h264: assembly version of get_cabac for x86_64 with PIC (v4)

This adds a hand-optimized assembly version for get_cabac much like the
existing one, but it works if the table offsets are RIP-relative.
Compared to the non-RIP-relative version this adds 2 lea instructions
and it needs one extra register.
There is a surprisingly large performance improvement over the c version (more
so than the generated assembly seems to suggest) just in get_cabac, I measured
roughly 40% faster for get_cabac on a K8. However, overall the difference is
not that big, I measured roughly 5% on a test clip on a K8 and a Core2.
Hopefully it still compiles on x86 32bit...
v2: incorporated feedback from Loren Merritt to avoid rip-relative movs
for every table, and got rid of unnecessary @GOTPCREL.
v3: apply similar fixes to the the decode_significance functions, and use
same macro arguments for non-pic case.
v4: prettify inline asm arguments, add a non-fast-cmov version (as I expect
the c code to be faster otherwise since both cmov and sbb suck hard on a
Prescott, even can't construct the mask with a 64bit shift as that's just as
terrible - it's quite difficult to find usable instructions on that chip...).
This is tested to work but not on a P4, in theory it _should_ be fast there.

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

3 files changed, 125 insertions, 25 deletions

diff --git a/libavcodec/h264_cabac.c b/libavcodec/h264_cabac.c
index 65f2cb4dc8..6c59d174a2 100644
--- a/libavcodec/h264_cabac.c
+++ b/libavcodec/h264_cabac.c
@@ -1659,7 +1659,7 @@ decode_cabac_residual_internal(H264Context *h, DCTELEM *block,
             index[coeff_count++] = last;\
         }
         const uint8_t *sig_off = significant_coeff_flag_offset_8x8[MB_FIELD];
-#if ARCH_X86 && HAVE_7REGS && !defined(BROKEN_RELOCATIONS)
+#if ARCH_X86 && HAVE_7REGS
         coeff_count= decode_significance_8x8_x86(CC, significant_coeff_ctx_base, index,
                                                  last_coeff_ctx_base, sig_off);
     } else {
diff --git a/libavcodec/x86/cabac.h b/libavcodec/x86/cabac.h
index 0c4419b4f3..5175daff70 100644
--- a/libavcodec/x86/cabac.h
+++ b/libavcodec/x86/cabac.h
@@ -24,8 +24,71 @@
 #include "libavcodec/cabac.h"
 #include "libavutil/attributes.h"
 #include "libavutil/x86_cpu.h"
+#include "libavutil/internal.h"
 #include "config.h"
 
