[Intel-gfx] [RFC v2] libdrm_intel: Add API for execbuf pad to size functionality
Tvrtko Ursulin
tvrtko.ursulin at linux.intel.com
Wed Apr 1 04:14:09 PDT 2015
From: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
New kernels add the ability to pad objects to specified size at execbuf time.
Add the drm_intel_bo_pad_to_size API via which this padded size can be set.
v2: Reject padded size smaller than object size. (Chris Wilson)
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
Cc: Chris Wilson <chris at chris-wilson.co.uk>
---
include/drm/i915_drm.h | 5 +++--
intel/intel_bufmgr.c | 6 ++++++
intel/intel_bufmgr.h | 2 ++
intel/intel_bufmgr_gem.c | 27 +++++++++++++++++++++++++--
intel/intel_bufmgr_priv.h | 17 +++++++++++++++++
5 files changed, 53 insertions(+), 4 deletions(-)
diff --git a/include/drm/i915_drm.h b/include/drm/i915_drm.h
index 0d07e6b..0476f09 100644
--- a/include/drm/i915_drm.h
+++ b/include/drm/i915_drm.h
@@ -662,10 +662,11 @@ struct drm_i915_gem_exec_object2 {
#define EXEC_OBJECT_NEEDS_FENCE (1<<0)
#define EXEC_OBJECT_NEEDS_GTT (1<<1)
#define EXEC_OBJECT_WRITE (1<<2)
-#define __EXEC_OBJECT_UNKNOWN_FLAGS -(EXEC_OBJECT_WRITE<<1)
+#define EXEC_OBJECT_PAD_TO_SIZE (1<<3)
+#define __EXEC_OBJECT_UNKNOWN_FLAGS -(EXEC_OBJECT_PAD_TO_SIZE<<1)
__u64 flags;
- __u64 rsvd1;
+ __u64 pad_to_size;
__u64 rsvd2;
};
diff --git a/intel/intel_bufmgr.c b/intel/intel_bufmgr.c
index 234cd13..a4a3e34 100644
--- a/intel/intel_bufmgr.c
+++ b/intel/intel_bufmgr.c
@@ -299,6 +299,12 @@ drm_intel_bo_references(drm_intel_bo *bo, drm_intel_bo *target_bo)
}
drm_public int
+drm_intel_bo_pad_to_size(drm_intel_bo *bo, uint64_t pad_to_size)
+{
+ return bo->bufmgr->bo_pad_to_size(bo, pad_to_size);
+}
+
+drm_public int
drm_intel_get_pipe_from_crtc_id(drm_intel_bufmgr *bufmgr, int crtc_id)
{
if (bufmgr->get_pipe_from_crtc_id)
diff --git a/intel/intel_bufmgr.h b/intel/intel_bufmgr.h
index 69d3743..2012f34 100644
--- a/intel/intel_bufmgr.h
+++ b/intel/intel_bufmgr.h
@@ -165,6 +165,8 @@ int drm_intel_bo_disable_reuse(drm_intel_bo *bo);
int drm_intel_bo_is_reusable(drm_intel_bo *bo);
int drm_intel_bo_references(drm_intel_bo *bo, drm_intel_bo *target_bo);
+int drm_intel_bo_pad_to_size(drm_intel_bo *bo, uint64_t pad_to_size);
+
/* drm_intel_bufmgr_gem.c */
drm_intel_bufmgr *drm_intel_bufmgr_gem_init(int fd, int batch_size);
drm_intel_bo *drm_intel_bo_gem_create_from_name(drm_intel_bufmgr *bufmgr,
diff --git a/intel/intel_bufmgr_gem.c b/intel/intel_bufmgr_gem.c
index c5c1600..f743fe3 100644
--- a/intel/intel_bufmgr_gem.c
+++ b/intel/intel_bufmgr_gem.c
@@ -255,6 +255,12 @@ struct _drm_intel_bo_gem {
drm_intel_aub_annotation *aub_annotations;
unsigned aub_annotation_count;
+
+ /**
+ * Size to pad the object to.
