Commit Graph

574 Commits

Author SHA1 Message Date
Diego Biurrun
81905088a1 x86: h264dsp: K&R formatting cosmetics 2012-08-02 20:20:21 +02:00
Ronald S. Bultje
c728518b3c x86: fft: fix imdct_half() for AVX
Some calculations were changed in b6a3849 to use mmsize, which was not correct
for the AVX version, which uses INIT_YMM and therefore has mmsize == 32.

Fixes Bug 341.

Signed-off-by: Justin Ruggles <justin.ruggles@gmail.com>
2012-08-02 13:40:11 -04:00
Mans Rullgard
ec7c501ed5 x86: remove libmpeg2 mmx(ext) idct functions
These functions are not faster than other mmx implementations on
any hardware I have been able to test on, and they are horribly
inaccurate.  There is thus no reason to ever use them.

Signed-off-by: Mans Rullgard <mans@mansr.com>
2012-08-02 12:14:52 +01:00
Ronald S. Bultje
b6a3849adb fft: port FFT/IMDCT 3dnow functions to yasm, and disable on x86-64.
64-bit CPUs always have SSE available, thus there is no need to compile
in the 3dnow functions. This results in smaller binaries.
2012-07-31 21:20:47 -07:00
Ronald S. Bultje
53dfaedc01 x86/dsputilenc: bury inline asm under HAVE_INLINE_ASM. 2012-07-31 20:28:52 -07:00
Diego Biurrun
6376a3ad24 x86: h264dsp: Remove unused variable ff_pb_3_1 2012-08-01 00:17:16 +02:00
Diego Biurrun
8728b381cb x86: h264dsp: Adjust YASM #ifdefs
This fixes compilation with YASM disabled.
2012-07-31 13:54:07 +02:00
Ronald S. Bultje
b829b4ce29 h264: convert loop filter strength dsp function to yasm.
This completes the conversion of h264dsp to yasm; note that h264 also
uses some dsputil functions, most notably qpel. Performance-wise, the
yasm-version is ~10 cycles faster (182->172) on x86-64, and ~8 cycles
faster (201->193) on x86-32.
2012-07-30 19:39:47 -07:00
Ronald S. Bultje
c83f44dba1 h264_idct_10bit: port x86 assembly to cpuflags. 2012-07-28 08:29:45 -07:00
Ronald S. Bultje
b3c5ae5607 fft: rename "z" to "zc" to prevent name collision.
Without this, cglobal will expand "z" to "zh" to access the high byte
in a register's word, which causes a name collision with the ZH(x) macro
further up in this file.
2012-07-28 08:29:44 -07:00
Ronald S. Bultje
4d777eedfd vp3: don't compile mmx IDCT functions on x86-64.
64-bit CPUs always have SSE2, and a SSE2 version exists, thus the MMX
version will never be used.
2012-07-27 20:12:30 -07:00
Ronald S. Bultje
a5bbb1242c h264_loopfilter: port x86 simd to cpuflags. 2012-07-27 20:12:11 -07:00
Ronald S. Bultje
d07ff3cd5a h264_chromamc_10bit: port x86 simd to cpuflags. 2012-07-27 17:35:49 -07:00
Ronald S. Bultje
4a26fdd852 vp3: port x86 SIMD to cpuflags. 2012-07-27 17:35:49 -07:00
Ronald S. Bultje
76888c64b0 rv34: port x86 SIMD to cpuflags. 2012-07-27 15:13:26 -07:00
Ronald S. Bultje
158744a4cd vp56: only compile MMX SIMD on x86-32.
All x86-64 CPUs have SSE2, so the MMX version will never be used. This
leads to smaller binaries.
2012-07-27 14:40:27 -07:00
Ronald S. Bultje
2734ba787b vp56: port x86 simd to cpuflags. 2012-07-27 14:39:07 -07:00
Ronald S. Bultje
5361e10a5e proresdsp: port x86 assembly to cpuflags. 2012-07-27 11:43:06 -07:00
Ronald S. Bultje
bde73f28af mpegaudio: bury inline asm under HAVE_INLINE_ASM. 2012-07-26 13:43:16 -07:00
Ronald S. Bultje
30b45d9c38 x86inc: automatically insert vzeroupper for YMM functions. 2012-07-26 13:43:16 -07:00
Ronald S. Bultje
a1878a88a1 vp3: don't use calls to inline asm in yasm code.
Mixing yasm and inline asm is a bad idea, since if either yasm or inline
asm is not supported by your toolchain, all of the asm stops working.
Thus, better to use either one or the other alone.