+#ifdef BROKEN_RELOCATIONS
+#define RIP_ARG , "r"(rip)
+
+#if HAVE_FAST_CMOV
+#define BRANCHLESS_GET_CABAC_UPDATE(ret, retq, low, range, tmp) \
+        "cmp    "low"       , "tmp"                        \n\t"\
+        "cmova  %%ecx       , "range"                      \n\t"\
+        "sbb    %%rcx       , %%rcx                        \n\t"\
+        "and    %%ecx       , "tmp"                        \n\t"\
+        "xor    %%rcx       , "retq"                       \n\t"\
+        "sub    "tmp"       , "low"                        \n\t"
+#else /* HAVE_FAST_CMOV */
+#define BRANCHLESS_GET_CABAC_UPDATE(ret, retq, low, range, tmp) \
+/* P4 Prescott has crappy cmov,sbb,64bit shift so avoid them */ \
+        "sub    "low"       , "tmp"                        \n\t"\
+        "sar    $31         , "tmp"                        \n\t"\
+        "sub    %%ecx       , "range"                      \n\t"\
+        "and    "tmp"       , "range"                      \n\t"\
+        "add    %%ecx       , "range"                      \n\t"\
+        "shl    $17         , %%ecx                        \n\t"\
+        "and    "tmp"       , %%ecx                        \n\t"\
+        "sub    %%ecx       , "low"                        \n\t"\
+        "xor    "tmp"       , "ret"                        \n\t"\
+        "movslq "ret"       , "retq"                       \n\t"
+#endif /* HAVE_FAST_CMOV */
+
+#define BRANCHLESS_GET_CABAC(ret, retq, statep, low, lowword, range, rangeq, tmp, tmpbyte, byte, end, rip) \
+        "movzbl "statep"    , "ret"                                     \n\t"\
+        "mov    "range"     , "tmp"                                     \n\t"\
+        "and    $0xC0       , "range"                                   \n\t"\
+        "lea    ("ret", "range", 2), %%ecx                              \n\t"\
+        "movzbl ff_h264_lps_range-1b("rip", %%rcx), "range"             \n\t"\
+        "sub    "range"     , "tmp"                                     \n\t"\
+        "mov    "tmp"       , %%ecx                                     \n\t"\
+        "shl    $17         , "tmp"                                     \n\t"\
+        BRANCHLESS_GET_CABAC_UPDATE(ret, retq, low, range, tmp)              \
+        "movzbl ff_h264_norm_shift-1b("rip", "rangeq"), %%ecx           \n\t"\
+        "shl    %%cl        , "range"                                   \n\t"\
+        "movzbl ff_h264_mlps_state-1b+128("rip", "retq"), "tmp"         \n\t"\
+        "shl    %%cl        , "low"                                     \n\t"\
+        "mov    "tmpbyte"   , "statep"                                  \n\t"\
+        "test   "lowword"   , "lowword"                                 \n\t"\
+        " jnz   2f                                                      \n\t"\
+        "mov    "byte"      , %%"REG_c"                                 \n\t"\
+        "add"OPSIZE" $2     , "byte"                                    \n\t"\
+        "movzwl (%%"REG_c") , "tmp"                                     \n\t"\
+        "lea    -1("low")   , %%ecx                                     \n\t"\
+        "xor    "low"       , %%ecx                                     \n\t"\
+        "shr    $15         , %%ecx                                     \n\t"\
+        "bswap  "tmp"                                                   \n\t"\
+        "shr    $15         , "tmp"                                     \n\t"\
+        "movzbl ff_h264_norm_shift-1b("rip", %%rcx), %%ecx              \n\t"\
+        "sub    $0xFFFF     , "tmp"                                     \n\t"\
+        "neg    %%ecx                                                   \n\t"\
+        "add    $7          , %%ecx                                     \n\t"\
+        "shl    %%cl        , "tmp"                                     \n\t"\
+        "add    "tmp"       , "low"                                     \n\t"\
+        "2:                                                             \n\t"
+
+#else /* BROKEN_RELOCATIONS */
+#define RIP_ARG
+
 #if HAVE_FAST_CMOV
 #define BRANCHLESS_GET_CABAC_UPDATE(ret, statep, low, range, tmp)\
         "mov    "tmp"       , %%ecx     \n\t"\
@@ -51,7 +114,7 @@
         "xor    "tmp"       , "ret"     \n\t"
 #endif /* HAVE_FAST_CMOV */
 
-#define BRANCHLESS_GET_CABAC(ret, statep, low, lowword, range, tmp, tmpbyte, byte, end) \
+#define BRANCHLESS_GET_CABAC(ret, retq, statep, low, lowword, range, rangeq, tmp, tmpbyte, byte, end, rip) \
         "movzbl "statep"    , "ret"                                     \n\t"\
         "mov    "range"     , "tmp"                                     \n\t"\
         "and    $0xC0       , "range"                                   \n\t"\
@@ -81,28 +144,39 @@
         "add    "tmp"       , "low"                                     \n\t"\
         "2:                                                             \n\t"
 
