xref: /aosp_15_r20/external/libaom/test/convolve_test.cc (revision 77c1e3ccc04c968bd2bc212e87364f250e820521)
1 /*
2  * Copyright (c) 2016, Alliance for Open Media. All rights reserved.
3  *
4  * This source code is subject to the terms of the BSD 2 Clause License and
5  * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
6  * was not distributed with this source code in the LICENSE file, you can
7  * obtain it at www.aomedia.org/license/software. If the Alliance for Open
8  * Media Patent License 1.0 was not distributed with this source code in the
9  * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
10  */
11 
12 #include <string.h>
13 #include <tuple>
14 
15 #include "gtest/gtest.h"
16 
17 #include "config/aom_config.h"
18 #include "config/aom_dsp_rtcd.h"
19 
20 #include "aom_dsp/aom_dsp_common.h"
21 #include "aom_dsp/aom_filter.h"
22 #include "aom_mem/aom_mem.h"
23 #include "aom_ports/aom_timer.h"
24 #include "aom_ports/mem.h"
25 #include "av1/common/filter.h"
26 #include "test/acm_random.h"
27 #include "test/register_state_check.h"
28 #include "test/util.h"
29 
30 namespace {
31 
32 static const unsigned int kMaxDimension = MAX_SB_SIZE;
33 static const int kDataAlignment = 16;
34 static const int kOuterBlockSize = 4 * kMaxDimension;
35 static const int kInputStride = kOuterBlockSize;
36 static const int kOutputStride = kOuterBlockSize;
37 static const int kInputBufferSize = kOuterBlockSize * kOuterBlockSize;
38 static const int kOutputBufferSize = kOuterBlockSize * kOuterBlockSize;
39 static const int16_t kInvalidFilter[8] = {};
40 static const int kNumFilterBanks = SWITCHABLE_FILTERS;
41 static const int kNumFilters = 16;
42 
43 typedef void (*ConvolveFunc)(const uint8_t *src, ptrdiff_t src_stride,
44                              uint8_t *dst, ptrdiff_t dst_stride,
45                              const int16_t *filter_x, int filter_x_stride,
46                              const int16_t *filter_y, int filter_y_stride,
47                              int w, int h);
48 
49 struct ConvolveFunctions {
ConvolveFunctions__anonf36255d80111::ConvolveFunctions50   ConvolveFunctions(ConvolveFunc h8, ConvolveFunc v8, int bd)
51       : h8_(h8), v8_(v8), use_highbd_(bd) {}
52 
53   ConvolveFunc h8_;
54   ConvolveFunc v8_;
55   int use_highbd_;  // 0 if high bitdepth not used, else the actual bit depth.
56 };
57 
58 typedef std::tuple<int, int, const ConvolveFunctions *> ConvolveParam;
59 
60 #define ALL_SIZES_64(convolve_fn)                                         \
61   make_tuple(4, 4, &convolve_fn), make_tuple(8, 4, &convolve_fn),         \
62       make_tuple(4, 8, &convolve_fn), make_tuple(8, 8, &convolve_fn),     \
63       make_tuple(16, 8, &convolve_fn), make_tuple(8, 16, &convolve_fn),   \
64       make_tuple(16, 16, &convolve_fn), make_tuple(32, 16, &convolve_fn), \
65       make_tuple(16, 32, &convolve_fn), make_tuple(32, 32, &convolve_fn), \
66       make_tuple(64, 32, &convolve_fn), make_tuple(32, 64, &convolve_fn), \
67       make_tuple(64, 64, &convolve_fn)
68 
69 #define ALL_SIZES(convolve_fn)                                          \
70   make_tuple(128, 64, &convolve_fn), make_tuple(64, 128, &convolve_fn), \
71       make_tuple(128, 128, &convolve_fn), ALL_SIZES_64(convolve_fn)
72 
73 // Reference 8-tap subpixel filter, slightly modified to fit into this test.
74 #define AV1_FILTER_WEIGHT 128
75 #define AV1_FILTER_SHIFT 7
clip_pixel(int x)76 uint8_t clip_pixel(int x) { return x < 0 ? 0 : x > 255 ? 255 : x; }
77 
filter_block2d_8_c(const uint8_t * src_ptr,unsigned int src_stride,const int16_t * HFilter,const int16_t * VFilter,uint8_t * dst_ptr,unsigned int dst_stride,unsigned int output_width,unsigned int output_height)78 void filter_block2d_8_c(const uint8_t *src_ptr, unsigned int src_stride,
79                         const int16_t *HFilter, const int16_t *VFilter,
80                         uint8_t *dst_ptr, unsigned int dst_stride,
81                         unsigned int output_width, unsigned int output_height) {
82   // Between passes, we use an intermediate buffer whose height is extended to
83   // have enough horizontally filtered values as input for the vertical pass.
84   // This buffer is allocated to be big enough for the largest block type we
85   // support.
86   const int kInterp_Extend = 4;
87   const unsigned int intermediate_height =
88       (kInterp_Extend - 1) + output_height + kInterp_Extend;
89   unsigned int i, j;
90 
91   assert(intermediate_height > 7);
92 
93   // Size of intermediate_buffer is max_intermediate_height * filter_max_width,
94   // where max_intermediate_height = (kInterp_Extend - 1) + filter_max_height
95   //                                 + kInterp_Extend
96   //                               = 3 + 16 + 4
97   //                               = 23
98   // and filter_max_width          = 16
99   //
100   uint8_t intermediate_buffer[(kMaxDimension + 8) * kMaxDimension];
101   const int intermediate_next_stride =
102       1 - static_cast<int>(intermediate_height * output_width);
103 
104   // Horizontal pass (src -> transposed intermediate).
105   uint8_t *output_ptr = intermediate_buffer;
106   const int src_next_row_stride = src_stride - output_width;
107   src_ptr -= (kInterp_Extend - 1) * src_stride + (kInterp_Extend - 1);
108   for (i = 0; i < intermediate_height; ++i) {
109     for (j = 0; j < output_width; ++j) {
110       // Apply filter...
111       const int temp = (src_ptr[0] * HFilter[0]) + (src_ptr[1] * HFilter[1]) +
112                        (src_ptr[2] * HFilter[2]) + (src_ptr[3] * HFilter[3]) +
113                        (src_ptr[4] * HFilter[4]) + (src_ptr[5] * HFilter[5]) +
114                        (src_ptr[6] * HFilter[6]) + (src_ptr[7] * HFilter[7]) +
115                        (AV1_FILTER_WEIGHT >> 1);  // Rounding
116 
117       // Normalize back to 0-255...
118       *output_ptr = clip_pixel(temp >> AV1_FILTER_SHIFT);
119       ++src_ptr;
120       output_ptr += intermediate_height;
121     }
122     src_ptr += src_next_row_stride;
123     output_ptr += intermediate_next_stride;
124   }
125 
126   // Vertical pass (transposed intermediate -> dst).
127   src_ptr = intermediate_buffer;
128   const int dst_next_row_stride = dst_stride - output_width;
129   for (i = 0; i < output_height; ++i) {
130     for (j = 0; j < output_width; ++j) {
131       // Apply filter...
