anv: Set memory types supported by Xe KMD

Due the lack of APIs to set mmap modes, Xe KMD can't support the same
memory types as i915.
So here adding a i915 and Xe function to set memory types supported
by each KMD.

Iris function iris_xe_bo_flags_to_mmap_mode() has a table with all the
mmaps modes of each type of placement.

Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22906>
This commit is contained in:
José Roberto de Souza 2023-04-27 14:14:22 -07:00 committed by Marge Bot
parent ffbbf23ef8
commit 4d4b0dfdb8
5 changed files with 151 additions and 79 deletions

View file

@ -871,7 +871,6 @@ anv_update_meminfo(struct anv_physical_device *device, int fd)
device->vram_non_mappable.available = devinfo->mem.vram.unmappable.free;
}
static VkResult
anv_physical_device_init_heaps(struct anv_physical_device *device, int fd)
{
@ -913,65 +912,6 @@ anv_physical_device_init_heaps(struct anv_physical_device *device, int fd)
.is_local_mem = true,
};
}
device->memory.type_count = 3;
device->memory.types[0] = (struct anv_memory_type) {
.propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT,
.heapIndex = 0,
};
device->memory.types[1] = (struct anv_memory_type) {
.propertyFlags = VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
VK_MEMORY_PROPERTY_HOST_COHERENT_BIT |
VK_MEMORY_PROPERTY_HOST_CACHED_BIT,
.heapIndex = 1,
};
device->memory.types[2] = (struct anv_memory_type) {
.propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
VK_MEMORY_PROPERTY_HOST_COHERENT_BIT,
/* This memory type either comes from heaps[0] if there is only
* mappable vram region, or from heaps[2] if there is both mappable &
* non-mappable vram regions.
*/
.heapIndex = device->vram_non_mappable.size > 0 ? 2 : 0,
};
} else if (device->info.has_llc) {
device->memory.heap_count = 1;
device->memory.heaps[0] = (struct anv_memory_heap) {
.size = device->sys.size,
.flags = VK_MEMORY_HEAP_DEVICE_LOCAL_BIT,
.is_local_mem = false,
};
/* Big core GPUs share LLC with the CPU and thus one memory type can be
* both cached and coherent at the same time.
*
* But some game engines can't handle single type well
* https://gitlab.freedesktop.org/mesa/mesa/-/issues/7360#note_1719438
*
* The second memory type w/out HOST_CACHED_BIT will get write-combining.
* See anv_AllocateMemory()).
*
* The Intel Vulkan driver for Windows also advertises these memory types.
*/
device->memory.type_count = 3;
device->memory.types[0] = (struct anv_memory_type) {
.propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT,
.heapIndex = 0,
};
device->memory.types[1] = (struct anv_memory_type) {
.propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
VK_MEMORY_PROPERTY_HOST_COHERENT_BIT,
.heapIndex = 0,
};
device->memory.types[2] = (struct anv_memory_type) {
.propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
VK_MEMORY_PROPERTY_HOST_COHERENT_BIT |
VK_MEMORY_PROPERTY_HOST_CACHED_BIT,
.heapIndex = 0,
};
} else {
device->memory.heap_count = 1;
device->memory.heaps[0] = (struct anv_memory_heap) {
@ -979,27 +919,21 @@ anv_physical_device_init_heaps(struct anv_physical_device *device, int fd)
.flags = VK_MEMORY_HEAP_DEVICE_LOCAL_BIT,
.is_local_mem = false,
};
/* The spec requires that we expose a host-visible, coherent memory
* type, but Atom GPUs don't share LLC. Thus we offer two memory types
* to give the application a choice between cached, but not coherent and
* coherent but uncached (WC though).
*/
device->memory.type_count = 2;
device->memory.types[0] = (struct anv_memory_type) {
.propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
VK_MEMORY_PROPERTY_HOST_CACHED_BIT,
.heapIndex = 0,
};
device->memory.types[1] = (struct anv_memory_type) {
.propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
VK_MEMORY_PROPERTY_HOST_COHERENT_BIT,
.heapIndex = 0,
};
}
switch (device->info.kmd_type) {
case INTEL_KMD_TYPE_XE:
result = anv_xe_physical_device_init_memory_types(device);
break;
case INTEL_KMD_TYPE_I915:
default:
result = anv_i915_physical_device_init_memory_types(device);
break;
}
if (result != VK_SUCCESS)
return result;
for (unsigned i = 0; i < device->memory.type_count; i++) {
VkMemoryPropertyFlags props = device->memory.types[i].propertyFlags;
if ((props & VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT) &&

