Loading...
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 | /*
* Support for Intel Camera Imaging ISP subsystem.
* Copyright (c) 2015, Intel Corporation.
*
* This program is free software; you can redistribute it and/or modify it
* under the terms and conditions of the GNU General Public License,
* version 2, as published by the Free Software Foundation.
*
* This program is distributed in the hope it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
* more details.
*/
#include "type_support.h"
#include "math_support.h"
#include "sh_css_defs.h"
#include "ia_css_types.h"
#ifdef ISP2401
#include "assert_support.h"
#endif
#include "ia_css_xnr3.host.h"
/* Maximum value for alpha on ISP interface */
#define XNR_MAX_ALPHA ((1 << (ISP_VEC_ELEMBITS - 1)) - 1)
/* Minimum value for sigma on host interface. Lower values translate to
* max_alpha.
*/
#define XNR_MIN_SIGMA (IA_CSS_XNR3_SIGMA_SCALE / 100)
/*
#ifdef ISP2401
* division look-up table
* Refers to XNR3.0.5
*/
#define XNR3_LOOK_UP_TABLE_POINTS 16
static const int16_t x[XNR3_LOOK_UP_TABLE_POINTS] = {
1024, 1164, 1320, 1492, 1680, 1884, 2108, 2352,
2616, 2900, 3208, 3540, 3896, 4276, 4684, 5120};
static const int16_t a[XNR3_LOOK_UP_TABLE_POINTS] = {
-7213, -5580, -4371, -3421, -2722, -2159, -6950, -5585,
-4529, -3697, -3010, -2485, -2070, -1727, -1428, 0};
static const int16_t b[XNR3_LOOK_UP_TABLE_POINTS] = {
4096, 3603, 3178, 2811, 2497, 2226, 1990, 1783,
1603, 1446, 1307, 1185, 1077, 981, 895, 819};
static const int16_t c[XNR3_LOOK_UP_TABLE_POINTS] = {
1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
/*
#endif
* Default kernel parameters. In general, default is bypass mode or as close
* to the ineffective values as possible. Due to the chroma down+upsampling,
* perfect bypass mode is not possible for xnr3 filter itself. Instead, the
* 'blending' parameter is used to create a bypass.
*/
const struct ia_css_xnr3_config default_xnr3_config = {
/* sigma */
{ 0, 0, 0, 0, 0, 0 },
/* coring */
{ 0, 0, 0, 0 },
/* blending */
{ 0 }
};
/*
* Compute an alpha value for the ISP kernel from sigma value on the host
* parameter interface as: alpha_scale * 1/(sigma/sigma_scale)
*/
static int32_t
compute_alpha(int sigma)
{
int32_t alpha;
#if defined(XNR_ATE_ROUNDING_BUG)
int32_t alpha_unscaled;
#else
int offset = sigma / 2;
#endif
if (sigma < XNR_MIN_SIGMA) {
alpha = XNR_MAX_ALPHA;
} else {
#if defined(XNR_ATE_ROUNDING_BUG)
/* The scale factor for alpha must be the same as on the ISP,
* For sigma, it must match the public interface. The code
* below mimics the rounding and unintended loss of precision
* of the ATE reference code. It computes an unscaled alpha,
* rounds down, and then scales it to get the required fixed
* point representation. It would have been more precise to
* round after scaling. */
alpha_unscaled = IA_CSS_XNR3_SIGMA_SCALE / sigma;
alpha = alpha_unscaled * XNR_ALPHA_SCALE_FACTOR;
#else
alpha = ((IA_CSS_XNR3_SIGMA_SCALE * XNR_ALPHA_SCALE_FACTOR) + offset)/ sigma;
#endif
if (alpha > XNR_MAX_ALPHA)
alpha = XNR_MAX_ALPHA;
}
return alpha;
}
/*
* Compute the scaled coring value for the ISP kernel from the value on the
* host parameter interface.
*/
static int32_t
compute_coring(int coring)
{
int32_t isp_coring;
int32_t isp_scale = XNR_CORING_SCALE_FACTOR;
int32_t host_scale = IA_CSS_XNR3_CORING_SCALE;
int32_t offset = host_scale / 2; /* fixed-point 0.5 */
/* Convert from public host-side scale factor to isp-side scale
* factor. Clip to [0, isp_scale-1).
*/
isp_coring = ((coring * isp_scale) + offset) / host_scale;
return min(max(isp_coring, 0), isp_scale - 1);
}
/*
* Compute the scaled blending strength for the ISP kernel from the value on
* the host parameter interface.
*/
static int32_t
compute_blending(int strength)
{
int32_t isp_strength;
int32_t isp_scale = XNR_BLENDING_SCALE_FACTOR;
int32_t host_scale = IA_CSS_XNR3_BLENDING_SCALE;
int32_t offset = host_scale / 2; /* fixed-point 0.5 */
/* Convert from public host-side scale factor to isp-side scale
* factor. The blending factor is positive on the host side, but
* negative on the ISP side because +1.0 cannot be represented
* exactly as s0.11 fixed point, but -1.0 can.
