-
Notifications
You must be signed in to change notification settings - Fork 0
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Added fast gaussian very fast path for neon
- Loading branch information
Showing
7 changed files
with
381 additions
and
146 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Large diffs are not rendered by default.
Oops, something went wrong.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,168 @@ | ||
| // Copyright (C) 2024 Radzivon Bartoshyk | ||
| // | ||
| // This file belongs to sparkyuv project | ||
| // | ||
| // Licensed under the Apache License, Version 2.0 (the "License"); | ||
| // you may not use this file except in compliance with the License. | ||
| // You may obtain a copy of the License at | ||
| // | ||
| // http://www.apache.org/licenses/LICENSE-2.0 | ||
| // | ||
| // Unless required by applicable law or agreed to in writing, software | ||
| // distributed under the License is distributed on an "AS IS" BASIS, | ||
| // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | ||
| // See the License for the specific language governing permissions and | ||
| // limitations under the License. | ||
|
|
||
| #include "FastGaussianNeon.h" | ||
|
|
||
| #if __aarch64__ | ||
|
|
||
| #include <arm_neon.h> | ||
| #include <algorithm> | ||
|
|
||
| namespace sparkyuv { | ||
|
|
||
| #if defined(__GNUC__) || defined(__clang__) | ||
| #define FAST_GAUSSIAN_INLINE __attribute__((flatten)) inline | ||
| #else | ||
| #define FAST_GAUSSIAN_INLINE inline | ||
| #endif | ||
|
|
||
| FAST_GAUSSIAN_INLINE int32x4_t vsld_u8_to_i32(const uint8_t *src, bool useVld) { | ||
| uint8_t safeStore[8] = {0}; | ||
| const uint8_t *origin; | ||
| if (useVld) { | ||
| origin = src; | ||
| } else { | ||
| std::copy(src, src + 3, safeStore); | ||
| origin = safeStore; | ||
| } | ||
| auto fullPx = vreinterpret_s16_u16(vget_low_u16(vmovl_u8(vld1_u8(origin)))); | ||
| return vmovl_s16(fullPx); | ||
| } | ||
|
|
||
| void VerticalGaussianPassRGBANeon(uint8_t *data, | ||
| const uint32_t stride, | ||
| const uint32_t width, | ||
| const uint32_t height, | ||
| const int radius, | ||
| const uint32_t start, | ||
| const uint32_t end) { | ||
| const float weight = 1.f / (static_cast<float>(radius) * static_cast<float>(radius)); | ||
|
|
||
| constexpr int64_t bufLength = 1023; | ||
| int32_t buffer[1024][4] = {{0}}; | ||
|
|
||
| auto radius64 = static_cast<int64_t>(radius); | ||
|
|
||
| auto initialSum = (radius * radius) >> 1; | ||
|
|
||
| for (int64_t x = start; x < static_cast<int64_t> (width) && x < static_cast<int64_t>(end); ++x) { | ||
| int32x4_t dif = vdupq_n_s32(0); | ||
| int32x4_t sum = vdupq_n_s32(initialSum); | ||
| int64_t px = x * 4; | ||
| for (int64_t y = 0 - 2 * radius64; y < static_cast<int64_t>(height); ++y) { | ||
| if (y >= 0) { | ||
| auto src = reinterpret_cast<uint8_t *>(reinterpret_cast<uint8_t *>(data) + static_cast<uint32_t >(y) * stride); | ||
| int64_t arrIndex = (y - radius64) & bufLength; | ||
| int64_t dArrIndex = y & bufLength; | ||
|
|
||
| auto p16 = vqmovun_s32(vcvtq_s32_f32(vrndq_f32(vmulq_n_f32(vcvtq_f32_s32(sum), weight)))); | ||
| auto p8 = vqmovn_u16(vcombine_u16(p16, p16)); | ||
|
|
||
| src[px] = vget_lane_u8(p8, 0); | ||
| src[px + 1] = vget_lane_u8(p8, 1); | ||
| src[px + 2] = vget_lane_u8(p8, 2); | ||
|
|
||
| int32x4_t bufferValue1 = vld1q_s32(reinterpret_cast<const int *>(&buffer[arrIndex][0])); | ||
|
|
||
| int32x4_t bufferValue2 = vld1q_s32(reinterpret_cast<const int *>(&buffer[dArrIndex][0])); | ||
| bufferValue2 = vshlq_n_s32(bufferValue2, 1); | ||
|
|
||
| dif = vaddq_s32(dif, vsubq_s32(bufferValue1, bufferValue2)); | ||
| } else if (y + radius >= 0) { | ||
| int64_t dArrIndex = y & bufLength; | ||
| int32x4_t bufferValue = vld1q_s32(reinterpret_cast<const int *>(&buffer[dArrIndex][0])); | ||
| bufferValue = vshlq_n_s32(bufferValue, 1); | ||
| dif = vsubq_s32(dif, bufferValue); | ||
| } | ||
|
|
||
| auto srcNext = | ||
| reinterpret_cast<uint8_t *>(reinterpret_cast<uint8_t *>(data) | ||
| + std::clamp(static_cast<int32_t >(y + radius), 0, static_cast<int32_t >(height - 1)) * stride); | ||
| int32x4_t p = vsld_u8_to_i32(&srcNext[px], x + 2 < width); | ||
| dif = vaddq_s32(dif, p); | ||