132       const int temp = (src_ptr[0] * VFilter[0]) + (src_ptr[1] * VFilter[1]) +
133                        (src_ptr[2] * VFilter[2]) + (src_ptr[3] * VFilter[3]) +
134                        (src_ptr[4] * VFilter[4]) + (src_ptr[5] * VFilter[5]) +
135                        (src_ptr[6] * VFilter[6]) + (src_ptr[7] * VFilter[7]) +
136                        (AV1_FILTER_WEIGHT >> 1);  // Rounding
137 
138       // Normalize back to 0-255...
139       *dst_ptr++ = clip_pixel(temp >> AV1_FILTER_SHIFT);
140       src_ptr += intermediate_height;
141     }
142     src_ptr += intermediate_next_stride;
143     dst_ptr += dst_next_row_stride;
144   }
145 }
146 
block2d_average_c(uint8_t * src,unsigned int src_stride,uint8_t * output_ptr,unsigned int output_stride,unsigned int output_width,unsigned int output_height)147 void block2d_average_c(uint8_t *src, unsigned int src_stride,
148                        uint8_t *output_ptr, unsigned int output_stride,
149                        unsigned int output_width, unsigned int output_height) {
150   unsigned int i, j;
151   for (i = 0; i < output_height; ++i) {
152     for (j = 0; j < output_width; ++j) {
153       output_ptr[j] = (output_ptr[j] + src[i * src_stride + j] + 1) >> 1;
154     }
155     output_ptr += output_stride;
156   }
157 }
158 
filter_average_block2d_8_c(const uint8_t * src_ptr,const unsigned int src_stride,const int16_t * HFilter,const int16_t * VFilter,uint8_t * dst_ptr,unsigned int dst_stride,unsigned int output_width,unsigned int output_height)159 void filter_average_block2d_8_c(const uint8_t *src_ptr,
160                                 const unsigned int src_stride,
161                                 const int16_t *HFilter, const int16_t *VFilter,
162                                 uint8_t *dst_ptr, unsigned int dst_stride,
163                                 unsigned int output_width,
164                                 unsigned int output_height) {
165   uint8_t tmp[kMaxDimension * kMaxDimension];
166 
167   assert(output_width <= kMaxDimension);
168   assert(output_height <= kMaxDimension);
169   filter_block2d_8_c(src_ptr, src_stride, HFilter, VFilter, tmp, kMaxDimension,
170                      output_width, output_height);
171   block2d_average_c(tmp, kMaxDimension, dst_ptr, dst_stride, output_width,
172                     output_height);
173 }
174 
highbd_filter_block2d_8_c(const uint16_t * src_ptr,const unsigned int src_stride,const int16_t * HFilter,const int16_t * VFilter,uint16_t * dst_ptr,unsigned int dst_stride,unsigned int output_width,unsigned int output_height,int bd)175 void highbd_filter_block2d_8_c(const uint16_t *src_ptr,
176                                const unsigned int src_stride,
177                                const int16_t *HFilter, const int16_t *VFilter,
178                                uint16_t *dst_ptr, unsigned int dst_stride,
179                                unsigned int output_width,
180                                unsigned int output_height, int bd) {
181   // Between passes, we use an intermediate buffer whose height is extended to
182   // have enough horizontally filtered values as input for the vertical pass.
183   // This buffer is allocated to be big enough for the largest block type we
184   // support.
185   const int kInterp_Extend = 4;
186   const unsigned int intermediate_height =
187       (kInterp_Extend - 1) + output_height + kInterp_Extend;
188 
189   /* Size of intermediate_buffer is max_intermediate_height * filter_max_width,
190    * where max_intermediate_height = (kInterp_Extend - 1) + filter_max_height
191    *                                 + kInterp_Extend
192    *                               = 3 + 16 + 4
193    *                               = 23
194    * and filter_max_width = 16
195    */
196   uint16_t intermediate_buffer[(kMaxDimension + 8) * kMaxDimension] = { 0 };
197   const int intermediate_next_stride =
198       1 - static_cast<int>(intermediate_height * output_width);
199 
200   // Horizontal pass (src -> transposed intermediate).
201   {
202     uint16_t *output_ptr = intermediate_buffer;
203     const int src_next_row_stride = src_stride - output_width;
204     unsigned int i, j;
205     src_ptr -= (kInterp_Extend - 1) * src_stride + (kInterp_Extend - 1);
206     for (i = 0; i < intermediate_height; ++i) {
207       for (j = 0; j < output_width; ++j) {
208         // Apply filter...
209         const int temp = (src_ptr[0] * HFilter[0]) + (src_ptr[1] * HFilter[1]) +
210                          (src_ptr[2] * HFilter[2]) + (src_ptr[3] * HFilter[3]) +
211                          (src_ptr[4] * HFilter[4]) + (src_ptr[5] * HFilter[5]) +
212                          (src_ptr[6] * HFilter[6]) + (src_ptr[7] * HFilter[7]) +
213                          (AV1_FILTER_WEIGHT >> 1);  // Rounding
214 
215         // Normalize back to 0-255...
216         *output_ptr = clip_pixel_highbd(temp >> AV1_FILTER_SHIFT, bd);
217         ++src_ptr;
218         output_ptr += intermediate_height;
219       }
220       src_ptr += src_next_row_stride;
221       output_ptr += intermediate_next_stride;
222     }
223   }
224 
225   // Vertical pass (transposed intermediate -> dst).
226   {
227     const uint16_t *interm_ptr = intermediate_buffer;
228     const int dst_next_row_stride = dst_stride - output_width;
229     unsigned int i, j;
230     for (i = 0; i < output_height; ++i) {
231       for (j = 0; j < output_width; ++j) {
232         // Apply filter...
233         const int temp =
234             (interm_ptr[0] * VFilter[0]) + (interm_ptr[1] * VFilter[1]) +
235             (interm_ptr[2] * VFilter[2]) + (interm_ptr[3] * VFilter[3]) +
236             (interm_ptr[4] * VFilter[4]) + (interm_ptr[5] * VFilter[5]) +
237             (interm_ptr[6] * VFilter[6]) + (interm_ptr[7] * VFilter[7]) +
238             (AV1_FILTER_WEIGHT >> 1);  // Rounding
239 
240         // Normalize back to 0-255...
241         *dst_ptr++ = clip_pixel_highbd(temp >> AV1_FILTER_SHIFT, bd);
242         interm_ptr += intermediate_height;
243       }
244       interm_ptr += intermediate_next_stride;
245       dst_ptr += dst_next_row_stride;
246     }
247   }
248 }
249 
highbd_block2d_average_c(uint16_t * src,unsigned int src_stride,uint16_t * output_ptr,unsigned int output_stride,unsigned int output_width,unsigned int output_height)250 void highbd_block2d_average_c(uint16_t *src, unsigned int src_stride,
251                               uint16_t *output_ptr, unsigned int output_stride,
252                               unsigned int output_width,
253                               unsigned int output_height) {
254   unsigned int i, j;
255   for (i = 0; i < output_height; ++i) {
256     for (j = 0; j < output_width; ++j) {
257       output_ptr[j] = (output_ptr[j] + src[i * src_stride + j] + 1) >> 1;
258     }
259     output_ptr += output_stride;
260   }
261 }
262 
highbd_filter_average_block2d_8_c(const uint16_t * src_ptr,unsigned int src_stride,const int16_t * HFilter,const int16_t * VFilter,uint16_t * dst_ptr,unsigned int dst_stride,unsigned int output_width,unsigned int output_height,int bd)263 void highbd_filter_average_block2d_8_c(
264     const uint16_t *src_ptr, unsigned int src_stride, const int16_t *HFilter,
265     const int16_t *VFilter, uint16_t *dst_ptr, unsigned int dst_stride,
266     unsigned int output_width, unsigned int output_height, int bd) {
267   uint16_t tmp[kMaxDimension * kMaxDimension];
268 
269   assert(output_width <= kMaxDimension);
270   assert(output_height <= kMaxDimension);
271   highbd_filter_block2d_8_c(src_ptr, src_stride, HFilter, VFilter, tmp,
272                             kMaxDimension, output_width, output_height, bd);
273   highbd_block2d_average_c(tmp, kMaxDimension, dst_ptr, dst_stride,
274                            output_width, output_height);
275 }
276 
277 class ConvolveTestBase : public ::testing::TestWithParam<ConvolveParam> {
278  public:
SetUpTestSuite()279   static void SetUpTestSuite() {
280     // Force input_ to be unaligned, output to be 16 byte aligned.
