mirror of
https://github.com/AuxXxilium/linux_dsm_epyc7002.git
synced 2024-12-22 22:11:22 +07:00
0d6fa53fd8
Modified Tiler placement to utilize bitmaps for bookkeeping and all placement algorithms. This resulted in a substantial savings in time for all Tiler reservation and free operations. Typical savings are in the range of 28% decrease in time taken with larger buffers showing a 80%+ decrease. Signed-off-by: Andy Gross <andy.gross@ti.com> Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
262 lines
6.2 KiB
C
262 lines
6.2 KiB
C
/*
|
|
* tcm-sita.c
|
|
*
|
|
* SImple Tiler Allocator (SiTA): 2D and 1D allocation(reservation) algorithm
|
|
*
|
|
* Authors: Ravi Ramachandra <r.ramachandra@ti.com>,
|
|
* Lajos Molnar <molnar@ti.com>
|
|
* Andy Gross <andy.gross@ti.com>
|
|
*
|
|
* Copyright (C) 2012 Texas Instruments, Inc.
|
|
*
|
|
* This package 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.
|
|
*
|
|
* THIS PACKAGE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR
|
|
* IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED
|
|
* WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR A PARTICULAR PURPOSE.
|
|
*
|
|
*/
|
|
#include <linux/init.h>
|
|
#include <linux/module.h>
|
|
#include <linux/errno.h>
|
|
#include <linux/sched.h>
|
|
#include <linux/wait.h>
|
|
#include <linux/bitmap.h>
|
|
#include <linux/slab.h>
|
|
#include "tcm.h"
|
|
|
|
static unsigned long mask[8];
|
|
/*
|
|
* pos position in bitmap
|
|
* w width in slots
|
|
* h height in slots
|
|
* map ptr to bitmap
|
|
* stride slots in a row
|
|
*/
|
|
static void free_slots(unsigned long pos, uint16_t w, uint16_t h,
|
|
unsigned long *map, uint16_t stride)
|
|
{
|
|
int i;
|
|
|
|
for (i = 0; i < h; i++, pos += stride)
|
|
bitmap_clear(map, pos, w);
|
|
}
|
|
|
|
/*
|
|
* w width in slots
|
|
* pos ptr to position
|
|
* map ptr to bitmap
|
|
* num_bits number of bits in bitmap
|
|
*/
|
|
static int r2l_b2t_1d(uint16_t w, unsigned long *pos, unsigned long *map,
|
|
size_t num_bits)
|
|
{
|
|
unsigned long search_count = 0;
|
|
unsigned long bit;
|
|
bool area_found = false;
|
|
|
|
*pos = num_bits - w;
|
|
|
|
while (search_count < num_bits) {
|
|
bit = find_next_bit(map, num_bits, *pos);
|
|
|
|
if (bit - *pos >= w) {
|
|
/* found a long enough free area */
|
|
bitmap_set(map, *pos, w);
|
|
area_found = true;
|
|
break;
|
|
}
|
|
|
|
search_count = num_bits - bit + w;
|
|
*pos = bit - w;
|
|
}
|
|
|
|
return (area_found) ? 0 : -ENOMEM;
|
|
}
|
|
|
|
/*
|
|
* w = width in slots
|
|
* h = height in slots
|
|
* a = align in slots (mask, 2^n-1, 0 is unaligned)
|
|
* offset = offset in bytes from 4KiB
|
|
* pos = position in bitmap for buffer
|
|
* map = bitmap ptr
|
|
* num_bits = size of bitmap
|
|
* stride = bits in one row of container
|
|
*/
|
|
static int l2r_t2b(uint16_t w, uint16_t h, uint16_t a, int16_t offset,
|
|
unsigned long *pos, unsigned long slot_bytes,
|
|
unsigned long *map, size_t num_bits, size_t slot_stride)
|
|
{
|
|
int i;
|
|
unsigned long index;
|
|
bool area_free;
|
|
unsigned long slots_per_band = PAGE_SIZE / slot_bytes;
|
|
unsigned long bit_offset = (offset > 0) ? offset / slot_bytes : 0;
|
|
unsigned long curr_bit = bit_offset;
|
|
|
|
/* reset alignment to 1 if we are matching a specific offset */
|
|
/* adjust alignment - 1 to get to the format expected in bitmaps */
|
|
a = (offset > 0) ? 0 : a - 1;
|
|
|
|
/* FIXME Return error if slots_per_band > stride */
|
|
|
|
while (curr_bit < num_bits) {
|
|
*pos = bitmap_find_next_zero_area(map, num_bits, curr_bit, w,
|
|
a);
|
|
|
|
/* skip forward if we are not at right offset */
|
|
if (bit_offset > 0 && (*pos % slots_per_band != bit_offset)) {
|
|
curr_bit = ALIGN(*pos, slots_per_band) + bit_offset;
