[Mesa-stable] [Mesa-dev] [PATCH] radv: Don't expose heaps with 0 memory.
Bas Nieuwenhuizen
bas at basnieuwenhuizen.nl
Tue Nov 7 20:54:21 UTC 2017
Hi Andres,
It is a fix, but yeah the changes are pretty extensive. I don't have a
strong opinion on this one.
- Bas
On Fri, Nov 3, 2017 at 4:10 PM, Andres Gomez <agomez at igalia.com> wrote:
> Bas, this commit landed without mentioning any specific stable queue.
>
> For 17.2, this is quite an extensive change and has several trivial
> conflicts. I think it is OK to merge if you really meant it for 17.2
> too, but just double checking.
>
> Thanks!
>
> On Wed, 2017-11-01 at 20:58 +0100, Bas Nieuwenhuizen wrote:
>> It confuses CTS. This pregenerates the heap info into the
>> physical device, so we can use it for translating contiguous
>> indices into our "standard" ones.
>>
>> This also makes the WSI a bit smarter in case the first preferred
>> heap does not exist.
>>
>> CC: <mesa-stable at lists.freedesktop.org>
>> ---
>> src/amd/vulkan/radv_device.c | 135 ++++++++++++++++++++++++++----------------
>> src/amd/vulkan/radv_private.h | 3 +
>> src/amd/vulkan/radv_wsi.c | 16 ++++-
>> 3 files changed, 101 insertions(+), 53 deletions(-)
>>
>> diff --git a/src/amd/vulkan/radv_device.c b/src/amd/vulkan/radv_device.c
>> index 0c2f6fa6312..e95f3ee32af 100644
>> --- a/src/amd/vulkan/radv_device.c
>> +++ b/src/amd/vulkan/radv_device.c
>> @@ -104,6 +104,75 @@ get_chip_name(enum radeon_family family)
>> }
>> }
>>
>> +static void
>> +radv_physical_device_init_mem_types(struct radv_physical_device *device)
>> +{
>> + STATIC_ASSERT(RADV_MEM_HEAP_COUNT <= VK_MAX_MEMORY_HEAPS);
>> + uint64_t visible_vram_size = MIN2(device->rad_info.vram_size,
>> + device->rad_info.vram_vis_size);
>> +
>> + int vram_index = -1, visible_vram_index = -1, gart_index = -1;
>> + device->memory_properties.memoryHeapCount = 0;
>> + if (device->rad_info.vram_size - visible_vram_size > 0) {
>> + vram_index = device->memory_properties.memoryHeapCount++;
>> + device->memory_properties.memoryHeaps[vram_index] = (VkMemoryHeap) {
>> + .size = device->rad_info.vram_size - visible_vram_size,
>> + .flags = VK_MEMORY_HEAP_DEVICE_LOCAL_BIT,
>> + };
>> + }
>> + if (visible_vram_size) {
>> + visible_vram_index = device->memory_properties.memoryHeapCount++;
>> + device->memory_properties.memoryHeaps[visible_vram_index] = (VkMemoryHeap) {
>> + .size = visible_vram_size,
>> + .flags = VK_MEMORY_HEAP_DEVICE_LOCAL_BIT,
>> + };
>> + }
>> + if (device->rad_info.gart_size > 0) {
>> + gart_index = device->memory_properties.memoryHeapCount++;
>> + device->memory_properties.memoryHeaps[gart_index] = (VkMemoryHeap) {
>> + .size = device->rad_info.gart_size,
>> + .flags = 0,
>> + };
>> + }
>> +
>> + STATIC_ASSERT(RADV_MEM_TYPE_COUNT <= VK_MAX_MEMORY_TYPES);
>> + unsigned type_count = 0;
>> + if (vram_index >= 0) {