281     input_ = reinterpret_cast<uint8_t *>(
282                  aom_memalign(kDataAlignment, kInputBufferSize + 1)) +
283              1;
284     ASSERT_NE(input_, nullptr);
285     ref8_ = reinterpret_cast<uint8_t *>(
286         aom_memalign(kDataAlignment, kOutputStride * kMaxDimension));
287     ASSERT_NE(ref8_, nullptr);
288     output_ = reinterpret_cast<uint8_t *>(
289         aom_memalign(kDataAlignment, kOutputBufferSize));
290     ASSERT_NE(output_, nullptr);
291     output_ref_ = reinterpret_cast<uint8_t *>(
292         aom_memalign(kDataAlignment, kOutputBufferSize));
293     ASSERT_NE(output_ref_, nullptr);
294     input16_ = reinterpret_cast<uint16_t *>(aom_memalign(
295                    kDataAlignment, (kInputBufferSize + 1) * sizeof(uint16_t))) +
296                1;
297     ASSERT_NE(input16_, nullptr);
298     ref16_ = reinterpret_cast<uint16_t *>(aom_memalign(
299         kDataAlignment, kOutputStride * kMaxDimension * sizeof(uint16_t)));
300     ASSERT_NE(ref16_, nullptr);
301     output16_ = reinterpret_cast<uint16_t *>(
302         aom_memalign(kDataAlignment, (kOutputBufferSize) * sizeof(uint16_t)));
303     ASSERT_NE(output16_, nullptr);
304     output16_ref_ = reinterpret_cast<uint16_t *>(
305         aom_memalign(kDataAlignment, (kOutputBufferSize) * sizeof(uint16_t)));
306     ASSERT_NE(output16_ref_, nullptr);
307   }
308 
TearDownTestSuite()309   static void TearDownTestSuite() {
310     aom_free(input_ - 1);
311     input_ = nullptr;
312     aom_free(ref8_);
313     ref8_ = nullptr;
314     aom_free(output_);
315     output_ = nullptr;
316     aom_free(output_ref_);
317     output_ref_ = nullptr;
318     aom_free(input16_ - 1);
319     input16_ = nullptr;
320     aom_free(ref16_);
321     ref16_ = nullptr;
322     aom_free(output16_);
323     output16_ = nullptr;
324     aom_free(output16_ref_);
325     output16_ref_ = nullptr;
326   }
327 
328  protected:
Width() const329   int Width() const { return GET_PARAM(0); }
Height() const330   int Height() const { return GET_PARAM(1); }
BorderLeft() const331   int BorderLeft() const {
332     const int center = (kOuterBlockSize - Width()) / 2;
333     return (center + (kDataAlignment - 1)) & ~(kDataAlignment - 1);
334   }
BorderTop() const335   int BorderTop() const { return (kOuterBlockSize - Height()) / 2; }
336 
IsIndexInBorder(int i)337   bool IsIndexInBorder(int i) {
338     return (i < BorderTop() * kOuterBlockSize ||
339             i >= (BorderTop() + Height()) * kOuterBlockSize ||
340             i % kOuterBlockSize < BorderLeft() ||
341             i % kOuterBlockSize >= (BorderLeft() + Width()));
342   }
343 
SetUp()344   void SetUp() override {
345     UUT_ = GET_PARAM(2);
346     if (UUT_->use_highbd_ != 0)
347       mask_ = (1 << UUT_->use_highbd_) - 1;
348     else
349       mask_ = 255;
350     /* Set up guard blocks for an inner block centered in the outer block */
351     for (int i = 0; i < kOutputBufferSize; ++i) {
352       if (IsIndexInBorder(i)) {
353         output_[i] = 255;
354         output16_[i] = mask_;
355       } else {
356         output_[i] = 0;
357         output16_[i] = 0;
358       }
359     }
360 
361     ::libaom_test::ACMRandom prng;
362     for (int i = 0; i < kInputBufferSize; ++i) {
363       if (i & 1) {
364         input_[i] = 255;
365         input16_[i] = mask_;
366       } else {
367         input_[i] = prng.Rand8Extremes();
368         input16_[i] = prng.Rand16() & mask_;
369       }
370     }
371   }
372 
SetConstantInput(int value)373   void SetConstantInput(int value) {
374     memset(input_, value, kInputBufferSize);
375     aom_memset16(input16_, value, kInputBufferSize);
376   }
377 
CopyOutputToRef()378   void CopyOutputToRef() {
379     memcpy(output_ref_, output_, kOutputBufferSize);
380     // Copy 16-bit pixels values. The effective number of bytes is double.
