lavu/x86: add FFT assembly
This commit adds a pure x86 assembly SIMD version of the FFT in libavutil/tx.
The design of this pure assembly FFT is pretty unconventional.
On the lowest level, instead of splitting the complex numbers into
real and imaginary parts, we keep complex numbers together but split
them in terms of parity. This saves a number of shuffles in each transform,
but more importantly, it splits each transform into two independent
paths, which we process using separate registers in parallel.
This allows us to keep all units saturated and lets us use all available
registers to avoid dependencies.
Moreover, it allows us to double the granularity of our per-load permutation,
skipping many expensive lookups and allowing us to use just 4 loads per register,
rather than 8, or in case FMA3 (and by extension, AVX2), use the vgatherdpd
instruction, which is at least as fast as 4 separate loads on old hardware,
and quite a bit faster on modern CPUs).
Higher up, we go for a bottom-up construction of large transforms, foregoing
the traditional per-transform call-return recursion chains. Instead, we always
start at the bottom-most basis transform (in this case, a 32-point transform),
and continue constructing larger and larger transforms until we return to the
top-most transform.
This way, we only touch the stack 3 times per a complete target transform:
once for the 1/2 length transform and two times for the 1/4 length transform.
The combination algorithm we use is a standard Split-Radix algorithm,
as used in our C code. Although a version with less operations exists
(Steven G. Johnson and Matteo Frigo's "A modified split-radix FFT with fewer
arithmetic operations", IEEE Trans. Signal Process. 55 (1), 111–119 (2007),
which is the one FFTW uses), it only has 2% less operations and requires at least 4x
the binary code (due to it needing 4 different paths to do a single transform).
That version also has other issues which prevent it from being implemented
with SIMD code as efficiently, which makes it lose the marginal gains it offered,
and cannot be performed bottom-up, requiring many recursive call-return chains,
whose overhead adds up.
We go through a lot of effort to minimize load/stores by keeping as much in
registers in between construcring transforms. This saves us around 32 cycles,
on paper, but in reality a lot more due to load/store aliasing (a load from a
memory location cannot be issued while there's a store pending, and there are
only so many (2 for Zen 3) load/store units in a CPU).
Also, we interleave coefficients during the last stage to save on a store+load
per register.
Each of the smallest, basis transforms (4, 8 and 16-point in our case)
has been extremely optimized. Our 8-point transform is barely 20 instructions
in total, beating our old implementation 8-point transform by 1 instruction.
Our 2x8-point transform is 23 instructions, beating our old implementation by
6 instruction and needing 50% less cycles. Our 16-point transform's combination
code takes slightly more instructions than our old implementation, but makes up
for it by requiring a lot less arithmetic operations.
Overall, the transform was optimized for the timings of Zen 3, which at the
time of writing has the most IPC from all documented CPUs. Shuffles were
preferred over arithmetic operations due to their 1/0.5 latency/throughput.
On average, this code is 30% faster than our old libavcodec implementation.
It's able to trade blows with the previously-untouchable FFTW on small transforms,
and due to its tiny size and better prediction, outdoes FFTW on larger transforms
by 11% on the largest currently supported size.
2021-04-10 03:54:22 +02:00
|
|
|
/*
|
|
|
|
* This file is part of FFmpeg.
|
|
|
|
*
|
|
|
|
* FFmpeg is free software; you can redistribute it and/or
|
|
|
|
* modify it under the terms of the GNU Lesser General Public
|
|
|
|
* License as published by the Free Software Foundation; either
|
|
|
|
* version 2.1 of the License, or (at your option) any later version.
|
|
|
|
*
|
|
|
|
* FFmpeg is distributed in the hope that it will be useful,
|
|
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
|
* Lesser General Public License for more details.
|
|
|
|
*
|
|
|
|
* You should have received a copy of the GNU Lesser General Public
|
|
|
|
* License along with FFmpeg; if not, write to the Free Software
|
|
|
|
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
|
|
|
*/
|
|
|
|
|
|
|
|
#define TX_FLOAT
|
|
|
|
#include "libavutil/tx_priv.h"
|
|
|
|
#include "libavutil/attributes.h"
|
|
|
|
#include "libavutil/x86/cpu.h"
|
|
|
|
|
2022-01-20 08:14:46 +02:00
|
|
|
#include "config.h"
|
lavu/x86: add FFT assembly
This commit adds a pure x86 assembly SIMD version of the FFT in libavutil/tx.
The design of this pure assembly FFT is pretty unconventional.
On the lowest level, instead of splitting the complex numbers into
real and imaginary parts, we keep complex numbers together but split
them in terms of parity. This saves a number of shuffles in each transform,
but more importantly, it splits each transform into two independent
paths, which we process using separate registers in parallel.
This allows us to keep all units saturated and lets us use all available
registers to avoid dependencies.
Moreover, it allows us to double the granularity of our per-load permutation,
skipping many expensive lookups and allowing us to use just 4 loads per register,
rather than 8, or in case FMA3 (and by extension, AVX2), use the vgatherdpd
instruction, which is at least as fast as 4 separate loads on old hardware,
and quite a bit faster on modern CPUs).
Higher up, we go for a bottom-up construction of large transforms, foregoing
the traditional per-transform call-return recursion chains. Instead, we always
start at the bottom-most basis transform (in this case, a 32-point transform),
and continue constructing larger and larger transforms until we return to the
top-most transform.
This way, we only touch the stack 3 times per a complete target transform:
once for the 1/2 length transform and two times for the 1/4 length transform.
The combination algorithm we use is a standard Split-Radix algorithm,
as used in our C code. Although a version with less operations exists
(Steven G. Johnson and Matteo Frigo's "A modified split-radix FFT with fewer
arithmetic operations", IEEE Trans. Signal Process. 55 (1), 111–119 (2007),
which is the one FFTW uses), it only has 2% less operations and requires at least 4x
the binary code (due to it needing 4 different paths to do a single transform).
That version also has other issues which prevent it from being implemented
with SIMD code as efficiently, which makes it lose the marginal gains it offered,
and cannot be performed bottom-up, requiring many recursive call-return chains,
whose overhead adds up.
We go through a lot of effort to minimize load/stores by keeping as much in
registers in between construcring transforms. This saves us around 32 cycles,
on paper, but in reality a lot more due to load/store aliasing (a load from a
memory location cannot be issued while there's a store pending, and there are
only so many (2 for Zen 3) load/store units in a CPU).
Also, we interleave coefficients during the last stage to save on a store+load
per register.
