1/*
2 * Scaler library
3 *
4 * Copyright (c) 2013 Texas Instruments Inc.
5 *
6 * David Griego, <dagriego@biglakesoftware.com>
7 * Dale Farnsworth, <dale@farnsworth.org>
8 * Archit Taneja, <archit@ti.com>
9 *
10 * This program is free software; you can redistribute it and/or modify it
11 * under the terms of the GNU General Public License version 2 as published by
12 * the Free Software Foundation.
13 */
14
15#include <linux/err.h>
16#include <linux/io.h>
17#include <linux/platform_device.h>
18#include <linux/slab.h>
19
20#include "sc.h"
21#include "sc_coeff.h"
22
23void sc_dump_regs(struct sc_data *sc)
24{
25	struct device *dev = &sc->pdev->dev;
26
27#define DUMPREG(r) dev_dbg(dev, "%-35s %08x\n", #r, \
28	ioread32(sc->base + CFG_##r))
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}
55
56/*
57 * set the horizontal scaler coefficients according to the ratio of output to
58 * input widths, after accounting for up to two levels of decimation
59 */
60void sc_set_hs_coeffs(struct sc_data *sc, void *addr, unsigned int src_w,
61		unsigned int dst_w)
62{
63	int sixteenths;
64	int idx;
65	int i, j;
66	u16 *coeff_h = addr;
67	const u16 *cp;
68
69	if (dst_w > src_w) {
70		idx = HS_UP_SCALE;
71	} else {
72		if ((dst_w << 1) < src_w)
73			dst_w <<= 1;	/* first level decimation */
74		if ((dst_w << 1) < src_w)
75			dst_w <<= 1;	/* second level decimation */
76
77		if (dst_w == src_w) {
78			idx = HS_LE_16_16_SCALE;
79		} else {
80			sixteenths = (dst_w << 4) / src_w;
81			if (sixteenths < 8)
82				sixteenths = 8;
83			idx = HS_LT_9_16_SCALE + sixteenths - 8;
84		}
85	}
86
87	if (idx == sc->hs_index)
88		return;
89
90	cp = scaler_hs_coeffs[idx];
91
92	for (i = 0; i < SC_NUM_PHASES * 2; i++) {
93		for (j = 0; j < SC_H_NUM_TAPS; j++)
94			*coeff_h++ = *cp++;
95		/*
96		 * for each phase, the scaler expects space for 8 coefficients
97		 * in it's memory. For the horizontal scaler, we copy the first
98		 * 7 coefficients and skip the last slot to move to the next
99		 * row to hold coefficients for the next phase
100		 */
101		coeff_h += SC_NUM_TAPS_MEM_ALIGN - SC_H_NUM_TAPS;
102	}
103
104	sc->hs_index = idx;
105
106	sc->load_coeff_h = true;
107}
108
109/*
110 * set the vertical scaler coefficients according to the ratio of output to
111 * input heights
112 */
113void sc_set_vs_coeffs(struct sc_data *sc, void *addr, unsigned int src_h,
114		unsigned int dst_h)
115{
116	int sixteenths;
117	int idx;
118	int i, j;
119	u16 *coeff_v = addr;
120	const u16 *cp;
121
122	if (dst_h > src_h) {
123		idx = VS_UP_SCALE;
124	} else if (dst_h == src_h) {
125		idx = VS_1_TO_1_SCALE;
126	} else {
127		sixteenths = (dst_h << 4) / src_h;
128		if (sixteenths < 8)
129			sixteenths = 8;
130		idx = VS_LT_9_16_SCALE + sixteenths - 8;
131	}
132
133	if (idx == sc->vs_index)
134		return;
135
136	cp = scaler_vs_coeffs[idx];
137
138	for (i = 0; i < SC_NUM_PHASES * 2; i++) {
139		for (j = 0; j < SC_V_NUM_TAPS; j++)
140			*coeff_v++ = *cp++;
141		/*
142		 * for the vertical scaler, we copy the first 5 coefficients and
143		 * skip the last 3 slots to move to the next row to hold
144		 * coefficients for the next phase
145		 */
146		coeff_v += SC_NUM_TAPS_MEM_ALIGN - SC_V_NUM_TAPS;
147	}
148
149	sc->vs_index = idx;
150	sc->load_coeff_v = true;
151}
152
153void sc_config_scaler(struct sc_data *sc, u32 *sc_reg0, u32 *sc_reg8,
154		u32 *sc_reg17, unsigned int src_w, unsigned int src_h,
155		unsigned int dst_w, unsigned int dst_h)
156{
157	struct device *dev = &sc->pdev->dev;
158	u32 val;
159	int dcm_x, dcm_shift;
160	bool use_rav;
161	unsigned long lltmp;
162	u32 lin_acc_inc, lin_acc_inc_u;
163	u32 col_acc_offset;
164	u16 factor = 0;
165	int row_acc_init_rav = 0, row_acc_init_rav_b = 0;
166	u32 row_acc_inc = 0, row_acc_offset = 0, row_acc_offset_b = 0;
167	/*
168	 * location of SC register in payload memory with respect to the first
169	 * register in the mmr address data block