>> + device->mem_type_indices[type_count] = RADV_MEM_TYPE_VRAM;
>> + device->memory_properties.memoryTypes[type_count++] = (VkMemoryType) {
>> + .propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT,
>> + .heapIndex = vram_index,
>> + };
>> + }
>> + if (gart_index >= 0) {
>> + device->mem_type_indices[type_count] = RADV_MEM_TYPE_GTT_WRITE_COMBINE;
>> + device->memory_properties.memoryTypes[type_count++] = (VkMemoryType) {
>> + .propertyFlags = VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
>> + VK_MEMORY_PROPERTY_HOST_COHERENT_BIT,
>> + .heapIndex = gart_index,
>> + };
>> + }
>> + if (visible_vram_index >= 0) {
>> + device->mem_type_indices[type_count] = RADV_MEM_TYPE_VRAM_CPU_ACCESS;
>> + device->memory_properties.memoryTypes[type_count++] = (VkMemoryType) {
>> + .propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
>> + VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
>> + VK_MEMORY_PROPERTY_HOST_COHERENT_BIT,
>> + .heapIndex = visible_vram_index,
>> + };
>> + }
>> + if (gart_index >= 0) {
>> + device->mem_type_indices[type_count] = RADV_MEM_TYPE_GTT_CACHED;
>> + device->memory_properties.memoryTypes[type_count++] = (VkMemoryType) {
>> + .propertyFlags = VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
>> + VK_MEMORY_PROPERTY_HOST_COHERENT_BIT |
>> + VK_MEMORY_PROPERTY_HOST_CACHED_BIT,
>> + .heapIndex = gart_index,
>> + };
>> + }
>> + device->memory_properties.memoryTypeCount = type_count;
>> +}
>> +
>> static VkResult
>> radv_physical_device_init(struct radv_physical_device *device,
>> struct radv_instance *instance,
>> @@ -190,6 +259,7 @@ radv_physical_device_init(struct radv_physical_device *device,
>> */
>> device->has_clear_state = device->rad_info.chip_class >= CIK;
>>
>> + radv_physical_device_init_mem_types(device);
>> return VK_SUCCESS;
>>
>> fail:
>> @@ -780,49 +850,7 @@ void radv_GetPhysicalDeviceMemoryProperties(
>> {
>> RADV_FROM_HANDLE(radv_physical_device, physical_device, physicalDevice);
>>
>> - STATIC_ASSERT(RADV_MEM_TYPE_COUNT <= VK_MAX_MEMORY_TYPES);
>> -
>> - pMemoryProperties->memoryTypeCount = RADV_MEM_TYPE_COUNT;
>> - pMemoryProperties->memoryTypes[RADV_MEM_TYPE_VRAM] = (VkMemoryType) {
>> - .propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT,
>> - .heapIndex = RADV_MEM_HEAP_VRAM,
>> - };
>> - pMemoryProperties->memoryTypes[RADV_MEM_TYPE_GTT_WRITE_COMBINE] = (VkMemoryType) {
>> - .propertyFlags = VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
>> - VK_MEMORY_PROPERTY_HOST_COHERENT_BIT,
>> - .heapIndex = RADV_MEM_HEAP_GTT,
>> - };
>> - pMemoryProperties->memoryTypes[RADV_MEM_TYPE_VRAM_CPU_ACCESS] = (VkMemoryType) {
>> - .propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
>> - VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
>> - VK_MEMORY_PROPERTY_HOST_COHERENT_BIT,
>> - .heapIndex = RADV_MEM_HEAP_VRAM_CPU_ACCESS,
>> - };