Each of the smallest, basis transforms (4, 8 and 16-point in our case)
has been extremely optimized. Our 8-point transform is barely 20 instructions
in total, beating our old implementation 8-point transform by 1 instruction.
Our 2x8-point transform is 23 instructions, beating our old implementation by
6 instruction and needing 50% less cycles. Our 16-point transform's combination
code takes slightly more instructions than our old implementation, but makes up
for it by requiring a lot less arithmetic operations.
Overall, the transform was optimized for the timings of Zen 3, which at the
time of writing has the most IPC from all documented CPUs. Shuffles were
preferred over arithmetic operations due to their 1/0.5 latency/throughput.
On average, this code is 30% faster than our old libavcodec implementation.
It's able to trade blows with the previously-untouchable FFTW on small transforms,
and due to its tiny size and better prediction, outdoes FFTW on larger transforms
by 11% on the largest currently supported size.
2021-04-10 03:54:22 +02:00
|
|
|
|
2022-02-07 04:30:27 +02:00
|
|
|
TX_DECL_FN(fft2, sse3)
|
|
|
|
TX_DECL_FN(fft4_fwd, sse2)
|
|
|
|
TX_DECL_FN(fft4_inv, sse2)
|
|
|
|
TX_DECL_FN(fft8, sse3)
|
|
|
|
TX_DECL_FN(fft8_ns, sse3)
|
|
|
|
TX_DECL_FN(fft8, avx)
|
|
|
|
TX_DECL_FN(fft8_ns, avx)
|
2022-09-28 06:46:57 +02:00
|
|
|
TX_DECL_FN(fft15, avx2)
|
|
|
|
TX_DECL_FN(fft15_ns, avx2)
|
2022-02-07 04:30:27 +02:00
|
|
|
TX_DECL_FN(fft16, avx)
|
|
|
|
TX_DECL_FN(fft16_ns, avx)
|
|
|
|
TX_DECL_FN(fft16, fma3)
|
|
|
|
TX_DECL_FN(fft16_ns, fma3)
|
|
|
|
TX_DECL_FN(fft32, avx)
|
|
|
|
TX_DECL_FN(fft32_ns, avx)
|
|
|
|
TX_DECL_FN(fft32, fma3)
|
|
|
|
TX_DECL_FN(fft32_ns, fma3)
|
2022-09-24 03:51:48 +02:00
|
|
|
TX_DECL_FN(fft_sr, avx)
|
|
|
|
TX_DECL_FN(fft_sr_ns, avx)
|
2022-05-21 01:39:12 +02:00
|
|
|
TX_DECL_FN(fft_sr, fma3)
|
|
|
|
TX_DECL_FN(fft_sr_ns, fma3)
|
2022-05-21 00:05:58 +02:00
|
|
|
TX_DECL_FN(fft_sr, avx2)
|
|
|
|
TX_DECL_FN(fft_sr_ns, avx2)
|
2022-01-25 23:39:09 +02:00
|
|
|
|
2022-09-19 05:53:01 +02:00
|
|
|
TX_DECL_FN(fft_pfa_15xM, avx2)
|
|
|
|
TX_DECL_FN(fft_pfa_15xM_ns, avx2)
|
|
|
|
|
2022-09-23 10:38:29 +02:00
|
|
|
TX_DECL_FN(mdct_inv, avx2)
|
x86/tx_float: implement inverse MDCT AVX2 assembly
This commit implements an iMDCT in pure assembly.
This is capable of processing any mod-8 transforms, rather than just
power of two, but since power of two is all we have assembly for
currently, that's what's supported.
It would really benefit if we could somehow use the C code to decide
which function to jump into, but exposing function labels from assebly
into C is anything but easy.
The post-transform loop could probably be improved.
This was somewhat annoying to write, as we must support arbitrary
strides during runtime. There's a fast branch for stride == 4 bytes
and a slower one which uses vgatherdps.
Zen 3 benchmarks for stride == 4 for old (av_imdct_half) vs new (av_tx):
128pt:
2811 decicycles in av_tx (imdct),16775916 runs, 1300 skips
3082 decicycles in av_imdct_half,16776751 runs, 465 skips
256pt:
4920 decicycles in av_tx (imdct),16775820 runs, 1396 skips
5378 decicycles in av_imdct_half,16776411 runs, 805 skips
512pt:
9668 decicycles in av_tx (imdct),16775774 runs, 1442 skips
10626 decicycles in av_imdct_half,16775647 runs, 1569 skips
1024pt:
19812 decicycles in av_tx (imdct),16777144 runs, 72 skips
23036 decicycles in av_imdct_half,16777167 runs, 49 skips
2022-09-03 03:36:40 +02:00
|
|
|
|
2022-09-19 04:14:52 +02:00
|
|
|
TX_DECL_FN(fft2_asm, sse3)
|
|
|
|
TX_DECL_FN(fft4_fwd_asm, sse2)
|
|
|
|
TX_DECL_FN(fft4_inv_asm, sse2)
|
2022-09-03 03:34:57 +02:00
|
|
|
TX_DECL_FN(fft8_asm, sse3)
|
|
|
|
TX_DECL_FN(fft8_asm, avx)
|
|
|
|
TX_DECL_FN(fft16_asm, avx)
|
|
|
|
TX_DECL_FN(fft16_asm, fma3)
|
|
|
|
TX_DECL_FN(fft32_asm, avx)
|
|
|
|
TX_DECL_FN(fft32_asm, fma3)
|
2022-09-24 03:51:48 +02:00
|
|
|
TX_DECL_FN(fft_sr_asm, avx)
|
2022-09-03 03:34:57 +02:00
|
|
|
TX_DECL_FN(fft_sr_asm, fma3)
|
|
|
|
TX_DECL_FN(fft_sr_asm, avx2)
|
|
|
|
|
2022-09-19 05:53:01 +02:00
|
|
|
TX_DECL_FN(fft_pfa_15xM_asm, avx2)
|
|
|
|
|
2022-01-20 08:14:46 +02:00
|
|
|
#define DECL_INIT_FN(basis, interleave) \
|
2022-02-07 04:30:27 +02:00
|
|
|
static av_cold int b ##basis## _i ##interleave(AVTXContext *s, \
|
|
|
|
const FFTXCodelet *cd, \
|
|
|
|
uint64_t flags, \
|
|
|
|
FFTXCodeletOptions *opts, \
|
|
|
|
int len, int inv, \
|
|
|
|
const void *scale) \
|
2022-01-20 08:14:46 +02:00
|
|
|
{ \
|
|
|
|
ff_tx_init_tabs_float(len); \
|
2022-01-25 23:39:09 +02:00
|
|
|
if (cd->max_len == 2) \
|
2022-11-19 01:47:45 +02:00
|
|
|
return ff_tx_gen_ptwo_revtab(s, opts); \
|
2022-01-25 23:39:09 +02:00
|
|
|
else \
|
2022-11-19 01:47:45 +02:00
|
|
|
return ff_tx_gen_split_radix_parity_revtab(s, len, inv, opts, \
|
2022-01-25 23:39:09 +02:00
|
|
|
basis, interleave); \
|
2022-01-20 08:14:46 +02:00
|
|
|
}
|
lavu/x86: add FFT assembly
This commit adds a pure x86 assembly SIMD version of the FFT in libavutil/tx.