Signed-off-by: Derek Buitenhuis <derek.buitenhuis@gmail.com>
2012-07-25 14:24:30 -04:00
Ronald S. Bultje
79195ce565 x86/dsputil: put inline asm under HAVE_INLINE_ASM.
This allows compiling with compilers that don't support gcc-style
inline assembly.

Signed-off-by: Derek Buitenhuis <derek.buitenhuis@gmail.com>
2012-07-25 14:24:27 -04:00
Yang Wang
845e92fd6a dsputil_mmx: fix incorrect assembly code
In ff_put_pixels_clamped_mmx(), there are two assembly code blocks.
In the first block (in the unrolled loop), the instructions
"movq 8%3, %%mm1 \n\t", and so forth, have problems.

From above instruction, it is clear what the programmer wants: a load from
p + 8. But this assembly code doesn’t guarantee that. It only works if the
compiler puts p in a register to produce an instruction like this:
"movq 8(%edi), %mm1". During compiler optimization, it is possible that the
compiler will be able to constant propagate into p. Suppose p = &x[10000].
Then operand 3 can become 10000(%edi), where %edi holds &x. And the instruction
becomes "movq 810000(%edx)". That is, it will stride by 810000 instead of 8.

This will cause a segmentation fault.

This error was fixed in the second block of the assembly code, but not in
the unrolled loop.

How to reproduce:
    This error is exposed when we build using Intel C++ Compiler, with
    IPO+PGO optimization enabled. Crashed when decoding an MJPEG video.

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Signed-off-by: Derek Buitenhuis <derek.buitenhuis@gmail.com>
2012-07-25 14:22:18 -04:00
Jason Garrett-Glaser
85a3c19ed1 dsputil: x86: add SHUFFLE_MASK_W macro
Simplifies pshufb masks that operate on words.
2012-07-22 16:56:58 -04:00
Diego Biurrun
9f97af2688 x86: dsputil: drop some unused CPU flag debug code 2012-07-19 10:17:56 +02:00
Mans Rullgard
28f9ab7029 vp3: move idct and loop filter pointers to new vp3dsp context
This moves all VP3-specific function pointers from dsputil to a
new vp3dsp context.  There is no reason to ever use the VP3 IDCT
where an MPEG2 IDCT is expected or vice versa.

Signed-off-by: Mans Rullgard <mans@mansr.com>
2012-07-18 10:32:19 +01:00
Mans Rullgard
ab9f987661 build: add CONFIG_VP3DSP, reduce repetition in OBJS lists
Signed-off-by: Mans Rullgard <mans@mansr.com>
2012-07-18 10:32:18 +01:00
Martin Storsjö
f27386cdc7 x86: h264_intrapred: Don't add the 'd' suffix to the SPLATB_REG macro
The SPLATB_REG macro already adds the 'd' suffix internally.

This fixes building on Win64, which has been broken since 878e66902.

This worked for unix, where r2 happened to be rdx in this case, which
with the first suffix rdxd was mapped to eax, and eaxd is defined back
to eax. On win64 however, r2 happened to be R8 in this case, and
R8d mapps to R8D just fine, but there's no mapping for R8Dd to anything.

Signed-off-by: Martin Storsjö <martin@martin.st>
2012-07-06 21:07:23 +03:00
Diego Biurrun
878e669029 x86: h264_intrapred: use newly introduced SPLAT* and PSHUFLW macros 2012-07-05 17:37:11 +02:00
Loren Merritt
4d4752366f x86inc: add SPLATB_LOAD, SPLATB_REG, PSHUFLW macros
Signed-off-by: Diego Biurrun <diego@biurrun.de>
2012-07-05 17:37:11 +02:00
Diego Biurrun
d20f133ef9 x86: h264_intrapred: port to cpuflag macros 2012-07-05 17:37:10 +02:00
Martin Storsjö
07eeeb1d4f vp8: Add ifdef guards around the sse2 loopfilter in the sse2slow branch too
This was missed in the the previous commit in 70a1c800.