+ *
+ */
+ uint64_t pad_to_size;
};
static unsigned int
@@ -504,8 +510,8 @@ drm_intel_add_validate_buffer2(drm_intel_bo *bo, int need_fence)
bufmgr_gem->exec2_objects[index].alignment = 0;
bufmgr_gem->exec2_objects[index].offset = 0;
bufmgr_gem->exec_bos[index] = bo;
- bufmgr_gem->exec2_objects[index].flags = 0;
- bufmgr_gem->exec2_objects[index].rsvd1 = 0;
+ bufmgr_gem->exec2_objects[index].flags = bo_gem->pad_to_size ? EXEC_OBJECT_PAD_TO_SIZE : 0;
+ bufmgr_gem->exec2_objects[index].pad_to_size = bo_gem->pad_to_size;
bufmgr_gem->exec2_objects[index].rsvd2 = 0;
if (need_fence) {
bufmgr_gem->exec2_objects[index].flags |=
@@ -1198,6 +1204,8 @@ drm_intel_gem_bo_unreference_final(drm_intel_bo *bo, time_t time)
DBG("bo_unreference final: %d (%s)\n",
bo_gem->gem_handle, bo_gem->name);
+ bo_gem->pad_to_size = 0;
+
/* release memory associated with this object */
if (bo_gem->reloc_target_info) {
free(bo_gem->reloc_target_info);
@@ -1855,6 +1863,20 @@ do_bo_emit_reloc(drm_intel_bo *bo, uint32_t offset,
}
static int
+drm_intel_gem_bo_pad_to_size(drm_intel_bo *bo, uint64_t pad_to_size)
+{
+ drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *) bo;
+
+ if (pad_to_size < bo->size)
+ return -EINVAL;
+
+ if (pad_to_size > bo->size)
+ bo_gem->pad_to_size = pad_to_size;
+
+ return 0;
+}
+
+static int
drm_intel_gem_bo_emit_reloc(drm_intel_bo *bo, uint32_t offset,
drm_intel_bo *target_bo, uint32_t target_offset,
uint32_t read_domains, uint32_t write_domain)
@@ -3603,6 +3625,7 @@ drm_intel_bufmgr_gem_init(int fd, int batch_size)
bufmgr_gem->bufmgr.bo_subdata = drm_intel_gem_bo_subdata;
bufmgr_gem->bufmgr.bo_get_subdata = drm_intel_gem_bo_get_subdata;
bufmgr_gem->bufmgr.bo_wait_rendering = drm_intel_gem_bo_wait_rendering;
+ bufmgr_gem->bufmgr.bo_pad_to_size = drm_intel_gem_bo_pad_to_size;
bufmgr_gem->bufmgr.bo_emit_reloc = drm_intel_gem_bo_emit_reloc;
bufmgr_gem->bufmgr.bo_emit_reloc_fence = drm_intel_gem_bo_emit_reloc_fence;
bufmgr_gem->bufmgr.bo_pin = drm_intel_gem_bo_pin;
diff --git a/intel/intel_bufmgr_priv.h b/intel/intel_bufmgr_priv.h
index 59ebd18..ec46876 100644
--- a/intel/intel_bufmgr_priv.h
+++ b/intel/intel_bufmgr_priv.h
@@ -152,6 +152,23 @@ struct _drm_intel_bufmgr {
void (*destroy) (drm_intel_bufmgr *bufmgr);
/**
+ * Sets buffer total padded size when buffer is used by the GPU.
+ *
+ * This enables dynamic padding to be added without using any backing
+ * storage. For example handling GPU padding requirements for buffers
+ * allocated by an entity unaware of the same.
+ *
+ * Set padded size remains active until reset (to zero or actual object
+ * size).
+ *
+ * Returns 0 on success or an error code.
+ *
+ * \param bo Buffer to set total padded size for
+ * \param pad_to_size Total size in bytes of object plus padding
+ */
+ int (*bo_pad_to_size) (drm_intel_bo *bo, uint64_t pad_to_size);
+
+ /**
* Add relocation entry in reloc_buf, which will be updated with the
* target buffer's real offset on on command submission.
*
--
2.3.2
More information about the Intel-gfx
mailing list