381     memcpy(output16_ref_, output16_, sizeof(output16_[0]) * kOutputBufferSize);
382   }
383 
CheckGuardBlocks()384   void CheckGuardBlocks() {
385     for (int i = 0; i < kOutputBufferSize; ++i) {
386       if (IsIndexInBorder(i)) {
387         EXPECT_EQ(255, output_[i]);
388       }
389     }
390   }
391 
input() const392   uint8_t *input() const {
393     const int offset = BorderTop() * kOuterBlockSize + BorderLeft();
394     if (UUT_->use_highbd_ == 0) {
395       return input_ + offset;
396     } else {
397       return CONVERT_TO_BYTEPTR(input16_) + offset;
398     }
399   }
400 
output() const401   uint8_t *output() const {
402     const int offset = BorderTop() * kOuterBlockSize + BorderLeft();
403     if (UUT_->use_highbd_ == 0) {
404       return output_ + offset;
405     } else {
406       return CONVERT_TO_BYTEPTR(output16_) + offset;
407     }
408   }
409 
output_ref() const410   uint8_t *output_ref() const {
411     const int offset = BorderTop() * kOuterBlockSize + BorderLeft();
412     if (UUT_->use_highbd_ == 0) {
413       return output_ref_ + offset;
414     } else {
415       return CONVERT_TO_BYTEPTR(output16_ref_) + offset;
416     }
417   }
418 
lookup(uint8_t * list,int index) const419   uint16_t lookup(uint8_t *list, int index) const {
420     if (UUT_->use_highbd_ == 0) {
421       return list[index];
422     } else {
423       return CONVERT_TO_SHORTPTR(list)[index];
424     }
425   }
426 
assign_val(uint8_t * list,int index,uint16_t val) const427   void assign_val(uint8_t *list, int index, uint16_t val) const {
428     if (UUT_->use_highbd_ == 0) {
429       list[index] = (uint8_t)val;
430     } else {
431       CONVERT_TO_SHORTPTR(list)[index] = val;
432     }
433   }
434 
wrapper_filter_average_block2d_8_c(const uint8_t * src_ptr,unsigned int src_stride,const int16_t * HFilter,const int16_t * VFilter,uint8_t * dst_ptr,unsigned int dst_stride,unsigned int output_width,unsigned int output_height)435   void wrapper_filter_average_block2d_8_c(
436       const uint8_t *src_ptr, unsigned int src_stride, const int16_t *HFilter,
437       const int16_t *VFilter, uint8_t *dst_ptr, unsigned int dst_stride,
438       unsigned int output_width, unsigned int output_height) {
439     if (UUT_->use_highbd_ == 0) {
440       filter_average_block2d_8_c(src_ptr, src_stride, HFilter, VFilter, dst_ptr,
441                                  dst_stride, output_width, output_height);
442     } else {
443       highbd_filter_average_block2d_8_c(
444           CONVERT_TO_SHORTPTR(src_ptr), src_stride, HFilter, VFilter,
445           CONVERT_TO_SHORTPTR(dst_ptr), dst_stride, output_width, output_height,
446           UUT_->use_highbd_);
447     }
448   }
449 
wrapper_filter_block2d_8_c(const uint8_t * src_ptr,unsigned int src_stride,const int16_t * HFilter,const int16_t * VFilter,uint8_t * dst_ptr,unsigned int dst_stride,unsigned int output_width,unsigned int output_height)450   void wrapper_filter_block2d_8_c(
451       const uint8_t *src_ptr, unsigned int src_stride, const int16_t *HFilter,
452       const int16_t *VFilter, uint8_t *dst_ptr, unsigned int dst_stride,
453       unsigned int output_width, unsigned int output_height) {
454     if (UUT_->use_highbd_ == 0) {
455       filter_block2d_8_c(src_ptr, src_stride, HFilter, VFilter, dst_ptr,
456                          dst_stride, output_width, output_height);
457     } else {
458       highbd_filter_block2d_8_c(CONVERT_TO_SHORTPTR(src_ptr), src_stride,
459                                 HFilter, VFilter, CONVERT_TO_SHORTPTR(dst_ptr),
460                                 dst_stride, output_width, output_height,
461                                 UUT_->use_highbd_);
462     }
463   }
464 
MatchesReferenceSubpixelFilter()465   void MatchesReferenceSubpixelFilter() {
466     uint8_t *const in = input();
467     uint8_t *const out = output();
468     uint8_t *ref;
469     if (UUT_->use_highbd_ == 0) {
470       ref = ref8_;
471     } else {
472       ref = CONVERT_TO_BYTEPTR(ref16_);
473     }
474     int subpel_search;
475     for (subpel_search = USE_2_TAPS; subpel_search <= USE_8_TAPS;
476          ++subpel_search) {
477       for (int filter_bank = 0; filter_bank < kNumFilterBanks; ++filter_bank) {
478         const InterpFilter filter = (InterpFilter)filter_bank;
479         const InterpKernel *filters =
480             (const InterpKernel *)av1_get_interp_filter_kernel(filter,
481                                                                subpel_search);
482         for (int filter_x = 0; filter_x < kNumFilters; ++filter_x) {
483           for (int filter_y = 0; filter_y < kNumFilters; ++filter_y) {
484             wrapper_filter_block2d_8_c(in, kInputStride, filters[filter_x],
485                                        filters[filter_y], ref, kOutputStride,
486                                        Width(), Height());
487 
488             if (filter_x && filter_y)
489               continue;
490             else if (filter_y)
491               UUT_->v8_(in, kInputStride, out, kOutputStride, kInvalidFilter,
492                         16, filters[filter_y], 16, Width(), Height());
493             else if (filter_x)
494               API_REGISTER_STATE_CHECK(UUT_->h8_(
495                   in, kInputStride, out, kOutputStride, filters[filter_x], 16,
496                   kInvalidFilter, 16, Width(), Height()));
497             else
498               continue;
499 
500             CheckGuardBlocks();
501 
502             for (int y = 0; y < Height(); ++y)
503               for (int x = 0; x < Width(); ++x)
504                 ASSERT_EQ(lookup(ref, y * kOutputStride + x),
505                           lookup(out, y * kOutputStride + x))
506                     << "mismatch at (" << x << "," << y << "), "
507                     << "filters (" << filter_bank << "," << filter_x << ","
508                     << filter_y << ")";
509           }
510         }
511       }
512     }
513   }
514 
FilterExtremes()515   void FilterExtremes() {
516     uint8_t *const in = input();
517     uint8_t *const out = output();
518     uint8_t *ref;
519     if (UUT_->use_highbd_ == 0) {
520       ref = ref8_;
521     } else {
522       ref = CONVERT_TO_BYTEPTR(ref16_);
523     }
524 
525     // Populate ref and out with some random data
526     ::libaom_test::ACMRandom prng;
527     for (int y = 0; y < Height(); ++y) {
528       for (int x = 0; x < Width(); ++x) {
529         uint16_t r;
530         if (UUT_->use_highbd_ == 0 || UUT_->use_highbd_ == 8) {
531           r = prng.Rand8Extremes();
532         } else {
533           r = prng.Rand16() & mask_;
534         }
535         assign_val(out, y * kOutputStride + x, r);
536         assign_val(ref, y * kOutputStride + x, r);
537       }
538     }
539 
540     for (int axis = 0; axis < 2; axis++) {
541       int seed_val = 0;
542       while (seed_val < 256) {
543         for (int y = 0; y < 8; ++y) {
544           for (int x = 0; x < 8; ++x) {
545             assign_val(in, y * kOutputStride + x - SUBPEL_TAPS / 2 + 1,
546                        ((seed_val >> (axis ? y : x)) & 1) * mask_);
547             if (axis) seed_val++;
548           }
549           if (axis)
550             seed_val -= 8;
551           else
552             seed_val++;
553         }
554         if (axis) seed_val += 8;
555         int subpel_search;
