Lines Matching +full:- +full:oc
6 // This source code is licensed under the BSD-style license found in the
31 return a > b ? a - b : T(0); in doz()
44 this->padding_top_ = padding; in padding()
45 this->padding_right_ = padding; in padding()
46 this->padding_bottom_ = padding; in padding()
47 this->padding_left_ = padding; in padding()
52 this->padding_top_ = padding_height; in padding_height()
53 this->padding_bottom_ = padding_height; in padding_height()
58 return this->padding_top_ + this->padding_bottom_; in padding_height()
62 this->padding_right_ = padding_width; in padding_width()
63 this->padding_left_ = padding_width; in padding_width()
68 return this->padding_left_ + this->padding_right_; in padding_width()
72 this->padding_top_ = padding_top; in padding_top()
76 inline uint32_t padding_top() const { return this->padding_top_; } in padding_top()
79 this->padding_right_ = padding_right; in padding_right()
83 inline uint32_t padding_right() const { return this->padding_right_; } in padding_right()
86 this->padding_bottom_ = padding_bottom; in padding_bottom()
90 inline uint32_t padding_bottom() const { return this->padding_bottom_; } in padding_bottom()
93 this->padding_left_ = padding_left; in padding_left()
97 inline uint32_t padding_left() const { return this->padding_left_; } in padding_left()
100 this->adjustment_height_ = adjustment_height; in adjustment_height()
105 return this->adjustment_height_; in adjustment_height()
109 this->adjustment_width_ = adjustment_width; in adjustment_width()
114 return this->adjustment_width_; in adjustment_width()
120 this->input_height_ = input_height; in input_size()
121 this->input_width_ = input_width; in input_size()
127 this->input_height_ = input_height; in input_height()
132 return this->input_height_; in input_height()
137 this->input_width_ = input_width; in input_width()
142 return this->input_width_; in input_width()
147 this->groups_ = groups; in groups()
152 return this->groups_; in groups()
157 this->group_input_channels_ = group_input_channels; in group_input_channels()
162 return this->group_input_channels_; in group_input_channels()
167 this->group_output_channels_ = group_output_channels; in group_output_channels()
172 return this->group_output_channels_; in group_output_channels()
177 this->batch_size_ = batch_size; in batch_size()
182 return this->batch_size_; in batch_size()
187 this->kernel_height_ = kernel_size; in kernel_size()
188 this->kernel_width_ = kernel_size; in kernel_size()
195 this->kernel_height_ = kernel_height; in kernel_size()
196 this->kernel_width_ = kernel_width; in kernel_size()
202 this->kernel_height_ = kernel_height; in kernel_height()
207 return this->kernel_height_; in kernel_height()
212 this->kernel_width_ = kernel_width; in kernel_width()
217 return this->kernel_width_; in kernel_width()
222 this->dilation_height_ = dilation; in dilation()
223 this->dilation_width_ = dilation; in dilation()
230 this->dilation_height_ = dilation_height; in dilation()
231 this->dilation_width_ = dilation_width; in dilation()
237 this->dilation_height_ = dilation_height; in dilation_height()
242 return this->dilation_height_; in dilation_height()
247 this->dilation_width_ = dilation_width; in dilation_width()
252 return this->dilation_width_; in dilation_width()
257 this->stride_height_ = stride; in stride()
258 this->stride_width_ = stride; in stride()
265 this->stride_height_ = stride_height; in stride()
266 this->stride_width_ = stride_width; in stride()
272 this->stride_height_ = stride_height; in stride_height()
277 return this->stride_height_; in stride_height()
282 this->stride_width_ = stride_width; in stride_width()
287 return this->stride_width_; in stride_width()
292 this->input_pixel_stride_ = input_pixel_stride; in input_pixel_stride()
297 if (this->input_pixel_stride_ == 0) { in input_pixel_stride()
