anv: Support multiple engines with DRM_IOCTL_I915_GEM_CONTEXT_CREATE_EXT

v2 (Jason Ekstrand):
 - Separate the anv_gem interface from anv_queue internals
 - Rework on top of the new anv_queue_family stuff

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8667>
This commit is contained in:
Jordan Justen
2019-03-24 01:00:37 -07:00
committed by Marge Bot
parent c5e7c91487
commit 4656be70dd
4 changed files with 165 additions and 14 deletions

View File

@@ -299,14 +299,37 @@ anv_physical_device_free_disk_cache(struct anv_physical_device *device)
static void
anv_physical_device_init_queue_families(struct anv_physical_device *pdevice)
{
pdevice->queue.family_count = 1;
pdevice->queue.families[0] = (struct anv_queue_family) {
.queueFlags = VK_QUEUE_GRAPHICS_BIT |
VK_QUEUE_COMPUTE_BIT |
VK_QUEUE_TRANSFER_BIT,
.queueCount = 1,
.engine_class = I915_ENGINE_CLASS_RENDER,
};
uint32_t family_count = 0;
if (pdevice->engine_info) {
int render_count = anv_gem_count_engines(pdevice->engine_info,
I915_ENGINE_CLASS_RENDER);
if (render_count > 0) {
pdevice->queue.families[family_count++] = (struct anv_queue_family) {
.queueFlags = VK_QUEUE_GRAPHICS_BIT |
VK_QUEUE_COMPUTE_BIT |
VK_QUEUE_TRANSFER_BIT,
.queueCount = render_count,
.engine_class = I915_ENGINE_CLASS_RENDER,
};
}
/* Increase count below when other families are added as a reminder to
* increase the ANV_MAX_QUEUE_FAMILIES value.
*/
STATIC_ASSERT(ANV_MAX_QUEUE_FAMILIES >= 1);
} else {
/* Default to a single render queue */
pdevice->queue.families[family_count++] = (struct anv_queue_family) {
.queueFlags = VK_QUEUE_GRAPHICS_BIT |
VK_QUEUE_COMPUTE_BIT |
VK_QUEUE_TRANSFER_BIT,
.queueCount = 1,
.engine_class = I915_ENGINE_CLASS_RENDER,
};
family_count = 1;
}
assert(family_count <= ANV_MAX_QUEUE_FAMILIES);
pdevice->queue.family_count = family_count;
}
static VkResult
@@ -2864,7 +2887,35 @@ VkResult anv_CreateDevice(
goto fail_device;
}
device->context_id = anv_gem_create_context(device);
uint32_t num_queues = 0;
for (uint32_t i = 0; i < pCreateInfo->queueCreateInfoCount; i++)
num_queues += pCreateInfo->pQueueCreateInfos[i].queueCount;
if (device->physical->engine_info) {
/* The kernel API supports at most 64 engines */
assert(num_queues <= 64);
uint16_t engine_classes[64];
int engine_count = 0;
for (uint32_t i = 0; i < pCreateInfo->queueCreateInfoCount; i++) {
const VkDeviceQueueCreateInfo *queueCreateInfo =
&pCreateInfo->pQueueCreateInfos[i];
assert(queueCreateInfo->queueFamilyIndex <
physical_device->queue.family_count);
struct anv_queue_family *queue_family =
&physical_device->queue.families[queueCreateInfo->queueFamilyIndex];
for (uint32_t j = 0; j < queueCreateInfo->queueCount; j++)
engine_classes[engine_count++] = queue_family->engine_class;
}
device->context_id =
anv_gem_create_context_engines(device,
physical_device->engine_info,
engine_count, engine_classes);
} else {
assert(num_queues == 1);
device->context_id = anv_gem_create_context(device);
}
if (device->context_id == -1) {
result = vk_error(VK_ERROR_INITIALIZATION_FAILED);
goto fail_fd;
@@ -2872,10 +2923,6 @@ VkResult anv_CreateDevice(
device->has_thread_submit = physical_device->has_thread_submit;
uint32_t num_queues = 0;
for (uint32_t i = 0; i < pCreateInfo->queueCreateInfoCount; i++)
num_queues += pCreateInfo->pQueueCreateInfos[i].queueCount;
device->queues =
vk_zalloc(&device->vk.alloc, num_queues * sizeof(*device->queues), 8,
VK_SYSTEM_ALLOCATION_SCOPE_DEVICE);
@@ -2890,8 +2937,15 @@ VkResult anv_CreateDevice(
&pCreateInfo->pQueueCreateInfos[i];
for (uint32_t j = 0; j < queueCreateInfo->queueCount; j++) {
/* When using legacy contexts, we use I915_EXEC_RENDER but, with
* engine-based contexts, the bottom 6 bits of exec_flags are used
* for the engine ID.
*/
uint32_t exec_flags = device->physical->engine_info ?
device->queue_count : I915_EXEC_RENDER;
result = anv_queue_init(device, &device->queues[device->queue_count],
I915_EXEC_RENDER, queueCreateInfo);
exec_flags, queueCreateInfo);
if (result != VK_SUCCESS)
goto fail_queues;