556         for (subpel_search = USE_2_TAPS; subpel_search <= USE_8_TAPS;
557              ++subpel_search) {
558           for (int filter_bank = 0; filter_bank < kNumFilterBanks;
559                ++filter_bank) {
560             const InterpFilter filter = (InterpFilter)filter_bank;
561             const InterpKernel *filters =
562                 (const InterpKernel *)av1_get_interp_filter_kernel(
563                     filter, subpel_search);
564             for (int filter_x = 0; filter_x < kNumFilters; ++filter_x) {
565               for (int filter_y = 0; filter_y < kNumFilters; ++filter_y) {
566                 wrapper_filter_block2d_8_c(in, kInputStride, filters[filter_x],
567                                            filters[filter_y], ref,
568                                            kOutputStride, Width(), Height());
569                 if (filter_x && filter_y)
570                   continue;
571                 else if (filter_y)
572                   API_REGISTER_STATE_CHECK(UUT_->v8_(
573                       in, kInputStride, out, kOutputStride, kInvalidFilter, 16,
574                       filters[filter_y], 16, Width(), Height()));
575                 else if (filter_x)
576                   API_REGISTER_STATE_CHECK(UUT_->h8_(
577                       in, kInputStride, out, kOutputStride, filters[filter_x],
578                       16, kInvalidFilter, 16, Width(), Height()));
579                 else
580                   continue;
581 
582                 for (int y = 0; y < Height(); ++y)
583                   for (int x = 0; x < Width(); ++x)
584                     ASSERT_EQ(lookup(ref, y * kOutputStride + x),
585                               lookup(out, y * kOutputStride + x))
586                         << "mismatch at (" << x << "," << y << "), "
587                         << "filters (" << filter_bank << "," << filter_x << ","
588                         << filter_y << ")";
589               }
590             }
591           }
592         }
593       }
594     }
595   }
596 
SpeedTest()597   void SpeedTest() {
598     uint8_t *const in = input();
599     uint8_t *const out = output();
600     uint8_t *ref;
601     if (UUT_->use_highbd_ == 0) {
602       ref = ref8_;
603     } else {
604       ref = CONVERT_TO_BYTEPTR(ref16_);
605     }
606 
607     // Populate ref and out with some random data
608     ::libaom_test::ACMRandom prng;
609     for (int y = 0; y < Height(); ++y) {
610       for (int x = 0; x < Width(); ++x) {
611         uint16_t r;
612         if (UUT_->use_highbd_ == 0 || UUT_->use_highbd_ == 8) {
613           r = prng.Rand8Extremes();
614         } else {
615           r = prng.Rand16() & mask_;
616         }
617         assign_val(out, y * kOutputStride + x, r);
618         assign_val(ref, y * kOutputStride + x, r);
619       }
620     }
621 
622     InterpFilter filter = (InterpFilter)1;
623     const InterpKernel *filters =
624         (const InterpKernel *)av1_get_interp_filter_kernel(filter, USE_8_TAPS);
625     wrapper_filter_average_block2d_8_c(in, kInputStride, filters[1], filters[1],
626                                        out, kOutputStride, Width(), Height());
627 
628     aom_usec_timer timer;
629     int tests_num = 1000;
630 
631     aom_usec_timer_start(&timer);
632     while (tests_num > 0) {
633       for (int filter_bank = 0; filter_bank < kNumFilterBanks; ++filter_bank) {
634         filter = (InterpFilter)filter_bank;
635         filters = (const InterpKernel *)av1_get_interp_filter_kernel(
636             filter, USE_8_TAPS);
637         for (int filter_x = 0; filter_x < kNumFilters; ++filter_x) {
638           for (int filter_y = 0; filter_y < kNumFilters; ++filter_y) {
639             if (filter_x && filter_y) continue;
640             if (filter_y)
641               API_REGISTER_STATE_CHECK(UUT_->v8_(
642                   in, kInputStride, out, kOutputStride, kInvalidFilter, 16,
643                   filters[filter_y], 16, Width(), Height()));
644             else if (filter_x)
645               API_REGISTER_STATE_CHECK(UUT_->h8_(
646                   in, kInputStride, out, kOutputStride, filters[filter_x], 16,
647                   kInvalidFilter, 16, Width(), Height()));
648           }
649         }
650       }
651       tests_num--;
652     }
653     aom_usec_timer_mark(&timer);
654 
655     const int elapsed_time =
656         static_cast<int>(aom_usec_timer_elapsed(&timer) / 1000);
657     printf("%dx%d (bitdepth %d) time: %5d ms\n", Width(), Height(),
658            UUT_->use_highbd_, elapsed_time);
659   }
660 
661   const ConvolveFunctions *UUT_;
662   static uint8_t *input_;
663   static uint8_t *ref8_;
664   static uint8_t *output_;
665   static uint8_t *output_ref_;
666   static uint16_t *input16_;
667   static uint16_t *ref16_;
668   static uint16_t *output16_;
669   static uint16_t *output16_ref_;
670   int mask_;
671 };
672 
673 uint8_t *ConvolveTestBase::input_ = nullptr;
674 uint8_t *ConvolveTestBase::ref8_ = nullptr;
675 uint8_t *ConvolveTestBase::output_ = nullptr;
676 uint8_t *ConvolveTestBase::output_ref_ = nullptr;
677 uint16_t *ConvolveTestBase::input16_ = nullptr;
678 uint16_t *ConvolveTestBase::ref16_ = nullptr;
679 uint16_t *ConvolveTestBase::output16_ = nullptr;
680 uint16_t *ConvolveTestBase::output16_ref_ = nullptr;
681 
682 using LowbdConvolveTest = ConvolveTestBase;
683 
TEST_P(LowbdConvolveTest,GuardBlocks)684 TEST_P(LowbdConvolveTest, GuardBlocks) { CheckGuardBlocks(); }
685 
FiltersWontSaturateWhenAddedPairwise()686 void FiltersWontSaturateWhenAddedPairwise() {
687   int subpel_search;
688   for (subpel_search = USE_2_TAPS; subpel_search <= USE_8_TAPS;
689        ++subpel_search) {
690     for (int filter_bank = 0; filter_bank < kNumFilterBanks; ++filter_bank) {
691       const InterpFilter filter = (InterpFilter)filter_bank;
692       const InterpKernel *filters =
693           (const InterpKernel *)av1_get_interp_filter_kernel(filter,
694                                                              subpel_search);
695       for (int i = 0; i < kNumFilters; i++) {
696         const int p0 = filters[i][0] + filters[i][1];
697         const int p1 = filters[i][2] + filters[i][3];
698         const int p2 = filters[i][4] + filters[i][5];
699         const int p3 = filters[i][6] + filters[i][7];
700         EXPECT_LE(p0, 128);
701         EXPECT_LE(p1, 128);
702         EXPECT_LE(p2, 128);
703         EXPECT_LE(p3, 128);
704         EXPECT_LE(p0 + p3, 128);
705         EXPECT_LE(p0 + p3 + p1, 128);
706         EXPECT_LE(p0 + p3 + p1 + p2, 128);
707         EXPECT_EQ(p0 + p1 + p2 + p3, 128);
708       }
709     }
710   }
711 }
712 
TEST(LowbdConvolveTest,FiltersWontSaturateWhenAddedPairwise)713 TEST(LowbdConvolveTest, FiltersWontSaturateWhenAddedPairwise) {
714   FiltersWontSaturateWhenAddedPairwise();
715 }
716 
TEST_P(LowbdConvolveTest,MatchesReferenceSubpixelFilter)717 TEST_P(LowbdConvolveTest, MatchesReferenceSubpixelFilter) {
718   MatchesReferenceSubpixelFilter();
719 }
720 
TEST_P(LowbdConvolveTest,FilterExtremes)721 TEST_P(LowbdConvolveTest, FilterExtremes) { FilterExtremes(); }
722 
TEST_P(LowbdConvolveTest,DISABLED_Speed)723 TEST_P(LowbdConvolveTest, DISABLED_Speed) { SpeedTest(); }
724 
725 using std::make_tuple;
726 
727 // WRAP macro is only used for high bitdepth build.