>> - pMemoryProperties->memoryTypes[RADV_MEM_TYPE_GTT_CACHED] = (VkMemoryType) {
>> - .propertyFlags = VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
>> - VK_MEMORY_PROPERTY_HOST_COHERENT_BIT |
>> - VK_MEMORY_PROPERTY_HOST_CACHED_BIT,
>> - .heapIndex = RADV_MEM_HEAP_GTT,
>> - };
>> -
>> - STATIC_ASSERT(RADV_MEM_HEAP_COUNT <= VK_MAX_MEMORY_HEAPS);
>> - uint64_t visible_vram_size = MIN2(physical_device->rad_info.vram_size,
>> - physical_device->rad_info.vram_vis_size);
>> -
>> - pMemoryProperties->memoryHeapCount = RADV_MEM_HEAP_COUNT;
>> - pMemoryProperties->memoryHeaps[RADV_MEM_HEAP_VRAM] = (VkMemoryHeap) {
>> - .size = physical_device->rad_info.vram_size -
>> - visible_vram_size,
>> - .flags = VK_MEMORY_HEAP_DEVICE_LOCAL_BIT,
>> - };
>> - pMemoryProperties->memoryHeaps[RADV_MEM_HEAP_VRAM_CPU_ACCESS] = (VkMemoryHeap) {
>> - .size = visible_vram_size,
>> - .flags = VK_MEMORY_HEAP_DEVICE_LOCAL_BIT,
>> - };
>> - pMemoryProperties->memoryHeaps[RADV_MEM_HEAP_GTT] = (VkMemoryHeap) {
>> - .size = physical_device->rad_info.gart_size,
>> - .flags = 0,
>> - };
>> + *pMemoryProperties = physical_device->memory_properties;
>> }
>>
>> void radv_GetPhysicalDeviceMemoryProperties2KHR(
>> @@ -2070,6 +2098,7 @@ VkResult radv_alloc_memory(VkDevice _device,
>> VkResult result;
>> enum radeon_bo_domain domain;
>> uint32_t flags = 0;
>> + enum radv_mem_type mem_type_index = device->physical_device->mem_type_indices[pAllocateInfo->memoryTypeIndex];
>>
>> assert(pAllocateInfo->sType == VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO);
>>
>> @@ -2112,18 +2141,18 @@ VkResult radv_alloc_memory(VkDevice _device,
>> }
>>
>> uint64_t alloc_size = align_u64(pAllocateInfo->allocationSize, 4096);
>> - if (pAllocateInfo->memoryTypeIndex == RADV_MEM_TYPE_GTT_WRITE_COMBINE ||
>> - pAllocateInfo->memoryTypeIndex == RADV_MEM_TYPE_GTT_CACHED)
>> + if (mem_type_index == RADV_MEM_TYPE_GTT_WRITE_COMBINE ||
>> + mem_type_index == RADV_MEM_TYPE_GTT_CACHED)
>> domain = RADEON_DOMAIN_GTT;
>> else
>> domain = RADEON_DOMAIN_VRAM;
>>
>> - if (pAllocateInfo->memoryTypeIndex == RADV_MEM_TYPE_VRAM)
>> + if (mem_type_index == RADV_MEM_TYPE_VRAM)
>> flags |= RADEON_FLAG_NO_CPU_ACCESS;
>> else
>> flags |= RADEON_FLAG_CPU_ACCESS;
>>
>> - if (pAllocateInfo->memoryTypeIndex == RADV_MEM_TYPE_GTT_WRITE_COMBINE)
>> + if (mem_type_index == RADV_MEM_TYPE_GTT_WRITE_COMBINE)
>> flags |= RADEON_FLAG_GTT_WC;
>>
>> if (mem_flags & RADV_MEM_IMPLICIT_SYNC)
>> @@ -2139,7 +2168,7 @@ VkResult radv_alloc_memory(VkDevice _device,
>> result = VK_ERROR_OUT_OF_DEVICE_MEMORY;
>> goto fail;
>> }
>> - mem->type_index = pAllocateInfo->memoryTypeIndex;
>> + mem->type_index = mem_type_index;
>> out_success:
>> *pMem = radv_device_memory_to_handle(mem);
>>
>> @@ -2232,13 +2261,14 @@ VkResult radv_InvalidateMappedMemoryRanges(
>> }
>>
>> void radv_GetBufferMemoryRequirements(
>> - VkDevice device,