+#endif /* BROKEN_RELOCATIONS */
 
-#if HAVE_7REGS && !defined(BROKEN_RELOCATIONS) && !(defined(__i386) && defined(__clang__) && (__clang_major__<2 || (__clang_major__==2 && __clang_minor__<10)))\
-                                               && !(defined(__i386) && !defined(__clang__) && defined(__llvm__) && __GNUC__==4 && __GNUC_MINOR__==2 && __GNUC_PATCHLEVEL__<=1)
+
+#if HAVE_7REGS && !(defined(__i386) && defined(__clang__) && (__clang_major__<2 || (__clang_major__==2 && __clang_minor__<10)))\
+               && !(defined(__i386) && !defined(__clang__) && defined(__llvm__) && __GNUC__==4 && __GNUC_MINOR__==2 && __GNUC_PATCHLEVEL__<=1)
 #define get_cabac_inline get_cabac_inline_x86
 static av_always_inline int get_cabac_inline_x86(CABACContext *c,
                                                  uint8_t *const state)
 {
     int bit, tmp;
+#ifdef BROKEN_RELOCATIONS
+    int *rip;
+
+    __asm__ volatile(
+        "1:                        \n\t"
+        "lea    1b(%%rip), %0      \n\t"
+        : "=&r"(rip)
+    );
+#endif
 
     __asm__ volatile(
-        BRANCHLESS_GET_CABAC("%0", "(%4)", "%1", "%w1",
-                             "%2", "%3", "%b3",
-                             "%a6(%5)", "%a7(%5)")
+        BRANCHLESS_GET_CABAC("%0", "%q0", "(%4)", "%1", "%w1",
+                             "%2", "%q2", "%3", "%b3",
+                             "%a6(%5)", "%a7(%5)", "%8")
         : "=&r"(bit), "+&r"(c->low), "+&r"(c->range), "=&q"(tmp)
         : "r"(state), "r"(c),
           "i"(offsetof(CABACContext, bytestream)),
-          "i"(offsetof(CABACContext, bytestream_end))
+          "i"(offsetof(CABACContext, bytestream_end)) RIP_ARG
         : "%"REG_c, "memory"
     );
     return bit & 1;
 }
-#endif /* HAVE_7REGS && !defined(BROKEN_RELOCATIONS) */
+#endif /* HAVE_7REGS */
 
 #define get_cabac_bypass_sign get_cabac_bypass_sign_x86
 static av_always_inline int get_cabac_bypass_sign_x86(CABACContext *c, int val)
diff --git a/libavcodec/x86/h264_i386.h b/libavcodec/x86/h264_i386.h
index 6aa2d073e3..20ffa25f08 100644
--- a/libavcodec/x86/h264_i386.h
+++ b/libavcodec/x86/h264_i386.h
@@ -36,7 +36,7 @@
 
 //FIXME use some macros to avoid duplicating get_cabac (cannot be done yet
 //as that would make optimization work hard)
-#if HAVE_7REGS && !defined(BROKEN_RELOCATIONS)
+#if HAVE_7REGS
 static int decode_significance_x86(CABACContext *c, int max_coeff,
                                    uint8_t *significant_coeff_ctx_base,
                                    int *index, x86_reg last_off){
@@ -45,20 +45,31 @@ static int decode_significance_x86(CABACContext *c, int max_coeff,
     int minusindex= 4-(intptr_t)index;
     int bit;
     x86_reg coeff_count;
+
+#ifdef BROKEN_RELOCATIONS
+    int *rip;
+
+    __asm__ volatile(
+        "1:                        \n\t"
+        "lea    1b(%%rip), %0      \n\t"
+        : "=&r"(rip)
+    );
+#endif
+
     __asm__ volatile(
         "3:                                     \n\t"
 
-        BRANCHLESS_GET_CABAC("%4", "(%1)", "%3", "%w3",
-                             "%5", "%k0", "%b0",
-                             "%a11(%6)", "%a12(%6)")
+        BRANCHLESS_GET_CABAC("%4", "%q4", "(%1)", "%3", "%w3",
+                             "%5", "%q5", "%k0", "%b0",
+                             "%a11(%6)", "%a12(%6)", "%13")
 
