X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libavcodec%2Fx86%2Fh264_i386.h;h=10ea32e0b866c2693da9571c53ab9894f44b7203;hb=40f81769aec24a93563cd547e02ba8478aa95849;hp=31ddaf6ee9e42776d695546fc5167cf4454d72a5;hpb=448dc42571edc5bc91da7b0b017daa61118ba2f5;p=ffmpeg

diff --git a/libavcodec/x86/h264_i386.h b/libavcodec/x86/h264_i386.h
index 31ddaf6ee9e..10ea32e0b86 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){
@@ -45,18 +45,30 @@ 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
+    void *tables;
+
+    __asm__ volatile(
+        "lea   "MANGLE(ff_h264_cabac_tables)", %0      \n\t"
+        : "=&r"(tables)
+    );
+#endif
+
     __asm__ volatile(
         "3:                                     \n\t"
 
-        BRANCHLESS_GET_CABAC("%4", "(%1)", "%3", "%w3",
-                             "%5", "%k0", "%b0", "%a11(%6)")
+        BRANCHLESS_GET_CABAC("%4", "%q4", "(%1)", "%3", "%w3",
+                             "%5", "%q5", "%k0", "%b0",
+                             "%a11(%6)", "%a12(%6)", "%a13", "%a14", "%a15", "%16")
 
         "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)")
+        BRANCHLESS_GET_CABAC("%4", "%q4", "(%1)", "%3", "%w3",
+                             "%5", "%q5", "%k0", "%b0",
+                             "%a11(%6)", "%a12(%6)", "%a13", "%a14", "%a15", "%16")
 
         "sub  %10, %1                           \n\t"
         "mov  %2, %0                            \n\t"
@@ -83,7 +95,11 @@ static int decode_significance_x86(CABACContext *c, int max_coeff,
         : "=&q"(coeff_count), "+r"(significant_coeff_ctx_base), "+m"(index),
           "+&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)),
+          "i"(offsetof(CABACContext, bytestream_end)),
+          "i"(H264_NORM_SHIFT_OFFSET),
+          "i"(H264_LPS_RANGE_OFFSET),
+          "i"(H264_MLPS_STATE_OFFSET) TABLES_ARG
         : "%"REG_c, "memory"
     );
     return coeff_count;
@@ -97,6 +113,16 @@ static int decode_significance_8x8_x86(CABACContext *c,
     x86_reg coeff_count;
     x86_reg last=0;
     x86_reg state;
+
+#ifdef BROKEN_RELOCATIONS
+    void *tables;
+
+    __asm__ volatile(
+        "lea    "MANGLE(ff_h264_cabac_tables)", %0      \n\t"
+        : "=&r"(tables)
+    );
+#endif
+
     __asm__ volatile(
         "mov %1, %6                             \n\t"
         "3:                                     \n\t"
@@ -105,18 +131,24 @@ 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)")
+        BRANCHLESS_GET_CABAC("%4", "%q4", "(%6)", "%3", "%w3",
+                             "%5", "%q5", "%k0", "%b0",
+                             "%a12(%7)", "%a13(%7)", "%a14", "%a15", "%a16", "%18")
 
         "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 %a17(%18, %q6), %k6\n\t"
+#else
+        "movzbl "MANGLE(ff_h264_cabac_tables)"+%a17(%k6), %k6\n\t"
+#endif
         "add %11, %6                            \n\t"
 
-        BRANCHLESS_GET_CABAC("%4", "(%6)", "%3", "%w3",
-                             "%5", "%k0", "%b0", "%a12(%7)")
+        BRANCHLESS_GET_CABAC("%4", "%q4", "(%6)", "%3", "%w3",
+                             "%5", "%q5", "%k0", "%b0",
+                             "%a12(%7)", "%a13(%7)", "%a14", "%a15", "%a16", "%18")
 
         "mov %2, %0                             \n\t"
         "mov %1, %k6                            \n\t"
@@ -141,7 +173,12 @@ static int decode_significance_8x8_x86(CABACContext *c,
           "=&r"(bit), "+&r"(c->range), "=&r"(state)
         : "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)),
+          "i"(offsetof(CABACContext, bytestream_end)),
+          "i"(H264_NORM_SHIFT_OFFSET),
+          "i"(H264_LPS_RANGE_OFFSET),
+          "i"(H264_MLPS_STATE_OFFSET),
+          "i"(H264_LAST_COEFF_FLAG_OFFSET_8x8_OFFSET) TABLES_ARG
         : "%"REG_c, "memory"
     );
     return coeff_count;