fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-08 02:38:04 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 4

panvk: enable 8x and 16x msaa when supported

Browse source 
This uses the helpers from the previous patch to calculate how many
attachments and MRT buffers we have space for.

In the case where we can support more MSAA samples for smaller formats,
we also add support for that.

The flaking test seems to be due to a CTS issue, see this ticket for
details:

https://gitlab.khronos.org/Tracker/vk-gl-cts/-/issues/5651

Reviewed-by: Lars-Ivar Hesselberg Simonsen <lars-ivar.simonsen@arm.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/33925>
This commit is contained in:
Erik Faye-Lund 2025-03-06 16:03:44 +01:00 • committed by Marge Bot
parent 329568b5eb
commit 1729409141
3 changed files with 49 additions and 6 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

1
docs/relnotes/new_features.txt
View file

@ -34,3 +34,4 @@ VK_KHR_float_controls on panvk
VK_KHR_float_controls2 on panvk
storageBuffer8BitAccess on panvk
uniformAndStorageBuffer8BitAccess on panvk
MSAA with 8 and 16 sample counts on panvk

4
src/panfrost/ci/panfrost-g610-flakes.txt
View file

@ -262,3 +262,7 @@ dEQP-VK.wsi.*.swapchain.simulate_oom.*
# With linux kernel 6.13-rc4 uprev
dEQP-GLES31.functional.copy_image.compressed.viewclass_astc_8x6_rgba.rgba_astc_8x6_khr_srgb8_alpha8_astc_8x6_khr.texture3d_to_cubemap
# CTS issue: https://gitlab.khronos.org/Tracker/vk-gl-cts/-/issues/5651
dEQP-VK.renderpass2.depth_stencil_resolve.image_2d_32_32.samples_16.d32_sfloat_s8_uint_separate_layouts.depth_zero_stencil_min_testing_stencil_samplemask
dEQP-VK.renderpass2.depth_stencil_resolve.image_2d_32_32.samples_16.d32_sfloat_s8_uint.depth_zero_stencil_min_testing_stencil_samplemask

50
src/panfrost/vulkan/panvk_physical_device.c
View file

@ -567,20 +567,43 @@ get_conformance_version(unsigned arch)
return (VkConformanceVersion){0, 0, 0, 0};
}
static VkSampleCountFlags
get_sample_counts(unsigned arch, unsigned max_tib_size,
unsigned max_cbuf_atts, unsigned format_size)
{
VkSampleCountFlags sample_counts =
VK_SAMPLE_COUNT_1_BIT | VK_SAMPLE_COUNT_4_BIT;
unsigned max_msaa =
pan_get_max_msaa(arch, max_tib_size, max_cbuf_atts, format_size);
assert(max_msaa >= 4);
if (max_msaa >= 8)
sample_counts |= VK_SAMPLE_COUNT_8_BIT;
if (max_msaa >= 16)
sample_counts |= VK_SAMPLE_COUNT_16_BIT;
return sample_counts;
}
static void
get_device_properties(const struct panvk_instance *instance,
const struct panvk_physical_device *device,
struct vk_properties *properties)
{
/* HW supports MSAA 4, 8 and 16, but we limit ourselves to MSAA 4 for now. */
unsigned arch = pan_arch(device->kmod.props.gpu_prod_id);
unsigned max_tib_size = pan_get_max_tib_size(arch, device->model);
const unsigned max_cbuf_format = 16; /* R32G32B32A32 */
unsigned max_cbuf_atts = pan_get_max_cbufs(arch, max_tib_size);
VkSampleCountFlags sample_counts =
VK_SAMPLE_COUNT_1_BIT | VK_SAMPLE_COUNT_4_BIT;
get_sample_counts(arch, max_tib_size, max_cbuf_atts, max_cbuf_format);
uint64_t os_page_size = 4096;
os_get_page_size(&os_page_size);
unsigned arch = pan_arch(device->kmod.props.gpu_prod_id);
/* Ensure that the max threads count per workgroup is valid for Bifrost */
assert(arch > 8 || device->kmod.props.max_threads_per_wg <= 1024);
@ -784,7 +807,7 @@ get_device_properties(const struct panvk_instance *instance,
.framebufferDepthSampleCounts = sample_counts,
.framebufferStencilSampleCounts = sample_counts,
.framebufferNoAttachmentsSampleCounts = sample_counts,
.maxColorAttachments = 8,
.maxColorAttachments = max_cbuf_atts,
.sampledImageColorSampleCounts = sample_counts,
.sampledImageIntegerSampleCounts = VK_SAMPLE_COUNT_1_BIT,
.sampledImageDepthSampleCounts = sample_counts,
@ -1431,6 +1454,20 @@ get_image_format_features(struct panvk_physical_device *physical_device,
return features;
}
static VkFormatFeatureFlags
get_image_format_sample_counts(struct panvk_physical_device *physical_device,
VkFormat format)
{
unsigned arch = pan_arch(physical_device->kmod.props.gpu_prod_id);
unsigned max_tib_size = pan_get_max_tib_size(arch, physical_device->model);
unsigned max_cbuf_atts = pan_get_max_cbufs(arch, max_tib_size);
assert(!vk_format_is_compressed(format));
unsigned format_size = vk_format_get_blocksize(format);
return get_sample_counts(arch, max_tib_size, max_cbuf_atts, format_size);
}
static VkFormatFeatureFlags
get_buffer_format_features(struct panvk_physical_device *physical_device,
VkFormat format)
@ -1604,7 +1641,8 @@ get_image_format_properties(struct panvk_physical_device *physical_device,
VK_FORMAT_FEATURE_DEPTH_STENCIL_ATTACHMENT_BIT)) &&
!(info->flags & VK_IMAGE_CREATE_CUBE_COMPATIBLE_BIT) &&
!(all_usage & VK_IMAGE_USAGE_STORAGE_BIT)) {
sampleCounts |= VK_SAMPLE_COUNT_4_BIT;
sampleCounts |=
get_image_format_sample_counts(physical_device, info->format);
}
/* From the Vulkan 1.2.199 spec: