* IN THE SOFTWARE.
*/
-#include <dlfcn.h>
#include <stdbool.h>
#include <string.h>
#include <unistd.h>
#include <fcntl.h>
-#include <sys/stat.h>
#include "radv_private.h"
#include "radv_cs.h"
+#include "util/disk_cache.h"
#include "util/strtod.h"
-
+#include "util/vk_util.h"
#include <xf86drm.h>
#include <amdgpu.h>
#include <amdgpu_drm.h>
#include "vk_format.h"
#include "sid.h"
#include "util/debug.h"
-struct radv_dispatch_table dtable;
-
-static int
-radv_get_function_timestamp(void *ptr, uint32_t* timestamp)
-{
- Dl_info info;
- struct stat st;
- if (!dladdr(ptr, &info) || !info.dli_fname) {
- return -1;
- }
- if (stat(info.dli_fname, &st)) {
- return -1;
- }
- *timestamp = st.st_mtim.tv_sec;
- return 0;
-}
static int
radv_device_get_cache_uuid(enum radeon_family family, void *uuid)
uint32_t mesa_timestamp, llvm_timestamp;
uint16_t f = family;
memset(uuid, 0, VK_UUID_SIZE);
- if (radv_get_function_timestamp(radv_device_get_cache_uuid, &mesa_timestamp) ||
- radv_get_function_timestamp(LLVMInitializeAMDGPUTargetInfo, &llvm_timestamp))
+ if (!disk_cache_get_function_timestamp(radv_device_get_cache_uuid, &mesa_timestamp) ||
+ !disk_cache_get_function_timestamp(LLVMInitializeAMDGPUTargetInfo, &llvm_timestamp))
return -1;
memcpy(uuid, &mesa_timestamp, 4);
.extensionName = VK_KHR_GET_PHYSICAL_DEVICE_PROPERTIES_2_EXTENSION_NAME,
.specVersion = 1,
},
+ {
+ .extensionName = VK_KHR_SHADER_DRAW_PARAMETERS_EXTENSION_NAME,
+ .specVersion = 1,
+ },
+ {
+ .extensionName = VK_NV_DEDICATED_ALLOCATION_EXTENSION_NAME,
+ .specVersion = 1,
+ },
};
static VkResult
assert(strlen(path) < ARRAY_SIZE(device->path));
strncpy(device->path, path, ARRAY_SIZE(device->path));
- device->ws = radv_amdgpu_winsys_create(fd);
+ device->ws = radv_amdgpu_winsys_create(fd, instance->debug_flags);
if (!device->ws) {
result = VK_ERROR_INCOMPATIBLE_DRIVER;
goto fail;
}
+
+ device->local_fd = fd;
device->ws->query_info(device->ws, &device->rad_info);
result = radv_init_wsi(device);
if (result != VK_SUCCESS) {
fprintf(stderr, "WARNING: radv is not a conformant vulkan implementation, testing use only.\n");
device->name = device->rad_info.name;
- close(fd);
+
return VK_SUCCESS;
fail:
radv_extensions_finish(device->instance, &device->extensions);
radv_finish_wsi(device);
device->ws->destroy(device->ws);
+ close(device->local_fd);
}
};
static const struct debug_control radv_debug_options[] = {
- {"fastclears", RADV_DEBUG_FAST_CLEARS},
+ {"nofastclears", RADV_DEBUG_NO_FAST_CLEARS},
{"nodcc", RADV_DEBUG_NO_DCC},
{"shaders", RADV_DEBUG_DUMP_SHADERS},
{"nocache", RADV_DEBUG_NO_CACHE},
{"nohiz", RADV_DEBUG_NO_HIZ},
{"nocompute", RADV_DEBUG_NO_COMPUTE_QUEUE},
{"unsafemath", RADV_DEBUG_UNSAFE_MATH},
+ {"allbos", RADV_DEBUG_ALL_BOS},
+ {"noibs", RADV_DEBUG_NO_IBS},
{NULL, 0}
};
{
RADV_FROM_HANDLE(radv_instance, instance, _instance);
+ if (!instance)
+ return;
+
for (int i = 0; i < instance->physicalDeviceCount; ++i) {
radv_physical_device_finish(instance->physicalDevices + i);
}
.shaderSampledImageArrayDynamicIndexing = true,
.shaderStorageBufferArrayDynamicIndexing = true,
.shaderStorageImageArrayDynamicIndexing = true,
- .shaderStorageImageReadWithoutFormat = false,
- .shaderStorageImageWriteWithoutFormat = false,
+ .shaderStorageImageReadWithoutFormat = true,
