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"
20 #include "assert_support.h"
22 #include "ia_css_xnr3.host.h"
24 /* Maximum value for alpha on ISP interface */
25 #define XNR_MAX_ALPHA ((1 << (ISP_VEC_ELEMBITS - 1)) - 1)
27 /* Minimum value for sigma on host interface. Lower values translate to
30 #define XNR_MIN_SIGMA (IA_CSS_XNR3_SIGMA_SCALE / 100)
34 * division look-up table
37 #define XNR3_LOOK_UP_TABLE_POINTS 16
39 static const int16_t x[XNR3_LOOK_UP_TABLE_POINTS] = {
40 1024, 1164, 1320, 1492, 1680, 1884, 2108, 2352,
41 2616, 2900, 3208, 3540, 3896, 4276, 4684, 5120};
43 static const int16_t a[XNR3_LOOK_UP_TABLE_POINTS] = {
44 -7213, -5580, -4371, -3421, -2722, -2159, -6950, -5585,
45 -4529, -3697, -3010, -2485, -2070, -1727, -1428, 0};
47 static const int16_t b[XNR3_LOOK_UP_TABLE_POINTS] = {
48 4096, 3603, 3178, 2811, 2497, 2226, 1990, 1783,
49 1603, 1446, 1307, 1185, 1077, 981, 895, 819};
51 static const int16_t 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 #if defined(XNR_ATE_ROUNDING_BUG)
79 int32_t alpha_unscaled;
81 int offset = sigma / 2;
83 if (sigma < XNR_MIN_SIGMA) {
84 alpha = XNR_MAX_ALPHA;
86 #if defined(XNR_ATE_ROUNDING_BUG)
87 /* The scale factor for alpha must be the same as on the ISP,
88 * For sigma, it must match the public interface. The code
89 * below mimics the rounding and unintended loss of precision
90 * of the ATE reference code. It computes an unscaled alpha,
91 * rounds down, and then scales it to get the required fixed
92 * point representation. It would have been more precise to
93 * round after scaling. */
94 alpha_unscaled = IA_CSS_XNR3_SIGMA_SCALE / sigma;
95 alpha = alpha_unscaled * XNR_ALPHA_SCALE_FACTOR;
97 alpha = ((IA_CSS_XNR3_SIGMA_SCALE * XNR_ALPHA_SCALE_FACTOR) + offset)/ sigma;
100 if (alpha > XNR_MAX_ALPHA)
101 alpha = XNR_MAX_ALPHA;
108 * Compute the scaled coring value for the ISP kernel from the value on the
109 * host parameter interface.
112 compute_coring(int coring)
115 int32_t isp_scale = XNR_CORING_SCALE_FACTOR;
116 int32_t host_scale = IA_CSS_XNR3_CORING_SCALE;
117 int32_t offset = host_scale / 2; /* fixed-point 0.5 */
119 /* Convert from public host-side scale factor to isp-side scale
120 * factor. Clip to [0, isp_scale-1).
122 isp_coring = ((coring * isp_scale) + offset) / host_scale;
123 return min(max(isp_coring, 0), isp_scale - 1);
127 * Compute the scaled blending strength for the ISP kernel from the value on
128 * the host parameter interface.
131 compute_blending(int strength)
133 int32_t isp_strength;
134 int32_t isp_scale = XNR_BLENDING_SCALE_FACTOR;
135 int32_t host_scale = IA_CSS_XNR3_BLENDING_SCALE;
136 int32_t offset = host_scale / 2; /* fixed-point 0.5 */
138 /* Convert from public host-side scale factor to isp-side scale
139 * factor. The blending factor is positive on the host side, but
140 * negative on the ISP side because +1.0 cannot be represented
141 * exactly as s0.11 fixed point, but -1.0 can.
