winsys/radeon: move GEM domains out of the drivers into winsys

The drivers don't need to care about the domains. All they need to set
are the bind and usage flags. This simplifies the winsys too.

This also fixes on r600g:
- fbo-depth-GL_DEPTH_COMPONENT32F-copypixels
- fbo-depth-GL_DEPTH_COMPONENT16-copypixels
- fbo-depth-GL_DEPTH_COMPONENT24-copypixels
- fbo-depth-GL_DEPTH_COMPONENT32-copypixels
- fbo-depth-GL_DEPTH24_STENCIL8-copypixels
I can't explain it.

Reviewed-by: Alex Deucher <[email protected]>

4 files changed, 58 insertions, 40 deletions

diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_bo.c b/src/gallium/winsys/radeon/drm/radeon_drm_bo.c
index 20139b42c68..2dca3b9d958 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_bo.c
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_bo.c
@@ -336,6 +336,12 @@ static struct pb_buffer *radeon_bomgr_create_bo(struct pb_manager *_mgr,
     struct drm_radeon_gem_create args = {};
     struct radeon_bo_desc *rdesc = (struct radeon_bo_desc*)desc;
 
+    assert(rdesc->initial_domains && rdesc->reloc_domains);
+    assert((rdesc->initial_domains &
+            ~(RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM)) == 0);
+    assert((rdesc->reloc_domains &
+            ~(RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM)) == 0);
+
     args.size = size;
     args.alignment = desc->alignment;
     args.initial_domain = rdesc->initial_domains;
@@ -361,6 +367,7 @@ static struct pb_buffer *radeon_bomgr_create_bo(struct pb_manager *_mgr,
     bo->mgr = mgr;
     bo->rws = mgr->rws;
     bo->handle = args.handle;
+    bo->reloc_domains = rdesc->reloc_domains;
     pipe_mutex_init(bo->map_mutex);
 
     return &bo->base;
@@ -505,8 +512,7 @@ static struct pb_buffer *
 radeon_winsys_bo_create(struct radeon_winsys *rws,
                         unsigned size,
                         unsigned alignment,
-                        unsigned bind,
-                        enum radeon_bo_domain domain)
+                        unsigned bind, unsigned usage)
 {
     struct radeon_drm_winsys *ws = radeon_drm_winsys(rws);
     struct radeon_bo_desc desc;
@@ -515,8 +521,27 @@ radeon_winsys_bo_create(struct radeon_winsys *rws,
 
     memset(&desc, 0, sizeof(desc));
     desc.base.alignment = alignment;
-    desc.base.usage = domain;
-    desc.initial_domains = domain;
+
+    /* Determine the memory domains. */
+    switch (usage) {
+    case PIPE_USAGE_STAGING:
+    case PIPE_USAGE_STREAM:
+    case PIPE_USAGE_DYNAMIC:
+            desc.initial_domains = RADEON_GEM_DOMAIN_GTT;
+            desc.reloc_domains = RADEON_GEM_DOMAIN_GTT;
+            break;
+    default:
+            if (bind & (PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER |
+                        PIPE_BIND_CONSTANT_BUFFER)) {
+                desc.initial_domains = RADEON_GEM_DOMAIN_GTT;
+            } else {
+                desc.initial_domains = RADEON_GEM_DOMAIN_VRAM;
+            }
+            desc.reloc_domains = RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM;
+    }
+
+    /* Additional criteria for the cache manager. */
+    desc.base.usage = desc.initial_domains;
 
     /* Assign a buffer manager. */
     if (bind & (PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER |
@@ -572,6 +597,7 @@ static struct pb_buffer *radeon_winsys_bo_from_handle(struct radeon_winsys *rws,
     }
     bo->handle = open_arg.handle;
     bo->name = whandle->handle;
+    bo->reloc_domains = RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM;
 
     /* Initialize it. */
     pipe_reference_init(&bo->base.reference, 1);
diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_bo.h b/src/gallium/winsys/radeon/drm/radeon_drm_bo.h
index a3dbde76dcd..ba71cfb3440 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_bo.h
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_bo.h
@@ -42,11 +42,13 @@ struct radeon_bo_desc {
     struct pb_desc base;
 
     unsigned initial_domains;
+    unsigned reloc_domains;
 };
 
 struct radeon_bo {
     struct pb_buffer base;
 