170	 */
171	u32 *sc_reg9 = sc_reg8 + 1;
172	u32 *sc_reg12 = sc_reg8 + 4;
173	u32 *sc_reg13 = sc_reg8 + 5;
174	u32 *sc_reg24 = sc_reg17 + 7;
175
176	val = sc_reg0[0];
177
178	/* clear all the features(they may get enabled elsewhere later) */
179	val &= ~(CFG_SELFGEN_FID | CFG_TRIM | CFG_ENABLE_SIN2_VER_INTP |
180		CFG_INTERLACE_I | CFG_DCM_4X | CFG_DCM_2X | CFG_AUTO_HS |
181		CFG_ENABLE_EV | CFG_USE_RAV | CFG_INVT_FID | CFG_SC_BYPASS |
182		CFG_INTERLACE_O | CFG_Y_PK_EN | CFG_HP_BYPASS | CFG_LINEAR);
183
184	if (src_w == dst_w && src_h == dst_h) {
185		val |= CFG_SC_BYPASS;
186		sc_reg0[0] = val;
187		return;
188	}
189
190	/* we only support linear scaling for now */
191	val |= CFG_LINEAR;
192
193	/* configure horizontal scaler */
194
195	/* enable 2X or 4X decimation */
196	dcm_x = src_w / dst_w;
197	if (dcm_x > 4) {
198		val |= CFG_DCM_4X;
199		dcm_shift = 2;
200	} else if (dcm_x > 2) {
201		val |= CFG_DCM_2X;
202		dcm_shift = 1;
203	} else {
204		dcm_shift = 0;
205	}
206
207	lltmp = dst_w - 1;
208	lin_acc_inc = div64_u64(((u64)(src_w >> dcm_shift) - 1) << 24, lltmp);
209	lin_acc_inc_u = 0;
210	col_acc_offset = 0;
211
212	dev_dbg(dev, "hs config: src_w = %d, dst_w = %d, decimation = %s, lin_acc_inc = %08x\n",
213		src_w, dst_w, dcm_shift == 2 ? "4x" :
214		(dcm_shift == 1 ? "2x" : "none"), lin_acc_inc);
215
216	/* configure vertical scaler */
217
218	/* use RAV for vertical scaler if vertical downscaling is > 4x */
219	if (dst_h < (src_h >> 2)) {
220		use_rav = true;
221		val |= CFG_USE_RAV;
222	} else {
223		use_rav = false;
224	}
225
226	if (use_rav) {
227		/* use RAV */
228		factor = (u16) ((dst_h << 10) / src_h);
229
230		row_acc_init_rav = factor + ((1 + factor) >> 1);
231		if (row_acc_init_rav >= 1024)
232			row_acc_init_rav -= 1024;
233
234		row_acc_init_rav_b = row_acc_init_rav +
235				(1 + (row_acc_init_rav >> 1)) -
236				(1024 >> 1);
237
238		if (row_acc_init_rav_b < 0) {
239			row_acc_init_rav_b += row_acc_init_rav;
240			row_acc_init_rav *= 2;
241		}
242
243		dev_dbg(dev, "vs config(RAV): src_h = %d, dst_h = %d, factor = %d, acc_init = %08x, acc_init_b = %08x\n",
244			src_h, dst_h, factor, row_acc_init_rav,
245			row_acc_init_rav_b);
246	} else {
247		/* use polyphase */
248		row_acc_inc = ((src_h - 1) << 16) / (dst_h - 1);
249		row_acc_offset = 0;
250		row_acc_offset_b = 0;
251
252		dev_dbg(dev, "vs config(POLY): src_h = %d, dst_h = %d,row_acc_inc = %08x\n",
253			src_h, dst_h, row_acc_inc);
254	}
255
256
257	sc_reg0[0] = val;
258	sc_reg0[1] = row_acc_inc;
259	sc_reg0[2] = row_acc_offset;
260	sc_reg0[3] = row_acc_offset_b;
261
262	sc_reg0[4] = ((lin_acc_inc_u & CFG_LIN_ACC_INC_U_MASK) <<
263			CFG_LIN_ACC_INC_U_SHIFT) | (dst_w << CFG_TAR_W_SHIFT) |
264			(dst_h << CFG_TAR_H_SHIFT);
265
266	sc_reg0[5] = (src_w << CFG_SRC_W_SHIFT) | (src_h << CFG_SRC_H_SHIFT);
267
268	sc_reg0[6] = (row_acc_init_rav_b << CFG_ROW_ACC_INIT_RAV_B_SHIFT) |
269		(row_acc_init_rav << CFG_ROW_ACC_INIT_RAV_SHIFT);
270
271	*sc_reg9 = lin_acc_inc;
272
273	*sc_reg12 = col_acc_offset << CFG_COL_ACC_OFFSET_SHIFT;
274
275	*sc_reg13 = factor;
276
277	*sc_reg24 = (src_w << CFG_ORG_W_SHIFT) | (src_h << CFG_ORG_H_SHIFT);
278}
279
280struct sc_data *sc_create(struct platform_device *pdev)
281{
282	struct sc_data *sc;
283
284	dev_dbg(&pdev->dev, "sc_create\n");
285
286	sc = devm_kzalloc(&pdev->dev, sizeof(*sc), GFP_KERNEL);
287	if (!sc) {
288		dev_err(&pdev->dev, "couldn't alloc sc_data\n");
289		return ERR_PTR(-ENOMEM);
290	}
291
292	sc->pdev = pdev;
293
294	sc->res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "sc");
295	if (!sc->res) {
296		dev_err(&pdev->dev, "missing platform resources data\n");
297		return ERR_PTR(-ENODEV);
298	}
299
300	sc->base = devm_ioremap_resource(&pdev->dev, sc->res);
301	if (IS_ERR(sc->base)) {
302		dev_err(&pdev->dev, "failed to ioremap\n");
303		return ERR_CAST(sc->base);
304	}
305
306	return sc;
307}
308