/* * Copyright 2011 The LibYuv Project Authors. All rights reserved. * * Use of this source code is governed by a BSD-style license * that can be found in the LICENSE file in the root of the source * tree. An additional intellectual property rights grant can be found * in the file PATENTS. All contributing project authors may * be found in the AUTHORS file in the root of the source tree.
*/
#if !defined(DISABLE_SLOW_TESTS) || defined(__x86_64__) || defined(__i386__) // SLOW TESTS are those that are unoptimized C code. // FULL TESTS are optimized but test many variations of the same code. #define ENABLE_FULL_TESTS #endif
// Test scaling with C vs Opt and return maximum pixel difference. 0 = exact. staticint ARGBTestFilter(int src_width, int src_height, int dst_width, int dst_height,
FilterMode f, int benchmark_iterations, int disable_cpu_flags, int benchmark_cpu_info) { if (!SizeValid(src_width, src_height, dst_width, dst_height)) { return 0;
}
int i, j; constint b = 0; // 128 to test for padding/stride.
int64_t src_argb_plane_size =
(Abs(src_width) + b * 2) * (Abs(src_height) + b * 2) * 4LL; int src_stride_argb = (b * 2 + Abs(src_width)) * 4;
// Warm up both versions for consistent benchmarks.
MaskCpuFlags(disable_cpu_flags); // Disable all CPU optimization.
ARGBScale(src_argb + (src_stride_argb * b) + b * 4, src_stride_argb,
src_width, src_height, dst_argb_c + (dst_stride_argb * b) + b * 4,
dst_stride_argb, dst_width, dst_height, f);
MaskCpuFlags(benchmark_cpu_info); // Enable all CPU optimization.
ARGBScale(src_argb + (src_stride_argb * b) + b * 4, src_stride_argb,
src_width, src_height, dst_argb_opt + (dst_stride_argb * b) + b * 4,
dst_stride_argb, dst_width, dst_height, f);
MaskCpuFlags(disable_cpu_flags); // Disable all CPU optimization. double c_time = get_time();
ARGBScale(src_argb + (src_stride_argb * b) + b * 4, src_stride_argb,
src_width, src_height, dst_argb_c + (dst_stride_argb * b) + b * 4,
dst_stride_argb, dst_width, dst_height, f);
c_time = (get_time() - c_time);
MaskCpuFlags(benchmark_cpu_info); // Enable all CPU optimization. double opt_time = get_time(); for (i = 0; i < benchmark_iterations; ++i) {
ARGBScale(src_argb + (src_stride_argb * b) + b * 4, src_stride_argb,
src_width, src_height,
dst_argb_opt + (dst_stride_argb * b) + b * 4, dst_stride_argb,
dst_width, dst_height, f);
}
opt_time = (get_time() - opt_time) / benchmark_iterations;
// Report performance of C vs OPT
printf("filter %d - %8d us C - %8d us OPT\n", f, static_cast<int>(c_time * 1e6), static_cast<int>(opt_time * 1e6));
// C version may be a little off from the optimized. Order of // operations may introduce rounding somewhere. So do a difference // of the buffers and look to see that the max difference isn't // over 2. int max_diff = 0; for (i = b; i < (dst_height + b); ++i) { for (j = b * 4; j < (dst_width + b) * 4; ++j) { int abs_diff = Abs(dst_argb_c[(i * dst_stride_argb) + j] -
dst_argb_opt[(i * dst_stride_argb) + j]); if (abs_diff > max_diff) {
max_diff = abs_diff;
}
}
}
staticint TileARGBScale(const uint8_t* src_argb, int src_stride_argb, int src_width, int src_height,
uint8_t* dst_argb, int dst_stride_argb, int dst_width, int dst_height,
FilterMode filtering) { for (int y = 0; y < dst_height; y += kTileY) { for (int x = 0; x < dst_width; x += kTileX) { int clip_width = kTileX; if (x + clip_width > dst_width) {
clip_width = dst_width - x;
} int clip_height = kTileY; if (y + clip_height > dst_height) {
clip_height = dst_height - y;
} int r = ARGBScaleClip(src_argb, src_stride_argb, src_width, src_height,
dst_argb, dst_stride_argb, dst_width, dst_height, x,
y, clip_width, clip_height, filtering); if (r) { return r;
}
}
} return 0;
}
staticint ARGBClipTestFilter(int src_width, int src_height, int dst_width, int dst_height,
FilterMode f, int benchmark_iterations) { if (!SizeValid(src_width, src_height, dst_width, dst_height)) { return 0;
}
constint b = 128;
int64_t src_argb_plane_size =
(Abs(src_width) + b * 2) * (Abs(src_height) + b * 2) * 4; int src_stride_argb = (b * 2 + Abs(src_width)) * 4;
// Do full image, no clipping. double c_time = get_time();
ARGBScale(src_argb + (src_stride_argb * b) + b * 4, src_stride_argb,
src_width, src_height, dst_argb_c + (dst_stride_argb * b) + b * 4,
dst_stride_argb, dst_width, dst_height, f);
c_time = (get_time() - c_time);
// Do tiled image, clipping scale to a tile at a time. double opt_time = get_time(); for (i = 0; i < benchmark_iterations; ++i) {
TileARGBScale(src_argb + (src_stride_argb * b) + b * 4, src_stride_argb,
src_width, src_height,
dst_argb_opt + (dst_stride_argb * b) + b * 4, dst_stride_argb,
dst_width, dst_height, f);
}
opt_time = (get_time() - opt_time) / benchmark_iterations;
// Report performance of Full vs Tiled.
