[Mesa-dev] [PATCH 3/8] gallium/radeon: move setting VRAM|GTT into winsyses
Marek Olšák
maraeo at gmail.com
Fri Dec 1 20:19:35 UTC 2017
From: Marek Olšák <marek.olsak at amd.com>
The combined VRAM|GTT heap will be removed.
---
src/gallium/drivers/r600/r600_buffer_common.c | 14 --------------
src/gallium/drivers/radeon/r600_buffer_common.c | 14 --------------
src/gallium/winsys/amdgpu/drm/amdgpu_bo.c | 14 +++++++++++++-
src/gallium/winsys/radeon/drm/radeon_drm_bo.c | 7 +++++++
src/gallium/winsys/radeon/drm/radeon_drm_cs.c | 8 ++++++++
5 files changed, 28 insertions(+), 29 deletions(-)
diff --git a/src/gallium/drivers/r600/r600_buffer_common.c b/src/gallium/drivers/r600/r600_buffer_common.c
index 5ff25ae..501b96f 100644
--- a/src/gallium/drivers/r600/r600_buffer_common.c
+++ b/src/gallium/drivers/r600/r600_buffer_common.c
@@ -169,34 +169,20 @@ void r600_init_resource_fields(struct r600_common_screen *rscreen,
/* Only displayable single-sample textures can be shared between
* processes. */
if (res->b.b.target == PIPE_BUFFER ||
res->b.b.nr_samples >= 2 ||
(rtex->surface.micro_tile_mode != RADEON_MICRO_MODE_DISPLAY &&
/* Raven doesn't use display micro mode for 32bpp, so check this: */
!(res->b.b.bind & PIPE_BIND_SCANOUT)))
res->flags |= RADEON_FLAG_NO_INTERPROCESS_SHARING;
- /* If VRAM is just stolen system memory, allow both VRAM and
- * GTT, whichever has free space. If a buffer is evicted from
- * VRAM to GTT, it will stay there.
- *
- * DRM 3.6.0 has good BO move throttling, so we can allow VRAM-only
- * placements even with a low amount of stolen VRAM.
- */
- if (!rscreen->info.has_dedicated_vram &&
- (rscreen->info.drm_major < 3 || rscreen->info.drm_minor < 6) &&
- res->domains == RADEON_DOMAIN_VRAM) {
- res->domains = RADEON_DOMAIN_VRAM_GTT;
- res->flags &= ~RADEON_FLAG_NO_CPU_ACCESS; /* disallowed with VRAM_GTT */
- }
-
if (rscreen->debug_flags & DBG_NO_WC)
res->flags &= ~RADEON_FLAG_GTT_WC;
if (res->b.b.bind & PIPE_BIND_SHARED)
res->flags |= RADEON_FLAG_NO_SUBALLOC;
/* Set expected VRAM and GART usage for the buffer. */
res->vram_usage = 0;
res->gart_usage = 0;
diff --git a/src/gallium/drivers/radeon/r600_buffer_common.c b/src/gallium/drivers/radeon/r600_buffer_common.c
index d162eea..ec282d5 100644
--- a/src/gallium/drivers/radeon/r600_buffer_common.c
+++ b/src/gallium/drivers/radeon/r600_buffer_common.c
@@ -164,34 +164,20 @@ void si_init_resource_fields(struct si_screen *sscreen,
res->flags |= RADEON_FLAG_NO_CPU_ACCESS |
RADEON_FLAG_GTT_WC;
}
/* Displayable and shareable surfaces are not suballocated. */
if (res->b.b.bind & (PIPE_BIND_SHARED | PIPE_BIND_SCANOUT))
res->flags |= RADEON_FLAG_NO_SUBALLOC; /* shareable */
else
res->flags |= RADEON_FLAG_NO_INTERPROCESS_SHARING;
- /* If VRAM is just stolen system memory, allow both VRAM and
- * GTT, whichever has free space. If a buffer is evicted from
- * VRAM to GTT, it will stay there.
- *
- * DRM 3.6.0 has good BO move throttling, so we can allow VRAM-only
- * placements even with a low amount of stolen VRAM.
