static void check_pred8x8l(H264PredContext *h, uint8_t *buf0, uint8_t *buf1,
int codec, int chroma_format, int bit_depth)
{
if (chroma_format == 1 && codec_ids[codec] == AV_CODEC_ID_H264) {
int pred_mode;
for (pred_mode = 0; pred_mode < 12; pred_mode++) {
if (check_pred_func(h->pred8x8l[pred_mode], "8x8l", pred4x4_modes[codec][pred_mode])) {
int neighbors;
for (neighbors = 0; neighbors <= 0xc000; neighbors += 0x4000) {
int has_topleft = neighbors & 0x8000;
int has_topright = neighbors & 0x4000;
if ((pred_mode == DIAG_DOWN_RIGHT_PRED || pred_mode == VERT_RIGHT_PRED) && !has_topleft)
continue; /* Those aren't allowed according to the spec */
randomize_buffers();
call_ref(src0, has_topleft, has_topright, (ptrdiff_t)24*SIZEOF_PIXEL);
call_new(src1, has_topleft, has_topright, (ptrdiff_t)24*SIZEOF_PIXEL);
if (memcmp(buf0, buf1, BUF_SIZE))
fail();
bench_new(src1, has_topleft, has_topright, (ptrdiff_t)24*SIZEOF_PIXEL);
}
}
}
}
}
static void check_add_res(HEVCDSPContext h, int bit_depth)
{
int i;
LOCAL_ALIGNED(32, int16_t, res0, [32 * 32]);
LOCAL_ALIGNED(32, int16_t, res1, [32 * 32]);
LOCAL_ALIGNED(32, uint8_t, dst0, [32 * 32 * 2]);
LOCAL_ALIGNED(32, uint8_t, dst1, [32 * 32 * 2]);
for (i = 2; i <= 5; i++) {
int block_size = 1 << i;
int size = block_size * block_size;
ptrdiff_t stride = block_size << (bit_depth > 8);
declare_func_emms(AV_CPU_FLAG_MMX, void, uint8_t *dst, int16_t *res, ptrdiff_t stride);
randomize_buffers(res0, size);
randomize_buffers2(dst0, size);
memcpy(res1, res0, sizeof(*res0) * size);
memcpy(dst1, dst0, size);
if (check_func(h.add_residual[i - 2], "add_res_%dx%d_%d", block_size, block_size, bit_depth)) {
call_ref(dst0, res0, stride);
call_new(dst1, res1, stride);
if (memcmp(dst0, dst1, size))
fail();
bench_new(dst1, res1, stride);
}
}
}
static void check_add_bytes(LLVidDSPContext c, int width)
{
uint8_t *src0 = av_mallocz(width);
uint8_t *src1 = av_mallocz(width);
uint8_t *dst0 = av_mallocz(width);
uint8_t *dst1 = av_mallocz(width);
declare_func_emms(AV_CPU_FLAG_MMX, void, uint8_t *dst, uint8_t *src, ptrdiff_t w);
if (!src0 || !src1 || !dst0 || !dst1)
fail();
randomize_buffers(src0, width);
memcpy(src1, src0, width);
if (check_func(c.add_bytes, "add_bytes")) {
call_ref(dst0, src0, width);
call_new(dst1, src1, width);
if (memcmp(dst0, dst1, width))
fail();
bench_new(dst1, src1, width);
}
av_free(src0);
av_free(src1);
av_free(dst0);
av_free(dst1);
}
static void check_pred8x8(H264PredContext *h, uint8_t *buf0, uint8_t *buf1,
int codec, int chroma_format, int bit_depth)
{
int pred_mode;
for (pred_mode = 0; pred_mode < 11; pred_mode++) {
if (check_pred_func(h->pred8x8[pred_mode], (chroma_format == 2) ? "8x16" : "8x8",
pred8x8_modes[codec][pred_mode])) {
randomize_buffers();
call_ref(src0, (ptrdiff_t)24*SIZEOF_PIXEL);
call_new(src1, (ptrdiff_t)24*SIZEOF_PIXEL);
if (memcmp(buf0, buf1, BUF_SIZE))
fail();
bench_new(src1, (ptrdiff_t)24*SIZEOF_PIXEL);
}
}
}
static void check_pred16x16(H264PredContext *h, uint8_t *buf0, uint8_t *buf1,
int codec, int chroma_format, int bit_depth)
{
if (chroma_format == 1) {
int pred_mode;
for (pred_mode = 0; pred_mode < 9; pred_mode++) {
if (check_pred_func(h->pred16x16[pred_mode], "16x16", pred16x16_modes[codec][pred_mode])) {
randomize_buffers();
call_ref(src0, (ptrdiff_t)48);
call_new(src1, (ptrdiff_t)48);
if (memcmp(buf0, buf1, BUF_SIZE))
fail();
bench_new(src1, (ptrdiff_t)48);
}
}
}
}
static void check_pred4x4(H264PredContext *h, uint8_t *buf0, uint8_t *buf1,
int codec, int chroma_format, int bit_depth)
{
if (chroma_format == 1) {
uint8_t *topright = buf0 + 2*16;
int pred_mode;
for (pred_mode = 0; pred_mode < 15; pred_mode++) {
if (check_pred_func(h->pred4x4[pred_mode], "4x4", pred4x4_modes[codec][pred_mode])) {
randomize_buffers();
call_ref(src0, topright, (ptrdiff_t)12*SIZEOF_PIXEL);
