mirror of
https://gitlab.freedesktop.org/mesa/mesa.git
synced 2026-05-05 09:38:07 +02:00
asahi: Dynamically configure tile size
We need to shrink the tile size when using small images (including due to mipmapping) or when using large block sizes. Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io> Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14903>
This commit is contained in:
parent
d103d64df6
commit
2028873ef6
3 changed files with 50 additions and 25 deletions
|
|
@ -29,7 +29,7 @@
|
|||
#include "util/macros.h"
|
||||
#include "tiling.h"
|
||||
|
||||
/* Z-order with 64x64 tiles:
|
||||
/* Z-order with square tiles, at most 64x64:
|
||||
*
|
||||
* [y5][x5][y4][x4][y3][x3][y2][x2][y1][x1][y0][x0]
|
||||
*
|
||||
|
|
@ -54,48 +54,46 @@
|
|||
* applying the two's complement identity, we are left with (X - mask) & mask
|
||||
*/
|
||||
|
||||
#define TILE_WIDTH 64
|
||||
#define TILE_HEIGHT 64
|
||||
#define TILE_SHIFT 6
|
||||
#define TILE_MASK ((1 << TILE_SHIFT) - 1)
|
||||
|
||||
/* mask of bits used for X coordinate in a tile */
|
||||
#define SPACE_MASK 0x555 // 0b010101010101
|
||||
|
||||
static uint32_t
|
||||
agx_space_bits(unsigned x)
|
||||
{
|
||||
assert(x < TILE_WIDTH);
|
||||
assert(x < 64);
|
||||
return ((x & 1) << 0) | ((x & 2) << 1) | ((x & 4) << 2) |
|
||||
((x & 8) << 3) | ((x & 16) << 4) | ((x & 32) << 5);
|
||||
}
|
||||
|
||||
#define TILED_UNALIGNED_TYPE(pixel_t, is_store) { \
|
||||
unsigned tiles_per_row = (width + TILE_WIDTH - 1) >> TILE_SHIFT;\
|
||||
unsigned y_offs = agx_space_bits(sy & TILE_MASK) << 1;\
|
||||
unsigned x_offs_start = agx_space_bits(sx & TILE_MASK);\
|
||||
#define TILED_UNALIGNED_TYPE(pixel_t, is_store, tile_shift) { \
|
||||
unsigned tile_size = (1 << tile_shift);\
|
||||
unsigned pixels_per_tile = tile_size * tile_size;\
|
||||
unsigned tiles_per_row = (width + tile_size - 1) >> tile_shift;\
|
||||
unsigned y_offs = agx_space_bits(sy & (tile_size - 1)) << 1;\
|
||||
unsigned x_offs_start = agx_space_bits(sx & (tile_size - 1));\
|
||||
unsigned space_mask = SPACE_MASK & (pixels_per_tile - 1);\
|
||||
\
|
||||
for (unsigned y = sy; y < smaxy; ++y) {\
|
||||
unsigned tile_y = (y >> TILE_SHIFT);\
|
||||
unsigned tile_y = (y >> tile_shift);\
|
||||
unsigned tile_row = tile_y * tiles_per_row;\
|
||||
unsigned x_offs = x_offs_start;\
|
||||
\
|
||||
pixel_t *linear_row = linear;\
|
||||
\
|
||||
for (unsigned x = sx; x < smaxx; ++x) {\
|
||||
unsigned tile_x = (x >> TILE_SHIFT);\
|
||||
unsigned tile_x = (x >> tile_shift);\
|
||||
unsigned tile_idx = (tile_row + tile_x);\
|
||||
unsigned tile_base = tile_idx * (TILE_WIDTH * TILE_HEIGHT);\
|
||||
unsigned tile_base = tile_idx * pixels_per_tile;\
|
||||
\
|
||||
pixel_t *ptiled = &tiled[tile_base + y_offs + x_offs];\
|
||||
pixel_t *plinear = (linear_row++);\
|
||||
pixel_t *outp = (pixel_t *) (is_store ? ptiled : plinear); \
|
||||
pixel_t *inp = (pixel_t *) (is_store ? plinear : ptiled); \
|
||||
*outp = *inp;\
|
||||
x_offs = (x_offs - SPACE_MASK) & SPACE_MASK;\
|
||||
x_offs = (x_offs - space_mask) & space_mask;\
|
||||
}\
|
||||
\
|
||||
y_offs = (((y_offs >> 1) - SPACE_MASK) & SPACE_MASK) << 1;\
|
||||
y_offs = (((y_offs >> 1) - space_mask) & space_mask) << 1;\
|
||||
linear += linear_pitch;\
|
||||
}\
|
||||
}
|
||||
|
|
@ -103,25 +101,26 @@ agx_space_bits(unsigned x)
|
|||
void
|
||||
agx_detile(void *_tiled, void *_linear,
|
||||
unsigned width, unsigned bpp, unsigned linear_pitch,
|
||||
unsigned sx, unsigned sy, unsigned smaxx, unsigned smaxy)
|
||||
unsigned sx, unsigned sy, unsigned smaxx, unsigned smaxy, unsigned tile_shift)
|
||||
{
|
