diff options
| -rwxr-xr-x | yuvbench/build-linux-x86_64-gcc.sh | 8 | ||||
| -rw-r--r-- | yuvbench/kbench.h | 14 | ||||
| -rw-r--r-- | yuvbench/yuvbench.c | 24 | ||||
| -rw-r--r-- | yuvbench/yuvbench.h | 1 | ||||
| -rw-r--r-- | yuvbench/yuvbench_libyuv.c | 21 |
5 files changed, 58 insertions, 10 deletions
diff --git a/yuvbench/build-linux-x86_64-gcc.sh b/yuvbench/build-linux-x86_64-gcc.sh index 555a195..19d8784 100755 --- a/yuvbench/build-linux-x86_64-gcc.sh +++ b/yuvbench/build-linux-x86_64-gcc.sh @@ -1,7 +1,9 @@ #!/bin/sh -CFLAGS="-Wall -Wextra -Wpedantic -O3 -g -DYUVBENCH_BAD" -LFLAGS="" +CFLAGS="-Wall -Wextra -Wpedantic -O3 -g -DYUVBENCH_BAD -DYUVBENCH_LIBYUV -DYUVBENCH_SWSCALE" +LFLAGS="$(pkg-config --libs libswscale) -lyuv" mkdir -p build gcc -o build/yuvbench.o $CFLAGS -c ./yuvbench.c gcc -o build/yuvbench_bad.o $CFLAGS -c ./yuvbench_bad.c -gcc -o build/yuvbench $LFLAGS build/yuvbench.o build/yuvbench_bad.o +gcc -o build/yuvbench_libyuv.o $CFLAGS -c ./yuvbench_libyuv.c +gcc -o build/yuvbench_swscale.o $CFLAGS $(pkg-config --cflags libswscale) -c ./yuvbench_swscale.c +gcc -o build/yuvbench $LFLAGS build/yuvbench.o build/yuvbench_bad.o build/yuvbench_libyuv.o build/yuvbench_swscale.o diff --git a/yuvbench/kbench.h b/yuvbench/kbench.h index c5d4149..42ec646 100644 --- a/yuvbench/kbench.h +++ b/yuvbench/kbench.h @@ -9,6 +9,10 @@ double KBenchElapsedTime(uintptr_t t0, uintptr_t t1); #ifdef KBENCH_IMPLEMENTATION +#ifdef __linux__ +#include <time.h> +#endif + uintptr_t KBenchTS(void) { uintptr_t result = 0; @@ -16,6 +20,12 @@ uintptr_t KBenchTS(void) #if defined(__APPLE__) && defined(__clang__) && defined(__aarch64__) __asm__ volatile("mrs %0, CNTVCT_EL0" : "=r"(result) :: "memory"); #endif + // linux: use clock_gettime +#if defined(__linux__) + struct timespec ts; + clock_gettime(CLOCK_MONOTONIC, &ts); + result = (uintptr_t)ts.tv_sec * 1e9 + (uintptr_t)ts.tv_nsec; +#endif return result; } @@ -35,6 +45,10 @@ double KBenchElapsedTime(uintptr_t t0, uintptr_t t1) __asm__ volatile("mrs %0, CNTFRQ_EL0" : "=r"(cntfreq_el0) :: "memory"); val = (double)elapsed / (double)cntfreq_el0; #endif + // linux: use clock_gettime +#if defined(__linux__) && defined(__GNUC__) && defined(__amd64__) + val = (double)elapsed / 1e9; +#endif return val; } diff --git a/yuvbench/yuvbench.c b/yuvbench/yuvbench.c index 095104e..3a92371 100644 --- a/yuvbench/yuvbench.c +++ b/yuvbench/yuvbench.c @@ -9,6 +9,9 @@ Backend yuvbench_accelerate(void); #ifdef YUVBENCH_BAD Backend yuvbench_bad(void); #endif +#ifdef YUVBENCH_LIBYUV +Backend yuvbench_libyuv(void); +#endif #ifdef YUVBENCH_SWSCALE Backend yuvbench_swscale(void); #endif @@ -44,14 +47,14 @@ static void run_backend(Backend b) return; } - printf("running warnups...\n"); - int warmup = 20; + printf(" running warnups...\n"); + int warmup = 100; for (int i = 0; i < warmup; ++i) { b.convert_fn(&ctx); } - printf("testing...\n"); - int tests = 1000; + printf(" testing...\n"); + int tests = 2500; double* tests_table = calloc(tests, sizeof(double)); assert(tests_table); for (int i = 0; i < tests; ++i) { uintptr_t t0 = KBenchTS(); @@ -79,9 +82,9 @@ static void run_backend(Backend b) } ts_avg += (tests_table[i] / (double)tests); } - printf("min result: %fms\n", ts_min * 1000.0f); - printf("max result: %fms\n", ts_max * 1000.0f); - printf("avg result: %fms\n", ts_avg * 1000.0f); + printf(" min result: %fms\n", ts_min * 1000.0f); + printf(" max result: %fms\n", ts_max * 1000.0f); + printf(" avg result: %fms\n", ts_avg * 1000.0f); if (G.show) { @@ -186,12 +189,19 @@ int main(int argc, char** argv) G.out_buf = calloc(1, G.out_len); #ifdef YUVBENCH_ACCELERATE + printf("YUVBENCH_ACCELERATE\n"); run_backend(yuvbench_accelerate()); #endif #ifdef YUVBENCH_BAD + printf("YUVBENCH_BAD\n"); run_backend(yuvbench_bad()); #endif +#ifdef YUVBENCH_LIBYUV + printf("YUVBENCH_LIBYUV\n"); + run_backend(yuvbench_libyuv()); +#endif #ifdef YUVBENCH_SWSCALE + printf("YUVBENCH_SWSCALE\n"); run_backend(yuvbench_swscale()); #endif } diff --git a/yuvbench/yuvbench.h b/yuvbench/yuvbench.h index ba6711a..0abdf15 100644 --- a/yuvbench/yuvbench.h +++ b/yuvbench/yuvbench.h @@ -4,6 +4,7 @@ #include <assert.h> #include <errno.h> #include <stdbool.h> +#include <stdint.h> #include <stdio.h> #include <stdlib.h> #include <string.h> diff --git a/yuvbench/yuvbench_libyuv.c b/yuvbench/yuvbench_libyuv.c new file mode 100644 index 0000000..64da249 --- /dev/null +++ b/yuvbench/yuvbench_libyuv.c @@ -0,0 +1,21 @@ +#include "yuvbench.h" + +#include <libyuv.h> +#include <libyuv/convert_argb.h> + +static bool yuvbench_libyuv_convert(Ctx* ctx) +{ + const uint32_t w = ctx->inp_w; + const uint32_t h = ctx->inp_h; + const uint8_t* Y = (const uint8_t*)ctx->inp_buf; + const uint8_t* Cb = Y + (w * h); + const uint8_t* Cr = Cb + (w / 2 * h / 2); + return I420ToRAW(Y, w, Cb, w / 2, Cr, w / 2, ctx->out_buf, w * 3, w, h) == 0; +} + +Backend yuvbench_libyuv(void) +{ + Backend b = { 0 }; + b.convert_fn = yuvbench_libyuv_convert; + return b; +} |