300 assert(this->input_pixel_stride_ >= group_input_channels() * groups()); in input_pixel_stride()
301 return this->input_pixel_stride_; in input_pixel_stride()
307 this->output_pixel_stride_ = output_pixel_stride; in output_pixel_stride()
312 if (this->output_pixel_stride_ == 0) { in output_pixel_stride()
315 assert(this->output_pixel_stride_ >= group_output_channels() * groups()); in output_pixel_stride()
316 return this->output_pixel_stride_; in output_pixel_stride()
321 return (kernel_height() - 1) * dilation_height() + 1; in dilated_kernel_height()
325 return (kernel_width() - 1) * dilation_width() + 1; in dilated_kernel_width()
329 …return stride_height() * (input_height() - 1) + adjustment_height() + dilated_kernel_height() - pa… in output_height()
333 …return stride_width() * (input_width() - 1) + adjustment_width() + dilated_kernel_width() - paddin… in output_width()
339 this->next_input_height_ = next_input_height; in next_input_size()
340 this->next_input_width_ = next_input_width; in next_input_size()
346 this->next_input_height_ = next_input_height; in next_input_height()
351 if (this->next_input_height_ == 0) { in next_input_height()
354 return this->next_input_height_; in next_input_height()
360 this->next_input_width_ = next_input_width; in next_input_width()
365 if (this->next_input_width_ == 0) { in next_input_width()
368 return this->next_input_width_; in next_input_width()
373 …return stride_height() * (next_input_height() - 1) + adjustment_height() + dilated_kernel_height()… in next_output_height()
377 …return stride_width() * (next_input_width() - 1) + adjustment_width() + dilated_kernel_width() - p… in next_output_width()
382 this->next_batch_size_ = next_batch_size; in next_batch_size()
387 if (this->next_batch_size_ == 0) { in next_batch_size()
390 return this->next_batch_size_; in next_batch_size()
395 this->qmin_ = qmin; in qmin()
400 return this->qmin_; in qmin()
404 this->qmax_ = qmax; in qmax()
409 return this->qmax_; in qmax()
413 this->has_bias_ = has_bias; in has_bias()
418 return this->has_bias_; in has_bias()
422 this->weights_type_ = weights_type; in weights_type()
427 return this->weights_type_; in weights_type()
431 this->use_weights_cache_ = use_weights_cache; in use_weights_cache()
436 return this->use_weights_cache_; in use_weights_cache()
440 this->iterations_ = iterations; in iterations()
445 return this->iterations_; in iterations()
453 std::uniform_int_distribution<int32_t> i32dist(-10000, 10000); in TestQS8()
457 -std::numeric_limits<int8_t>::max(), std::numeric_limits<int8_t>::max()); in TestQS8()
460 …(batch_size() * input_height() * input_width() - 1) * input_pixel_stride() + groups() * group_inpu… in TestQS8()
463 …std::vector<int8_t> output((batch_size() * output_height() * output_width() - 1) * output_pixel_st… in TestQS8()
481 for (size_t oc = 0; oc < group_output_channels(); oc++) { in TestQS8() local
482 …(i * output_height() + oy) * output_width() + ox) * groups() + g) * group_output_channels() + oc] = in TestQS8()
483 bias[g * group_output_channels() + oc]; in TestQS8()
496 const size_t y = oy + padding_top() - ky * dilation_height(); in TestQS8()
500 const size_t x = ox + padding_left() - kx * dilation_width(); in TestQS8()
504 for (size_t oc = 0; oc < group_output_channels(); oc++) { in TestQS8() local
506 …i * output_height() + oy) * output_width() + ox) * groups() + g) * group_output_channels() + oc] += in TestQS8()
507 …ut_width() + ix) * input_pixel_stride() + g * group_input_channels() + ic]) - int32_t(input_zero_p… in TestQS8()
508 …int32_t(kernel[(((g * group_output_channels() + oc) * kernel_height() + ky) * kernel_width() + kx)… in TestQS8()
524 const double output_scale = double(uint32_t(accumulated_max - accumulated_min)) / 255.0; in TestQS8()
526 lrint(-0.5 - 0.5 * double(accumulated_min + accumulated_max) / output_scale), in TestQS8()
531 [this, output_scale, output_zero_point](int32_t x) -> double { in TestQS8()