+ .shaderStorageImageWriteWithoutFormat = true,
.shaderClipDistance = true,
.shaderCullDistance = true,
- .shaderFloat64 = false,
+ .shaderFloat64 = true,
.shaderInt64 = false,
.shaderInt16 = false,
- .alphaToOne = true,
.variableMultisampleRate = false,
.inheritedQueries = false,
};
return radv_GetPhysicalDeviceFeatures(physicalDevice, &pFeatures->features);
}
+static uint32_t radv_get_driver_version()
+{
+ const char *minor_string = strchr(VERSION, '.');
+ const char *patch_string = minor_string ? strchr(minor_string + 1, ','): NULL;
+ int major = atoi(VERSION);
+ int minor = minor_string ? atoi(minor_string + 1) : 0;
+ int patch = patch_string ? atoi(patch_string + 1) : 0;
+ if (strstr(VERSION, "devel")) {
+ if (patch == 0) {
+ patch = 99;
+ if (minor == 0) {
+ minor = 99;
+ --major;
+ } else
+ --minor;
+ } else
+ --patch;
+ }
+ uint32_t version = VK_MAKE_VERSION(major, minor, patch);
+ return version;
+}
+
void radv_GetPhysicalDeviceProperties(
VkPhysicalDevice physicalDevice,
VkPhysicalDeviceProperties* pProperties)
};
*pProperties = (VkPhysicalDeviceProperties) {
- .apiVersion = VK_MAKE_VERSION(1, 0, 5),
- .driverVersion = 1,
+ .apiVersion = VK_MAKE_VERSION(1, 0, 42),
+ .driverVersion = radv_get_driver_version(),
.vendorID = 0x1002,
.deviceID = pdevice->rad_info.pci_id,
.deviceType = VK_PHYSICAL_DEVICE_TYPE_DISCRETE_GPU,
return radv_GetPhysicalDeviceProperties(physicalDevice, &pProperties->properties);
}
-void radv_GetPhysicalDeviceQueueFamilyProperties(
- VkPhysicalDevice physicalDevice,
+static void radv_get_physical_device_queue_family_properties(
+ struct radv_physical_device* pdevice,
uint32_t* pCount,
- VkQueueFamilyProperties* pQueueFamilyProperties)
+ VkQueueFamilyProperties** pQueueFamilyProperties)
{
- RADV_FROM_HANDLE(radv_physical_device, pdevice, physicalDevice);
int num_queue_families = 1;
int idx;
if (pdevice->rad_info.compute_rings > 0 &&
idx = 0;
if (*pCount >= 1) {
- pQueueFamilyProperties[idx] = (VkQueueFamilyProperties) {
+ *pQueueFamilyProperties[idx] = (VkQueueFamilyProperties) {
.queueFlags = VK_QUEUE_GRAPHICS_BIT |
VK_QUEUE_COMPUTE_BIT |
VK_QUEUE_TRANSFER_BIT,
pdevice->rad_info.chip_class >= CIK &&
!(pdevice->instance->debug_flags & RADV_DEBUG_NO_COMPUTE_QUEUE)) {
if (*pCount > idx) {
- pQueueFamilyProperties[idx] = (VkQueueFamilyProperties) {
+ *pQueueFamilyProperties[idx] = (VkQueueFamilyProperties) {
.queueFlags = VK_QUEUE_COMPUTE_BIT | VK_QUEUE_TRANSFER_BIT,
.queueCount = pdevice->rad_info.compute_rings,
.timestampValidBits = 64,
*pCount = idx;
}
+void radv_GetPhysicalDeviceQueueFamilyProperties(
+ VkPhysicalDevice physicalDevice,
+ uint32_t* pCount,
+ VkQueueFamilyProperties* pQueueFamilyProperties)
+{
+ RADV_FROM_HANDLE(radv_physical_device, pdevice, physicalDevice);
+ if (!pQueueFamilyProperties) {
+ return radv_get_physical_device_queue_family_properties(pdevice, pCount, NULL);
+ return;
+ }
+ VkQueueFamilyProperties *properties[] = {
+ pQueueFamilyProperties + 0,
+ pQueueFamilyProperties + 1,
+ pQueueFamilyProperties + 2,
+ };
+ radv_get_physical_device_queue_family_properties(pdevice, pCount, properties);
+ assert(*pCount <= 3);
+}
+
void radv_GetPhysicalDeviceQueueFamilyProperties2KHR(
VkPhysicalDevice physicalDevice,
uint32_t* pCount,
VkQueueFamilyProperties2KHR *pQueueFamilyProperties)
{