         "test $1, %4                            \n\t"
         " jz 4f                                 \n\t"
         "add  %10, %1                           \n\t"
 
-        BRANCHLESS_GET_CABAC("%4", "(%1)", "%3", "%w3",
-                             "%5", "%k0", "%b0",
-                             "%a11(%6)", "%a12(%6)")
+        BRANCHLESS_GET_CABAC("%4", "%q4", "(%1)", "%3", "%w3",
+                             "%5", "%q5", "%k0", "%b0",
+                             "%a11(%6)", "%a12(%6)", "%13")
 
         "sub  %10, %1                           \n\t"
         "mov  %2, %0                            \n\t"
@@ -86,7 +97,7 @@ static int decode_significance_x86(CABACContext *c, int max_coeff,
           "+&r"(c->low), "=&r"(bit), "+&r"(c->range)
         : "r"(c), "m"(minusstart), "m"(end), "m"(minusindex), "m"(last_off),
           "i"(offsetof(CABACContext, bytestream)),
-          "i"(offsetof(CABACContext, bytestream_end))
+          "i"(offsetof(CABACContext, bytestream_end)) RIP_ARG
         : "%"REG_c, "memory"
     );
     return coeff_count;
@@ -100,6 +111,17 @@ static int decode_significance_8x8_x86(CABACContext *c,
     x86_reg coeff_count;
     x86_reg last=0;
     x86_reg state;
+
+#ifdef BROKEN_RELOCATIONS
+    int *rip;
+
+    __asm__ volatile(
+        "1:                        \n\t"
+        "lea    1b(%%rip), %0      \n\t"
+        : "=&r"(rip)
+    );
+#endif
+
     __asm__ volatile(
         "mov %1, %6                             \n\t"
         "3:                                     \n\t"
@@ -108,20 +130,24 @@ static int decode_significance_8x8_x86(CABACContext *c,
         "movzbl (%0, %6), %k6                   \n\t"
         "add %9, %6                             \n\t"
 
-        BRANCHLESS_GET_CABAC("%4", "(%6)", "%3", "%w3",
-                             "%5", "%k0", "%b0",
-                             "%a12(%7)", "%a13(%7)")
+        BRANCHLESS_GET_CABAC("%4", "%q4", "(%6)", "%3", "%w3",
+                             "%5", "%q5", "%k0", "%b0",
+                             "%a12(%7)", "%a13(%7)", "%14")
 
         "mov %1, %k6                            \n\t"
         "test $1, %4                            \n\t"
         " jz 4f                                 \n\t"
 
-        "movzbl "MANGLE(last_coeff_flag_offset_8x8)"(%k6), %k6\n\t"
+#ifdef BROKEN_RELOCATIONS
+        "movzbl last_coeff_flag_offset_8x8-1b(%14, %q6), %k6\n\t"
+#else
+        "movzbl last_coeff_flag_offset_8x8(%k6), %k6\n\t"
+#endif
         "add %11, %6                            \n\t"
 
-        BRANCHLESS_GET_CABAC("%4", "(%6)", "%3", "%w3",
-                             "%5", "%k0", "%b0",
-                             "%a12(%7)", "%a13(%7)")
+        BRANCHLESS_GET_CABAC("%4", "%q4", "(%6)", "%3", "%w3",
+                             "%5", "%q5", "%k0", "%b0",
+                             "%a12(%7)", "%a13(%7)", "%14")
 
         "mov %2, %0                             \n\t"
         "mov %1, %k6                            \n\t"
@@ -147,7 +173,7 @@ static int decode_significance_8x8_x86(CABACContext *c,
         : "r"(c), "m"(minusindex), "m"(significant_coeff_ctx_base),
           "m"(sig_off), "m"(last_coeff_ctx_base),
           "i"(offsetof(CABACContext, bytestream)),
-          "i"(offsetof(CABACContext, bytestream_end))
+          "i"(offsetof(CABACContext, bytestream_end)) RIP_ARG
         : "%"REG_c, "memory"
     );
     return coeff_count;