Add correctness tests for the SSE2 warp filter
Also rename warp_affine() to av1_warp_affine()
Change-Id: I945baff6be8a1ea942ce88dfcfa5344af6b3a966
diff --git a/av1/common/av1_rtcd_defs.pl b/av1/common/av1_rtcd_defs.pl
index 891d817..0805fd7 100644
--- a/av1/common/av1_rtcd_defs.pl
+++ b/av1/common/av1_rtcd_defs.pl
@@ -770,8 +770,8 @@
if ((aom_config("CONFIG_WARPED_MOTION") eq "yes") ||
(aom_config("CONFIG_GLOBAL_MOTION") eq "yes")) {
- add_proto qw/void warp_affine/, "int32_t *mat, uint8_t *ref, int width, int height, int stride, uint8_t *pred, int p_col, int p_row, int p_width, int p_height, int p_stride, int subsampling_x, int subsampling_y, int ref_frm, int32_t alpha, int32_t beta, int32_t gamma, int32_t delta";
- specialize qw/warp_affine sse2/;
+ add_proto qw/void av1_warp_affine/, "int32_t *mat, uint8_t *ref, int width, int height, int stride, uint8_t *pred, int p_col, int p_row, int p_width, int p_height, int p_stride, int subsampling_x, int subsampling_y, int ref_frm, int32_t alpha, int32_t beta, int32_t gamma, int32_t delta";
+ specialize qw/av1_warp_affine sse2/;
}
1;
diff --git a/av1/common/warped_motion.c b/av1/common/warped_motion.c
index d6c5279..6056914 100644
--- a/av1/common/warped_motion.c
+++ b/av1/common/warped_motion.c
@@ -908,11 +908,12 @@
return v;
}
-void warp_affine_c(int32_t *mat, uint8_t *ref, int width, int height,
- int stride, uint8_t *pred, int p_col, int p_row, int p_width,
- int p_height, int p_stride, int subsampling_x,
- int subsampling_y, int ref_frm, int32_t alpha, int32_t beta,
- int32_t gamma, int32_t delta) {
+void av1_warp_affine_c(int32_t *mat, uint8_t *ref, int width, int height,
+ int stride, uint8_t *pred, int p_col, int p_row,
+ int p_width, int p_height, int p_stride,
+ int subsampling_x, int subsampling_y, int ref_frm,
+ int32_t alpha, int32_t beta, int32_t gamma,
+ int32_t delta) {
int16_t tmp[15 * 8];
int i, j, k, l, m;
@@ -1070,9 +1071,9 @@
return;
}
- warp_affine(mat, ref, width, height, stride, pred, p_col, p_row, p_width,
- p_height, p_stride, subsampling_x, subsampling_y, ref_frm,
- alpha, beta, gamma, delta);
+ av1_warp_affine(mat, ref, width, height, stride, pred, p_col, p_row,
+ p_width, p_height, p_stride, subsampling_x, subsampling_y,
+ ref_frm, alpha, beta, gamma, delta);
} else {
warp_plane_old(wm, ref, width, height, stride, pred, p_col, p_row, p_width,
p_height, p_stride, subsampling_x, subsampling_y, x_scale,
diff --git a/av1/common/x86/warp_plane_sse2.c b/av1/common/x86/warp_plane_sse2.c
index 395797a..42c25c9 100644
--- a/av1/common/x86/warp_plane_sse2.c
+++ b/av1/common/x86/warp_plane_sse2.c
@@ -17,12 +17,12 @@
const __m128i *const filter = (const __m128i *const)warped_filter;
/* SSE2 version of the rotzoom/affine warp filter */
-void warp_affine_sse2(int32_t *mat, uint8_t *ref, int width, int height,
- int stride, uint8_t *pred, int p_col, int p_row,
- int p_width, int p_height, int p_stride,
- int subsampling_x, int subsampling_y, int ref_frm,
- int32_t alpha, int32_t beta, int32_t gamma,
- int32_t delta) {
+void av1_warp_affine_sse2(int32_t *mat, uint8_t *ref, int width, int height,
+ int stride, uint8_t *pred, int p_col, int p_row,
+ int p_width, int p_height, int p_stride,
+ int subsampling_x, int subsampling_y, int ref_frm,
+ int32_t alpha, int32_t beta, int32_t gamma,
+ int32_t delta) {
__m128i tmp[15];
int i, j, k;
diff --git a/test/test.mk b/test/test.mk
index 85a90f1..c071480 100644
--- a/test/test.mk
+++ b/test/test.mk
@@ -202,6 +202,9 @@
LIBAOM_TEST_SRCS-$(CONFIG_AV1) += av1_inv_txfm2d_test.cc
LIBAOM_TEST_SRCS-$(CONFIG_AV1) += av1_convolve_test.cc
LIBAOM_TEST_SRCS-$(CONFIG_AV1) += av1_convolve_optimz_test.cc
+ifneq ($(findstring yes,$(CONFIG_GLOBAL_MOTION) $(CONFIG_WARPED_MOTION)),)
+LIBAOM_TEST_SRCS-$(HAVE_SSE2) += warp_filter_test.cc
+endif
TEST_INTRA_PRED_SPEED_SRCS-yes := test_intra_pred_speed.cc