The design of this pure assembly FFT is pretty unconventional.
On the lowest level, instead of splitting the complex numbers into
real and imaginary parts, we keep complex numbers together but split
them in terms of parity. This saves a number of shuffles in each transform,
but more importantly, it splits each transform into two independent
paths, which we process using separate registers in parallel.
This allows us to keep all units saturated and lets us use all available
registers to avoid dependencies.
Moreover, it allows us to double the granularity of our per-load permutation,
skipping many expensive lookups and allowing us to use just 4 loads per register,
rather than 8, or in case FMA3 (and by extension, AVX2), use the vgatherdpd
instruction, which is at least as fast as 4 separate loads on old hardware,
and quite a bit faster on modern CPUs).
Higher up, we go for a bottom-up construction of large transforms, foregoing
the traditional per-transform call-return recursion chains. Instead, we always
start at the bottom-most basis transform (in this case, a 32-point transform),
and continue constructing larger and larger transforms until we return to the
top-most transform.
This way, we only touch the stack 3 times per a complete target transform:
once for the 1/2 length transform and two times for the 1/4 length transform.
The combination algorithm we use is a standard Split-Radix algorithm,
as used in our C code. Although a version with less operations exists
(Steven G. Johnson and Matteo Frigo's "A modified split-radix FFT with fewer
arithmetic operations", IEEE Trans. Signal Process. 55 (1), 111–119 (2007),
which is the one FFTW uses), it only has 2% less operations and requires at least 4x
the binary code (due to it needing 4 different paths to do a single transform).
That version also has other issues which prevent it from being implemented
with SIMD code as efficiently, which makes it lose the marginal gains it offered,
and cannot be performed bottom-up, requiring many recursive call-return chains,
whose overhead adds up.
We go through a lot of effort to minimize load/stores by keeping as much in
registers in between construcring transforms. This saves us around 32 cycles,
on paper, but in reality a lot more due to load/store aliasing (a load from a
memory location cannot be issued while there's a store pending, and there are
only so many (2 for Zen 3) load/store units in a CPU).
Also, we interleave coefficients during the last stage to save on a store+load
per register.
Each of the smallest, basis transforms (4, 8 and 16-point in our case)
has been extremely optimized. Our 8-point transform is barely 20 instructions
in total, beating our old implementation 8-point transform by 1 instruction.
Our 2x8-point transform is 23 instructions, beating our old implementation by
6 instruction and needing 50% less cycles. Our 16-point transform's combination
code takes slightly more instructions than our old implementation, but makes up
for it by requiring a lot less arithmetic operations.
Overall, the transform was optimized for the timings of Zen 3, which at the
time of writing has the most IPC from all documented CPUs. Shuffles were
preferred over arithmetic operations due to their 1/0.5 latency/throughput.
On average, this code is 30% faster than our old libavcodec implementation.
It's able to trade blows with the previously-untouchable FFTW on small transforms,
and due to its tiny size and better prediction, outdoes FFTW on larger transforms
by 11% on the largest currently supported size.
2021-04-10 03:54:22 +02:00
|
|
|
|
2022-01-20 08:14:46 +02:00
|
|
|
DECL_INIT_FN(8, 0)
|
|
|
|
DECL_INIT_FN(8, 2)
|
lavu/x86: add FFT assembly
This commit adds a pure x86 assembly SIMD version of the FFT in libavutil/tx.
The design of this pure assembly FFT is pretty unconventional.
On the lowest level, instead of splitting the complex numbers into
real and imaginary parts, we keep complex numbers together but split
them in terms of parity. This saves a number of shuffles in each transform,
but more importantly, it splits each transform into two independent
paths, which we process using separate registers in parallel.
This allows us to keep all units saturated and lets us use all available
registers to avoid dependencies.
Moreover, it allows us to double the granularity of our per-load permutation,
skipping many expensive lookups and allowing us to use just 4 loads per register,
rather than 8, or in case FMA3 (and by extension, AVX2), use the vgatherdpd
instruction, which is at least as fast as 4 separate loads on old hardware,
and quite a bit faster on modern CPUs).
Higher up, we go for a bottom-up construction of large transforms, foregoing
the traditional per-transform call-return recursion chains. Instead, we always
start at the bottom-most basis transform (in this case, a 32-point transform),
and continue constructing larger and larger transforms until we return to the
top-most transform.
This way, we only touch the stack 3 times per a complete target transform:
once for the 1/2 length transform and two times for the 1/4 length transform.
The combination algorithm we use is a standard Split-Radix algorithm,
as used in our C code. Although a version with less operations exists
(Steven G. Johnson and Matteo Frigo's "A modified split-radix FFT with fewer
arithmetic operations", IEEE Trans. Signal Process. 55 (1), 111–119 (2007),
which is the one FFTW uses), it only has 2% less operations and requires at least 4x
the binary code (due to it needing 4 different paths to do a single transform).
That version also has other issues which prevent it from being implemented
with SIMD code as efficiently, which makes it lose the marginal gains it offered,
and cannot be performed bottom-up, requiring many recursive call-return chains,
whose overhead adds up.
We go through a lot of effort to minimize load/stores by keeping as much in
registers in between construcring transforms. This saves us around 32 cycles,
on paper, but in reality a lot more due to load/store aliasing (a load from a
memory location cannot be issued while there's a store pending, and there are
only so many (2 for Zen 3) load/store units in a CPU).
Also, we interleave coefficients during the last stage to save on a store+load
per register.
Each of the smallest, basis transforms (4, 8 and 16-point in our case)
has been extremely optimized. Our 8-point transform is barely 20 instructions
in total, beating our old implementation 8-point transform by 1 instruction.
Our 2x8-point transform is 23 instructions, beating our old implementation by
6 instruction and needing 50% less cycles. Our 16-point transform's combination
code takes slightly more instructions than our old implementation, but makes up
for it by requiring a lot less arithmetic operations.
