1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Support for Intel Camera Imaging ISP subsystem.
4 * Copyright (c) 2015, Intel Corporation.
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms and conditions of the GNU General Public License,
8 * version 2, as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13 * more details.
14 */
15
16#include "type_support.h"
17#include "math_support.h"
18#include "sh_css_defs.h"
19#include "ia_css_types.h"
20#include "assert_support.h"
21#include "ia_css_xnr3.host.h"
22
23/* Maximum value for alpha on ISP interface */
24#define XNR_MAX_ALPHA  ((1 << (ISP_VEC_ELEMBITS - 1)) - 1)
25
26/* Minimum value for sigma on host interface. Lower values translate to
27 * max_alpha.
28 */
29#define XNR_MIN_SIGMA  (IA_CSS_XNR3_SIGMA_SCALE / 100)
30
31/*
32 * division look-up table
33 * Refers to XNR3.0.5
34 */
35#define XNR3_LOOK_UP_TABLE_POINTS 16
36
37static const s16 x[XNR3_LOOK_UP_TABLE_POINTS] = {
38	1024, 1164, 1320, 1492, 1680, 1884, 2108, 2352,
39	2616, 2900, 3208, 3540, 3896, 4276, 4684, 5120
40};
41
42static const s16 a[XNR3_LOOK_UP_TABLE_POINTS] = {
43	-7213, -5580, -4371, -3421, -2722, -2159, -6950, -5585,
44	    -4529, -3697, -3010, -2485, -2070, -1727, -1428, 0
45    };
46
47static const s16 b[XNR3_LOOK_UP_TABLE_POINTS] = {
48	4096, 3603, 3178, 2811, 2497, 2226, 1990, 1783,
49	1603, 1446, 1307, 1185, 1077, 981, 895, 819
50};
51
52static const s16 c[XNR3_LOOK_UP_TABLE_POINTS] = {
53	1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
54};
55
56/*
57 * Default kernel parameters. In general, default is bypass mode or as close
58 * to the ineffective values as possible. Due to the chroma down+upsampling,
59 * perfect bypass mode is not possible for xnr3 filter itself. Instead, the
60 * 'blending' parameter is used to create a bypass.
61 */
62const struct ia_css_xnr3_config default_xnr3_config = {
63	/* sigma */
64	{ 0, 0, 0, 0, 0, 0 },
65	/* coring */
66	{ 0, 0, 0, 0 },
67	/* blending */
68	{ 0 }
69};
70
71/*
72 * Compute an alpha value for the ISP kernel from sigma value on the host
73 * parameter interface as: alpha_scale * 1/(sigma/sigma_scale)
74 */
75static int32_t
76compute_alpha(int sigma)
77{
78	s32 alpha;
79	int offset = sigma / 2;
80
81	if (sigma < XNR_MIN_SIGMA) {
82		alpha = XNR_MAX_ALPHA;
83	} else {
84		alpha = ((IA_CSS_XNR3_SIGMA_SCALE * XNR_ALPHA_SCALE_FACTOR) + offset) / sigma;
85
86		if (alpha > XNR_MAX_ALPHA)
87			alpha = XNR_MAX_ALPHA;
88	}
89
90	return alpha;
91}
92
93/*
94 * Compute the scaled coring value for the ISP kernel from the value on the
95 * host parameter interface.
96 */
97static int32_t
98compute_coring(int coring)
99{
100	s32 isp_coring;
101	s32 isp_scale = XNR_CORING_SCALE_FACTOR;
102	s32 host_scale = IA_CSS_XNR3_CORING_SCALE;
103	s32 offset = host_scale / 2; /* fixed-point 0.5 */
104
105	/* Convert from public host-side scale factor to isp-side scale
106	 * factor. Clip to [0, isp_scale-1).
107	 */
108	isp_coring = ((coring * isp_scale) + offset) / host_scale;
109	return min(max(isp_coring, 0), isp_scale - 1);
110}
111
112/*
113 * Compute the scaled blending strength for the ISP kernel from the value on
114 * the host parameter interface.
115 */
116static int32_t
117compute_blending(int strength)
118{
119	s32 isp_strength;
120	s32 isp_scale = XNR_BLENDING_SCALE_FACTOR;
121	s32 host_scale = IA_CSS_XNR3_BLENDING_SCALE;
122	s32 offset = host_scale / 2; /* fixed-point 0.5 */
123
124	/* Convert from public host-side scale factor to isp-side scale
125	 * factor. The blending factor is positive on the host side, but
126	 * negative on the ISP side because +1.0 cannot be represented
127	 * exactly as s0.11 fixed point, but -1.0 can.