printf("filter %d - %8d us Full - %8d us Tiled\n", f, static_cast<int>(c_time * 1e6), static_cast<int>(opt_time * 1e6));
// Compare full scaled image vs tiled image. int max_diff = 0; for (i = b; i < (dst_height + b); ++i) { for (j = b * 4; j < (dst_width + b) * 4; ++j) { int abs_diff = Abs(dst_argb_c[(i * dst_stride_argb) + j] -
dst_argb_opt[(i * dst_stride_argb) + j]); if (abs_diff > max_diff) {
max_diff = abs_diff;
}
}
}
#ifdefined(ENABLE_FULL_TESTS) // Test scale with swapped width and height with all 3 filters.
TEST_SCALESWAPXY1(ARGBScale, None, 0)
TEST_SCALESWAPXY1(ARGBScale, Linear, 0)
TEST_SCALESWAPXY1(ARGBScale, Bilinear, 0) #else
TEST_SCALESWAPXY1(ARGBScale, Bilinear, 0) #endif #undef TEST_SCALESWAPXY1
// Scale with YUV conversion to ARGB and clipping. // TODO(fbarchard): Add fourcc support. All 4 ARGB formats is easy to support. staticint YUVToARGBScaleReference2(const uint8_t* src_y, int src_stride_y, const uint8_t* src_u, int src_stride_u, const uint8_t* src_v, int src_stride_v,
uint32_t /* src_fourcc */, int src_width, int src_height,
uint8_t* dst_argb, int dst_stride_argb,
uint32_t /* dst_fourcc */, int dst_width, int dst_height, int clip_x, int clip_y, int clip_width, int clip_height, enum FilterMode filtering) {
uint8_t* argb_buffer = static_cast<uint8_t*>(malloc(src_width * src_height * 4)); int r;
I420ToARGB(src_y, src_stride_y, src_u, src_stride_u, src_v, src_stride_v,
argb_buffer, src_width * 4, src_width, src_height);
staticvoid FillRamp(uint8_t* buf, int width, int height, int v, int dx, int dy) { int rv = v; for (int y = 0; y < height; ++y) { for (int x = 0; x < width; ++x) {
*buf++ = v;
v += dx; if (v < 0 || v > 255) {
dx = -dx;
v += dx;
}
}
v = rv + dy; if (v < 0 || v > 255) {
dy = -dy;
v += dy;
}
rv = v;
}
}
// Test scaling with C vs Opt and return maximum pixel difference. 0 = exact. staticint YUVToARGBTestFilter(int src_width, int src_height, int dst_width, int dst_height,
FilterMode f, int benchmark_iterations) {
int64_t src_y_plane_size = Abs(src_width) * Abs(src_height);
int64_t src_uv_plane_size =
((Abs(src_width) + 1) / 2) * ((Abs(src_height) + 1) / 2); int src_stride_y = Abs(src_width); int src_stride_uv = (Abs(src_width) + 1) / 2;
Die Informationen auf dieser Webseite wurden
nach bestem Wissen sorgfältig zusammengestellt. Es wird jedoch weder Vollständigkeit, noch Richtigkeit,
noch Qualität der bereit gestellten Informationen zugesichert.
Bemerkung:
Die farbliche Syntaxdarstellung und die Messung sind noch experimentell.