>> + VkDevice _device,
>> VkBuffer _buffer,
>> VkMemoryRequirements* pMemoryRequirements)
>> {
>> + RADV_FROM_HANDLE(radv_device, device, _device);
>> RADV_FROM_HANDLE(radv_buffer, buffer, _buffer);
>>
>> - pMemoryRequirements->memoryTypeBits = (1u << RADV_MEM_TYPE_COUNT) - 1;
>> + pMemoryRequirements->memoryTypeBits = (1u << device->physical_device->memory_properties.memoryTypeCount) - 1;
>>
>> if (buffer->flags & VK_BUFFER_CREATE_SPARSE_BINDING_BIT)
>> pMemoryRequirements->alignment = 4096;
>> @@ -2272,13 +2302,14 @@ void radv_GetBufferMemoryRequirements2KHR(
>> }
>>
>> void radv_GetImageMemoryRequirements(
>> - VkDevice device,
>> + VkDevice _device,
>> VkImage _image,
>> VkMemoryRequirements* pMemoryRequirements)
>> {
>> + RADV_FROM_HANDLE(radv_device, device, _device);
>> RADV_FROM_HANDLE(radv_image, image, _image);
>>
>> - pMemoryRequirements->memoryTypeBits = (1u << RADV_MEM_TYPE_COUNT) - 1;
>> + pMemoryRequirements->memoryTypeBits = (1u << device->physical_device->memory_properties.memoryTypeCount) - 1;
>>
>> pMemoryRequirements->size = image->size;
>> pMemoryRequirements->alignment = image->alignment;
>> diff --git a/src/amd/vulkan/radv_private.h b/src/amd/vulkan/radv_private.h
>> index 381afb777f3..bee851c64b4 100644
>> --- a/src/amd/vulkan/radv_private.h
>> +++ b/src/amd/vulkan/radv_private.h
>> @@ -282,6 +282,9 @@ struct radv_physical_device {
>> * the pipeline cache defined by apps.
>> */
>> struct disk_cache * disk_cache;
>> +
>> + VkPhysicalDeviceMemoryProperties memory_properties;
>> + enum radv_mem_type mem_type_indices[RADV_MEM_TYPE_COUNT];
>> };
>>
>> struct radv_instance {
>> diff --git a/src/amd/vulkan/radv_wsi.c b/src/amd/vulkan/radv_wsi.c
>> index b65ef27351d..e07c5028c5b 100644
>> --- a/src/amd/vulkan/radv_wsi.c
>> +++ b/src/amd/vulkan/radv_wsi.c
>> @@ -194,12 +194,26 @@ radv_wsi_image_create(VkDevice device_h,
>> .image = image_h
>> };
>>
>> + /* Find the first VRAM memory type, or GART for PRIME images. */
>> + int memory_type_index = -1;
>> + for (int i = 0; i < device->physical_device->memory_properties.memoryTypeCount; ++i) {
>> + bool is_local = !!(device->physical_device->memory_properties.memoryTypes[i].propertyFlags & VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT);
>> + if ((linear && !is_local) || (!linear && is_local)) {
>> + memory_type_index = i;
>> + break;
>> + }
>> + }
>> +
>> + /* fallback */
>> + if (memory_type_index == -1)
>> + memory_type_index = 0;
>> +
>> result = radv_alloc_memory(device_h,
>> &(VkMemoryAllocateInfo) {
>> .sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO,
>> .pNext = &ded_alloc,
>> .allocationSize = image->size,
>> - .memoryTypeIndex = linear ? 1 : 0,
>> + .memoryTypeIndex = memory_type_index,
>> },
>> NULL /* XXX: pAllocator */,
>> RADV_MEM_IMPLICIT_SYNC,
> --
> Br,
>
> Andres
More information about the mesa-stable
mailing list