- */
- if (!sscreen->info.has_dedicated_vram &&
- (sscreen->info.drm_major < 3 || sscreen->info.drm_minor < 6) &&
- res->domains == RADEON_DOMAIN_VRAM) {
- res->domains = RADEON_DOMAIN_VRAM_GTT;
- res->flags &= ~RADEON_FLAG_NO_CPU_ACCESS; /* disallowed with VRAM_GTT */
- }
-
if (sscreen->debug_flags & DBG(NO_WC))
res->flags &= ~RADEON_FLAG_GTT_WC;
/* Set expected VRAM and GART usage for the buffer. */
res->vram_usage = 0;
res->gart_usage = 0;
res->max_forced_staging_uploads = 0;
res->b.max_forced_staging_uploads = 0;
if (res->domains & RADEON_DOMAIN_VRAM) {
diff --git a/src/gallium/winsys/amdgpu/drm/amdgpu_bo.c b/src/gallium/winsys/amdgpu/drm/amdgpu_bo.c
index 6ec7cb7..9ab8f67 100644
--- a/src/gallium/winsys/amdgpu/drm/amdgpu_bo.c
+++ b/src/gallium/winsys/amdgpu/drm/amdgpu_bo.c
@@ -379,36 +379,48 @@ static struct amdgpu_winsys_bo *amdgpu_create_bo(struct amdgpu_winsys *ws,
unsigned pb_cache_bucket)
{
struct amdgpu_bo_alloc_request request = {0};
amdgpu_bo_handle buf_handle;
uint64_t va = 0;
struct amdgpu_winsys_bo *bo;
amdgpu_va_handle va_handle;
unsigned va_gap_size;
int r;
- assert(initial_domain & RADEON_DOMAIN_VRAM_GTT);
+ /* VRAM or GTT must be specified, but not both at the same time. */
+ assert(util_bitcount(initial_domain & RADEON_DOMAIN_VRAM_GTT) == 1);
+
bo = CALLOC_STRUCT(amdgpu_winsys_bo);
if (!bo) {
return NULL;
}
pb_cache_init_entry(&ws->bo_cache, &bo->u.real.cache_entry, &bo->base,
pb_cache_bucket);
request.alloc_size = size;
request.phys_alignment = alignment;
if (initial_domain & RADEON_DOMAIN_VRAM)
request.preferred_heap |= AMDGPU_GEM_DOMAIN_VRAM;
if (initial_domain & RADEON_DOMAIN_GTT)
request.preferred_heap |= AMDGPU_GEM_DOMAIN_GTT;
+ /* If VRAM is just stolen system memory, allow both VRAM and
+ * GTT, whichever has free space. If a buffer is evicted from
+ * VRAM to GTT, it will stay there.
+ *
+ * DRM 3.6.0 has good BO move throttling, so we can allow VRAM-only
+ * placements even with a low amount of stolen VRAM.
+ */
+ if (!ws->info.has_dedicated_vram && ws->info.drm_minor < 6)
+ request.preferred_heap |= AMDGPU_GEM_DOMAIN_GTT;
+
if (flags & RADEON_FLAG_NO_CPU_ACCESS)
request.flags |= AMDGPU_GEM_CREATE_NO_CPU_ACCESS;
if (flags & RADEON_FLAG_GTT_WC)
request.flags |= AMDGPU_GEM_CREATE_CPU_GTT_USWC;
if (flags & RADEON_FLAG_NO_INTERPROCESS_SHARING &&
ws->info.drm_minor >= 20)
request.flags |= AMDGPU_GEM_CREATE_VM_ALWAYS_VALID;
r = amdgpu_bo_alloc(ws->dev, &request, &buf_handle);
if (r) {
diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_bo.c b/src/gallium/winsys/radeon/drm/radeon_drm_bo.c
index 87c3f1e..fc95a98 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_bo.c
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_bo.c
@@ -602,20 +602,27 @@ static struct radeon_bo *radeon_create_bo(struct radeon_drm_winsys *rws,
assert(initial_domains);
assert((initial_domains &
~(RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM)) == 0);
args.size = size;
args.alignment = alignment;
args.initial_domain = initial_domains;
args.flags = 0;
+ /* If VRAM is just stolen system memory, allow both VRAM and
+ * GTT, whichever has free space. If a buffer is evicted from
+ * VRAM to GTT, it will stay there.
+ */
+ if (!rws->info.has_dedicated_vram)
+ args.initial_domain |= RADEON_DOMAIN_GTT;
+
if (flags & RADEON_FLAG_GTT_WC)
args.flags |= RADEON_GEM_GTT_WC;
if (flags & RADEON_FLAG_NO_CPU_ACCESS)
args.flags |= RADEON_GEM_NO_CPU_ACCESS;
if (drmCommandWriteRead(rws->fd, DRM_RADEON_GEM_CREATE,
&args, sizeof(args))) {
fprintf(stderr, "radeon: Failed to allocate a buffer:\n");
fprintf(stderr, "radeon: size : %u bytes\n", size);
fprintf(stderr, "radeon: alignment : %u bytes\n", alignment);
diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_cs.c b/src/gallium/winsys/radeon/drm/radeon_drm_cs.c
index c3398d0..92452e4 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_cs.c
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_cs.c
@@ -331,20 +331,28 @@ static int radeon_lookup_or_add_slab_buffer(struct radeon_drm_cs *cs,
static unsigned radeon_drm_cs_add_buffer(struct radeon_winsys_cs *rcs,
struct pb_buffer *buf,
enum radeon_bo_usage usage,
enum radeon_bo_domain domains,
enum radeon_bo_priority priority)
{
struct radeon_drm_cs *cs = radeon_drm_cs(rcs);
struct radeon_bo *bo = (struct radeon_bo*)buf;
enum radeon_bo_domain added_domains;
+
+ /* If VRAM is just stolen system memory, allow both VRAM and
+ * GTT, whichever has free space. If a buffer is evicted from
+ * VRAM to GTT, it will stay there.
+ */
+ if (!cs->ws->info.has_dedicated_vram)
+ domains |= RADEON_DOMAIN_GTT;
+
enum radeon_bo_domain rd = usage & RADEON_USAGE_READ ? domains : 0;
enum radeon_bo_domain wd = usage & RADEON_USAGE_WRITE ? domains : 0;
struct drm_radeon_cs_reloc *reloc;
int index;
if (!bo->handle) {
index = radeon_lookup_or_add_slab_buffer(cs, bo);
if (index < 0)
return 0;
--
2.7.4
More information about the mesa-dev
mailing list