+    /* Don't move these! */
     unsigned last_flush;
     unsigned binding;
 
@@ -56,6 +58,7 @@ struct radeon_bo {
     void *ptr;
     pipe_mutex map_mutex;
 
+    uint32_t reloc_domains;
     uint32_t handle;
     uint32_t name;
 
diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_cs.c b/src/gallium/winsys/radeon/drm/radeon_drm_cs.c
index a3d0be5c1f3..8d5a6b33a7e 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_cs.c
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_cs.c
@@ -170,15 +170,14 @@ static struct radeon_winsys_cs *radeon_drm_cs_create(struct radeon_winsys *rws)
 
 #define OUT_CS(cs, value) (cs)->buf[(cs)->cdw++] = (value)
 
-static INLINE void update_domains(struct drm_radeon_cs_reloc *reloc,
-                                  enum radeon_bo_domain rd,
-                                  enum radeon_bo_domain wd,
-                                  enum radeon_bo_domain *added_domains)
+static INLINE void update_reloc_domains(struct drm_radeon_cs_reloc *reloc,
+                                        enum radeon_bo_usage usage,
+                                        unsigned domains)
 {
-    *added_domains = (rd | wd) & ~(reloc->read_domains | reloc->write_domain);
-
-    reloc->read_domains |= rd;
-    reloc->write_domain |= wd;
+    if (usage & RADEON_USAGE_READ)
+        reloc->read_domains |= domains;
+    if (usage & RADEON_USAGE_WRITE)
+        reloc->write_domain |= domains;
 }
 
 int radeon_get_reloc(struct radeon_cs_context *csc, struct radeon_bo *bo)
@@ -220,9 +219,8 @@ int radeon_get_reloc(struct radeon_cs_context *csc, struct radeon_bo *bo)
 
 static unsigned radeon_add_reloc(struct radeon_cs_context *csc,
                                  struct radeon_bo *bo,
-                                 enum radeon_bo_domain rd,
-                                 enum radeon_bo_domain wd,
-                                 enum radeon_bo_domain *added_domains)
+                                 enum radeon_bo_usage usage,
+                                 unsigned *added_domains)
 {
     struct drm_radeon_cs_reloc *reloc;
     unsigned i;
@@ -231,7 +229,7 @@ static unsigned radeon_add_reloc(struct radeon_cs_context *csc,
     if (csc->is_handle_added[hash]) {
         reloc = csc->relocs_hashlist[hash];
         if (reloc->handle == bo->handle) {
-            update_domains(reloc, rd, wd, added_domains);
+            update_reloc_domains(reloc, usage, bo->reloc_domains);
             return csc->reloc_indices_hashlist[hash];
         }
 
@@ -240,7 +238,7 @@ static unsigned radeon_add_reloc(struct radeon_cs_context *csc,
             --i;
             reloc = &csc->relocs[i];
             if (reloc->handle == bo->handle) {
-                update_domains(reloc, rd, wd, added_domains);
+                update_reloc_domains(reloc, usage, bo->reloc_domains);
 
                 csc->relocs_hashlist[hash] = reloc;
                 csc->reloc_indices_hashlist[hash] = i;
@@ -270,8 +268,10 @@ static unsigned radeon_add_reloc(struct radeon_cs_context *csc,
     p_atomic_inc(&bo->num_cs_references);
     reloc = &csc->relocs[csc->crelocs];
     reloc->handle = bo->handle;
-    reloc->read_domains = rd;
-    reloc->write_domain = wd;
+    if (usage & RADEON_USAGE_READ)
+        reloc->read_domains = bo->reloc_domains;
+    if (usage & RADEON_USAGE_WRITE)
+        reloc->write_domain = bo->reloc_domains;
     reloc->flags = 0;
 
     csc->is_handle_added[hash] = TRUE;
@@ -280,24 +280,23 @@ static unsigned radeon_add_reloc(struct radeon_cs_context *csc,
 
     csc->chunks[1].length_dw += RELOC_DWORDS;
 
