Blame - src/core/GLES_COMPUTE/cs_shaders/normalization_layer.cs - ml/ComputeLibrary

Anthony Barbier

7068f99

2017-10-26 15:23:08 +0100

[diff] [blame^]

/*

*

* SPDX-License-Identifier: MIT

5

*

6

* Permission is hereby granted, free of charge, to any person obtaining a copy

7

* of this software and associated documentation files (the "Software"), to

8

* deal in the Software without restriction, including without limitation the

9

* rights to use, copy, modify, merge, publish, distribute, sublicense, and/or

10

* sell copies of the Software, and to permit persons to whom the Software is

11

* furnished to do so, subject to the following conditions:

12

*

13

* The above copyright notice and this permission notice shall be included in all

14

* copies or substantial portions of the Software.

15

*

16

* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR

17

* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,

18

* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE

19

* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER

20

* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,

21

* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE

* SOFTWARE.

*/

layout(local_size_x = LOCAL_SIZE_X, local_size_y = LOCAL_SIZE_Y, local_size_z = LOCAL_SIZE_Z) in;

#include "helpers.h"

layout(std140) uniform shader_params

30

{

31

TENSOR3D_PARAM_DECLARATION(src1);

32

TENSOR3D_PARAM_DECLARATION(src2);

33

TENSOR3D_PARAM_DECLARATION(dst);

34

};

35

36

BUFFER_DECLARATION(src1, 1, float, readonly);

37

BUFFER_DECLARATION(src2, 2, float, readonly);

38

BUFFER_DECLARATION(dst, 3, float, writeonly);

39

40

#ifdef CROSS_MAP

41

/** Apply cross map normalization.

42

*

43

* @note Alpha parameter / norm_size should be given as a preprocessor argument using "#define COEFF x"

44

* @note BETA parameter in the normalization equation should be given as a preprocessor argument using "#define BETA x"

45

* @note KAPPA parameter in the normalization equation should be given as a preprocessor argument using "#define KAPPA x"

46

* @note Number of elements on the right or left side to normalize across should be given as a preprocessor argument using "#define RADIUS x"

47

*

48

* @param[in] src1_ptr Pointer to the first source tensor. Supported data types: F32

49

* @param[in] src1_stride_x Stride of the first source tensor in X dimension (in bytes)

50

* @param[in] src1_step_x src1_stride_x * number of elements along X processed per workitem(in bytes)

51

* @param[in] src1_stride_y Stride of the first source tensor in Y dimension (in bytes)

52

* @param[in] src1_step_y src1_stride_y * number of elements along Y processed per workitem(in bytes)

53

* @param[in] src1_stride_z Stride of the first source tensor in Z dimension (in bytes)

54

* @param[in] src1_step_z src1_stride_z * number of elements along Z processed per workitem(in bytes)

55

* @param[in] src1_offset_first_element_in_bytes The offset of the first element in the first source tensor

56

* @param[in] src2_ptr Pointer to the second source tensor. Supported data types: Same as @p src1_ptr

57

* @param[in] src2_stride_x Stride of the second source tensor in X dimension (in bytes)

58

* @param[in] src2_step_x src2_stride_x * number of elements along X processed per workitem(in bytes)

59

* @param[in] src2_stride_y Stride of the second source tensor in Y dimension (in bytes)

60

* @param[in] src2_step_y src2_stride_y * number of elements along Y processed per workitem(in bytes)

61

* @param[in] src2_stride_z Stride of the second source tensor in Z dimension (in bytes)

62

* @param[in] src2_step_z src2_stride_z * number of elements along Z processed per workitem(in bytes)

63

* @param[in] src2_offset_first_element_in_bytes The offset of the second element in the second source tensor

64

* @param[out] dst_ptr Pointer to the destination tensor. Supported data types: Same as @p src1_ptr

65

* @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes)

66

* @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes)

67

* @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes)

68

* @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes)

69

* @param[in] dst_stride_z Stride of the destination tensor in Z dimension (in bytes)

70

* @param[in] dst_step_z dst_stride_z * number of elements along Z processed per workitem(in bytes)

71

* @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor

*/

void main(void)

{

Tensor3D src1 = CONVERT_TO_TENSOR3D_STRUCT(src1);

76

Tensor3D src2 = CONVERT_TO_TENSOR3D_STRUCT(src2);

77

Tensor3D dst = CONVERT_TO_TENSOR3D_STRUCT(dst);

float acc = 0.0;

int num_of_slices = int(gl_NumWorkGroups.z * gl_WorkGroupSize.z);

82

int current_slice = int(gl_GlobalInvocationID.z);

83

84

int left_slice = max(current_slice - int(RADIUS), int(0));

85

int right_slice = min(current_slice + int(RADIUS), int(num_of_slices - 1));

86

87

for(int i = left_slice; i <= right_slice; i++)

88

{

89

acc += src2_ptr[tensor3D_offset(src2, 0, 0, i - current_slice)];

90

}

91

92

float normalized = pow(float(KAPPA) + float(COEFF) * acc, float(BETA));

93

94

float normalized_pixel = (src1_ptr[src1.current_offset]) / normalized;

95

96

dst_ptr[dst.current_offset] = normalized_pixel;

97

}

98

99

#elif defined(IN_MAP_1D)

100

/** Apply in map normalization.

101

*

102