mirror of
https://code.videolan.org/videolan/dav1d
synced 2026-06-11 04:03:05 +00:00
Niklas Haas
3a2a874994
There are a number of benefits tied to the upstream / third-party checkasm version, including: - Improved long-term maintainability, code reuse with other projects, etc. - Vastly improved overall performance / runtime for benchmarking, due primarily to the ability to scale the runtime of each test to that test's complexity. - Much more robust statistical analysis of benchmarking results; including robust outlier rejection, an estimation of the histogram, and the ability to report the variance / stddev in addition to the (trimmed) mean. - Interactive HTML and JSON output formats in addition to CSV/TSV. - More readable and user-friendly output across the board, especially for failures and data dumps (e.g. also showing errors inside padding bytes). - Better cross-platform support, including dynamic fallback of timer implementations on ARM platforms, a better RISC-V harness, and more. There are multiple approaches to how we can solve the problem of integrating this third party checkasm into dav1d, but I think the hybrid approach of loading it as an external dependency, falling back to a meson wrap file, provides the best overall compromise. This avoids the messiness of git e.g. git submodules, while still allowing us to pin individual tags.
215 lines
9.0 KiB
C
215 lines
9.0 KiB
C
/*
|
|
* Copyright © 2018, VideoLAN and dav1d authors
|
|
* Copyright © 2018, Two Orioles, LLC
|
|
* All rights reserved.
|
|
*
|
|
* Redistribution and use in source and binary forms, with or without
|
|
* modification, are permitted provided that the following conditions are met:
|
|
*
|
|
* 1. Redistributions of source code must retain the above copyright notice, this
|
|
* list of conditions and the following disclaimer.
|
|
*
|
|
* 2. Redistributions in binary form must reproduce the above copyright notice,
|
|
* this list of conditions and the following disclaimer in the documentation
|
|
* and/or other materials provided with the distribution.
|
|
*
|
|
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
|
|
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
|
|
* WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
|
|
* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
|
|
* ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
|
|
* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
|
|
* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
|
|
* ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
|
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
|
|
* SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
|
*/
|
|
|
|
#include "tests/checkasm/internal.h"
|
|
|
|
#include <stdio.h>
|
|
#include <string.h>
|
|
|
|
#include "src/levels.h"
|
|
#include "src/looprestoration.h"
|
|
#include "src/tables.h"
|
|
|
|
static int to_binary(int x) { /* 0-15 -> 0000-1111 */
|
|
return (x & 1) + 5 * (x & 2) + 25 * (x & 4) + 125 * (x & 8);
|
|
}
|
|
|
|
static void init_tmp(pixel *buf, const ptrdiff_t stride,
|
|
const int w, const int h, const int bitdepth_max)
|
|
{
|
|
const int noise_mask = bitdepth_max >> 4;
|
|
const int x_off = rnd() & 7, y_off = rnd() & 7;
|
|
|
|
for (int y = 0; y < h; y++) {
|
|
for (int x = 0; x < w; x++) {
|
|
buf[x] = (((x + x_off) ^ (y + y_off)) & 8 ? bitdepth_max : 0) ^
|
|
(rnd() & noise_mask);
|
|
}
|
|
buf += PXSTRIDE(stride);
|
|
}
|
|
}
|
|
|
|
static void check_wiener(Dav1dLoopRestorationDSPContext *const c, const int bpc) {
|
|
PIXEL_RECT(c_dst, 384, 64);
|
|
PIXEL_RECT(a_dst, 384, 64);
|
|
PIXEL_RECT(h_edge, 384, 8);
|
|
pixel left[64][4];
|
|
LooprestorationParams params;
|
|
int16_t (*const filter)[8] = params.filter;
|
|
|
|
declare_func(void, pixel *dst, ptrdiff_t dst_stride,
|
|
const pixel (*const left)[4],
|
|
const pixel *lpf, int w, int h,
|
|
const LooprestorationParams *params,
|
|
enum LrEdgeFlags edges HIGHBD_DECL_SUFFIX);
|
|
|
|
for (int t = 0; t < 2; t++) {
|
|
if (check_func(c->wiener[t], "wiener_%dtap_%dbpc", t ? 5 : 7, bpc)) {
|
|
filter[0][0] = filter[0][6] = t ? 0 : (rnd() & 15) - 5;
|
|
filter[0][1] = filter[0][5] = (rnd() & 31) - 23;
|
|
filter[0][2] = filter[0][4] = (rnd() & 63) - 17;
|
|
filter[0][3] = -(filter[0][0] + filter[0][1] + filter[0][2]) * 2;
|
|
#if BITDEPTH != 8
|
|
filter[0][3] += 128;
|
|
#endif
|
|
|
|
filter[1][0] = filter[1][6] = t ? 0 : (rnd() & 15) - 5;
|
|
filter[1][1] = filter[1][5] = (rnd() & 31) - 23;
|
|
filter[1][2] = filter[1][4] = (rnd() & 63) - 17;
|
|
filter[1][3] = 128 - (filter[1][0] + filter[1][1] + filter[1][2]) * 2;
|
|
|
|
const int base_w = 1 + (rnd() % 384);
|
|
const int base_h = 1 + (rnd() & 63);
|
|
const int bitdepth_max = (1 << bpc) - 1;
|
|
|
|
CLEAR_PIXEL_RECT(c_dst);
|
|
/* We potentially read 3 pixels to the left of the input
|
|
* pointer, and up to the max width, horizontally.