- return radv_GetPhysicalDeviceQueueFamilyProperties(physicalDevice,
- pCount,
- &pQueueFamilyProperties->queueFamilyProperties);
+ RADV_FROM_HANDLE(radv_physical_device, pdevice, physicalDevice);
+ if (!pQueueFamilyProperties) {
+ return radv_get_physical_device_queue_family_properties(pdevice, pCount, NULL);
+ return;
+ }
+ VkQueueFamilyProperties *properties[] = {
+ &pQueueFamilyProperties[0].queueFamilyProperties,
+ &pQueueFamilyProperties[1].queueFamilyProperties,
+ &pQueueFamilyProperties[2].queueFamilyProperties,
+ };
+ radv_get_physical_device_queue_family_properties(pdevice, pCount, properties);
+ assert(*pCount <= 3);
}
void radv_GetPhysicalDeviceMemoryProperties(
if (queue->hw_ctx)
queue->device->ws->ctx_destroy(queue->hw_ctx);
- if (queue->preamble_cs)
- queue->device->ws->cs_destroy(queue->preamble_cs);
+ if (queue->initial_preamble_cs)
+ queue->device->ws->cs_destroy(queue->initial_preamble_cs);
+ if (queue->continue_preamble_cs)
+ queue->device->ws->cs_destroy(queue->continue_preamble_cs);
if (queue->descriptor_bo)
queue->device->ws->buffer_destroy(queue->descriptor_bo);
if (queue->scratch_bo)
break;
}
device->ws->cs_finalize(device->empty_cs[family]);
+
+ device->flush_cs[family] = device->ws->cs_create(device->ws, family);
+ switch (family) {
+ case RADV_QUEUE_GENERAL:
+ case RADV_QUEUE_COMPUTE:
+ si_cs_emit_cache_flush(device->flush_cs[family],
+ device->physical_device->rad_info.chip_class,
+ family == RADV_QUEUE_COMPUTE && device->physical_device->rad_info.chip_class >= CIK,
+ RADV_CMD_FLAG_INV_ICACHE |
+ RADV_CMD_FLAG_INV_SMEM_L1 |
+ RADV_CMD_FLAG_INV_VMEM_L1 |
+ RADV_CMD_FLAG_INV_GLOBAL_L2);
+ break;
+ }
+ device->ws->cs_finalize(device->flush_cs[family]);
}
if (getenv("RADV_TRACE_FILE")) {
goto fail;
}
+ if (device->physical_device->rad_info.chip_class >= CIK)
+ cik_create_gfx_config(device);
+
*pDevice = radv_device_to_handle(device);
return VK_SUCCESS;
if (device->trace_bo)
device->ws->buffer_destroy(device->trace_bo);
+ if (device->gfx_init)
+ device->ws->buffer_destroy(device->gfx_init);
+
for (unsigned i = 0; i < RADV_MAX_QUEUE_FAMILIES; i++) {
for (unsigned q = 0; q < device->queue_count[i]; q++)
radv_queue_finish(&device->queues[i][q]);
{
RADV_FROM_HANDLE(radv_device, device, _device);
+ if (!device)
+ return;
+
if (device->trace_bo)
device->ws->buffer_destroy(device->trace_bo);
+ if (device->gfx_init)
+ device->ws->buffer_destroy(device->gfx_init);
+
for (unsigned i = 0; i < RADV_MAX_QUEUE_FAMILIES; i++) {
for (unsigned q = 0; q < device->queue_count[i]; q++)
radv_queue_finish(&device->queues[i][q]);
if (device->queue_count[i])
vk_free(&device->alloc, device->queues[i]);
+ if (device->empty_cs[i])
+ device->ws->cs_destroy(device->empty_cs[i]);
+ if (device->flush_cs[i])
+ device->ws->cs_destroy(device->flush_cs[i]);
}
radv_device_finish_meta(device);
uint32_t compute_scratch_size,
uint32_t esgs_ring_size,
uint32_t gsvs_ring_size,
- struct radeon_winsys_cs **preamble_cs)
+ struct radeon_winsys_cs **initial_preamble_cs,
+ struct radeon_winsys_cs **continue_preamble_cs)
{
struct radeon_winsys_bo *scratch_bo = NULL;
struct radeon_winsys_bo *descriptor_bo = NULL;
struct radeon_winsys_bo *compute_scratch_bo = NULL;
struct radeon_winsys_bo *esgs_ring_bo = NULL;
struct radeon_winsys_bo *gsvs_ring_bo = NULL;
