switch (device->gpu_id) {
case 618:
- device->tile_align_w = 64;
- device->tile_align_h = 16;
device->magic.RB_UNKNOWN_8E04_blit = 0x00100000;
- device->magic.RB_CCU_CNTL_gmem = 0x3e400004;
+ device->ccu_offset_gmem = 0x7c000; /* 0x7e000 in some cases? */
+ device->ccu_offset_bypass = 0x10000;
device->magic.PC_UNKNOWN_9805 = 0x0;
device->magic.SP_UNKNOWN_A0F8 = 0x0;
break;
case 630:
case 640:
- device->tile_align_w = 64;
- device->tile_align_h = 16;
device->magic.RB_UNKNOWN_8E04_blit = 0x01000000;
- device->magic.RB_CCU_CNTL_gmem = 0x7c400004;
+ device->ccu_offset_gmem = 0xf8000;
+ device->ccu_offset_bypass = 0x20000;
device->magic.PC_UNKNOWN_9805 = 0x1;
device->magic.SP_UNKNOWN_A0F8 = 0x1;
break;
*pFeatures = (VkPhysicalDeviceFeatures) {
.robustBufferAccess = false,
.fullDrawIndexUint32 = true,
- .imageCubeArray = false,
+ .imageCubeArray = true,
.independentBlend = true,
.geometryShader = true,
.tessellationShader = false,
VkPhysicalDeviceProperties *pProperties)
{
TU_FROM_HANDLE(tu_physical_device, pdevice, physicalDevice);
- VkSampleCountFlags sample_counts = VK_SAMPLE_COUNT_1_BIT |
- VK_SAMPLE_COUNT_2_BIT | VK_SAMPLE_COUNT_4_BIT | VK_SAMPLE_COUNT_8_BIT;
-
- /* make sure that the entire descriptor set is addressable with a signed
- * 32-bit int. So the sum of all limits scaled by descriptor size has to
- * be at most 2 GiB. the combined image & samples object count as one of
- * both. This limit is for the pipeline layout, not for the set layout, but
- * there is no set limit, so we just set a pipeline limit. I don't think
- * any app is going to hit this soon. */
- size_t max_descriptor_set_size =
- ((1ull << 31) - 16 * MAX_DYNAMIC_BUFFERS) /
- (32 /* uniform buffer, 32 due to potential space wasted on alignment */ +
- 32 /* storage buffer, 32 due to potential space wasted on alignment */ +
- 32 /* sampler, largest when combined with image */ +
- 64 /* sampled image */ + 64 /* storage image */);
+ VkSampleCountFlags sample_counts =
+ VK_SAMPLE_COUNT_1_BIT | VK_SAMPLE_COUNT_2_BIT | VK_SAMPLE_COUNT_4_BIT;
+
+ /* I have no idea what the maximum size is, but the hardware supports very
+ * large numbers of descriptors (at least 2^16). This limit is based on
+ * CP_LOAD_STATE6, which has a 28-bit field for the DWORD offset, so that
+ * we don't have to think about what to do if that overflows, but really
+ * nothing is likely to get close to this.