Overall, the transform was optimized for the timings of Zen 3, which at the
time of writing has the most IPC from all documented CPUs. Shuffles were
preferred over arithmetic operations due to their 1/0.5 latency/throughput.
On average, this code is 30% faster than our old libavcodec implementation.
It's able to trade blows with the previously-untouchable FFTW on small transforms,
and due to its tiny size and better prediction, outdoes FFTW on larger transforms
by 11% on the largest currently supported size.
2021-04-10 03:54:22 +02:00
|
|
|
|
2022-09-28 06:46:57 +02:00
|
|
|
static av_cold int factor_init(AVTXContext *s, const FFTXCodelet *cd,
|
|
|
|
uint64_t flags, FFTXCodeletOptions *opts,
|
|
|
|
int len, int inv, const void *scale)
|
|
|
|
{
|
2022-11-19 01:47:45 +02:00
|
|
|
int ret;
|
|
|
|
|
|
|
|
/* The transformations below are performed in the gather domain,
|
|
|
|
* so override the option and let the infrastructure convert the map
|
|
|
|
* to SCATTER if needed. */
|
|
|
|
FFTXCodeletOptions sub_opts = { .map_dir = FF_TX_MAP_GATHER };
|
|
|
|
|
2022-09-28 06:46:57 +02:00
|
|
|
TX_TAB(ff_tx_init_tabs)(len);
|
|
|
|
|
2022-11-19 01:47:45 +02:00
|
|
|
if (len == 15)
|
|
|
|
ret = ff_tx_gen_pfa_input_map(s, &sub_opts, 3, 5);
|
2022-09-28 06:46:57 +02:00
|
|
|
else
|
2022-11-19 01:47:45 +02:00
|
|
|
ret = ff_tx_gen_default_map(s, &sub_opts);
|
|
|
|
|
|
|
|
if (ret < 0)
|
|
|
|
return ret;
|
2022-09-28 06:46:57 +02:00
|
|
|
|
|
|
|
if (len == 15) {
|
|
|
|
int cnt = 0, tmp[15];
|
|
|
|
|
2022-11-19 01:47:45 +02:00
|
|
|
/* Special permutation to simplify loads in the pre-permuted version */
|
2022-09-28 06:46:57 +02:00
|
|
|
memcpy(tmp, s->map, 15*sizeof(*tmp));
|
|
|
|
for (int i = 1; i < 15; i += 3) {
|
|
|
|
s->map[cnt] = tmp[i];
|
|
|
|
cnt++;
|
|
|
|
}
|
|
|
|
for (int i = 2; i < 15; i += 3) {
|
|
|
|
s->map[cnt] = tmp[i];
|
|
|
|
cnt++;
|
|
|
|
}
|
|
|
|
for (int i = 0; i < 15; i += 3) {
|
|
|
|
s->map[cnt] = tmp[i];
|
|
|
|
cnt++;
|
|
|
|
}
|
|
|
|
memmove(&s->map[7], &s->map[6], 4*sizeof(int));
|
|
|
|
memmove(&s->map[3], &s->map[1], 4*sizeof(int));
|
|
|
|
s->map[1] = tmp[2];
|
|
|
|
s->map[2] = tmp[0];
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
x86/tx_float: implement inverse MDCT AVX2 assembly
This commit implements an iMDCT in pure assembly.
This is capable of processing any mod-8 transforms, rather than just
power of two, but since power of two is all we have assembly for
currently, that's what's supported.
It would really benefit if we could somehow use the C code to decide
which function to jump into, but exposing function labels from assebly
into C is anything but easy.
The post-transform loop could probably be improved.
This was somewhat annoying to write, as we must support arbitrary
strides during runtime. There's a fast branch for stride == 4 bytes
and a slower one which uses vgatherdps.
Zen 3 benchmarks for stride == 4 for old (av_imdct_half) vs new (av_tx):
128pt:
2811 decicycles in av_tx (imdct),16775916 runs, 1300 skips
3082 decicycles in av_imdct_half,16776751 runs, 465 skips
256pt:
4920 decicycles in av_tx (imdct),16775820 runs, 1396 skips
5378 decicycles in av_imdct_half,16776411 runs, 805 skips
512pt:
9668 decicycles in av_tx (imdct),16775774 runs, 1442 skips
10626 decicycles in av_imdct_half,16775647 runs, 1569 skips
1024pt:
19812 decicycles in av_tx (imdct),16777144 runs, 72 skips
23036 decicycles in av_imdct_half,16777167 runs, 49 skips
2022-09-03 03:36:40 +02:00
|
|
|
static av_cold int m_inv_init(AVTXContext *s, const FFTXCodelet *cd,
|
|
|
|
uint64_t flags, FFTXCodeletOptions *opts,
|
|
|
|
int len, int inv, const void *scale)
|
|
|
|
{
|
|
|
|
int ret;
|
2022-11-19 01:47:45 +02:00
|
|
|
FFTXCodeletOptions sub_opts = { .map_dir = FF_TX_MAP_GATHER };
|
x86/tx_float: implement inverse MDCT AVX2 assembly
This commit implements an iMDCT in pure assembly.
This is capable of processing any mod-8 transforms, rather than just
power of two, but since power of two is all we have assembly for
currently, that's what's supported.
It would really benefit if we could somehow use the C code to decide
which function to jump into, but exposing function labels from assebly
into C is anything but easy.
The post-transform loop could probably be improved.
This was somewhat annoying to write, as we must support arbitrary
strides during runtime. There's a fast branch for stride == 4 bytes
and a slower one which uses vgatherdps.