532 …n<double>(double(x) / output_scale, double(qmax() - 0x80) - output_zero_point), double(qmin() - 0x… in TestQS8()
559 output_scale, int8_t(qmin() - 0x80), int8_t(qmax() - 0x80), in TestQS8()
596 output_scale, int8_t(qmin() - 0x80), int8_t(qmax() - 0x80), in TestQS8()
630 …t_width() + x) * output_pixel_stride() + g * group_output_channels() + c]), int32_t(qmax() - 0x80)) in VerifyQS8()
632 …t_width() + x) * output_pixel_stride() + g * group_output_channels() + c]), int32_t(qmin() - 0x80)) in VerifyQS8()
636 …ut_width() + x) * output_pixel_stride() + g * group_output_channels() + c]) - double(output_zero_p… in VerifyQS8()
647 ASSERT_EQ(weights_cache->cache.hits, 1); in VerifyWeightsCache()
649 ASSERT_EQ(old_size, weights_cache->cache.weights.size); in VerifyWeightsCache()
657 std::uniform_int_distribution<int32_t> i32dist(-10000, 10000); in TestQU8()
662 …(batch_size() * input_height() * input_width() - 1) * input_pixel_stride() + groups() * group_inpu… in TestQU8()
665 …std::vector<uint8_t> output((batch_size() * output_height() * output_width() - 1) * output_pixel_s… in TestQU8()
684 for (size_t oc = 0; oc < group_output_channels(); oc++) { in TestQU8() local
685 …(i * output_height() + oy) * output_width() + ox) * groups() + g) * group_output_channels() + oc] = in TestQU8()
686 bias[g * group_output_channels() + oc]; in TestQU8()
699 const size_t y = oy + padding_top() - ky * dilation_height(); in TestQU8()
703 const size_t x = ox + padding_left() - kx * dilation_width(); in TestQU8()
707 for (size_t oc = 0; oc < group_output_channels(); oc++) { in TestQU8() local
709 …i * output_height() + oy) * output_width() + ox) * groups() + g) * group_output_channels() + oc] += in TestQU8()
710 …ut_width() + ix) * input_pixel_stride() + g * group_input_channels() + ic]) - int32_t(input_zero_p… in TestQU8()
711 …[(((g * group_output_channels() + oc) * kernel_height() + ky) * kernel_width() + kx) * group_input… in TestQU8()
727 const double output_scale = double(uint32_t(accumulated_max - accumulated_min)) / 255.0; in TestQU8()
729 lrint(127.5 - 0.5 * double(accumulated_min + accumulated_max) / output_scale), in TestQU8()
734 [this, output_scale, output_zero_point](int32_t x) -> double { in TestQU8()
735 …td::min<double>(double(x) / output_scale, double(qmax()) - output_zero_point), double(qmin()) - ou… in TestQU8()
841 …ut_width() + x) * output_pixel_stride() + g * group_output_channels() + c]) - double(output_zero_p… in VerifyQU8()
866 …(batch_size() * input_height() * input_width() - 1) * input_pixel_stride() + groups() * group_inpu… in TestF16()
871 …std::vector<uint16_t> output((batch_size() * output_height() * output_width() - 1) * output_pixel_… in TestF16()
888 for (size_t oc = 0; oc < group_output_channels(); oc++) { in TestF16() local
889 …(i * output_height() + oy) * output_width() + ox) * groups() + g) * group_output_channels() + oc] = in TestF16()
890 bias_as_float[g * group_output_channels() + oc]; in TestF16()
903 const size_t y = oy + padding_top() - ky * dilation_height(); in TestF16()
907 const size_t x = ox + padding_left() - kx * dilation_width(); in TestF16()
911 for (size_t oc = 0; oc < group_output_channels(); oc++) { in TestF16() local
913 …i * output_height() + oy) * output_width() + ox) * groups() + g) * group_output_channels() + oc] += in TestF16()
915 …kernel_as_float[(((g * group_output_channels() + oc) * kernel_height() + ky) * kernel_width() + kx… in TestF16()
930 const float accumulated_range = accumulated_max - accumulated_min; in TestF16()
932 float output_max = accumulated_max - accumulated_range / 255.0f * float(255 - qmax()); in TestF16()
936 output_min = -std::numeric_limits<float>::infinity(); in TestF16()
940 output_min = -std::numeric_limits<float>::infinity(); in TestF16()
1064 …1.0e-2f * std::abs(output_ref[(((i * output_height() + y) * output_width() + x) * groups() + g) * … in VerifyF16()
