
+        punpckhwd  %1, %1
+    %else
+        punpcklwd  %1, %1
+    %endif
+%endif
+%endmacro
+
+%macro SPLATD 1
+%if mmsize == 8
+    punpckldq  %1, %1
+%elif cpuflag(sse2)
+    pshufd  %1, %1, 0
+%elif cpuflag(sse)
+    shufps  %1, %1, 0
+%endif
+%endmacro
+
+%macro CLIPW 3 ;(dst, min, max)
+    pmaxsw %1, %2
+    pminsw %1, %3
+%endmacro
+
+%macro PMINSD_MMX 3 ; dst, src, tmp
+    mova      %3, %2
+    pcmpgtd   %3, %1
+    pxor      %1, %2
+    pand      %1, %3
+    pxor      %1, %2
+%endmacro
+
+%macro PMAXSD_MMX 3 ; dst, src, tmp
+    mova      %3, %1
+    pcmpgtd   %3, %2
+    pand      %1, %3
+    pandn     %3, %2
+    por       %1, %3
+%endmacro
+
+%macro CLIPD_MMX 3-4 ; src/dst, min, max, tmp
+    PMINSD_MMX %1, %3, %4
+    PMAXSD_MMX %1, %2, %4
+%endmacro
+
+%macro CLIPD_SSE2 3-4 ; src/dst, min (float), max (float), unused
+    cvtdq2ps  %1, %1
+    minps     %1, %3
+    maxps     %1, %2
+    cvtps2dq  %1, %1
+%endmacro
+
+%macro CLIPD_SSE41 3-4 ;  src/dst, min, max, unused
+    pminsd  %1, %3
+    pmaxsd  %1, %2
+%endmacro
+
+%macro VBROADCASTSS 2 ; dst xmm/ymm, src m32
+%if cpuflag(avx)
+    vbroadcastss %1, %2
+%else ; sse
+    movss        %1, %2
+    shufps       %1, %1, 0
+%endif
+%endmacro
+
+%macro VBROADCASTSD 2 ; dst xmm/ymm, src m64
+%if cpuflag(avx) && mmsize == 32
+    vbroadcastsd %1, %2
+%elif cpuflag(sse3)
+    movddup      %1, %2
+%else ; sse2
+    movsd        %1, %2
+    movlhps      %1, %1
+%endif
+%endmacro
+
+%macro SHUFFLE_MASK_W 8
+    %rep 8
+        %if %1>=0x80
+            db %1, %1
+        %else
+            db %1*2
+            db %1*2+1
+        %endif
+        %rotate 1
+    %endrep
+%endmacro
+
+%macro PMOVSXWD 2; dst, src
+%if cpuflag(sse4)
+    pmovsxwd     %1, %2
+%else
+    %ifnidn %1, %2
+    mova         %1, %2
+    %endif
+    punpcklwd    %1, %1
+    psrad        %1, 16
+%endif
+%endmacro
+
+; Wrapper for non-FMA version of fmaddps
+%macro FMULADD_PS 5
+    %if cpuflag(fma3) || cpuflag(fma4)
+        fmaddps %1, %2, %3, %4
+    %elifidn %1, %4
+        mulps   %5, %2, %3
+        addps   %1, %4, %5
+    %else
+        mulps   %1, %2, %3
+        addps   %1, %4
+    %endif
+%endmacro