|
|
* (In the case of LR_HAVE_RIGHT we read 3 pixels past the
|
|
* input dimensions, but if LR_HAVE_RIGHT we have w == 256.)
|
|
* Therefore, initialize (384+4) x 64 pixels. */
|
|
init_tmp(c_dst - 4, c_dst_stride, 384 + 4, 64, bitdepth_max);
|
|
init_tmp(h_edge - 4, h_edge_stride, 384 + 4, 8, bitdepth_max);
|
|
init_tmp((pixel *) left, 4 * sizeof(pixel), 4, 64, bitdepth_max);
|
|
|
|
for (enum LrEdgeFlags edges = 0; edges <= 0xf; edges++) {
|
|
const int w = edges & LR_HAVE_RIGHT ? 256 : base_w;
|
|
const int h = edges & LR_HAVE_BOTTOM ? 64 : base_h;
|
|
|
|
assert(c_dst_stride == a_dst_stride);
|
|
assert(c_dst_buf_h == a_dst_buf_h);
|
|
memcpy(a_dst_buf, c_dst_buf, a_dst_stride * a_dst_buf_h);
|
|
|
|
assert(c_dst_stride == h_edge_stride);
|
|
call_ref(c_dst, c_dst_stride, left, h_edge,
|
|
w, h, ¶ms, edges HIGHBD_TAIL_SUFFIX);
|
|
call_new(a_dst, a_dst_stride, left, h_edge,
|
|
w, h, ¶ms, edges HIGHBD_TAIL_SUFFIX);
|
|
if (checkasm_check_pixel_padded_align(c_dst, c_dst_stride,
|
|
a_dst, a_dst_stride,
|
|
w, h, "dst", 64, 1))
|
|
{
|
|
fprintf(stderr, "size = %dx%d, edges = %04d\n",
|
|
w, h, to_binary(edges));
|
|
break;
|
|
}
|
|
}
|
|
bench_new(alternate(c_dst, a_dst), a_dst_stride, left,
|
|
h_edge, 256, 64, ¶ms, 0xf HIGHBD_TAIL_SUFFIX);
|
|
}
|
|
}
|
|
}
|
|
|
|
static void check_sgr(Dav1dLoopRestorationDSPContext *const c, const int bpc) {
|
|
PIXEL_RECT(c_dst, 384, 64);
|
|
PIXEL_RECT(a_dst, 384, 64);
|
|
PIXEL_RECT(h_edge, 384, 8);
|
|
pixel left[64][4];
|
|
LooprestorationParams params;
|
|
|
|
declare_func(void, pixel *dst, ptrdiff_t dst_stride,
|
|
const pixel (*const left)[4],
|
|
const pixel *lpf, int w, int h,
|
|
const LooprestorationParams *params,
|
|
enum LrEdgeFlags edges HIGHBD_DECL_SUFFIX);
|
|
|
|
static const struct { char name[4]; uint8_t idx; } sgr_data[3] = {
|
|
{ "5x5", 14 },
|
|
{ "3x3", 10 },
|
|
{ "mix", 0 },
|
|
};
|
|
|
|
for (int i = 0; i < 3; i++) {
|
|
if (check_func(c->sgr[i], "sgr_%s_%dbpc", sgr_data[i].name, bpc)) {
|
|
const uint16_t *const sgr_params = dav1d_sgr_params[sgr_data[i].idx];
|
|
params.sgr.s0 = sgr_params[0];
|
|
params.sgr.s1 = sgr_params[1];
|
|
params.sgr.w0 = sgr_params[0] ? (rnd() & 127) - 96 : 0;
|
|
params.sgr.w1 = (sgr_params[1] ? 160 - (rnd() & 127) : 33) - params.sgr.w0;
|
|
|
|
const int base_w = 1 + (rnd() % 384);
|
|
const int base_h = 1 + (rnd() & 63);
|
|
const int bitdepth_max = (1 << bpc) - 1;
|
|
|
|
CLEAR_PIXEL_RECT(c_dst);
|
|
/* We potentially read 3 pixels to the left of the input
|
|
* pointer, and up to the max width, horizontally.
