[FFmpeg-devel] [PATCH v4 2/2][GSoC 2024] tests/checkasm: Add check_vvc_sad to vvc_mc.c

[Rémi Denis-Courmont]

Hi,

Le 20 mai 2024 03:42:03 GMT+03:00, Stone Chen <chen.stonechen at gmail.com> a écrit :
>Adds checkasm for DMVR SAD AVX2 implementation.
>
>Benchmarks ( AMD 7940HS )
>vvc_sad_8x8_c: 70.0
>vvc_sad_8x8_avx2: 10.0
>vvc_sad_16x16_c: 280.0
>vvc_sad_16x16_avx2: 20.0
>vvc_sad_32x32_c: 1020.0
>vvc_sad_32x32_avx2: 70.0
>vvc_sad_64x64_c: 3560.0
>vvc_sad_64x64_avx2: 270.0
>vvc_sad_128x128_c: 13760.0
>vvc_sad_128x128_avx2: 1070.0
>---
> tests/checkasm/vvc_mc.c | 38 ++++++++++++++++++++++++++++++++++++++
> 1 file changed, 38 insertions(+)

VVC benchmarks have increased checksam runtime by at least an order of magnitude. It's become so prohibitively slow that I could not even get to the end.

This is not an acceptable situation and impedes non-VVC assembler work 

Please fix this before you add any new VVC tests. In the mean time:

-1 / Nack all VVC checksam from my behalf.


>diff --git a/tests/checkasm/vvc_mc.c b/tests/checkasm/vvc_mc.c
>index 97f57cb401..e251400bfc 100644
>--- a/tests/checkasm/vvc_mc.c
>+++ b/tests/checkasm/vvc_mc.c
>@@ -322,8 +322,46 @@ static void check_avg(void)
>     report("avg");
> }
> 
>+static void check_vvc_sad(void)
>+{
>+    const int bit_depth = 10;
>+    VVCDSPContext c;
>+    LOCAL_ALIGNED_32(uint16_t, src0, [MAX_CTU_SIZE * MAX_CTU_SIZE * 4]);
>+    LOCAL_ALIGNED_32(uint16_t, src1, [MAX_CTU_SIZE * MAX_CTU_SIZE * 4]);
>+    declare_func(int, const int16_t *src0, const int16_t *src1, int dx, int dy, int block_w, int block_h);
>+
>+    ff_vvc_dsp_init(&c, bit_depth);
>+    memset(src0, 0, MAX_CTU_SIZE * MAX_CTU_SIZE * 4);
>+    memset(src1, 0, MAX_CTU_SIZE * MAX_CTU_SIZE * 4);
>+
>+    randomize_pixels(src0, src1, MAX_CTU_SIZE * MAX_CTU_SIZE * 2);
>+     for (int h = 8; h <= MAX_CTU_SIZE; h *= 2) {
>+        for (int w = 8; w <= MAX_CTU_SIZE; w *= 2) {
>+            for(int offy = 0; offy <= 4; offy++) {
>+                for(int offx = 0; offx <= 4; offx++) {
>+                    if(check_func(c.inter.sad, "vvc_sad_%dx%d", w, h)) {
>+                        int result0;
>+                        int result1;
>+
>+                        result0 =  call_ref(src0 + PIXEL_STRIDE * 2 + 2, src1 + PIXEL_STRIDE * 2 + 2, offx, offy, w, h);
>+                        result1 =  call_new(src0 + PIXEL_STRIDE * 2 + 2, src1 + PIXEL_STRIDE * 2 + 2, offx, offy, w, h);
>+
>+                        if (result1 != result0)
>+                            fail();
>+                        if(w == h && offx == 0 && offy == 0)
>+                            bench_new(src0 + PIXEL_STRIDE * 2 + 2, src1 + PIXEL_STRIDE * 2 + 2, offx, offy, w, h);
>+                    }
>+                }
>+            }
>+        }
>+     }
>+
>+    report("check_vvc_sad");
>+}
>+
> void checkasm_check_vvc_mc(void)
> {
>+    check_vvc_sad();
>     check_put_vvc_luma();
>     check_put_vvc_luma_uni();
>     check_put_vvc_chroma();