ilo: replace domains by reloc flags

It is simpler and is supported by the kernel.  It cannot be used with
libdrm_intel yet though.

2 files changed, 43 insertions, 32 deletions

diff --git a/src/gallium/winsys/intel/drm/intel_drm_winsys.c b/src/gallium/winsys/intel/drm/intel_drm_winsys.c
index f94a1cb9f33..c8ae6c16a8c 100644
--- a/src/gallium/winsys/intel/drm/intel_drm_winsys.c
+++ b/src/gallium/winsys/intel/drm/intel_drm_winsys.c
@@ -195,12 +195,7 @@ intel_winsys_create_for_fd(int fd)
 
    /*
     * No need to implicitly set up a fence register for each non-linear reloc
-    * entry.  When a fence register is needed for a reloc entry,
-    * drm_intel_bo_emit_reloc_fence() will be called explicitly.
-    *
-    * intel_bo_add_reloc() currently lacks "bool fenced" for this to work.
-    * But we never need a fence register on GEN4+ so we do not need to worry
-    * about it yet.
+    * entry.  INTEL_RELOC_FENCE will be set on reloc entries that need them.
     */
    drm_intel_bufmgr_gem_enable_fenced_relocs(winsys->bufmgr);
 
@@ -266,10 +261,8 @@ intel_winsys_alloc_bo(struct intel_winsys *winsys,
                       enum intel_tiling_mode tiling,
                       unsigned long pitch,
                       unsigned long height,
-                      uint32_t initial_domain)
+                      bool cpu_init)
 {
-   const bool for_render =
-      (initial_domain & (INTEL_DOMAIN_RENDER | INTEL_DOMAIN_INSTRUCTION));
    const unsigned int alignment = 4096; /* always page-aligned */
    unsigned long size;
    drm_intel_bo *bo;
@@ -292,12 +285,12 @@ intel_winsys_alloc_bo(struct intel_winsys *winsys,
 
    size = pitch * height;
 
-   if (for_render) {
-      bo = drm_intel_bo_alloc_for_render(winsys->bufmgr,
-            name, size, alignment);
+   if (cpu_init) {
+      bo = drm_intel_bo_alloc(winsys->bufmgr, name, size, alignment);
    }
    else {
-      bo = drm_intel_bo_alloc(winsys->bufmgr, name, size, alignment);
+      bo = drm_intel_bo_alloc_for_render(winsys->bufmgr,
+            name, size, alignment);
    }
 
    if (bo && tiling != INTEL_TILING_NONE) {
@@ -558,14 +551,37 @@ intel_bo_pread(struct intel_bo *bo, unsigned long offset,
 int
 intel_bo_add_reloc(struct intel_bo *bo, uint32_t offset,
                    struct intel_bo *target_bo, uint32_t target_offset,
-                   uint32_t read_domains, uint32_t write_domain,
-                   uint64_t *presumed_offset)
+                   uint32_t flags, uint64_t *presumed_offset)
 {
+   uint32_t read_domains, write_domain;
    int err;
 
-   err = drm_intel_bo_emit_reloc(gem_bo(bo), offset,
-         gem_bo(target_bo), target_offset,
-         read_domains, write_domain);
+   if (flags & INTEL_RELOC_WRITE) {
+      /*
+       * Because of the translation to domains, INTEL_RELOC_GGTT should only
+       * be set on GEN6 when the bo is written by MI_* or PIPE_CONTROL.  The
+       * kernel will translate it back to INTEL_RELOC_GGTT.
+       */
+      write_domain = (flags & INTEL_RELOC_GGTT) ?
+         I915_GEM_DOMAIN_INSTRUCTION : I915_GEM_DOMAIN_RENDER;
+      read_domains = write_domain;
+   } else {
+      write_domain = 0;
+      read_domains = I915_GEM_DOMAIN_RENDER |
+                     I915_GEM_DOMAIN_SAMPLER |
+                     I915_GEM_DOMAIN_INSTRUCTION |
+                     I915_GEM_DOMAIN_VERTEX;
+   }
+
+   if (flags & INTEL_RELOC_FENCE) {
+      err = drm_intel_bo_emit_reloc_fence(gem_bo(bo), offset,
+            gem_bo(target_bo), target_offset,
+            read_domains, write_domain);
+   } else {
+      err = drm_intel_bo_emit_reloc(gem_bo(bo), offset,
+            gem_bo(target_bo), target_offset,
+            read_domains, write_domain);
+   }
 
    *presumed_offset = gem_bo(target_bo)->offset64 + target_offset;
 
diff --git a/src/gallium/winsys/intel/intel_winsys.h b/src/gallium/winsys/intel/intel_winsys.h
index 679f24817b8..bc18bb69be5 100644
--- a/src/gallium/winsys/intel/intel_winsys.h
+++ b/src/gallium/winsys/intel/intel_winsys.h
@@ -44,14 +44,10 @@ enum intel_exec_flag {
 };
 
 /* this is compatible with i915_drm.h's definitions */
-enum intel_domain_flag {
-   INTEL_DOMAIN_CPU           = 0x00000001,
-   INTEL_DOMAIN_RENDER        = 0x00000002,
-   INTEL_DOMAIN_SAMPLER       = 0x00000004,
-   INTEL_DOMAIN_COMMAND	      = 0x00000008,
-   INTEL_DOMAIN_INSTRUCTION   = 0x00000010,
-   INTEL_DOMAIN_VERTEX        = 0x00000020,
-   INTEL_DOMAIN_GTT           = 0x00000040,
+enum intel_reloc_flag {
+   INTEL_RELOC_FENCE          = 1 << 0,
+   INTEL_RELOC_GGTT           = 1 << 1,
+   INTEL_RELOC_WRITE          = 1 << 2,
 };
 
 /* this is compatible with i915_drm.h's definitions */
@@ -126,7 +122,7 @@ intel_winsys_read_reg(struct intel_winsys *winsys,
  * \param tiling           Tiling mode.
  * \param pitch            Pitch of the bo.
  * \param height           Height of the bo.
- * \param initial_domain   Initial (write) domain.
+ * \param cpu_init         Will be initialized by CPU.
  */
 struct intel_bo *
 intel_winsys_alloc_bo(struct intel_winsys *winsys,
@@ -134,7 +130,7 @@ intel_winsys_alloc_bo(struct intel_winsys *winsys,
                       enum intel_tiling_mode tiling,
                       unsigned long pitch,
                       unsigned long height,
-                      uint32_t initial_domain);
+                      bool cpu_init);
 
 /**
  * Allocate a linear buffer object.
@@ -143,10 +139,10 @@ static inline struct intel_bo *
 intel_winsys_alloc_buffer(struct intel_winsys *winsys,
                           const char *name,
                           unsigned long size,
-                          uint32_t initial_domain)
+                          bool cpu_init)
 {
    return intel_winsys_alloc_bo(winsys, name,
-         INTEL_TILING_NONE, size, 1, initial_domain);
+         INTEL_TILING_NONE, size, 1, cpu_init);
 }
 
 /**
@@ -275,8 +271,7 @@ intel_bo_pread(struct intel_bo *bo, unsigned long offset,
 int
 intel_bo_add_reloc(struct intel_bo *bo, uint32_t offset,
                    struct intel_bo *target_bo, uint32_t target_offset,
-                   uint32_t read_domains, uint32_t write_domain,
-                   uint64_t *presumed_offset);
+                   uint32_t flags, uint64_t *presumed_offset);
 
 /**
  * Return the current number of relocations.