call_new(src1, topright, (ptrdiff_t)12*SIZEOF_PIXEL);
if (memcmp(buf0, buf1, BUF_SIZE))
fail();
bench_new(src1, topright, (ptrdiff_t)12*SIZEOF_PIXEL);
}
}
}
}
static void check_qpel(HEVCDSPContext *h, int16_t *dst0, int16_t *dst1,
uint8_t *src, int16_t *mcbuffer, int bit_depth)
{
int i, j, k, l, mx, my;
declare_func(void, int16_t *dst, ptrdiff_t dststride, uint8_t *src, ptrdiff_t srcstride,
int height, int mx, int my, int16_t *mcbuffer);
randomize_buffers(src, BUF_SIZE, bit_depth);
memset(dst0, 0, BUF_SIZE * sizeof(*dst0));
memset(dst1, 0, BUF_SIZE * sizeof(*dst1));
for (i = 0; i < 2; i++) {
for (j = 0; j < 2; j++) {
for (k = 0; k < FF_ARRAY_ELEMS(h->put_hevc_qpel[i][j]); k++) {
int width = pred_widths[k];
int dststride = FFALIGN(width, 16) * sizeof(*dst0);
int srcstride = FFALIGN(width + 7, 8) * PIXEL_SIZE(bit_depth);
if (!check_func(h->put_hevc_qpel[i][j][k], "qpel_%s_%d_%d", interp_names[i][j], width, bit_depth))
continue;
for (l = 0; l < FF_ARRAY_ELEMS(pred_heights[0]); l++) {
int height = pred_heights[width][l];
if (!height)
continue;
for (my = i; my < (i ? 2 : 1); my++)
for (mx = j; mx < (j ? 2 : 1); mx++) {
call_ref(dst0, dststride, src + 3 * srcstride + 3 * PIXEL_SIZE(bit_depth), srcstride, height, mx, my, mcbuffer);
call_new(dst1, dststride, src + 3 * srcstride + 3 * PIXEL_SIZE(bit_depth), srcstride, height, mx, my, mcbuffer);
if (memcmp(dst0, dst1, dststride * height * sizeof(*dst0)))
fail();
bench_new(dst1, dststride, src + 3 * srcstride + 3 * PIXEL_SIZE(bit_depth), srcstride, height, mx, my, mcbuffer);
}
}
}
}
}
}
void test_ref()
{
for(int i = 0; i < g_repeat; i++) {
CALL_SUBTEST_1( ref_vector(Matrix<float, 1, 1>()) );
CALL_SUBTEST_1( check_const_correctness(Matrix<float, 1, 1>()) );
CALL_SUBTEST_2( ref_vector(Vector4d()) );
CALL_SUBTEST_2( check_const_correctness(Matrix4d()) );
CALL_SUBTEST_3( ref_vector(Vector4cf()) );
CALL_SUBTEST_4( ref_vector(VectorXcf(8)) );
CALL_SUBTEST_5( ref_vector(VectorXi(12)) );
CALL_SUBTEST_5( check_const_correctness(VectorXi(12)) );
CALL_SUBTEST_1( ref_matrix(Matrix<float, 1, 1>()) );
CALL_SUBTEST_2( ref_matrix(Matrix4d()) );
CALL_SUBTEST_1( ref_matrix(Matrix<float,3,5>()) );
CALL_SUBTEST_4( ref_matrix(MatrixXcf(internal::random<int>(1,10),internal::random<int>(1,10))) );
CALL_SUBTEST_4( ref_matrix(Matrix<std::complex<double>,10,15>()) );
CALL_SUBTEST_5( ref_matrix(MatrixXi(internal::random<int>(1,10),internal::random<int>(1,10))) );
CALL_SUBTEST_6( call_ref() );
}
}
static void check_idct_dc(HEVCDSPContext h, int bit_depth)
{
int i;
LOCAL_ALIGNED(32, int16_t, coeffs0, [32 * 32]);
LOCAL_ALIGNED(32, int16_t, coeffs1, [32 * 32]);
for (i = 2; i <= 5; i++) {
int block_size = 1 << i;
int size = block_size * block_size;
declare_func_emms(AV_CPU_FLAG_MMXEXT, void, int16_t *coeffs);
randomize_buffers(coeffs0, size);
memcpy(coeffs1, coeffs0, sizeof(*coeffs0) * size);
if (check_func(h.idct_dc[i - 2], "idct_%dx%d_dc_%d", block_size, block_size, bit_depth)) {
call_ref(coeffs0);
call_new(coeffs1);
if (memcmp(coeffs0, coeffs1, sizeof(*coeffs0) * size))
fail();
bench_new(coeffs1);
}
}
}
static void test_fcmul_add(const float *src0, const float *src1, const float *src2)
{
LOCAL_ALIGNED_32(float, cdst, [LEN*2+8]);
LOCAL_ALIGNED_32(float, odst, [LEN*2+8]);
int i;
declare_func(void, float *sum, const float *t, const float *c,
ptrdiff_t len);
memcpy(cdst, src0, (LEN*2+8) * sizeof(float));
memcpy(odst, src0, (LEN*2+8) * sizeof(float));
call_ref(cdst, src1, src2, LEN);
call_new(odst, src1, src2, LEN);
for (i = 0; i <= LEN*2; i++) {
if (!float_near_abs_eps(cdst[i], odst[i], FLT_EPSILON)) {
fprintf(stderr, "%d: %- .12f - %- .12f = % .12g\n",
i, cdst[i], odst[i], cdst[i] - odst[i]);
fail();
break;
}
}
memcpy(odst, src0, (LEN*2+8) * sizeof(float));
bench_new(odst, src1, src2, LEN);
}
