fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-07 04:58:05 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 4

pvr: Fix allocating the required scratch buffer space for tile buffers

Browse source 
When calculating the dwords per pixel the output registers should
always be taken into account in addition to the number of tile buffers.

Fixes incorrect scratch buffer space calculation when both output
registers and tile buffers are emitted by a render.

Partial fix for:
dEQP-VK.renderpass.*.attachment_allocation.input_output.71

Fixes: 3457f8083a ("pvr: Acquire scratch buffer on framebuffer creation.")
Signed-off-by: Nick Hamilton <nick.hamilton@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
(cherry picked from commit df445dc9b9)

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40092>
This commit is contained in:
Jarred Davies 2026-02-11 09:04:03 +00:00 • committed by Eric Engestrom
parent 9ad2e48819
commit fb1ba13c57
2 changed files with 13 additions and 8 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
.pick_status.json
View file

@ -524,7 +524,7 @@
"description": "pvr: Fix allocating the required scratch buffer space for tile buffers",
"nominated": true,
"nomination_type": 2,
"resolution": 0,
"resolution": 1,
"main_sha": null,
"because_sha": "3457f8083a656524ffe7ce572a7d09b0c7279cf3",
"notes": null

19
src/imagination/vulkan/pvr_arch_spm.c
View file

@ -46,20 +46,25 @@ uint64_t pvr_arch_spm_scratch_buffer_calc_required_size(
uint64_t dwords_per_pixel;
uint64_t buffer_size;
/* If we're allocating an SPM scratch buffer we'll have a minimum of 1 output
* reg and/or tile_buffer.
uint32_t max_used_tile_buffers = 0;
/* The maximum number of output registers / tile buffer dwords it ever
* encountered when allocating space in the output registers and any tile
* buffers.
*/
uint32_t nr_tile_buffers = 1;
uint32_t nr_output_regs = 1;
uint32_t max_used_dwords = 1;
for (uint32_t i = 0; i < render_count; i++) {
const struct pvr_renderpass_hwsetup_render *hw_render = &renders[i];
nr_tile_buffers = MAX2(nr_tile_buffers, hw_render->tile_buffers_count);
nr_output_regs = MAX2(nr_output_regs, hw_render->output_regs_count);
max_used_tile_buffers = MAX2(max_used_tile_buffers, hw_render->tile_buffers_count);
max_used_dwords = MAX2(max_used_dwords, hw_render->output_regs_count);
}
dwords_per_pixel = (uint64_t)sample_count * nr_output_regs * nr_tile_buffers;
/* Add one to the number of tile buffers to account for the output registers,
* which are always used
*/
dwords_per_pixel = (uint64_t)sample_count * max_used_dwords * (1 + max_used_tile_buffers);
buffer_size = ALIGN_POT((uint64_t)framebuffer_width,
ROGUE_CR_PBE_WORD0_MRT0_LINESTRIDE_ALIGNMENT);