Add support for tiled textures #61
Conversation
Signed-off-by: Rohan Gard <[email protected]>
Signed-off-by: Vasily Khoruzhick <[email protected]>
Signed-off-by: Vasily Khoruzhick <[email protected]>
Fixes gbm_surface_create_with_modifiers() failure in kmscube Signed-off-by: Vasily Khoruzhick <[email protected]>
| wb[0].pitch = res->stride / 8; | ||
| if (res->tiled) { | ||
| wb[0].pixel_layout = 0x2; | ||
| wb[0].pitch = align(ctx->framebuffer.width, 16) / 16; |
| res->base = *templat; | ||
| res->base.screen = pscreen; | ||
| pipe_reference_init(&res->base.reference, 1); | ||
|
|
||
| /* TODO: mipmap */ | ||
| pres = &res->base; | ||
| res->stride = util_format_get_stride(pres->format, width); | ||
| res->tiled = should_tile; | ||
| res->stride = util_format_get_stride(pres->format, should_tile ? align(width, 16) : width); |
There was a problem hiding this comment.
I'd like add this:
if (should_tile) {
width = align(width, 16);
height = align(height, 16);
}
There was a problem hiding this comment.
No, that is not correct. Width and height don't have to be multiply of 16. But stride has to.
There was a problem hiding this comment.
width & height passed in is already not the logical one but the real one used to alloc buffer.
| ptrans->box.y = box_y1; | ||
| ptrans->box.width = box_x2 - box_x1; | ||
| ptrans->box.height = box_y2 - box_y1; | ||
| trans->map = malloc(ptrans->stride * ptrans->box.height * ptrans->box.depth); |
There was a problem hiding this comment.
Can trans->map only contain the box size data? Currently the stride is the whole texture stride. Besides I'd like to use "staging" instead of "map" for trans. Maybe u_transfer_helper_transfer_map is a reference.
There was a problem hiding this comment.
Not really, but we can limit height. I'll change map to staging, it sounds reasonable.
There was a problem hiding this comment.
Why not? Seems u_transfer_helper_transfer_map does like this.
There was a problem hiding this comment.
OK, I'll look into u_transfer_helper_transfer_map implementation.
| struct pipe_resource *pres; | ||
|
|
||
| if (trans->map) { | ||
| pres = &res->base; |
There was a problem hiding this comment.
Yes, we certainly can, will fix.
| /* | ||
| * Copyright (c) 2011-2013 Luc Verhaegen <[email protected]> | ||
| * Copyright (c) 2018 Alyssa Rosenzweig <[email protected]> | ||
| * Copyright (c) 2018 Lima Project |
There was a problem hiding this comment.
You may sign your name as the feedback in drm-lima RFC "Lima Project" is not a real org.
| struct pipe_resource tmpl = *templat; | ||
|
|
||
| /* | ||
| * We currently assume that all buffers allocated through this interface |
There was a problem hiding this comment.
Why we assume this? Can't we just return NULL for not support or add a modifier for "tiled"?
There was a problem hiding this comment.
Just mimicking etnaviv behavior. If we don't do it here kmscube doesn't work.
This is initial support for tiled textures - tiling/untiling routines aren't optimized yet.
lima-18.1 branch can't compile fragment shader of "kmscube -M rgba" due to ppir regalloc failure, I'm using this patch as a workaround for testing: anarsoul/mesa-lima@14e0665
Tested with kmscube -M rgba from my repo (it uses 500x500 texture) and gbm-surface-fbo from your gfx repo with 260x260 fbo.