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;
close(device->master_fd);
}
-static void *
+static VKAPI_ATTR void *
default_alloc_func(void *pUserData,
size_t size,
size_t align,
return malloc(size);
}
-static void *
+static VKAPI_ATTR void *
default_realloc_func(void *pUserData,
void *pOriginal,
size_t size,
return realloc(pOriginal, size);
}
-static void
+static VKAPI_ATTR void
default_free_func(void *pUserData, void *pMemory)
{
free(pMemory);
*pFeatures = (VkPhysicalDeviceFeatures) {
.robustBufferAccess = false,
.fullDrawIndexUint32 = true,
- .imageCubeArray = false,
+ .imageCubeArray = true,
.independentBlend = true,
- .geometryShader = false,
+ .geometryShader = true,
.tessellationShader = false,
.sampleRateShading = true,
.dualSrcBlend = true,
.logicOp = true,
.multiDrawIndirect = false,
.drawIndirectFirstInstance = false,
- .depthClamp = false,
+ .depthClamp = true,
.depthBiasClamp = false,
.fillModeNonSolid = false,
.depthBounds = 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,
.maxTessellationControlTotalOutputComponents = 4096,
.maxTessellationEvaluationInputComponents = 128,
.maxTessellationEvaluationOutputComponents = 128,
- .maxGeometryShaderInvocations = 127,
+ .maxGeometryShaderInvocations = 32,
.maxGeometryInputComponents = 64,
.maxGeometryOutputComponents = 128,
.maxGeometryOutputVertices = 256,
.maxGeometryTotalOutputComponents = 1024,
- .maxFragmentInputComponents = 128,
+ .maxFragmentInputComponents = 124,
.maxFragmentOutputAttachments = 8,
.maxFragmentDualSrcAttachments = 1,
.maxFragmentCombinedOutputResources = 8,
.maxComputeWorkGroupCount = { 65535, 65535, 65535 },
.maxComputeWorkGroupInvocations = 2048,
.maxComputeWorkGroupSize = { 2048, 2048, 2048 },
- .subPixelPrecisionBits = 4 /* FIXME */,
+ .subPixelPrecisionBits = 8,
.subTexelPrecisionBits = 4 /* FIXME */,
.mipmapPrecisionBits = 4 /* FIXME */,
.maxDrawIndexedIndexValue = UINT32_MAX,
.viewportSubPixelBits = 8,
.minMemoryMapAlignment = 4096, /* A page */
.minTexelBufferOffsetAlignment = 64,
- .minUniformBufferOffsetAlignment = 4,
- .minStorageBufferOffsetAlignment = 4,
+ .minUniformBufferOffsetAlignment = 64,
+ .minStorageBufferOffsetAlignment = 64,
.minTexelOffset = -32,
.maxTexelOffset = 31,
.minTexelGatherOffset = -32,
.sampledImageStencilSampleCounts = sample_counts,
.storageImageSampleCounts = VK_SAMPLE_COUNT_1_BIT,
.maxSampleMaskWords = 1,
- .timestampComputeAndGraphics = false, /* FINISHME */
- .timestampPeriod = 1,
+ .timestampComputeAndGraphics = true,
+ .timestampPeriod = 1000000000.0 / 19200000.0, /* CP_ALWAYS_ON_COUNTER is fixed 19.2MHz */
.maxClipDistances = 8,
.maxCullDistances = 8,
.maxCombinedClipAndCullDistances = 8,
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;
}
.queueFlags =
VK_QUEUE_GRAPHICS_BIT | VK_QUEUE_COMPUTE_BIT | VK_QUEUE_TRANSFER_BIT,
.queueCount = 1,
- .timestampValidBits = 0, /* FINISHME */
+ .timestampValidBits = 48,
.minImageTransferGranularity = { 1, 1, 1 },
};
return -1;
}
+struct PACKED bcolor_entry {
+ uint32_t fp32[4];
+ uint16_t ui16[4];
+ int16_t si16[4];
+ uint16_t fp16[4];
+ uint16_t rgb565;
+ uint16_t rgb5a1;
+ uint16_t rgba4;
+ uint8_t __pad0[2];
+ uint8_t ui8[4];
+ int8_t si8[4];
+ uint32_t rgb10a2;
+ uint32_t z24; /* also s8? */
+ uint16_t srgb[4]; /* appears to duplicate fp16[], but clamped, used for srgb */
+ uint8_t __pad1[56];
+} border_color[] = {
+ [VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK] = {},
+ [VK_BORDER_COLOR_INT_TRANSPARENT_BLACK] = {},
+ [VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK] = {
+ .fp32[3] = 0x3f800000,
+ .ui16[3] = 0xffff,
+ .si16[3] = 0x7fff,
+ .fp16[3] = 0x3c00,
+ .rgb5a1 = 0x8000,
+ .rgba4 = 0xf000,
+ .ui8[3] = 0xff,
+ .si8[3] = 0x7f,
+ .rgb10a2 = 0xc0000000,
+ .srgb[3] = 0x3c00,
+ },
+ [VK_BORDER_COLOR_INT_OPAQUE_BLACK] = {
+ .fp32[3] = 1,
+ .fp16[3] = 1,
+ },
+ [VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE] = {
+ .fp32[0 ... 3] = 0x3f800000,
