2 * Support for Intel Camera Imaging ISP subsystem.
3 * Copyright (c) 2015, Intel Corporation.
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 #include "type_support.h"
16 #include "math_support.h"
17 #include "sh_css_defs.h"
18 #include "ia_css_types.h"
19 #include "assert_support.h"
20 #include "ia_css_xnr3.host.h"
22 /* Maximum value for alpha on ISP interface */
23 #define XNR_MAX_ALPHA ((1 << (ISP_VEC_ELEMBITS - 1)) - 1)
25 /* Minimum value for sigma on host interface. Lower values translate to
28 #define XNR_MIN_SIGMA (IA_CSS_XNR3_SIGMA_SCALE / 100)
31 * division look-up table
34 #define XNR3_LOOK_UP_TABLE_POINTS 16
36 static const s16 x[XNR3_LOOK_UP_TABLE_POINTS] = {
37 1024, 1164, 1320, 1492, 1680, 1884, 2108, 2352,
38 2616, 2900, 3208, 3540, 3896, 4276, 4684, 5120
41 static const s16 a[XNR3_LOOK_UP_TABLE_POINTS] = {
42 -7213, -5580, -4371, -3421, -2722, -2159, -6950, -5585,
43 -4529, -3697, -3010, -2485, -2070, -1727, -1428, 0
46 static const s16 b[XNR3_LOOK_UP_TABLE_POINTS] = {
47 4096, 3603, 3178, 2811, 2497, 2226, 1990, 1783,
48 1603, 1446, 1307, 1185, 1077, 981, 895, 819
51 static const s16 c[XNR3_LOOK_UP_TABLE_POINTS] = {
52 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
56 * Default kernel parameters. In general, default is bypass mode or as close
57 * to the ineffective values as possible. Due to the chroma down+upsampling,
58 * perfect bypass mode is not possible for xnr3 filter itself. Instead, the
59 * 'blending' parameter is used to create a bypass.
61 const struct ia_css_xnr3_config default_xnr3_config = {
71 * Compute an alpha value for the ISP kernel from sigma value on the host
72 * parameter interface as: alpha_scale * 1/(sigma/sigma_scale)
75 compute_alpha(int sigma)
78 int offset = sigma / 2;
80 if (sigma < XNR_MIN_SIGMA) {
81 alpha = XNR_MAX_ALPHA;
83 alpha = ((IA_CSS_XNR3_SIGMA_SCALE * XNR_ALPHA_SCALE_FACTOR) + offset) / sigma;
85 if (alpha > XNR_MAX_ALPHA)
86 alpha = XNR_MAX_ALPHA;
93 * Compute the scaled coring value for the ISP kernel from the value on the
94 * host parameter interface.
97 compute_coring(int coring)
100 s32 isp_scale = XNR_CORING_SCALE_FACTOR;
101 s32 host_scale = IA_CSS_XNR3_CORING_SCALE;
102 s32 offset = host_scale / 2; /* fixed-point 0.5 */
104 /* Convert from public host-side scale factor to isp-side scale
105 * factor. Clip to [0, isp_scale-1).
107 isp_coring = ((coring * isp_scale) + offset) / host_scale;
108 return min(max(isp_coring, 0), isp_scale - 1);
112 * Compute the scaled blending strength for the ISP kernel from the value on
113 * the host parameter interface.
116 compute_blending(int strength)
119 s32 isp_scale = XNR_BLENDING_SCALE_FACTOR;
120 s32 host_scale = IA_CSS_XNR3_BLENDING_SCALE;
121 s32 offset = host_scale / 2; /* fixed-point 0.5 */
123 /* Convert from public host-side scale factor to isp-side scale
124 * factor. The blending factor is positive on the host side, but
125 * negative on the ISP side because +1.0 cannot be represented
126 * exactly as s0.11 fixed point, but -1.0 can.