TEST_INTRA_PRED_SPEED_SRCS-yes += ../md5_utils.h ../md5_utils.c
diff --git a/test/warp_filter_test.cc b/test/warp_filter_test.cc
new file mode 100644
index 0000000..914ee74
--- /dev/null
+++ b/test/warp_filter_test.cc
@@ -0,0 +1,130 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "third_party/googletest/src/include/gtest/gtest.h"
+
+#include "./av1_rtcd.h"
+#include "./aom_dsp_rtcd.h"
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "test/util.h"
+
+#include "av1/common/mv.h"
+
+using std::tr1::tuple;
+using std::tr1::make_tuple;
+using libaom_test::ACMRandom;
+
+typedef tuple<int, int, int> WarpTestParam;
+
+namespace {
+
+class AV1WarpFilterTest : public ::testing::TestWithParam<WarpTestParam> {
+ public:
+ virtual ~AV1WarpFilterTest() {}
+ virtual void SetUp() { rnd_.Reset(ACMRandom::DeterministicSeed()); }
+
+ virtual void TearDown() { libaom_test::ClearSystemState(); }
+
+ protected:
+ int32_t random_param(int bits) {
+ // 1 in 8 chance of generating zero (arbitrarily chosen)
+ if (((rnd_.Rand8()) & 7) == 0) return 0;
+ // Otherwise, enerate uniform values in the range
+ // [-(1 << bits), 1] U [1, 1<<bits]
+ int32_t v = 1 + (rnd_.Rand16() & ((1 << bits) - 1));
+ if ((rnd_.Rand8()) & 1) return -v;
+ return v;
+ }
+ void generate_model(int32_t *mat, int32_t *alpha, int32_t *beta,
+ int32_t *gamma, int32_t *delta) {
+ while (1) {
+ mat[0] = random_param(WARPEDMODEL_PREC_BITS + 6);
+ mat[1] = random_param(WARPEDMODEL_PREC_BITS + 6);
+ mat[2] = (random_param(WARPEDMODEL_PREC_BITS - 3)) +
+ (1 << WARPEDMODEL_PREC_BITS);
+ mat[3] = random_param(WARPEDMODEL_PREC_BITS - 3);
+ // 50/50 chance of generating ROTZOOM vs. AFFINE models
+ if (rnd_.Rand8() & 1) {
+ // AFFINE
+ mat[4] = random_param(WARPEDMODEL_PREC_BITS - 3);
+ mat[5] = (random_param(WARPEDMODEL_PREC_BITS - 3)) +
+ (1 << WARPEDMODEL_PREC_BITS);
+ } else {
+ mat[4] = -mat[3];
+ mat[5] = mat[2];
+ }
+
+ // Calculate the derived parameters and check that they are suitable
+ // for the warp filter.
+ assert(mat[2] != 0);
+
+ *alpha = mat[2] - (1 << WARPEDMODEL_PREC_BITS);
+ *beta = mat[3];
+ *gamma = ((int64_t)mat[4] << WARPEDMODEL_PREC_BITS) / mat[2];
+ *delta = mat[5] - (((int64_t)mat[3] * mat[4] + (mat[2] / 2)) / mat[2]) -
+ (1 << WARPEDMODEL_PREC_BITS);
+
+ if ((4 * abs(*alpha) + 7 * abs(*beta) > (1 << WARPEDMODEL_PREC_BITS)) ||
+ (4 * abs(*gamma) + 7 * abs(*delta) > (1 << WARPEDMODEL_PREC_BITS)))
+ continue;
+
+ // We have a valid model, so finish
+ return;
+ }
+ }
+
+ void RunCheckOutput() {
+ const int w = 128, h = 128;
+ const int out_w = GET_PARAM(0), out_h = GET_PARAM(1);
+ const int num_iters = GET_PARAM(2);
+ int i, j;
+ uint8_t *input = new uint8_t[w * h];
+ uint8_t *output = new uint8_t[out_w * out_h];
+ uint8_t *output2 = new uint8_t[out_w * out_h];
+ int32_t mat[8], alpha, beta, gamma, delta;
+
+ for (i = 0; i < h; ++i)
+ for (j = 0; j < w; ++j) input[i * w + j] = rnd_.Rand8();
+
+ /* Try different sizes of prediction block */
+ for (i = 0; i < num_iters; ++i) {
+ generate_model(mat, &alpha, &beta, &gamma, &delta);
+ av1_warp_affine_c(mat, input, w, h, w, output, 32, 32, out_w, out_h,
+ out_w, 0, 0, 0, alpha, beta, gamma, delta);
+ av1_warp_affine_sse2(mat, input, w, h, w, output2, 32, 32, out_w, out_h,
+ out_w, 0, 0, 0, alpha, beta, gamma, delta);
+
+ for (j = 0; j < out_w * out_h; ++j)
+ ASSERT_EQ(output[j], output2[j])
+ << "Pixel mismatch at index " << j << " = (" << (j % out_w) << ", "
+ << (j / out_w) << ") on iteration " << i;
+ }
+
+ delete[] input;
+ delete[] output;
+ delete[] output2;
+ }
+
+ ACMRandom rnd_;
+};
+
+TEST_P(AV1WarpFilterTest, CheckOutput) { RunCheckOutput(); }
+
+const WarpTestParam params[] = {
+ make_tuple(4, 4, 50000), make_tuple(8, 8, 50000), make_tuple(64, 64, 1000),
+ make_tuple(4, 16, 20000), make_tuple(32, 8, 10000),
+};
+
+INSTANTIATE_TEST_CASE_P(SSE2, AV1WarpFilterTest, ::testing::ValuesIn(params));
+
+} // namespace
