fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-09 04:38:03 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

radv/video: Prefer visible VRAM for host visible context buffers
Some checks are pending
macOS-CI / macOS-CI (dri) (push) Waiting to run
Details
macOS-CI / macOS-CI (xlib) (push) Waiting to run
Details

Browse source 
In most cases this has little to no effect, but AV1 decode on VCN5
is up to 40% slower when the context buffer is in GTT.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35684>
This commit is contained in:
David Rosca 2025-06-23 12:15:59 +02:00 • committed by Marge Bot
parent f5498fda2a
commit 189bf9fe9a
1 changed files with 14 additions and 8 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

22
src/amd/vulkan/radv_video.c
View file

@ -1053,6 +1053,18 @@ radv_GetVideoSessionMemoryRequirementsKHR(VkDevice _device, VkVideoSessionKHR vi
const struct radv_physical_device *pdev = radv_device_physical(device);
uint32_t memory_type_bits = (1u << pdev->memory_properties.memoryTypeCount) - 1;
uint32_t memory_type_bits_visible = 0;
/* These buffers are only mapped once during session reset, for performance reasons
* we should prefer visible VRAM when available.
*/
for (unsigned i = 0; i < pdev->memory_properties.memoryTypeCount; i++) {
VkMemoryPropertyFlags flags = pdev->memory_properties.memoryTypes[i].propertyFlags;
if ((flags & VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT) &&
(flags & (VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT | VK_MEMORY_PROPERTY_HOST_CACHED_BIT)))
memory_type_bits_visible |= (1 << i);
}
if (vid->encode) {
return radv_video_get_encode_session_memory_requirements(device, vid, pMemoryRequirementsCount,
@ -1099,10 +1111,7 @@ radv_GetVideoSessionMemoryRequirementsKHR(VkDevice _device, VkVideoSessionKHR vi
m->memoryBindIndex = RADV_BIND_DECODER_CTX;
m->memoryRequirements.size = align(calc_ctx_size_vp9(pdev, vid), 4096);
m->memoryRequirements.alignment = 0;
m->memoryRequirements.memoryTypeBits = 0;
for (unsigned i = 0; i < pdev->memory_properties.memoryTypeCount; i++)
if (pdev->memory_properties.memoryTypes[i].propertyFlags & VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT)
m->memoryRequirements.memoryTypeBits |= (1 << i);
m->memoryRequirements.memoryTypeBits = memory_type_bits_visible;
}
if (vid->vk.max_dpb_slots == 0) {
@ -1120,10 +1129,7 @@ radv_GetVideoSessionMemoryRequirementsKHR(VkDevice _device, VkVideoSessionKHR vi
m->memoryBindIndex = RADV_BIND_DECODER_CTX;
m->memoryRequirements.size = align(calc_ctx_size_av1(device, vid), 4096);
m->memoryRequirements.alignment = 0;
m->memoryRequirements.memoryTypeBits = 0;
for (unsigned i = 0; i < pdev->memory_properties.memoryTypeCount; i++)
if (pdev->memory_properties.memoryTypes[i].propertyFlags & VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT)
m->memoryRequirements.memoryTypeBits |= (1 << i);
m->memoryRequirements.memoryTypeBits = memory_type_bits_visible;
}
}
return vk_outarray_status(&out);