- struct radeon_winsys_cs *cs = NULL;
-
- if (!scratch_size && !compute_scratch_size && !esgs_ring_size && !gsvs_ring_size) {
- *preamble_cs = NULL;
- return VK_SUCCESS;
- }
+ struct radeon_winsys_cs *dest_cs[2] = {0};
if (scratch_size <= queue->scratch_size &&
compute_scratch_size <= queue->compute_scratch_size &&
esgs_ring_size <= queue->esgs_ring_size &&
- gsvs_ring_size <= queue->gsvs_ring_size) {
- *preamble_cs = queue->preamble_cs;
+ gsvs_ring_size <= queue->gsvs_ring_size &&
+ queue->initial_preamble_cs) {
+ *initial_preamble_cs = queue->initial_preamble_cs;
+ *continue_preamble_cs = queue->continue_preamble_cs;
+ if (!scratch_size && !compute_scratch_size && !esgs_ring_size && !gsvs_ring_size)
+ *continue_preamble_cs = NULL;
return VK_SUCCESS;
}
} else
descriptor_bo = queue->descriptor_bo;
- cs = queue->device->ws->cs_create(queue->device->ws,
- queue->queue_family_index ? RING_COMPUTE : RING_GFX);
- if (!cs)
- goto fail;
+ for(int i = 0; i < 2; ++i) {
+ struct radeon_winsys_cs *cs = NULL;
+ cs = queue->device->ws->cs_create(queue->device->ws,
+ queue->queue_family_index ? RING_COMPUTE : RING_GFX);
+ if (!cs)
+ goto fail;
+ dest_cs[i] = cs;
- if (scratch_bo)
- queue->device->ws->cs_add_buffer(cs, scratch_bo, 8);
+ if (scratch_bo)
+ queue->device->ws->cs_add_buffer(cs, scratch_bo, 8);
- if (esgs_ring_bo)
- queue->device->ws->cs_add_buffer(cs, esgs_ring_bo, 8);
+ if (esgs_ring_bo)
+ queue->device->ws->cs_add_buffer(cs, esgs_ring_bo, 8);
- if (gsvs_ring_bo)
- queue->device->ws->cs_add_buffer(cs, gsvs_ring_bo, 8);
+ if (gsvs_ring_bo)
+ queue->device->ws->cs_add_buffer(cs, gsvs_ring_bo, 8);
- if (descriptor_bo)
- queue->device->ws->cs_add_buffer(cs, descriptor_bo, 8);
+ if (descriptor_bo)
+ queue->device->ws->cs_add_buffer(cs, descriptor_bo, 8);
- if (descriptor_bo != queue->descriptor_bo) {
- uint32_t *map = (uint32_t*)queue->device->ws->buffer_map(descriptor_bo);
+ if (descriptor_bo != queue->descriptor_bo) {
+ uint32_t *map = (uint32_t*)queue->device->ws->buffer_map(descriptor_bo);
- if (scratch_bo) {
- uint64_t scratch_va = queue->device->ws->buffer_get_va(scratch_bo);
- uint32_t rsrc1 = S_008F04_BASE_ADDRESS_HI(scratch_va >> 32) |
- S_008F04_SWIZZLE_ENABLE(1);
- map[0] = scratch_va;
- map[1] = rsrc1;
+ if (scratch_bo) {
+ uint64_t scratch_va = queue->device->ws->buffer_get_va(scratch_bo);
+ uint32_t rsrc1 = S_008F04_BASE_ADDRESS_HI(scratch_va >> 32) |
+ S_008F04_SWIZZLE_ENABLE(1);
+ map[0] = scratch_va;
+ map[1] = rsrc1;
+ }
+
+ if (esgs_ring_bo || gsvs_ring_bo)
+ fill_geom_rings(queue, map, esgs_ring_size, esgs_ring_bo, gsvs_ring_size, gsvs_ring_bo);
+
+ queue->device->ws->buffer_unmap(descriptor_bo);
}
- if (esgs_ring_bo || gsvs_ring_bo)
- fill_geom_rings(queue, map, esgs_ring_size, esgs_ring_bo, gsvs_ring_size, gsvs_ring_bo);
+ if (esgs_ring_bo || gsvs_ring_bo) {
+ radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
+ radeon_emit(cs, EVENT_TYPE(V_028A90_VS_PARTIAL_FLUSH) | EVENT_INDEX(4));
+ radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
+ radeon_emit(cs, EVENT_TYPE(V_028A90_VGT_FLUSH) | EVENT_INDEX(0));
+
+ if (queue->device->physical_device->rad_info.chip_class >= CIK) {
+ radeon_set_uconfig_reg_seq(cs, R_030900_VGT_ESGS_RING_SIZE, 2);
+ radeon_emit(cs, esgs_ring_size >> 8);