Signed-off-by: Martin Storsjö <martin@martin.st>
2012-07-05 09:39:01 +03:00
Martin Storsjö
70a1c8000f vp8: loopfilter >=sse2 functions need aligned stack on x86-32.
Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
2012-07-04 08:25:50 -07:00
Ronald S. Bultje
723b266d72 dsputilenc: group yasm and inline asm function pointer assignment. 2012-07-04 07:46:27 -07:00
Ronald S. Bultje
ceabc13f12 dsputilenc_mmx: split assignment of ff_sse16_sse2 to SSE2 section. 2012-06-30 09:24:52 -07:00
Ronald S. Bultje
66a02159ea x86: fmtconvert: add special asm for float_to_int16_interleave_misc_*
This gets rid of a variable-length array and a for loop in C code.

Signed-off-by: Martin Storsjö <martin@martin.st>
2012-06-30 19:10:36 +03:00
Mans Rullgard
f2fd167835 x86: vc1: fix and enable optimised loop filter
The problem is that the ssse3 psign instruction does the wrong
thing here.  Commit ea60dfe incorrectly removed a macro emulating
this instruction for pre-ssse3 code.  However, the emulation is
incorrect, and the code relies on the behaviour of the macro.
Specifically, the psign sets destination elements to zero where
the corresponding source element is zero, whereas the emulation
only negates destination elements where the source is negative.

Furthermore, the PSIGNW_MMX macro in x86util.asm is totally bogus,
which is why the original VC-1 code had an additional right shift
when using it.  Since the psign instruction cannot be used here,
skip all the macro hell and use the working instruction sequence
directly.

None of this was noticed due a stray return statement in
ff_vc1dsp_init_mmx() which meant that only the mmx version of the
loop filter was ever used (before being removed in ea60dfe).

Signed-off-by: Mans Rullgard <mans@mansr.com>
2012-06-30 00:12:05 +01:00
Christophe Gisquet
a5bfa66df5 x86: fft: replace call to memcpy by a loop
The function call was a mess to handle, and memcpy cannot make
the assumptions we do in the new code.

Tested on an IMC sample: 430c -> 370c.

Signed-off-by: Mans Rullgard <mans@mansr.com>
2012-06-27 12:49:33 +01:00
Mans Rullgard
0595334892 x86: fft: elf64: fix PIC build
In a 64-bit PIC build, external functions must be called
through the PLT.

Signed-off-by: Mans Rullgard <mans@mansr.com>
2012-06-25 22:58:18 +01:00
Mans Rullgard
8725da49a2 x86: fft: win64: fix stack alignment for memcpy() call 2012-06-25 15:10:39 +01:00
Mans Rullgard
8299260470 x86: fft: convert sse inline asm to yasm 2012-06-25 13:31:00 +01:00
Ronald S. Bultje
8123e0901f x86: place some inline asm under #if HAVE_INLINE_ASM
Signed-off-by: Mans Rullgard <mans@mansr.com>
2012-06-25 13:23:12 +01:00
Mans Rullgard
0b6f973635 h264: use asm cabac reader under a generic condition
This removes a dependency on implementation details from generic
code and allows easy addition of the equivalent optimisation for
other architectures than x86.

Signed-off-by: Mans Rullgard <mans@mansr.com>
2012-06-23 22:14:21 +01:00
Diego Biurrun
fe07c9c6b5 x86: Only use optimizations with cmov if the CPU supports the instruction 2012-06-23 16:21:50 +02:00
Mans Rullgard
29686d6ea3 x86: remove unused inline asm macros from dsputil_mmx.h
Signed-off-by: Mans Rullgard <mans@mansr.com>
2012-06-23 14:14:06 +01:00
Mans Rullgard
685f5438bb x86: move some inline asm macros to the only places they are used
Signed-off-by: Mans Rullgard <mans@mansr.com>
2012-06-23 14:14:06 +01:00
Diego Biurrun
a5a93fa8f5 cosmetics: do not use full path for local headers 2012-06-22 10:49:40 +02:00
Ronald S. Bultje
d9669eab0b dwt: remove variable-length arrays
Signed-off-by: Mans Rullgard <mans@mansr.com>
2012-06-17 23:20:10 +01:00
Justin Ruggles
d5a7229ba4 Add a float DSP framework to libavutil
Move vector_fmul() from DSPContext to AVFloatDSPContext.
2012-06-08 13:14:38 -04:00
Vitor Sessak
bac0729d9e x86: use new schema for ASM macros
Signed-off-by: Janne Grunau <janne-libav@jannau.net>
2012-05-29 14:49:45 +02:00