728 #if CONFIG_AV1_HIGHBITDEPTH
729 #define WRAP(func, bd)                                                       \
730   static void wrap_##func##_##bd(                                            \
731       const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,                \
732       ptrdiff_t dst_stride, const int16_t *filter_x, int filter_x_stride,    \
733       const int16_t *filter_y, int filter_y_stride, int w, int h) {          \
734     aom_highbd_##func(src, src_stride, dst, dst_stride, filter_x,            \
735                       filter_x_stride, filter_y, filter_y_stride, w, h, bd); \
736   }
737 #if HAVE_SSE2 && AOM_ARCH_X86_64
738 WRAP(convolve8_horiz_sse2, 8)
739 WRAP(convolve8_vert_sse2, 8)
740 WRAP(convolve8_horiz_sse2, 10)
741 WRAP(convolve8_vert_sse2, 10)
742 WRAP(convolve8_horiz_sse2, 12)
743 WRAP(convolve8_vert_sse2, 12)
744 #endif  // HAVE_SSE2 && AOM_ARCH_X86_64
745 
746 WRAP(convolve8_horiz_c, 8)
747 WRAP(convolve8_vert_c, 8)
748 WRAP(convolve8_horiz_c, 10)
749 WRAP(convolve8_vert_c, 10)
750 WRAP(convolve8_horiz_c, 12)
751 WRAP(convolve8_vert_c, 12)
752 
753 #if HAVE_AVX2
754 WRAP(convolve8_horiz_avx2, 8)
755 WRAP(convolve8_vert_avx2, 8)
756 
757 WRAP(convolve8_horiz_avx2, 10)
758 WRAP(convolve8_vert_avx2, 10)
759 
760 WRAP(convolve8_horiz_avx2, 12)
761 WRAP(convolve8_vert_avx2, 12)
762 #endif  // HAVE_AVX2
763 
764 #if HAVE_NEON
765 WRAP(convolve8_horiz_neon, 8)
766 WRAP(convolve8_vert_neon, 8)
767 
768 WRAP(convolve8_horiz_neon, 10)
769 WRAP(convolve8_vert_neon, 10)
770 
771 WRAP(convolve8_horiz_neon, 12)
772 WRAP(convolve8_vert_neon, 12)
773 #endif  // HAVE_NEON
774 
775 #if HAVE_SVE
776 WRAP(convolve8_horiz_sve, 8)
777 WRAP(convolve8_vert_sve, 8)
778 
779 WRAP(convolve8_horiz_sve, 10)
780 WRAP(convolve8_vert_sve, 10)
781 
782 WRAP(convolve8_horiz_sve, 12)
783 WRAP(convolve8_vert_sve, 12)
784 #endif  // HAVE_SVE
785 #endif  // CONFIG_AV1_HIGHBITDEPTH
786 
787 #undef WRAP
788 
789 #if CONFIG_AV1_HIGHBITDEPTH
790 
791 using HighbdConvolveTest = ConvolveTestBase;
792 
TEST_P(HighbdConvolveTest,GuardBlocks)793 TEST_P(HighbdConvolveTest, GuardBlocks) { CheckGuardBlocks(); }
794 
TEST(HighbdConvolveTest,FiltersWontSaturateWhenAddedPairwise)795 TEST(HighbdConvolveTest, FiltersWontSaturateWhenAddedPairwise) {
796   FiltersWontSaturateWhenAddedPairwise();
797 }
798 
TEST_P(HighbdConvolveTest,MatchesReferenceSubpixelFilter)799 TEST_P(HighbdConvolveTest, MatchesReferenceSubpixelFilter) {
800   MatchesReferenceSubpixelFilter();
801 }
802 
TEST_P(HighbdConvolveTest,FilterExtremes)803 TEST_P(HighbdConvolveTest, FilterExtremes) { FilterExtremes(); }
804 
TEST_P(HighbdConvolveTest,DISABLED_Speed)805 TEST_P(HighbdConvolveTest, DISABLED_Speed) { SpeedTest(); }
806 
807 const ConvolveFunctions wrap_convolve8_c(wrap_convolve8_horiz_c_8,
808                                          wrap_convolve8_vert_c_8, 8);
809 const ConvolveFunctions wrap_convolve10_c(wrap_convolve8_horiz_c_10,
810                                           wrap_convolve8_vert_c_10, 10);
811 const ConvolveFunctions wrap_convolve12_c(wrap_convolve8_horiz_c_12,
812                                           wrap_convolve8_vert_c_12, 12);
813 const ConvolveParam kArrayHighbdConvolve_c[] = { ALL_SIZES(wrap_convolve8_c),
814                                                  ALL_SIZES(wrap_convolve10_c),
815                                                  ALL_SIZES(wrap_convolve12_c) };
816 
817 INSTANTIATE_TEST_SUITE_P(C, HighbdConvolveTest,
818                          ::testing::ValuesIn(kArrayHighbdConvolve_c));
819 #endif  // CONFIG_AV1_HIGHBITDEPTH
820 
821 const ConvolveFunctions convolve8_c(aom_convolve8_horiz_c, aom_convolve8_vert_c,
822                                     0);
823 const ConvolveParam kArrayConvolve_c[] = { ALL_SIZES(convolve8_c) };
824 
825 INSTANTIATE_TEST_SUITE_P(C, LowbdConvolveTest,
826                          ::testing::ValuesIn(kArrayConvolve_c));
827 
828 #if HAVE_SSE2 && AOM_ARCH_X86_64
829 #if CONFIG_AV1_HIGHBITDEPTH
830 const ConvolveFunctions wrap_convolve8_sse2(wrap_convolve8_horiz_sse2_8,
831                                             wrap_convolve8_vert_sse2_8, 8);
832 const ConvolveFunctions wrap_convolve10_sse2(wrap_convolve8_horiz_sse2_10,
833                                              wrap_convolve8_vert_sse2_10, 10);
834 const ConvolveFunctions wrap_convolve12_sse2(wrap_convolve8_horiz_sse2_12,
835                                              wrap_convolve8_vert_sse2_12, 12);
836 const ConvolveParam kArrayHighbdConvolve_sse2[] = {
837   ALL_SIZES(wrap_convolve8_sse2), ALL_SIZES(wrap_convolve10_sse2),
838   ALL_SIZES(wrap_convolve12_sse2)
839 };
840 
841 INSTANTIATE_TEST_SUITE_P(SSE2, HighbdConvolveTest,
842                          ::testing::ValuesIn(kArrayHighbdConvolve_sse2));
843 #endif
844 #endif
845 
846 #if HAVE_SSSE3
847 const ConvolveFunctions convolve8_ssse3(aom_convolve8_horiz_ssse3,
848                                         aom_convolve8_vert_ssse3, 0);
849 
850 const ConvolveParam kArrayConvolve8_ssse3[] = { ALL_SIZES(convolve8_ssse3) };
851 
852 INSTANTIATE_TEST_SUITE_P(SSSE3, LowbdConvolveTest,
853                          ::testing::ValuesIn(kArrayConvolve8_ssse3));
854 #endif
855 
856 #if HAVE_AVX2
857 #if CONFIG_AV1_HIGHBITDEPTH
858 const ConvolveFunctions wrap_convolve8_avx2(wrap_convolve8_horiz_avx2_8,
859                                             wrap_convolve8_vert_avx2_8, 8);
860 const ConvolveFunctions wrap_convolve10_avx2(wrap_convolve8_horiz_avx2_10,