| sum = vaddq_s32(sum, dif); | ||
|
|
||
| auto arrIndex = (y + radius64) & bufLength; | ||
|
|
||
| vst1q_s32(&buffer[arrIndex][0], p); | ||
| } | ||
| } | ||
| } | ||
|
|
||
| void HorizontalGaussianPassRGBANeon(uint8_t *data, | ||
| const uint32_t stride, | ||
| const uint32_t width, | ||
| const uint32_t height, | ||
| const int radius, | ||
| const uint32_t start, | ||
| const uint32_t end) { | ||
| const float weight = 1.f / (static_cast<float>(radius) * static_cast<float>(radius)); | ||
|
|
||
| constexpr int64_t bufLength = 1023; | ||
| int32_t buffer[1024][4] = {{0}}; | ||
|
|
||
| auto radius64 = static_cast<int64_t>(radius); | ||
|
|
||
| auto initialSum = (radius * radius) >> 1; | ||
|
|
||
| for (auto y = static_cast<int64_t>(start); y < static_cast<int64_t>(height) && y < static_cast<int64_t>(end); ++y) { | ||
| int32x4_t dif = vdupq_n_s32(0); | ||
| int32x4_t sum = vdupq_n_s32(initialSum); | ||
| for (int64_t x = 0 - 2 * radius64; x < static_cast<int64_t>(width); ++x) { | ||
| if (x >= 0) { | ||
| const int64_t px = x * 4; | ||
|
|
||
| auto src = reinterpret_cast<uint8_t *>(reinterpret_cast<uint8_t *>(data) + static_cast<uint32_t >(y) * stride); | ||
| int64_t arrIndex = (x - radius64) & bufLength; | ||
| int64_t dArrIndex = x & bufLength; | ||
|
|
||
| auto p16 = vqmovun_s32(vcvtq_s32_f32(vrndq_f32(vmulq_n_f32(vcvtq_f32_s32(sum), weight)))); | ||
| auto p8 = vqmovn_u16(vcombine_u16(p16, p16)); | ||
|
|
||
| src[px] = vget_lane_u8(p8, 0); | ||
| src[px + 1] = vget_lane_u8(p8, 1); | ||
| src[px + 2] = vget_lane_u8(p8, 2); | ||
|
|
||
| int32x4_t bufferValue1 = vld1q_s32(reinterpret_cast<const int *>(&buffer[arrIndex][0])); | ||
|
|
||
| int32x4_t bufferValue2 = vld1q_s32(reinterpret_cast<const int *>(&buffer[dArrIndex][0])); | ||
| bufferValue2 = vshlq_n_s32(bufferValue2, 1); | ||
|
|
||
| dif = vaddq_s32(dif, vsubq_s32(bufferValue1, bufferValue2)); | ||
| } else if (x + radius >= 0) { | ||
| int64_t dArrIndex = x & bufLength; | ||
| int32x4_t bufferValue = vld1q_s32(reinterpret_cast<const int *>(&buffer[dArrIndex][0])); | ||
| bufferValue = vshlq_n_s32(bufferValue, 1); | ||
| dif = vsubq_s32(dif, bufferValue); | ||
| } | ||
|
|
||
| auto srcNext = reinterpret_cast<uint8_t *>(reinterpret_cast<uint8_t *>(data) + y * stride); | ||
| int64_t px = | ||
| std::clamp(static_cast<int64_t >(x + radius), static_cast<int64_t >(0), static_cast<int64_t>(width - 1)); | ||
| int32x4_t p = vsld_u8_to_i32(&srcNext[px*4], px + 2 < width); | ||
| dif = vaddq_s32(dif, p); | ||
| sum = vaddq_s32(sum, dif); | ||
|
|
||
| auto arrIndex = (x + radius64) & bufLength; | ||
|
|
||
| vst1q_s32(&buffer[arrIndex][0], p); | ||
| } | ||
| } | ||
| } | ||
|
|
||
| } | ||
|
|
||
| #endif |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,44 @@ | ||
| // Copyright (C) 2024 Radzivon Bartoshyk | ||
| // | ||
| // This file belongs to sparkyuv project | ||
| // | ||
| // Licensed under the Apache License, Version 2.0 (the "License"); | ||
| // you may not use this file except in compliance with the License. | ||
| // You may obtain a copy of the License at | ||
| // | ||
| // http://www.apache.org/licenses/LICENSE-2.0 | ||
| // | ||
| // Unless required by applicable law or agreed to in writing, software | ||
| // distributed under the License is distributed on an "AS IS" BASIS, | ||
| // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | ||
| // See the License for the specific language governing permissions and | ||
| // limitations under the License. | ||
|
|
||
| #ifndef YUV_SRC_FASTGAUSSIANNEON_H_ | ||
| #define YUV_SRC_FASTGAUSSIANNEON_H_ | ||
|
|
||
| #if __aarch64__ | ||
|
|
||
| #include <cstdint> | ||
|
|
||
| namespace sparkyuv { | ||
| void VerticalGaussianPassRGBANeon(uint8_t *data, | ||
| uint32_t stride, | ||
| uint32_t width, | ||
| uint32_t height, | ||
| int radius, | ||
| uint32_t start, | ||
| uint32_t end); | ||
|
|
||
| void HorizontalGaussianPassRGBANeon(uint8_t *data, | ||
| uint32_t stride, | ||
| uint32_t width, | ||
| uint32_t height, | ||
| int radius, | ||
| uint32_t start, | ||
| uint32_t end); | ||
| } | ||
|
|
||
| #endif | ||
|
|
||
| #endif //YUV_SRC_FASTGAUSSIANNEON_H_ |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.