+ radeon_emit(cs, gsvs_ring_size >> 8);
+ } else {
+ radeon_set_config_reg_seq(cs, R_0088C8_VGT_ESGS_RING_SIZE, 2);
+ radeon_emit(cs, esgs_ring_size >> 8);
+ radeon_emit(cs, gsvs_ring_size >> 8);
+ }
+ }
- queue->device->ws->buffer_unmap(descriptor_bo);
- }
+ if (descriptor_bo) {
+ uint32_t regs[] = {R_00B030_SPI_SHADER_USER_DATA_PS_0,
+ R_00B130_SPI_SHADER_USER_DATA_VS_0,
+ R_00B230_SPI_SHADER_USER_DATA_GS_0,
+ R_00B330_SPI_SHADER_USER_DATA_ES_0,
+ R_00B430_SPI_SHADER_USER_DATA_HS_0,
+ R_00B530_SPI_SHADER_USER_DATA_LS_0};
- if (esgs_ring_bo || gsvs_ring_bo) {
- radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
- radeon_emit(cs, EVENT_TYPE(V_028A90_VS_PARTIAL_FLUSH) | EVENT_INDEX(4));
- radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
- radeon_emit(cs, EVENT_TYPE(V_028A90_VGT_FLUSH) | EVENT_INDEX(0));
+ uint64_t va = queue->device->ws->buffer_get_va(descriptor_bo);
- if (queue->device->physical_device->rad_info.chip_class >= CIK) {
- radeon_set_uconfig_reg_seq(cs, R_030900_VGT_ESGS_RING_SIZE, 2);
- radeon_emit(cs, esgs_ring_size >> 8);
- radeon_emit(cs, gsvs_ring_size >> 8);
- } else {
- radeon_set_config_reg_seq(cs, R_0088C8_VGT_ESGS_RING_SIZE, 2);
- radeon_emit(cs, esgs_ring_size >> 8);
- radeon_emit(cs, gsvs_ring_size >> 8);
+ for (int i = 0; i < ARRAY_SIZE(regs); ++i) {
+ radeon_set_sh_reg_seq(cs, regs[i], 2);
+ radeon_emit(cs, va);
+ radeon_emit(cs, va >> 32);
+ }
}
- }
- if (descriptor_bo) {
- uint32_t regs[] = {R_00B030_SPI_SHADER_USER_DATA_PS_0,
- R_00B130_SPI_SHADER_USER_DATA_VS_0,
- R_00B230_SPI_SHADER_USER_DATA_GS_0,
- R_00B330_SPI_SHADER_USER_DATA_ES_0,
- R_00B430_SPI_SHADER_USER_DATA_HS_0,
- R_00B530_SPI_SHADER_USER_DATA_LS_0};
+ if (compute_scratch_bo) {
+ uint64_t scratch_va = queue->device->ws->buffer_get_va(compute_scratch_bo);
+ uint32_t rsrc1 = S_008F04_BASE_ADDRESS_HI(scratch_va >> 32) |
+ S_008F04_SWIZZLE_ENABLE(1);
- uint64_t va = queue->device->ws->buffer_get_va(descriptor_bo);
+ queue->device->ws->cs_add_buffer(cs, compute_scratch_bo, 8);
- for (int i = 0; i < ARRAY_SIZE(regs); ++i) {
- radeon_set_sh_reg_seq(cs, regs[i], 2);
- radeon_emit(cs, va);
- radeon_emit(cs, va >> 32);
+ radeon_set_sh_reg_seq(cs, R_00B900_COMPUTE_USER_DATA_0, 2);
+ radeon_emit(cs, scratch_va);
+ radeon_emit(cs, rsrc1);
}
- }
- if (compute_scratch_bo) {
- uint64_t scratch_va = queue->device->ws->buffer_get_va(compute_scratch_bo);
- uint32_t rsrc1 = S_008F04_BASE_ADDRESS_HI(scratch_va >> 32) |
- S_008F04_SWIZZLE_ENABLE(1);
-
- queue->device->ws->cs_add_buffer(cs, compute_scratch_bo, 8);
+ if (!i) {
+ si_cs_emit_cache_flush(cs,
+ queue->device->physical_device->rad_info.chip_class,
+ queue->queue_family_index == RING_COMPUTE &&
+ queue->device->physical_device->rad_info.chip_class >= CIK,
+ RADV_CMD_FLAG_INV_ICACHE |
+ RADV_CMD_FLAG_INV_SMEM_L1 |
+ RADV_CMD_FLAG_INV_VMEM_L1 |
+ RADV_CMD_FLAG_INV_GLOBAL_L2);
+ }
- radeon_set_sh_reg_seq(cs, R_00B900_COMPUTE_USER_DATA_0, 2);
- radeon_emit(cs, scratch_va);
- radeon_emit(cs, rsrc1);
+ if (!queue->device->ws->cs_finalize(cs))
+ goto fail;
}
- if (!queue->device->ws->cs_finalize(cs))
- goto fail;
+ if (queue->initial_preamble_cs)
+ queue->device->ws->cs_destroy(queue->initial_preamble_cs);
- if (queue->preamble_cs)
- queue->device->ws->cs_destroy(queue->preamble_cs);
+ if (queue->continue_preamble_cs)