+ */
+ const size_t max_descriptor_set_size = (1 << 28) / A6XX_TEX_CONST_DWORDS;
VkPhysicalDeviceLimits limits = {
.maxImageDimension1D = (1 << 14),
.maxImageDimensionCube = (1 << 14),
.maxImageArrayLayers = (1 << 11),
.maxTexelBufferElements = 128 * 1024 * 1024,
- .maxUniformBufferRange = UINT32_MAX,
+ .maxUniformBufferRange = MAX_UNIFORM_BUFFER_RANGE,
.maxStorageBufferRange = MAX_STORAGE_BUFFER_RANGE,
.maxPushConstantsSize = MAX_PUSH_CONSTANTS_SIZE,
.maxMemoryAllocationCount = UINT32_MAX,
.maxPerStageDescriptorStorageBuffers = max_descriptor_set_size,
.maxPerStageDescriptorSampledImages = max_descriptor_set_size,
.maxPerStageDescriptorStorageImages = max_descriptor_set_size,
- .maxPerStageDescriptorInputAttachments = max_descriptor_set_size,
+ .maxPerStageDescriptorInputAttachments = MAX_RTS,
.maxPerStageResources = max_descriptor_set_size,
.maxDescriptorSetSamplers = max_descriptor_set_size,
.maxDescriptorSetUniformBuffers = max_descriptor_set_size,
.maxDescriptorSetStorageBuffersDynamic = MAX_DYNAMIC_STORAGE_BUFFERS,
.maxDescriptorSetSampledImages = max_descriptor_set_size,
.maxDescriptorSetStorageImages = max_descriptor_set_size,
- .maxDescriptorSetInputAttachments = max_descriptor_set_size,
+ .maxDescriptorSetInputAttachments = MAX_RTS,
.maxVertexInputAttributes = 32,
.maxVertexInputBindings = 32,
- .maxVertexInputAttributeOffset = 2047,
+ .maxVertexInputAttributeOffset = 4095,
.maxVertexInputBindingStride = 2048,
.maxVertexOutputComponents = 128,
.maxTessellationGenerationLevel = 64,
.maxGeometryOutputComponents = 128,
.maxGeometryOutputVertices = 256,
.maxGeometryTotalOutputComponents = 1024,
- .maxFragmentInputComponents = 128,
+ .maxFragmentInputComponents = 124,
.maxFragmentOutputAttachments = 8,
.maxFragmentDualSrcAttachments = 1,
.maxFragmentCombinedOutputResources = 8,
.viewportSubPixelBits = 8,
.minMemoryMapAlignment = 4096, /* A page */
.minTexelBufferOffsetAlignment = 64,
- .minUniformBufferOffsetAlignment = 4,
- .minStorageBufferOffsetAlignment = 4,
+ .minUniformBufferOffsetAlignment = 64,
+ .minStorageBufferOffsetAlignment = 64,
.minTexelOffset = -32,
.maxTexelOffset = 31,
.minTexelGatherOffset = -32,
properties->maxTransformFeedbackStreamDataSize = 512;
properties->maxTransformFeedbackBufferDataSize = 512;
properties->maxTransformFeedbackBufferDataStride = 512;
- /* TODO: enable xfb query */
- properties->transformFeedbackQueries = false;
+ properties->transformFeedbackQueries = true;
properties->transformFeedbackStreamsLinesTriangles = false;
properties->transformFeedbackRasterizationStreamSelect = false;
properties->transformFeedbackDraw = true;
break;
}
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLE_LOCATIONS_PROPERTIES_EXT: {
+ VkPhysicalDeviceSampleLocationsPropertiesEXT *properties =
+ (VkPhysicalDeviceSampleLocationsPropertiesEXT *)ext;
+ properties->sampleLocationSampleCounts = 0;
+ if (pdevice->supported_extensions.EXT_sample_locations) {
+ properties->sampleLocationSampleCounts =
+ VK_SAMPLE_COUNT_1_BIT | VK_SAMPLE_COUNT_2_BIT | VK_SAMPLE_COUNT_4_BIT;
+ }
+ properties->maxSampleLocationGridSize = (VkExtent2D) { 1 , 1 };
+ properties->sampleLocationCoordinateRange[0] = 0.0f;
+ properties->sampleLocationCoordinateRange[1] = 0.9375f;
+ properties->sampleLocationSubPixelBits = 4;
+ properties->variableSampleLocations = true;
+ break;
+ }
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLER_FILTER_MINMAX_PROPERTIES: {
+ VkPhysicalDeviceSamplerFilterMinmaxProperties *properties =
