Welcome to mirror list, hosted at ThFree Co, Russian Federation.

github.com/FFmpeg/FFmpeg.git - Unnamed repository; edit this file 'description' to name the repository.
summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorRoland Scheidegger <rscheidegger_lists@hispeed.ch>2012-04-20 04:10:57 +0400
committerMichael Niedermayer <michaelni@gmx.at>2012-04-21 02:27:06 +0400
commita812b599b504b39a8021827da89d5e23fb361cc9 (patch)
tree34b3f8ce41b8cdca8df4f3b84682731b09c94f34 /libavcodec/x86/h264_i386.h
parent3194ab78a6c4ea0a4c60c91c4d0ea34028ca408f (diff)
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>
Diffstat (limited to 'libavcodec/x86/h264_i386.h')
-rw-r--r--libavcodec/x86/h264_i386.h58
1 files changed, 42 insertions, 16 deletions
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;