||||
/* TODO: parametrize with macro magic */
|
||||
assert(bpp == 32);
|
||||
|
||||
uint32_t *linear = _linear;
|
||||
uint32_t *tiled = _tiled;
|
||||
TILED_UNALIGNED_TYPE(uint32_t, false);
|
||||
TILED_UNALIGNED_TYPE(uint32_t, false, tile_shift);
|
||||
}
|
||||
|
||||
void
|
||||
agx_tile(void *_tiled, void *_linear,
|
||||
unsigned width, unsigned bpp, unsigned linear_pitch,
|
||||
unsigned sx, unsigned sy, unsigned smaxx, unsigned smaxy)
|
||||
unsigned sx, unsigned sy, unsigned smaxx, unsigned smaxy,
|
||||
unsigned tile_shift)
|
||||
{
|
||||
/* TODO: parametrize with macro magic */
|
||||
assert(bpp == 32);
|
||||
|
||||
uint32_t *linear = _linear;
|
||||
uint32_t *tiled = _tiled;
|
||||
TILED_UNALIGNED_TYPE(uint32_t, true);
|
||||
TILED_UNALIGNED_TYPE(uint32_t, true, tile_shift);
|
||||
}
|
||||
|
|
|
|||
|
|
@ -24,12 +24,38 @@
|
|||
#ifndef __AGX_TILING_H
|
||||
#define __AGX_TILING_H
|
||||
|
||||
#include "util/u_math.h"
|
||||
|
||||
void agx_detile(void *tiled, void *linear,
|
||||
unsigned width, unsigned bpp, unsigned linear_pitch,
|
||||
unsigned sx, unsigned sy, unsigned smaxx, unsigned smaxy);
|
||||
unsigned sx, unsigned sy, unsigned smaxx, unsigned smaxy, unsigned tile_shift);
|
||||
|
||||
void agx_tile(void *tiled, void *linear,
|
||||
unsigned width, unsigned bpp, unsigned linear_pitch,
|
||||
unsigned sx, unsigned sy, unsigned smaxx, unsigned smaxy);
|
||||
unsigned sx, unsigned sy, unsigned smaxx, unsigned smaxy, unsigned tile_shift);
|
||||
|
||||
/* Select effective tile size given texture dimensions */
|
||||
static inline unsigned
|
||||
agx_select_tile_shift(unsigned width, unsigned height, unsigned level, unsigned blocksize)
|
||||
{
|
||||
/* Calculate effective width/height due to mipmapping */
|
||||
width = u_minify(width, level);
|
||||
height = u_minify(height, level);
|
||||
|
||||
/* Select the largest square power-of-two tile fitting in the image */
|
||||
unsigned shift = util_logbase2_ceil(MIN3(width, height, 64));
|
||||
|
||||
/* Shrink based on block size */
|
||||
if (blocksize > 4)
|
||||
return MAX2(shift - 1, 0);
|
||||
else
|
||||
return shift;
|
||||
}
|
||||
|
||||
static inline unsigned
|
||||
agx_select_tile_size(unsigned width, unsigned height, unsigned level, unsigned blocksize)
|
||||
{
|
||||
return 1 << agx_select_tile_shift(width, height, level, blocksize);
|
||||
}
|
||||
|
||||
#endif
|
||||
|
|
|
|||
|
|
@ -312,7 +312,7 @@ agx_transfer_map(struct pipe_context *pctx,
|
|||
agx_detile(map, dst,
|
||||
u_minify(resource->width0, level), bytes_per_pixel * 8,
|
||||
transfer->base.stride / bytes_per_pixel,
|
||||
box->x, box->y, box->x + box->width, box->y + box->height);
|
||||
box->x, box->y, box->x + box->width, box->y + box->height, 6);
|
||||
}
|
||||
}
|
||||
|
||||
|
|
@ -365,7 +365,7 @@ agx_transfer_unmap(struct pipe_context *pctx,
|
|||
transfer->stride / bytes_per_pixel,
|
||||
transfer->box.x, transfer->box.y,
|
||||
transfer->box.x + transfer->box.width,
|
||||
transfer->box.y + transfer->box.height);
|
||||
transfer->box.y + transfer->box.height, 6);
|
||||
}
|
||||
}
|
||||
|
||||
|
|
@ -660,7 +660,7 @@ agx_flush_frontbuffer(struct pipe_screen *_screen,
|
|||
if (rsrc->modifier == DRM_FORMAT_MOD_APPLE_64X64_MORTON_ORDER) {
|
||||
agx_detile(rsrc->bo->ptr.cpu, map,
|
||||
rsrc->base.width0, 32, rsrc->dt_stride / 4,
|
||||
0, 0, rsrc->base.width0, rsrc->base.height0);
|
||||
0, 0, rsrc->base.width0, rsrc->base.height0, 6);
|
||||
} else {
|
||||
memcpy(map, rsrc->bo->ptr.cpu, rsrc->dt_stride * rsrc->base.height0);
|
||||
}
|
||||
|
|
|
|||
Loading…
Add table
Reference in a new issue