1// SPDX-License-Identifier: GPL-2.0-only 2/* 3 * Scaler library 4 * 5 * Copyright (c) 2013 Texas Instruments Inc. 6 * 7 * David Griego, <dagriego@biglakesoftware.com> 8 * Dale Farnsworth, <dale@farnsworth.org> 9 * Archit Taneja, <archit@ti.com> 10 */ 11 12#include <linux/err.h> 13#include <linux/io.h> 14#include <linux/module.h> 15#include <linux/platform_device.h> 16#include <linux/slab.h> 17 18#include "sc.h" 19#include "sc_coeff.h" 20 21void sc_dump_regs(struct sc_data *sc) 22{ 23 struct device *dev = &sc->pdev->dev; 24 25#define DUMPREG(r) dev_dbg(dev, "%-35s %08x\n", #r, \ 26 ioread32(sc->base + CFG_##r)) 27 28 dev_dbg(dev, "SC Registers @ %pa:\n", &sc->res->start); 29 30 DUMPREG(SC0); 31 DUMPREG(SC1); 32 DUMPREG(SC2); 33 DUMPREG(SC3); 34 DUMPREG(SC4); 35 DUMPREG(SC5); 36 DUMPREG(SC6); 37 DUMPREG(SC8); 38 DUMPREG(SC9); 39 DUMPREG(SC10); 40 DUMPREG(SC11); 41 DUMPREG(SC12); 42 DUMPREG(SC13); 43 DUMPREG(SC17); 44 DUMPREG(SC18); 45 DUMPREG(SC19); 46 DUMPREG(SC20); 47 DUMPREG(SC21); 48 DUMPREG(SC22); 49 DUMPREG(SC23); 50 DUMPREG(SC24); 51 DUMPREG(SC25); 52 53#undef DUMPREG 54} 55EXPORT_SYMBOL(sc_dump_regs); 56 57/* 58 * set the horizontal scaler coefficients according to the ratio of output to 59 * input widths, after accounting for up to two levels of decimation 60 */ 61void sc_set_hs_coeffs(struct sc_data *sc, void *addr, unsigned int src_w, 62 unsigned int dst_w) 63{ 64 int sixteenths; 65 int idx; 66 int i, j; 67 u16 *coeff_h = addr; 68 const u16 *cp; 69 70 if (dst_w > src_w) { 71 idx = HS_UP_SCALE; 72 } else { 73 if ((dst_w << 1) < src_w) 74 dst_w <<= 1; /* first level decimation */ 75 if ((dst_w << 1) < src_w) 76 dst_w <<= 1; /* second level decimation */ 77 78 if (dst_w == src_w) { 79 idx = HS_LE_16_16_SCALE; 80 } else { 81 sixteenths = (dst_w << 4) / src_w; 82 if (sixteenths < 8) 83 sixteenths = 8; 84 idx = HS_LT_9_16_SCALE + sixteenths - 8; 85 } 86 } 87 88 cp = scaler_hs_coeffs[idx]; 89 90 for (i = 0; i < SC_NUM_PHASES * 2; i++) { 91 for (j = 0; j < SC_H_NUM_TAPS; j++) 92 *coeff_h++ = *cp++; 93 /* 94 * for each phase, the scaler expects space for 8 coefficients 95 * in it's memory. For the horizontal scaler, we copy the first 96 * 7 coefficients and skip the last slot to move to the next 97 * row to hold coefficients for the next phase 98 */ 99 coeff_h += SC_NUM_TAPS_MEM_ALIGN - SC_H_NUM_TAPS; 100 } 101 102 sc->load_coeff_h = true; 103} 104EXPORT_SYMBOL(sc_set_hs_coeffs); 105 106/* 107 * set the vertical scaler coefficients according to the ratio of output to 108 * input heights 109 */ 110void sc_set_vs_coeffs(struct sc_data *sc, void *addr, unsigned int src_h, 111 unsigned int dst_h) 112{ 113 int sixteenths; 114 int idx; 115 int i, j; 116 u16 *coeff_v = addr; 117 const u16 *cp; 118 119 if (dst_h > src_h) { 120 idx = VS_UP_SCALE; 121 } else if (dst_h == src_h) { 122 idx = VS_1_TO_1_SCALE; 123 } else { 124 sixteenths = (dst_h << 4) / src_h; 125 if (sixteenths < 8) 126 sixteenths = 8; 127 idx = VS_LT_9_16_SCALE + sixteenths - 8; 128 } 129 130 cp = scaler_vs_coeffs[idx]; 131 132 for (i = 0; i < SC_NUM_PHASES * 2; i++) { 133 for (j = 0; j < SC_V_NUM_TAPS; j++) 134 *coeff_v++ = *cp++; 135 /* 136 * for the vertical scaler, we copy the first 5 coefficients and 137 * skip the last 3 slots to move to the next row to hold 138 * coefficients for the next phase 139 */ 140 coeff_v += SC_NUM_TAPS_MEM_ALIGN - SC_V_NUM_TAPS; 141 } 142 143 sc->load_coeff_v = true; 144} 145EXPORT_SYMBOL(sc_set_vs_coeffs); 146 147void sc_config_scaler(struct sc_data *sc, u32 *sc_reg0, u32 *sc_reg8, 148 u32 *sc_reg17, unsigned int src_w, unsigned int src_h, 149 unsigned int dst_w, unsigned int dst_h) 150{ 151 struct device *dev = &sc->pdev->dev; 152 u32 val; 153 int dcm_x, dcm_shift; 154 bool use_rav; 155 unsigned long lltmp; 156 u32 lin_acc_inc, lin_acc_inc_u; 157 u32 col_acc_offset; 158 u16 factor = 0; 159 int row_acc_init_rav = 0, row_acc_init_rav_b = 0; 160 u32 row_acc_inc = 0, row_acc_offset = 0, row_acc_offset_b = 0; 161 /* 162 * location of SC register in payload memory with respect to the first 163 * register in the mmr address data block 164 */ 165 u32 *sc_reg9 = sc_reg8 + 1; 166 u32 *sc_reg12 = sc_reg8 + 4; 167 u32 *sc_reg13 = sc_reg8 + 5; 168 u32 *sc_reg24 = sc_reg17 + 7; 169 170 val = sc_reg0[0]; 171 172 /* clear all the features(they may get enabled elsewhere later) */ 173 val &= ~(CFG_SELFGEN_FID | CFG_TRIM | CFG_ENABLE_SIN2_VER_INTP | 174 CFG_INTERLACE_I | CFG_DCM_4X | CFG_DCM_2X | CFG_AUTO_HS | 175 CFG_ENABLE_EV | CFG_USE_RAV | CFG_INVT_FID | CFG_SC_BYPASS | 176 CFG_INTERLACE_O | CFG_Y_PK_EN | CFG_HP_BYPASS | CFG_LINEAR); 177 178 if (src_w == dst_w && src_h == dst_h) { 179 val |= CFG_SC_BYPASS; 180 sc_reg0[0] = val; 181 return; 182 } 183 184 /* we only support linear scaling for now */ 185 val |= CFG_LINEAR; 186 187 /* configure horizontal scaler */ 188 189 /* enable 2X or 4X decimation */ 190 dcm_x = src_w / dst_w; 191 if (dcm_x > 4) { 192 val |= CFG_DCM_4X; 193 dcm_shift = 2; 194 } else if (dcm_x > 2) { 195 val |= CFG_DCM_2X; 196 dcm_shift = 1; 197 } else { 198 dcm_shift = 0; 199 } 200 201 lltmp = dst_w - 1; 202 lin_acc_inc = div64_u64(((u64)(src_w >> dcm_shift) - 1) << 24, lltmp); 203 lin_acc_inc_u = 0; 204 col_acc_offset = 0; 205 206 dev_dbg(dev, "hs config: src_w = %d, dst_w = %d, decimation = %s, lin_acc_inc = %08x\n", 207 src_w, dst_w, dcm_shift == 2 ? "4x" : 208 (dcm_shift == 1 ? "2x" : "none"), lin_acc_inc); 