128	 */
129	isp_strength = -(((strength * isp_scale) + offset) / host_scale);
130	return MAX(MIN(isp_strength, 0), -isp_scale);
131}
132
133void
134ia_css_xnr3_encode(
135    struct sh_css_isp_xnr3_params *to,
136    const struct ia_css_xnr3_config *from,
137    unsigned int size)
138{
139	int kernel_size = XNR_FILTER_SIZE;
140	/* The adjust factor is the next power of 2
141	   w.r.t. the kernel size*/
142	int adjust_factor = ceil_pow2(kernel_size);
143	s32 max_diff = (1 << (ISP_VEC_ELEMBITS - 1)) - 1;
144	s32 min_diff = -(1 << (ISP_VEC_ELEMBITS - 1));
145
146	s32 alpha_y0 = compute_alpha(from->sigma.y0);
147	s32 alpha_y1 = compute_alpha(from->sigma.y1);
148	s32 alpha_u0 = compute_alpha(from->sigma.u0);
149	s32 alpha_u1 = compute_alpha(from->sigma.u1);
150	s32 alpha_v0 = compute_alpha(from->sigma.v0);
151	s32 alpha_v1 = compute_alpha(from->sigma.v1);
152	s32 alpha_ydiff = (alpha_y1 - alpha_y0) * adjust_factor / kernel_size;
153	s32 alpha_udiff = (alpha_u1 - alpha_u0) * adjust_factor / kernel_size;
154	s32 alpha_vdiff = (alpha_v1 - alpha_v0) * adjust_factor / kernel_size;
155
156	s32 coring_u0 = compute_coring(from->coring.u0);
157	s32 coring_u1 = compute_coring(from->coring.u1);
158	s32 coring_v0 = compute_coring(from->coring.v0);
159	s32 coring_v1 = compute_coring(from->coring.v1);
160	s32 coring_udiff = (coring_u1 - coring_u0) * adjust_factor / kernel_size;
161	s32 coring_vdiff = (coring_v1 - coring_v0) * adjust_factor / kernel_size;
162
163	s32 blending = compute_blending(from->blending.strength);
164
165	(void)size;
166
167	/* alpha's are represented in qN.5 format */
168	to->alpha.y0 = alpha_y0;
169	to->alpha.u0 = alpha_u0;
170	to->alpha.v0 = alpha_v0;
171	to->alpha.ydiff = min(max(alpha_ydiff, min_diff), max_diff);
172	to->alpha.udiff = min(max(alpha_udiff, min_diff), max_diff);
173	to->alpha.vdiff = min(max(alpha_vdiff, min_diff), max_diff);
174
175	/* coring parameters are expressed in q1.NN format */
176	to->coring.u0 = coring_u0;
177	to->coring.v0 = coring_v0;
178	to->coring.udiff = min(max(coring_udiff, min_diff), max_diff);
179	to->coring.vdiff = min(max(coring_vdiff, min_diff), max_diff);
180
181	/* blending strength is expressed in q1.NN format */
182	to->blending.strength = blending;
183}
184
185/* ISP2401 */
186/* (void) = ia_css_xnr3_vmem_encode(*to, *from)
187 * -----------------------------------------------
188 * VMEM Encode Function to translate UV parameters from userspace into ISP space
189*/
190void
191ia_css_xnr3_vmem_encode(
192    struct sh_css_isp_xnr3_vmem_params *to,
193    const struct ia_css_xnr3_config *from,
194    unsigned int size)
195{
196	unsigned int i, j, base;
197	const unsigned int total_blocks = 4;
198	const unsigned int shuffle_block = 16;
199
200	(void)from;
201	(void)size;
202
203	/* Init */
204	for (i = 0; i < ISP_VEC_NELEMS; i++) {
205		to->x[0][i] = 0;
206		to->a[0][i] = 0;
207		to->b[0][i] = 0;
208		to->c[0][i] = 0;
209	}
210
211	/* Constraints on "x":
212	 * - values should be greater or equal to 0.
213	 * - values should be ascending.
214	 */
215	assert(x[0] >= 0);
216
217	for (j = 1; j < XNR3_LOOK_UP_TABLE_POINTS; j++) {
218		assert(x[j] >= 0);
219		assert(x[j] > x[j - 1]);
220	}
221
222	/* The implementation of the calulating 1/x is based on the availability
223	 * of the OP_vec_shuffle16 operation.
224	 * A 64 element vector is split up in 4 blocks of 16 element. Each array is copied to
225	 * a vector 4 times, (starting at 0, 16, 32 and 48). All array elements are copied or
226	 * initialised as described in the KFS. The remaining elements of a vector are set to 0.
227	 */
228	/* TODO: guard this code with above assumptions */
229	for (i = 0; i < total_blocks; i++) {
230		base = shuffle_block * i;
231
232		for (j = 0; j < XNR3_LOOK_UP_TABLE_POINTS; j++) {
233			to->x[0][base + j] = x[j];
234			to->a[0][base + j] = a[j];
235			to->b[0][base + j] = b[j];
236			to->c[0][base + j] = c[j];
237		}
238	}
239}
240
241/* Dummy Function added as the tool expects it*/
242void
243ia_css_xnr3_debug_dtrace(
244    const struct ia_css_xnr3_config *config,
245    unsigned int level)
246{
247	(void)config;
248	(void)level;
249}
250