|
|
* (In the case of LR_HAVE_RIGHT we read 3 pixels past the
|
|
* input dimensions, but if LR_HAVE_RIGHT we have w == 256.)
|
|
* Therefore, initialize (384+4) x 64 pixels. */
|
|
init_tmp(c_dst - 4, c_dst_stride, 384 + 4, 64, bitdepth_max);
|
|
init_tmp(h_edge - 4, h_edge_stride, 384 + 4, 8, bitdepth_max);
|
|
init_tmp((pixel *) left, 4 * sizeof(pixel), 4, 64, bitdepth_max);
|
|
|
|
for (enum LrEdgeFlags edges = 0; edges <= 0xf; edges++) {
|
|
const int w = edges & LR_HAVE_RIGHT ? 256 : base_w;
|
|
const int h = edges & LR_HAVE_BOTTOM ? 64 : base_h;
|
|
|
|
assert(c_dst_stride == a_dst_stride);
|
|
assert(c_dst_buf_h == a_dst_buf_h);
|
|
memcpy(a_dst_buf, c_dst_buf, a_dst_stride * a_dst_buf_h);
|
|
|
|
assert(c_dst_stride == h_edge_stride);
|
|
call_ref(c_dst, c_dst_stride, left, h_edge,
|
|
w, h, ¶ms, edges HIGHBD_TAIL_SUFFIX);
|
|
call_new(a_dst, a_dst_stride, left, h_edge,
|
|
w, h, ¶ms, edges HIGHBD_TAIL_SUFFIX);
|
|
if (checkasm_check_pixel_padded_align(c_dst, c_dst_stride,
|
|
a_dst, a_dst_stride,
|
|
w, h, "dst", 64, 1))
|
|
{
|
|
fprintf(stderr, "size = %dx%d, edges = %04d\n",
|
|
w, h, to_binary(edges));
|
|
break;
|
|
}
|
|
}
|
|
bench_new(alternate(c_dst, a_dst), a_dst_stride, left,
|
|
h_edge, 256, 64, ¶ms, 0xf HIGHBD_TAIL_SUFFIX);
|
|
}
|
|
}
|
|
}
|
|
|
|
void bitfn(checkasm_check_looprestoration)(void) {
|
|
#if BITDEPTH == 16
|
|
const int bpc_min = 10, bpc_max = 12;
|
|
#else
|
|
const int bpc_min = 8, bpc_max = 8;
|
|
#endif
|
|
for (int bpc = bpc_min; bpc <= bpc_max; bpc += 2) {
|
|
Dav1dLoopRestorationDSPContext c;
|
|
bitfn(dav1d_loop_restoration_dsp_init)(&c, bpc);
|
|
check_wiener(&c, bpc);
|
|
}
|
|
report("wiener");
|
|
for (int bpc = bpc_min; bpc <= bpc_max; bpc += 2) {
|
|
Dav1dLoopRestorationDSPContext c;
|
|
bitfn(dav1d_loop_restoration_dsp_init)(&c, bpc);
|
|
check_sgr(&c, bpc);
|
|
}
|
|
report("sgr");
|
|
}
|