Zen 3 benchmarks for stride == 4 for old (av_imdct_half) vs new (av_tx):
128pt:
2811 decicycles in av_tx (imdct),16775916 runs, 1300 skips
3082 decicycles in av_imdct_half,16776751 runs, 465 skips
256pt:
4920 decicycles in av_tx (imdct),16775820 runs, 1396 skips
5378 decicycles in av_imdct_half,16776411 runs, 805 skips
512pt:
9668 decicycles in av_tx (imdct),16775774 runs, 1442 skips
10626 decicycles in av_imdct_half,16775647 runs, 1569 skips
1024pt:
19812 decicycles in av_tx (imdct),16777144 runs, 72 skips
23036 decicycles in av_imdct_half,16777167 runs, 49 skips
2022-09-03 03:36:40 +02:00
|
|
|
|
|
|
|
s->scale_d = *((SCALE_TYPE *)scale);
|
|
|
|
s->scale_f = s->scale_d;
|
|
|
|
|
|
|
|
flags &= ~FF_TX_OUT_OF_PLACE; /* We want the subtransform to be */
|
|
|
|
flags |= AV_TX_INPLACE; /* in-place */
|
|
|
|
flags |= FF_TX_PRESHUFFLE; /* This function handles the permute step */
|
|
|
|
flags |= FF_TX_ASM_CALL; /* We want an assembly function, not C */
|
|
|
|
|
|
|
|
if ((ret = ff_tx_init_subtx(s, TX_TYPE(FFT), flags, &sub_opts, len >> 1,
|
|
|
|
inv, scale)))
|
|
|
|
return ret;
|
|
|
|
|
2022-09-23 10:38:29 +02:00
|
|
|
s->map = av_malloc(len*sizeof(*s->map));
|
|
|
|
if (!s->map)
|
|
|
|
return AVERROR(ENOMEM);
|
|
|
|
|
|
|
|
memcpy(s->map, s->sub->map, (len >> 1)*sizeof(*s->map));
|
|
|
|
/* Invert lookup table for unstrided path */
|
|
|
|
for (int i = 0; i < (len >> 1); i++)
|
|
|
|
s->map[(len >> 1) + s->map[i]] = i;
|
|
|
|
|
|
|
|
if ((ret = ff_tx_mdct_gen_exp_float(s, s->map)))
|
x86/tx_float: implement inverse MDCT AVX2 assembly
This commit implements an iMDCT in pure assembly.
This is capable of processing any mod-8 transforms, rather than just
power of two, but since power of two is all we have assembly for
currently, that's what's supported.
It would really benefit if we could somehow use the C code to decide
which function to jump into, but exposing function labels from assebly
into C is anything but easy.
The post-transform loop could probably be improved.
This was somewhat annoying to write, as we must support arbitrary
strides during runtime. There's a fast branch for stride == 4 bytes
and a slower one which uses vgatherdps.
Zen 3 benchmarks for stride == 4 for old (av_imdct_half) vs new (av_tx):
128pt:
2811 decicycles in av_tx (imdct),16775916 runs, 1300 skips
3082 decicycles in av_imdct_half,16776751 runs, 465 skips
256pt:
4920 decicycles in av_tx (imdct),16775820 runs, 1396 skips
5378 decicycles in av_imdct_half,16776411 runs, 805 skips
512pt:
9668 decicycles in av_tx (imdct),16775774 runs, 1442 skips
10626 decicycles in av_imdct_half,16775647 runs, 1569 skips
1024pt:
19812 decicycles in av_tx (imdct),16777144 runs, 72 skips
23036 decicycles in av_imdct_half,16777167 runs, 49 skips
2022-09-03 03:36:40 +02:00
|
|
|
return ret;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2022-09-19 05:53:01 +02:00
|
|
|
static av_cold int fft_pfa_init(AVTXContext *s,
|
|
|
|
const FFTXCodelet *cd,
|
|
|
|
uint64_t flags,
|
|
|
|
FFTXCodeletOptions *opts,
|
|
|
|
int len, int inv,
|
|
|
|
const void *scale)
|
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
int sub_len = len / cd->factors[0];
|
2022-11-19 01:47:45 +02:00
|
|
|
FFTXCodeletOptions sub_opts = { .map_dir = FF_TX_MAP_SCATTER };
|
2022-09-19 05:53:01 +02:00
|
|
|
|
|
|
|
flags &= ~FF_TX_OUT_OF_PLACE; /* We want the subtransform to be */
|
|
|
|
flags |= AV_TX_INPLACE; /* in-place */
|
|
|
|
flags |= FF_TX_PRESHUFFLE; /* This function handles the permute step */
|
|
|
|
flags |= FF_TX_ASM_CALL; /* We want an assembly function, not C */
|
|
|
|
|
|
|
|
if ((ret = ff_tx_init_subtx(s, TX_TYPE(FFT), flags, &sub_opts,
|
|
|
|
sub_len, inv, scale)))
|
|
|
|
return ret;
|
|
|
|
|
2022-11-19 01:47:45 +02:00
|
|
|
if ((ret = ff_tx_gen_compound_mapping(s, opts, s->inv, cd->factors[0], sub_len)))
|
2022-09-19 05:53:01 +02:00