|
|
continue;
|
|
}
|
|
|
|
/* skip forward to next row if we overlap end of row */
|
|
if ((*pos % slot_stride) + w > slot_stride) {
|
|
curr_bit = ALIGN(*pos, slot_stride) + bit_offset;
|
|
continue;
|
|
}
|
|
|
|
/* TODO: Handle overlapping 4K boundaries */
|
|
|
|
/* break out of look if we will go past end of container */
|
|
if ((*pos + slot_stride * h) > num_bits)
|
|
break;
|
|
|
|
/* generate mask that represents out matching pattern */
|
|
bitmap_clear(mask, 0, slot_stride);
|
|
bitmap_set(mask, (*pos % BITS_PER_LONG), w);
|
|
|
|
/* assume the area is free until we find an overlap */
|
|
area_free = true;
|
|
|
|
/* check subsequent rows to see if complete area is free */
|
|
for (i = 1; i < h; i++) {
|
|
index = *pos / BITS_PER_LONG + i * 8;
|
|
if (bitmap_intersects(&map[index], mask,
|
|
(*pos % BITS_PER_LONG) + w)) {
|
|
area_free = false;
|
|
break;
|
|
}
|
|
}
|
|
|
|
if (area_free)
|
|
break;
|
|
|
|
/* go forward past this match */
|
|
if (bit_offset > 0)
|
|
curr_bit = ALIGN(*pos, slots_per_band) + bit_offset;
|
|
else
|
|
curr_bit = *pos + a + 1;
|
|
}
|
|
|
|
if (area_free) {
|
|
/* set area as in-use. iterate over rows */
|
|
for (i = 0, index = *pos; i < h; i++, index += slot_stride)
|
|
bitmap_set(map, index, w);
|
|
}
|
|
|
|
return (area_free) ? 0 : -ENOMEM;
|
|
}
|
|
|
|
static s32 sita_reserve_1d(struct tcm *tcm, u32 num_slots,
|
|
struct tcm_area *area)
|
|
{
|
|
unsigned long pos;
|
|
int ret;
|
|
|
|
spin_lock(&(tcm->lock));
|
|
ret = r2l_b2t_1d(num_slots, &pos, tcm->bitmap, tcm->map_size);
|
|
if (!ret) {
|
|
area->p0.x = pos % tcm->width;
|
|
area->p0.y = pos / tcm->width;
|
|
area->p1.x = (pos + num_slots - 1) % tcm->width;
|
|
area->p1.y = (pos + num_slots - 1) / tcm->width;
|
|
}
|
|
spin_unlock(&(tcm->lock));
|
|
|
|
return ret;
|
|
}
|
|
|
|
static s32 sita_reserve_2d(struct tcm *tcm, u16 h, u16 w, u16 align,
|
|
int16_t offset, uint16_t slot_bytes,
|
|
struct tcm_area *area)
|
|
{
|
|
unsigned long pos;
|
|
int ret;
|
|
|
|
spin_lock(&(tcm->lock));
|
|
ret = l2r_t2b(w, h, align, offset, &pos, slot_bytes, tcm->bitmap,
|
|
tcm->map_size, tcm->width);
|
|
|
|
if (!ret) {
|
|
area->p0.x = pos % tcm->width;
|
|
area->p0.y = pos / tcm->width;
|
|
area->p1.x = area->p0.x + w - 1;
|
|
area->p1.y = area->p0.y + h - 1;
|
|
}
|
|
spin_unlock(&(tcm->lock));
|
|
|
|
return ret;
|
|
}
|
|
|
|
static void sita_deinit(struct tcm *tcm)
|
|
{
|
|
kfree(tcm);
|
|
}
|
|
|
|
static s32 sita_free(struct tcm *tcm, struct tcm_area *area)
|
|
{
|
|
unsigned long pos;
|
|
uint16_t w, h;
|
|
|
|
pos = area->p0.x + area->p0.y * tcm->width;
|
|
if (area->is2d) {
|
|
w = area->p1.x - area->p0.x + 1;
|
|
h = area->p1.y - area->p0.y + 1;
|
|
} else {
|
|
w = area->p1.x + area->p1.y * tcm->width - pos + 1;
|
|
h = 1;
|
|
}
|
|
|
|
spin_lock(&(tcm->lock));
|
|
free_slots(pos, w, h, tcm->bitmap, tcm->width);
|
|
spin_unlock(&(tcm->lock));
|
|
return 0;
|
|
}
|
|
|
|
struct tcm *sita_init(u16 width, u16 height)
|
|
{
|
|
struct tcm *tcm;
|
|
size_t map_size = BITS_TO_LONGS(width*height) * sizeof(unsigned long);
|
|
|
|
if (width == 0 || height == 0)
|
|
return NULL;
|
|
|
|
tcm = kzalloc(sizeof(*tcm) + map_size, GFP_KERNEL);
|
|
if (!tcm)
|
|
goto error;
|
|
|
|
/* Updating the pointers to SiTA implementation APIs */
|
|
tcm->height = height;
|
|
tcm->width = width;
|
|
tcm->reserve_2d = sita_reserve_2d;
|
|
tcm->reserve_1d = sita_reserve_1d;
|
|
tcm->free = sita_free;
|
|
tcm->deinit = sita_deinit;
|
|
|
|
spin_lock_init(&tcm->lock);
|
|
tcm->bitmap = (unsigned long *)(tcm + 1);
|
|
bitmap_clear(tcm->bitmap, 0, width*height);
|
|
|
|
tcm->map_size = width*height;
|
|
|
|
return tcm;
|
|
|
|
error:
|
|
kfree(tcm);
|
|
return NULL;
|
|
}
|