1081 …(batch_size() * input_height() * input_width() - 1) * input_pixel_stride() + groups() * group_inpu… in TestF32()
1084 …std::vector<float> output((batch_size() * output_height() * output_width() - 1) * output_pixel_str… in TestF32()
1099 for (size_t oc = 0; oc < group_output_channels(); oc++) { in TestF32() local
1100 …(i * output_height() + oy) * output_width() + ox) * groups() + g) * group_output_channels() + oc] = in TestF32()
1101 bias[g * group_output_channels() + oc]; in TestF32()
1114 const size_t y = oy + padding_top() - ky * dilation_height(); in TestF32()
1118 const size_t x = ox + padding_left() - kx * dilation_width(); in TestF32()
1122 for (size_t oc = 0; oc < group_output_channels(); oc++) { in TestF32() local
1124 …i * output_height() + oy) * output_width() + ox) * groups() + g) * group_output_channels() + oc] += in TestF32()
1126 …kernel[(((g * group_output_channels() + oc) * kernel_height() + ky) * kernel_width() + kx) * group… in TestF32()
1142 const float output_min = qmin() == 0 ? -std::numeric_limits<float>::infinity() : in TestF32()
1143 accumulated_min + (accumulated_max - accumulated_min) / 255.0f * float(qmin()); in TestF32()
1145 accumulated_max - (accumulated_max - accumulated_min) / 255.0f * float(255 - qmax()); in TestF32()
1268 …(batch_size() * input_height() * input_width() - 1) * input_pixel_stride() + groups() * group_inpu… in StressWeightsCacheTestF32()
1271 …std::vector<float> output((batch_size() * output_height() * output_width() - 1) * output_pixel_str… in StressWeightsCacheTestF32()
1285 for (size_t oc = 0; oc < group_output_channels(); oc++) { in StressWeightsCacheTestF32() local
1286 …(i * output_height() + oy) * output_width() + ox) * groups() + g) * group_output_channels() + oc] = in StressWeightsCacheTestF32()
1287 bias[g * group_output_channels() + oc]; in StressWeightsCacheTestF32()
1300 const size_t y = oy + padding_top() - ky * dilation_height(); in StressWeightsCacheTestF32()
1304 const size_t x = ox + padding_left() - kx * dilation_width(); in StressWeightsCacheTestF32()
1308 for (size_t oc = 0; oc < group_output_channels(); oc++) { in StressWeightsCacheTestF32() local
1310 …i * output_height() + oy) * output_width() + ox) * groups() + g) * group_output_channels() + oc] += in StressWeightsCacheTestF32()
1312 …kernel[(((g * group_output_channels() + oc) * kernel_height() + ky) * kernel_width() + kx) * group… in StressWeightsCacheTestF32()
1328 const float output_min = qmin() == 0 ? -std::numeric_limits<float>::infinity() : in StressWeightsCacheTestF32()
1329 accumulated_min + (accumulated_max - accumulated_min) / 255.0f * float(qmin()); in StressWeightsCacheTestF32()
1331 accumulated_max - (accumulated_max - accumulated_min) / 255.0f * float(255 - qmax()); in StressWeightsCacheTestF32()
1412 …1.0e-4 * std::abs(output_ref[(((i * output_height() + y) * output_width() + x) * groups() + g) * g… in VerifyF32()
1426 std::uniform_int_distribution<int32_t> i32dist(-10000, 10000); in TestSetupQS8()
1430 -std::numeric_limits<int8_t>::max(), std::numeric_limits<int8_t>::max()); in TestSetupQS8()
1433 …(batch_size() * input_height() * input_width() - 1) * input_pixel_stride() + groups() * group_inpu… in TestSetupQS8()
1434 …(next_batch_size() * next_input_height() * next_input_width() - 1) * input_pixel_stride() + groups… in TestSetupQS8()
1438 …(batch_size() * output_height() * output_width() - 1) * output_pixel_stride() + groups() * group_o… in TestSetupQS8()
1439 …(next_batch_size() * next_output_height() * next_output_width() - 1) * output_pixel_stride() + gro… in TestSetupQS8()
1459 for (size_t oc = 0; oc < group_output_channels(); oc++) { in TestSetupQS8() local
1460 …(i * output_height() + oy) * output_width() + ox) * groups() + g) * group_output_channels() + oc] = in TestSetupQS8()
1461 bias[g * group_output_channels() + oc]; in TestSetupQS8()
1474 const size_t y = oy + padding_top() - ky * dilation_height(); in TestSetupQS8()