+ queue->device->ws->cs_destroy(queue->continue_preamble_cs);
- queue->preamble_cs = cs;
+ queue->initial_preamble_cs = dest_cs[0];
+ queue->continue_preamble_cs = dest_cs[1];
if (scratch_bo != queue->scratch_bo) {
if (queue->scratch_bo)
queue->descriptor_bo = descriptor_bo;
}
- *preamble_cs = cs;
+ *initial_preamble_cs = queue->initial_preamble_cs;
+ *continue_preamble_cs = queue->continue_preamble_cs;
+ if (!scratch_size && !compute_scratch_size && !esgs_ring_size && !gsvs_ring_size)
+ *continue_preamble_cs = NULL;
return VK_SUCCESS;
fail:
- if (cs)
- queue->device->ws->cs_destroy(cs);
+ for (int i = 0; i < ARRAY_SIZE(dest_cs); ++i)
+ if (dest_cs[i])
+ queue->device->ws->cs_destroy(dest_cs[i]);
if (descriptor_bo && descriptor_bo != queue->descriptor_bo)
queue->device->ws->buffer_destroy(descriptor_bo);
if (scratch_bo && scratch_bo != queue->scratch_bo)
uint32_t scratch_size = 0;
uint32_t compute_scratch_size = 0;
uint32_t esgs_ring_size = 0, gsvs_ring_size = 0;
- struct radeon_winsys_cs *preamble_cs = NULL;
+ struct radeon_winsys_cs *initial_preamble_cs = NULL, *continue_preamble_cs = NULL;
VkResult result;
+ bool fence_emitted = false;
/* Do this first so failing to allocate scratch buffers can't result in
* partially executed submissions. */
}
}
- result = radv_get_preamble_cs(queue, scratch_size, compute_scratch_size, esgs_ring_size, gsvs_ring_size, &preamble_cs);
+ result = radv_get_preamble_cs(queue, scratch_size, compute_scratch_size,
+ esgs_ring_size, gsvs_ring_size,
+ &initial_preamble_cs, &continue_preamble_cs);
if (result != VK_SUCCESS)
return result;
for (uint32_t i = 0; i < submitCount; i++) {
struct radeon_winsys_cs **cs_array;
- bool can_patch = true;
+ bool has_flush = !submitCount;
+ bool can_patch = !has_flush;
uint32_t advance;
- if (!pSubmits[i].commandBufferCount)
+ if (!pSubmits[i].commandBufferCount) {
+ if (pSubmits[i].waitSemaphoreCount || pSubmits[i].signalSemaphoreCount) {
+ ret = queue->device->ws->cs_submit(ctx, queue->queue_idx,
+ &queue->device->empty_cs[queue->queue_family_index],
+ 1, NULL, NULL,
+ (struct radeon_winsys_sem **)pSubmits[i].pWaitSemaphores,
+ pSubmits[i].waitSemaphoreCount,
+ (struct radeon_winsys_sem **)pSubmits[i].pSignalSemaphores,
+ pSubmits[i].signalSemaphoreCount,
+ false, base_fence);
+ if (ret) {
+ radv_loge("failed to submit CS %d\n", i);
+ abort();
+ }
+ fence_emitted = true;
+ }
continue;
+ }
cs_array = malloc(sizeof(struct radeon_winsys_cs *) *
- pSubmits[i].commandBufferCount);
+ (pSubmits[i].commandBufferCount + has_flush));
+
+ if(has_flush)
+ cs_array[0] = queue->device->flush_cs[queue->queue_family_index];
for (uint32_t j = 0; j < pSubmits[i].commandBufferCount; j++) {
RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer,
pSubmits[i].pCommandBuffers[j]);
assert(cmd_buffer->level == VK_COMMAND_BUFFER_LEVEL_PRIMARY);
- cs_array[j] = cmd_buffer->cs;
+ cs_array[j + has_flush] = cmd_buffer->cs;
if ((cmd_buffer->usage_flags & VK_COMMAND_BUFFER_USAGE_SIMULTANEOUS_USE_BIT))
can_patch = false;
}
- for (uint32_t j = 0; j < pSubmits[i].commandBufferCount; j += advance) {
+ for (uint32_t j = 0; j < pSubmits[i].commandBufferCount + has_flush; j += advance) {
advance = MIN2(max_cs_submission,
- pSubmits[i].commandBufferCount - j);