View file

@ -125,6 +125,85 @@ anv_i915_physical_device_get_parameters(struct anv_physical_device *device)
return result;
}
VkResult
anv_i915_physical_device_init_memory_types(struct anv_physical_device *device)
{
if (anv_physical_device_has_vram(device)) {
device->memory.type_count = 3;
device->memory.types[0] = (struct anv_memory_type) {
.propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT,
.heapIndex = 0,
};
device->memory.types[1] = (struct anv_memory_type) {
.propertyFlags = VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
VK_MEMORY_PROPERTY_HOST_COHERENT_BIT |
VK_MEMORY_PROPERTY_HOST_CACHED_BIT,
.heapIndex = 1,
};
device->memory.types[2] = (struct anv_memory_type) {
.propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
VK_MEMORY_PROPERTY_HOST_COHERENT_BIT,
/* This memory type either comes from heaps[0] if there is only
* mappable vram region, or from heaps[2] if there is both mappable &
* non-mappable vram regions.
*/
.heapIndex = device->vram_non_mappable.size > 0 ? 2 : 0,
};
} else if (device->info.has_llc) {
/* Big core GPUs share LLC with the CPU and thus one memory type can be
* both cached and coherent at the same time.
*
* But some game engines can't handle single type well
* https://gitlab.freedesktop.org/mesa/mesa/-/issues/7360#note_1719438
*
* The second memory type w/out HOST_CACHED_BIT will get write-combining.
* See anv_AllocateMemory()).
*
* The Intel Vulkan driver for Windows also advertises these memory types.
*/
device->memory.type_count = 3;
device->memory.types[0] = (struct anv_memory_type) {
.propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT,
.heapIndex = 0,
};
device->memory.types[1] = (struct anv_memory_type) {
.propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
VK_MEMORY_PROPERTY_HOST_COHERENT_BIT,
.heapIndex = 0,
};
device->memory.types[2] = (struct anv_memory_type) {
.propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
VK_MEMORY_PROPERTY_HOST_COHERENT_BIT |
VK_MEMORY_PROPERTY_HOST_CACHED_BIT,
.heapIndex = 0,
};
} else {
/* The spec requires that we expose a host-visible, coherent memory
* type, but Atom GPUs don't share LLC. Thus we offer two memory types
* to give the application a choice between cached, but not coherent and
* coherent but uncached (WC though).
*/
device->memory.type_count = 2;
device->memory.types[0] = (struct anv_memory_type) {
.propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
VK_MEMORY_PROPERTY_HOST_CACHED_BIT,
.heapIndex = 0,
};
device->memory.types[1] = (struct anv_memory_type) {
.propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
VK_MEMORY_PROPERTY_HOST_COHERENT_BIT,
.heapIndex = 0,
};
}
return VK_SUCCESS;
}
VkResult
anv_i915_device_setup_context(struct anv_device *device,
const VkDeviceCreateInfo *pCreateInfo,

View file

@ -30,6 +30,8 @@ struct anv_physical_device;
VkResult
anv_i915_physical_device_get_parameters(struct anv_physical_device *device);
VkResult
anv_i915_physical_device_init_memory_types(struct anv_physical_device *device);
VkResult
anv_i915_device_setup_context(struct anv_device *device,

View file

@ -118,6 +118,60 @@ anv_xe_physical_device_get_parameters(struct anv_physical_device *device)
return VK_SUCCESS;
}
VkResult
anv_xe_physical_device_init_memory_types(struct anv_physical_device *device)
{
if (anv_physical_device_has_vram(device)) {
device->memory.type_count = 3;
device->memory.types[0] = (struct anv_memory_type) {
.propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT,
.heapIndex = 0,
};
device->memory.types[1] = (struct anv_memory_type) {
.propertyFlags = VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
VK_MEMORY_PROPERTY_HOST_COHERENT_BIT |
VK_MEMORY_PROPERTY_HOST_CACHED_BIT,
.heapIndex = 1,
};
device->memory.types[2] = (struct anv_memory_type) {
.propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
VK_MEMORY_PROPERTY_HOST_COHERENT_BIT,
/* This memory type either comes from heaps[0] if there is only
* mappable vram region, or from heaps[2] if there is both mappable &
* non-mappable vram regions.
*/
.heapIndex = device->vram_non_mappable.size > 0 ? 2 : 0,
};
} else if (device->info.has_llc) {
/* Big core GPUs share LLC with the CPU and thus one memory type can be
* both cached and coherent at the same time.
*
* But some game engines can't handle single type well
* https://gitlab.freedesktop.org/mesa/mesa/-/issues/7360#note_1719438
*
* TODO: But with current UAPI we can't change the mmap mode in Xe, so
* here only supporting two memory types.
*/
device->memory.type_count = 2;
device->memory.types[0] = (struct anv_memory_type) {
.propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT,
.heapIndex = 0,
};
device->memory.types[1] = (struct anv_memory_type) {
.propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
VK_MEMORY_PROPERTY_HOST_COHERENT_BIT |
VK_MEMORY_PROPERTY_HOST_CACHED_BIT,
.heapIndex = 0,
};
} else {
return vk_errorf(device, VK_ERROR_INITIALIZATION_FAILED,
"No memory heaps types set for non llc devices yet on Xe");
}
return VK_SUCCESS;
}
VkResult
anv_xe_device_check_status(struct vk_device *vk_device)
{

View file

@ -38,5 +38,8 @@ VkResult anv_xe_device_check_status(struct vk_device *vk_device);
VkResult
anv_xe_physical_device_get_parameters(struct anv_physical_device *device);
VkResult
anv_xe_physical_device_init_memory_types(struct anv_physical_device *device);
enum drm_sched_priority
anv_vk_priority_to_drm_sched_priority(VkQueueGlobalPriorityKHR vk_priority);