请问ACL中有没求两幅图平方和开根号的现成的库？

terry007

January 9, 2018

Less than one minute read time.

请问ACL中有没求两幅图平方和开根号的现成的库？最好是GPU的

Sheri Zhang 张晓霞 over 6 years ago in reply to terry007

Also, worth checking pooling_layer_2().

226 __kernel void pooling_layer_2(

227 TENSOR3D_DECLARATION(input),

228 TENSOR3D_DECLARATION(output))

229 {

230 // Get pixels pointer

231 Tensor3D input = CONVERT_TO_TENSOR3D_STRUCT(input);

232 Tensor3D output = CONVERT_TO_TENSOR3D_STRUCT(output);

233

234 // Load data

235 VEC_DATA_TYPE(DATA_TYPE, 2)

236 data0 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 0, 0));

237 VEC_DATA_TYPE(DATA_TYPE, 2)

238 data1 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 1, 0));

239

240 #if defined(POOL_L2)

241 // Raise to power of 2 for L2 Pooling

242 data0 = POW2_OP(data0, 2);

243 data1 = POW2_OP(data1, 2);

244 #endif /* defined(POOL_L2) */

245

246 // Perform calculations

247 data0 = POOL_OP(data0, data1);

248 DATA_TYPE res = POOL_OP(data0.s0, data0.s1);

249

250 #if defined(POOL_AVG) || defined(POOL_L2)

251 // Divide by pool region in case of average or l2 pooling

252 res = DIV_OP(res, calculate_avg_scale(2, MAX_WIDTH, MAX_HEIGHT, PAD_X, PAD_Y, STRIDE_X, STRIDE_Y));

253 #endif /* defined(POOL_AVG) || defined(POOL_L2) */

254

255 #if defined(POOL_L2)

256 // Take square root of the result in L2 pooling

257 res = SQRT_OP(res);

258 #endif /* defined(POOL_L2) */

259

260 // Store result

261 *(__global DATA_TYPE *)output.ptr = res;

262 }

263
- Cancel
- Up 0 Down
- Reply
- More
- Cancel
Sheri Zhang 张晓霞 over 6 years ago in reply to terry007

Some additional information, there's one kernel named accumulate_squared(), you can modified it a bit and use it.

112 __kernel void accumulate_squared(

113 IMAGE_DECLARATION(input),

114 IMAGE_DECLARATION(accu),

115 const uint shift)

116 {

117 // Get pixels pointer

118 Image input = CONVERT_TO_IMAGE_STRUCT(input);

119 Image accu = CONVERT_TO_IMAGE_STRUCT(accu);

120

121 // Load data

122 ushort16 in_data = convert_ushort16(vload16(0, input.ptr));

123 uint16 accu_data = convert_uint16(vload16(0, (__global short *)accu.ptr));

124

125 // Calculate squared accumulation

126 short16 res = convert_short16_sat(accu_data + convert_uint16((in_data * in_data) >> shift));

127

128 // Store result

129 vstore16(res, 0, (__global short *)accu.ptr);

130 }

You can refer to the convolution example for sobel filer: arm-software.github.io/.../cl__convolution_8cpp.xhtml
- Cancel
- Up 0 Down
- Reply
- More
- Cancel
terry007 over 6 years ago in reply to Sheri Zhang 张晓霞

Ok, thanks! Does ACL have the example for sobel filter?
- Cancel
- Up 0 Down
- Reply
- More
- Cancel
Sheri Zhang 张晓霞 over 6 years ago

Hi Terry,

Currently, we don't have such function in ACL.

BR,

Sheri
- Cancel
- Up 0 Down
- Reply
- More
- Cancel

中文社区博客

Arm A-Profile构架2022扩展

Zenon Xiu (修志龙）

原文：Arm A-Profile Architecture Developments 2022 - Architectures and Processors blog - Arm Community blogs - Arm Community 作者：Martin Weidmann翻译：修志龙（Zenon Xiu) 与arm构架授权和生态伙伴一起，arm持续演进其构架，开发新功能以满足现有和新市场的要求…
- October 17, 2022
深入理解 Arm A-profile的non-maskable interrupt -NMI

Zenon Xiu (修志龙）

原文： https://community.arm.com/arm-community-blogs/b/architectures-and-processors-blog/posts/a-profile-non-maskable-interrupts 翻译：修志龙 Zenon Xiu Arm A-profile构架一个长久以来的局限性是：缺乏对non-maskable interrupt (NMI…
- August 24, 2022
Arm A-Profile 构架2021扩展

Zenon Xiu (修志龙）

原文： https://community.arm.com/arm-community-blogs/b/architectures-and-processors-blog/posts/arm-a-profile-architecture-developments-2021 Martin Weidmann September 8, 2021 翻译注释：Zenon Xiu Arm与arm构架授权公司及生态伙伴一起…
- August 17, 2022

ARM中国大学计划博客

Arm新闻

中文mbed博客

中文社区博客

恩智浦汽车电子MCU讨论区博客

请问ACL中有没求两幅图平方和开根号的现成的库？

Arm A-Profile构架2022扩展

深入理解 Arm A-profile的non-maskable interrupt -NMI

Arm A-Profile 构架2021扩展