View File

@@ -374,6 +374,90 @@ anv_gem_create_context(struct anv_device *device)
return create.ctx_id;
}
int
anv_gem_create_context_engines(struct anv_device *device,
const struct drm_i915_query_engine_info *info,
int num_engines, uint16_t *engine_classes)
{
const size_t engine_inst_sz = 2 * sizeof(__u16); /* 1 class, 1 instance */
const size_t engines_param_size =
sizeof(__u64) /* extensions */ + num_engines * engine_inst_sz;
void *engines_param = malloc(engines_param_size);
assert(engines_param);
*(__u64*)engines_param = 0;
__u16 *class_inst_ptr = (__u16*)(((__u64*)engines_param) + 1);
/* For each type of drm_i915_gem_engine_class of interest, we keep track of
* the previous engine instance used.
*/
int last_engine_idx[] = {
[I915_ENGINE_CLASS_RENDER] = -1,
};
int i915_engine_counts[] = {
[I915_ENGINE_CLASS_RENDER] =
anv_gem_count_engines(info, I915_ENGINE_CLASS_RENDER),
};
/* For each queue, we look for the next instance that matches the class we
* need.
*/
for (int i = 0; i < num_engines; i++) {
uint16_t engine_class = engine_classes[i];
if (i915_engine_counts[engine_class] <= 0) {
free(engines_param);
return -1;
}
/* Run through the engines reported by the kernel looking for the next
* matching instance. We loop in case we want to create multiple
* contexts on an engine instance.
*/
int engine_instance = -1;
for (int i = 0; i < info->num_engines; i++) {
int *idx = &last_engine_idx[engine_class];
if (++(*idx) >= info->num_engines)
*idx = 0;
if (info->engines[*idx].engine.engine_class == engine_class) {
engine_instance = info->engines[*idx].engine.engine_instance;
break;
}
}
if (engine_instance < 0) {
free(engines_param);
return -1;
}
*class_inst_ptr++ = engine_class;
*class_inst_ptr++ = engine_instance;
}
assert((uintptr_t)engines_param + engines_param_size ==
(uintptr_t)class_inst_ptr);
struct drm_i915_gem_context_create_ext_setparam set_engines = {
.base = {
.name = I915_CONTEXT_CREATE_EXT_SETPARAM,
},
.param = {
.param = I915_CONTEXT_PARAM_ENGINES,
.value = (uintptr_t)engines_param,
.size = engines_param_size,
}
};
struct drm_i915_gem_context_create_ext create = {
.flags = I915_CONTEXT_CREATE_FLAGS_USE_EXTENSIONS,
.extensions = (uintptr_t)&set_engines,
};
int ret = gen_ioctl(device->fd, DRM_IOCTL_I915_GEM_CONTEXT_CREATE_EXT, &create);
free(engines_param);
if (ret == -1)
return -1;
return create.ctx_id;
}
int
anv_gem_destroy_context(struct anv_device *device, int context)
{

View File

@@ -252,6 +252,15 @@ anv_i915_query(int fd, uint64_t query_id, void *buffer,
unreachable("Unused");
}
int
anv_gem_create_context_engines(struct anv_device *device,
const struct drm_i915_query_engine_info *info,
int num_engines,
uint16_t *engine_classes)
{
unreachable("Unused");
}
struct drm_i915_query_engine_info *
anv_gem_get_engine_info(int fd)
{

View File

@@ -1619,6 +1619,10 @@ int anv_gem_execbuffer(struct anv_device *device,
int anv_gem_set_tiling(struct anv_device *device, uint32_t gem_handle,
uint32_t stride, uint32_t tiling);
int anv_gem_create_context(struct anv_device *device);
int anv_gem_create_context_engines(struct anv_device *device,
const struct drm_i915_query_engine_info *info,
int num_engines,
uint16_t *engine_classes);
bool anv_gem_has_context_priority(int fd);
int anv_gem_destroy_context(struct anv_device *device, int context);
int anv_gem_set_context_param(int fd, int context, uint32_t param,