+ .ui16[0 ... 3] = 0xffff,
+ .si16[0 ... 3] = 0x7fff,
+ .fp16[0 ... 3] = 0x3c00,
+ .rgb565 = 0xffff,
+ .rgb5a1 = 0xffff,
+ .rgba4 = 0xffff,
+ .ui8[0 ... 3] = 0xff,
+ .si8[0 ... 3] = 0x7f,
+ .rgb10a2 = 0xffffffff,
+ .z24 = 0xffffff,
+ .srgb[0 ... 3] = 0x3c00,
+ },
+ [VK_BORDER_COLOR_INT_OPAQUE_WHITE] = {
+ .fp32[0 ... 3] = 1,
+ .fp16[0 ... 3] = 1,
+ },
+};
+
+
VkResult
tu_CreateDevice(VkPhysicalDevice physicalDevice,
const VkDeviceCreateInfo *pCreateInfo,
if (result != VK_SUCCESS)
goto fail_vsc_data2;
+ STATIC_ASSERT(sizeof(struct bcolor_entry) == 128);
+ result = tu_bo_init_new(device, &device->border_color, sizeof(border_color));
+ if (result != VK_SUCCESS)
+ goto fail_border_color;
+
+ result = tu_bo_map(device, &device->border_color);
+ if (result != VK_SUCCESS)
+ goto fail_border_color_map;
+
+ memcpy(device->border_color.map, border_color, sizeof(border_color));
+
VkPipelineCacheCreateInfo ci;
ci.sType = VK_STRUCTURE_TYPE_PIPELINE_CACHE_CREATE_INFO;
ci.pNext = NULL;
return VK_SUCCESS;
fail_pipeline_cache:
+fail_border_color_map:
+ tu_bo_finish(device, &device->border_color);
+
+fail_border_color:
tu_bo_finish(device, &device->vsc_data2);
fail_vsc_data2:
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);
}
pMemoryRequirements->memoryTypeBits = 1;
pMemoryRequirements->size = image->layout.size;
- pMemoryRequirements->alignment = image->alignment;
+ pMemoryRequirements->alignment = image->layout.base_align;
}
void
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);
}
static enum a6xx_tex_clamp
-tu6_tex_wrap(VkSamplerAddressMode address_mode, bool *needs_border)
+tu6_tex_wrap(VkSamplerAddressMode address_mode)
{
switch (address_mode) {
case VK_SAMPLER_ADDRESS_MODE_REPEAT:
case VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE:
return A6XX_TEX_CLAMP_TO_EDGE;
case VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER:
- *needs_border = true;
return A6XX_TEX_CLAMP_TO_BORDER;
case VK_SAMPLER_ADDRESS_MODE_MIRROR_CLAMP_TO_EDGE:
/* only works for PoT.. need to emulate otherwise! */
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);
- bool needs_border = false;
- sampler->state[0] =
+ sampler->descriptor[0] =
COND(miplinear, A6XX_TEX_SAMP_0_MIPFILTER_LINEAR_NEAR) |
A6XX_TEX_SAMP_0_XY_MAG(tu6_tex_filter(pCreateInfo->magFilter, aniso)) |
A6XX_TEX_SAMP_0_XY_MIN(tu6_tex_filter(pCreateInfo->minFilter, aniso)) |
A6XX_TEX_SAMP_0_ANISO(aniso) |
- A6XX_TEX_SAMP_0_WRAP_S(tu6_tex_wrap(pCreateInfo->addressModeU, &needs_border)) |
- A6XX_TEX_SAMP_0_WRAP_T(tu6_tex_wrap(pCreateInfo->addressModeV, &needs_border)) |
- A6XX_TEX_SAMP_0_WRAP_R(tu6_tex_wrap(pCreateInfo->addressModeW, &needs_border)) |
+ A6XX_TEX_SAMP_0_WRAP_S(tu6_tex_wrap(pCreateInfo->addressModeU)) |
+ A6XX_TEX_SAMP_0_WRAP_T(tu6_tex_wrap(pCreateInfo->addressModeV)) |
+ A6XX_TEX_SAMP_0_WRAP_R(tu6_tex_wrap(pCreateInfo->addressModeW)) |
A6XX_TEX_SAMP_0_LOD_BIAS(pCreateInfo->mipLodBias);
- sampler->state[1] =
+ sampler->descriptor[1] =
/* COND(!cso->seamless_cube_map, A6XX_TEX_SAMP_1_CUBEMAPSEAMLESSFILTOFF) | */
COND(pCreateInfo->unnormalizedCoordinates, A6XX_TEX_SAMP_1_UNNORM_COORDS) |
A6XX_TEX_SAMP_1_MIN_LOD(pCreateInfo->minLod) |
A6XX_TEX_SAMP_1_MAX_LOD(pCreateInfo->maxLod) |
COND(pCreateInfo->compareEnable,
A6XX_TEX_SAMP_1_COMPARE_FUNC(tu6_compare_func(pCreateInfo->compareOp)));
- sampler->state[2] = 0;
- sampler->state[3] = 0;
+ /* This is an offset into the border_color BO, which we fill with all the
+ * possible Vulkan border colors in the correct order, so we can just use
+ * the Vulkan enum with no translation necessary.
+ */
+ sampler->descriptor[2] =
+ A6XX_TEX_SAMP_2_BCOLOR_OFFSET((unsigned) pCreateInfo->borderColor *
+ 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?
- * border color
*/
-
- sampler->needs_border = needs_border;
- sampler->border = pCreateInfo->borderColor;
}
VkResult
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 };
+}
projects / mesa.git / blobdiff