143 isp_strength = -(((strength * isp_scale) + offset) / host_scale);
144 return max(min(isp_strength, 0), -XNR_BLENDING_SCALE_FACTOR);
149 struct sh_css_isp_xnr3_params *to,
150 const struct ia_css_xnr3_config *from,
153 int kernel_size = XNR_FILTER_SIZE;
154 /* The adjust factor is the next power of 2
155 w.r.t. the kernel size*/
156 int adjust_factor = ceil_pow2(kernel_size);
157 int32_t max_diff = (1 << (ISP_VEC_ELEMBITS - 1)) - 1;
158 int32_t min_diff = -(1 << (ISP_VEC_ELEMBITS - 1));
160 int32_t alpha_y0 = compute_alpha(from->sigma.y0);
161 int32_t alpha_y1 = compute_alpha(from->sigma.y1);
162 int32_t alpha_u0 = compute_alpha(from->sigma.u0);
163 int32_t alpha_u1 = compute_alpha(from->sigma.u1);
164 int32_t alpha_v0 = compute_alpha(from->sigma.v0);
165 int32_t alpha_v1 = compute_alpha(from->sigma.v1);
166 int32_t alpha_ydiff = (alpha_y1 - alpha_y0) * adjust_factor / kernel_size;
167 int32_t alpha_udiff = (alpha_u1 - alpha_u0) * adjust_factor / kernel_size;
168 int32_t alpha_vdiff = (alpha_v1 - alpha_v0) * adjust_factor / kernel_size;
170 int32_t coring_u0 = compute_coring(from->coring.u0);
171 int32_t coring_u1 = compute_coring(from->coring.u1);
172 int32_t coring_v0 = compute_coring(from->coring.v0);
173 int32_t coring_v1 = compute_coring(from->coring.v1);
174 int32_t coring_udiff = (coring_u1 - coring_u0) * adjust_factor / kernel_size;
175 int32_t coring_vdiff = (coring_v1 - coring_v0) * adjust_factor / kernel_size;
177 int32_t blending = compute_blending(from->blending.strength);
181 /* alpha's are represented in qN.5 format */
182 to->alpha.y0 = alpha_y0;
183 to->alpha.u0 = alpha_u0;
184 to->alpha.v0 = alpha_v0;
185 to->alpha.ydiff = min(max(alpha_ydiff, min_diff), max_diff);
186 to->alpha.udiff = min(max(alpha_udiff, min_diff), max_diff);
187 to->alpha.vdiff = min(max(alpha_vdiff, min_diff), max_diff);
189 /* coring parameters are expressed in q1.NN format */
190 to->coring.u0 = coring_u0;
191 to->coring.v0 = coring_v0;
192 to->coring.udiff = min(max(coring_udiff, min_diff), max_diff);
193 to->coring.vdiff = min(max(coring_vdiff, min_diff), max_diff);
195 /* blending strength is expressed in q1.NN format */
196 to->blending.strength = blending;
200 /* (void) = ia_css_xnr3_vmem_encode(*to, *from)
201 * -----------------------------------------------
202 * VMEM Encode Function to translate UV parameters from userspace into ISP space
205 ia_css_xnr3_vmem_encode(
206 struct sh_css_isp_xnr3_vmem_params *to,
207 const struct ia_css_xnr3_config *from,
211 const unsigned total_blocks = 4;
212 const unsigned shuffle_block = 16;
218 for (i = 0; i < ISP_VEC_NELEMS; i++) {
225 /* Constraints on "x":
226 * - values should be greater or equal to 0.
227 * - values should be ascending.
231 for (j = 1; j < XNR3_LOOK_UP_TABLE_POINTS; j++) {
233 assert(x[j] > x[j - 1]);
237 /* The implementation of the calulating 1/x is based on the availability
238 * of the OP_vec_shuffle16 operation.
239 * A 64 element vector is split up in 4 blocks of 16 element. Each array is copied to
240 * a vector 4 times, (starting at 0, 16, 32 and 48). All array elements are copied or
241 * initialised as described in the KFS. The remaining elements of a vector are set to 0.
243 /* TODO: guard this code with above assumptions */
244 for (i = 0; i < total_blocks; i++) {
245 base = shuffle_block * i;
247 for (j = 0; j < XNR3_LOOK_UP_TABLE_POINTS; j++) {
248 to->x[0][base + j] = x[j];
249 to->a[0][base + j] = a[j];
250 to->b[0][base + j] = b[j];
251 to->c[0][base + j] = c[j];
257 /* Dummy Function added as the tool expects it*/
259 ia_css_xnr3_debug_dtrace(
260 const struct ia_css_xnr3_config *config,