861                                              wrap_convolve8_vert_avx2_10, 10);
862 const ConvolveFunctions wrap_convolve12_avx2(wrap_convolve8_horiz_avx2_12,
863                                              wrap_convolve8_vert_avx2_12, 12);
864 const ConvolveParam kArray_HighbdConvolve8_avx2[] = {
865   ALL_SIZES_64(wrap_convolve8_avx2), ALL_SIZES_64(wrap_convolve10_avx2),
866   ALL_SIZES_64(wrap_convolve12_avx2)
867 };
868 
869 INSTANTIATE_TEST_SUITE_P(AVX2, HighbdConvolveTest,
870                          ::testing::ValuesIn(kArray_HighbdConvolve8_avx2));
871 #endif
872 const ConvolveFunctions convolve8_avx2(aom_convolve8_horiz_avx2,
873                                        aom_convolve8_vert_avx2, 0);
874 const ConvolveParam kArray_Convolve8_avx2[] = { ALL_SIZES(convolve8_avx2) };
875 
876 INSTANTIATE_TEST_SUITE_P(AVX2, LowbdConvolveTest,
877                          ::testing::ValuesIn(kArray_Convolve8_avx2));
878 #endif  // HAVE_AVX2
879 
880 #if HAVE_NEON
881 #if CONFIG_AV1_HIGHBITDEPTH
882 const ConvolveFunctions wrap_convolve8_neon(wrap_convolve8_horiz_neon_8,
883                                             wrap_convolve8_vert_neon_8, 8);
884 const ConvolveFunctions wrap_convolve10_neon(wrap_convolve8_horiz_neon_10,
885                                              wrap_convolve8_vert_neon_10, 10);
886 const ConvolveFunctions wrap_convolve12_neon(wrap_convolve8_horiz_neon_12,
887                                              wrap_convolve8_vert_neon_12, 12);
888 const ConvolveParam kArray_HighbdConvolve8_neon[] = {
889   ALL_SIZES_64(wrap_convolve8_neon), ALL_SIZES_64(wrap_convolve10_neon),
890   ALL_SIZES_64(wrap_convolve12_neon)
891 };
892 
893 INSTANTIATE_TEST_SUITE_P(NEON, HighbdConvolveTest,
894                          ::testing::ValuesIn(kArray_HighbdConvolve8_neon));
895 #endif
896 const ConvolveFunctions convolve8_neon(aom_convolve8_horiz_neon,
897                                        aom_convolve8_vert_neon, 0);
898 const ConvolveParam kArray_Convolve8_neon[] = { ALL_SIZES(convolve8_neon) };
899 
900 INSTANTIATE_TEST_SUITE_P(NEON, LowbdConvolveTest,
901                          ::testing::ValuesIn(kArray_Convolve8_neon));
902 #endif  // HAVE_NEON
903 
904 #if HAVE_NEON_DOTPROD
905 const ConvolveFunctions convolve8_neon_dotprod(aom_convolve8_horiz_neon_dotprod,
906                                                aom_convolve8_vert_neon_dotprod,
907                                                0);
908 const ConvolveParam kArray_Convolve8_neon_dotprod[] = { ALL_SIZES(
909     convolve8_neon_dotprod) };
910 
911 INSTANTIATE_TEST_SUITE_P(NEON_DOTPROD, LowbdConvolveTest,
912                          ::testing::ValuesIn(kArray_Convolve8_neon_dotprod));
913 #endif  // HAVE_NEON_DOTPROD
914 
915 #if HAVE_NEON_I8MM
916 const ConvolveFunctions convolve8_neon_i8mm(aom_convolve8_horiz_neon_i8mm,
917                                             aom_convolve8_vert_neon_i8mm, 0);
918 const ConvolveParam kArray_Convolve8_neon_i8mm[] = { ALL_SIZES(
919     convolve8_neon_i8mm) };
920 
921 INSTANTIATE_TEST_SUITE_P(NEON_I8MM, LowbdConvolveTest,
922                          ::testing::ValuesIn(kArray_Convolve8_neon_i8mm));
923 #endif  // HAVE_NEON_I8MM
924 
925 #if HAVE_SVE
926 #if CONFIG_AV1_HIGHBITDEPTH
927 const ConvolveFunctions wrap_convolve8_sve(wrap_convolve8_horiz_sve_8,
928                                            wrap_convolve8_vert_sve_8, 8);
929 const ConvolveFunctions wrap_convolve10_sve(wrap_convolve8_horiz_sve_10,
930                                             wrap_convolve8_vert_sve_10, 10);
931 const ConvolveFunctions wrap_convolve12_sve(wrap_convolve8_horiz_sve_12,
932                                             wrap_convolve8_vert_sve_12, 12);
933 const ConvolveParam kArray_HighbdConvolve8_sve[] = {
934   ALL_SIZES_64(wrap_convolve8_sve), ALL_SIZES_64(wrap_convolve10_sve),
935   ALL_SIZES_64(wrap_convolve12_sve)
936 };
937 
938 INSTANTIATE_TEST_SUITE_P(SVE, HighbdConvolveTest,
939                          ::testing::ValuesIn(kArray_HighbdConvolve8_sve));
940 #endif
941 #endif  // HAVE_SVE
942 
943 typedef void (*ConvolveScale2DFunc)(const uint8_t *src, ptrdiff_t src_stride,
944                                     uint8_t *dst, ptrdiff_t dst_stride,
945                                     const InterpKernel *filter, int x0_q4,
946                                     int x_step_q4, int y0_q4, int y_step_q4,
947                                     int w, int h);
948 
949 typedef std::tuple<int, int, ConvolveScale2DFunc> ConvolveScale2DParam;
950 
951 class ConvolveScale2DTest
952     : public ::testing::TestWithParam<ConvolveScale2DParam> {
953  public:
Width() const954   int Width() const { return GET_PARAM(0); }
Height() const955   int Height() const { return GET_PARAM(1); }
BorderLeft() const956   int BorderLeft() const {
957     const int center = (kOuterBlockSize - Width()) / 2;
958     return (center + (kDataAlignment - 1)) & ~(kDataAlignment - 1);
959   }
BorderTop() const960   int BorderTop() const { return (kOuterBlockSize - Height()) / 2; }
961 
IsIndexInBorder(int i)962   bool IsIndexInBorder(int i) {
963     return (i < BorderTop() * kOuterBlockSize ||
964             i >= (BorderTop() + Height()) * kOuterBlockSize ||
965             i % kOuterBlockSize < BorderLeft() ||
966             i % kOuterBlockSize >= (BorderLeft() + Width()));
967   }
968 
SetUp()969   void SetUp() override {
970     // Force input_ to be unaligned, output to be 16 byte aligned.