1478 const size_t x = ox + padding_left() - kx * dilation_width(); in TestSetupQS8()
1482 for (size_t oc = 0; oc < group_output_channels(); oc++) { in TestSetupQS8() local
1484 …i * output_height() + oy) * output_width() + ox) * groups() + g) * group_output_channels() + oc] += in TestSetupQS8()
1485 …ut_width() + ix) * input_pixel_stride() + g * group_input_channels() + ic]) - int32_t(input_zero_p… in TestSetupQS8()
1486 …int32_t(kernel[(((g * group_output_channels() + oc) * kernel_height() + ky) * kernel_width() + kx)… in TestSetupQS8()
1502 const double output_scale = double(uint32_t(accumulated_max - accumulated_min)) / 255.0; in TestSetupQS8()
1504 lrint(-0.5 - 0.5 * double(accumulated_min + accumulated_max) / output_scale), in TestSetupQS8()
1509 [this, output_scale, output_zero_point](int32_t x) -> double { in TestSetupQS8()
1510 …n<double>(double(x) / output_scale, double(qmax() - 0x80) - output_zero_point), double(qmin() - 0x… in TestSetupQS8()
1528 output_zero_point, output_scale, int8_t(qmin() - 0x80), int8_t(qmax() - 0x80), in TestSetupQS8()
1551 …t_width() + x) * output_pixel_stride() + g * group_output_channels() + c]), int32_t(qmax() - 0x80)) in TestSetupQS8()
1553 …t_width() + x) * output_pixel_stride() + g * group_output_channels() + c]), int32_t(qmin() - 0x80)) in TestSetupQS8()
1557 …ut_width() + x) * output_pixel_stride() + g * group_output_channels() + c]) - double(output_zero_p… in TestSetupQS8()
1566 // Re-generate data for the second run. in TestSetupQS8()
1576 for (size_t oc = 0; oc < group_output_channels(); oc++) { in TestSetupQS8() local
1577 …output_height() + oy) * next_output_width() + ox) * groups() + g) * group_output_channels() + oc] = in TestSetupQS8()
1578 bias[g * group_output_channels() + oc]; in TestSetupQS8()
1591 const size_t y = oy + padding_top() - ky * dilation_height(); in TestSetupQS8()
1595 const size_t x = ox + padding_left() - kx * dilation_width(); in TestSetupQS8()
1599 for (size_t oc = 0; oc < group_output_channels(); oc++) { in TestSetupQS8() local
1601 …utput_height() + oy) * next_output_width() + ox) * groups() + g) * group_output_channels() + oc] += in TestSetupQS8()
1602 …ut_width() + ix) * input_pixel_stride() + g * group_input_channels() + ic]) - int32_t(input_zero_p… in TestSetupQS8()
1603 …int32_t(kernel[(((g * group_output_channels() + oc) * kernel_height() + ky) * kernel_width() + kx)… in TestSetupQS8()
1615 [this, output_scale, output_zero_point](int32_t x) -> double { in TestSetupQS8()
1616 …n<double>(double(x) / output_scale, double(qmax() - 0x80) - output_zero_point), double(qmin() - 0x… in TestSetupQS8()
1637 …t_width() + x) * output_pixel_stride() + g * group_output_channels() + c]), int32_t(qmax() - 0x80)) in TestSetupQS8()
1639 …t_width() + x) * output_pixel_stride() + g * group_output_channels() + c]), int32_t(qmin() - 0x80)) in TestSetupQS8()
1643 …ut_width() + x) * output_pixel_stride() + g * group_output_channels() + c]) - double(output_zero_p… in TestSetupQS8()
1659 std::uniform_int_distribution<int32_t> i32dist(-10000, 10000); in TestSetupQU8()
1664 …(batch_size() * input_height() * input_width() - 1) * input_pixel_stride() + groups() * group_inpu… in TestSetupQU8()
1665 …(next_batch_size() * next_input_height() * next_input_width() - 1) * input_pixel_stride() + groups… in TestSetupQU8()
1669 …(batch_size() * output_height() * output_width() - 1) * output_pixel_stride() + groups() * group_o… in TestSetupQU8()
1670 …(next_batch_size() * next_output_height() * next_output_width() - 1) * output_pixel_stride() + gro… in TestSetupQU8()
1691 for (size_t oc = 0; oc < group_output_channels(); oc++) { in TestSetupQU8() local
1692 …(i * output_height() + oy) * output_width() + ox) * groups() + g) * group_output_channels() + oc] = in TestSetupQU8()
1693 bias[g * group_output_channels() + oc]; in TestSetupQU8()