*/
isp_strength = -(((strength * isp_scale) + offset) / host_scale);
return max(min(isp_strength, 0), -XNR_BLENDING_SCALE_FACTOR);
}
void
ia_css_xnr3_encode(
struct sh_css_isp_xnr3_params *to,
const struct ia_css_xnr3_config *from,
unsigned size)
{
int kernel_size = XNR_FILTER_SIZE;
/* The adjust factor is the next power of 2
w.r.t. the kernel size*/
int adjust_factor = ceil_pow2(kernel_size);
int32_t max_diff = (1 << (ISP_VEC_ELEMBITS - 1)) - 1;
int32_t min_diff = -(1 << (ISP_VEC_ELEMBITS - 1));
int32_t alpha_y0 = compute_alpha(from->sigma.y0);
int32_t alpha_y1 = compute_alpha(from->sigma.y1);
int32_t alpha_u0 = compute_alpha(from->sigma.u0);
int32_t alpha_u1 = compute_alpha(from->sigma.u1);
int32_t alpha_v0 = compute_alpha(from->sigma.v0);
int32_t alpha_v1 = compute_alpha(from->sigma.v1);
int32_t alpha_ydiff = (alpha_y1 - alpha_y0) * adjust_factor / kernel_size;
int32_t alpha_udiff = (alpha_u1 - alpha_u0) * adjust_factor / kernel_size;
int32_t alpha_vdiff = (alpha_v1 - alpha_v0) * adjust_factor / kernel_size;
int32_t coring_u0 = compute_coring(from->coring.u0);
int32_t coring_u1 = compute_coring(from->coring.u1);
int32_t coring_v0 = compute_coring(from->coring.v0);
int32_t coring_v1 = compute_coring(from->coring.v1);
int32_t coring_udiff = (coring_u1 - coring_u0) * adjust_factor / kernel_size;
int32_t coring_vdiff = (coring_v1 - coring_v0) * adjust_factor / kernel_size;
int32_t blending = compute_blending(from->blending.strength);
(void)size;
/* alpha's are represented in qN.5 format */
to->alpha.y0 = alpha_y0;
to->alpha.u0 = alpha_u0;
to->alpha.v0 = alpha_v0;
to->alpha.ydiff = min(max(alpha_ydiff, min_diff), max_diff);
to->alpha.udiff = min(max(alpha_udiff, min_diff), max_diff);
to->alpha.vdiff = min(max(alpha_vdiff, min_diff), max_diff);
/* coring parameters are expressed in q1.NN format */
to->coring.u0 = coring_u0;
to->coring.v0 = coring_v0;
to->coring.udiff = min(max(coring_udiff, min_diff), max_diff);
to->coring.vdiff = min(max(coring_vdiff, min_diff), max_diff);
/* blending strength is expressed in q1.NN format */
to->blending.strength = blending;
}
#ifdef ISP2401
/* (void) = ia_css_xnr3_vmem_encode(*to, *from)
* -----------------------------------------------
* VMEM Encode Function to translate UV parameters from userspace into ISP space
*/
void
ia_css_xnr3_vmem_encode(
struct sh_css_isp_xnr3_vmem_params *to,
const struct ia_css_xnr3_config *from,
unsigned size)
{
unsigned i, j, base;
const unsigned total_blocks = 4;
const unsigned shuffle_block = 16;
(void)from;
(void)size;
/* Init */
for (i = 0; i < ISP_VEC_NELEMS; i++) {
to->x[0][i] = 0;
to->a[0][i] = 0;
to->b[0][i] = 0;
to->c[0][i] = 0;
}
/* Constraints on "x":
* - values should be greater or equal to 0.
* - values should be ascending.
*/
assert(x[0] >= 0);
for (j = 1; j < XNR3_LOOK_UP_TABLE_POINTS; j++) {
assert(x[j] >= 0);
assert(x[j] > x[j - 1]);
}
/* The implementation of the calulating 1/x is based on the availability
* of the OP_vec_shuffle16 operation.
* A 64 element vector is split up in 4 blocks of 16 element. Each array is copied to
* a vector 4 times, (starting at 0, 16, 32 and 48). All array elements are copied or
* initialised as described in the KFS. The remaining elements of a vector are set to 0.
*/
/* TODO: guard this code with above assumptions */
for (i = 0; i < total_blocks; i++) {
base = shuffle_block * i;
for (j = 0; j < XNR3_LOOK_UP_TABLE_POINTS; j++) {
to->x[0][base + j] = x[j];
to->a[0][base + j] = a[j];
to->b[0][base + j] = b[j];
to->c[0][base + j] = c[j];
}
}
}
#endif
/* Dummy Function added as the tool expects it*/
void
ia_css_xnr3_debug_dtrace(
const struct ia_css_xnr3_config *config,
unsigned level)
{
(void)config;
(void)level;
}
|