}
}
- declare_func_emms(AV_CPU_FLAG_MMX, void, uint8_t *pix, int stride,
+
+ static void check_loop_filter(void)
+ {
+ LOCAL_ALIGNED_16(uint8_t, dst, [32 * 16 * 2]);
+ LOCAL_ALIGNED_16(uint8_t, dst0, [32 * 16 * 2]);
+ LOCAL_ALIGNED_16(uint8_t, dst1, [32 * 16 * 2]);
+ H264DSPContext h;
+ int bit_depth;
+ int alphas[36], betas[36];
+ int8_t tc0[36][4];
+
- declare_func_emms(AV_CPU_FLAG_MMX, void, uint8_t *pix, int stride,
++ declare_func_emms(AV_CPU_FLAG_MMX, void, uint8_t *pix, ptrdiff_t stride,
+ int alpha, int beta, int8_t *tc0);
+
+ for (bit_depth = 8; bit_depth <= 10; bit_depth++) {
+ int i, j, a, c;
+ uint32_t mask = pixel_mask_lf[bit_depth - 8];
+ ff_h264dsp_init(&h, bit_depth, 1);
+ for (i = 35, a = 255, c = 250; i >= 0; i--) {
+ alphas[i] = a << (bit_depth - 8);
+ betas[i] = (i + 1) / 2 << (bit_depth - 8);
+ tc0[i][0] = tc0[i][3] = (c + 6) / 10;
+ tc0[i][1] = (c + 7) / 15;
+ tc0[i][2] = (c + 9) / 20;
+ a = a*9/10;
+ c = c*9/10;
+ }
+
+ #define CHECK_LOOP_FILTER(name, align, ...) \
+ do { \
+ if (check_func(h.name, #name "_%dbpp", bit_depth)) { \
+ for (j = 0; j < 36; j++) { \
+ intptr_t off = 8 * 32 + (j & 15) * 4 * !align; \
+ for (i = 0; i < 1024; i+=4) { \
+ AV_WN32A(dst + i, rnd() & mask); \
+ } \
+ memcpy(dst0, dst, 32 * 16 * 2); \
+ memcpy(dst1, dst, 32 * 16 * 2); \
+ \
+ call_ref(dst0 + off, 32, alphas[j], betas[j], tc0[j]); \
+ call_new(dst1 + off, 32, alphas[j], betas[j], tc0[j]); \
+ if (memcmp(dst0, dst1, 32 * 16 * SIZEOF_PIXEL)) { \
+ fprintf(stderr, #name ": j:%d, alpha:%d beta:%d " \
+ "tc0:{%d,%d,%d,%d}\n", j, alphas[j], betas[j], \
+ tc0[j][0], tc0[j][1], tc0[j][2], tc0[j][3]); \
+ fail(); \
+ } \
+ bench_new(dst1, 32, alphas[j], betas[j], tc0[j]); \
+ } \
+ } \
+ } while (0)
+
+ CHECK_LOOP_FILTER(h264_v_loop_filter_luma, 1);
+ CHECK_LOOP_FILTER(h264_h_loop_filter_luma, 0);
+ CHECK_LOOP_FILTER(h264_h_loop_filter_luma_mbaff, 0);
+ CHECK_LOOP_FILTER(h264_v_loop_filter_chroma, 1);
+ CHECK_LOOP_FILTER(h264_h_loop_filter_chroma, 0);
+ CHECK_LOOP_FILTER(h264_h_loop_filter_chroma_mbaff, 0);
+ #undef CHECK_LOOP_FILTER
+ }
+ }
+
+ static void check_loop_filter_intra(void)
+ {
+ LOCAL_ALIGNED_16(uint8_t, dst, [32 * 16 * 2]);
+ LOCAL_ALIGNED_16(uint8_t, dst0, [32 * 16 * 2]);
+ LOCAL_ALIGNED_16(uint8_t, dst1, [32 * 16 * 2]);
+ H264DSPContext h;
+ int bit_depth;
+ int alphas[36], betas[36];
+
++ declare_func_emms(AV_CPU_FLAG_MMX, void, uint8_t *pix, ptrdiff_t stride,
+ int alpha, int beta);
+
+ for (bit_depth = 8; bit_depth <= 10; bit_depth++) {
+ int i, j, a;
+ uint32_t mask = pixel_mask_lf[bit_depth - 8];
+ ff_h264dsp_init(&h, bit_depth, 1);
+ for (i = 35, a = 255; i >= 0; i--) {
+ alphas[i] = a << (bit_depth - 8);
+ betas[i] = (i + 1) / 2 << (bit_depth - 8);
+ a = a*9/10;
+ }
+
+ #define CHECK_LOOP_FILTER(name, align) \
+ do { \
+ if (check_func(h.name, #name "_%dbpp", bit_depth)) { \
+ for (j = 0; j < 36; j++) { \
+ intptr_t off = 8 * 32 + (j & 15) * 4 * !align; \
+ for (i = 0; i < 1024; i+=4) { \
+ AV_WN32A(dst + i, rnd() & mask); \
+ } \
+ memcpy(dst0, dst, 32 * 16 * 2); \
+ memcpy(dst1, dst, 32 * 16 * 2); \
+ \
+ call_ref(dst0 + off, 32, alphas[j], betas[j]); \
+ call_new(dst1 + off, 32, alphas[j], betas[j]); \
+ if (memcmp(dst0, dst1, 32 * 16 * SIZEOF_PIXEL)) { \
+ fprintf(stderr, #name ": j:%d, alpha:%d beta:%d\n", \
+ j, alphas[j], betas[j]); \
+ fail(); \
+ } \
+ bench_new(dst1, 32, alphas[j], betas[j]); \
+ } \
+ } \
+ } while (0)
+
+ CHECK_LOOP_FILTER(h264_v_loop_filter_luma_intra, 1);
+ CHECK_LOOP_FILTER(h264_h_loop_filter_luma_intra, 0);
+ CHECK_LOOP_FILTER(h264_h_loop_filter_luma_mbaff_intra, 0);
+ CHECK_LOOP_FILTER(h264_v_loop_filter_chroma_intra, 1);
+ CHECK_LOOP_FILTER(h264_h_loop_filter_chroma_intra, 0);
+ CHECK_LOOP_FILTER(h264_h_loop_filter_chroma_mbaff_intra, 0);
+ #undef CHECK_LOOP_FILTER
+ }
+ }
+
void checkasm_check_h264dsp(void)
{
check_idct();