+ pSubmits[i].commandBufferCount + has_flush - j);
bool b = j == 0;
- bool e = j + advance == pSubmits[i].commandBufferCount;
+ bool e = j + advance == pSubmits[i].commandBufferCount + has_flush;
if (queue->device->trace_bo)
*queue->device->trace_id_ptr = 0;
ret = queue->device->ws->cs_submit(ctx, queue->queue_idx, cs_array + j,
- advance, preamble_cs,
+ advance, initial_preamble_cs, continue_preamble_cs,
(struct radeon_winsys_sem **)pSubmits[i].pWaitSemaphores,
b ? pSubmits[i].waitSemaphoreCount : 0,
(struct radeon_winsys_sem **)pSubmits[i].pSignalSemaphores,
radv_loge("failed to submit CS %d\n", i);
abort();
}
+ fence_emitted = true;
if (queue->device->trace_bo) {
bool success = queue->device->ws->ctx_wait_idle(
queue->hw_ctx,
}
if (fence) {
- if (!submitCount)
+ if (!fence_emitted)
ret = queue->device->ws->cs_submit(ctx, queue->queue_idx,
&queue->device->empty_cs[queue->queue_family_index],
- 1, NULL, NULL, 0, NULL, 0,
+ 1, NULL, NULL, NULL, 0, NULL, 0,
false, base_fence);
fence->submitted = true;
return radv_lookup_entrypoint(pName);
}
+bool radv_get_memory_fd(struct radv_device *device,
+ struct radv_device_memory *memory,
+ int *pFD)
+{
+ struct radeon_bo_metadata metadata;
+
+ if (memory->image) {
+ radv_init_metadata(device, memory->image, &metadata);
+ device->ws->buffer_set_metadata(memory->bo, &metadata);
+ }
+
+ return device->ws->buffer_get_fd(device->ws, memory->bo,
+ pFD);
+}
+
VkResult radv_AllocateMemory(
VkDevice _device,
const VkMemoryAllocateInfo* pAllocateInfo,
VkResult result;
enum radeon_bo_domain domain;
uint32_t flags = 0;
+ const VkDedicatedAllocationMemoryAllocateInfoNV *dedicate_info = NULL;
assert(pAllocateInfo->sType == VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO);
if (pAllocateInfo->allocationSize == 0) {
return VK_SUCCESS;
}
+ vk_foreach_struct(ext, pAllocateInfo->pNext) {
+ switch (ext->sType) {
+ case VK_STRUCTURE_TYPE_DEDICATED_ALLOCATION_MEMORY_ALLOCATE_INFO_NV:
+ dedicate_info = (const VkDedicatedAllocationMemoryAllocateInfoNV *)ext;
+ break;
+ default:
+ break;
+ }
+ }
+
mem = vk_alloc2(&device->alloc, pAllocator, sizeof(*mem), 8,
VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
if (mem == NULL)
return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
+ if (dedicate_info) {
+ mem->image = radv_image_from_handle(dedicate_info->image);
+ mem->buffer = radv_buffer_from_handle(dedicate_info->buffer);
+ } else {
+ mem->image = NULL;
+ mem->buffer = NULL;
+ }
+
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 (pAllocateInfo->memoryTypeIndex == RADV_MEM_TYPE_GTT_WRITE_COMBINE)
flags |= RADEON_FLAG_GTT_WC;
- mem->bo = device->ws->buffer_create(device->ws, alloc_size, 32768,
+ mem->bo = device->ws->buffer_create(device->ws, alloc_size, 65536,
domain, flags);
if (!mem->bo) {
if (!sem)
return VK_ERROR_OUT_OF_HOST_MEMORY;
- *pSemaphore = (VkSemaphore)sem;
+ *pSemaphore = radeon_winsys_sem_to_handle(sem);
return VK_SUCCESS;
}
const VkAllocationCallbacks* pAllocator)
{
RADV_FROM_HANDLE(radv_device, device, _device);
- struct radeon_winsys_sem *sem;
+ RADV_FROM_HANDLE(radeon_winsys_sem, sem, _semaphore);
if (!_semaphore)
return;
- sem = (struct radeon_winsys_sem *)_semaphore;
device->ws->destroy_sem(sem);
}
return image->surface.tiling_index[level];
}
+static uint32_t radv_surface_layer_count(struct radv_image_view *iview)