971     input_ = reinterpret_cast<uint8_t *>(
972                  aom_memalign(kDataAlignment, kInputBufferSize + 1)) +
973              1;
974     output_ = reinterpret_cast<uint8_t *>(
975         aom_memalign(kDataAlignment, kOutputBufferSize));
976     output_ref_ = reinterpret_cast<uint8_t *>(
977         aom_memalign(kDataAlignment, kOutputBufferSize));
978 
979     ASSERT_NE(input_, nullptr);
980     ASSERT_NE(output_, nullptr);
981     ASSERT_NE(output_ref_, nullptr);
982 
983     test_func_ = GET_PARAM(2);
984     /* Set up guard blocks for an inner block centered in the outer block */
985     for (int i = 0; i < kOutputBufferSize; ++i) {
986       if (IsIndexInBorder(i)) {
987         output_[i] = 255;
988       } else {
989         output_[i] = 0;
990       }
991     }
992 
993     ::libaom_test::ACMRandom prng;
994     for (int i = 0; i < kInputBufferSize; ++i) {
995       if (i & 1) {
996         input_[i] = 255;
997       } else {
998         input_[i] = prng.Rand8Extremes();
999       }
1000     }
1001   }
1002 
TearDown()1003   void TearDown() override {
1004     aom_free(input_ - 1);
1005     input_ = nullptr;
1006     aom_free(output_);
1007     output_ = nullptr;
1008     aom_free(output_ref_);
1009     output_ref_ = nullptr;
1010   }
1011 
SetConstantInput(int value)1012   void SetConstantInput(int value) { memset(input_, value, kInputBufferSize); }
1013 
CopyOutputToRef()1014   void CopyOutputToRef() { memcpy(output_ref_, output_, kOutputBufferSize); }
1015 
CheckGuardBlocks()1016   void CheckGuardBlocks() {
1017     for (int i = 0; i < kOutputBufferSize; ++i) {
1018       if (IsIndexInBorder(i)) {
1019         EXPECT_EQ(255, output_[i]);
1020       }
1021     }
1022   }
1023 
input() const1024   uint8_t *input() const {
1025     const int offset = BorderTop() * kOuterBlockSize + BorderLeft();
1026     return input_ + offset;
1027   }
1028 
output() const1029   uint8_t *output() const {
1030     const int offset = BorderTop() * kOuterBlockSize + BorderLeft();
1031     return output_ + offset;
1032   }
1033 
output_ref() const1034   uint8_t *output_ref() const {
1035     const int offset = BorderTop() * kOuterBlockSize + BorderLeft();
1036     return output_ref_ + offset;
1037   }
1038 
lookup(uint8_t * list,int index) const1039   uint16_t lookup(uint8_t *list, int index) const { return list[index]; }
1040 
assign_val(uint8_t * list,int index,uint16_t val) const1041   void assign_val(uint8_t *list, int index, uint16_t val) const {
1042     list[index] = (uint8_t)val;
1043   }
1044 
1045   ConvolveScale2DFunc test_func_;
1046   uint8_t *input_;
1047   uint8_t *output_;
1048   uint8_t *output_ref_;
1049 };
1050 
TEST_P(ConvolveScale2DTest,DISABLED_Speed)1051 TEST_P(ConvolveScale2DTest, DISABLED_Speed) {
1052   const uint8_t *const in = input();
1053   uint8_t *const out = output();
1054   const InterpKernel *const filter =
1055       (const InterpKernel *)av1_get_interp_filter_kernel(EIGHTTAP_REGULAR,
1056                                                          USE_8_TAPS);
1057   const int kNumTests = 10000;
1058   const int width = Width();
1059   const int height = Height();
1060   const int frac = 8;
1061   const int step = 16;
1062   aom_usec_timer timer;
1063 
1064   aom_usec_timer_start(&timer);
1065   for (int n = 0; n < kNumTests; ++n) {
1066     test_func_(in, kInputStride, out, kOutputStride, filter, frac, step, frac,
1067                step, width, height);
1068   }
1069   aom_usec_timer_mark(&timer);
1070 
1071   const int elapsed_time = static_cast<int>(aom_usec_timer_elapsed(&timer));
1072   printf("convolve_scale_2d_%dx%d_%d: %d us\n", width, height, 8, elapsed_time);
1073 }
1074 
TEST_P(ConvolveScale2DTest,Correctness)1075 TEST_P(ConvolveScale2DTest, Correctness) {
1076   uint8_t *const in = input();
1077   uint8_t *const out = output();
1078   uint8_t ref[kOutputStride * kMaxDimension];
1079 
1080   ::libaom_test::ACMRandom prng;
1081   for (int y = 0; y < Height(); ++y) {
1082     for (int x = 0; x < Width(); ++x) {
1083       const uint16_t r = prng.Rand8Extremes();
1084       assign_val(in, y * kInputStride + x, r);
1085     }
1086   }
1087 
1088   for (int subpel_search = USE_2_TAPS; subpel_search <= USE_8_TAPS;
1089        ++subpel_search) {
1090     for (int filter_bank = 0; filter_bank < kNumFilterBanks; ++filter_bank) {
1091       const InterpFilter filter = static_cast<InterpFilter>(filter_bank);
1092       const InterpKernel *filters =
1093           (const InterpKernel *)av1_get_interp_filter_kernel(filter,
1094                                                              subpel_search);
1095       for (int frac = 0; frac < 16; ++frac) {
1096         for (int step = 1; step <= 32; ++step) {
1097           aom_scaled_2d_c(in, kInputStride, ref, kOutputStride, filters, frac,
1098                           step, frac, step, Width(), Height());
1099           API_REGISTER_STATE_CHECK(
1100               test_func_(in, kInputStride, out, kOutputStride, filters, frac,
1101                          step, frac, step, Width(), Height()));
1102 
1103           CheckGuardBlocks();
1104 
1105           for (int y = 0; y < Height(); ++y) {
1106             for (int x = 0; x < Width(); ++x) {
1107               ASSERT_EQ(lookup(ref, y * kOutputStride + x),
1108                         lookup(out, y * kOutputStride + x))
1109                   << "x == " << x << ", y == " << y << ", frac == " << frac
1110                   << ", step == " << step;
1111             }
1112           }
1113         }
1114       }
1115     }
1116   }
1117 }
1118 
1119 INSTANTIATE_TEST_SUITE_P(C, ConvolveScale2DTest,
1120                          ::testing::Values(ALL_SIZES_64(aom_scaled_2d_c)));
1121 
1122 #if HAVE_NEON
1123 INSTANTIATE_TEST_SUITE_P(NEON, ConvolveScale2DTest,
1124                          ::testing::Values(ALL_SIZES_64(aom_scaled_2d_neon)));
1125 #endif  // HAVE_NEON
1126 
1127 #if HAVE_NEON_DOTPROD
1128 INSTANTIATE_TEST_SUITE_P(
1129     NEON_DOTPROD, ConvolveScale2DTest,
1130     ::testing::Values(ALL_SIZES_64(aom_scaled_2d_neon_dotprod)));
1131 #endif  // HAVE_NEON_DOTPROD
1132 
1133 #if HAVE_NEON_I8MM
1134 INSTANTIATE_TEST_SUITE_P(
1135     NEON_I8MM, ConvolveScale2DTest,
1136     ::testing::Values(ALL_SIZES_64(aom_scaled_2d_neon_i8mm)));
1137 #endif  // HAVE_NEON_I8MM
1138 
1139 #if HAVE_SSSE3
1140 INSTANTIATE_TEST_SUITE_P(SSSE3, ConvolveScale2DTest,
1141                          ::testing::Values(ALL_SIZES_64(aom_scaled_2d_ssse3)));
1142 #endif  // HAVE_SSSE3
1143 
1144 }  // namespace
1145