|
|
|
return ret;
|
|
|
|
|
|
|
|
if (cd->factors[0] == 15) {
|
2022-11-19 01:47:45 +02:00
|
|
|
int tmp[15];
|
|
|
|
|
|
|
|
/* Our 15-point transform is also a compound one, so embed its input map */
|
|
|
|
TX_EMBED_INPUT_PFA_MAP(s->map, len, 3, 5);
|
|
|
|
|
|
|
|
/* Special permutation to simplify loads in the pre-permuted version */
|
2022-09-19 05:53:01 +02:00
|
|
|
for (int k = 0; k < s->sub[0].len; k++) {
|
|
|
|
int cnt = 0;
|
|
|
|
memcpy(tmp, &s->map[k*15], 15*sizeof(*tmp));
|
|
|
|
for (int i = 1; i < 15; i += 3) {
|
|
|
|
s->map[k*15 + cnt] = tmp[i];
|
|
|
|
cnt++;
|
|
|
|
}
|
|
|
|
for (int i = 2; i < 15; i += 3) {
|
|
|
|
s->map[k*15 + cnt] = tmp[i];
|
|
|
|
cnt++;
|
|
|
|
}
|
|
|
|
for (int i = 0; i < 15; i += 3) {
|
|
|
|
s->map[k*15 + cnt] = tmp[i];
|
|
|
|
cnt++;
|
|
|
|
}
|
|
|
|
memmove(&s->map[k*15 + 7], &s->map[k*15 + 6], 4*sizeof(int));
|
|
|
|
memmove(&s->map[k*15 + 3], &s->map[k*15 + 1], 4*sizeof(int));
|
|
|
|
s->map[k*15 + 1] = tmp[2];
|
|
|
|
s->map[k*15 + 2] = tmp[0];
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!(s->tmp = av_malloc(len*sizeof(*s->tmp))))
|
|
|
|
return AVERROR(ENOMEM);
|
|
|
|
|
|
|
|
TX_TAB(ff_tx_init_tabs)(len / sub_len);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2022-01-20 08:14:46 +02:00
|
|
|
const FFTXCodelet * const ff_tx_codelet_list_float_x86[] = {
|
2022-02-07 04:30:27 +02:00
|
|
|
TX_DEF(fft2, FFT, 2, 2, 2, 0, 128, NULL, sse3, SSE3, AV_TX_INPLACE, 0),
|
2022-09-19 04:14:52 +02:00
|
|
|
TX_DEF(fft2_asm, FFT, 2, 2, 2, 0, 192, b8_i0, sse3, SSE3,
|
|
|
|
AV_TX_INPLACE | FF_TX_PRESHUFFLE | FF_TX_ASM_CALL, 0),
|
2022-02-07 04:30:27 +02:00
|
|
|
TX_DEF(fft2, FFT, 2, 2, 2, 0, 192, b8_i0, sse3, SSE3, AV_TX_INPLACE | FF_TX_PRESHUFFLE, 0),
|
|
|
|
TX_DEF(fft4_fwd, FFT, 4, 4, 2, 0, 128, NULL, sse2, SSE2, AV_TX_INPLACE | FF_TX_FORWARD_ONLY, 0),
|
2022-09-19 04:14:52 +02:00
|
|
|
TX_DEF(fft4_fwd_asm, FFT, 4, 4, 2, 0, 192, b8_i0, sse2, SSE2,
|
|
|
|
AV_TX_INPLACE | FF_TX_PRESHUFFLE | FF_TX_ASM_CALL, 0),
|
|
|
|
TX_DEF(fft4_inv_asm, FFT, 4, 4, 2, 0, 128, NULL, sse2, SSE2,
|
|
|
|
AV_TX_INPLACE | FF_TX_INVERSE_ONLY | FF_TX_ASM_CALL, 0),
|
2022-02-07 04:30:27 +02:00
|
|
|
TX_DEF(fft4_fwd, FFT, 4, 4, 2, 0, 192, b8_i0, sse2, SSE2, AV_TX_INPLACE | FF_TX_PRESHUFFLE, 0),
|
|
|
|
TX_DEF(fft4_inv, FFT, 4, 4, 2, 0, 128, NULL, sse2, SSE2, AV_TX_INPLACE | FF_TX_INVERSE_ONLY, 0),
|
|
|
|
TX_DEF(fft8, FFT, 8, 8, 2, 0, 128, b8_i0, sse3, SSE3, AV_TX_INPLACE, 0),
|
2022-09-03 03:34:57 +02:00
|
|
|
TX_DEF(fft8_asm, FFT, 8, 8, 2, 0, 192, b8_i0, sse3, SSE3,
|
|
|
|
AV_TX_INPLACE | FF_TX_PRESHUFFLE | FF_TX_ASM_CALL, 0),
|
2022-02-07 04:30:27 +02:00
|
|
|
TX_DEF(fft8_ns, FFT, 8, 8, 2, 0, 192, b8_i0, sse3, SSE3, AV_TX_INPLACE | FF_TX_PRESHUFFLE, 0),
|
2022-05-21 00:11:45 +02:00
|
|
|
TX_DEF(fft8, FFT, 8, 8, 2, 0, 256, b8_i0, avx, AVX, AV_TX_INPLACE, AV_CPU_FLAG_AVXSLOW),
|
2022-09-03 03:34:57 +02:00
|
|
|
TX_DEF(fft8_asm, FFT, 8, 8, 2, 0, 320, b8_i0, avx, AVX,
|
|
|
|
AV_TX_INPLACE | FF_TX_PRESHUFFLE | FF_TX_ASM_CALL, AV_CPU_FLAG_AVXSLOW),
|
2022-05-21 00:11:45 +02:00
|
|
|
TX_DEF(fft8_ns, FFT, 8, 8, 2, 0, 320, b8_i0, avx, AVX, AV_TX_INPLACE | FF_TX_PRESHUFFLE,
|
|
|
|
AV_CPU_FLAG_AVXSLOW),
|
|
|
|
TX_DEF(fft16, FFT, 16, 16, 2, 0, 256, b8_i2, avx, AVX, AV_TX_INPLACE, AV_CPU_FLAG_AVXSLOW),
|
2022-09-03 03:34:57 +02:00
|
|
|
TX_DEF(fft16_asm, FFT, 16, 16, 2, 0, 320, b8_i2, avx, AVX,
|
|
|
|
AV_TX_INPLACE | FF_TX_PRESHUFFLE | FF_TX_ASM_CALL, AV_CPU_FLAG_AVXSLOW),
|
2022-05-21 00:11:45 +02:00
|
|
|
TX_DEF(fft16_ns, FFT, 16, 16, 2, 0, 320, b8_i2, avx, AVX, AV_TX_INPLACE | FF_TX_PRESHUFFLE,
|
|
|
|
AV_CPU_FLAG_AVXSLOW),
|
|
|
|
TX_DEF(fft16, FFT, 16, 16, 2, 0, 288, b8_i2, fma3, FMA3, AV_TX_INPLACE, AV_CPU_FLAG_AVXSLOW),
|
2022-09-03 03:34:57 +02:00
|
|
|
TX_DEF(fft16_asm, FFT, 16, 16, 2, 0, 352, b8_i2, fma3, FMA3,
|
|
|
|
