linux_dsm_epyc7002/drivers/media/platform/ti-vpe/sc.c
Archit Taneja 0df20f9657 [media] v4l: ti-vpe: support loading of scaler coefficients
The SC block in VPE/VIP contains a SRAM within it. This internal memory
requires to be loaded with appropriate scaler coefficients from a contiguous
block of memory through VPDMA.

The horizontal and vertical scaler each require 2 sets of scaler coefficients
for luma and chroma scaling. The horizontal polyphase scaler requires
coefficients for a 32 phase and 8 tap filter. Similarly, the vertical scaler
requires coefficients for a 5 tap filter.

The choice of the scaler coefficients depends on the scaling ratio. Add
coefficient tables for different scaling ratios in sc_coeffs.h. In the case of
horizontal downscaling, we need to consider the change in ratio caused by
decimation performed by the horizontal scaler.

In order to load the scaler coefficients via VPDMA, a configuration descriptor
is used in block mode. The payload for the descriptor is the scaler coefficients
copied to memory. Coefficients for each phase have to be placed in memory in a
particular order understood by the scaler hardware.

The choice of the scaler coefficients, and the loading of the coefficients from
our tables to a contiguous buffer is managed by the functions
sc_set_hs_coefficients and sc_set_vs_coefficients.

The sc_data handle is now added with some parameters to describe the state of
the coefficients loaded in the SC block. 'loaded_coeff_h' and 'loaded_coeff_v'
hold the address of the last dma buffer which was used by VPDMA to copy
coefficients. This information can be used by a vpe mem-to-mem context to decide
whether it should load coefficients or not. 'hs_index' and 'vs_index' provide
some optimization by preventing loading of coefficients if the scaling ratio
didn't change between 2 contexts. 'load_coeff_h' and 'load_coeff_v' tell the
vpe/vip driver whether we need to load the coefficients through VPDMA or not.

Signed-off-by: Archit Taneja <archit@ti.com>
Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: Mauro Carvalho Chehab <m.chehab@samsung.com>
2014-01-07 06:54:48 -02:00

190 lines
4.0 KiB
C

/*
* Scaler library
*
* Copyright (c) 2013 Texas Instruments Inc.
*
* David Griego, <dagriego@biglakesoftware.com>
* Dale Farnsworth, <dale@farnsworth.org>
* Archit Taneja, <archit@ti.com>
*
* This program is free software; you can redistribute it and/or modify it
* under the terms of the GNU General Public License version 2 as published by
* the Free Software Foundation.
*/
#include <linux/err.h>
#include <linux/io.h>
#include <linux/platform_device.h>
#include <linux/slab.h>
#include "sc.h"
#include "sc_coeff.h"
void sc_set_regs_bypass(struct sc_data *sc, u32 *sc_reg0)
{
*sc_reg0 |= CFG_SC_BYPASS;
}
void sc_dump_regs(struct sc_data *sc)
{
struct device *dev = &sc->pdev->dev;
u32 read_reg(struct sc_data *sc, int offset)
{
return ioread32(sc->base + offset);
}
#define DUMPREG(r) dev_dbg(dev, "%-35s %08x\n", #r, read_reg(sc, CFG_##r))
DUMPREG(SC0);
DUMPREG(SC1);
DUMPREG(SC2);
DUMPREG(SC3);
DUMPREG(SC4);
DUMPREG(SC5);
DUMPREG(SC6);
DUMPREG(SC8);
DUMPREG(SC9);
DUMPREG(SC10);
DUMPREG(SC11);
DUMPREG(SC12);
DUMPREG(SC13);
DUMPREG(SC17);
DUMPREG(SC18);
DUMPREG(SC19);
DUMPREG(SC20);
DUMPREG(SC21);
DUMPREG(SC22);
DUMPREG(SC23);
DUMPREG(SC24);
DUMPREG(SC25);
#undef DUMPREG
}
/*
* set the horizontal scaler coefficients according to the ratio of output to
* input widths, after accounting for up to two levels of decimation
*/
void sc_set_hs_coeffs(struct sc_data *sc, void *addr, unsigned int src_w,
unsigned int dst_w)
{
int sixteenths;
int idx;
int i, j;
u16 *coeff_h = addr;
const u16 *cp;
if (dst_w > src_w) {
idx = HS_UP_SCALE;
} else {
if ((dst_w << 1) < src_w)
dst_w <<= 1; /* first level decimation */
if ((dst_w << 1) < src_w)
dst_w <<= 1; /* second level decimation */
if (dst_w == src_w) {
idx = HS_LE_16_16_SCALE;
} else {
sixteenths = (dst_w << 4) / src_w;
if (sixteenths < 8)
sixteenths = 8;
idx = HS_LT_9_16_SCALE + sixteenths - 8;
}
}
if (idx == sc->hs_index)
return;
cp = scaler_hs_coeffs[idx];
for (i = 0; i < SC_NUM_PHASES * 2; i++) {
for (j = 0; j < SC_H_NUM_TAPS; j++)
*coeff_h++ = *cp++;
/*
* for each phase, the scaler expects space for 8 coefficients
* in it's memory. For the horizontal scaler, we copy the first
* 7 coefficients and skip the last slot to move to the next
* row to hold coefficients for the next phase
*/
coeff_h += SC_NUM_TAPS_MEM_ALIGN - SC_H_NUM_TAPS;
}
sc->hs_index = idx;
sc->load_coeff_h = true;
}
/*
* set the vertical scaler coefficients according to the ratio of output to
* input heights
*/
void sc_set_vs_coeffs(struct sc_data *sc, void *addr, unsigned int src_h,
unsigned int dst_h)
{
int sixteenths;
int idx;
int i, j;
u16 *coeff_v = addr;
const u16 *cp;
if (dst_h > src_h) {
idx = VS_UP_SCALE;
} else if (dst_h == src_h) {
idx = VS_1_TO_1_SCALE;
} else {
sixteenths = (dst_h << 4) / src_h;
if (sixteenths < 8)
sixteenths = 8;
idx = VS_LT_9_16_SCALE + sixteenths - 8;
}
if (idx == sc->vs_index)
return;
cp = scaler_vs_coeffs[idx];
for (i = 0; i < SC_NUM_PHASES * 2; i++) {
for (j = 0; j < SC_V_NUM_TAPS; j++)
*coeff_v++ = *cp++;
/*
* for the vertical scaler, we copy the first 5 coefficients and
* skip the last 3 slots to move to the next row to hold
* coefficients for the next phase
*/
coeff_v += SC_NUM_TAPS_MEM_ALIGN - SC_V_NUM_TAPS;
}
sc->vs_index = idx;
sc->load_coeff_v = true;
}
struct sc_data *sc_create(struct platform_device *pdev)
{
struct sc_data *sc;
dev_dbg(&pdev->dev, "sc_create\n");
sc = devm_kzalloc(&pdev->dev, sizeof(*sc), GFP_KERNEL);
if (!sc) {
dev_err(&pdev->dev, "couldn't alloc sc_data\n");
return ERR_PTR(-ENOMEM);
}
sc->pdev = pdev;
sc->res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "sc");
if (!sc->res) {
dev_err(&pdev->dev, "missing platform resources data\n");
return ERR_PTR(-ENODEV);
}
sc->base = devm_ioremap_resource(&pdev->dev, sc->res);
if (!sc->base) {
dev_err(&pdev->dev, "failed to ioremap\n");
return ERR_PTR(-ENOMEM);
}
return sc;
}