#include "libavcodec/cabac.h"
#include "libavutil/attributes.h"
-#include "libavutil/x86_cpu.h"
+#include "libavutil/macros.h"
+#include "libavutil/x86/asm.h"
#include "config.h"
+#if HAVE_INLINE_ASM
+
+#ifdef BROKEN_RELOCATIONS
+#define TABLES_ARG , "r"(tables)
+
+#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, norm_off, lps_off, mlps_off, tables) \
+ "movzbl "statep" , "ret" \n\t"\
+ "mov "range" , "tmp" \n\t"\
+ "and $0xC0 , "range" \n\t"\
+ "lea ("ret", "range", 2), %%ecx \n\t"\
+ "movzbl "lps_off"("tables", %%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 "norm_off"("tables", "rangeq"), %%ecx \n\t"\
+ "shl %%cl , "range" \n\t"\
+ "movzbl "mlps_off"+128("tables", "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"\
+ "cmp "end" , %%"REG_c" \n\t"\
+ "jge 1f \n\t"\
+ "add"OPSIZE" $2 , "byte" \n\t"\
+ "1: \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 "norm_off"("tables", %%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 TABLES_ARG
+
#if HAVE_FAST_CMOV
#define BRANCHLESS_GET_CABAC_UPDATE(ret, low, range, tmp)\
"mov "tmp" , %%ecx \n\t"\
"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, norm_off, lps_off, mlps_off, tables) \
"movzbl "statep" , "ret" \n\t"\
"mov "range" , "tmp" \n\t"\
"and $0xC0 , "range" \n\t"\
- "movzbl "MANGLE(ff_h264_lps_range)"("ret", "range", 2), "range" \n\t"\
+ "movzbl "MANGLE(ff_h264_cabac_tables)"+"lps_off"("ret", "range", 2), "range" \n\t"\
"sub "range" , "tmp" \n\t"\
BRANCHLESS_GET_CABAC_UPDATE(ret, low, range, tmp) \
- "movzbl " MANGLE(ff_h264_norm_shift) "("range"), %%ecx \n\t"\
+ "movzbl "MANGLE(ff_h264_cabac_tables)"+"norm_off"("range"), %%ecx \n\t"\
"shl %%cl , "range" \n\t"\
- "movzbl "MANGLE(ff_h264_mlps_state)"+128("ret"), "tmp" \n\t"\
+ "movzbl "MANGLE(ff_h264_cabac_tables)"+"mlps_off"+128("ret"), "tmp" \n\t"\
"shl %%cl , "low" \n\t"\
"mov "tmpbyte" , "statep" \n\t"\
"test "lowword" , "lowword" \n\t"\
"shr $15 , %%ecx \n\t"\
"bswap "tmp" \n\t"\
"shr $15 , "tmp" \n\t"\
- "movzbl " MANGLE(ff_h264_norm_shift) "(%%ecx), %%ecx \n\t"\
+ "movzbl "MANGLE(ff_h264_cabac_tables)"+"norm_off"(%%ecx), %%ecx \n\t"\
"sub $0xFFFF , "tmp" \n\t"\
"neg %%ecx \n\t"\
"add $7 , %%ecx \n\t"\
"add "tmp" , "low" \n\t"\
"2: \n\t"
-#if HAVE_7REGS && !defined(BROKEN_RELOCATIONS)
+#endif /* BROKEN_RELOCATIONS */
+
+
+#if HAVE_7REGS
#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
+ void *tables;