+{
+ return iview->type == VK_IMAGE_VIEW_TYPE_3D ? iview->extent.depth : iview->layer_count;
+}
+
static void
radv_initialise_color_surface(struct radv_device *device,
struct radv_color_buffer_info *cb,
va += iview->image->dcc_offset;
cb->cb_dcc_base = va >> 8;
- uint32_t max_slice = iview->type == VK_IMAGE_VIEW_TYPE_3D ? iview->extent.depth : iview->layer_count;
+ uint32_t max_slice = radv_surface_layer_count(iview);
cb->cb_color_view = S_028C6C_SLICE_START(iview->base_layer) |
S_028C6C_SLICE_MAX(iview->base_layer + max_slice - 1);
cb->cb_color_info |= S_028C70_COMPRESSION(1);
if (iview->image->cmask.size &&
- (device->debug_flags & RADV_DEBUG_FAST_CLEARS))
+ !(device->debug_flags & RADV_DEBUG_NO_FAST_CLEARS))
cb->cb_color_info |= S_028C70_FAST_CLEAR(1);
if (iview->image->surface.dcc_size && level_info->dcc_enabled)
z_offs += iview->image->surface.level[level].offset;
s_offs += iview->image->surface.stencil_level[level].offset;
- uint32_t max_slice = iview->type == VK_IMAGE_VIEW_TYPE_3D ? iview->extent.depth : iview->layer_count;
+ uint32_t max_slice = radv_surface_layer_count(iview);
ds->db_depth_view = S_028008_SLICE_START(iview->base_layer) |
S_028008_SLICE_MAX(iview->base_layer + max_slice - 1);
ds->db_depth_info = S_02803C_ADDR5_SWIZZLE_MASK(1);
ds->db_stencil_info |= S_028044_TILE_MODE_INDEX(tile_mode_index);
}
- if (iview->image->htile.size && !level) {
+ if (iview->image->surface.htile_size && !level) {
ds->db_z_info |= S_028040_TILE_SURFACE_ENABLE(1) |
S_028040_ALLOW_EXPCLEAR(1);
ds->db_stencil_info |= S_028044_TILE_STENCIL_DISABLE(1);
va = device->ws->buffer_get_va(iview->bo) + iview->image->offset +
- iview->image->htile.offset;
+ iview->image->htile_offset;
ds->db_htile_data_base = va >> 8;
ds->db_htile_surface = S_028ABC_FULL_CACHE(1);
} else {
return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
framebuffer->attachment_count = pCreateInfo->attachmentCount;
+ framebuffer->width = pCreateInfo->width;
+ framebuffer->height = pCreateInfo->height;
+ framebuffer->layers = pCreateInfo->layers;
for (uint32_t i = 0; i < pCreateInfo->attachmentCount; i++) {
VkImageView _iview = pCreateInfo->pAttachments[i];
struct radv_image_view *iview = radv_image_view_from_handle(_iview);
} else if (iview->aspect_mask & (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) {
radv_initialise_ds_surface(device, &framebuffer->attachments[i].ds, iview);
}
+ framebuffer->width = MIN2(framebuffer->width, iview->extent.width);
+ framebuffer->height = MIN2(framebuffer->height, iview->extent.height);
+ framebuffer->layers = MIN2(framebuffer->layers, radv_surface_layer_count(iview));
}
- framebuffer->width = pCreateInfo->width;
- framebuffer->height = pCreateInfo->height;
- framebuffer->layers = pCreateInfo->layers;
-
*pFramebuffer = radv_framebuffer_to_handle(framebuffer);
return VK_SUCCESS;
}
S_008F38_XY_MAG_FILTER(radv_tex_filter(pCreateInfo->magFilter, max_aniso)) |
S_008F38_XY_MIN_FILTER(radv_tex_filter(pCreateInfo->minFilter, max_aniso)) |
S_008F38_MIP_FILTER(radv_tex_mipfilter(pCreateInfo->mipmapMode)) |
- S_008F38_MIP_POINT_PRECLAMP(1) |
+ S_008F38_MIP_POINT_PRECLAMP(0) |
S_008F38_DISABLE_LSB_CEIL(1) |
S_008F38_FILTER_PREC_FIX(1) |
S_008F38_ANISO_OVERRIDE(is_vi));
projects / mesa.git / blobdiff