+ (VkPhysicalDeviceSamplerFilterMinmaxProperties *)ext;
+ properties->filterMinmaxImageComponentMapping = true;
+ properties->filterMinmaxSingleComponentFormats = true;
+ break;
+ }
+
default:
break;
}
return VK_SUCCESS;
}
+VkResult
+tu_ImportSemaphoreFdKHR(VkDevice _device,
+ const VkImportSemaphoreFdInfoKHR *pImportSemaphoreFdInfo)
+{
+ tu_stub();
+
+ return VK_SUCCESS;
+}
+
+VkResult
+tu_GetSemaphoreFdKHR(VkDevice _device,
+ const VkSemaphoreGetFdInfoKHR *pGetFdInfo,
+ int *pFd)
+{
+ tu_stub();
+
+ return VK_SUCCESS;
+}
+
+VkResult
+tu_ImportFenceFdKHR(VkDevice _device,
+ const VkImportFenceFdInfoKHR *pImportFenceFdInfo)
+{
+ tu_stub();
+
+ return VK_SUCCESS;
+}
+
+VkResult
+tu_GetFenceFdKHR(VkDevice _device,
+ const VkFenceGetFdInfoKHR *pGetFdInfo,
+ int *pFd)
+{
+ tu_stub();
+
+ return VK_SUCCESS;
+}
+
VkResult
tu_EnumerateInstanceExtensionProperties(const char *pLayerName,
uint32_t *pPropertyCount,
TU_FROM_HANDLE(tu_buffer, buffer, _buffer);
pMemoryRequirements->memoryTypeBits = 1;
- pMemoryRequirements->alignment = 16;
+ pMemoryRequirements->alignment = 64;
pMemoryRequirements->size =
align64(buffer->size, pMemoryRequirements->alignment);
}
vk_free2(&device->alloc, pAllocator, buffer);
}
-static uint32_t
-tu_surface_max_layer_count(struct tu_image_view *iview)
-{
- return iview->type == VK_IMAGE_VIEW_TYPE_3D
- ? iview->extent.depth
- : (iview->base_layer + iview->layer_count);
-}
-
VkResult
tu_CreateFramebuffer(VkDevice _device,
const VkFramebufferCreateInfo *pCreateInfo,
VkImageView _iview = pCreateInfo->pAttachments[i];
struct tu_image_view *iview = tu_image_view_from_handle(_iview);
framebuffer->attachments[i].attachment = iview;
-
- framebuffer->width = MIN2(framebuffer->width, iview->extent.width);
- framebuffer->height = MIN2(framebuffer->height, iview->extent.height);
- framebuffer->layers =
- MIN2(framebuffer->layers, tu_surface_max_layer_count(iview));
}
*pFramebuffer = tu_framebuffer_to_handle(framebuffer);
return A6XX_TEX_NEAREST;
case VK_FILTER_LINEAR:
return aniso ? A6XX_TEX_ANISO : A6XX_TEX_LINEAR;
- case VK_FILTER_CUBIC_IMG:
+ case VK_FILTER_CUBIC_EXT:
+ return A6XX_TEX_CUBIC;
default:
unreachable("illegal texture filter");
break;
struct tu_sampler *sampler,
const VkSamplerCreateInfo *pCreateInfo)
{
+ const struct VkSamplerReductionModeCreateInfo *reduction =
+ vk_find_struct_const(pCreateInfo->pNext, SAMPLER_REDUCTION_MODE_CREATE_INFO);
+
unsigned aniso = pCreateInfo->anisotropyEnable ?
util_last_bit(MIN2((uint32_t)pCreateInfo->maxAnisotropy >> 1, 8)) : 0;
bool miplinear = (pCreateInfo->mipmapMode == VK_SAMPLER_MIPMAP_MODE_LINEAR);
sizeof(struct bcolor_entry));
sampler->descriptor[3] = 0;
+ if (reduction) {
+ /* note: vulkan enum matches hw */
+ sampler->descriptor[2] |= A6XX_TEX_SAMP_2_REDUCTION_MODE(reduction->reductionMode);
+ }
+
/* TODO:
* A6XX_TEX_SAMP_1_MIPFILTER_LINEAR_FAR disables mipmapping, but vk has no NONE mipfilter?
*/
VK_PEER_MEMORY_FEATURE_GENERIC_SRC_BIT |
VK_PEER_MEMORY_FEATURE_GENERIC_DST_BIT;
}
+
+void tu_GetPhysicalDeviceMultisamplePropertiesEXT(
+ VkPhysicalDevice physicalDevice,
+ VkSampleCountFlagBits samples,
+ VkMultisamplePropertiesEXT* pMultisampleProperties)
+{
+ TU_FROM_HANDLE(tu_physical_device, pdevice, physicalDevice);
+
+ if (samples <= VK_SAMPLE_COUNT_4_BIT && pdevice->supported_extensions.EXT_sample_locations)
+ pMultisampleProperties->maxSampleLocationGridSize = (VkExtent2D){ 1, 1 };
+ else
+ pMultisampleProperties->maxSampleLocationGridSize = (VkExtent2D){ 0, 0 };
+}