128 isp_strength = -(((strength * isp_scale) + offset) / host_scale);
129 return MAX(MIN(isp_strength, 0), -XNR_BLENDING_SCALE_FACTOR);
134 struct sh_css_isp_xnr3_params *to,
135 const struct ia_css_xnr3_config *from,
138 int kernel_size = XNR_FILTER_SIZE;
139 /* The adjust factor is the next power of 2
140 w.r.t. the kernel size*/
141 int adjust_factor = ceil_pow2(kernel_size);
142 s32 max_diff = (1 << (ISP_VEC_ELEMBITS - 1)) - 1;
143 s32 min_diff = -(1 << (ISP_VEC_ELEMBITS - 1));
145 s32 alpha_y0 = compute_alpha(from->sigma.y0);
146 s32 alpha_y1 = compute_alpha(from->sigma.y1);
147 s32 alpha_u0 = compute_alpha(from->sigma.u0);
148 s32 alpha_u1 = compute_alpha(from->sigma.u1);
149 s32 alpha_v0 = compute_alpha(from->sigma.v0);
150 s32 alpha_v1 = compute_alpha(from->sigma.v1);
151 s32 alpha_ydiff = (alpha_y1 - alpha_y0) * adjust_factor / kernel_size;
152 s32 alpha_udiff = (alpha_u1 - alpha_u0) * adjust_factor / kernel_size;
153 s32 alpha_vdiff = (alpha_v1 - alpha_v0) * adjust_factor / kernel_size;
155 s32 coring_u0 = compute_coring(from->coring.u0);
156 s32 coring_u1 = compute_coring(from->coring.u1);
157 s32 coring_v0 = compute_coring(from->coring.v0);
158 s32 coring_v1 = compute_coring(from->coring.v1);
159 s32 coring_udiff = (coring_u1 - coring_u0) * adjust_factor / kernel_size;
160 s32 coring_vdiff = (coring_v1 - coring_v0) * adjust_factor / kernel_size;
162 s32 blending = compute_blending(from->blending.strength);
166 /* alpha's are represented in qN.5 format */
167 to->alpha.y0 = alpha_y0;
168 to->alpha.u0 = alpha_u0;
169 to->alpha.v0 = alpha_v0;
170 to->alpha.ydiff = min(max(alpha_ydiff, min_diff), max_diff);
171 to->alpha.udiff = min(max(alpha_udiff, min_diff), max_diff);
172 to->alpha.vdiff = min(max(alpha_vdiff, min_diff), max_diff);
174 /* coring parameters are expressed in q1.NN format */
175 to->coring.u0 = coring_u0;
176 to->coring.v0 = coring_v0;
177 to->coring.udiff = min(max(coring_udiff, min_diff), max_diff);
178 to->coring.vdiff = min(max(coring_vdiff, min_diff), max_diff);
180 /* blending strength is expressed in q1.NN format */
181 to->blending.strength = blending;
185 /* (void) = ia_css_xnr3_vmem_encode(*to, *from)
186 * -----------------------------------------------
187 * VMEM Encode Function to translate UV parameters from userspace into ISP space
190 ia_css_xnr3_vmem_encode(
191 struct sh_css_isp_xnr3_vmem_params *to,
192 const struct ia_css_xnr3_config *from,
195 unsigned int i, j, base;
196 const unsigned int total_blocks = 4;
197 const unsigned int shuffle_block = 16;
203 for (i = 0; i < ISP_VEC_NELEMS; i++) {
210 /* Constraints on "x":
211 * - values should be greater or equal to 0.
212 * - values should be ascending.
216 for (j = 1; j < XNR3_LOOK_UP_TABLE_POINTS; j++) {
218 assert(x[j] > x[j - 1]);
221 /* The implementation of the calulating 1/x is based on the availability
222 * of the OP_vec_shuffle16 operation.
223 * A 64 element vector is split up in 4 blocks of 16 element. Each array is copied to
224 * a vector 4 times, (starting at 0, 16, 32 and 48). All array elements are copied or
225 * initialised as described in the KFS. The remaining elements of a vector are set to 0.
227 /* TODO: guard this code with above assumptions */
228 for (i = 0; i < total_blocks; i++) {
229 base = shuffle_block * i;
231 for (j = 0; j < XNR3_LOOK_UP_TABLE_POINTS; j++) {
232 to->x[0][base + j] = x[j];
233 to->a[0][base + j] = a[j];
234 to->b[0][base + j] = b[j];
235 to->c[0][base + j] = c[j];
240 /* Dummy Function added as the tool expects it*/
242 ia_css_xnr3_debug_dtrace(
243 const struct ia_css_xnr3_config *config,