1706 const size_t y = oy + padding_top() - ky * dilation_height(); in TestSetupQU8()
1710 const size_t x = ox + padding_left() - kx * dilation_width(); in TestSetupQU8()
1714 for (size_t oc = 0; oc < group_output_channels(); oc++) { in TestSetupQU8() local
1716 …i * output_height() + oy) * output_width() + ox) * groups() + g) * group_output_channels() + oc] += in TestSetupQU8()
1717 …ut_width() + ix) * input_pixel_stride() + g * group_input_channels() + ic]) - int32_t(input_zero_p… in TestSetupQU8()
1718 …[(((g * group_output_channels() + oc) * kernel_height() + ky) * kernel_width() + kx) * group_input… in TestSetupQU8()
1734 const double output_scale = double(uint32_t(accumulated_max - accumulated_min)) / 255.0; in TestSetupQU8()
1736 lrint(127.5 - 0.5 * double(accumulated_min + accumulated_max) / output_scale), in TestSetupQU8()
1741 [this, output_scale, output_zero_point](int32_t x) -> double { in TestSetupQU8()
1742 …td::min<double>(double(x) / output_scale, double(qmax()) - output_zero_point), double(qmin()) - ou… in TestSetupQU8()
1789 …ut_width() + x) * output_pixel_stride() + g * group_output_channels() + c]) - double(output_zero_p… in TestSetupQU8()
1798 // Re-generate data for the second run. in TestSetupQU8()
1808 for (size_t oc = 0; oc < group_output_channels(); oc++) { in TestSetupQU8() local
1809 …output_height() + oy) * next_output_width() + ox) * groups() + g) * group_output_channels() + oc] = in TestSetupQU8()
1810 bias[g * group_output_channels() + oc]; in TestSetupQU8()
1823 const size_t y = oy + padding_top() - ky * dilation_height(); in TestSetupQU8()
1827 const size_t x = ox + padding_left() - kx * dilation_width(); in TestSetupQU8()
1831 for (size_t oc = 0; oc < group_output_channels(); oc++) { in TestSetupQU8() local
1833 …utput_height() + oy) * next_output_width() + ox) * groups() + g) * group_output_channels() + oc] += in TestSetupQU8()
1834 …ut_width() + ix) * input_pixel_stride() + g * group_input_channels() + ic]) - int32_t(input_zero_p… in TestSetupQU8()
1835 …[(((g * group_output_channels() + oc) * kernel_height() + ky) * kernel_width() + kx) * group_input… in TestSetupQU8()
1847 [this, output_scale, output_zero_point](int32_t x) -> double { in TestSetupQU8()
1848 …td::min<double>(double(x) / output_scale, double(qmax()) - output_zero_point), double(qmin()) - ou… in TestSetupQU8()
1875 …ut_width() + x) * output_pixel_stride() + g * group_output_channels() + c]) - double(output_zero_p… in TestSetupQU8()
1894 …(batch_size() * input_height() * input_width() - 1) * input_pixel_stride() + groups() * group_inpu… in TestSetupF16()
1895 …(next_batch_size() * next_input_height() * next_input_width() - 1) * input_pixel_stride() + groups… in TestSetupF16()
1899 …(batch_size() * output_height() * output_width() - 1) * output_pixel_stride() + groups() * group_o… in TestSetupF16()
1900 …(next_batch_size() * next_output_height() * next_output_width() - 1) * output_pixel_stride() + gro… in TestSetupF16()
1916 for (size_t oc = 0; oc < group_output_channels(); oc++) { in TestSetupF16() local
1917 …(i * output_height() + oy) * output_width() + ox) * groups() + g) * group_output_channels() + oc] = in TestSetupF16()
1918 fp16_ieee_to_fp32_value(bias[g * group_output_channels() + oc]); in TestSetupF16()
1931 const size_t y = oy + padding_top() - ky * dilation_height(); in TestSetupF16()
1935 const size_t x = ox + padding_left() - kx * dilation_width(); in TestSetupF16()
1939 for (size_t oc = 0; oc < group_output_channels(); oc++) { in TestSetupF16() local
1941 …i * output_height() + oy) * output_width() + ox) * groups() + g) * group_output_channels() + oc] += in TestSetupF16()
1943 …fp16_ieee_to_fp32_value(kernel[(((g * group_output_channels() + oc) * kernel_height() + ky) * kern… in TestSetupF16()
1958 const float accumulated_range = accumulated_max - accumulated_min; in TestSetupF16()
1960 float output_max = accumulated_max - accumulated_range / 255.0f * float(255 - qmax()); in TestSetupF16()