209 210 /* configure vertical scaler */ 211 212 /* use RAV for vertical scaler if vertical downscaling is > 4x */ 213 if (dst_h < (src_h >> 2)) { 214 use_rav = true; 215 val |= CFG_USE_RAV; 216 } else { 217 use_rav = false; 218 } 219 220 if (use_rav) { 221 /* use RAV */ 222 factor = (u16) ((dst_h << 10) / src_h); 223 224 row_acc_init_rav = factor + ((1 + factor) >> 1); 225 if (row_acc_init_rav >= 1024) 226 row_acc_init_rav -= 1024; 227 228 row_acc_init_rav_b = row_acc_init_rav + 229 (1 + (row_acc_init_rav >> 1)) - 230 (1024 >> 1); 231 232 if (row_acc_init_rav_b < 0) { 233 row_acc_init_rav_b += row_acc_init_rav; 234 row_acc_init_rav *= 2; 235 } 236 237 dev_dbg(dev, "vs config(RAV): src_h = %d, dst_h = %d, factor = %d, acc_init = %08x, acc_init_b = %08x\n", 238 src_h, dst_h, factor, row_acc_init_rav, 239 row_acc_init_rav_b); 240 } else { 241 /* use polyphase */ 242 row_acc_inc = ((src_h - 1) << 16) / (dst_h - 1); 243 row_acc_offset = 0; 244 row_acc_offset_b = 0; 245 246 dev_dbg(dev, "vs config(POLY): src_h = %d, dst_h = %d,row_acc_inc = %08x\n", 247 src_h, dst_h, row_acc_inc); 248 } 249 250 251 sc_reg0[0] = val; 252 sc_reg0[1] = row_acc_inc; 253 sc_reg0[2] = row_acc_offset; 254 sc_reg0[3] = row_acc_offset_b; 255 256 sc_reg0[4] = ((lin_acc_inc_u & CFG_LIN_ACC_INC_U_MASK) << 257 CFG_LIN_ACC_INC_U_SHIFT) | (dst_w << CFG_TAR_W_SHIFT) | 258 (dst_h << CFG_TAR_H_SHIFT); 259 260 sc_reg0[5] = (src_w << CFG_SRC_W_SHIFT) | (src_h << CFG_SRC_H_SHIFT); 261 262 sc_reg0[6] = (row_acc_init_rav_b << CFG_ROW_ACC_INIT_RAV_B_SHIFT) | 263 (row_acc_init_rav << CFG_ROW_ACC_INIT_RAV_SHIFT); 264 265 *sc_reg9 = lin_acc_inc; 266 267 *sc_reg12 = col_acc_offset << CFG_COL_ACC_OFFSET_SHIFT; 268 269 *sc_reg13 = factor; 270 271 *sc_reg24 = (src_w << CFG_ORG_W_SHIFT) | (src_h << CFG_ORG_H_SHIFT); 272} 273EXPORT_SYMBOL(sc_config_scaler); 274 275struct sc_data *sc_create(struct platform_device *pdev, const char *res_name) 276{ 277 struct sc_data *sc; 278 279 dev_dbg(&pdev->dev, "sc_create\n"); 280 281 sc = devm_kzalloc(&pdev->dev, sizeof(*sc), GFP_KERNEL); 282 if (!sc) { 283 dev_err(&pdev->dev, "couldn't alloc sc_data\n"); 284 return ERR_PTR(-ENOMEM); 285 } 286 287 sc->pdev = pdev; 288 289 sc->res = platform_get_resource_byname(pdev, IORESOURCE_MEM, res_name); 290 if (!sc->res) { 291 dev_err(&pdev->dev, "missing '%s' platform resources data\n", 292 res_name); 293 return ERR_PTR(-ENODEV); 294 } 295 296 sc->base = devm_ioremap_resource(&pdev->dev, sc->res); 297 if (IS_ERR(sc->base)) 298 return ERR_CAST(sc->base); 299 300 return sc; 301} 302EXPORT_SYMBOL(sc_create); 303 304MODULE_DESCRIPTION("TI VIP/VPE Scaler"); 305MODULE_AUTHOR("Texas Instruments Inc."); 306MODULE_LICENSE("GPL v2"); 307