-    *added_domains = rd | wd;
+    *added_domains = bo->reloc_domains;
     return csc->crelocs++;
 }
 
 static unsigned radeon_drm_cs_add_reloc(struct radeon_winsys_cs *rcs,
                                         struct radeon_winsys_cs_handle *buf,
-                                        enum radeon_bo_domain rd,
-                                        enum radeon_bo_domain wd)
+                                        enum radeon_bo_usage usage)
 {
     struct radeon_drm_cs *cs = radeon_drm_cs(rcs);
     struct radeon_bo *bo = (struct radeon_bo*)buf;
-    enum radeon_bo_domain added_domains;
+    unsigned added_domains = 0;
 
-    unsigned index = radeon_add_reloc(cs->csc, bo, rd, wd, &added_domains);
+    unsigned index = radeon_add_reloc(cs->csc, bo, usage, &added_domains);
 
-    if (added_domains & RADEON_DOMAIN_GTT)
+    if (added_domains & RADEON_GEM_DOMAIN_GTT)
         cs->csc->used_gart += bo->base.size;
-    if (added_domains & RADEON_DOMAIN_VRAM)
+    if (added_domains & RADEON_GEM_DOMAIN_VRAM)
         cs->csc->used_vram += bo->base.size;
 
     return index;
diff --git a/src/gallium/winsys/radeon/drm/radeon_winsys.h b/src/gallium/winsys/radeon/drm/radeon_winsys.h
index 31d39e9bdcb..8bfbc6cd2ef 100644
--- a/src/gallium/winsys/radeon/drm/radeon_winsys.h
+++ b/src/gallium/winsys/radeon/drm/radeon_winsys.h
@@ -56,11 +56,6 @@ enum radeon_bo_layout {
     RADEON_LAYOUT_UNKNOWN
 };
 
-enum radeon_bo_domain { /* bitfield */
-    RADEON_DOMAIN_GTT  = 2,
-    RADEON_DOMAIN_VRAM = 4
-};
-
 enum radeon_bo_usage { /* bitfield */
     RADEON_USAGE_READ = 2,
     RADEON_USAGE_WRITE = 4,
@@ -140,14 +135,13 @@ struct radeon_winsys {
      * \param size      The size to allocate.
      * \param alignment An alignment of the buffer in memory.
      * \param bind      A bitmask of the PIPE_BIND_* flags.
-     * \param domain    A bitmask of the RADEON_DOMAIN_* flags.
+     * \param usage     A bitmask of the PIPE_USAGE_* flags.
      * \return          The created buffer object.
      */
     struct pb_buffer *(*buffer_create)(struct radeon_winsys *ws,
                                        unsigned size,
                                        unsigned alignment,
-                                       unsigned bind,
-                                       enum radeon_bo_domain domain);
+                                       unsigned bind, unsigned usage);
 
     struct radeon_winsys_cs_handle *(*buffer_get_cs_handle)(
             struct pb_buffer *buf);
@@ -275,14 +269,12 @@ struct radeon_winsys {
      *
      * \param cs  A command stream to add buffer for validation against.
      * \param buf A winsys buffer to validate.
-     * \param rd  A read domain containing a bitmask of the RADEON_DOMAIN_* flags.
-     * \param wd  A write domain containing a bitmask of the RADEON_DOMAIN_* flags.
+     * \param usage  Whether the buffer is used for read and/or write.
      * \return Relocation index.
      */
     unsigned (*cs_add_reloc)(struct radeon_winsys_cs *cs,
                              struct radeon_winsys_cs_handle *buf,
-                             enum radeon_bo_domain rd,
-                             enum radeon_bo_domain wd);
+                             enum radeon_bo_usage usage);
 
     /**
      * Return TRUE if there is enough memory in VRAM and GTT for the relocs
@@ -299,8 +291,6 @@ struct radeon_winsys {
      *
      * \param cs        A command stream the relocation is written to.
      * \param buf       A winsys buffer to write the relocation for.
-     * \param rd        A read domain containing a bitmask of the RADEON_DOMAIN_* flags.
-     * \param wd        A write domain containing a bitmask of the RADEON_DOMAIN_* flags.
      */
     void (*cs_write_reloc)(struct radeon_winsys_cs *cs,
                            struct radeon_winsys_cs_handle *buf);