1964 output_min = -std::numeric_limits<float>::infinity(); in TestSetupF16()
1968 output_min = -std::numeric_limits<float>::infinity(); in TestSetupF16()
2026 …1.0e-2f * std::abs(output_ref[(((i * output_height() + y) * output_width() + x) * groups() + g) * … in TestSetupF16()
2034 // Re-generate data for the second run. in TestSetupF16()
2044 for (size_t oc = 0; oc < group_output_channels(); oc++) { in TestSetupF16() local
2045 …output_height() + oy) * next_output_width() + ox) * groups() + g) * group_output_channels() + oc] = in TestSetupF16()
2046 fp16_ieee_to_fp32_value(bias[g * group_output_channels() + oc]); in TestSetupF16()
2059 const size_t y = oy + padding_top() - ky * dilation_height(); in TestSetupF16()
2063 const size_t x = ox + padding_left() - kx * dilation_width(); in TestSetupF16()
2067 for (size_t oc = 0; oc < group_output_channels(); oc++) { in TestSetupF16() local
2069 …utput_height() + oy) * next_output_width() + ox) * groups() + g) * group_output_channels() + oc] += in TestSetupF16()
2071 …fp16_ieee_to_fp32_value(kernel[(((g * group_output_channels() + oc) * kernel_height() + ky) * kern… in TestSetupF16()
2111 …1.0e-2f * std::abs(next_output_ref[(((i * next_output_height() + y) * next_output_width() + x) * g… in TestSetupF16()
2129 …(batch_size() * input_height() * input_width() - 1) * input_pixel_stride() + groups() * group_inpu… in TestSetupF32()
2130 …(next_batch_size() * next_input_height() * next_input_width() - 1) * input_pixel_stride() + groups… in TestSetupF32()
2134 …(batch_size() * output_height() * output_width() - 1) * output_pixel_stride() + groups() * group_o… in TestSetupF32()
2135 …(next_batch_size() * next_output_height() * next_output_width() - 1) * output_pixel_stride() + gro… in TestSetupF32()
2151 for (size_t oc = 0; oc < group_output_channels(); oc++) { in TestSetupF32() local
2152 …(i * output_height() + oy) * output_width() + ox) * groups() + g) * group_output_channels() + oc] = in TestSetupF32()
2153 bias[g * group_output_channels() + oc]; in TestSetupF32()
2166 const size_t y = oy + padding_top() - ky * dilation_height(); in TestSetupF32()
2170 const size_t x = ox + padding_left() - kx * dilation_width(); in TestSetupF32()
2174 for (size_t oc = 0; oc < group_output_channels(); oc++) { in TestSetupF32() local
2176 …i * output_height() + oy) * output_width() + ox) * groups() + g) * group_output_channels() + oc] += in TestSetupF32()
2178 …kernel[(((g * group_output_channels() + oc) * kernel_height() + ky) * kernel_width() + kx) * group… in TestSetupF32()
2194 …const float output_min = accumulated_min + (accumulated_max - accumulated_min) / 255.0f * float(qm… in TestSetupF32()
2195 …const float output_max = accumulated_max - (accumulated_max - accumulated_min) / 255.0f * float(25… in TestSetupF32()
2245 …1.0e-4 * std::abs(output_ref[(((i * output_height() + y) * output_width() + x) * groups() + g) * g… in TestSetupF32()
2253 // Re-generate data for the second run. in TestSetupF32()
2263 for (size_t oc = 0; oc < group_output_channels(); oc++) { in TestSetupF32() local
2264 …output_height() + oy) * next_output_width() + ox) * groups() + g) * group_output_channels() + oc] = in TestSetupF32()
2265 bias[g * group_output_channels() + oc]; in TestSetupF32()
2278 const size_t y = oy + padding_top() - ky * dilation_height(); in TestSetupF32()
2282 const size_t x = ox + padding_left() - kx * dilation_width(); in TestSetupF32()
2286 for (size_t oc = 0; oc < group_output_channels(); oc++) { in TestSetupF32() local
2288 …utput_height() + oy) * next_output_width() + ox) * groups() + g) * group_output_channels() + oc] += in TestSetupF32()
2290 …kernel[(((g * group_output_channels() + oc) * kernel_height() + ky) * kernel_width() + kx) * group… in TestSetupF32()
2330 …1.0e-4 * std::abs(next_output_ref[(((i * next_output_height() + y) * next_output_width() + x) * gr… in TestSetupF32()