AV_TX_INPLACE | FF_TX_PRESHUFFLE | FF_TX_ASM_CALL, AV_CPU_FLAG_AVXSLOW),
|
2022-05-21 00:11:45 +02:00
|
|
|
TX_DEF(fft16_ns, FFT, 16, 16, 2, 0, 352, b8_i2, fma3, FMA3, AV_TX_INPLACE | FF_TX_PRESHUFFLE,
|
|
|
|
AV_CPU_FLAG_AVXSLOW),
|
2021-04-24 18:45:02 +02:00
|
|
|
|
|
|
|
#if ARCH_X86_64
|
2022-05-21 00:11:45 +02:00
|
|
|
TX_DEF(fft32, FFT, 32, 32, 2, 0, 256, b8_i2, avx, AVX, AV_TX_INPLACE, AV_CPU_FLAG_AVXSLOW),
|
2022-09-03 03:34:57 +02:00
|
|
|
TX_DEF(fft32_asm, FFT, 32, 32, 2, 0, 320, b8_i2, avx, AVX,
|
|
|
|
AV_TX_INPLACE | FF_TX_PRESHUFFLE | FF_TX_ASM_CALL, AV_CPU_FLAG_AVXSLOW),
|
2022-05-21 00:11:45 +02:00
|
|
|
TX_DEF(fft32_ns, FFT, 32, 32, 2, 0, 320, b8_i2, avx, AVX, AV_TX_INPLACE | FF_TX_PRESHUFFLE,
|
|
|
|
AV_CPU_FLAG_AVXSLOW),
|
|
|
|
TX_DEF(fft32, FFT, 32, 32, 2, 0, 288, b8_i2, fma3, FMA3, AV_TX_INPLACE, AV_CPU_FLAG_AVXSLOW),
|
2022-09-03 03:34:57 +02:00
|
|
|
TX_DEF(fft32_asm, FFT, 32, 32, 2, 0, 352, b8_i2, fma3, FMA3,
|
|
|
|
AV_TX_INPLACE | FF_TX_PRESHUFFLE | FF_TX_ASM_CALL, AV_CPU_FLAG_AVXSLOW),
|
2022-05-21 00:11:45 +02:00
|
|
|
TX_DEF(fft32_ns, FFT, 32, 32, 2, 0, 352, b8_i2, fma3, FMA3, AV_TX_INPLACE | FF_TX_PRESHUFFLE,
|
|
|
|
AV_CPU_FLAG_AVXSLOW),
|
2024-02-09 16:21:37 +02:00
|
|
|
TX_DEF(fft_sr, FFT, 64, 2097152, 2, 0, 256, b8_i2, avx, AVX, 0, AV_CPU_FLAG_AVXSLOW),
|
|
|
|
TX_DEF(fft_sr_asm, FFT, 64, 2097152, 2, 0, 320, b8_i2, avx, AVX,
|
2022-09-24 03:51:48 +02:00
|
|
|
AV_TX_INPLACE | FF_TX_PRESHUFFLE | FF_TX_ASM_CALL, AV_CPU_FLAG_AVXSLOW),
|
2024-02-09 16:21:37 +02:00
|
|
|
TX_DEF(fft_sr_ns, FFT, 64, 2097152, 2, 0, 320, b8_i2, avx, AVX, AV_TX_INPLACE | FF_TX_PRESHUFFLE,
|
2022-09-24 03:51:48 +02:00
|
|
|
AV_CPU_FLAG_AVXSLOW),
|
2024-02-09 16:21:37 +02:00
|
|
|
TX_DEF(fft_sr, FFT, 64, 2097152, 2, 0, 288, b8_i2, fma3, FMA3, 0, AV_CPU_FLAG_AVXSLOW),
|
|
|
|
TX_DEF(fft_sr_asm, FFT, 64, 2097152, 2, 0, 352, b8_i2, fma3, FMA3,
|
2022-09-03 03:34:57 +02:00
|
|
|
AV_TX_INPLACE | FF_TX_PRESHUFFLE | FF_TX_ASM_CALL, AV_CPU_FLAG_AVXSLOW),
|
2024-02-09 16:21:37 +02:00
|
|
|
TX_DEF(fft_sr_ns, FFT, 64, 2097152, 2, 0, 352, b8_i2, fma3, FMA3, AV_TX_INPLACE | FF_TX_PRESHUFFLE,
|
2022-05-21 00:05:58 +02:00
|
|
|
AV_CPU_FLAG_AVXSLOW),
|
2022-09-19 05:53:01 +02:00
|
|
|
|
2022-05-21 01:39:12 +02:00
|
|
|
#if HAVE_AVX2_EXTERNAL
|
2022-09-28 06:46:57 +02:00
|
|
|
TX_DEF(fft15, FFT, 15, 15, 15, 0, 320, factor_init, avx2, AVX2,
|
|
|
|
AV_TX_INPLACE, AV_CPU_FLAG_AVXSLOW),
|
|
|
|
TX_DEF(fft15_ns, FFT, 15, 15, 15, 0, 384, factor_init, avx2, AVX2,
|
|
|
|
AV_TX_INPLACE | FF_TX_PRESHUFFLE, AV_CPU_FLAG_AVXSLOW),
|
|
|
|
|
2024-02-09 16:21:37 +02:00
|
|
|
TX_DEF(fft_sr, FFT, 64, 2097152, 2, 0, 320, b8_i2, avx2, AVX2, 0,
|
2022-05-21 00:11:45 +02:00
|
|
|
AV_CPU_FLAG_AVXSLOW | AV_CPU_FLAG_SLOW_GATHER),
|
2024-02-09 16:21:37 +02:00
|
|
|
TX_DEF(fft_sr_asm, FFT, 64, 2097152, 2, 0, 384, b8_i2, avx2, AVX2,
|
2022-09-03 03:34:57 +02:00
|
|
|
AV_TX_INPLACE | FF_TX_PRESHUFFLE | FF_TX_ASM_CALL, AV_CPU_FLAG_AVXSLOW | AV_CPU_FLAG_SLOW_GATHER),
|
2024-02-09 16:21:37 +02:00
|
|
|
TX_DEF(fft_sr_ns, FFT, 64, 2097152, 2, 0, 384, b8_i2, avx2, AVX2, AV_TX_INPLACE | FF_TX_PRESHUFFLE,
|
2022-05-21 00:11:45 +02:00
|
|
|
AV_CPU_FLAG_AVXSLOW | AV_CPU_FLAG_SLOW_GATHER),
|
x86/tx_float: implement inverse MDCT AVX2 assembly
This commit implements an iMDCT in pure assembly.
This is capable of processing any mod-8 transforms, rather than just
power of two, but since power of two is all we have assembly for
currently, that's what's supported.
It would really benefit if we could somehow use the C code to decide
which function to jump into, but exposing function labels from assebly
into C is anything but easy.
The post-transform loop could probably be improved.
This was somewhat annoying to write, as we must support arbitrary
strides during runtime. There's a fast branch for stride == 4 bytes
and a slower one which uses vgatherdps.