__asm__ volatile(
- BRANCHLESS_GET_CABAC("%0", "(%4)", "%1", "%w1",
- "%2", "%3", "%b3",
- "%a6(%5)", "%a7(%5)")
+ "lea "MANGLE(ff_h264_cabac_tables)", %0 \n\t"
+ : "=&r"(tables)
+ );
+#endif
+
+ __asm__ volatile(
+ BRANCHLESS_GET_CABAC("%0", "%q0", "(%4)", "%1", "%w1",
+ "%2", "%q2", "%3", "%b3",
+ "%c6(%5)", "%c7(%5)",
+ AV_STRINGIFY(H264_NORM_SHIFT_OFFSET),
+ AV_STRINGIFY(H264_LPS_RANGE_OFFSET),
+ AV_STRINGIFY(H264_MLPS_STATE_OFFSET),
+ "%8")
: "=&r"(bit), "+&r"(c->low), "+&r"(c->range), "=&q"(tmp)
: "r"(state), "r"(c),
"i"(offsetof(CABACContext, bytestream)),
"i"(offsetof(CABACContext, bytestream_end))
+ TABLES_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)
{
x86_reg tmp;
__asm__ volatile(
- "movl %a6(%2), %k1 \n\t"
- "movl %a3(%2), %%eax \n\t"
+ "movl %c6(%2), %k1 \n\t"
+ "movl %c3(%2), %%eax \n\t"
"shl $17, %k1 \n\t"
"add %%eax, %%eax \n\t"
"sub %k1, %%eax \n\t"
"sub %%edx, %%ecx \n\t"
"test %%ax, %%ax \n\t"
"jnz 1f \n\t"
- "mov %a4(%2), %1 \n\t"
+ "mov %c4(%2), %1 \n\t"
"subl $0xFFFF, %%eax \n\t"
"movzwl (%1), %%edx \n\t"
"bswap %%edx \n\t"
"shrl $15, %%edx \n\t"
"addl %%edx, %%eax \n\t"
- "cmp %a5(%2), %1 \n\t"
+ "cmp %c5(%2), %1 \n\t"
"jge 1f \n\t"
- "add"OPSIZE" $2, %a4(%2) \n\t"
+ "add"OPSIZE" $2, %c4(%2) \n\t"
"1: \n\t"
- "movl %%eax, %a3(%2) \n\t"
+ "movl %%eax, %c3(%2) \n\t"
: "+c"(val), "=&r"(tmp)
: "r"(c),
return val;
}
+#define get_cabac_bypass get_cabac_bypass_x86
+static av_always_inline int get_cabac_bypass_x86(CABACContext *c)
+{
+ x86_reg tmp;
+ int res;
+ __asm__ volatile(
+ "movl %c6(%2), %k1 \n\t"
+ "movl %c3(%2), %%eax \n\t"
+ "shl $17, %k1 \n\t"
+ "add %%eax, %%eax \n\t"
+ "sub %k1, %%eax \n\t"
+ "cltd \n\t"
+ "and %%edx, %k1 \n\t"
+ "add %k1, %%eax \n\t"
+ "inc %%edx \n\t"
+ "test %%ax, %%ax \n\t"
+ "jnz 1f \n\t"
+ "mov %c4(%2), %1 \n\t"
+ "subl $0xFFFF, %%eax \n\t"
+ "movzwl (%1), %%ecx \n\t"
+ "bswap %%ecx \n\t"
+ "shrl $15, %%ecx \n\t"
+ "addl %%ecx, %%eax \n\t"
+ "cmp %c5(%2), %1 \n\t"
+ "jge 1f \n\t"
+ "add"OPSIZE" $2, %c4(%2) \n\t"
+ "1: \n\t"
+ "movl %%eax, %c3(%2) \n\t"
+
+ : "=&d"(res), "=&r"(tmp)
+ : "r"(c),
+ "i"(offsetof(CABACContext, low)),
+ "i"(offsetof(CABACContext, bytestream)),
+ "i"(offsetof(CABACContext, bytestream_end)),
+ "i"(offsetof(CABACContext, range))
+ : "%eax", "%ecx", "memory"
+ );
+ return res;
+}
+
+#endif /* HAVE_INLINE_ASM */
#endif /* AVCODEC_X86_CABAC_H */
projects / ffmpeg / blobdiff