Zen 3 benchmarks for stride == 4 for old (av_imdct_half) vs new (av_tx):
128pt:
2811 decicycles in av_tx (imdct),16775916 runs, 1300 skips
3082 decicycles in av_imdct_half,16776751 runs, 465 skips
256pt:
4920 decicycles in av_tx (imdct),16775820 runs, 1396 skips
5378 decicycles in av_imdct_half,16776411 runs, 805 skips
512pt:
9668 decicycles in av_tx (imdct),16775774 runs, 1442 skips
10626 decicycles in av_imdct_half,16775647 runs, 1569 skips
1024pt:
19812 decicycles in av_tx (imdct),16777144 runs, 72 skips
23036 decicycles in av_imdct_half,16777167 runs, 49 skips
2022-09-03 03:36:40 +02:00
|
|
|
|
2022-10-01 12:44:06 +02:00
|
|
|
TX_DEF(fft_pfa_15xM, FFT, 60, TX_LEN_UNLIMITED, 15, 2, 320, fft_pfa_init, avx2, AVX2,
|
2022-09-19 05:53:01 +02:00
|
|
|
AV_TX_INPLACE, AV_CPU_FLAG_AVXSLOW | AV_CPU_FLAG_SLOW_GATHER),
|
2022-10-01 12:44:06 +02:00
|
|
|
TX_DEF(fft_pfa_15xM_asm, FFT, 60, TX_LEN_UNLIMITED, 15, 2, 384, fft_pfa_init, avx2, AVX2,
|
2022-09-19 05:53:01 +02:00
|
|
|
AV_TX_INPLACE | FF_TX_PRESHUFFLE | FF_TX_ASM_CALL, AV_CPU_FLAG_AVXSLOW | AV_CPU_FLAG_SLOW_GATHER),
|
2022-10-01 12:44:06 +02:00
|
|
|
TX_DEF(fft_pfa_15xM_ns, FFT, 60, TX_LEN_UNLIMITED, 15, 2, 384, fft_pfa_init, avx2, AVX2,
|
2022-09-19 05:53:01 +02:00
|
|
|
AV_TX_INPLACE | FF_TX_PRESHUFFLE, AV_CPU_FLAG_AVXSLOW | AV_CPU_FLAG_SLOW_GATHER),
|
|
|
|
|
2022-09-23 10:38:29 +02:00
|
|
|
TX_DEF(mdct_inv, MDCT, 16, TX_LEN_UNLIMITED, 2, TX_FACTOR_ANY, 384, m_inv_init, avx2, AVX2,
|
x86/tx_float: implement inverse MDCT AVX2 assembly
This commit implements an iMDCT in pure assembly.
This is capable of processing any mod-8 transforms, rather than just
power of two, but since power of two is all we have assembly for
currently, that's what's supported.
It would really benefit if we could somehow use the C code to decide
which function to jump into, but exposing function labels from assebly
into C is anything but easy.
The post-transform loop could probably be improved.
This was somewhat annoying to write, as we must support arbitrary
strides during runtime. There's a fast branch for stride == 4 bytes
and a slower one which uses vgatherdps.
Zen 3 benchmarks for stride == 4 for old (av_imdct_half) vs new (av_tx):
128pt:
2811 decicycles in av_tx (imdct),16775916 runs, 1300 skips
3082 decicycles in av_imdct_half,16776751 runs, 465 skips
256pt:
4920 decicycles in av_tx (imdct),16775820 runs, 1396 skips
5378 decicycles in av_imdct_half,16776411 runs, 805 skips
512pt:
9668 decicycles in av_tx (imdct),16775774 runs, 1442 skips
10626 decicycles in av_imdct_half,16775647 runs, 1569 skips
1024pt:
19812 decicycles in av_tx (imdct),16777144 runs, 72 skips
23036 decicycles in av_imdct_half,16777167 runs, 49 skips
2022-09-03 03:36:40 +02:00
|
|
|
FF_TX_INVERSE_ONLY, AV_CPU_FLAG_AVXSLOW | AV_CPU_FLAG_SLOW_GATHER),
|
2022-05-21 00:05:58 +02:00
|
|
|
#endif
|
2022-01-20 08:14:46 +02:00
|
|
|
#endif
|
lavu/x86: add FFT assembly
This commit adds a pure x86 assembly SIMD version of the FFT in libavutil/tx.
The design of this pure assembly FFT is pretty unconventional.
On the lowest level, instead of splitting the complex numbers into
real and imaginary parts, we keep complex numbers together but split
them in terms of parity. This saves a number of shuffles in each transform,
but more importantly, it splits each transform into two independent
paths, which we process using separate registers in parallel.
This allows us to keep all units saturated and lets us use all available
registers to avoid dependencies.
Moreover, it allows us to double the granularity of our per-load permutation,
skipping many expensive lookups and allowing us to use just 4 loads per register,
rather than 8, or in case FMA3 (and by extension, AVX2), use the vgatherdpd
instruction, which is at least as fast as 4 separate loads on old hardware,
and quite a bit faster on modern CPUs).
Higher up, we go for a bottom-up construction of large transforms, foregoing
the traditional per-transform call-return recursion chains. Instead, we always
start at the bottom-most basis transform (in this case, a 32-point transform),
and continue constructing larger and larger transforms until we return to the
top-most transform.
This way, we only touch the stack 3 times per a complete target transform:
once for the 1/2 length transform and two times for the 1/4 length transform.
The combination algorithm we use is a standard Split-Radix algorithm,
as used in our C code. Although a version with less operations exists
(Steven G. Johnson and Matteo Frigo's "A modified split-radix FFT with fewer
arithmetic operations", IEEE Trans. Signal Process. 55 (1), 111–119 (2007),
which is the one FFTW uses), it only has 2% less operations and requires at least 4x
the binary code (due to it needing 4 different paths to do a single transform).
That version also has other issues which prevent it from being implemented
with SIMD code as efficiently, which makes it lose the marginal gains it offered,
and cannot be performed bottom-up, requiring many recursive call-return chains,
whose overhead adds up.
We go through a lot of effort to minimize load/stores by keeping as much in
registers in between construcring transforms. This saves us around 32 cycles,
on paper, but in reality a lot more due to load/store aliasing (a load from a
memory location cannot be issued while there's a store pending, and there are
only so many (2 for Zen 3) load/store units in a CPU).
Also, we interleave coefficients during the last stage to save on a store+load
per register.
Each of the smallest, basis transforms (4, 8 and 16-point in our case)
has been extremely optimized. Our 8-point transform is barely 20 instructions
in total, beating our old implementation 8-point transform by 1 instruction.
Our 2x8-point transform is 23 instructions, beating our old implementation by
6 instruction and needing 50% less cycles. Our 16-point transform's combination
code takes slightly more instructions than our old implementation, but makes up
for it by requiring a lot less arithmetic operations.
Overall, the transform was optimized for the timings of Zen 3, which at the
time of writing has the most IPC from all documented CPUs. Shuffles were
preferred over arithmetic operations due to their 1/0.5 latency/throughput.
On average, this code is 30% faster than our old libavcodec implementation.
It's able to trade blows with the previously-untouchable FFTW on small transforms,
and due to its tiny size and better prediction, outdoes FFTW on larger transforms
by 11% on the largest currently supported size.
2021-04-10 03:54:22 +02:00
|
|
|
|
2022-01-20 08:14:46 +02:00
|
|
|
NULL,
|
|
|
|
};
|